Download as pdf or txt
Download as pdf or txt
You are on page 1of 334

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/282942564

Development of a Transmission-line Model Considering the Skin and Corona


Effects for Power Systems Transient Analysis

Thesis · March 1997

CITATIONS READS

26 1,015

1 author:

Taku Noda
Central Research Institute of Electric Power Industry
156 PUBLICATIONS 2,395 CITATIONS

SEE PROFILE

All content following this page was uploaded by Taku Noda on 18 October 2015.

The user has requested enhancement of the downloaded file.


DEVELOPMENT OF A TRANSMISSION-LINE MODEL

CONSIDERING THE SKIN AND CORONA EFFECTS

FOR POWER SYSTEMS TRANSIENT ANALYSIS

by

TAKU NODA

A Thesis Submitted to Doshisha University, Kyoto, Japan

for the Degree of Doctor of Electrical Engineering

December 10th, 1996

© Taku Noda
ACKNOWLEDGMENTS

The author is grateful to Profs. Akihiro Ametani, Naoto Nagaoka, and Mr.
Nobutaka Mori of Doshisha University, Kyoto, Japan for their valuable technical advice
and support all through the author’s work at Doshisha University. The author deeply
appreciates Laurent Dubé of DEI Simulation Software, who always provides important
suggestions and support. Especially, his support during author’s stay in Neskowin,
Oregon, USA in 1995 has opened the author’s eyes to the world-wide engineering
societies, and collaborations such as at Meidensha Corp., Tokyo, Japan give precious
experience of real industrial simulations. The author is also grateful to Drs. W.S. Meyer
and T.H. Liu of the Bonneville Power Administration (BPA) for their assistance
implementing the new line model in the ATP version of EMTP and also for their support
during author’s visits at BPA. Others at BPA who contributed include D.L
Goldsworthy, J.L. Hall, and R.M. Hassibar. Prof. M.T. Correia de Barros of Instituto
Superior Tecnico (IST) of the Technical University of Lisbon, Portugal has given the
start point of the corona model and gave the opportunity to participate the first
International Conference on Power Systems Transients ’95. Prof. M. Kizilcay of FH
Osnabrueck in Germany is thanked for his useful E-mail discussions. Dr. Shozo Sekioka
of Kansai Tech, Osaka, Japan is deeply thanked for providing the experimental setup of
the corona tests and for valuable discussions on surge phenomena. Others who
contributed as their undergraduate research include Mitsuhiro Kubota of the Chubu
Electric Power Company, Nobuya Miwa of Nihon Denso, Takashi Sawada of the
Fujikura Electric Line Company, Keisuke Fujii of the Kyoto City Office, Tamaki
Matsuura, Kouichi Yamabuki, Jun Takami, Naoko Nose, Koji Yoshida, and Kazuo
Yamamoto of Doshisha University. Finally, the author is deeply grateful to his family
and Masae Mikuni of the Tokyo Electric Power Company for their persevering support
all through the research activities and for ongoing software development projects.

Taku Noda
ABSTRACT

This thesis proposes a transmission-line model to analyze power system


transients. The line model accurately takes into account the frequency dependence and
the nonlinearity of a transmission line. The frequency dependence is due to the skin
effects of conductors and earth soil, and the nonlinearity is caused by corona discharge
along the conductors. To accurately evaluate the frequency-dependent parameters, a
closed-form formula of the overhead-line earth-return impedance is derived using a
double-exponential approximation. The formula is much simpler than the rigorous
Carson-Pollaczeck one, and far more accurate than Dubanton-Deri’s approximate
formula. In order to realize the frequency dependence in the time-domain, direct phase-
domain modeling is proposed, and lines are modeled in the phase domain rather than in
the modal domain. This avoids convolution due to the modal transformation and
possible numerical instability due to mode crossing. In this approach, time-domain
convolutions are replaced by an IARMA (Interpolated Auto-Regressive Moving
Average) model that minimizes computation, and modal-traveling-time differences in a
phase-domain response is directly modeled taking advantage of the one-sample-delay
nature of the Z-operator. The proposed phase-domain approach is fully investigated in
terms of the fitting, approximation order, stability, and steady-state initialization of the
line model. To study corona discharge on a transmission line, charge - voltage (q - v)
curves are measured using a novel measurement method that requires no corona cage.
The method has solved a problem that the presence of the corona cage disturbs the
electric field, and comprehensive data are accumulated and investigated using the method.
Coupling-factor curves to determine the increase of electrostatic coupling due to corona
are also measured and investigated for multiphase modeling. The transmission-line
corona is wave-steepness dependent, because the q - v curve of a lightning surge is
different from that of a switching surge. A wave-steepness-dependent corona model is
proposed, and it can express the dependence by a simple calculation procedure as
accurately as a rigorous finite-difference method of which the calculation time is
expensive. In order to introduce the corona model in the proposed line model by
discretization, an efficient method to deal with nonlinear corona branches using the
trapezoidal rule of integration and the predictor-corrector method is proposed. In this
thesis, the line model is fully validated compared with field tests and rigorous solutions.
TABLE OF CONTENTS

ACKNOWLEDGMENTS

ABSTRACT

1 INTRODUCTION 1-1
1.1 Series-Impedance Calculation 1-2
1.2 Time-Domain Realization of Frequency Dependence 1-3
1.3 Surge Corona Measurement 1-5
1.4 Corona Modeling 1-6
1.5 Inclusion of Corona Effects 1-7

2 FREQUENCY-DEPENDENT AND NONLINEAR WAVE EQUATIONS 2-1


DESCRIBING TRANSMISSION-LINE DYNAMICS

2.1 Transmission Lines: Overhead Lines and Underground Cables 2-2


2.2 Frequency-Dependent and Nonlinear Wave Equations 2-6

3 FREQUENCY-DEPENDENT SERIES-IMPEDANCE MATRIX 3-1


CONSIDERING SKIN EFFECTS

3.1 Skin Effects and Frequency-Dependent Series-Impedance Matrix 3-2


3.2 Earth-Return Impedance 3-7
3.2.1 Historical review of earth-return impedance formulas 3-7
3.2.2 Closed-form formula of an overhead line earth-return 3-15
impedance
A. Derivation 3-15
B. Numerical validation 3-22
3.3 Conductor-Internal Impedance 3-32

4 TIME-DOMAIN REALIZATION OF FREQUENCY DEPENDENCE 4-1

4.1 Modal-Domain Modeling and Problems of Frequency-Dependent 4-2


Modal-Transformation Matrices
4.2 Phase-Domain Formulation 4-10
4.3 Equivalent Circuit for Time-Domain Simulation 4-13
4.4 Consideration of Conductor Configuration Symmetry to Reduce Fitting 4-16
4.5 IARMA Convolution 4-18
4.5.1 Interpolated ARMA (IARMA) model 4-18
4.5.2 Time-step determination 4-21
4.5.3 Modeling of modal traveling-time differences 4-24
4.5.4 Linearized least-squares method for ARMA identification 4-27
A. Mathematical background 4-27
B. Relative-error evaluation 4-28
C. Samples in the z-domain 4-29
4.5.5 Stability check by Jury’s method 4-31
4.5.6 Order determination by AIC (Akaike’s Information Criterion) 4-32
4.5.7 Phase-domain matrix-vector convolution 4-33
4.6 Steady-State Initialization 4-35
4.6.1 Frequency response of individual IARMA model 4-36
4.6.2 Steady-state initialization of individual IARMA model 4-36
4.6.3 Admittance matrix of line model 4-37
4.6.4 Initialization of line-model internal state 4-38
4.7 Matrix Stability Criteria 4-39
4.8 Switching- and Fault-Surge Calculations 4-41
4.8.1 Switching-surge calculations 4-41
A. 500-kV single-circuit horizontal line 4-41
B. 500-kV double-circuit vertical line 4-48
C. 275-kV pipe-type oil-filled cable 4-56
4.8.2 Fault-surge calculations 4-62
A. 500-kV double-circuit vertical line 4-62
B. 1100-kV double-circuit vertical line 4-64
C. Asymmetrical double-circuit line 4-66
D. 400-kV bipolar HVDC line 4-68
E. 275-kV pipe-type oil-filled cable 4-70

5 NONLINEAR SHUNT-ADMITTANCE MATRIX CONSIDERING 5-1


CORONA EFFECTS

5.1 Corona Discharge and Nonlinear Shunt-Admittance Matrix 5-2


5.2 A Novel q - v Curve Measurement Method without Corona Cage 5-7
5.2.1 Conventional q - v curve measurement methods 5-7
5.2.2 Experimental setup 5-9
5.2.3 Postprocessing of measured data for generating q - v curves 5-18
5.2.4 Corona and leader development 5-20
5.2.5 Measured q - v curves 5-23
5.2.6 Effects of conductor radius 5-30
5.2.7 Effects of line length 5-39
5.2.8 Effects of adjacent conductor 5-42
5.3 Coupling-Factor Curve Measurement 5-45
5.3.1 Measured coupling-factor curves for multiphase cases 5-45
5.3.2 Effects of conductor separation and radius on 5-54
coupling-factor curves
5.4 Historical Review of Corona Models 5-62
5.5 Wave-Steepness-Dependent Corona Model 5-71
5.5.1 Calculation procedure of quasi-static q - v curve 5-73
5.5.2 Introduction of wave-steepness dependence 5-76
5.6 Simulation of q - v and Coupling-Factor Curves 5-81
5.6.1 Simulation of q - v curves 5-81
5.6.2 Simulation of coupling-factor curves 5-95
6 REALIZATION OF NONLINEARITY DUE TO CORONA 6-1
DISCHARGE

6.1 State-Space Approach to Deal with Nonlinear Corona Branches 6-2


6.1.1 Space discretization 6-2
6.1.2 Norton equivalent circuit of linear portions of line model 6-3
6.1.3 State-space equation of corona branch 6-5
6.1.4 Numerical solution by trapezoidal rule of integration and 6-6
predictor-corrector scheme
6.2 Effects of Corona on Wave Propagation 6-8
6.3 Wave-Propagation Simulations 6-14
6.3.1 Single-phase case 6-14
6.3.2 Multiphase case 6-15

7 CONCLUSIONS 7-1
APPENDIX A LINEARIZED LEAST-SQUARES METHOD A-1

APPENDIX B A PREDICTOR-CORRECTOR SCHEME FOR B-1


SOLVING A NONLINEAR CIRCUIT

B.1 Introduction B-2


B.2 Formulation of Nonlinear Circuit B-4
A. Optimum ordering of nodes B-4
B. Predictor-corrector iteration B-5
C. Relation with ODE formulation B-6
D. Comparison with compensation method B-8
E. Modeling of nonlinear component B-9
B.3 Examples B-10
A. Single-phase diode-bridge rectifier circuit B-10
B. Transistor switching circuit B-14
C. Transformer model B-17
B.4 Conclusions B-20

APPENDIX C INVESTIGATION OF LIGHTNING-INDUCED C-1


VOLTAGES TO ELECTRONIC APPLIANCE

C.1 Introduction C-2


C.2 Calculation Method C-3
A. Lightning-stroke model C-3
B. Equivalent circuit of printed circuit C-5
C. Vector potential generated by lightning stroke C-8
D. Induced voltage for arbitrary waveform current C-10
C.3 Experimental Setup C-11
A. Modeling of lightning-stroke current C-11
B. Modeling of electronic appliance by loop circuit C-14
C.4 Experimental and Calculated Results C-15
A. Induced voltage by step-function current C-15
B. Induced voltage by slow-front current C-16
C.5 Conclusions C-19

APPENDIX D LIST OF PAPERS PRESENTED BY THE AUTHOR D-1

_________________________________________________

In this thesis, MKSA units are employed throughout.


1 INTRODUCTION

Stable supply of electric power is extremely important as a basis of the modern


society, and thus the reliable operation of a power system is highly required. Nowadays,
even a momentary service interruption is not allowed, because electronic computers play
an inevitable role in the society. Insulation design, which includes lightning, switching,
and temporary overvoltage protections, is one of the most important technologies
supporting the reliability of a power system. To predict those overvoltages, transients
of the power system are calculated using a digital computer. Analogue computers had
been used until late sixties, but the use of a digital computer has become more common
due to its recent rapid progress. The Electro-Magnetic Transients Program (EMTP)
developed by the Bonneville Power Administration (BPA), US Department of Energy is
the most widely used computer program for the transient simulations [1]. Modern
insulation design especially tends to require accurate transient simulations using a
computer program such as the EMTP, in order to minimize the cost of construction. To
achieve an accurate simulation, each element of a power system has to be modeled
accurately in the program. Especially, the dynamics of a transmission line has to be
modeled accurately, because traveling waves on the line are a dominant cause of the
overvoltages.

The present thesis proposes an accurate transmission-line model for power


system transient calculations. Traveling waves on a transmission line are deformed by
the frequency dependence and the nonlinearity. The frequency dependence is due to the
skin effects of conductors and earth soil, and the nonlinearity is due to corona discharge
along the conductors. The proposed line model accurately takes into account both the
frequency dependence and the nonlinearity in time-domain simulations, especially in an
EMTP-type program.

1-1
1.1 Series-Impedance Calculation

For modeling the frequency dependence of a transmission line accurately, the


frequency response of line parameters, i.e. earth-return and conductor-internal imped-
ances, needs to be accurately evaluated. Also, the calculated frequency response should
not include numerical discontinuities with respect to frequency, which affect the time-
domain realization of the frequency dependence. Therefore, an accurate closed-form
formula has been desired.

The problem of wave propagation along an overhead wire parallel to the earth
surface was first consistently solved by J.R. Carson and F. Pollaczek [2,3], and an
accurate earth-return impedance was given in an infinite-integral form. Carson gave its
series expansion formula [2], and then the formula was rearranged to suite computer
implementation by R.H. Galloway et al. [4]. But Carson-Galloway’s expansion formula
is complicated and have different forms for large and small arguments, which cause a
numerical discontinuity [5]. Later, Dubanton proposed a simple approximate formula
without proofs, and the theoretical derivation was given by A. Deri et al. using the
concept of the complex ground-return plane [5]. In this thesis, a closed-form formula,
which can be regarded as practically exact, is derived applying a double-exponential
approximation to Carson-Pollaczek’s infinite-integral, and the derived formula is far more
accurate than Dubanton-Deri’s one in practical configurations. A physical
representation of the proposed double-exponential approximation is given using the
concept of the complex ground-return plane.

1-2
1.2 Time-Domain Realization of Frequency Dependence

In order to realize the frequency dependence in the time domain, Wedepohl’s


modal theory [6] has been used in EMTP for the distributed representation of transmis-
sion lines. For this reason, weighting functions [7], exponential recursive convolution
[8,11], linear recursive convolution [9], Z-transforms [10], and extended recursive
convolution [12,13] all have been proposed. In those methods that relate the phase and
modal domains by constant transformation matrices, the frequency dependence of the
matrices is ignored. But it should not be ignored for vertical overhead lines or
underground cables of which the modal-transformation matrices depend heavily on
frequency [14-16]. The frequency dependence of the transformation matrices was
originally introduced into a time-domain simulation using convolution by A. Ametani [17],
and later by L. Marti [18]. But the practical implementation is enormously complicated
in terms of eigenvalue tracing by mode crossing (at some frequency, two or more
eigenvalues become equal) [19]. On the other hand, H. Nakanishi and A. Ametani
proposed a phase-domain method rather than in the modal domain, so as to avoid
possible numerical instability due to mode crossing [20]. Following the method, direct
phase-domain models, which are based on a nodal-conductance representation, have
recently been proposed for purposing EMTP implementation [21-23]. Phase-domain
responses include discontinuities due to modal propagation in the time-domain, and the
most difficult part of the direct phase-domain modeling is to efficiently reproduce the
discontinuous responses using a recursive convolution. One of the solutions has been
given by the author [23-28] using an IARMA (Interpolated Auto-Regressive Moving
Average) convolution methodology [28], and implemented in the ATP version of EMTP
at BPA as the NODA SETUP [29]. The IARMA model accurately reproduces the
discontinuous responses taking advantage of the one-sample-delay nature of the Z-
operator, and greatly saves the computation time. To purpose an efficient procedure,
each convolution in the line model is allowed to use its own economical time step, and

1-3
thus, a slow response (long rise time) such as an earth-return effect can use a large time
step to avoid excessive computation, and a fast response (short rise time) can use a small
one not to lose its peak value. A linear interpolation technique was used to interface
with the external circuit, to allow a different time step for each convolution. The
convolution technique is equivalent to ARMA (Auto-Regressive Moving Average) model
of which the theory is well-established in the field of time-series modeling, and the
interpolation operation is considered to be a part of the ARMA model, so is called an
IARMA (Interpolated ARMA) model. With the interpolation, it could be used even for
accurate HVDC and power-electronic simulations which often require an on/off-timing
adjustment for switching devices [30]. A fast and stable optimization method to identify
the IARMA model is also developed, by applying Householder's transformation [31-35].
Because the procedure is linearized, the previous time-consuming, sometimes-unstable,
nonlinear optimizations are no longer required. The optimization method includes : 1)
the determination procedure of appropriate order of an IARMA model using the theory
of Akaike's Information Criterion (AIC) [36] ; and 2) the stability evaluation of each
IARMA model using Jury's method [37]. But even if all the IARMA elements of the
line model are stable, it is still possible that the line model behaves as an active circuit, and
thus as a numerically unstable model. To assure the complete numerical stability, the
real part of the admittance matrix of the line model has to be positive-definite [38] (matrix
stability criterion). But the criterion is difficult to be incorporated in the optimization
procedure, because the wave-deformation and the characteristic admittance matrices are
fitted separately in the procedure. Thus, each criterion for the wave-deformation matrix
and for the characteristic admittance matrix is derived in this thesis, for purposing robust
fitting [28].

In most cases, steady-state initialization is necessary to make simulations starting


from a steady state, e.g. fault-surge simulation, feasible, and a steady-state initialization
method of the line model is also described in this thesis [28,39].

Using the line model, switching-surge simulations are carried out and compared
with field tests and rigorous frequency-transform solutions on a 500-kV single-circuit

1-4
horizontal line, a 500-kV double-circuit vertical line, and a 275-kV pipe-type oil-filled
cable [23,40-42]. Also, fault-surge simulations are carried out and compared with
rigorous solutions on a 500-kV double-circuit vertical line, a 1000-kV double-circuit
vertical line, a double-circuit line with asymmetrical configuration, a 400-kV bipolar
HVDC line, and a 275-kV pipe-type oil-filled cable, especially using the steady-state
initialization method [28,39]. The above comprehensive comparison with field tests and
rigorous frequency-transform solutions shows, here without corona, the practical validity
of the line model.

1.3 Surge Corona Measurement

The inclusion of the nonlinearity due to corona discharge along conductors is


very important for overhead lines, when the conductor voltages exceed the corona onset
voltage. The corona discharge causes attenuation and distortion of a traveling wave.
Compared with the skin effects for the series-impedance matrix, the mechanism of the
corona discharge itself has not sufficiently been made clear. Therefore, experimental
investigations are quite important, and an accurate measurement technique to observe the
corona discharge has to be established as the basis. Normally, transmission-line corona
is characterized by a family of charge - voltage (q - v) curves, because the corona
discharge increases space charges around a conductor. Conventional q - v curve
measurement methods use a corona cage which surrounds the specimen conductor for
space-charge measuring [43,44], or place a conductor plate nearby the specimen
conductor for the same purpose [43]. But a charge-measuring conductor such as the
cage or the plate disturbs the original electric field and causes an error. To achieve more
accurate q - v curve measurement, a method that require no charge-measuring conductor
is developed in this thesis [45]. The proposed method measures the amount of space
charges by numerically integrating the injected current, and thus requires no charge-
measuring conductor. This enables the q - v curve measurement under the real electric-
field distribution of a transmission-line configuration. In this thesis, comprehensive q - v

1-5
curve measurement has been carried out using the proposed method, and it is shown that
q - v curves measured in the presence of a charge-measuring conductor are different from
those measured in the absence of the conductor. The effects of conductor radius and
line length are also investigated using the proposed method, and it is shown that the shape
of q - v curves depends heavily on the radius. The increase of electrostatic coupling due
to corona is necessary to be investigated, because it corresponds to a practical case that
the voltage of a phase wire is increased by a lightning surge propagating along a ground
wire. Thus, field data of coupling-factor curves are obtained, and the effects of
conductor separation and radius are discussed.

1.4 Corona Modeling

In the early days of transmission-line corona studies, the following two theories
had been dominant: 1) The equivalent capacitance of a conductor increases due to the
generation of space charges; and 2) Peek’s quadratic law, which is determined by
experiments under AC voltages, can be applied to surge corona. As for the former
theory, Wagner and Lloyd explained that corona discharge causes wave delay from their
measurement of q - v curves [43]. As for the latter, Umoto and Hara proposed
numerical techniques for solving the nonlinear equation, assuming that corona discharge
can be represented by a nonlinear capacitance and a conductance using Peek’s quadratic
law [46]. Based on those theories, comprehensive models have been proposed [47-52].
But the former theory requires the measurement of q - v curves for simulations, and the
latter raises a question whether Peek’s quadratic law is valid even for surge corona or not.
To reproduce q - v curves without measurement and Peek’s quadratic law, physical
models have recently been proposed such as a rigorous finite-difference model [53] and
corona-shell models [54-57]. The finite-difference method first discretizes the space
around a conductor in coaxial regions, and the physical equations of space-charge
generations and drift are established in each region and are solved simultaneously. On
the other hand, the concept of the corona shell takes notice of a fact that the space

1-6
charges form a shell around the conductor. According to ref. [44], the q - v curve of a
lightning surge is different from that of a switching surge, and therefore a corona model
should reproduce different q - v curves for different wave-front times (i.e. wave-steepness
dependent). So far, only the finite-difference method can reproduce the wave-steepness
dependence, but the method is time-consuming due to the discretization. Therefore, this
thesis develops a wave-steepness-dependent corona model, which can express the
dependence by a simple calculation procedure as accurately as the rigorous finite-
difference method [58]. The corona model is based on the concept of a quasi-static q - v
curve, which is calculated by Guillier’s simple corona-shell model [57], and (a) inception
delay, (b) streamer development, and (c) ionization process are introduced to reproduce
the dependence. The simplicity enhances the implementation of the corona model into
the proposed line model by discretization, because a large number of the corona model is
required to be inserted as corona branches. The calculated q - v and coupling-factor
curves agree well with field tests, and the wave-steepness-dependence is accurately
reproduced by the proposed model. Pure coupling-factor curves which are difficult to
be obtained by measurement are also reproduced by the model.

1.5 Inclusion of Corona Effects

This thesis also proposes an efficient method to deal with nonlinear corona
branches in the proposed line model using the trapezoidal rule of integration combined
with a predictor-corrector scheme [58,59]. Each of the corona branches is the proposed
wave-steepness-dependent corona model. The method solves the corona branches with
the same degree of accuracy as other electrical parts in an EMTP-type program by use of
the trapezoidal rule. And the predictor-corrector scheme avoids the Jacobian
calculation, which is difficult to be evaluated by a physical corona model, for the
nonlinear iteration. Transient calculations are carried out by the proposed method
including the skin and corona effects, and the results agree well with field tests. Also,
the effect of the wave-steepness dependence of corona on wave propagation is made

1-7
clear.

The present line model has been implemented in the ATP version of EMTP at
BPA, Portland, Oregon, USA, first in 1995, and continuously developed by the author.
The model is expected to be used for insulation designs all over the world.

Other contributions to power system transient analysis made by the author is


given as appendices in this thesis. Appendix B describes a predictor-corrector scheme
for solving a nonlinear circuit [60]. The scheme extends EMTP’s nodal-conductance
approach to include arbitrary number and configuration of nonlinear elements in a
network. Because the proposed method avoids the Jacobian calculation of iteration by
use of the predictor-corrector scheme, any nonlinearity can be included in a simulation by
its own suitable representation unlike the compensation method [61]. Also, the clear
logic of the trivial automatic formulation of circuit equations is available in the method.
Transient calculations of a diode-bridge rectifier circuit, a transistor switching circuit, and
a transformer model are demonstrated, and the calculated results show that the method is
stable for a large time step and also for simultaneous abrupt changes of the characteristics
of two or more nonlinear elements.

Appendix C presents basic investigation of lightning-induced voltages to an


electronic appliance [62]. A printed-board circuit in the electronic appliance is modeled
as a loop circuit, and the induced voltage is analytically calculated from the interlinked
magnetic flux generated by a lightning-stroke current. An experiment has been carried
out to confirm the accuracy of the proposed method, and a good agreement is obtained.

1-8
References of Chapter 1

[1] H.W. Dommel, “Digital Computer Solution of Electromagnetic Transients in Single- and Multi-
phase Networks,” IEEE Trans., Power Apparatus and Systems, Vol. PAS-88 (4), pp.388-398,
1969.
[2] J.R. Carson, “Wave Propagation in Overhead Wires with Ground Return,” Bell Syst. Tech., J.,
Vol. 5, pp. 539-554, 1926.
[3] F. Pollaczek, “Über das Feld einer unendlich langen wechselstromdurch-flossen Einfachleitung,”
E.N.T., Band 3 (Heft 9), pp. 339-360, 1926.
[4] R.H. Galloway, W.B. Shorrocks, and L.M. Wedepohl, “Calculation of Electrical Parameters for
Short and Long Polyphase Transmission Lines,” Proc. IEE, Vol. 111, No. 12, 1964.
[5] A. Deri, G. Tevan, A. Semlyen, and A. Castanheira, “The Complex Ground Return Plane, A
Simplified Model for Homogeneous and Multi-layer Earth Return,” IEEE Trans., Power
Apparatus and Systems, Vol. PAS-100, pp. 3686-3693, 1981.
[6] L.M. Wedepohl and S.E.T. Mohamed, “Multiconductor Transmission Lines : Theory of Natural
Modes and Fourier Integrals Applied to Transient Analysis,” Proc. IEE, Vol. 116, pp.1553-1563,
1969.
[7] W.S. Meyer and H.W. Dommel, “Numerical Modeling of Frequency-Dependent Transmission
Parameters in an Electromagnetic Transient Program,” IEEE Trans., Power Apparatus and
Systems, Vol. PAS-93, pp.1401-1409, 1974.
[8] A. Semlyen and A. Dabuleau, “Fast and Accurate Switching Transient Calculations on
Transmission Lines with Ground Return using Recursive Convolutions,” IEEE Trans., Power
Apparatus and Systems, Vol. PAS-94(2), pp.561-571, 1975.
[9] A. Ametani, “A Highly Efficient Method for Calculating Transmission Line Transients,” IEEE
Trans., Power Apparatus and Systems, Vol. PAS-95 (5), pp.1545-1549, 1976.
[10] W.D. Humpage, K.P. Wong, and T.T. Nguyen, “Z-transform Electromagnetic Transient Analysis
in Power Systems,” IEE Proc., Vol. 127, Pt. C, No.6, pp.370-378, 1980.
[11] A. Semlyen, “Contributions to the Theory of Calculation of Electromagnetic Transients on
Transmission Line with Frequency Dependent Parameters,” IEEE Trans., Power Apparatus and
Systems, Vol. PAS-100 (2), pp.848-856, 1981.
[12] J.F. Hauer, “State-Space Modeling of Transmission Line Dynamics via Nonlinear Optimization,”
IEEE Trans., Power Apparatus and Systems, Vol. PAS-100(12), pp.4918-4925, 1981.
[13] J.R. Marti, “Accurate Modelling of Frequency-Dependent Transmission Lines in Electromagnetic
Transient Simulations,” IEEE Trans., Power Apparatus and Systems, Vol. PAS-101 (1), pp.147-
155, 1982.
[14] A. Ametani, Refraction Coefficient Theory and Surge Phenomena in Power Systems, PhD Thesis,
The University of Manchester, 1973.
[15] A. Ametani, Distributed-Parameter Circuit Theory, Corona Pub. Co., Tokyo, 1992. (in Japanese:
雨谷,分布定数回路論,コロナ社,1992.)
[16] A. Ametani, “A Study of Cable Transient Calculations,” Sci. & Eng. Rev., Doshisha University,
Vol. 24, No. 2, pp. 110-127, 1983.
[17] A. Ametani, “Refraction Coefficient Method for Switching-Surge Calculations on Untransposed

1-9
Transmission Lines (Accurate and Approximate Inclusion of Frequency Dependency),” IEEE PES
Summer Meeting, C 73-444-7, 1973.
[18] L. Marti, “Simulation of Transients in Underground Cables with Frequency Dependent Modal
Transformation Matrices,” IEEE Trans., Power Delivery, vol. PWD-3 (3), pp.1099-1110, 1988.
[19] Tsu-huei Liu and Li Jin-gui, “Call for Help with Rational Function Approximations to
Frequency-Dependent Transformation Matrices of Cables and Lines,” EMTP News, Leuven
EMTP Center, March, 1988.
[20] H. Nakanishi, and A. Ametani, “Transient Calculation of a Transmission Line using
Superposition Law,” IEE Proc., Vol. 133, Pt. C, No. 5, pp.263-269, 1986.
[21] G. Angelidis and A. Semlyen, “Direct Phase-Domain Calculation of Transmission Line
Transients Using Two-Sided Recursions,” IEEE Trans., Power Delivery, Vol. 10, No. 2, pp. 941-
949, April 1995.
[22] B. Gustavsen, J. Sletbak, and T. Henriksen, “Calculation of Electromagnetic Transients in
Transmission Cables and Lines Taking Frequency Dependent Effects Accurately into Account,”
IEEE Trans., Power Delivery, Vol. 10, No. 2, pp. 1076-1084, April 1995.
[23] T. Noda, N. Nagaoka, and A. Ametani, “Phase Domain Modeling of Frequency-Dependent
Transmission Lines by Means of an ARMA Model,” IEEE Trans., Power Delivery, Vol. 11, No. 1,
pp. 401-411, January 1996.
[24] T. Noda, T. Masakawa, and N. Nagaoka, “A Transmission-Line Surge Calculation Method using
the Z Transform,” Proc. Annual Meeting of IEE of Japan, Paper No. 1151, 1992. (in Japanese: 野
田,政川, 長岡, 「Z 変換による線路サージ計算法」,平成 4 年電気学会全国大会,1151, 1992.)
[25] T. Noda, M. Kubota, and N. Nagaoka, “A Phase Domain Surge Calculation Method,” Proc.
Annual Meeting of IEE of Japan, Paper No. 1387, 1993. (in Japanese: 野田,窪田,長岡,「実
領域多相線路サージ計算法」,平成 5 年電気学会全国大会,1387, 1993.)
[26] T. Noda and N. Nagaoka, “A Phase Domain Surge Calculation Method (Part II) - A Surge
Calculation on a Cable -,” Proc. Kansai-Branch Meeting, IEE of Japan, Paper No. G4-31, 1993.
(in Japanese: 野田,長岡,「実領域多相線路サージ計算法 第 2 報−ケーブルサージ計算へ
の応用」,平成 5 年電気関係学会関西支部連合大会,G4-31, 1993.)
[27] T. Noda and N. Nagaoka, “A Time-Domain Surge Calculation Method with Frequency-
Dependent Modal-Transformation Matrices,” Conference of Electric Power System Technology,
IEE of Japan, PE-93-153, 1993. (in Japanese: 野田,長岡,「変換行列の周波数依存性を考慮
した時間領域線路サージ計算法」,電気学会電力技術研究会資料,PE-93-153, 1993.)
[28] T. Noda, N. Nagaoka, A. Ametani, “Further Improvements to a Phase-Domain ARMA Line
Model in Terms of Convolution, Steady-State Initialization, and Stability,” IEEE Power
Engineering Society Summer Meeting, Denver, Colorado, USA, 1996. (to be published in IEEE
Trans.)
[29] W.S. Meyer, “Taku Noda Frequency Dependence,” Can/Am EMTP News, Canadian/American
EMTP User Group, Vol. 95-1, January, 1995.
[30] P. Kuffel, K. Kent, and G. Irwin, “The Implementation and Effectiveness of Linear Interpolation
within Digital Simulation,” Proc. IPST '95, pp. 449-504, 1995.
[31] T. Noda and N. Nagaoka, “Development of ARMA Models for a Transient Calculation using
Linearized Least-Squares Method,” Trans. IEE of Japan, Vol. 114-B, No. 4, pp. 396-402, 1994.
(in Japanese: 野田,長岡,「線形化最小 2 乗法による ARMA 過渡現象計算モデルの開発」,
電気学会論文誌 B, Vol. 114-B, No. 4, pp. 396-402, 1994.)
[32] T. Noda and N. Nagaoka, “An Estimation Method of Linear Network's Transfer Function,”
Conference of Electric Power System Technology, IEE of Japan, PE-92-150, 1992. (in Japanese:
野田, 長岡, 「線形回路網伝達関数推定の一手法」,電気学会電力技術研究会資料,PE-92-150,

1-10
1992.)
[33] T. Noda and N. Nagaoka, “An Estimation Method of Linear Network's Transfer Function,” Proc.
Kansai-Branch Meeting, IEE of Japan, Paper No. G4-8, 1992. (in Japanese: 野田,長岡,「線形
回路網伝達関数推定の一手法」,平成 4 年電気関係学会関西支部連合大会,G4-8, 1992.)
[34] T. Noda and N. Nagaoka, “A Linearized Least-Squares Method for Estimating Linear Network's
Transfer Function - Relative Error Evaluation -,” Proc. Annual Meeting of IEE of Japan, Paper
No. 1397, 1993. (in Japanese: 野田,長岡,「線形回路網伝達関数推定に適した線形化最小 2
乗法―相対誤差評価法―」,平成 5 年電気学会全国大会,1397, 1993.)
[35] T. Noda, T. Sawada, K. Fujii, and N. Nagaoka, “ARMA Models for Transient Calculations and its
Identification - Identification of Coefficients and Orders -,” Proc. Annual Meeting of IEE of
Japan, Paper No. 1373, 1994. (in Japanese: 野田,澤田,藤井,長岡,「ARMA 過渡現象計算
モデルとその作成法」,平成 6 年電気学会全国大会,1373, 1994.)
[36] J.S. Lim and A.V. Oppenheim, Advanced Topics in Signal Processing, Prentice-Hall, 1988.
[37] E.I. Jury, Theory and Application of the z-Transform Method, Wieley, New York, 1964.
[38] L. Weinberg, Network Analysis and Synthesis, McGraw-Hill, 1962.
[39] T. Noda, N. Nagaoka, and A. Ametani, “Fault-Surge Calculations using the Phase-Domain
ARMA Line Model,” Trans. IEE of Japan, Vol. 116-B, No. 11, pp. 1409-1414, 1996. (in
Japanese: 野田,長岡,雨谷,「相領域 ARMA 線路モデルを用いた地絡サージ計算」,電
気学会論文誌 B, Vol. 116-B, No. 11, pp. 1409-1414, 1996.)
[40] A. Ametani, T. Ono, and A. Honga, “Surge Propagation on Japanese 500kV Untransposed
Transmission Line,” Proc. IEE, Vol. 121, No.2, 1974.
[41] A. Ametani, E. Osaki, and Y. Honaga, “Surge Characteristics on an Untransposed Vertical Line,”
Trans. IEE Japan, Vol. 103-B, pp.117-124, 1983. (in Japanese: 雨谷,大崎,補永,「非撚架垂
直配列線路におけるサージ特性」,電気学会論文誌 B, Vol. 103-B, pp.117-124, 1983.)
[42] N. Nagaoka, M. Yamamoto, and A. Ametani, “Surge Propagation Characteristics of a POF
Cable,” Trans. IEE Japan, Vol. 105-B, pp.645-652, 1985. (in Japanese: 長岡,山本,雨谷,「POF
ケーブルにおけるサージ波形伝搬特性」,電気学会論文誌 B, Vol. 105-B, pp.645-652, 1985.)
[43] C.F. Wagner, and B.L. Lloyd, “Effect of Corona on Traveling Waves”, AIEE Trans. on Power
Apparatus and Systems, Vol.74, part Ⅲ, 1955.
[44] P.S. Maruvada, H. Menemenlis, R. Malewski, “Corona Characteristics of Conductor Bundles
under Impulse Voltages”, IEEE Trans. on Power Apparatus and Systems, Vol. PAS-96, No. 1,
1977.
[45] T. Noda, H. Fukuzono, S. Sekioka, N. Nagaoka, and A. Ametani, “Charge - Voltage Measurement
of Transmission-Line Corona without a Corona Cage,” Proc. of the Seventh Annual Conference of
Power & Energy Society, IEE of Japan, Session II, Paper No. 427, p. 451, 1996. (in Japanese: 野
田,福園,関岡,長岡,雨谷,「コロナケージを用いない送電線コロナ電荷−電圧曲線の
測定」,平成 8 年電気学会電力・エネルギー部門大会,論文 II, 427, p. 451, 1996.)
[46] J. Umoto, and T. Hara, “Numerical Analysis of Line Equations Considering Corona Loss on
Single-Conductor System,” Vol. 89, No. 968, 1969. (in Japanese: 卯本,原,「コロナ損を考慮
した単導線系伝搬方程式の数値解析」,電学誌,Vol. 89, No. 968, 1969.)
[47] M.M. Suliciu and I. Suliciu, “A Rate Type Constitutive Equation for the Description of the Corona
Effect”, IEEE Trans. on Power Apparatus and Systems, Vol. PAS-100, 1981.
[48] H.M. Kudyan, C.H. Shiu, “A Nonlinear Circuit Model for Transmission Line in Corona”, IEEE
Trans. on Power Apparatus and Systems, Vol. PAS-100, 1981.
[49] K.C. Lee, “Nonlinear Corona Models in an Electromagnetic Transients Program (EMTP)”, IEEE
Trans. on Power Apparatus and Systems, Vol. PAS-102, No. 9, 1983.

1-11
[50] R.J. Harrington, “Implementation of Computer Model to Include the Effects of Corona in
Transient Overvoltage Calculations”, IEEE Trans. on Power Apparatus and Systems, Vol. PAS-
102, No. 4, 1983.
[51] A. Inoue, “Propagation Analysis of Overvoltage Surges with Corona Based upon Charge Versus
Voltage Curve”, IEEE Trans. on Power Apparatus and Systems, Vol. PAS-104, No. 3, 1985.
[52] P.S. Maruvada, D.H. Nguyen, H. Hamadani-Zadeh, “Studies on Modeling Corona Attenuation of
Dynamic Overvoltages”, IEEE Trans. on Power Delivery, Vol.4, No. 2, 1989.
[53] M.T. Correia de Barros, “Wide Bandwidth Modeling of Corona on High Voltage Transmission
Lines”, IEEE Dielectries and Electrical Insulation Society Conference, Electrical Insulation and
Partial Discharges, Paper No. 78, 1994. Or the same theory can be found from :
Célia de Jesus and M.T. Correia de Barros, “Modelling of Corona Dynamics for Surge
Propagation Studies,” IEEE Trans. on Power Delivery, Vol. 9, No. 3, 1994.
[54] J.J. Claudé, C.H. Gary, and C.A.Lefévre, “Calculation of Corona Losses Beyond the Critical
Gradient in Alternating Voltage,” IEEE Trans. on Power Apparatus and Systems, Vol. PAS-88,
No. 5, 1969.
[55] A. Semlyen and H. Wei-Gang, “Corona Modelling for the Calculation of Transmission Lines,”
IEEE Trans. on Power Delivery, Vol. PWRD-1, No. 3, 1986.
[56] M.A. Al-Tai, H.S.B. Elayyan, D.M. German, A. Haddad, N. Harid, and R.T. Waters, “The
Simulation of Surge Corona on Transmission Lines,” IEEE Trans. on Power Delivery,Vol.4, No.
2, 1989.
[57] J.F. Guillier, M. Poloujadoff, and M. Rioul, “Damping Model of Traveling Waves by Corona
Effect along Extra High Voltage Three Phase Lines,” IEEE Trans. on Power Deliverry, Vol. 10,
No. 4, 1995.
[58] T. Noda, N. Nose, N. Nagaoka, and A. Ametani, “A Wave-Front-Time Dependent Corona Model
for Transmission-Line Surge Calculations,” Proc. of the Seventh Annual Conference of Power &
Energy Society, IEE of Japan, Session I-J, Paper No. 36, pp. 209-214, 1996. (to be published in
Trans. IEE of Japan, in Japanese: 野田,能勢,長岡,雨谷,「線路サージ計算に用いる波頭
長依存コロナモデル」,平成 8 年電気学会電力・エネルギー部門大会,論文 I-J, 36, pp.
209-214, 1996.)
[59] T. Noda, N. Nose, N. Nagaoka, and A. Ametani, “An EMTP Compatible Corona Model (Part 1),”
Proc. Kansai-Branch Meeting, IEE of Japan, Paper No. G5-31, 1995. (in Japanese: 野田,能勢,
長岡,雨谷,「EMTP 線路モデルに導入容易なコロナモデル 第 1 報」,平成 7 年電気関係
学会関西支部連合大会, G5-31, 1995.)
[60] T. Noda, K. Yamamoto, N. Nagaoka, and A. Ametani, “A Predictor-Corrector Scheme for
Solving a Nonlinear Circuit,” Proc. of International Conference on Power Systems Transients ‘97
(IPST), Seattle, Washington USA, 1997. (provisionally accepted)
[61] H.W. Dommel, “Nonlinear and time-varying elements in digital simulation of electromagnetic
transients,” IEEE Trans., Power Apparatus and Systems, Vol. PAS-90, pp. 2561-2567, 1971.
[62] T. Noda, J. Takami, N. Nagaoka, and A. Ametani, "Basic Investigation of a Lightning-Induced
Voltage to an Electronic Appliance," Proc. 23rd International Conference on Lightning
Protection (ICLP), Volume II, pp. 690-695, Firenze, Italy, 1996.

1-12
2 FREQUENCY-DEPENDENT AND NONLINEAR WAVE
EQUATIONS DESCRIBING TRANSMISSION-LINE DYNAMICS

Summary

This chapter gives wave equations describing the dynamics of a transmission line
including its frequency dependence and nonlinearity. The dynamics of a frequency-
independent linear transmission line is fully described by a pair of first-order partial-
differential equations which is called telegrapher’s equations, and d’Alembert’s solution is
well-known as its time-domain solution explaining traveling-wave behavior. D’Alembert’s
solution is very useful to obtain an approximate solution by hand. But, in practice, the
resistance and inductance (impedance) of a line vary with respect to frequency, and the
capacitance (admittance) varies with respect to voltage. The frequency dependence of the
impedance is due to the skin effects of conductors and earth soil, and the nonlinearity of the
admittance is caused by corona discharge along the conductors. To obtain an accurate
solution, the frequency dependence and the nonlinearity should be included in equations
describing the dynamics of a transmission line. In this chapter, a general description of
overhead lines and underground cables is briefly given, and then, matrix partial-differential
equations (wave equations) taking into account the frequency dependence and the nonlinearity
are derived.

2-1
2.1 Transmission Lines :
Overhead Lines and Underground Cables [1-4]

A transmission line mainly plays a role of the transmission of electric power


energy from a hydroelectric power generator in a mountainous area or from a thermal/
nuclear power plant on a coast to a substation near consumers. Also, it is used in order
to interlink between two or more power systems. Transmission lines are roughly
classified into two categories : overhead lines and underground cables. The overhead
lines are advantageous compared with the underground cables in terms of construction
cost, transmission capacity, and maintainability. On the other hand, the underground
cables seldom suffer from lightning, storm, and snow accidents, and thus the reliability is
relatively high. Also, the impact on the view of a city is quite small. The recent
increase of electric power demand in urban areas and the difficulty to obtain sites for
transmission towers have increased underground cable installation. The demand of
electric power and the capacity of generators is rapidly growing, and a distance between a
generator and consumers becomes much longer. Thus, to reduce the ohmic loss in
conductors, transmission voltages have been raised considerably. In Japan, 1,000-kV
class UHV (Ultra High Voltage) transmission systems which use overhead lines have
already been constructed and ready for commercial operation. Also, 500-kV class
underground/submarine cables are operating in many places in Japan.

Fig. 2.1 illustrates a typical structure of an overhead line system used in Japan,
which consists of phase and ground wires, insulator strings, and support. A double-
circuit arrangement is usually used, and phase conductors are suspended by the insulator
strings from three pairs of arms. At the top of the tower, one or two ground wires are
installed for lightning protection. The foundation of the support and a grounding
electrode to discharge lightning currents induced on the ground wire(s) are laid below the
earth surface. An Aluminum Conductor Steel Reinforced (ACSR) is usually used as a

2-2
ground wire

insulator string

phase wire

arm

support

earth

Fig. 2.1 Typical structure of overhead line system

aluminum
wires

galvanized
steel wires

Fig. 2.2 Aluminum Conductor Steel Reinforced (ACSR)

phase wire, and its typical cross section is illustrated in Fig. 2.2. The ACSR consists of
galvanized wires as its core and enclosing aluminum wires stranded. The galvanized
wires reinforce mechanical strength to endure the conductor weight, storm, and snow.
Electric currents mainly flow through the aluminum wires. A Hard Drawn Copper
Cable (HDCC) had been used under 154-kV systems, but has been replaced with the
ACSR because of economical reasons. In EHV (Extra High Voltage) and UHV systems,
a conductor bundle is used as each phase wire, in order to reduce corona interference and
to increase transmission capacity. Fig. 2.3 illustrates examples of the conductor bundle.

2-3
In Japan, a 275-kV system uses a bundle of two conductors, a 500-kV system four or six
conductors, and a 1,000-kV system eight conductors.

(a) bundle of two conductors (b) bundle of four conductors (c) bundle of six conductors

Fig. 2.3 Bundle of conductors

Because the insulation layer of an underground cable system is compact, the


capacitance is much larger than that of an overhead line, and the inductance is much
smaller. Thus, the reactive power generated by a cable affects the voltage regulation of
a power system, and the charging currents affect the transmission capacity. As for
transient phenomena, the surge impedance and propagation velocity of a cable is much
smaller than those of an overhead line. As illustrated in Fig. 2.4, underground cables are
classified into five types in view of structure :

(a) Single-Core Cable consists of a core and an enclosing sheath. A three-phase circuit consists of
three of the single-core cables. (e.g. single-core EV, CV, and OF cables)

(b) Three-Core Cable consists of three cores with individual sheaths and also with an enclosing sheath.
(e.g. three-core EV, CV, and OF cables)

(c) Triplex-Type Cable consists of three single-core cables stranded each other. (e.g. stranded single-
core CV cable )

(d) Pipe-Type Cable consists of three cores with individual sheaths and an enclosing pipe. The
inside of the enclosing pipe is filled up with insulating oil or gas. (e.g. POF, PGF,
and PGC cables)

(e) Compressed Gas consists of three cores insulated by common insulating gas and an enclosing
Insulated Cable pipe.

2-4
(a) (b) (c)

(d) (e)

Fig. 2.4 Structural classification of cables

In view of main insulator medium, cables are classified into the following types :

(a) OF Cable (Oil-Filled Cable) uses insulting paper filled with pressured oil as main
insulator.

(b) EV / CV Cable uses polyethylene / crosslinked polyethylene as main insulator.

(c) POF / PGF / PGC uses insulating oil / gas / pressured gas as main insulator.

Recently, the research and development of CV cables are quite active, and they
are applied even to 500-kV transmission systems. Especially, cables are suitable for DC
transmission, because there are no problems of charging currents, dielectric losses, and
sheath losses. Thus, CV cables are often used in an HVDC (High Voltage DC)
transmission system.

2-5
2.2 Frequency-Dependent and Nonlinear Wave Equations

The per unit length resistance R and inductance L (impedance Z = R + jωL) of a


transmission line vary with respect to frequency due to the skin effects of conductors and
earth soil, and can be expressed as follows in a single-phase case :

Z ( jω ) = R( jω ) + jωL( jω ). (2.1)

Also, the per unit length capacitance C (admittance Y = jωC) varies with respect to
voltage due to corona discharge along the conductors. In a single-phase overhead-line
case :

C = C (v ). (2.2)

Because corona discharge is not observed in the insulating layer(s) of a cable in normal
cases, the capacitance is not nonlinear, i.e. the cable can be treated as a “linear”
transmission line. On the other hand, the dielectric loss of a cable insulator may be
represented as a frequency-dependent conductance G(jω) connected in parallel to the
capacitance (Y = G + jωC) [5]. But the contribution of the conductance is quite small,
and there is no appropriate theoretical formula to calculate the frequency dependence.
Thus, the conductance is usually ignored. From eqs. (2.1) and (2.2), an equivalent
circuit of a single-phase line is derived as shown in Fig. 2.5, assuming plane-wave
propagation which is a quite accurate approximation in a frequency range of our interest.
When ∆x is approaching zero, the following set of partial differential equations (wave
equations) are obtained :

2-6
x
∆x
R(ω)∆x L(ω)∆x

i
C(v)∆x
v ....

Fig. 2.5 Circuit equivalent of single-phase line considering


frequency dependence and nonlinearity

 ∂v ∂i
− = Ri + L ,
 ∂x ∂t
 (2.3)
− ∂i ∂q
= ,

 ∂x ∂t

where x : distance from the sending end, t : time, v = v(x,t) : voltage, i = i(x,t) :
current, and q = q(x,t) : per unit length charge on the conductor, all at distance x
at time t.

Fig. 2.6 shows a typical charge-voltage (q - v) curve representing the nonlinear


capacitance C(v). As explained in Sec. 5.2, the nonlinear capacitance is defined as the
slope of the q - v curve as :

∂q
C( v ) = , (2.4)
∂v

2-7
charge:
C(v)= ∂q /∂v
q

(v, q)

C0

voltage: v

Fig. 2.6 Typical q - v curve

and thus the time derivative of q is calculated as :

∂q ∂q ∂v ∂v
= = C( v ) . (2.5)
∂t ∂v ∂t ∂t

The nonlinear capacitance can be decomposed into the geometrical capacitance C0


(natural geometrical capacitance in the absence of corona) and the contribution of corona
discharge (increase of space charges due to corona) ∆C(v) as :

C( v ) = C0 + ∆C( v ), (2.6)

and thus eq. (2.5) becomes

∂q ∂v ∂v
= C0 + ∆C ( v ) . (2.7)
∂t ∂t ∂t

2-8
From eqs. (2.3) and (2.7), the wave equations are modified into the following form :

 ∂v ∂i
− = Ri + L ,
 ∂x ∂t
 (2.8)
− ∂i ∂v ∂v
= C0 + ∆C( v ) .

 ∂x ∂t ∂t

We now decompose ∂i/∂x into the contributions of the geometrical capacitance ∂i0/∂x

and the nonlinear capacitance ∂ic/∂x as :

∂i ∂i0 ∂ic
= + (2.9)
∂x ∂x ∂x

As a result, the final form of the wave equations is given as :

 ∂v ∂i
− ∂x
= Ri + L ,
∂t
(a)
 ∂i0 ∂v
− = C0 , (b)
 ∂x ∂t
 ∂ic ∂v
(2.10)
− = ∆C (v ) , (c)
 ∂x ∂t
− ∂i ∂i0 ∂ic
 =− − (d)
 ∂x ∂x ∂x

Therefore, the circuit equivalent shown in Fig. 2.5 is modified as shown in Fig. 2.7 based
on the above wave equations. In the figure, the frequency-dependent but linear part is
represented by R(ω), L(ω), and C0, and the nonlinear part by ∆C(v). It should be noted
that the linear and nonlinear parts are separated from each other. The linear part is
modeled based on a phase-domain formulation to accurately take into account the
frequency dependence of modal-transformation matrices in Sec. 4 [6], and the nonlinear
part is modeled based on a state-space equation in Sec. 6 [7,8].

2-9
x
∆x nonlinear part
linear part
(freq.-dep.)
R(ω)∆x L(ω)∆x

i
....

v C0∆x ∆C(v)∆x

Fig. 2.7 Circuit equivalent of single-phase line considering


frequency dependence and nonlinearity (final form)

Eqs. (a) and (b) of (2.10) are modeled in the frequency domain because of the frequency
dependence, and (c) is modeled in the time domain because of the nonlinearity. Thus,

 ∂V
− = {R( jω ) + jω L( jω )}I = Z ( jω ) I , (a)
∂x
 ∂I 0
− = jω C0V = Y0V , (b)
 ∂x
 ∂ic ∂v (2.11)
− = ∆C( v ) , (c)
 ∂x ∂t
− ∂i -1  ∂I 0  ∂i
= − F  − c , (d)

 ∂x  ∂x  ∂x

where frequency-domain counterparts are denoted using uppercase letters, and F− 1


denotes inverse Fourier Transform. In eq. (d), the interface between the time and
frequency domains is formally expressed by the inverse Fourier Transform, but practically
realized by an efficient IARMA method in Sec. 4. In a multiphase case, the wave
equations are expressed in the following matrix form :

2-10
 ∂V
− = {R ( jω ) + jω L( jω )}I = Z ( jω ) I , (a)
∂x
 ∂I 0
− = jω C 0V = Y0V , (b)
 ∂x
 ∂ic ∂v (2.12)
− = ∆C ( v ) , (c)
 ∂x ∂t
− ∂i -1  ∂I 0  ∂ic
=−F  − , (d)

 ∂x  ∂x  ∂x

and used in the subsequent modeling described in the following chapters.

2-11
References of Chapter 2

[1] Electrical Engineering Handbook, New Edition, The Institute of Electrical Engineers of Japan,
1988. (in Japanese: 電気工学ハンドブック,電気学会,1988.)
[2] A. Ametani, K. Kaneko, H. Tsujimura, Electric Energy System Engineering, Nikkan Kougyou
Shinbun-sha, Tokyo, 1988. (in Japanese: 雨谷,金子,辻村,電気エネルギーシステム工学,
日刊工業新聞社,1988.)
[3] T. Kouno, J. Toyoda, T. Kawase, K. Matsuura, Power Transmission and Distribution Engineering,
Ohm-sha, Tokyo, 1969. (in Japanese: 河野,豊田,川瀬,松浦,送配電工学,オーム社,1969.)
[4] T. Kouno, Insulation of Power System, Corona Pub. Co., Tokyo, 1984. (in Japanese: 河野,系統
絶縁論,コロナ社,1984.)
[5] G. Corti, G. Tizzi, “Calculation of Transmission Parameters and Design of Coaxial Structures by
Computer,” ALTA FREQUENZA, Vol. XLVI - N. 9, pp. 389-399, 1977.
[6] T. Noda, N. Nagaoka, and A. Ametani, “Phase Domain Modeling of Frequency-Dependent
Transmission Lines by Means of an ARMA Model,” IEEE Trans., Power Delivery, Vol. 11, No. 1,
pp. 401-411, January 1996.
[7] T. Noda, N. Nose, N. Nagaoka, and A. Ametani, “A Wave-Front-Time Dependent Corona Model
for Transmission-Line Surge Calculations,” Proc. of the Seventh Annual Conference of Power &
Energy Society, IEE of Japan, Session I-J, Paper No. 36, pp. 209-214, 1996. (to be published in
Trans. IEE of Japan, in Japanese: 野田,能勢,長岡,雨谷,「線路サージ計算に用いる波頭
長依存コロナモデル」,平成 8 年電気学会電力・エネルギー部門大会,論文 I-J, 36, pp.
209-214, 1996.)
[8] T. Noda, N. Nose, N. Nagaoka, and A. Ametani, “An EMTP Compatible Corona Model (Part 1),”
Proc. Kansai-Branch Meeting, IEE of Japan, Paper No. G5-31, 1995. (in Japanese: 野田,能勢,
長岡,雨谷,「EMTP 線路モデルに導入容易なコロナモデル 第 1 報」,平成 7 年電気関係
学会関西支部連合大会, G5-31, 1995.)

2-12
3 FREQUENCY-DEPENDENT SERIES-IMPEDANCE MATRIX
CONSIDERING SKIN EFFECTS

Summary

The purpose of this chapter is to calculate the series-impedance matrix of a


transmission line taking into account the skin effects of conductors and earth soil. Due to the
skin effects, current distribution in the conductors and in the earth soil varies with respect to
frequency, and thus the equivalent resistance and inductance (impedance) are frequency-
dependent. In this chapter, a general description of the skin effects is first given, and then a
closed-form formula of the overhead-line earth-return impedance is derived using a double
exponential fitting. The physical representation of the double-exponential approximation is
given by use of the concept of the complex ground-return plane proposed by Deri et al. For the
time-domain realization of the frequency dependence, the frequency response of the series
impedance must be smooth with respect to frequency. This is a very important consideration,
when a line is modeled using a modern numerical fitting method. Compared with conventional
Carson-Pollaczek’s expansion formula, the proposed formula does not generate discontinuities
with respect to frequency. Also, the proposed formula is far more accurate than Dubanton-
Deri’s approximate formula.

3-1
3.1 Skin Effects and Frequency-Dependent Series-Impedance Matrix

Consider an electromagnetic plane wave propagating in a vacuum space along


the x axis to an imperfectly conducting plate as shown in Fig. 3.1. The incident wave is
assumed to be polarized in the z-axis direction, and the magnetic-field intensity is given as
Hz = H0 exp( jωt) in the vacuum. According to the resistivity of the plate, some of the
incident wave reflects but the rest penetrates into the plate. In the conductor region, the
magnetic-flux desity is calculated by Maxwell’s equations as [1] :

 ωµσ   ωµσ 
Bz = µH 0 exp− x  cosω t − x , (3.1)
 2   2 

where µ, σ : permeability and conductivity of the conductor plate, and ω :


angular frequency.

z
incident

reflected

penetrating

y x

imperfectly conducting
vacuum plate

Fig. 3.1 Electromagnetic plane wave inciding to imperfectly conducting plate

3-2
The current distribution in the conductor region is given as :

 ωµσ   ωµσ π
J y = ωησH 0 exp− x  cosω t − x + . (3.2)
 2   2 4

The following value p gives a depth where the amplitude of the penetrating wave is
attenuated by a factor of e− 1, and is called the penetration depth :

2
p= . (3.3)
ωµσ

From eqs. (3.2) and (3.3), it is observed that currents gather on the conductor surface at a
high frequency, and the equivalent resistance of the conductor becomes larger as
frequency increases. From eqs. (3.1) and (3.3), it is also observed that magnetic flux
gathers on the surface at a high frequency, and the equivalent inductance becomes smaller.
This phenomenon is called a skin effect, and the equivalent resistance and inductance
(impedance) are frequency-dependent due to the resistivity of a material.

Consider a single-phase transmission line illustrated in Fig. 3.2. Because the

transmission wire penetration

penetration

earth soil

Fig. 3.2 Single-phase transmission line

3-3
transmission wire and the earth soil below the line are imperfect conductors, an analogy
of the above skin-effect theory can be applied to those imperfect conductors. When a
traveling wave is propagating along the line, electromagnetic waves penetrate into the
imperfect conductors, and the penetration depth varies with respect to frequency.
Therefore, the series impedance per unit length of a transmission line is frequency-
dependent due to the skin effects of the wire and the earth soil, and expressed as :

Z ( jω ) = R( jω ) + jωL( jω ). (3.4)

where R, L : per unit length resistive and inductive components.

The above impedance consists of the contributions of the physical geometry, the earth-
return path, and the conductor as :

Z = Zs + Ze + Zc = jω Ls + ( Re + jω Le ) + ( Rc + ω Lc ) (3.5)

where Zs = jωLs : contribution of physical geometry (space impedance), Ze = Re


+ jωLe : contribution of earth-return path (earth-return impedance), and Zc = Rc
+ jωLc : contribution of wire (conductor-internal impedance).

Because the space impedance Zs is due to the magnetic flux outside the conductors, it
consists only of an inductive component and is frequency-independent. On the other
hand, because Ze and Zc are due to the magnetic flux and currents inside the imperfect
conductors (earth soil and wire), they consist both of resistive and inductive components
and are frequency-dependent by the skin effects. As a numerical example, Fig. 3.3
shows the frequency dependence of the earth-return impedance Ze of a single-phase
overhead line : conductor height = 5 [m], conductor radius = 1 [cm], earth resistivity =
100 [Ω m], and the conductor permeability is assumed to be equal to that of vacuum.
Also, Fig. 3.4 shows the frequency dependence of the conductor-internal impedance Zc of

the same line with conductor resistivity = 5 × 10− 8 [Ω m]. Those curves are calculated
using J.R. Carson’s earth-return impedance formula [2] and S.A. Schelkunoff’s

3-4
2 1.5

resistance Re
inductance Le

inductance [µH/m]
resistance [Ω /m]

0.5

0 0 0
10 101 102 103 104 105 106 107
frequency [Hz]

Fig. 3.3 Frequency dependence of earth-return impedance


Ze(jω) = Re(jω) + jωLe(jω)

0.03 0.06

resistance Rc inductance [µH/m]


resistance [Ω /m]

inductance Lc
0.02 0.04

0.01 0.02

0 0
100 10 1
10 2
10 3
10 4
10 5
10 6
107

frequency, Hz

Fig. 3.4 Frequency dependence of conductor-internal impedance


Zc(jω) = Rc(jω) + jωLc(jω)

3-5
conductor-internal impedance formula [3] respectively. In both the figures, it is
observed that the resistive components increase and the inductive components decrease
as frequency increases due to the skin effects. But the curves are different because of
the different electromagnetic-wave distribution. In a multiphase case, eq. (3.5) can be
rewritten in the following matrix form :

Z = Z s + Z e + Z c = jω Ls + ( Re + jω Le ) + ( Rc + ω Lc ) (3.6)

and the above frequency-dependent series-impedance matrix Z is what we need to


evaluate accurately, in this chapter.

In the proposed line model, the frequency dependence of the series impedance
results in that of the wave propagation function and of the characteristic admittance,
because the line model uses a traveling-wave approach [4]. On the other hand, the
frequency dependence can directly be synthesized as a lumped-circuit equivalent as
shown in Fig. 3.5 [5-17]. Normally, Cauer’s circuit is used, because it satisfies the
physical analogy of a skin effect, but some use other circuits, such as Foster’s circuit, for
the numerical identification of the lumped parameters of the circuit. The present author
also takes part in one of the research projects as presented in refs. [12-17].

Z ....

Fig. 3.5 Lumped-circuit equivalent of skin effect (Cauer’s circuit)

3-6
3.2 Earth-Return Impedance

3.2.1 Historical review of earth-return impedance formulas

Fig. 3.6 (a) shows an infinitely long cylindrical conductor parallel to an


imperfectly conducting earth. The conductor is placed at the height of h above the earth,
and the radius is r. Normally, an earth-return impedance formula is derived under this
configuration. Deriving the earth-return impedance formula is equivalent to solving the
electromagnetic waves propagating along the conductor. In 1925, R. Rüdenberg solved
the problem by assuming that the shape of the earth surface can equivalently be modified
as shown in Fig. 3.6 (b) [18]. His assumption makes significantly easy to solve the
problem, because the distribution of the magnetic flux becomes cylindrical. In 1926, J.R.
Carson [2] and F. Pollaczek [19] first consistently solved the problem under the
configuration of Fig. 3.6 (a). The work has been carried out almost simultaneously and
independently by both the authors. The derived self-impedance formula is given as an
infinite-integral form (in MKSA units):

µ 0 2 hi µ
Zeii = jω ln + jω J eii , (3.7a)
2π ri π

∞ e − 2 hiλ
J eii = ∫0 d λ, (3.7b)
λ+ λ2 + jωµσ

and the mutual-impedance formula is

µ 0 Dij′ µ
Z eij = jω ln + jω J eij , (3.8a)
2 π Dij π
− ( hi + h j )λ
∞ e
J eij = ∫0 cos d ij λd λ, (3.8b)
λ+ λ2 + jωµσ

3-7
r

(a) conductor above imperfectly conducting earth

r
h

(b) Rüdenberg’s assumption

Fig. 3.6 Conductor configuration and Rüdenberg’s assumption

3-8
where σ : earth conductivity, i, j: subscripts to denote i-th and j-th conductors,
dij : horizontal separation between i-th and j-th conductors, Dij =

( hi − h j )2 + d ij 2 , and Dij′= ( hi + h j ) 2 + d ij 2 .

Nowadays, this solution is considered as the standard solution in the field of power
systems transient analysis. It is interesting to note that Rüdenberg's formula gives larger

value than Carson-Pollaczek’s by a factor of 2 at a high frequency where the skin


effect is considerable, because Rüdenberg's assumption shown in Fig. 3.6 (b) cannot
represent the true earth-return current distribution correctly. Later in 1934, W.H. Wise
derived the same result as Carson-Pollaczek's using the Herzian potential of a horizontal
current-element dipole formulated as an infinite integral by H. von Hoerschelmann [20].
In order to evaluate the infinite integral in eq. (3.8b) (eq. (3.7b) becomes identical to
(3.8b) by setting dij = 0), Carson derived a series expansion for small argument

r = Dij′ ωµσ ≤5 , and an asymptotic expansion for r > 5 [2]. Then, R.H. Galloway et

al. rearranged the expansions to suite computation as follows [21]:

µω
Zeij = ( Pij + jQij ), (3.9)
π

for r ≤5 :

π 1 2 1 σ σ σ
Pij = (1 − S 4 ) + log  S 2 + θij S 2′− 1 + 2 + 3 (3.10a)
8 2 γ
r 2 2 2 2

1 1 2 1 σ πS 2 σ3 σ4
Qij = + log (1 − S 4 ) − θij S 4′+ 1 − + − (3.10b)
4 2 γ r 2 2 8 2 2

where

∞ ∞
S 2 = ∑ a n cos(4n + 2 )θij , S 2′= ∑ a n sin( 4 n + 2)θij ,
0 0

3-9
∞ ∞
S 4 = ∑ c n cos( 4 n + 4)θij , S 4 ' = ∑ c n sin( 4n + 4 )θij ,
0 0

∞ ∞
σ1 = ∑ en cos(4n + 1)θij , σ2 = ∑ gn ( S2 )n ,
0 0

∞ ∞ ∞
σ3 = ∑ f n cos( 4n + 3)θij , σ4 = ∑ hn ( S4 )n , σ4 = ∑ hn ( S4 )n ,
0 0 0

4
− a n− 1 r  r2
an =   , a0 = ,
2n( 2 n + 1) ( 2n + 2 )  2 
2 8
4
− c n− 1 r  r4
cn =   , c0 = ,
( 2n + 1)( 2n + 2 ) ( 2 n + 3) 2 
2 192

− en− 1 r
en = r4 , e0 = ,
(4 − 1)( 4 + 1) ( 4 + 3)
n n 2
n 3

− f n− 1 r3
fn = r4 , f0 = ,
( 4n + 1)( 4n + 3)2 ( 4 n + 5) 45

1 1 1 1 5
gn = gn − 1 + + + − , g0 = ,
4 n 2 n + 1 2n + 2 4 n + 4 4
1 1 1 1 5
hn = hn − 1 + + + − h0 =
4n + 2 2n + 2 2n + 3 4 n + 6 , 3,

for r > 5 :

cos θij cos 2θij cos 3θij 3 cos 5θij


Pij = − 2
+ 3
+ (3.11a)
2r r 2r 2r 5
cos θij cos 3θij 3 cos 5θij
Qij = − +
2r 2r 3 2r 5 (3.11b)

Because the above expansions include a switch of formulas at r = 5, a numerical


discontinuity is sometimes observed (examples in the subsequent section show the
discontinuity). In stead of calculating the above tedious expansions, a simple closed-
form formula was intuitively derived by Dubanton [22] (published by Gary), and later a
theoretical derivation was given by A. Deri et al. using the concept of the complex

3-10
ground-return plane [23]. Because Dubanton-Deri’s formula is closely related to the
proposed closed-form formula in the subsequent section, the detailed description is given
in the section. R. Schinzinger and A. Ametani proposed a smart method to avoid the
infinite integral in eqs. (3.7b) and (3.8b) by use of a conformal transformation approach
[24]. In the method, the bilinear Moebius function :

p − js
q=R (3.12)
p + js

transforms the original plane p = x + jy into the concentric cylinder plane q = u + jv,
where the impedance can be derived easily, as shown in Fig. 3.7.

r
bilinear Moebius
transformation

r1
h

r2

(a) original plane p = x + jy (b) concentric cylinder plane q = u + jv

Fig. 3.7 Conformal transformation approach

All the above approaches assume that electromagnetic waves propagate in the
TEM (Transverse Electric and Magnetic) mode, that is, quasi plane-wave propagation.
Thus, displacement currents are neglected. In order to take into account the effects of
the displacement currents, there are two main approaches. One is to introduce the

3-11
frequency-dependence of the shunt admittance, what is called an admittance correction.
And the other is to solve the complete field solution considering TE (Transverse Electric)
and TM (Transverse Magnetic) waves. As for the admittance correction, Wise derived
a frequency-dependent shunt-admittance formula using the same Herzian potential
approach as he derived his series-impedance formula [25]. Subsequent studies are
carried out by M. Nakagawa to make clear the effects of the admittance correction
[26,27]. On the other hand, in 1955, H. Kikuchi derived a complete field solution for a
single phase case starting from Maxwell’s equations, and finally resulted in formulas of
frequency-dependent impedance and admittance [28,29]. He then gave the detailed
series expansions of the formulas in the same references. Later, L.M. Wedepohl et al.
derived a complete field solution for a mutiphase case, which is given in a nonlinear
infinite-integral equation, and they used an iteration scheme to numerically obtain the
propagation constant and the characteristic impedance [30,31]. Recently, M. D’Amore
et al. derived another complete solution using the logarithmic approximation of the
Sommerfeld integrals and Bessel functions [32]. All those studies regarding the
inclusion of the displacement currents indicates that the wave attenuation due to the earth

Carson-Pollaczek
attenuation
constant α

including
displacement currents

10 MHz frequency

Fig. 3.8 Attenuation correction due to displacement currents

3-12
conduction becomes smaller than that obtained by Carson-Pollaczek’s above 10 [MHz]
as shown in Fig. 3.8. Kikuchi pointed out that this is because of the transition from the
TEM mode to the surface-wave mode in which electromagnetic waves propagating along
a conductor consider the earth as a dielectric as explained by Sommerfeld-Goubau’s
theory [28].

It should be noted that practical simulations still use Carson-Pollaczek’s earth-


return impedance, because it gives a fairly good accuracy up to 10 [MHz] which is of
power engineers’ common interest. Also, if the displacement currents are considered,
then the definition of voltage becomes ambiguous, and makes it difficult to implement a
line model in an R-L-C based transient program such as the EMTP. Thus, Carson-
Pollaczek’s earth-return impedance can be regarded as the rigorous solution in terms of a
practical simulation. To avoid the tedious expansions and the discontinuity, and also to
gain more accuracy than Dubanton-Deri’s formula, a very accurate closed-form
approximation is proposed in the subsequent section.

Based on a fact that a ground usually consists of mutilayers of resistivity


(stratified earth), earth-return impedance formulas considering the multilayer earth are
proposed by different authors [33-37]. But resistivity data for the multilayer earth are
often not available, and Carson-Pollaczek’s homogeneous model seems to be used
normally.

As for the earth-return impedance of an underground conductor such as a cable


(which must be insulated from the earth soil), Pollaczek first gave a solution in 1931 [38].
The solution is given in a similar form to the overhead line case, and its self-impedance
formula is

Zeii = jω
µ
2π { (
K 0 ( pri ) − K 0 p ri2 + 4 hi2 + J eii , ) } (3.13a)

λ2 + p2
∞ e − 2 hi
J eii = ∫− ∞ e jλri d λ, p= jωµσ, (3.13b)
λ+ λ+ p
2 2

3-13
and the mutual-impedance formula is

Zeij = jω
µ

{ (
K 0 ( pDij ) − K 0 pDij′+ J eij , ) } (3.14a)

− ( hi + h j ) λ2 + p 2
∞ e jλd ij
J eij = ∫− ∞ e d λ, p= jωµσ, (3.14b)
λ+ λ+ p
2 2

where r : cable radius, h : cable depth, σ : earth conductivity, i, j: subscripts to


denote i-th and j-th conductors, dij : horizontal separation between i-th and j-th

conductors, Dij = ( hi − h j )2 + d ij 2 , and Dij′= ( hi + h j ) 2 + d ij 2 .

The solution is normally used as the standard solution in the field of power systems
transient analysis and is evaluated by similar expansions. An approximate closed-form
formula is proposed by Wedepohl et al. in 1973 [39], and later O. Saad et al. applied a
similar approximation methodology as Deri’s to derive a closed-form formula [40].
Detailed wave-propagation characteristics are presented by Ametani in [41].

Normally, an earth-return impedance formula includes the space impedance Zs =

jωLs in eq. (3.6), and thus the series-impedance matrix Z is given as the matrix sum of Ze

and Zc.

3-14
3.2.2 Closed-form formula of overhead-line earth-return impedance

In this section, a very accurate closed-form formula of overhead-line earth-


return impedance is derived using a double-exponential approximation, and validated by
numerical examples.

A. Derivation

The self-impedance formula derived by Carson and Pollaczek is rewritten in the


following :

µ 0 2h µ
Z es = jω ln + jω J es , (3.15a)
2π r π

∞ e − 2 hλ
J es = ∫0 d λ, (3.15b)
λ+ λ2 + jωµσ

where subscripts ii are abbreviated, but s denotes “self” impedance. In the same manner
as Deri et al., the distance of the complex ground-return plane

1
p= (3.16)
jωµσ

is substituted into eq (3.16b) :

e − 2 qw h
J es = ∫L d w, w = pλ, q= , (3.17)
w+ w2 + 1 p

where the integration path is shown in Fig. 3.9.

3-15
Im{w}

Re{w}

− π/4

C
L

Fig. 3.9 Integration path

The above integration can be performed along the real axis from zero to infinity,
because the integrand has no poles in the dashed region and is zero along path “C” at
infinity in Fig. 3.9 :

∞ e − 2 qw
J es = ∫0 d w, (3.18)
w+ w + 1
2

The derivative of eq. (3.18) with respect to q is given as :

∂J es ∞
=− ∫0 f ( w ) e − 2qw d w , (3.19a)
∂q
2w
f (w ) = . (3.19b)
w+ w + 1
2

To perform the above integration analytically, Deri et al. use an approximation :

f (w ) ≅1 − e − 2 w , (3.20)

3-16
and derived the following formula :

µ 0 2h µ  p
Z es( Deri ) = jω ln + jω ln1 + 
2π r π  h
(3.21)
µ 0 2(h + p)
= jω ln (µ = µ 0 ).
2π r

On the other hand, a much better accuracy can be gained by a double-


exponential approximation :

f ( w ) ≅ 1 − A e − 2αw − (1 − A) e − 2βw , (3.22)

where A = 0.131836, α = 0.26244, β = 1.12385.

The above parameter values are determined using a Newton-Raphson based nonlinear
optimization with 500 samples in the range of w = 0 to 10. Fig. 3.10 shows the
approximated results, and it is clear that the proposed double-exponential approximation
is far more accurate than Deri’s one. It should be noted that the proposed
approximation satisfies f(0) = 0 and f(∞ ) = 1. Thus, the infinite integration should be
performed accurately and can be regarded as practically exact. Substituting eq. (3.22)
into eq. (3.19a), the infinite integral now can be performed analytically. Finally, the
following closed-form formula is obtained :

1−
µ 0 2h µ  
A A
 αp   βp  
Z es ( proposed ) = jω ln + jω ln 1 +  1 +  
2π r 2π   h  h 
 

µ0  2(h + αp) 2(h + βp) 


= jω  A ln + (1 − A) ln  (µ = µ 0 ).
2π  r r 
(3.23)

3-17
1.2

0.8

0.6 given function f(w)

0.4 proposed approximation

Deri's approximation
0.2

0
0 2 4 6 8 10
w

Fig. 3.10 Approximations of function f(w)

According to Deri et al., eq. (3.21) can also be derived by the concept of the
complex ground-return plane which is a perfectly conducting current-return plane placed
below the earth surface at a complex depth equal to p [23]. Thus, as illustrated in Fig.
3.11 (a), all the earth-return current is assumed to flow into the complex ground-return
plane instead of considering the distribution of the return current below the earth surface.
This concept can also be applied to the proposed closed-form formula of eq. (3.23) in
order to derive the physical representation of the double-exponential approximation.
Compared with eq. (3.21), eq. (3.23) consists of two terms, when µ = µ0. The first term
represents a component of the earth-return current flowing into a complex ground-return
plane placed at a complex depth αp, and the second term is the rest of the earth-return
current in another complex ground-return plane at a complex depth βp. Thus, the total
earth-return current I is distributed into the two return planes with ratios A and 1 − A, as
shown in Fig. 3.11 (b). It should be clear that the proposed formula gains its accuracy
by use of the multiple complex ground-return planes for a better representation of the
earth-return cutrrent distribution.

3-18
I conductor I conductor

earth surface earth surface


αp
p
AI complex ground-
return plane #1 βp
I complex ground-
return plane (1 − A)I complex ground-
return plane #2

(a) Deri’s model (b) proposed model

Fig. 3.11 Deri’s and proposed models of complex ground-return plane

The mutual-impedance formula derived by Carson and Pollaczek is also


rewritten as follows (subscript m denotes “mutual” impedance) :

µ 0 Dij′ µ
Z em = jω ln + jω J em , (3.24a)
2π Dij π
− ( hi + h j )λ
∞ e
J em = ∫0 cos d ij λd λ, (3.24b)
λ+ λ2 + jωµσ

Here, variables w and q are defined as :

hi + h j
w = pλ, q= , (3.25)
2p

and a new variable ϕ ij is defined as :

3-19
d ij
ϕ ij = (3.26)
hi + h j

Hereafter the subscript ij of ϕ ij is omitted in the derivation. Thus, eq. (3.24b) can be
modified into :

e − 2qw
J em = ∫L cos(2ϕ qw ) d w
w+ w + 1
2
(3.27)
− 2 (1+ jϕ ) qw
1 e + e − 2(1− jϕ ) qw
= ∫L d w,
2 w+ w2 + 1

where the integration path is shown in Fig. 3.9.

As in the derivation of the self impedance, the integration path L is required to


be moved to the real axis. But, in this mutual-impedance case, a condition is imposed on
ϕ, because the integrand of eq. (3.27) has to be zero on the path C in Fig. 3.9. In order
to satisfy this, the real parts of exponents in eq. (3.27) Re{− 2(1 + jϕ)qw} and Re{− 2(1 −
jϕ)qw} have to be negative, i.e. arg(1 + jϕ) < π/4, because arg(q) = π/4 and − π/4 <arg(w)
< 0. Therefore, the condition of ϕ is

ϕ < 1. (3.28)

Under the above condition, the integration path is move to the real axis :

1 ∞ e − 2(1+ jϕ ) qw + e − 2 (1− jϕ ) qw
2 ∫0
J em = d w. (3.29)
w + w2 + 1

The derivative of the above equation with respect to q is :

∂J em 1 ∞
= − ∫0 f ( w ){(1 + jϕ ) e − 2(1+ jϕ )qw
+ (1 − jϕ ) e − 2(1− jϕ ) qw
}d w , (3.30)
∂q 2

3-20
where f(w) is the same function as the self impedance case in (3.19b).

The identical double-exponential approximation of eq. (3.22) can be applied to


perform the above integration analytically. After algebraic manipulations, the following
formula is derived :

µ0 (hi + h j ) 2 + d ij2
Z em( proposed ) = jω ln
2π (hi − h j ) 2 + d ij2
1− A
 
A
 (h + h + 2βp) 2 + d 2  
µ  (hi + h j + 2αp) + d ij
2 2
  i j ij  
+ jω ln  
2 π  (hi + h j ) 2 + d ij2
 
(hi + h j ) 2 + d ij2


   
 
(3.31)
  ( h + h + 2α p) 2 + d 2 
µ0   i j ij 
= jω  A ln
2π  
 (hi − h j ) 2 + d ij2 

 (h + h + 2βp) 2 + d 2 
ij 
+ (1 − A) ln
i j
(µ = µ 0 ),
 
 ( h − h ) 2
+ d 2

i j ij 

where A = 0.131836, α = 0.26244, β = 1.12385.

If the approximation of eq. (3.20) is used, Deri’s mutual-impedance formula is obtained :

(hi + h j ) 2 + d ij2  2 
µ  (hi + h j + 2 p) + d ij 
2
µ0
Z em( Deri ) = jω ln + jω ln
2π (hi − h j ) 2 + d ij2 2π  (hi + h j ) 2 + d ij2 

 
µ 0  (hi + h j + 2 p) + d ij
2 2
= jω ln  (µ = µ 0 ).
2π  (hi − h j ) 2 + d ij2

 
(3.32)

A comparison of eq. (3.31) with eq. (3.32) (when µ = µ0) indicates that the first
term of eq. (3.31) represents a component of the earth-return current with amplitude A

3-21
flowing into a complex ground-return plane placed at depth αp, and the second term
represents the rest of the return current with amplitude 1 − A in another complex
ground-return plane at depth βp, as shown in Fig. 3.11 (b).

As described above, the proposed formulas are derived in a similar manner to


Deri’s formulas, but uses a far more accurate approximation for the analytical evaluation
of the infinite integrals.

B. Numerical Validation

Fig. 3.12 shows the earth-return impedance of a single-phase line (h = 5 [m], r =


1 [cm], and earth resistivity ρ = 100 [Ω m]) calculated by the proposed self-impedance
formula eq. (3.23) and by a numerical integration of Carson-Pollaczek’s infinite integral
(3.15). The calculated result by the numerical integration can be regarded as the
rigorous solution, and the integration is performed by a modified Romberg method which
recursively discretizes an integration range until a given accuracy is achieved (in this case,
10− 6 in relative error is used). From the figure, the calculated resistance and inductance
values by the proposed formula agree well with the rigorous solution.

Fig. 3.13 shows the error of the proposed formula compared with the rigorous
solution, together with the errors of Deri’s formula eq. (3.21) and Carson-Galloway’s
expansions eqs. (3.9) - (3.11). The error of the proposed formula is less than 0.2 % in
resistance and less than 0.7 % in inductance, and it is clear that the proposed formula is
far more accurate than Deri’s one. It should be noted that Carson-Galloway’s
expansions show a numerical discontinuity at which two different expansions are
switched (r = 5).

3-22
2 1.5

proposed formula
numerical integration

inductance [µH/m]
resistance [Ω /m]

inductance
1 resistance

0.5

0 0 0
10 101 102 103 104 105 106 107
frequency [Hz]

Fig. 3.12 Self earth-return impedance


(h = 5 [m], r = 1 [cm], ρ = 100 [Ω m] )

Recent high-voltage transmission systems use a quite tall towers and a bundle of
conductors of which the equivalent radius is large. Fig. 3.14 shows the earth-return
impedance of a single-phase line of which the height, equivalent radius, earth resistivity
are 50 [m], 38 [cm], and 100 [Ω m] respectively (values are taken from a typical 500-kV
line in Japan). It is calculated by the proposed formula and also by the numerical
integration, and a good agreement is found. Fig. 3.15 shows the error curves of this
case obtained by the proposed and Deri’s formulas and Carson-Galloway’s expansions.
The error of the proposed formula is less than 0.9 % in resistance and less than 1% in
inductance, which is far less than that of Deri’s formula. A numerical discontinuity is
found in Carson-Galloway’s expansions also in this case.

In order to validates the proposed formula in a high earth-resistivity case, the


earth-return impedance of a single-phase line of which the earth resistivity is 500 [Ω m] (h
= 5 [m], r = 1 [cm]) is calculated and the error curves of those three formulas are shown

3-23
4

proposed formula
3
Carson-Galloway's expansion
relative error [%]

Deri's formula
2

1 discontinuity

-1 0
10 101 102 103 104 105 106 107
frequency [Hz]
(a) resistance error

4
proposed formula
3 Carson-Galloway's expansion
relative error [%]

Deri's formula
2
discontinuity

-1 0
10 101 102 103 104 105 106 107
frequency [Hz]
(b) inductance error

Fig. 3.13 Errors of self earth-return impedance formulas


(h = 5 [m], r = 1 [cm], ρ = 100 [Ω m] )

3-24
0.2 1

proposed formula
numerical integration
0.8

inductance [µH/m]
resistance [Ω /m]

0.6
0.1 inductance resistance
0.4

0.2

0 0 0
10 101 102 103 104 105 106 107
frequency [Hz]

Fig. 3.14 Self earth-return impedance


(h = 50 [m], r = 38 [cm], ρ = 100 [Ω m] )

in Fig. 3.16. The error curves of Fig. 3.16 can be obtained by shifting those of Fig. 3.13
rightward by a factor of 5. This is explained by eq. (3.16). In the equation, if the earth
resistivity ρ = 1 / σ is increased by a factor of 5, then a same value of p is obtained by
increasing frequency by the same factor. In this high-resistivity case, the error of the
proposed formula is less than 0.2 % in resistance and 0.3 % in inductance, which is far
more accurate than Deri’s formula.

Fig. 3.17 shows the mutual earth-return impedance of a two conductor system
(hi = hj = dij = 20 [m], ρ = 100 [Ω m]) calculated by the proposed mutual-impedance
formula eq. (3.31) and by the Romberg-method based numerical integration of Carson-
Pollaczek’s infinite integral eq. (3.24) as the rigorous solution. In the mutual-impedance
case, the integrand of eq. (3.24) is oscillating due to cosine function cos dijλ, and thus the
numerical integration has been carefully performed considering the oscillation. The
calculated result by the proposed formula agrees well with the rigorous numerical-

3-25
4
proposed formula, Carson-Galloway's expansion

3 Deri's formula
relative error [%]

2
discontinuity
1

-1 0
10 101 102 103 104 105 106 107
frequency [Hz]
(a) resistance error

4
proposed formula, Carson-Galloway's expansion
3 Deri's formula
relative error [%]

2
discontinuity
1

-1 0
10 101 102 103 104 105 106 107
frequency [Hz]
(b) inductance error

Fig. 3.15 Errors of self earth-return impedance formulas


(h = 50 [m], r = 38 [cm], ρ = 100 [Ω m] )

3-26
4

proposed formula
3
Carson-Galloway's expansion
relative error [%]

Deri's formula
2

-1 0
10 101 102 103 104 105 106 107
frequency [Hz]
(a) resistance error

proposed formula
3
Carson-Galloway's expansion
relative error [%]

Deri's formula
2

-1 0
10 101 102 103 104 105 106 107
frequency [Hz]
(b) inductance error

Fig. 3.16 Errors of self earth-return impedance formulas


(h = 5 [m], r = 1 [cm], ρ = 500 [Ω m] )

3-27
0.5 1

proposed formula
0.4 numerical integration
0.8

inductance [µH/m]
resistance [Ω /m]

0.3 0.6
inductance
resistance
0.2 0.4

0.1 0.2

0 0 0
10 101 102 103 104 105 106 107
frequency [Hz]

Fig. 3.17 Mutual earth-return impedance


(hi = hj = dij = 20 [m], ρ = 100 [Ω m] )

integration solution. Fig. 3.18 shows the error curves of the proposed formula with
varying a parameter ϕ ij in eq. (3.26). It is varied as 0.1, 0.5, 1, 5, and 10, and this is
equivalent to varying the horizontal separation of the two lines dij as 4, 20, 40, 200, and
400 [m] with line height hi = hj = 20 [m]. In the practical conductor configuration of a

transmission line, ϕ ij ≤ 1 (dij ≤ hi + hj) may be satisfied, thus the error of the proposed

formula is less than 1 % both in resistance and inductance. If a greater separation ϕ ij > 1
is considered, then the error becomes noticeable. This is because that a condition
eq.(3.28) is imposed to move the integration path from L to the real axis in the derivation
process of the proposed mutual-impedance formula. But it should be noted that the
proposed formula still shows a good accuracy less than 3.5 % in resistance and 8 % in
inductance even if the horizontal separation dij is equal to the 5 times of hi + hj. On the
other hand, Fig. 3.19 shows the error curves of Deri’s mutual-impedance formula eq.
(3.32), under the same condition. It is clear that the proposed formula shows a higher

3-28
10

ϕij = 10
ϕij = 5
relative error [%]

5
ϕij = 1

ϕij = 0.5
ϕij = 0.1
-5 0
10 101 102 103 104 105 106 107
frequency [Hz]
(a) resistance error

20
ϕij = 10
relative error [%]

ϕij = 5

10
ϕij = 1

0
ϕij = 0.5 ϕij = 0.1

100 101 102 103 104 105 106 107


frequency [Hz]
(b) inductance error

Fig. 3.18 Error of proposed mutual earth-return impedance formula


(h = 5 [m], r = 1 [cm], ρ = 100 [Ω m] )

3-29
10
ϕij = 5
ϕij = 10 ϕij = 1
ϕij = 0.5
relative error [%]

ϕij = 0.1

-10 0
10 101 102 103 104 105 106 107
frequency [Hz]
(a) resistance error

20
ϕij = 10

10
relative error [%]

ϕij = 0.1
-10 ϕij = 0.5

ϕij = 1
ϕij = 5
-20 0 1 2
10 10 10 103 104 105 106 107
frequency [Hz]
(b) inductance error

Fig. 3.19 Error of Deri’s mutual earth-return impedance formula


(h = 5 [m], r = 1 [cm], ρ = 100 [Ω m] )

3-30
accuracy than Deri’s one in the practical configuration. In case of ϕ ij = 10, Deri’s
formula shows a better accuracy than the proposed. This is estimated by that Deri’s
approximation eq. (3.20) has a higher rate of convergence than the proposed double-
exponential approximation eq. (3.22), as shown in Fig. 3.10, and the higher rate of
convergence helps to satisfy Re{− 2(1 + jϕ)qw} < 0 and Re{− 2(1 − jϕ)qw} < 0.
Because the error of Deri’s formula also exceeds 10 % in case of ϕ ij = 10, and thus
another method such as the numerical integration, which is enormously time-consuming
however, should be used to obtain a better accuracy.

The error curves of the proposed formulas presented in Figs. 13, 15, 16, 18, and
19 converge into about 1 % as frequency goes high around 100 [MHz], although they
seem to increase. In any case, because displacement currents have to be considered in
such a high frequency region as mentioned in Sec. 3.2.1, the accuracy in the frequency
region physically does not make any sense.

3-31
3.3 Conductor-Internal Impedance

The problem of wave propagation along a cylindrical coaxial conductor system


was solved by S.A. Schelkunoff, taking into account the skin effects of the conductors [3].
The result is given as the following impedance formulas. In the case of a solid
cylindrical conductor, the conductor-internal impedance is

jωµ c I 0 ( z )
Zc = , (3.33)
2 πz I1 ( z )

where µc : conductor permeability, z = r jωµ c σc , r : conductor radius, and

σc : conductor conductivity.

In the case of a tubular conductor,

jωµ c I 0 ( z 2 ) K1 ( z1 ) + K 0 ( z 2 ) I1 ( z1 )
Zc = , (3.34)
2 πz 2 I 1 ( z 2 ) K1 ( z1 ) − K1 ( z 2 ) I1 ( z1 )

where z1 = r1 jωµ c σc , z 2 = r2 jωµ c σc , and r1, r2 : inner and outer radii.

It should be noted that an ACSR mentioned in Sec. 2.1 (see Fig. 2.2) is treated as the
tubular conductor, because the resistivity of the enclosing aluminum wires is much
smaller than that of the core galvanized steel wires. To deal with an overhead line, the
above two formulas are sufficient. But to deal with a cable with a sheath considering
ground-return currents (either overhead or underground), a mutual impedance (transfer
impedance) between the internal- and external-return currents of the sheath is required to
represent the penetration of electromagnetic waves into the sheath. The mutual
impedance is given as :

3-32
1 1
Z cm = , (3.35)
2 πσc r1r2 I 1 ( z 2 ) K1 ( z1 ) − K1 ( z1 ) I 1 ( z 2 )

For a detailed description, a general formulation of the impedance and the admittance of a
cable is given by Ametani [42].

In an overhead-line case, the conductor-internal matrix Zc of eq. (3.6) is a


diagonal matrix, and the diagonal elements are given in eq. (3.33) or (3.34). On the
other hand, the conductor-internal matrix of a cable is not a diagonal matrix due to the
presence of the mutual impedance given in eq. (3.35).

3-33
References of Chapter 3

[1] N. Yamada, Electromagnetic Theory, The Institute of Electrical Engineers of Japan, 1950. (in
Japanese: 山田,電気磁気学,電気学会,1950.)
[2] J.R. Carson, “Wave Propagation in Overhead Wires with Ground Return,” Bell Syst. Tech., J.,
Vol. 5, pp. 539-554, 1926.
[3] S.A. Schelkunoff, “The Electromagnetic Theory of Coaxial Transmission Line and Cylindrical
Shields,” Bell Syst. Tech. J., Vol. 13, pp. 532-579, 1934.
[4] L.M. Wedepohl, “Application of Matrix Methods to the Solution of Travelling-Wave Phenomena
in Polyphase Systems,” Proc. IEE, Vol. 110, No. 12, pp. 2200-2212, 1963.
[5] K. Iwamoto, Traveling Waves and Introduction to their Calculations, Denki-Shoin, 1958. (in
Japanese: 岩本,進行波とその計算入門,電気書院,1958.)
[6] H.A. Wheeler, “Formulas for the Skin Effect,” Proc. IRE, pp. 412-424, September, 1942.
[7] C.S. Yen, Z. Fazarinc, and R.L. Wheeler, “Time-Domain Skin Effect Model for Transient
Analysis of Lossy Transmission Lines,” Proc. IEEE, Vol. 70, No. 7, pp. 750-757, 1982.
[8] A. Semlyen and A. Deri, “Time Domain Modelling of Frequency Dependent Three-Phase
Transmission Line Impedance,” IEEE Trans. on Power Apparatus and Systems, Vol. PAS-104,
No. 6, pp. 1549-1555, 1985.
[9] T. Henriksen, “First Order Approximation to the Frequency Dependency of Parameters in
EMTP,” Proc. First European Conference on Power Systems Transients (EPST’93), pp. 241-244,
1993.
[10] A.E. Davies, Y.K. Tong, P.L. Lewin, and D.M. German, “A Frequency-Independent Circuit
Model of a Conductor under Surge Conditions,” International Journal of Numerical Modeling :
Electronic Networks, Devices and Fields, Vol. 8, pp. 139-147, 1995.
[11] R.J. Meredith, “EMTP Modeling of Electromagnetic Transients in Multi-Mode Coaxial Cables by
Finite Sections,” IEEE PES Summer Meeting, 96 SM 437-4 PWRD, 1996.
[12] T. Matsuura, T. Noda, N. Nagaoka, and A. Ametani, “A Simplified Calculation Method of a Line
Surge Considering the Frequency-Dependent Effect,” Trans. IEE of Japan, Vol. 116-B, No. 6, pp.
706-711, 1996. (in Japanese: 松浦,野田,長岡,雨谷,「周波数依存効果を考慮した線路サ
ージ簡易計算」,電気学会論文誌 B, Vol. 116-B, No. 6, pp. 706-711, 1996.)
[13] A. Ametani, N. Nagaoka, T. Noda, and T. Matsuura, "A Simple and Efficient Method for
Including a Frequency-Dependent Effect in a Transmission Line Transient Analysis," Proc.
International Conference on Power Systems Transients ‘95 (IPST), pp. 11-16, , Lisbon, Portugal,
1995.
[14] A. Ametani, N. Nagaoka, T. Noda, and T. Matsuura, "A Simple and Efficient Method for
Including Frequency-Dependent Effects in a Transmission Line Transient Analysis,"
International Journal of Electrical Power & Energy Systems, October/November, 1996.
[15] T. Matsuura, T. Noda, and N. Nagaoka, “A Simple Method for Including Frequency-Dependent
Effects in a Transmission-Line Surge Calculation (Part 1), ” Proc. Annual Meeting of IEE of
Japan, Paper No. 1367, 1994. (in Japanese: 松浦,野田,長岡,「線路サージ計算における周
波数依存効果の簡略導入法 第 1 報」,平成 6 年電気学会全国大会,1367, 1994.)
[16] T. Matsuura, T. Noda, N. Nagaoka, and A. Ametani, “A Simple Method for Including

3-34
Frequency-Dependent Effects in a Transmission-Line Surge Calculation (Part 2), ” Proc. Annual
Meeting of IEE of Japan, Paper No. 1699, 1995. (in Japanese: 松浦,野田,長岡,雨谷,「線
路サージ計算における周波数依存効果の簡略導入法 第2報」,平成7年電気学会全国大
会,1699, 1995.)
[17] T. Matsuura, T. Noda, N. Nagaoka, and A. Ametani, “A Simplified Transmission-Line Surge
Calculation Method Considering Frequency-Dependent Effects,” Conference of Discharge and
High Voltage Technology, IEE of Japan, ED-95-188/HV-95-59, 1995, (in Japanese: 松浦,野田,
長岡,雨谷,「周波数依存効果を考慮した線路サージ簡略計算法」,電気学会放電高電圧
合同研究会, ED-95-188/ HV-95-59, 1995.)
[18] R. Rüdenberg, Z. ang. Math. Mech. Bd. 5, p. 361, 1925. or can be found in
R. Rüdenberg, Transient Performance of Electric Power Systems, p. 393, 1950.
[19] F. Pollaczek, “Über das Feld einer unendlich langen wechselstromdurch-flossen Einfachleitung,”
E.N.T., Band 3 (Heft 9), pp. 339-360, 1926.
[20] W.H. Wise, “Propagation of High-Frequency Current in Ground Return Circuits,” Proc. IRE, 22,
pp. 522-527, 1934.
[21] R.H. Galloway, W.B. Shorrocks, and L.M. Wedepohl, “Calculation of Electrical Parameters for
Short and Long Polyphase Transmission Lines,” Proc. IEE, Vol. 111, No. 12, 1964.
[22] C. Gary, “Approche Complète de la Propagation Multifilaire en Haute Fréquence par Utilisation
des Matrices Complexes,” EDF Bulletin de la Direction des Études et Recherches-Série B, No. 3/4,
pp. 5-20, 1976.
[23] A. Deri, G. Tevan, A. Semlyen, and A. Castanheira, “The Complex Ground Return Plane, A
Simplified Model for Homogeneous and Multi-layer Earth Return,” IEEE Trans., Power
Apparatus and Systems, Vol. PAS-100, pp. 3686-3693, 1981.
[24] R. Schinzinger and A. Ametani, “ Characteristics of a Transmission Line with Earth Return: A
Conformal Transformation Approach,” Proc. IEE, Vol. 121, No. 7, pp. 677-683, 1974.
[25] W.H. Wise, “Potential Coefficients for Ground Return Circuits,” Bell System Technical Journal,
Vol. 27, pp. 365-371, 1948.
[26] M. Nakagawa, “Admittance Correction Effects of a Single Overhead Line,” IEEE Trans. on Power
Apparatus and Systems, Vol. PAS-100, No. 3, pp. 1154-1161, 1981.
[27] M. Nakagawa, “Further Studies on Wave Propagation along Overhead Transmission Lines:
Effects of Admittance Correction,” IEEE Trans. on Power Apparatus and Systems, Vol. PAS-100,
No. 7, pp. 3626-3632, 1981.
[28] H. Kikuchi, “Wave Propagation on the Ground Return Circuit in High Frequency Regions,” J.
IEE of Japan, Vol. 75, No. 805, pp. 1176-1187, 1955. (in Japanese: 菊池,「高周波における大
地帰路回路について」,電気学会雑誌,75 巻,805 号,pp. 1176-1187, 1955.)
[29] H. Kikuchi, “Electromagnetic Fields on Infinite Wire at High Frequencies above Plane-Earth,” J.
IEE of Japan, Vol. 77, No. 825, pp. 721-733, 1957. (in Japanese: 菊池,「平面大地上の無限円
筒導体に関する高周波電磁界」,電気学会雑誌,77 巻,825 号,pp. 721-733, 1957.)
[30] L.M. Wedepohl and A.E. Efthymiadis, “Wave Propagation in Transmission Lines over Lossy
Ground: A New, Complete Field Solution,” Proc. IEE, Vol. 125, No. 6, pp. 505-510, 1978.
[31] A.E. Efthymiadis and L.M. Wedepohl, “Propagation Characteristics of Infinitely-Long Single-
Conductor Lines by the Complete Field Solution Method,” Proc. IEE, Vol. 125, No. 6, pp. 511-
517, 1978.
[32] M. D’Amore and M.S. Sarto, “A New Formulation of Lossy Ground Return Parameters for
Transient Analysis of Multiconductor Dissipative Lines,” IEEE PES Winter Meeting, 96 WM
074-5 PWRD, 1996.

3-35
[33] L.M. Wedepohl and R.G. Wasley, “Wave Propagation in Multiconductor Overhead Lines,
Calculation of Series Impedance for Multilayer Earth,” Proc. IEE, Vol. 113, No. 4, 1966.
[34] M. Nakagawa, A. Ametani, and K. Iwamoto, “Further Studies on Wave Propagation in Overhead
Lines with Earth Return: Impedance of Stratified Earth,” Proc. IEE, Vol. 120, No. 12, 1973.
[35] M. Nakagawa and K. Iwamoto, “Earth-Return Impedance for the Multi-Layer Case,” IEEE Trans.
on Power Apparatus and Systems, Vol. PAS-95, No. 2, pp. 671-676, 1976.
[36] A. Ametani, “Stratified Earth Effects on Wave Propagation − Frequency-Dependent Parameters,”
IEEE Trans. on Power Apparatus and Systems, Vol. PAS-93, pp. 1233-1239, 1974.
[37] A. Ametani, “Equations for Surge Impedance and Propagation Constant of Transmission Lines
above Stratified Earth,” IEEE Trans. on Power Apparatus and Systems, Vol. PAS-95, No. 3, pp.
773-781, 1976.
[38] F. Pollaczek, “Sur le champ produit par un conducteur simple infiniment long parcouru par un
courant alternatif,” Revue Gén. Elec., 29, pp. 851-867, 1931.
[39] L.M. Wedepohl and D.J. Wilcox, “Transient Analysis of Underground Power-Transmission
Systems: System-Model and Wave-Propagation Characteristics,” Proc. IEE, Vol. 120, No. 2, pp.
253-260, 1973.
[40] O. Saad, G. Gaba, and M. Giroux, “A Closed-Form Approximation for Ground Return
Impedance of Underground Cables,” IEEE Trans., Power Delivery, Vol. 11, No. 3, pp. 1536-1545,
1996.
[41] A. Ametani, “Wave Propagation Characteristics of Cables,” IEEE Trans. on Power Apparatus and
Systems, Vol. PAS-99, No. 2, pp. 499-505, 1980.
[42] A. Ametani, “A General Formulation of Impedance and Admittance of Cables,” IEEE Trans. on
Power Apparatus and Systems, Vol. PAS-99, No. 3, pp. 902-910, 1980.

3-36
4 TIME-DOMAIN REALIZATION OF
FREQUENCY DEPENDENCE

Summary

The purpose of this chapter is to realize the frequency dependence of a transmission


line in a time-domain transient calculation. Because a modal theory proposed by L.M.
Wedepohl reduces computation, the theory has been used to model the line, and time-domain
models based on the modal theory normally assume constant real modal-transformation
matrices. But when modeling a vertical overhead line or an underground cable, the
transformation matrices heavily depend on frequency, and the inclusion of the frequency-
dependent transformation matrices has been one of the main problems of time-domain line
modeling. The problem is first reviewed, and then a solution is given: phase-domain modeling.
Starting from the phase-domain formulation, an equivalent circuit to implement the line model
in an EMTP-type program is presented. This avoids convolution due to the modal
transformation, and possible numerical instability due to mode crossing. In the new approach,
time-domain convolutions are replaced by an IARMA (Interpolated Auto-Regressive Moving
Average) model that minimizes computation, and a fast and stable method to produce the
IARMA model is developed and sufficiently investigated in terms of error evaluation, stability,
and approximation order. Even if all convolution elements are stable, it is still possible that a
line model behaves as an active circuit due to a small amount of error, thus a numerically
unstable model. This is avoided by deriving a set of matrix-stability criteria. A steady-state
initialization method of the proposed line model is also developed to make feasible a simulation
starting from a steady-state, e.g. a fault calculation. In order to validate the proposed line
model, comprehensive switching and fault calculations are carried out and compared with field
tests and rigorous frequency-domain simulations at the end of this chapter.

4-1
4.1 Modal-Domain Modeling and Problems of Frequency-Dependent
Modal-Transformation Matrices

The Electro-Magnetic Transients Program (EMTP) is the most widely used


time-domain transient analysis program [1]. Because a modal theory [2,3] has been
used in the EMTP for the distributed representation of transmission lines, weighting
functions [4], exponential recursive convolution [5,8], linear recursive convolution [6],
Z-transforms [7], and modified recursive convolution [9,10] all have been developed
based on the modal theory. In the following, the modal theory is briefly summarized to
discuss the problems of the modal-domain modeling of a transmission line [2,11].

Consider an n-phase transmission line with length l as shown in Fig. 4.1. In

1 i1 i2 2
#1

#2

#n
v1 v2
l

Fig. 4.1 n-phase transmission line

4-2
the frequency domain, voltages and currents on the line at distance x [m] from the sending
end are expressed in the following column vectors :

V ( x , ω ) = (V1 , V2 , L , Vn )T , I ( x , ω ) = ( I1 , I2 , L , I n )T , (4.1)

where Vi, Ii : voltage and current on i-th conductor.

A set of vector partial-differential equations describing the dynamics of the line, called
Telegraphers’equations, is

dV ( x , ω )
− = Z (ω ) I ( x , ω ), (4.2a)
dx
dI ( x , ω )
− = Y (ω )V ( x , ω ), (4.2b)
dx

where Z(ω), Y(ω) : series-impedance and shunt-admittance matrices per unit


length of the line.

Algebraic manipulation of the above two equations gives the following set of wave
equations of voltage and current :

d 2V ( x , ω )
= P (ω )V ( x , ω ), (4.3a)
dx 2
d 2 I ( x, ω )
2
= P T (ω ) I ( x , ω ), (4.3b)
dx

where P (ω ) = Z (ω )Y (ω ) .

The general solution of the above set of equations is

V ( x , ω ) = e − Γ( ω ) x V f (ω ) + e Γ( ω ) x Vb (ω ), (4.4a)

{ }
I ( x , ω ) = Y0 (ω ) e − Γ( ω ) x V f ( ω ) − e Γ( ω ) x Vb (ω ) , (4.4b)

4-3
where Vf(ω), Vb(ω): forward and backward voltage-traveling-wave vectors,

Γ(ω ) = P 1/ 2 (ω ) : propagation-constant matrix, and Y0(ω) = Z− 1(ω)Γ(ω) :


characteristic-admittance matrix.

The above propagation function e− Γ(ω) cannot be directly evaluated, because Γ(ω) is a
matrix. In order to evaluate the propagation function, the following eigenvalue
decomposition may be used :

P (ω ) = A(ω ) diag{ pk (ω )} A − 1 (ω ) (k = 1, 2, ..., n), (4.5)

where diag(pk) denotes diagonal matrix with k-th diagonal element pk, pk(ω) :

eigenvalues of P(ω), A(ω) : matrix consisting of eigenvectors of P(ω) as its


columns.

Then, the propagation-constant matrix is given by :

Γ(ω ) = P 1/ 2 (ω ) = A(ω ) diag{γk (ω )} A− 1 (ω ) , (4.6)

where γk = p1k/ 2 = pk (k = 1, 2, ..., n).

Finally, the propagation function is given by :

e − Γ( ω ) x = A(ω ) diag{e − γk ( ω ) x } A − 1 (ω ). (4.7)

It is a common practice to analyze a transmission-line transient in the modal (eigenvalue)


domain rather than in the phase domain. Substituting eq. (4.5) into eqs. (4.3a) and
(4.3b), we obtain

d 2V ( x , ω )
2
= A(ω ) diag{ p k (ω )} A − 1 (ω )V ( x , ω ), (4.8a)
dx

4-4
d 2 I ( x, ω )
2
= B (ω ) diag{ pk (ω )} B − 1 (ω ) I ( x , ω ), (4.8b)
dx

where B(ω) = {A− 1(ω)}T.

The above equations can be modified as follows by multiplying A− 1(ω) and B− 1(ω) from
the left side respectively.

d 2 A − 1 (ω )V ( x , ω )
= diag{ pk (ω )} A − 1 (ω )V ( x , ω ), (4.9a)
dx 2
d 2 B − 1 (ω ) I ( x , ω )
2
= diag{ pk (ω )} B − 1 (ω ) I ( x , ω ). (4.9b)
dx

Therefore, n decoupled wave equations are obtained :

d 2 v( x, ω ) d 2vk ( x,ω )
= diag{ pk (ω )}v ( x , ω ) ⇒ = pk (ω )v k ( x , ω ), (4.10a)
dx 2 dx 2

d 2 i ( x, ω ) d 2ik ( x, ω )
= diag{ pk (ω )}i ( x , ω ) ⇒ = p k (ω )i k ( x , ω ), (4.10b)
dx 2 dx 2

where k = 1, 2, ..., n, and v(x,ω) = A− 1(ω)V(x,ω), i(x,ω) = B− 1(ω) I(x,ω) : modal-


domain voltage and current vectors.

In eq. (4.6), the propagation-constant matrix is decomposed into n scalar values


γ
k’s, which are given as the square root of pk’s in the above equations. Each set of the
above voltage and current wave equations represents a single-phase line of which the
propagation constant is γ
k. Thus, the mathematical eigenvalue decomposition
corresponds to the decomposition of a coupled n-phase line into n decoupled single-phase
lines, and the decoupled domain is called the modal domain. The interface between the
phase and modal domains is due to the following matrices :

voltage modal-transformation matrices

4-5
A(ω) : modal → phase, A− 1(ω) : phase → modal
current modal-transformation matrices
B(ω) : modal → phase, B− 1(ω) : phase → modal

As is clear from eqs. (4.10a) and (4.10b), the n by n coupled problem is reduced into an
n-decoupled problem which requires 1/n computation in order to perform a transient
calculation. If this approach is used in an EMTP-type program, which is a time-domain
transient calculation program, it is convenient to assume that the transformation matrices
are real and constant (frequency-independent) for the time-domain implementation.
Otherwise the convolution of those matrices may be required to take into account their
frequency dependence.

As mentioned at the beginning of this section, all the frequency-dependent line


models installed in the EMTP have been developed based on the above modal theory [4-
10], and real constant modal-transformation matrices are assumed in those models. In
the case of a horizontally arranged overhead line, its transformation matrices are almost
real and constant. For example, Fig. 4.2 shows the first-column elements of the voltage
transformation matrix A(ω) of a single-circuit 500-kV horizontal line (Azumi Trunk of
the Tokyo Electric Power Co., see Sec. 4.8.1-A and ref. [33]). On the other hand, the
transformation matrices of a vertically arranged overhead line or an underground cable
have been known to depend heavily on frequency. In the case of the vertical line,
because its ground wires conduct more return current than the earth above a certain
frequency due to the difference of the ground-wire and earth resistivity and due to the
skin effects, the modal voltage and current distribution changes at the frequency. Fig.
4.3 shows the third-column elements of the transformation matrix A(ω) of a double-
circuit 500-kV vertical line (New Chichibu-Tochigi Trunk of the Tokyo Electric Power
Co., see Sec. 4.8.1-B and ref. [34]), and the modal-distribution change is observed at 10
[kHz]. In case of the underground cable, when the penetration depth which varies with

4-6
1

0.8

0.6

0.4

A11, A31
0.2
(A21 = 1: normalization)

0 -1
10 100 101 102 103 104 105 106 107
frequency [Hz]

Fig. 4.2 First-column elements of voltage transformation matrix A(ω)


of a single-circuit 500-kV horizontal line

1
A23, A53
0.8 A33, A63
(A13, A43 = 1: normalization)
0.6

0.4

0.2

0 -1
10 100 101 102 103 104 105 106 107
frequency [Hz]

Fig. 4.3 Third-column elements of transformation matrix A(ω)


of a double-circuit 500-kV vertical line

4-7
respect to frequency is smaller than the thickness of its sheath or pipe, shield effects are
significant. When larger, earth-return effects can be observed. Therefore, the modal
distribution largely varies with respect to frequency [11]. Fig. 4.4 shows the frequency
dependence of the first-column elements of the modal-transformation matrix A(ω) of a
275-kV pipe-type oil-filled cable (an underground line of Chubu Electric Power Co., see
Sec. 4.8.1-C and ref. [35]), and the modal distribution abruptly changes at 20 [kHz] due
to the completion of the shield effects (the completion of the coaxial modes). Therefore,
in the vertical overhead line and underground cable cases, the frequency dependence of
the modal-transformation matrices have to be taken into account for an accurate
simulation.

Theoretically, the frequency dependence of the modal-transformation matrices


can be introduced into a time-domain simulation by convolution [12,13]. But the

0.8

0.6

0.4
A11, A31
0.2
A21
0
10-1 100 101 102 103 104 105 106 107
frequency

Fig. 4.4 First-column elements of modal-transformation matrix A(ω)


of a 275-kV pipe-type cable

4-8
practical implementation of the transformation-matrix convolution seems enormously
complicated in terms of fitting and eigenvalue tracing by mode crossing. It is sometimes
observed that an element of the transformation matrices is a non-causal function having a
response in the negative time region. If the element is a non-causal function, numerical
fitting to apply a recursive convolution is impossible. Also, a small numerical error may
cause eigenvalue switchovers (= mode crossing), especially in case of a cable, and the
practical implementation of the reformation may be numerically difficult. Recently,
investigation regarding the above issue is quite active (for example, see ref. [15]).

As described above, if the transformation matrices can be assumed real and


constant, the modal-domain approach is numerically efficient. But, if heavily frequency-
dependent transformation matrices, i.e. if the frequency dependence has to be considered,
then the approach has the above problems to establish an accurate and solid model. One
may continuously pursue the modal-domain approach to develop a robust eigenvalue
calculation/tracing method such as one in ref. [15]. It should be noted that, if the
convolution of the transformation matrices is introduced, there is no computational
advantage compared with a direct phase-domain approach which performs a transient
calculation using eqs. (4.4a) and (4.4b), because both the approach requires 2n2
convolution operations. Thus, in the following sections, the author proposes a phase-
domain line model in which the modal transformation itself is avoided.

4-9
4.2 Phase-Domain Formulation [16]

The preceding section shows basic equations of both the modal-domain and
phase-domain approaches. In order to derive a set of equations which describe the
complete dynamics of a transmission line in the phase domain, including the frequency-
dependent effects of modal-transformation matrices, we start from the phase-domain
general solution eqs. (4.4a) and (4.4b). The solution is rewritten here :

V ( x , ω ) = e − Γ( ω ) x V f (ω ) + e Γ( ω ) x Vb (ω ) (4.11a)

I ( x , ω ) = Y0 (ω ){e − Γ( ω ) x V f (ω ) − e Γ( ω ) x Vb (ω )} . (4.11b)

Eliminating Vb(ω) from eqs. (4.11a) and (4.11b), we obtain

Y0 (ω )V ( x , ω ) + I ( x , ω ) = 2Y0 (ω )e − Γ( ω ) xV f (ω ). (4.12)

Let a vector of the sending end voltages be V1(ω) and the receiving end V2(ω). At the
sending end (x = 0) and at the receiving end (x = l), the following two equations are
obtained from eq. (4.12).

x = 0 : Y0 (ω )V1 (ω ) + I1 (ω ) = 2Y0 (ω )V f (ω ) (4.13a)

x = l : Y0 (ω )V2 (ω ) + I 2 (ω ) = 2Y0 (ω )e − Γ( ω )lV f (ω ) (4.13b)

Eq. (4.13a) can be modified as

2V f (ω ) = V1(ω ) + Z 0 (ω ) I1(ω ). (4.14)

Substituting eq. (4.14) into eq. (4.13b), we obtain

4-10
I1 (ω ) = Y0 (ω )V1 (ω )
− Y0 (ω )e − jωτ
H (ω ){V2 ( ω ) + Z 0 (ω ) I 2 (ω )}. (4.15a)

By eliminating Vf(ω) from eqs. (4.11a) and (4.11b) in the same manner, we obtain

I 2 ( ω ) = Y0 (ω )V2 (ω )
(4.15b)
− Y0 (ω )e − jωτ
H (ω ){V1( ω ) + Z 0 (ω ) I1 (ω )},

where H(ω) = ejωτe− Γ(ω)l : phase-domain wave-deformation matrix, τ : minimum


traveling time (see footnote *1 on the next page), and Z0(ω) = Y0− 1(ω) :
characteristic-impedance matrix.

If the above equations are transformed into the time domain, they require 8 convolutions.
To reduce the computation time, the following relation (footnote *2) :

Y0 (ω ) H (ω ) = H T (ω )Y0 (ω ) (4.16)

is applied to eqs. (4.15a) and (4.15b), and we obtain

I 1 (ω ) = Y 0 (ω )V1 (ω ) − e − jωτ
H T (ω ){Y 0 (ω )V 2 (ω ) + I 2 (ω )} (4.17a)

I 2 (ω ) = Y0 (ω )V2 (ω ) − e − jωτ
H T (ω ){Y0 ( ω )V1 (ω ) + I 1 (ω )}. (4.17b)

Transforming eqs. (4.17a) and (4.17b) into the time domain (lower case letters are used
to denote their time domain counterparts),

i1 ( t ) = y 0 (t )∗ v1 ( t ) − i p1 ( t ) (4.18a)

i p1 ( t ) = hT ( t )∗{ y 0 ( t )∗ v 2 ( t − τ) + i2 ( t − τ)} (4.18b)

i 2 ( t ) = y 0 ( t )∗ v 2 ( t ) − i p2 ( t ) (4.18c)

i p 2 (t ) = hT ( t )∗{ y 0 ( t )∗ v1 ( t − τ) + i1 ( t − τ)}, (4.18d)

4-11
where ### : symbol to indicate matrix-vector convolution, and h(t) = F− 1
{H(ω)} (inverse Fourier transform).

The above equations are compatible with Schnyder-Bergeron’s expression [1]. As the
terms underlined with − and = are the same, the number of convolutions is reduced to 4.
Consequently, the computation time is greatly reduced using eqs. (4.18a) to (4.18d).
The set of the equations is the basis of the present phase-domain modeling, and solves the
linear and frequency-dependent part of the proposed line model enclosed by a broken line
in Fig. 2.7 corresponding to eqs. (2.12) (a) and (b).

__________________________________________________________

(*1) The minimum traveling time τ is the traveling time of the fastest mode. If
the k-th mode is the fastest, it is given as τ = βk(ω)l/ω |ω = ω max, where βk : phase-constant
k = α k + jβk), and ω max : highest frequency of interest.
of the k-th mode (γ

(*2) Let A = A(ω) and B = B(ω) be a voltage and current modal-transformation


matrix respectively, as used in the preceding section.

Y0 (ω )e − Γ( ω )l = {B diag( y 0 k ) A − 1}{ A diag(e − γk l ) A − 1}


= B diag( y 0 k ) diag(e − γk l ) A − 1 = B diag(e − γk l ) diag( y 0 k ) A − 1
= {B diag(e − γk l ) B − 1}{B diag( y 0 k ) A − 1}

Using the relation B = (A− 1)Tand B− 1=AT,

= {( A − 1 ) T diag(e − γk l ) A T }{B diag( y 0k ) A − 1}


= { A diag(e − γk l ) A − 1}T {B diag( y 0 k ) A − 1} = {e − Γ( ω ) l } T Y0 (ω ).

Therefore, Y0(ω)H(ω) = HT(ω)Y0(ω).

4-12
4.3 Equivalent Circuit for Time-Domain Simulation [16]

The set of equations (4.18a) to (4.18d) indicates that the equivalent circuit for a
time-domain simulation is expressed as an n-terminal admittance paired with an n-
terminal current source as illustrated in Fig. 4.5 (a). It can be seen that the first term of
eqs. (4.18a) and (4.18c) corresponds to the n-terminal admittance, and the second term to
the n-terminal current source corresponding to past history obtained from eqs. (4.18b)
and (4.18d). When including the equivalent circuit into a nodal-conductance
representation, how to model the frequency-dependent characteristic admittance Y0(ω) is

important. To model Y0(ω) with ARMA models, the convolution operation y0(t)∗v(t) is
easily decomposed as :

y0 ( t )∗ v ( t ) = y00 v ( t ) + y01( t )∗ v (t − ∆t ). (4.19)

Matrix y00 is real and constant, where each element is the first-term coefficient of the
numerator of an ARMA model representing the frequency dependence of the element,
and y01(t)∗v(t − ∆t) is a matrix-vector convolution without instantaneous responses.
From eq. (4.19), eqs. (4.18a) to (4.18d) lead to :

i1 ( t ) = y 00 v1 ( t ) − i ′
p1 ( t ) (4.20a)

i p′
1 ( t ) = i p1 ( t ) − y 01 ( t )∗ v1 ( t − ∆t ) (4.20b)

i 2 ( t ) = y 00 v 2 ( t ) − i p′(
2 t) (4.20c)

i p′
2 ( t ) = i p 2 ( t ) − y 01 ( t )∗ v 2 ( t − ∆t ). (4.20d)

where i p1 ( t ) = hT ( t )∗{ y 0 ( t )∗ v 2 ( t − τ) + i2 ( t − τ)} ,

i p2 (t ) = hT (t )∗{ y 0 (t )∗ v1 (t − τ) + i1 (t − τ)} .

4-13
1 2
i1 i2

τ
v1 y0 ( t ) y0 ( t ) v2
ip1( t ) ip2( t )
(a)

1 2
i1 i2

τ
v1 y00 y00 v2
ip1'( t ) ip2'( t )

(b)

Fig. 4.5 Equivalent circuits for time-domain simulation by phase domain modeling

4-14
This set of equations corresponds to the equivalent circuit illustrated in Fig. 4.5
(b). It can easily be introduced into programs based on the nodal-conductance
representation, such as the EMTP. First, the constant matrix y00 is added directly to the
nodal-conductance matrix of the whole system before the transient calculation. During
the simulation, the past-history current-source vectors i'p1(t) and i'p2(t) are then added to
the current-source vector of the system at each time step.

4-15
4.4 Consideration of Conductor Configuration Symmetry to Reduce
Fitting [17]

Because the proposed line model is a phase-domain model, the computation time
of the frequency-dependence synthesis is in proportional to n2. Thus, the reduction of
the synthesis time (fitting time) is important, although a linearized least-squares fitting
method, proposed in the subsequent section, is quite fast. A reduction of the fitting time
can be achieved by considering the symmetry of a line configuration. As illustrated in
Fig. 4.6, the conductor configuration of an overhead line is normally symmetrical with
respect to its tower as the axis. A function “sym(i)” is introduced for the reduction
considering the symmetry. The function returns a corresponding phase number i’to a
given phase number i, in the meaning of the symmetry :

i ′= sym(i ), (4.21)

if there is no corresponding number, i itself is returned.

For example, as shown in the table in Fig. 4.6, if i = 2 (phase b) is given, then i’
= sym(i) = 5 (phase b’) is returned, because phases b and b’are symmetrical to each other.
Let us assume that the fitting of the wave-deformation matrix H(ω) is performed in the
order of (i, j) = (1,1), (1,2), ..., (1, n), (2,1), (2,2), ..., (n, n). If the following
condition is met, an element Hi’j’(ω) having the identical frequency response to Hij(ω) has
already been synthesized :

i ⋅n + j > i ′
⋅n + j ′
, (4.22)

where i ′= sym(i ), and j ′= sym( j ) .

4-16
g g'
sym関数対応表
sym function
a a' i sym( i )
1 (a) 4
b b' 2 (b) 5
3 (c) 6
c c' 4 (a') 1
axis of 5 (b') 2
対称線
symmetry 6 (c') 3

Fig. 4.6 Line configuration having symmetry

Thus, the fitting of Hij(ω) can be omitted, and the fitting result of Hi’j’(ω) is used. As for

the characteristic-admittance matrix Y0(ω), a further reduction can be achieved


considering the symmetry of the matrix itself :

Y0ij (ω ) = Y0 ji (ω ) . (4.23)

Consider a double-circuit overhead line illustrated in Fig. 4.6. The number of elements,
H(ω) and Y0(ω), to be fitted is 36 respectively, because the number of phases is 6
assuming that the voltages of the ground wires are zero by a matrix manipulation. If the
proposed fitting-reduction method is applied, the number of elements of H(ω) to be fitted
becomes 18 (half), and that of Y0(ω) 16 by considering the matrix symmetry. Also, most
of cable systems have symmetry with respect to an vertical axis, and thus the proposed
fitting-reduction method can be applied.

4-17
4.5 IARMA Convoluation [16,18]

4.5.1 Interpolated ARMA (IARMA) model

An ARMA (Auto-Regressive Moving-Average) model is a well-established


numerical model used in the field of system identification and signal processing. The
present thesis extends the model to an IARMA (Interpolated ARMA) model and utilizes
for the convolution of the wave-deformation matrix H(ω) and the characteristic-
admittance matrix Y0(ω) in the phase domain. First, the ARMA model is summarized as
the basis.

An ARMA model essentially represents a discrete-time system. Input x(t) and


output y(t) are sampled at the calculation interval ∆t. Using Z-transform theory, an
ARMA model is defined by the following rational function of z− 1 (= e− jω ∆t) in the z-
domain :

a 0 + a1 z − 1 + L + a N z − N
G( z ) = , (4.24)
1 + b1 z − 1 + L + bN z − N

where an, bn : coefficients and N : order of ARMA model.

Because the operator z− n denotes a delay of n samples, eq. (4.24) is transformed into the
time domain as :

y (t ) = a 0 x (t ) + a1 x (t − ∆t ) + L + a N x (t − N∆t )
(4.25)
− b1 y (t − ∆t ) − L − b N y (t − N∆t ).

The above equation is the time-domain representation of the ARMA model, and is
equivalent to applying the recursive convolution [5-10]. Using this equation, output y(t)

4-18
can be calculated by only 2N + 1 multiplications and 2N additions.

To calculate a system transient accurately and efficiently, a small time step


should be selected for a fast response (short rise time) in order not to lose its peak value,
and a large time step for a slow response (long rise time) to avoid an excessive
computation. This idea was introduced in the MODELS language, which was
implemented in the ATP version of EMTP, for providing an efficient modeling
methodology [19]. To apply the idea to the line model, each ARMA model, which
represents the frequency dependence of an element of the wave-deformation matrix or of
the characteristic admittance matrix, should use an economical time step considering its
frequency characteristic. An interpolation technique may be used to interface with the
external circuit at intervals of the simulation time step ∆ts used as the common time step
in the host transient calculation program such as the EMTP. In this thesis, the linear
interpolation technique is applied to get an instantaneous input to the ARMA model and
also an instantaneous output from the model at intervals of ∆ts. For a simplification, it
is convenient that the interpolation procedure is considered as a part of the ARMA model
as illustrated in Fig. 4.7 (a), and thus the whole procedure is named as an IARMA
(Interpolated ARMA) model. Fig. 4.7 (b) shows the interpolation timing of the IARMA
model. When the simulation time step ∆ts is greater than the time step of an IARMA
model ∆t, then ∆ts corresponds ∆t1 and ∆t corresponds ∆t2. When ∆ts < ∆t, vice
versa. It should be noted that an ARMA model represents a discrete-time system with a
fixed time step whereas an IARMA model represents a continuous-time system.
Therefore, an IARMA model does not have the restriction of the time step, which is
sometimes an inconvenience of the ARMA model. Accurate HVDC and power-
electronics simulations often need a variable time step approach for an on/off-timing
adjustment of switching devices [20], and thus it may be advantageous that the new
IARMA model is fully compatible with an EMTP-type program having a variable time
step.

When a frequency characteristic is approximated by a rational function of s


(Laplace operator), it is independent of the simulation time step, because it can be

4-19
IARMA model
linear linear
x (t ) inter- ARMA inter- y (t )
polation model polation

(a) calculation procedure of IARMA model

interpolation interpolation
0 ∆ t1 ∆ t1
t
∆ t2

(b) interpolation timing

Fig. 4.7 Concept of IARMA model

transformed into the recursive-convolution form using the BLT (Bi-Linear Transform),
just before a transient calculation. Therefore, the s-rational function has a flexibility in
terms of the time step. This may be a good reason to use the s-rational function instead
of the ARMA model (z-rational function). But because all the coefficients of an s-
rational function have to be positive, assuring that it is physically realizable, the restriction
makes the least-squares fitting procedure numerically unstable. An asymptotic tracing
approach by allocating poles directly on the s-plane is proposed [10], and it is successful
to reproduce a modal-domain response being a minimum-phase-shift function. But it
may be difficult to adopt the approach to a phase-domain response which is often not a
minimum-phase-shift function. Moreover, a phase-domain response, including
discontinuities due to modal traveling-time differences, cannot easily be approximated by

4-20
the s-rational function. On the other hand, a direct Z-domain fitting method using the
ARMA model, which takes advantage of the one-sample-delay nature of the Z-operator
to model the discontinuous response, is developed in the following sections with a high
level of perfection. Because the coefficients of an ARMA model can be negative, i.e. no
restriction of the sign, the simple least-squares method (but linearized for a z-rational
function) is suitable for the fitting procedure. And, the least-squares fitting can produce
a non minimum-phase-shift function, because not only the magnitude of the given
frequency response but also the phase are taken into account. Furthermore, the time
step restriction, which is an inconvenience of the ARMA model, is practically overcome
by use of the IARMA model in the present line model. As a result, the IARMA model
approach may be one of the most efficient approaches for phase-domain transmission-line
modeling.

4.5.2 Time-step determination

For single-core cables, because a sheath self-propagation response consists only


of the earth-return mode, the response is comparatively slow with respect to the coaxial
mode and a large time step can be used. Also, a large time step can generally be used
for the convolution of a characteristic admittance, because the instantaneous response is
already included in the conductance matrix of the host transient calculation program and
the frequency dependence is observed only in a low-frequency region. Usually, a user
provides an error tolerance ε, which is the maximum permitted error of the fitting
procedure. For a phase-domain wave-deformation response Hij(ω), when the response
is smaller than ε as illustrated in Fig. 4.8 (a) :

H ij (ω ) < ε , (4.26)

then the response at f = ω/2π can be neglected. Thus, the highest frequency fmax which
meets the above criterion determines an economical time step considering the well-known

4-21
sampling theorem :

∆t = 1 / 2 f max . (4.27)

The above ∆t is the maximum time step without losing its accuracy defined by ε. When
the response is very fast and we do not find fmax, about some MHz (∆t ≅ 0.1µs) may be
recommended considering the accuracy of impedance formulas such as Carson-
Pollaczek’s. Those formulas may not have enough accuracy above some MHz, because
displacement currents are ignored, and an admittance correction or a complete field
solution may be needed for a better accuracy as mentioned in Sec. 3.2.1. But when the
simulation time step is much smaller than ∆t ≅0.1µs, such as for surge calculation of a
GIS (Gas Insulated Substation), then ∆t should be set to the smallest possible simulation
time step to avoid ∆t > τ (traveling time). For a characteristic admittance element Y0ij,
because the instantaneous response is already included in the conductance matrix of the
host program, the following criterion is used as illustrated in Fig. 4.8 (b) :

Y0ij (ω ) − Y0 ij ( ∞ ) < ε . (4.28)

The above time-step determination is performed before the fitting procedure, and the time
step is hidden inside the IARMA model.

To show the effectiveness of the proposed IARMA model, (1,1) and (2,2)
elements of the wave-deformation matrix of an underground cable are fitted using the
IARMA model, and their step responses are shown in Fig. 4.9. In this case, the error
constant is set to ε = 1%. Because the time-domain counterpart of the (1,1) element has
a short rise time due to the coaxial wave, the selected time step is small enough to
represent its sharp response accurately. On the other hand, the (2,2) element consists
only of the earth-return component, and the selected time step is larger by about 20 times
than that of (1,1), but the response is reproduced accurately.

4-22
|Hij(jω)|

less than ε

frequency fmax

(a) wave-deformation response Hij(ω)

|Y0ij(jω)| |Y0ij(j∞ )|
less than ε

fmax

frequency

(b) characteristic admittance response Y0ij(ω)

Fig. 4.8 Determination of fmax to determine ∆t

1.5
voltage, p.u.

h11(t), △t = 0.447 μs
1

0.5 h22(t), △t = 8.91μs

0
0 100 200 300 400 500
time, microseconds

Fig. 4.9 Step responses of phase-domain wave-deformation using IARMA model

4-23
4.5.3 Modeling of modal traveling-time differences

Because an element of the phase-domain wave-deformation matrix Hij(ω)


consists of n modal components, its time-domain response may have discontinuities every
∆τk = τk − τ (see footnote below) as illustrated in Fig. 4.10. Thus, it is difficult to
approximate this discontinuous response with a conventional ARMA model expressed in
(4.24), of which the impulse response is given as a sum of N complex exponentials with
denominator order N. In order to express the discontinuous response, a modification is
made in the expression of the ARMA model. The following condition determines
whether the k-th mode component is dominant or not in Hij(ω) (see the next paragraph for
the derivation) :

Aik Akjinv e − γk l > ε M . (4.29)

where Aik : (i, k) element of transformation matrix A, Ainvkj : (k, j) element of


inverse transformation matrix A− 1, γ
k : modal propagation constant of k-th mode,

and εM : a small constant.

The above Aik, Ainvkj, and γ


k are obtained directly when evaluating the propagation-

function matrix e− Γ(ω). When the condition eq. (4.29) is met, the impulse response of Hij(
ω) has a discontinuity at t = ∆τk. In this case, the ARMA model can be modified in
order to represent the discontinuity using the one sample delay nature of the Z-operator :
the delay operator z− qk (qk = integer(∆τk/∆t)), which corresponds to the traveling-time
difference, is applied to the corresponding numerator term. For example, supposing that

__________________________________________________________

Minimum traveling time τ is the traveling time of the fastest mode. The
traveling time of each mode k is given as τk = βk(ω)l/ω |ω = ω max, where βk : phase-constant
k = α k + jβk), and ω max : highest frequency of interest
of the k-th mode (γ

4-24
earth-return mode 0

aerial modes
1, 2

discontinuity

time

Fig. 4.10 Time-domain discontinuous response due to modal traveling-time differences

Hij(ω) has a discontinuity due to the k-th mode, the modified ARMA model with order N
= 5 is expressed as :

( a0 + a1 z − 1 + a 2 z − 2 ) + z − qk ( a qk + a qk + 1 z − 1 + a qk + 2 z − 2 )
H ij (ω ) = , (4.30)
1 + b1 z − 1 + b2 z − 2 + b3 z − 3 + b4 z − 4 + b5 z − 5

where qk = integer(∆τk / ∆t).

This representation is extended to include each mode. Note that when both the k1-th
and k2-th modes are involved dominantly in Hij(ω) and the traveling time differences ∆τk1

and ∆τk2 are almost the same, the two modes can be represented with one qk. The
general form is

K  Mk

∑ z − qk ∑ aqk + m z − m 
k =0 m= 0 K 
H ij (ω ) = ,  ∑ ( M k + 1) = N + 1 (4.31)
N
k = 0 
1+ ∑ bn z −n

n =1

where q0 = 0, and K : number of modes of which the traveling-time difference


is considered.

4-25
The condition of the mode detection eq. (4.29) is derived as follows. Phase-
domain wave-deformation with respect to a traveling wave is expressed as Vd(ω) =

H(ω)Vs(ω), where Vd(ω) is a deformed voltage vector at distance l, and Vs(ω) is an

applied voltage at the sending end. Thus, an element of H(ω) can be expressed in the
following equation (ej : the j-th unit vector).

H ij (ω ) = {Vd (ω )}i − th , when Vs (ω ) = e j (4.32)

On the other hand, the applied modal voltage being given as vs(ω) = A− 1(ω)Vs(ω) =

A− 1(ω)ej, the deformed modal voltage is expressed in the following modal propagation.

{v d (ω )}k -th = { A− 1 (ω )e j }k -th e − γk l = Akjinv e − γk l (4.33)

Transforming the above equation into the phase domain, the phase-domain deformed
voltage is written as :

n
{Vd (ω )}i-th = { A(ω )v d (ω )}i-th = ∑ Aik Akjinv e − γk l . (4.34)
k =1

Substituting (4.34) into (4.32), we obtain

n
H ij (ω ) = ∑ Aik Akjinv e − γk l . (4.35)
k =1

Consequently, it is possible to estimate whether the k-th mode component is dominant or


not in Hij(ω) with eq. (4.29).

4-26
4.5.4 Linearized least-squares method for ARMA identification

A. Mathematical background

This section describes a highly efficient method called “Linearized Least-Squares


(LS) Method”, which synthesizes an ARMA model reproducing a frequency response

given as a set of data (ωk,Gk) (k = 1, 2,… , K) defined at a discrete angular frequency [21-
26]. The linearized LS method is applied to fit each element of the wave-deformation
matrix H(ω) and of the characteristic-admittance matrix Y0(ω) with an ARMA model, in
order to apply the IARMA convolution for the time-domain realization of the frequency
dependence. Let us represent each element of H(ω) or Y0(ω) as the discrete frequency

response (ωk,Gk) (k = 1, 2,… , K). The relation :

z = e jω∆t (4.36)

replaces the discrete frequency response with a discrete z-domain function (zk,Gk). If
(zk,Gk) is fitted by an ARMA model using a conventional least-squares method, the error
function for the k-th datum is

errk = G( z k ) − Gk (4.37)

Substitution of eq. (4.24) into the above errk gives an error function that is nonlinear due
to its rational polynomial form (denominator coefficients b1, b2, ..., bN are included in errk
as nonlinear parameters). This case requires the application of a nonlinear optimization
method to identify the parameters of the ARMA model a0, a1, ..., aN, b1, b2, ..., bN. An
initial value and a differential coefficient would be required, because the nonlinear
optimization method searches for an optimum value using an iterative calculation.
Moreover, the assurance of convergence to an optimum value would be highly dependent
on the choice of the initial value.

On the other hand, the proposed linearized LS method chooses the following

4-27
error function :

err' k = ( a0 + a1 z k− 1 + L + a N z k− N ) − (1 + b1 z k− 1 + L + bN z k− N )Gk (4.38)

The above error function includes all the parameters as linear parameters. This is why
the present method is called “Linearized LS Method”, and a time-consuming and
sometimes unstable iterative calculation is no longer required.

To obtain the least-squares solution of eq. (4.38), the conventional normal


equation method is well-known. But the accuracy will get worse as the order N
increases. Therefore, Householder’s transformation is applied to improve the accuracy.
As a result, the linearization of the error function and the application of Householder’s
transformation make the calculation stable, fast and accurate. The detailed formulation
of the linearized least-squares procedure is described in Appendix A, or can be found in
refs. [21-26].

B. Relative-error evaluation

A relative type of error evaluation is selected, in order to improve the accuracy


of the cut-off band. This evaluation can easily be introduced into the least-squares
calculation with weighting values. Let N(z) and D(z) be the numerator and denominator
polynomials of a transfer function G(z) respectively. The error function for the k-th
datum (ωk,Gk) of the linearized LS method can be modified as

err' k = N ( z k ) − D( z k )Gk . (4.39)

For the relative-error evaluation, the following equation is used to determine the
weighting value of the k-th datum wk.

4-28
G( z k ) − Gk
= w k N ( z k ) − D( z k )G k (4.40)
Gk

From the above equation, the weighting value for the relative error evaluation is

w k = 1 / D( z k )G k , D( z ) = 1 + b1 z − 1 + L + bN z − N .
2
(4.41)

Mathematically, the above weighting function is correct. But, in practical


implementation, the weighting values, of course, must be calculated in advance of the
parameter identification, although the calculation of wk requires the denominator

parameters b1, … , bN which have not been calculated at this time yet. This problem can

be solved as follows. The parameters b1, … , bN for wk can roughly be evaluated without
using the relative-error evaluation, because a high accuracy for wk is not required.

Therefore, at first, the parameters b1, … , bN are approximately estimated without the
relative-error evaluation in order to calculate wk, and then an accurate solution of the
parameters a0, a1, ..., aN, b1, b2, ..., bN are calculated with the weighting values for the
relative-error evaluation.

C. Samples in the z-domain

Angular frequency ωk should be sampled logarithmically to cover a wide


frequency range with a small number of samples K. Because G(z) is a rational function
of z− 1, the following condition is always satisfied :

G ( z ) = G ( z ), (4.42)

where − denotes complex conjugate.

4-29
From eq. (4.36), frequency samples ωk (k = 1, 2,… , K) correspond to black dots on the
unit circle on the z-plane as illustrated in Fig. 4.11. Eq. (4.42) implies that if function
G(z) is determined on the upper-half plane, its values on the lower-half plane are
automatically determined, that is, we can determine the values only on the upper-half
plane not on the lower-half plane. This condition in turn determines the maximum
frequency in the following equation.

z K = e jω max ∆t ⇒ f max = 1 / 2∆t. (4.43)

The above equation corresponds to the well-known sampling theorem, and determines

the maximum frequency of the data set (ωk,Gk) (k = 1, 2,… , K). Therefore, when
calculating the frequency dependence of a transmission line, i.e. the frequency response of
H(ω) and Y0(ω), a frequency higher than fmax in eq. (4.43) should not be sampled.

sampling

f = fmax 0 f=0

unit circle
z - plane

Fig. 4.11 Samples on the z-plane

4-30
4.5.5 Stability check by Jury’s method

An ARMA model to be used in a transient calculation has to be stable. The


numerical stability of an ARMA model can be evaluated by determining its poles. If all
the poles are in the unit circle on the z-plane, the model is assured to be stable. The
determination of the poles results in that of all the solutions of D(z) = 0, where D(z) is the
denominator polynomial of the ARMA model. Thus, the straight-forward
implementation of the above pole determination is numerically ill-conditioned. E.I. Jury
derived a convenient scheme to determine the stability with a simple calculation [27].
The following table called “Jury’s table” illustrates the scheme :

line number
1 b0 b1 b2 … bN
2 bN bN− 1 bN− 2 … b1
3 c0 c1 c2 … cN− 1
4 cN− 1 cN− 2 cN− 3 … c0
5 d0 d1 d2 … dN− 2
6 dN− 2 dN− 3 dN− 4 … d0
:
:
2N − 3 q0 q1 q2

where ck = b0bk − bN− kbN , dk = c0ck− cN− k− 1cN− 1, …

Jury proved that the following conditions : D(1)> 0, (− 1)ND(− 1) > 0, b0 > |bN|, c0 > |cN|, d0

> |dN|, … , and q0 > |qN| assures that all the poles of the ARMA model are in the unit
circle, i.e. the model is numerically stable. The present line model uses this scheme to
assure that each IARMA element representing its frequency dependence is stable, because
the method only requires the above simple algebraic computations and is easy to
implement in computer code.

4-31
4.5.6 Order determination by AIC (Akaike’s Information Criterion)

In ref. [28], it is pointed out that the order of an ARMA model cannot be
determined without trial and error. Following this, the proposed order-determination
method basically increases the order one by one until a best fit is found. A simple error
evaluation uses standard deviation (SD) :

K
1

2
SD( N ) = G ( z k , N ) − Gk . (4.44)
K k =1

For a permitted constant error εA and maximum order Nmax, an order N meeting the
conditions SD(N) < εA and N ≤ N max is determined.

But it cannot be guaranteed that the above condition is met in the range of
N ≤ N max . In this case, Akaike's Information Criterion (AIC) [29] provides a criterion

for determining the order. The values provided in the data set (ωk, Gk) (k = 1, 2, … , K)
involve approximation and numerical-calculation errors. The approximation errors are
due to approximations and assumptions in the derivation process of series-impedance
formulas, and the numerical-calculation errors are due to numerical eigenvalue evaluation,
matrix inversion, and etc. When the errors are small, the SD will decrease
monotonically as the order increases, and achieve the condition SD(N) < εA at a
comparatively small order. When the errors are large, the model will approximate not
only the true electromagnetic characteristics but also the effect of the introduced error,
thus requiring too large an order. In this case, an order N which minimizes AIC(N) will
give an appropriate order by estimating the error characteristics involved in the data set.
AIC(N) is given in the following equation.

K 2
AIC( N ) = K ln ∑ G( z k , N ) − G k  + 4 N + 2 (4.45)
k =1 

Considering the above including the model stability mentioned in the previous

4-32
section, an order meeting the following conditions is selected.

1) A minimum order which meets the condition SD(N) < εA.

2) An order which minimizes AIC(N), when SD(N) < εA cannot be achieved.

3) In the above conditions, the resulting model has to be stable according to


Jury’s method.

4.5.7 Phase domain matrix-vector convolution

A transient calculation based on the present line model, eqs. (4.20a) to (4.20d),
requires matrix-vector convolutions in the following form.

y( t ) = g ( t )∗ x ( t ), (4.46)

where g(t) : transfer-function matrix such as hT(t) or y01(t), x(t) : input vector,
y(t) : output vector.

Each element of g(t) is replaced by an IARMA model using an optimization method


described in the preceding sections. Once the elements of G(z)=Z{g(t)} (Z{} denotes
Z-transform) are replaced by an ARMA model, the relation between the z-domain input
vector X(z) and the output vector Y(z) is expressed as Y(z) = G(z)X(z). Therefore, the i-th
element of the output vector Y(z) is described as :

n
Yi ( z ) = ∑ Gij ( z ) X j ( z ). (4.47)
j =1

Substituting eq. (4.25) into eq. (4.47), the inverse Z-transformation gives

4-33
n
y i ( t ) = ∑ { a ij 0 x j ( t )
j =1

+ a ij1 x j ( t − ∆t )+ L + a ij N ij x j ( t − N ij ∆t ) (4.48)
− b ij1uij ( t − ∆t )+ L + b ij N ij uij ( t − N ij ∆t ) },

where aijn, bijn : coefficients of (i, j) IARMA element, Nij : order of (i, j) IARMA
element, uij(n) = Z− 1{Gij(z)Xj(z)}.

4-34
4.6 Steady-State Initialization

In order to calculate a transient starting from a steady state, e.g. a fault


calculation, the nodal-admittance matrix of the whole circuit is formed first, and the
steady-state solution is calculated. Then, using the solution, the internal state of all the
models are initialized, and the transient calculation will start. Therefore, the admittance
matrix of the present line model has to be calculated to form the admittance matrix of the
whole circuit. According to the steady-state solution calculated by the host transient
calculation program such as the EMTP, the internal state of the line model has to be
initialized preceding the transient calculation.

EMTP
(host program)
Line Model

Formation of the Nordal-


Formation of the Admittance
Admittance Matrix of the
Matrix of the Line Model Y
Whole Circuit

Steady-State Calculation of the Steady-


Initialization State Solution

Initialization of the Internal


Initialization of the Internal
State of all the Models in
State of the Line Model
the Whole Circuit

Transient Calculation

Fig. 4.12 Steady-state initialization procedure

4-35
4.6.1 Frequency response of individual IARMA model

In the first stage of the steady-state initialization, the admittance matrix Y of the
present line model is required to be formed. For this, the response of each IARMA
element inside the line model at the steady-state frequency has to be calculated.
Because the operator z is closely related to the Laplace operator s = jω as in eq. (4.36),
the frequency response of an ARMA model (In this stage, IARMA is equivalent to
ARMA by replacing its own time step.) is easily obtained by substituting eq. (4.36) into
eq. (4.24) :

a 0 + a1 e − jω∆t + L + a N e − jωN∆t
G( z ) = , (4.49)
1 + b1 e − jω∆t + L + bN e − jωN∆t

Using the above response, Y is then formed to be passed to the host program as described
in Sec. 4.6.3.

4.6.2 Steady-state initialization of individual IARMA model

Once a steady-state solution is calculated by the host program including the


admittance matrix Y of the present line model, the internal state of each IARMA element
inside the line model is required to be initialized according to the steady-state solution.
Knowing the steady-state solution, i.e. the steady-state input x(t) and output y(t) to the
IARMA model, the internal state of the individual IARMA model ( x(0), x(− ∆t), ...,
x(− N∆t), y(− ∆t), ..., y(− N∆t) in its time-domain equation of eq. (4.25) ) can be initialized.

4-36
4.6.3 Admittance matrix of line model

The admittance matrix of a distributed-parameter line is given in the following


matrix form.

 Y coth(Γl ) − Y0 cosech(Γl )
Y = 0
Y0 coth(Γl ) 
, (4.50)
− Y0 cosech(Γl ) 

The above matrix-hyperbolic functions can be decomposed into matrix-exponential


functions as follows :

coth(Γl ) = [exp( Γl ) + exp( − Γl )]⋅[exp( Γl ) − exp( − Γl )]


−1

(4.51)
cosech(Γl ) = 2[exp( Γl ) − exp( − Γl )] .
−1

Furthermore, the above matrix-exponential functions can be obtained from the phase-
domain wave-deformation matrix H(ω) as :

exp( − Γl ) = exp( − jωτ) H (ω ), exp( Γl ) = {exp( − Γl )} .


−1
(4.52)

Each element of the phase-domain wave-deformation matrix H(ω) and of the


characteristic-admittance matrix Y0(ω) can be evaluated by the frequency response of the
corresponding ARMA model using eq. (4.49). Once H(ω) is known, exp(− Γl) and
exp(Γl) are calculated by eq. (4.52), and thus coth(Γl) and cosech(Γl) are also by (4.51).
Substituting coth(Γl), cosech(Γl), and Y0(ω) into (4.50), the admittance matrix Y is
finally obtained.

4-37
4.6.4 Initialization of line-model internal state

At the sending end, the time-domain convolution procedure of the phase-domain


wave-deformation matrix is performed in the form of eq. (4.18b). Transforming into the
frequency domain, we obtain :

H T (ω ) ⋅[e − jωτ
{Y0 (ω )V2 (ω ) + I 2 (ω )}]. (4.53)

The second term e− jωτ{Y0(ω)V2(ω) + I2(ω)} is the steady-state input to HT(ω) at f = ω/2π,
and it can be calculated by a steady-state solution of V2(ω) and I2(ω). The steady-state
output from each element of HT(ω) can be calculated by eq. (4.49). Because now we
have the input and output at the steady-state, the internal state of each ARMA element
can be initialized as described in Sec. 4.6.2. The same initialization procedure can be
applied to the wave-deformation matrix at the receiving end.

To initialize the internal state of the characteristic-admittance matrix Y0(ω) and


the past-history current source I'p1(ω) and I'p2(ω), the same procedure as for the wave-
deformation can be applied by evaluating the following values (see eq. (4.20) for
symbols) :

for Y0 (ω ) at both ends : Y0 (ω )V1 ( ω ), Y0 (ω )V2 (ω ) , (4.54)

for I ′
p1 and I ′
p2 : I ′
p1 ( ω ) = Y00V1 ( ω ) − I1 ( ω )
(4.55)
I′
p 2 ( ω ) = Y00V2 ( ω ) − I 2 ( ω ).

4-38
4.7 Matrix Stability Criteria

As described in Sec. 4.5.5, the stability of an ARMA model can be evaluated


using Jury's method. But, even if all the ARMA elements of both the wave-deformation
matrix H(ω) and the characteristic admittance matrix Y0(ω) are stable, it is still possible
that the line model behaves as an active circuit, and thus as a numerically unstable model.
In order to design a line model to be a passive circuit, the admittance matrix of the line
model Y = G + jB has to meet the criterion that G is positive definite, namely all the
eigenvalues are positive (criterion A) [30]. But it is convenient if each stability criterion
of both H(ω) and Y0(ω) is independently derived for the purpose of stabilization. For
Y0(ω), because it is just a multi-terminal admittance, the criterion A can be applied
(criterion B-1). Thus, the derivation of H(ω)'s criterion is necessary. Because H(ω)
= exp(jωτ)exp(− Γl) is a voltage matrix-transfer function with respect to traveling-wave
voltage Vf (not terminal voltage), output power Po should be smaller than input power Pi,
i.e. :

2 2
Pi − Po = V1 f − V2 f = V1 f *V1 f − V2 f *V2 f
= V1 f *V1 f − ( HV1 f )* ( HV1 f ) = V1 f *V1 f − V *1 f H * HV1 f (4.56)
= V1 f * (U − H * H )V1 f > 0

where * denotes complex-conjugate transposed matrix, U is the unit matrix.

To meet the above condition for arbitrary Vf, therefore, a real symmetrical matrix Q
defined by :

Q = Re{U − H * H } (4.57)

should be positive definite (criterion B-2). For now, the set of criteria B-1 and B-2 are

4-39
used only for stability checking, and thus the advantage compared to criterion A is that an
eigenvalue calculation of an n by n matrix is numerically easier than 2n by 2n. But it
should be noted that the set of criteria could be incorporated in the fitting procedure in
the future, because B-1 and B-2 are independent of each other.

4-40
4.8 Switching- and Fault-Surge Calculations

Using the present phase-domain IARMA line model, various switching-surge


and fault-surge calculations are carried out to confirm its accuracy. In this chapter,
corona discharge along conductors is not included for the comparison with rigorous
solutions. The rigorous solutions are obtained using a numerical inverse Laplace
transform method based on a frequency-transform theory [31,32], and thus frequency-
dependent effects are rigorously taken into account, although nonlinear effects such as
the corona may be difficult to be dealt with by the frequency-transform. In the
switching-surge calculations, the calculated results are compared with field-test results as
well as the rigorous solutions. In the fault-surge calculations, the simulations have to
start from a steady state, and thus the steady-state initialization method described in Sec.
4.6 is used to initialize the present line model. The calculated results are compared with
the rigorous solutions.

4.8.1 Switching-surge calculations

A. 500-kV single-circuit horizontal line

Fig. 4.13 shows a 500-kV single-circuit untransposed horizontal overhead line


consisting of 3 phase wires and 2 ground wires (Azumi Trunk of the Tokyo Electric
Power Co.). In this switching-surge study, because the voltages on the ground wires are
not interested, the impedance and admittance matrices are reduced from 5 by 5 to 3 by 3
assuming zero voltages. Thus, the line is treated as a three-phase line. Fig. 4.14 shows
the (1,1) element of the wave-deformation matrix H11(ω) fitted with an IARMA model
using the linearized LS method described in Sec. 4.5. The order of the IARMA model is
N = 11, and the fitted result is accurate in both amplitude and phase from a very low
frequency to a high frequency. Table 4.1 shows the coefficients of the model, and it is
observed that the modal traveling-time differences are taken into account by q1 = 6 and q2

4-41
= 74 as described in Sec. 4.5.3. The first aerial mode determines the minimum traveling
time τ, and the second aerial mode and the earth-return mode determine q1 and q2

respectively. The step response of H11(ω) reproduced by the fitted IARMA model is
shown in Fig. 4.15, and the result agrees well with the rigorous step response obtained by
the frequency-transform method. The order of the model is determined by the proposed
method described in Sec. 4.5.6, and the SD and AIC characteristics versus order N of the
fitting is shown in Fig. 4.16. In this case, both the SD and the AIC indicate that N = 11
is the appropriate order. The number of fitting is reduced considering the conductor
symmetry as described in Sec. 4.4, from 9 to 5 for the wave-deformation matrix, and from
9 to 4 for the characteristic-admittance matrix. Fig. 4.13 (b) illustrates an actual test
circuit where the receiving-end voltages were measured in case of switching as shown in
Fig. 4.17 (b) [33]. The switching transients reproduced by the proposed line model
together with the ATP version of EMTP are shown in Fig. 4.17 (a), and the calculated
result agrees well with the field-test result and with rigorous frequency-transform solution.
Table 4.2 compares the order of produced convolution elements between the proposed
linearized LS method and an asymptotic tracing method which is widely-used as a
standard method in the EMTP (J.Marti model) [10]. It is clear that the proposed fitting
method uses lower orders to reproduce given characteristics. To confirm the
computation efficiency, the execution time of both the fitting and transient calculation are
also compared in Table 4.3 with those of the J.Marti model. Both the simulation results
were obtained using the ATP. Because the proposed linearized LS method requires no
iterative calculation, the fitting procedure is faster than the asymptotic tracing method,
although the proposed method performs the full-matrix fitting of H(ω) and Y0(ω) in the
phase domain, but the symmetry of conductor configuration is considered. The
transient-calculation time of the proposed line model is longer by 25 % than the J.Marti
model, but it should be noted that the proposed line model takes into account the full
frequency dependence, while the J.Marti model ignores the frequency dependence of the
modal transformation matrices.

4-42
G.W. 22m
2
120mm ACSR t = 0 415Ω
83km
10m 14m
1pu
a b c
25m P.W.
2
240mm ACSR
(a) conductor 200Ωm
arrangement 0.4m (b) test circuit

Fig. 4.13 500-kV single-circuit untransposed horizontal overhead line


(Azumi trunk of the Tokyo Electric Power Co.)

Table 4.1 Coefficients of IARMA model (Hij(ω), order N = 11)

time step 2.51511E-07 minimum traveling time 2.77245E-04

a0 -2.312093441906782E-03 − −

a1 4.686240038630644E-02 b1 -4.747602145152051E+00

a2 -1.340901988860374E-01 b2 9.294536877750321E+00

a3 1.158411405126180E-01 b3 -1.050250461133035E+01

a6 -4.538081446028246E-02 b4 9.684832210928064E+00

a7 -1.403913056461285E-02 b5 -9.626600012165724E+00

a8 5.371659895911445E-02 b6 8.493078540627986E+00

a9 -2.059785586918279E-02 b7 -5.576776976971257E+00

a74 8.736459373947917E-06 b8 3.079230404395348E+00

a75 -2.723742568040881E-05 b9 -1.580636907846814E+00

a76 2.849171911159558E-05 b10 5.763042836445534E-01

a77 -1.003204476963157E-05 b11 -9.386165850094114E-02

4-43
1

0.5
amplitude

0.1

0.05
theoretical
fitted with ARMA

0.01 0
10 101 102 103 104 105 106
frequency [Hz]
(a) amplitude

90
[degree]

0
phase

-90
theoretical
fitted with ARMA

-180 0
10 101 102 103 104 105 106
frequency [Hz]
(b) phase

Fig. 4.14 Fitted results of wave deformation H11(ω) of


500-kV single-circuit untransposed horizontal overhead line

4-44
1

earth-return mode
h11(t)

0.5 IARMA
frequency transform

second aerial mode


first aerial mode
0

0 50 100 150
time µs]
[

Fig. 4.15 Step response reproduced by IARMA convolution h11(t) = F− 1{H11(ω)}

1.5 400

SD
300
AIC
1
200
AIC
SD

selected
100
0.5

0 -100
3 4 5 6 7 8 9 10 11 12 13
order

Fig. 4.16 SD and AIC versus model order

4-45
1

a
[p.u.]

0.5
voltage

0
b
proposed model
c
frequency transform
-0.5
260 280 300 320 340 360 380 400
time [µs]
(a) calculated results

(b) field-test results

Fig. 4.17 Switching transients on 500-kV single-circuit


untransposed horizontal overhead line

4-46
Table 4.2 Order comparison

Proposed Phase-Domain Model J.Marti Model (modal domain)

phase H(ω) Y0(ω) mode exp(− γ


kl) y0k(ω)

(1,1), (3,3) 11 5 1 12 22

(1,2), (3,2) 12 6 2 16 8

(1,3), (3,1) 12 8 3 21 27

(2,1), (2,3) 12 − − − −

(2,2) 10 5 − − −

Table 4.3 Computation time comparison

Proposed Phase-Domain Model J.Marti Model


(modal domain)
fitting 14.69 sec 20.71 sec
transient calculation 5.99 sec 4.51 sec

GATEWAY2000 P5-90 (Intel Pentium 90 MHz), under Microsoft Windows 95

4-47
B. 500-kV double-circuit vertical line

As shown in Fig. 4.3, the modal-transformation matrices of a vertical overhead


line are frequency-dependent. Fig. 4.18 (a) shows a 500-kV double-circuit
untransposed vertical overhead line consisting of 6 phase wires and 2 ground wires (New
Chichibu-Tochigi Trunk of the Tokyo Electric Power Co.). The impedance and
admittance matrix reduction is also applied to neglect the ground wires assuming zero
voltages, and thus the line is treated as a six-phase line. The number of fitting is reduced
from 36 to 18 for the wave-deformation matrix and from 36 to 15 for the characteristic-
admittance matrix by the symmetrical configuration. Fig. 4.18 (b) shows an actual test
circuit in which the sending- and receiving-end voltages were measured [34]. The
switching transients reproduced by the proposed line model agree well with the field-test
result and also with frequency-transform solution as shown in Fig.4.19.

22.4m
G.W.
92.1m AS 160
a 15.9m c' 304Ω t = 0
c 101.13km
78.6m P.W.
TACSR 810 × 6
b 16.7m b'
66.8m 1 pu
c 17.5m a'
55.0m
0.5m
300Ωm
(a) conductor
arrangement (b) test circuit

Fig. 4.18 500-kV double-circuit untransposed vertical overhead line

4-48
1
[p.u.]

0.5
voltage

proposed model, frequency transform


0
0 0.2 0.4 0.6 0.8 1
time [ms]

(a) sending-end voltage, phase c

Fig. 4.19 Switching transients on 500-kV double-circuit untransposed vertical


overhead line (continues to the next page)

4-49
0.4
[p.u.]

0.2
0
voltage

-0.2
proposed model, frequency transform
-0.4
0 0.2 0.4 0.6 0.8 1
time [ms]

(b) receiving-end voltage, phase a

Fig. 4.19 Switching transients on 500-kV double-circuit untransposed vertical


overhead line (continues to the next page)

4-50
0.6
[p.u.]

proposed model, frequency transform


0.4
0.2
voltage

0
-0.2
0 0.2 0.4 0.6 0.8 1
time [ms]

(c) receiving-end voltage, phase b

Fig. 4.19 Switching transients on 500-kV double-circuit untransposed vertical


overhead line (continues to the next page)

4-51
[p.u.]

0.5
voltage

proposed model
0 frequency transform
0 0.2 0.4 0.6 0.8 1
time [ms]

(d) receiving-end voltage, phase c

Fig. 4.19 Switching transients on 500-kV double-circuit untransposed vertical


overhead line (continues to the next page)

4-52
0.4
[p.u.]

proposed model, frequency transform


0.2
voltage

-0.2
0 0.2 0.4 0.6 0.8 1
time [ms]

(e) receiving-end voltage, phase c’

Fig. 4.19 Switching transients on 500-kV double-circuit untransposed vertical


overhead line (continues to the next page)

4-53
0.6
[p.u.]

proposed model, frequency transform


0.4
0.2
voltage

0 Receiving end, phase b’

-0.2
0 0.2 0.4 0.6 0.8 1
time [ms]

(f) receiving-end voltage, phase b’

Fig. 4.19 Switching transients on 500-kV double-circuit untransposed vertical


overhead line (continues to the next page)

4-54
0.6
[p.u.]

0.4 proposed model, frequency transform


0.2
voltage

0
-0.2
-0.4
0 0.2 0.4 0.6 0.8 1
time [ms]

(g) receiving-end voltage, phase a’

Fig. 4.19 Switching transients on 500-kV double-circuit untransposed vertical


overhead line (continued from the previous page)

4-55
C. 275-kV pipe-type oil-filled cable

Fig. 4.20 (a) shows a 275-kV pipe-type oil-filled cable (of the Chubu Electric
Power Co.). The dimensions and the physical constants of the cable are given in Table
4.4. Because of its thin sheaths, the frequency dependence of the cable is considerable.
As the sheaths and the enclosing pipe touch each other, the matrix reduction can be
applied, assuming that the voltages of the pipe and the sheaths are the same. Fig. 4.20
(b) shows an actual test circuit in which the sending- and receiving-end voltages were
measured [35]. The calculated result by the proposed line model agrees well with the
field-test result and with the rigorous frequency-transform solution as shown in Fig. 4.21.
In Fig. 4.21 (c), a difference is observed between the solutions by the proposed line model
and the frequency-transform method. This is because the eigenvalue calculation to
obtain the propagation constants is unstable when the above particular matrix reduction is
applied. Thus, the data sets to be fitted by an IARMA model include errors as shown in
Fig. 4.22, and this causes the error in Fig. 4.21 (c). In any case, the error is relatively
small considering the voltage scale of the figure.

0.12 µF 80 Ω 360 Ω
Rs
Equivalent
circuit of IG E 800 Ω

insulating tube
POF cable
pipe Rs 22 Ω
insulating paper pipe
oil
sheath
b core
IG

a c

(a) conductor arrangement (b) test circuit

Fig. 4.20 275-kV pipe-type oil-filled cable

4-56
0

phase a, c
[p.u.]

-0.5
voltage

phase b

-1
proposed model
frequency transform

-1.5
0 20 40 60 80 100
time [µs]

(a) sending-end voltages

Fig. 4.21 Switching transients on 275-kV pipe-type oil-filled cable


(continues to the next page)

4-57
0
[p.u.]

-0.5
voltage

-1
proposed model

frequency transform

-1.5
0 20 40 60 80 100
time [µs]

(b) receiving-end voltage, phase b

Fig. 4.21 Switching transients on 275-kV pipe-type oil-filled cable


(continues to the next page)

4-58
0.1
[p.u.]

0
voltage

proposed model
frequency transform

-0.1
0 20 40 60 80 100
time [µs]

(c) receiving-end voltages, phase a (=c)

Fig. 4.21 Switching transients on 275-kV pipe-type oil-filled cable


(continued from the previous page)

4-59
10-1
amplitude

theoretical
fitted with ARMA

10-2 3
10 104 105 106
frequency [Hz]
(a) amplitude

-140

theoretical
-150
fitted with ARMA
amplitude

-160

-170

-180 3
10 104 105 106
frequency [Hz]
(b) phase

Fig. 4.22 Fitted results of characteristic admittance Y041(ω) of


275-kV pipe-type oil-filled cable

4-60
Table 4.4 Dimensions and physical constants of 275-kV pipe-type oil-filled cable

core radius 31.10 mm core resistivity 0.17 µΩ m


insulator thickness 20.45 mm sheath resistivity 1.73 µΩ m
sheath thickness 0.15 mm pipe resistivity 0.1 µΩ m
pipe inner radius 150.85 mm earth resistivity 100 µΩ m
pipe outer radius 159.25 mm pipe relative 100
permeability
armor thickness 4.5 mm insulator relative 3.83
permittivity
cable depth 3.22 m oil relative 2.3
permittivity
cable length 3382 m armor relative 3.5
permeability

4-61
4.8.2 Fault-surge calculations

A. 500-kV double-circuit vertical line

The same 500-kV double-circuit line, as used in a switching-surge calculation in


Sec. 4.8.1-B (Fig. 4.18), is used to calculate a fault-surge calculation. The voltages at
the receiving end are calculated for a single-line-to-ground fault. At the sending end,
each phase conductor is connected to a double-circuit three-phase voltage source through
a 50 [mH] inductance. At the receiving end, phase conductors are open-circuited, and
at t = 10 [ms], phase a is closed to the ground through an 1 [Ω ] resistance which
represents a grounding impedance of the fault. The calculated result shown in Fig. 4.23
agrees well with a rigorous frequency-transform solution [18]. Because the frequency-
transform method is based on a numerical Laplace transform, a transient calculation
starting from a steady-state cannot be dealt with. Thus, the rigorous solutions presented
in this section are obtained by adding the corresponding transient solution to the steady-
state solution by use of the superposition law of the circuit theory. It should be noted
that the frequency-transform method is difficult to be used in a practical fault calculaion,
because nonlinear elements, such as circuit breakers, surge arresters, and etc., may
operate. In such cases, the proposed line model used in a time-domain EMTP-type
program is more advantageous than the frequency-transform method, because those
nonlinear elements can easily be modeled and used in the time-domain program.

4-62
2

b
1
voltage, p.u.

-1
c a
proposed model, frequency-transform
-2
0 10 20 30 40
time, milliseconds
(a) voltages of first (faulted) circuit at receiving end

b'
1
voltage, p.u.

-1
c' a'
proposed model, frequency-transform
-2
0 10 20 30 40
time, milliseconds
(b) voltages of second circuit at receiving end

Fig. 4.23 Fault transients on 500-kV double-circuit untransposed vertical


overhead line

4-63
B. 1100-kV double-circuit vertical line

Fig. 4.24 shows the conductor configuration of a 1100-kV (UHV : Ultra-High


Voltage) double-circuit untransposed vertical line, and a fault-surge calculation is carried
out under the same conditions as the preceding calculation. The calculated result shown
in Fig. 4.25 agrees well with rigorous frequency-transform solution and is almost the
same as the preceding result.

39 m
G.W. 2
109 m ACSR 810 mm
a 32 m c' ×1
0.5 m
92 m P.W.
b 33 m b' ACSR 810 mm2
72 m ×8
c 34 m a'
52 m length = 100 km, earth resistivity = 100 ohm m

Fig. 4.24 1100-kV double-circuit untransposed vertical line

4-64
1
voltage, p.u.

0 10 20 30 40

-1
c a

(a) voltages of first (faulted) circuit at receiving end

1
voltage, p.u.

b'

0 10 20 30 40

-1
c' a' time, milliseconds
proposed, frequency transform

(b) voltages of second circuit at receiving end

Fig. 4.25 Fault transients on 1000-kV double-circuit untransposed vertical


overhead line

4-65
C. Asymmetrical double-circuit line

The discussions that followed ref. [16] raised a question about the ability of the
present line model to simulate a line with strong asymmetry. In practice, this
configuration corresponds to a case that two or more different voltage lines share the
same right-of-way. Fig. 4.26 (a) shows the configuration of a double-circuit line, where
the left circuit is horizontal and the right one is vertical. To demonstrate the ability, the
voltages at the receiving end are calculated for a single-line-to-ground fault. The
calculation circuit is shown in Fig. 5.26 (b). The calculated result shown in Fig. 4.27 is
accurate, and it is clear that the present line model can be applied to asymmetrical
configurations.

G.W. 5m G.W.
22 m a'
IACSR 2 4 m IACSR 2
10 m 120 mm 8m 120 mm
a b c P.W.
8m b'
14 m 14 m TACSR 2
2 c' 240 mm
25 m P.W. TACSR 240 mm
×4
20 m
17 m
earth resistivity = 100 ohm m

(a) configuration

Fig. 5.26 Asymmetrical double-circuit line (continues to the next page)

4-66
50 mH × 3 length = 30 km a

1 p.u. ∼ b Rg

c = 1 ohm
25 mH × 3 a'

0.5 p.u. ∼ b'
∼ c'
(b) calculation circuit

Fig. 5.26 Asymmetrical double-circuit line (continued from the previous page)

1
voltage, p.u.

a b

0 10 20 30
c
-1
(a) voltages of first (faulted) circuit at receiving end

1
voltage, p.u.

a' b'

0 10 20 30
c' time, milliseconds
-1
proposed, frequency transform
(b) voltages of second circuit at receiving end

Fig. 4.27 Fault transients on asymmetrical double-circuit line

4-67
D. 400-kV bipolar HVDC line

Fig. 4.28 (a) shows a 400-kV bipolar HVDC transmission system, in which a
transmission line consisting of two phase wires and a ground wire is used. The line
length is 400 [km], and the values of the line-surge capacitors and the dc filters are given
in Fig. 4.28 (b). A fault-surge calculation of the HVDC system is carried out for a
single-line-to-ground fault. It is reported based on field tests that a mid-point fault gives
the most sever overvoltage [36]. Following this, the positive phase at the mid point is
short-circuited to the ground through a 5-Ω resistance representing a tower-footing
impedance. In order to start from a dc steady state, the steady-state initialization is
performed with frequency 0.001 [Hz] to approximate the dc steady state. Because it is
very difficult for the frequency-transform method to deal with the ac-dc converters at the
both ends of the line, the converters are replaced by a dc voltage source. Thus, the
operational characteristics of the converters are ignored, but this enables to compare the
results by the proposed model and by the frequency-transform method. The calculated
results are shown in Fig. 4.29, and the result by the proposed method agrees well with
that by the frequency-transform method. Because the negative spike voltage appeared
on the negative phase is caused by the aerial-mode traveling wave reflected by the
capacitance of the line-surge capacitors and the dc filters, the peak voltage is not affected
by the modeling of the converters. But it is estimated that the voltage waveform after
the peak may be different due to the operational characteristic of the converters, in reality.
The proposed line model can be used for transient calculations of an HVDC system with
detailed converter models in an EMTP-type program, for the proposed model is a time-
domain model.

4-68
MID+

DC DC
FILTER Cs Cs FILTER
FAULT

DC Cs Cs DC
FILTER FILTER
400km

earth resistivity = 170 ohm m MID-


200km

(a) calculation circuit

covnerter 500 mH line

2.5 µF
6.3 Ω
100 7 mH
280 mH

0.7 µF
0.7 µF

0.01 Ω

5 µF

(b) line-surge capacitor and dc filter

Fig. 4.28 400-kV bipolar HVDC transmission system

4-69
2
proposed model

1 frequency-transform
voltage, p.u.

(MID+)
0
(MID-)
-1

-2
0 2 4 6 8 10
time, milliseconds

Fig. 4.29 Fault transient on 400-kV bipolar HVDC transmission system

E. 275-kV pipe-type oil-filled cable

A fault-surge calculation is carried out using a 275-kV pipe-type oil-filled cable


shown in 4.20 (a), of which the frequency dependence is quite heavy. When the
penetration depth which varies with respect to frequency is smaller than the thickness of
the sheathes or of the pipe, shield effects are significant. When larger, earth-return
effects can be observed. Therefore, the modal-voltage distribution varies with respect
to frequency, and thus the frequency dependence of the modal-transformation matrices is
heavy. Fig 4.4 shows the frequency dependence of the first column of the modal-
transformation matrix of the cable. Because the modal-voltage distribution at the power
frequency is very different from that in a surge-frequency region, this calculation should

4-70
show the necessity of the frequency-dependent transformation matrices. Thus, the
proposed line model, which takes into account the complete frequency dependence, is
expected to show a better accuracy than a modal-domain model, which assumes constant
transformation matrices. Fig. 4.30 shows the fault calculation circuit, and the fault is
simulated by phase a short-circuit to the ground through an 1-Ω resistance at the
receiving end. The calculated result by the proposed line model shows an improved
accuracy compared to a modal-domain frequency-dependent line model (J.Marti model)
as shown in Fig. 4.31. It is clear that the errors of the proposed method are due to
fitting, but negligible. Using the modal-domain model, a different transition from the
spike voltage to the sinusoidal wave is observed due to the use of constant transformation
matrices (in this case, evaluated at f = 5 kHz). It should be noted that the frequency-
dependent transformation matrices affect the calculated result of the modal-domain model,
and that the proposed line model reproduces the frequency dependence accurately in the
time domain.

POF cable
a pipe
b
c
Rg
∼ ∼ ∼

Fig. 4.30 Fault calculation circuit on POF cable

4-71
2

1
a
voltage, p.u.

0
b

-1 c
proposed, frequency transform
-2
0 5 10 15
time, milliseconds
(a) proposed line model

1
a
voltage, p.u.

0
b

-1 c modal-domain model
frequency transform
-2
0 5 10 15
time, milliseconds
(a) proposed line model

Fig. 4.31 Fault transients on 275-kV pipe-type oil-filled cable

4-72
References of Chapter 4

[1] H.W. Dommel, “Digital Computer Solution of Electromagnetic Transients in Single- and Multi-
phase Networks,” IEEE Trans., Power Apparatus and Systems, Vol. PAS-88 (4), pp.388-398,
1969.
[2] L.M. Wedepohl, “Application of Matrix Methods to the Solution of Travelling-Wave Phenomena
in Polyphase Systems,” Proc. IEE, Vol. 110, No. 12, pp. 2200-2212, 1963.
[3] L.M. Wedepohl and S.E.T. Mohamed, “Multiconductor Transmission Lines : Theory of Natural
Modes and Fourier Integrals Applied to Transient Analysis,” Proc. IEE, Vol. 116, pp.1553-1563,
1969.
[4] W.S. Meyer and H.W. Dommel, “Numerical Modeling of Frequency-Dependent Transmission
Parameters in an Electromagnetic Transient Program,” IEEE Trans., Power Apparatus and
Systems, Vol. PAS-93, pp.1401-1409, 1974.
[5] A. Semlyen and A. Dabuleau, “Fast and Accurate Switching Transient Calculations on
Transmission Lines with Ground Return using Recursive Convolutions,” IEEE Trans., Power
Apparatus and Systems, Vol. PAS-94 (2), pp.561-571, 1975.
[6] A. Ametani, “A Highly Efficient Method for Calculating Transmission Line Transients,” IEEE
Trans., Power Apparatus and Systems, Vol. PAS-95 (5), pp.1545-1549, 1976.
[7] W.D. Humpage, K.P. Wong, and T.T. Nguyen, “Z-transform Electromagnetic Transient Analysis
in Power Systems,” IEE Proc., Vol. 127, Pt. C, No.6, pp.370-378, 1980.
[8] A. Semlyen, “Contributions to the Theory of Calculation of Electromagnetic Transients on
Transmission Line with Frequency Dependent Parameters,” IEEE Trans., Power Apparatus and
Systems, Vol. PAS-100 (2), pp.848-856, 1981.
[9] J.F. Hauer, “State-Space Modeling of Transmission Line Dynamics via Nonlinear Optimization,”
IEEE Trans., Power Apparatus and Systems, Vol. PAS-100 (12), pp.4918-4925, 1981.
[10] J.R. Marti, “Accurate Modelling of Frequency-Dependent Transmission Lines in Electromagnetic
Transient Simulations,” IEEE Trans., Power Apparatus and Systems, Vol. PAS-101 (1), pp.147-
155, 1982.
[11] A. Ametani, Distributed-Parameter Circuit Theory, Corona Pub. Co., Tokyo, 1992. (in Japanese:
雨谷,分布定数回路論,コロナ社,1992.) Especially for the frequency-dependent modal-
transformation matrices of cables: A. Ametani, “A Study of Cable Transient Calculations,” Sci. &
Eng. Rev., Doshisha University, Vol. 24, No. 2, pp. 110-127, 1983.
[12] A. Ametani, “Refraction Coefficient Method for Switching-Surge Calculations on Untransposed
Transmission Lines (Accurate and Approximate Inclusion of Frequency Dependency),” IEEE PES
Summer Meeting, C 73-444-7, 1973.
[13] L. Marti, “Simulation of Transients in Underground Cables with Frequency Dependent Modal
Transformation Matrices,” IEEE Trans., Power Delivery, vol. PWD-3 (3), pp.1099-1110, 1988.
[14] Tsu-huei Liu and Li Jin-gui, “Call for Help with Rational Function Approximations to
Frequency-Dependent Transformation Matrices of Cables and Lines,” EMTP News, Leuven
EMTP Center, March, 1988.
[15] L.M. Wedepohl, H.V. Nguyen, and G.D. Irwin, “Frequency-Dependent Transformation Matrices
for Untransposed Transmission Lines using Newton-Raphson Method,” IEEE PES Summer
Meeting, 95 SM 602-3 PWRS, 1995.

4-73
[16] T. Noda, N. Nagaoka, and A. Ametani, “Phase Domain Modeling of Frequency-Dependent
Transmission Lines by Means of an ARMA Model,” IEEE Trans., Power Delivery, Vol. 11, No. 1,
pp. 401-411, January 1996.
[17] T. Noda, N. Nagaoka, and A. Ametani, “Fault-Surge Calculations using the Phase-Domain
ARMA Line Model,” Trans. IEE of Japan, Vol. 116-B, No. 11, pp. 1409-1414, 1996. (in
Japanese: 野田,長岡,雨谷,「相領域 ARMA 線路モデルを用いた地絡サージ計算」,電
気学会論文誌 B, Vol. 116-B, No. 11, pp. 1409-1414, 1996.)
[18] T. Noda, N. Nagaoka, A. Ametani, “Further Improvements to a Phase-Domain ARMA Line
Model in Terms of Convolution, Steady-State Initialization, and Stability,” IEEE Power
Engineering Society Summer Meeting, Denver, Colorado, USA, 1996. (to be published in IEEE
Trans.)
[19] L. Dubé and I. Bonfanti, "MODELS: A new simulation tool in the EMTP", European
Transactions on Electrical Power Engineering (ETEP), Vol.2, no.1, pp.45-50, January/February
1992. For a Japanese reference: T. Noda and L. Dubé, “Simulation Model Description Language
MODELS, Tutorial Examples,” Japanese EMTP Committee, 1995. (in Japanese: 野田 琢, L.
Dubé, 「Simulation Model Description Language MODELS 例題集」,日本 EMTP 委員会,
1995.)
[20] P. Kuffel, K. Kent, and G. Irwin, “The Implementation and Effectiveness of Linear Interpolation
within Digital Simulation,” Proc. IPST '95, pp. 449-504, 1995.
[21] T. Noda and N. Nagaoka, “Development of ARMA Models for a Transient Calculation using
Linearized Least-Squares Method,” Trans. IEE of Japan, Vol. 114-B, No. 4, pp. 396-402, 1994.
(in Japanese: 野田,長岡,「線形化最小 2 乗法による ARMA 過渡現象計算モデルの開発」,
電気学会論文誌 B, Vol. 114-B, No. 4, pp. 396-402, 1994.)
[22] T. Noda and N. Nagaoka, “An Estimation Method of Linear Network's Transfer Function,”
Conference of Electric Power System Technology, IEE of Japan, PE-92-150, 1992. (in Japanese:
野田, 長岡, 「線形回路網伝達関数推定の一手法」,電気学会電力技術研究会資料,PE-92-150,
1992.)
[23] T. Noda and N. Nagaoka, “An Estimation Method of Linear Network's Transfer Function,” Proc.
Kansai-Branch Meeting, IEE of Japan, Paper No. G4-8, 1992. (in Japanese: 野田,長岡,「線形
回路網伝達関数推定の一手法」,平成 4 年電気関係学会関西支部連合大会,G4-8, 1992.)
[24] T. Noda and N. Nagaoka, “A Linearized Least-Squares Method for Estimating Linear Network's
Transfer Function - Relative Error Evaluation -,” Proc. Annual Meeting of IEE of Japan, Paper
No. 1397, 1993. (in Japanese: 野田,長岡,「線形回路網伝達関数推定に適した線形化最小 2
乗法―相対誤差評価法―」,平成 5 年電気学会全国大会,1397, 1993.)
[25] T. Noda, T. Sawada, K. Fujii, and N. Nagaoka, “ARMA Models for Transient Calculations and its
Identification - Identification of Coefficients and Orders -,” Proc. Annual Meeting of IEE of
Japan, Paper No. 1373, 1994. (in Japanese: 野田,澤田,藤井,長岡,「ARMA 過渡現象計算
モデルとその作成法」,平成 6 年電気学会全国大会,1373, 1994.)
[26] K. Yoshida, T. Noda, N. Nagaoka, and A. Ametani, “An Immitance Synthesis Method of a
Network using the Linearized Least-Squares Method,” Proc. Kansai-Branch Meeting, IEE of
Japan, Paper No. G5-11, 1996. (in Japanese: 吉田,野田,長岡,雨谷,「線形化最小 2 乗法
を用いた回路網イミタンス合成法」,平成 8 年電気関係学会関西支部連合大会, G5-11,
1996.)
[27] E.I. Jury, Theory and Application of the z-Transform Method, Wieley, New York, 1964.
[28] T. Yahagi, Theory of Digital Signal Processing, vol. 1-3, Corona Publishing, 1986.
[29] J.S. Lim and A.V. Oppenheim, Advanced Topics in Signal Processing, Prentice-Hall, 1988.
[30] L. Weinberg, Network Analysis and Synthesis, McGraw-Hill, 1962.

4-74
[31] A. Ametani, “The Application of the Fast Fourier Transform to Electrical Transient Phenomena,”
Int. J. Elect. Eng. Educ. Vol. 10 (4), pp. 277-286, 1972.
[32] N.Nagaoka and A.Ametani, "A development of a generalized frequency-domain transient
program− FTP," IEEE Trans., Power Delivery, Vol. PWRD-3 (4), pp.1996-2004, 1988.
[33] A. Ametani, T. Ono, and A. Honga, “Surge Propagation on Japanese 500kV Untransposed
Transmission Line,” Proc. IEE, Vol. 121, No.2, 1974.
[34] A. Ametani, E. Osaki, and Y. Honaga, “Surge Characteristics on an Untransposed Vertical Line,”
Trans. IEE Japan, Vol. 103-B, pp.117-124, 1983. (in Japanese: 雨谷,大崎,補永,「非撚架垂
直配列線路におけるサージ特性」,電気学会論文誌 B, Vol. 103-B, pp.117-124, 1983.)
[35] N. Nagaoka, M. Yamamoto, and A. Ametani, “Surge Propagation Characteristics of a POF
Cable,” Trans. IEE Japan, Vol. 105-B, pp.645-652, 1985. (in Japanese: 長岡,山本,雨谷,「POF
ケーブルにおけるサージ波形伝搬特性」,電気学会論文誌 B, Vol. 105-B, pp.645-652, 1985.)
[36] D.J. Melvold, P.C. Odam, and J.J. Vithayathil, “Transient overvoltages on an HVDC bipolar line
during monopolar line faults,” IEEE Trans., Power Apparatus and Systems, Vol. PAS-96 (2),
pp.591-601, 1977.

4-75
5 NONLINEAR SHUNT-ADMITTANCE MATRIX CONSIDERING
CORONA EFFECTS

Summary

The purpose of this chapter is to identify the accurate shunt-admittance matrix Y of a


transmission line including the nonlinearity due to corona discharge along transmission wires.
Compared with the skin effects for the series-impedance matrix, the mechanism of corona
discharge itself has not sufficiently been made clear. In this chapter, corona discharge is first
described, and then a novel q - v curve measurement method is presented to study the discharge
mechanism. Because the proposed method requires no charge-measuring conductor unlike
conventional methods, a q - v curve is measured in the real electric-field distribution of a
transmission-line configuration. Comprehensive q - v curves are measured and investigations
are made using the method. Coupling-factor curves to determine the increase of electrostatic
coupling between wires due to corona are also measured for mutiphase modeling. Then, a
wave-steepness-dependent corona model is proposed, for accurately reproducing q - v curves
(coupling-factor curves for mutual elements) not only for lightning-surge studies but also for
switching-surge studies. Simulations of q - v and coupling-factor curves are carried out, and
the validity is confirmed. The shunt-admittance matrix Y is given by those simulations.

5-1
5.1 Corona Discharge and Nonlinear Shunt-Admittance Matrix

Under a uniform electric field, faint currents are observed between a pair of
electrodes when applying a voltage below the sparking criterion. The currents are non
self-sustaining and called dark currents. When the condition of self-sustaining discharge
is met by raising the voltage, a complete breakdown is formed and spark discharge is
immediately observed. Under a nonuniform electric field, on the other hand, even if the
self-sustaining condition is met, spark discharge is not observed immediately, and partial
self-sustaining discharge is observed before sparking. That is, the transformation from
electron avalanches to a streamer does not necessarily lead to a complete breakdown.
The self-sustaining discharge before sparking is called corona discharge, and also called
partial discharge, because it is observed only in places where the electric field is
concentrated.

Corona discharge steadily sustains, when space charges generated by itself


reduces the maximum electric field strength. Therefore, the form of corona discharge
highly depends on electric field distribution, applied voltage waveform, and polarity.
Especially, remarkable polarity effects are observed, and positive-point corona is different
from negative-point corona in many points. The polarity effects may come from the
difference of the mobility of positive ions and electrons. The nonreversible discharge
form of corona, namely histeresis, is also an important characteristic, because the
histeresis causes corona losses and attenuation of traveling waves.

Because the radius of a transmission wire is small enough to make such a


nonuniform electric field, corona discharge is observed along the wire when a high
voltage is applied. At a power frequency, the corona discharge causes corona losses
and radio interference in the band of 0.5 ∼ 1 MHz. A number of efforts are made to
reduce the losses and interference, for example, by use of a bundle of conductors.
Those efforts are especially made in Japan, for Ultra-High Voltage (UHV) transmission

5-2
systems. The losses and interference are apparently disadvantageous at the power
frequency. But, under transient conditions, corona discharge causes attenuation of
traveling waves. Therefore, the inclusion of corona in a transmission-line model is
important to predict accurate maximum voltages for reducing insulation levels. And
simulations using the accurate line model minimizes the cost of construction, and thus a
cost-effective insulation design.

The transmission-line corona is normally characterized by a (charge - voltage) q


- v curve as illustrated in Fig. 5.1, because the generation of space charges with respect to
the applied voltage is well expressed by the curve [1]. Below the corona onset voltage
vCR, there are no space charges and the amount of total charges is given by q = C0v, where
C0 is the geometrical capacitance of the wire to the ground per unit length. Above vCR,
q is larger than C0v because of space charges generated by the corona discharge. When
v starts decreasing, the gradient of the curve is given by C0, and thus a histeresis
phenomenon is observed. The area enclosed by the curve is equal to corona losses. In
a multiphase case, electrostatic coupling between two wires is also increased by corona
due to the space-charge generation. The increase of the electrostatic coupling has been
represented by a coupling factor, which is defined as the ratio of the crest values of
applied and induced voltages when an impulse is applied [2]. In the present thesis, the
coupling factor is extended to a coupling-factor curve which is defined as the ratio of the
instant values of applied and induced voltages, that is, the curve of an instant coupling
factor. The coupling-factor curve achieves a better representation of the variation of the
instant coupling factor with respect to the applied voltage.

Consider an n-phase transmission line. Let q = (q1, q2, ..., qn)T and v = (v1,

v2, ..., vn)T be the charge and voltage vectors, where each element denotes the amount of
charges per unit length and the voltage on the corresponding wire respectively. In the
absence of corona, the relationship between q and v is expressed as :

q = C0v, (5.1)

5-3
charge: space-charge
q generation

geometrical
capacitance

corona onset voltage: v


voltage: vCR

Fig. 5.1 Typical q - v curve

where C0 is the geometrical capacitance matrix per unit length, obtained by inversion of
the potential matrix. The time derivative of eq. (5.1) gives :

∂q ∂v
= C0 . (5.2)
∂t ∂t

Ref. [3] gives a clear theory for generalizing eqs. (5.1) and (5.2) into the following
equations, in the presence of corona. The generalization of eq. (5.1) is :

q = Γv, (5.3)

where Γ represents an arithmetic relation between q and v. On the other hand, the
generalization of eq. (5.2) is :

5-4
∂q ∂v
=C , (5.4)
∂t ∂t

where C represents the incremental relation. Especially, in a single-phase case, Γ gives


the ratio of q/v, and C the gradient of the q - v curve ( C = ∂q / ∂v ) as illustrated in Fig.

5.2. This gives a clear distinction between the arithmetic capacitance Γ and the
incremental capacitance C.

charge: C
q
Γ

(v, q)

voltage: v

Fig. 5.2 Arithmetic capacitance Γ and incremental capacitance C

Because the time derivative of eq. (5.3) becomes :

∂q ∂v ∂Γ
=Γ + v, (5.5)
∂t ∂t ∂t

5-5
∂Γ
the additional term v complicates the practical implementation of the transient
∂t
calculation subroutine. Thus, eq. (5.4) is used for transient simulations in Chapter 6,
and the incremental capacitance matrix C is identified by a wave-steepness-dependent
corona model developed in Sec. 5.5. The shunt-admittance matrix Y is then obtained by
C. A further theoretical background regarding the identification of each element of the
incremental capacitance C is described in Sec. 5.6.2.

5-6
5.2 A Novel q - v Curve Measurement Method without Corona Cage
[4]

5.2.1 Conventional q - v curve measurement methods

In comparison with the theory of the frequency dependence of the series-


impedance matrix due to skin effects, the mechanism of corona discharge along a
transmission wire has not sufficiently been made clear. Therefore, experimental
investigations have to be accumulated, and especially an accurate q - v curve
measurement method is needed to be developed. To measure the amount of generated
space charges around a wire, conventional q - v curve measurement methods place
another conductor nearby the wire. Normally, a corona cage, which surrounds the wire,
is used as the space-charge measuring conductor as illustrated in Fig. 5.3 [1,5]. The
cage, which is made of a conductor, encloses the whole corona layer, and therefore the
amount of charges is obtained by q = Ccagev. Another method uses a conductor plate for

corona cage

specimen conductor

IG v
q Ccage

(IG: impulse generator) earth

Fig. 5.3 Corona cage for measuring space charges

5-7
the same purpose as also illustrated in Fig. 5.4 [1]. But in those methods, the presence
of a charge-measuring conductor disturbs the original electric-field distribution of a
transmission line, and could cause an error. A q - v curve should be measured under a
real transmission-line configuration.

specimen conductor

IG v conductor plate

(IG: impulse generator) earth

Fig. 5.4 Conductor plate for measuring space charges

For purposing more accurate q - v curve measurement, the author has developed
a method that requires no charge-measuring conductor [4]. In the method, in stead of
using the charge-measuring conductor, the injected-current waveform is recorded in
digital and then numerically integrated to obtain the amount of charges. Therefore, the
method measures a q - v curve under the identical electric-field distribution to a real
transmission-line configuration, because it does not use any additional conductor which
disturbs the field distribution. Using the new measurement method, field tests have been
carried out, and comprehensive q - v curves have been measured under the real field
distribution for the first time. Coupling-factor curves, which represent the increase of
electrostatic coupling between applied and induced conductors due to space-charge

5-8
generation, have also been measured in multiphase cases. Effects of conductor radius,
line length, and conductor separation are then investigated from the obtained data.

5.2.2 Experimental setup

The author has carried out two groups of field tests to validate the proposed
method and to accumulate field data, under support of Kansai Tech. Co. The first group
of field tests was performed on the flat ground as shown in Fig. 5.5, at the General
Education Center of Kansai Tech Co., Takatsuki-shi, Osaka-pref., Japan. The purpose
of this group is to validate the proposed method, and single-phase tests have been carried
out. Fig. 5.5 (a) illustrates the cross section of the experimental setup. An insulating
pipe is settled on a concrete pole, and a copper wire (OE60sq) is installed on the
insulating pipe. The OE60sq wire consists of 19 hard-drawn copper conductors, of
which the individual diameter is 2 [mm], and the total diameter is 10 [mm] as illustrated in
Fig. 5.5 (b). The OE60sq wire is normally used as a distribution wire with an insulating
cover, but the cover is removed to simulate a scaled transmission wire. The length of
the line is 44 [m], and it is supported by eight of the insulating pipes and concrete poles as
illustrated in Fig. 5.5 (c). The installation height is 2 [m], and the sag was measured to
be 25 [cm]. An impulse generator (IG) is connected to an end of the line to apply high-
voltage surges. The IG can generate 1 [MV] in case of no load and can storage 12 [kJ]
of energy at maximum, and its total capacitance is 0.0167 [µF]. A lead wire, that
connects the IG to the specimen conductor, consists of a bundle of four conductors to
avoid corona along the lead wire itself by increasing the equivalent radius. At the
sending end, the applied voltage v is measured using a resistance-type potential divider
(PD), and the injected current i is measured using a Rogowski-coil-type current
transformer (CT). The rise times of the PD and CT are 100 [ns] and 20 [ns] respectively,
and fast enough for this purpose. For high-voltage measuring, induced voltages to a
measuring wire is significant and causes an error in measured data. A coaxial cable may
be used as the measuring wire, because it has a shielding sheath. But the shielding effect
is not sufficient for this kind of MV-class test. To achieve more accurate measuring, an
optical system is used to transmit the output signals of the PD and CT to a digital-storage

5-9
oscilloscope as illustrated in Fig. 5.5 (d). The electric output signals of the PD and CT
are converted into optical signals using E/O (electric to optical) converters, and then
transmitted using optical-fiber cables to the oscilloscope. At the oscilloscope end, the
optical signals are restored into the original electric signals using O/E (optical to electric)
converters, and the oscilloscope finally records the original electric signals in digital.
The sampling frequency of the oscilloscope is 1 [GHz] and also fast enough for this
purpose. Photographs of the experimental setup are shown in Fig. 5.5 (e). The earth
resistivity was measured as shown in Table 5.1, and a fairly good resistivity was observed.

The second group of field tests was performed on a slope as illustrated in Fig.
5.6, at a property of Kansai Tech Co. near its iris park, in Yamasaki, Hyogo-pref., Japan.
The purpose of this group is to measure coupling-factor curves in multiphase cases as
well as q - v curves, and also to investigate the effects of conductor radius, line length,
and adjacent conductor on q - v curves and the effects of conductor separation and radius
on coupling-factor curves. Two wires are installed in parallel to measure coupling-
factor curves. Fig 5.6 (a) illustrates the cross section of the experimental setup, and a
thick polyvinyl-chloride pipe is used to support an insulating pipe. On the insulating
pipe, two specimen conductors are installed in parallel. The separation between the two
wires is varied to investigate the effects of conductor separation. This time, two types
of conductor are used also to investigate the effect of conductor radius : HDCC8sq and
OE60sq. The HDCC8sq wire consists of 7 hard-drawn copper conductors, of which the
individual diameter is 1.2 [mm] as illustrated in Fig. 5.6 (b). The OE60sq wire is the
same one used in the first group of field tests. Some tests were performed on the line
with length 111.3 [m], and others with length 22.1 [m], to investigate the effect of line
length. The side view is given in Fig. 5.6 (c). The same IG, PD, CT, lead wire,
optical-signal transmission system, and digital-storage oscilloscope as the first group are
used. Some photographs of the experimental setup are shown in Fig. 5.6 (d). The
earth resistivity was measured as shown in Table 5.2. The ground is electrically highly
lossy, because the slope is on a rocky mountain. The installed height and sag for both
the HDCC8sq and OE60sq wires are given in Table 5.3.

5-10
specimen
conductor concrete pole
insulating pipe
(plastic)

2m
2m

earth

(a) cross section

19 hard-drawn
copper conductors

10 mm

OE60sq 2 mm

(b) specimen conductor (OE60sq)

concrete pole specimen conductor


44 m
i lead wire
CT
2m 1.75 m PD IG
v

earth

(c) side view

Fig. 5.5 Experimental setup at the General Education Center of Kansai Tech Co.
(continues to the next page)

5-11
electrical optical electrical

E/O O/E
PD
converter converter digital-
optical-fiber cables storage
E/O oscilloscope
O/E
CT converter converter

(d) optical-signal transmission system of measured data

(e) photographs (continues to the next page)

Fig. 5.5 Experimental setup at the General Education Center of Kansai Tech Co.
(continues to the next page)

5-12
(e) photographs (continued from the previous page)

Fig. 5.5 Experimental setup at the General Education Center of Kansai Tech Co.

5-13
specimen
conductors

insulating
pipe (plastic) 2m

polyvinyl-
chloride pipe

earth

(a) cross section

19 hard-drawn
7 hard-drawn copper conductors
copper conductors
3.6 mm 10 mm

HDCC8sq 1.2 mm OE60sq 2 mm

(b) specimen conductors

polyvinyl-
chloride pipe

.....

specimen wire

CT i lead wire

PD IG
v
22.1
(c) side view 111.3mm earth

Fig. 5.6 Experimental setup at a property of Kansai Tech Co. in Yamasaki


(continues to the next page)

5-14
(d) photographs (continues to the next page)

Fig. 5.6 Experimental setup at a property of Kansai Tech Co. in Yamasaki


(continues to the next page)

5-15
(d) photographs (continued from the previous page)

Fig. 5.6 Experimental setup at a property of Kansai Tech Co. in Yamasaki

5-16
Table 5.1 Earth resistivity at the General Education Center of Kansai Tech Co.

depth [m] 0.5 1.0 2.0


resistivity [Ω m] 103 69.0 46.8

Table 5.2 Earth resistivity at a property of Kansai Tech Co. in Yamasaki

depth [m] 0.5 1.0 1.5 2.0 2.5


resistivity [Ω m] 898 741 503 462 436

Table 5.3 Installed height and sag of the second group of tests

conductor HDCC8sq OE60sq


phase applied induced applied induced
installed height
1.96 2.00 2.01 2.03
hT [m]

sag d [cm] 16 14 57 57

5-17
5.2.3 Postprocessing of measured data for generating q - v curves

A q - v curve is plotted considering the amount of total charges q, which is given


as the sum of conductor charges qc and space charges qs, as the y axis, and an applied
voltage v as the x axis. The voltage waveform can be measured directly, using a probe
and a storage oscilloscope (In the high voltage case, a potential divider may be used
additionally). But the amount of charges can not be measured directly, and thus an
intermediate quantity is to be measured to obtain the amount of charges. Then,
postprocessing of measured data is required. When a charge-measuring conductor such
as a corona cage or plate is used as described in 5.2.1, the voltage of the charge-
measuring conductor is recorded as the intermediate quantity, and the amount of charges
is calculated from the voltage. On the other hand, the proposed method records the
injected current as the intermediate quantity. Because the injected current is measured
using a CT at a point where the leading wire consists of a bundle of four conductors,
there is no corona around the CT and all the current flow is inside the conductors. Thus,
the amount of charges q(t) is obtained by integrating the injected current i(t) in the
leading wire with respect to time :

t
q( t ) = ∫0 i( τ)dτ. (5.6)

The above integration could be performed using a real circuit, namely an integration
circuit. But the injected-current waveform is recorded in digital and can numerically be
integrated easily. Furthermore, the recent rapid progress of digital-storage systems and
digital computers makes the numerical-integration method more advantageous, and the
numerical integration is not affected by induced noise which requires a great care of
high-voltage measurement. For computer implementation, the following trapezoidal
rule of integration is used :

∆t
q( t ) ≅ q( t − ∆t ) + {i(t − ∆t ) + i(t )}. (5.7)
2

5-18
Eq. (5.6) is derived assuming that an open-ended transmission line can be
represented by a pure capacitance C to the ground as illustrated in Fig. 5.7 (a). This
assumption is true, if a time region of our interest is longer than twice of the traveling
time of the line (round-trip time). The same is assumed also in any other conventional q
- v curve measurement methods. But this assumption is not true in a time region shorter
than the round-trip time, because an open-ended transmission line should be represented
by a pure resistance of which the value is given as the real part of the surge impedance Z0
as illustrated in Fig. 5.7 (b). The test-line length for measuring its q - v curves is chosen
to be short enough, in order to neglect a high-frequency oscillation caused by the multiple
reflections of traveling waves at the wave front of an injected-current waveform (further
investigation is given is Sec. 5.2.7). The oscillating frequency is easily given as 1/4τ,
where τ is the traveling time. The high-frequency oscillation vanishes immediately and
the injected-current waveform is then dominated by the equivalent capacitance C. For
we are interested not in such high-frequency traveling waves but in the generation of

IG

C Z0
IG IG

(a) t > 2τ (b) t < 2τ


(τ : traveling time)

Fig. 5.7 Equivalent circuits of an open-ended transmission line

5-19
space charges, we have to measure the injected current dominated by C not by Z0.
Therefore, the high-frequency oscillation is removed using a moving-average technique in
this thesis. The time window of the moving-average procedure should be selected as the
inverse of the oscillating frequency T = 4τ, and the procedure is given as :

1 t+ T /2
T ∫t − T / 2
iAVG (t ) = i( τ)dτ. (5.8)

For computer implementation, the above equation is discretized as :

k = N / 2− 1
1
iAVG ( n∆t ) =
N
∑ i(( n − k )∆t ), (5.9)
k =− N /2

where ∆t : time step and N = integer(T/∆t).

Because the injected-current waveform is recorded as digital data, the above procedure
can easily be implemented in a computer program together with the trapezoidal rule of
integration of eq. (5.7). A recorded injected-current waveform is first averaged using eq.
(5.9), and then the charge waveform is calculated using eq. (5.7) to plot a q - v curve.

5.2.4 Corona and leader development

As described in Sec. 5.1, corona discharge is also called partial discharge.


Under a nonuniform field, corona discharge is first observed in the place where the
electric-filed strength is maximum, and then, if the sparking criterion is met, a complete
breakdown is observed. The complete breakdown is usually called “flashover” in the
field of electric-power transmission. In the process of the flashover, electron avalanches
form a leader, which develops from an electrode to the other one, i.e. from a transmission
wire to the ground in a single-phase configuration. The leader development also causes
the increase of space charges as well as corona, because it consists of a plasma channel.
To measure a q - v curve due to corona, we need a proof that a measured injected current
is increased by corona not by leader development. Fig. 5.8 shows two cases of

5-20
500
applied voltage
400 200
injected current
voltage, kV

300

current, A
corona current
200 100

100

0 0

-100
0 5 10 15

time, microseconds

(a) with corona, without flashover

600 300
applied voltage
injected current
400 corona current 200
voltage, kV

current, A
200 100
leader current

0 0

0 5 10 15

time, microseconds

(b) with corona and flashover

Fig. 5.8 Applied-voltage and injected-current waveforms

5-21
measured applied-voltage and injected-current waveforms. Those waveforms are
recorded using an experimental setup illustrated in Fig. 5.6, and an OE60sq wire with
length 22.1 [m] is used as the specimen conductor. The charging voltage of the IG in
Fig. 5.8 (a) is 500 [kV], and the voltage in Fig. 5.8 (b) is 600 [kV]. In both the cases,
we observe a corona current at the wave front, because both the applied voltages exceed
160 [kV] that is the corona onset voltage of the specimen conductor OE60sq installed at
a height of 2 [m]. The comparison of Fig. 5.8 (a) and (b) shows a gradual increase of
the injected current after t ≅ 5 [µs] in (b) but not in (a). A longer time record of Fig. 5.8
(b) is shown in Fig. 5.9, and it shows that the increase of the injected current leads to a
flashover at t ≅ 20 [µs]. Therefore, it should be concluded that the increase of the
injected current after t ≅ 5 [µs] is due to leader development. Because a leader current
leads to a flashover, a clear difference between the currents due to corona and leader
development can be made. Thus, we can easily determine whether a measured q - v
curve is affected by a leader current or not, looking at the current waveform. And we
use q - v curves of which the injected-current waveform does not involve a leader current.

600 300

flashover applied voltage


400 injected current 200
voltage, kV

current, A

200 100

0 leader current 0
corona current
0 20 40 60 80

time, microseconds

Fig, 5.9 Applied-voltage and injected-current waveforms in case of a flashover

5-22
5.2.5 Measured q - v curves

This section validates the practical use of the proposed q - v curve measurement
method. Using an experimental setup illustrated in Fig. 5.5, single-phase tests have been
carried out. An OE60sq wire is used as illustrated in Fig. 5.5 (b). The line length is 44
[m], of which the traveling time is 0.147 [µs]. Applied-voltage and injected-current
waveforms are recorded in cases that the charging voltage of the IG is set to 100, 200,
300, 400, 500, 600, and 700 [kV], under a fair-weather condition. Corresponding
negative-polarity cases are also recorded under the same condition. The q - v curves of
±700 [kV] are not used, because a flashover was observed. Fig. 5.10 shows the
applied-voltage and injected-current waveforms of the positive-polarity cases, and Fig.
5.11 shows those of the negative-polarity cases. The wave-front time of the applied
voltages is approximately 3 [µs]. Using the postprocessing procedure described in Sec.
5.2.3, the corresponding q - v curves are produced and shown in Fig. 5.12 (a) and (b) in
the positive and negative cases respectively.

In the q - v curves, when the voltage is lower than vCR = 160 [kV] that is the
corona-onset voltage of the OE60sq wire installed at a height of 2 [m], the q - v curves
follow q = C0v, where C0 is the geometrical capacitance (the linear relation of q = C0v is
plotted as a broken line in the figures). This is clearly confirmed by Fig. 5.13 which
shows the q - v curves of the ±100 [kV] cases, where corona is not observed. The value
of the geometrical capacitance C0 is given by :

2πε0 2
C0 = ≅ 8.43 [pF / m ], he = hT − d, (5.10)
 2he  3
ln  
 r 

where he : mean height considering the catenary curve of wire, hT : installed


height (= 2 [m]), d : sag (= 25 [cm]), and r : wire radius (= 5 [mm])

The above value agrees well with the measured one as shown in Fig. 5.13. A very small

5-23
600

500

400
voltage, kV

300 A

200

100
B
0
0 2 4 6
time, µs

(a) applied-voltage waveforms

200
current, A

100

0 2 4 6
time, µs

(b) injected-current waveforms

Fig. 5.10 Applied-voltage and injected-current waveforms


(OE60sq, line length = 44 [m], fair weather, positive)

5-24
-600

-500

-400
voltage, kV

-300

-200

-100

0
0 2 4 6
time, µs

(a) applied-voltage waveforms

-200
current, A

-100

0 2 4 6
time, µs

(b) injected-current waveforms

Fig. 5.11 Applied-voltage and injected-current waveforms


(OE60sq, line length = 44 [m], fair weather, negative)

5-25
12

10 (a) positive

8
charge, µC

2
geometrical capacitance
0
0 200 400 600
voltage, kV

-12

-10 (b) negative

-8
charge, µC

-6

-4

-2
geometrical capacitance
0
0 -200 -400 -600
voltage, kV

Fig. 5.12 Measured q - v curves (OE60sq, line length = 44 [m], fair weather)

5-26
2

1
charge, µC

-1

geometrical capacitance
-2
-200 -100 0 100 200
voltage, kV

Fig. 5.13 Measured q - v curves in the absence of corona


(OE60sq, line length = 44 [m], fair weather)

difference is observed between the measured and calculated geometrical capacitance.


This is estimated due to the presence of supporting concrete poles which are grounded
through a low resistance, and also of the response-time difference of the PD and CT. In
any case, the difference is negligible, and it should be noted that the slope of the q - v
curves in the absence of corona agrees well with the calculated value of C0. Thus, the
proposed method is said to measure a q - v curve under the real electric-field distribution
of a transmission-line configuration. On the other hand, in the conventional methods
mentioned in Sec. 5.2.1, the slope of a q - v curve does not agree with the calculated
value because of the presence of a charge-measuring conductor such as a corona cage or
a plate which disturbs the electric-field distribution. Therefore, the agreement of C0 is
one of the advantages of the proposed method, and the present q - v curves are measured

5-27
in a real transmission-line configuration for the first time.

When the voltage exceeds the corona-onset voltage vCR, the amount of charges
is increased by corona, and the slope becomes greater than C0. The amount of
generated space charges for the positive-polarity is much larger than that for the negative,
because the slope in the positive cases is larger than the negative cases. The mechanism
of the space-charge generation is going to be discussed and simulated by a wave-
steepness dependent corona model in Sec. 5.5. It is experimentally well-known that
corona immediately disappears when the applied voltage starts decreasing. This can be
confirmed by the q - v curves in Fig. 5.12, where the slope of the q - v curves becomes
equal to the geometrical capacitance C0 after the applied voltage peak.

A steeper-front impulse gives a higher corona-onset voltage, as indicated in ref.


[5]. That is, varying the peak value of applied voltages with keeping the wave-front
time constant, the corona-onset voltage becomes higher for a higher-peak impulse. This
is difficult to be confirmed from Fig. 5.12, but can be confirmed from Fig. 5.10 (a), where
the corona-onset voltage in the 600 [kV] case (indicated by A) is higher than that in the
200 [kV] case (indicated by B). This looks like that corona-onset voltage depends on
the wave steepness of an applied voltage. In actuality, the formation of initial electrons
of a corona onset takes a time less than 1 [µs], that is, there is the inception time delay
between times when an applied voltage exceeds the true corona-onset voltage and when
streamer development begins [6]. Therefore, the variation of corona-onset voltages
observed from Fig. 5.10 (a) is due to the inception time delay, which makes an onset
voltage higher because of the increase of an applied voltage during the delay. It is also
observed that each curve has its own path on the q - v plane during the space-charge
generation especially just before its own peak voltage. Ref. [6] explains that there are
three reasons why each curve can have its own path: (a) inception time delay, (b) process
of streamer development, and (c) relaxation time of ionization. The inception time delay
(a) is explained above, but it is not significantly observed in the q - v curves. As for (b),
just after the inception, streamers start developing outward from the surface of a
conductor until it reaches the corona radius. If the velocity of the streamer development

5-28
is independent of the wave steepness of an applied voltage, this can make difference of q -
v curves for different wave steepness, just after an inception. The region swept by the
streamers is called a corona region, and space charges are generated with a relaxation
time (c) inside the region. Thus, (c) makes difference of q - v curves just before a peak
voltage. In Fig. 5.12, we observe the effects of the relaxation time of ionization (c).
These effects (a), (b), and (c) vary with respect to wave steepness, and thus are called the
“wave-steepness-dependence” of a q - v curve, in this thesis. And the wave-steepness-
dependent effects are modeled and discussed in Sec. 5.5.

In case of a flashover, a q - v curve becomes as shown in Fig. 5.14. The


amount of space charges increases rapidly due to leader development primary and also
due to the formation of a complete-breakdown path finally. As mentioned in Sec. 5.2.4,
it is clearly observed that the rapid increase of space charges is not due to corona, and this
sort of data cannot be accepted as a corona q - v curve.

As a conclusion of this section, the following points are confirmed :

1) The practical use of the proposed q - v curve measurement method, that requires no
charge-measuring conductor, has been validated. q - v curves are measured under
the real electric-field distribution of a transmission-line configuration for the first
time.

2) Below the corona-onset voltage of a specimen conductor, the slope of the q - v


curve is equal to the geometrical capacitance.

3) Above the corona-onset voltage, the slope becomes larger than the geometrical
capacitance.

4) The amount of generated space charges in a positive-polarity case is much larger


than that in a negative case.

5) When an applied voltage starts decreasing, corona immediately disappears.

6) A steeper-front impulse gives a higher corona-onset voltage because of the inception

5-29
time delay of corona. Each curve has its own path on the q - v plane during the
space-charge generation due to the relaxation time of ionization. Those effects are
called the wave-steepness-dependence of a q - v curve.

100

80
charge, µC

60

complete breakdown
40

leader development
20

0
0 200 400 600 800
voltage, kV

Fig. 5.14 q - v curves in presence of a flashover


(OE60sq, line length = 22.1 [m], fair weather)

5.2.6 Effects of conductor radius

To observe the effects of conductor radius, q - v curves of a HDCC8sq wire


have been measured using an experimental setup illustrated in Fig. 5.6. The radius of
the HDCC8sq wire is 1.8 [mm], which is thinner than the OE60sq wire used in the

5-30
previous section. Therefore, we now obtain q - v curves of a thinner conductor, and are
able to compare results for different radii. The line length is 22.1 [m], of which the
traveling time is 0.074 [µs] (the effects of line length is going to be discussed in the next
section). In the same manner as described in the previous section, applied-voltage and
injected-current waveforms are recorded for eight IG charging voltages (100, 200, ...,
800 [kV]), and corresponding negative-polarity cases are also recorded under a fair-
weather condition. Because the corona-onset voltage of the HDCC8sq wire installed at
a height of 2 [m] is about 62 [kV] which is much smaller than that of the OE60sq wire,
the lead wire is connected to the middle of the discharge resistance of the IG to reduce
applied voltages as illustrated in Fig. 5.15. The discharge resistance consists of 24
elements (each element is 150 [Ω ]), and the total resistance is 3.6 [kΩ ]. Connecting the
lead wire to the middle of the discharge resistance is equivalent to dividing the charging
voltage using two 1.8 [kΩ ] resistances. Thus, a voltage slightly smaller than the half the
charging voltage is applied to the lead wire, considering the impedance of the HDCC8sq
wire seen from the lead wire.

IG

Lf
3.6kΩ
discharge resistance

to HDCC8sq
bank
capacitor

lead wire

Fig. 5.15 Reduction of applied voltages

5-31
Fig. 5.16 and 5.17 show the applied-voltage and injected-current waveforms in
both the positive and negative polarity cases respectively. As for the positive cases, the
wave-front time of the applied voltages varies approximately between 2.3 and 3.0 [µs].
On the other hand, the wave-front time in the negative cases varies between 2.0 and 3.5
[µs]. The peak voltage in the positive cases are lower than that in the negative cases.
This means that the input impedance of the wire in the positive cases is smaller than that
in the negative cases, in other words, the increase of the wire capacitance due to corona
in the positive cases is lager than that in the negative cases. Thus, the generation of
space charges due to corona increases the capacitance of the wire to the ground and
suppresses the peak voltage, and this effect is more noticeable in the positive-polarity
cases. Compared with the waveforms of the OE60sq wire cases shown in Figs. 5.10 and
5.11, the suppression of the peak voltages is more noticeable in the HDCC8sq cases.

When the charging voltage of the IG is ±100 [kV], corona is not observed
because the applied voltage is about ±35 [kV] which is smaller than the corona-onset
voltage. Thus, the q - v curves in the ±100 [kV] cases are first shown in Fig. 5.18 to
confirm that the measured geometrical capacitance agrees well with calculated one given
by eq. (5.10) (C0 = 7.29 [pF/m], hT = 1.96 [m], d = 16 [cm], r = 1.8 [mm]). As shown in
the case of the OE60sq wire, the measured capacitance is slightly larger than the
calculated value, but the difference is negligible.

The q - v curves for the other charging voltages, in the presence of corona, are
plotted in Fig. 5.19. No flashover is observed for all the IG charging voltages, and all
the curves are plotted in the figure. Compared with the q - v curves of the HDCC8sq
(Fig. 5.19) and OE60sq (Fig. 5.12) wires, namely compared with the q - v curves for
different radii, several differences are observed. The corona-onset voltage of the
HDCC8sq wire is obviously lower than that of the OE60sq wire, i.e. the onset voltage of
a thinner conductor is lower. This agrees with an experimental formula of the corona-
onset voltage established by Peek, and the experimental formula is widely accepted (for
example, see page 89 of ref. [7]). Later, Hartmann theoretically derived the Peek

5-32
400

300
voltage, kV

200

100

0
0 2 4 6
time, µs

(a) applied-voltage waveforms

80

60
current, A

40

20

-20
0 2 4 6
time, µs

(b) injected-current waveforms

Fig. 5.16 Applied-voltage and injected-current waveforms


(HDCC8sq, line length = 22.1 [m], fair weather, positive)

5-33
-400

-300
voltage, kV

-200

-100

0
0 2 4 6
time, µs

(a) applied-voltage waveforms

-80

-60
current, A

-40

-20

20
0 2 4 6
time, µs

(b) injected-current waveforms

Fig. 5.17 Applied-voltage and injected-current waveforms


(HDCC8sq, line length = 22.1 [m], fair weather, negative)

5-34
0.4

0.2
charge, µC

-0.2

-0.4
geometrical capacitance
-40 -20 0 20 40
voltage, kV

Fig. 5.18 Measured q - v curves in the absence of corona


(HDCC8sq, line length = 22.1 [m], fair weather)

formula [8], but this is not well-known unfortunately. The Hartmann formula is used in
the wave-steepness-dependent corona model proposed in Sec. 5.5. As for polarity
effects, it is obvious that the amount of generated space charges is larger in the positive
polarity cases than in the negative cases.

The shape of the curves is found very different. As for the curves of the
OE60sq cases, after exceeding the onset voltage, the amount of charges increases first
gradually and then acceleratedly, as voltage increases. On the other hand, the amount of
charges of the HDCC8sq cases rapidly increases just after exceeding the onset voltage,
and then it increases in the same way as the OE60sq cases. All the q - v tracings of the

5-35
6

5 (a) positive

4
charge, µC

1
geometrical capacitance
0
0 100 200 300
voltage, kV

-6

-5 (b) negative

-4
charge, µC

-3

-2

-1
geometrical capacitance
0
0 -100 -200 -300
voltage, kV

Fig. 5.19 Measured q - v curves (HDCC8sq, line length = 22.1 [m], fair weather)

5-36
HDCC8sq cases are identical before the applied voltage begins decreasing, although each
tracing of the OE60sq cases has its own path. It should be noted that no significant
wave-steepness-dependent effects are observed in the HDCC8sq (thinner conductor) case,
in comparison with the OE60sq case. This can be explained by the process of streamer
development that broadens the corona region as follows. The velocity of streamers is
experimentally observed to be 1.0 [m/µs] and almost constant (for example, see page 74
of ref. [9]). If the radius of a conductor is small, then the corona radius, at which the
streamers stop, is also small, and the streamers reach the corona radius immediately after
a corona onset. On the other hand, if the radius of a conductor is large, then the corona
radius is large, and it takes time for the streamers to reach the corona radius. If the time
constant of the streamer development to reach the corona radius is of the order of 0.1 ∼
1.0 [µs] and if the wave-front time of an applied voltage is some [µs], q - v curves for
different wave steepness are observed to have different q - v paths because of the
streamer-development time, that is, wave-steepness-dependent effects. Therefore, no
significant wave-steepness-dependent effects are observed in case of a thin conductor, of
which the radius is small enough compared with the speed of the streamers.
Furthermore, a difference is observed after the applied voltage starts decreasing. Fig.
5.20 shows longer tracings of Fig. 5.21. In the longer tracing case, the sampling time
step of the digital oscilloscope has to be greater due to the limitation of its storage
memory. Thus, the time resolution of the q - v curves shown in Fig. 5.20 becomes
worse than that in Fig. 19. The accuracy of the wave-front part (rising part) of the q - v
curves is insufficient, but the sampling time step is set to a value which is small enough to
obtain a good accuracy for the wave-tail part (falling part). Therefore, we can at least
rely upon the slope of the curves after the applied voltage starts decreasing (in the wave-
tail part), although we cannot expect a good accuracy for the wave-front part of the q - v
curves and also for the peak value of the amount of charges (integrated value). As
mentioned in the previous section, the slope of a q - v curve in the OE60sq cases
immediately becomes equal to the geometrical capacitance in the wave-tail part. On the
other hand, from Fig. 5.20, the slope of a q - v curve of the HDCC8sq wire cases
approaches the value of the geometrical capacitance gradually, and it slightly takes time.

5-37
6

5 (a) positive

4
charge, µC

1
geometrical capacitance
0
0 100 200 300
voltage, kV
-6

-5 (b) negative

-4
charge, µC

-3

-2

-1
geometrical capacitance
0
0 -100 -200 -300
voltage, kV

Fig. 5.20 Measured long-tail q - v curves


(HDCC8sq, line length = 22.1 [m], fair weather)

5-38
As a conclusion of this section, we have confirmed the following effects of
conductor radius using the new q - v curve measurement method :

1) The corona-onset voltage for a thinner conductor is lower, as widely-known.

2) The shape of q - v curves of an HDCC8q (thin) wire is different from an OE60sq


(thick) wire.

- In the wave-front part of an applied voltage after the onset, the amount of charges
increases first acceleratedly and then gradually, as the applied voltage increases.
All the q - v tracings are identical before the applied voltage begins decreasing.

- In the wave-tail part, the slope of q - v curves approaches the value of the
geometrical capacitance gradually.

3) All the q - v tracings in the HDCC8sq cases are identical before the applied voltage
begins decreasing, although each tracing of the OE60sq cases has its own path.
Thus, significant wave-steepness-dependent effects are not observed in the
HDCC8sq cases, compared with the OE60sq cases. This can be explained by the
process of streamer development, and no significant wave-steepness-dependent
effects are observed in case of a thin conductor, of which the radius is small enough
compared with the speed of the streamers.

5.2.7 Effects of line length

Theoretically, the length of a test line for measuring its q - v curve can be
arbitrary. However, in practice, we should consider the following two points for
obtaining an accurate q - v curve : 1) the length should be long to obtain reproducible
corona discharge and to accurately measure the capacitance value; but 2) the length
should be short to obtain a good time resolution which is affected by the moving-average
calculation of the postpocessing procedure. Those two points are contrary to each
other, and thus we have to make a compromise. As for the point 1), discharge

5-39
phenomena are often governed by probability effects due to initial electrons, and the
conductor-surface condition of a line may be uneven with respect to the longitudinal
direction. Also, the geometrical capacitance of a line is quite small compared with the
stray capacitances of a measurement system. A longer line averages the uneven
discharge and gives a larger capacitance value, and thus we obtain a reproducible q - v
curve and a reliable capacitance value. On the other hand, as for the point 2), the
postprocessing procedure to produce the q - v curve described in Sec. 5.2.3 includes a
moving-average calculation to remove high-frequency oscillations due to traveling waves
reflecting at both ends. The moving-average filtering smoothes the curve, but fast
changes are lost. Because the time constant of the moving-average calculation, which
corresponds to the period of the oscillations, is determined as :

4l
T = 4τ = , (5.11)
c0

where τ: traveling time, l: line length, and c0: speed of light.

A shorter line gives a smaller time constant. Thus, the shorter line gives a good time
resolution, and fast changes such as wave-steepness effects are accurately observed.

Fig. 5.21 shows q - v curves of an HDCC8sq wire with length 111.3 [m], which
are measured to investigate the effect of the line length using an experimental setup
illustrated in Fig. 5.6. The q - v curves are compared with those of Fig. 5.19
corresponding to the case with line length 22.1 [m]. In Fig. 5.21, the curves are smooth
and the geometrical capacitance agrees well with the calculated value plotted with a
broken line. This is because the length 111.3 [m] is long enough to average the corona
discharge and to measure the capacitance value accurately. But we cannot observe the
rapid increase of space charges just after exceeding the corona-onset voltage, which is
observed in Fig. 5.19. The time constant of 111.3 [m] is 1.48 [µs] which is too large to
observe such the rapid increase, and it is removed by the moving-average calculation.
On the other hand, that of 22.1 [m] is 0.3 [ms], thus the rapid increase is clearly observed.

5-40
6

5 (a) positive

4
charge, µC

1
geometrical capacitance
0
0 100 200 300
voltage, kV

-6

-5 (b) negative

-4
charge, µC

-3

-2

-1
geometrical capacitance
0
0 -100 -200 -300
voltage, kV

Fig. 5.21 Measured q - v curves (HDCC8sq, line length = 111.3 [m], fair weather)

5-41
Therefore, the length of a test line can be determined by the time constant of
interest T as :

c0 T
l= . (5.12)
4

5.2.8 Effects of adjacent conductor

Conventional q - v curve measurement methods mentioned in Sec. 5.2.1 use a


charge-measuring conductor such as a corona cage or a conductor plate placing nearby a
specimen conductor. The charge-measuring conductor may disturb the original
electric-field distribution of a real transmission-line configuration. Thus, to determine
the influence of the charge-measuring conductor to the accuracy, q - v curves have been
measured under a condition that another HDCC8sq wire simulating the charge-measuring
conductor is placed nearby a specimen conductor, using an experimental setup illustrated
in Fig. 5.6. The specimen and charge-measuring conductors are installed as shown in
Fig. 5.6 (a), and the separation is 0.6 [m]. At the sending end, a resistance-type voltage
divider is connected to the charge-measuring conductor to measure the voltage, but the
voltage information is actually not used, and thus the charge-measuring conductor should
be called an adjacent conductor. The charging voltage of the IG is set to 200, 400, 600,
and 800 [kV] and also to corresponding negative values. Fig. 5.22 shows measured q -
v curves. In the figure, curves measured in the presence of the adjacent conductor is
plotted using solid lines, and curves in the absence of the adjacent conductor using broken
lines. The maximum amount of generated space charges are different, and the errors are
6.4 % for positive polarity and 3.5 % for negative. The slope of the rising part is
different in the positive cases, and also the slope when an applied voltage begins
decreasing is also different in both the polarity cases. Those error may be caused by the

5-42
6

5 (a) positive
with adjacent conductor
4
charge, µC
without adjacent conductor

0
0 100 200 300
voltage, kV

-6

-5 (b) negative
with adjacent conductor
-4
charge, µC

without adjacent conductor

-3

-2

-1

0
0 -100 -200 -300
voltage, kV

Fig. 5.22 Effects of adjacent conductor on q - v curves


(HDCC8sq, line length = 22.1 [m], fair weather)

5-43
presence of the adjacent conductor.

It has been demonstrated that the presence of an adjacent conductor affects the
accuracy of a q - v curve, and that the proposed method should have a high accuracy
compared with the conventional methods.

5-44
5.3 Coupling-Factor Curve Measurement

5.3.1 Measured coupling-factor curves for multiphase cases

A transmission line normally consists of multiphase conductors: some sets of


three phase wires and one or two ground wires. Corona discharge along a wire
increases the amount of space charges around it, and the electrostatic coupling to another
wire is increased by the space charges. For lightning-surge studies, this corresponds to a
very practical case that a phase wire voltage is increased by a lightning surge propagating
along a ground wire above the corona onset voltage, and thus the electrostatic coupling
between the ground and phase wires in the presence of corona has to be predicted
precisely. Therefore, the electrostatic coupling between two wires in the presence of
corona is necessary to be investigated. A coupling factor, which is defined as the ratio
of the crest applied and induced phase voltages, has been used to characterize the
increase of electrostatic coupling due to corona [2]. The following formula gives the
geometrical coupling factor (no corona) in the configuration of Fig. 5.23 [10].

b
log
K= a , r : radius of conductor #1 (5.13)
2h
log
r

The conventional coupling factor does not represent the time variation of the
electrostatic coupling. To observe the time variation which may be wave-steepness-
dependent, the coupling factor is extended to a coupling-factor curve which plots with
the applied voltage as the x axis and with the instant coupling factor as the y axis.
Therefore, the coupling-factor curve K = K(v1) is expressed as:

5-45
conductor #1 conductor #2

a
h
b

earth
h

image

Fig. 5.23 Configuration showing dimensions to calculate coupling factor

v2 (t )
K ( v1 ) = K ( v1 ( t )) = , (5.14)
v1 ( t )

where v1(t), v2(t): instant values of applied and induced phase voltages.

Coupling-factor curves between two HDCC8sq wires have been measured using
an experimental setup in Fig. 5.6. The line configuration is illustrated in Fig. 5.6 (a), and
the installation height and sag of the wires are given in Table 5.3. The conductor
separation is 1.2 [m]. The charging voltage of the IG is set to 100, 200, 400, 600, 800
[kV] and also to corresponding negative values except -100 [kV]. Because corona is
not observed in case of -100 [kV], the waveforms should be identical to the positive 100
[kV] case except the polarity, and thus the case has been omitted. A lead wire to apply
the charging voltage is connected to the middle of the discharge resistance of the IG to
reduce the applied voltage as shown in Fig. 5.15. The voltage waveforms of both the
applied and induced phases are recorded at the sending end using resistance-type
potential dividers (PDs). The resistance values of the PDs are 15 [kΩ ] for the applied
phase and 20 [kΩ ] for the induced phase. It should be noted that the resistance value of

5-46
the PD used for the induced phase is not large enough to avoid a leakage current through
the PD. Although a high-resistance potentiometer or a capacity divider may be
appropriate as used in ref. [2] to avoid the leakage current, one of those high-impedance
voltage dividers which can measure up to 1 [MV] could not be prepared according to
financial reasons. Thus, the equivalent circuit of the measurement system becomes as
shown in Fig. 5.24 (a), and the resistance of the PD Rdiv has to be considered. Assuming
that the IG is a voltage source of a step function with peak value V1, then an equivalent
circuit becomes Fig. 5.24 (b). The time response of the PD is to be evaluated. In the
Laplace s-domain, the voltage of conductor #2 is given as :

1
sC2 + 1 / Rdiv V1 Cm Rdiv 1
V2 ( s ) = = V, (5.15)
1
+
1 s τ l s+
1 1
sCm sC2 + 1 / Rdiv τl

where τl = CmRdiv(1 + C2 / Cm) : the time constant of the measurement system.

The inverse Laplace transform of the above equation gives the time response of the
measurement system :

Cm Rdiv  t
v 2 ( t ) = V1 exp− , (5.16)
τl  τl 

and we obtain

Cm
lim v 2 (t ) = V1 . (5.17)
t→ 0 Cm + C2

The above limiting value indicates that the voltage of conductor #2 at time zero is given
as the pure capacitive voltage division of Cm and C2, and that the presence of Rdiv does
not affect the measured value at time zero. But eq. (5.16) indicates that the induced
charges on conductor #2 discharge through Rdiv as a leakage current, and the measured

5-47
conductor #1 conductor #2

Cm
IG C1 C2 Rdiv

earth
(a)

conductor #1 conductor #2

IG
Cm
V1 C2 Rdiv

earth
(b)

Fig. 5.24 Equivalent circuits of measurement system

Cm/(C2+Cm)
voltage v2(t)

0
τl 2τl 3τl

time

Fig. 5.25 Time response of measurement system

5-48
value finally approaches zero as shown in Fig. 5.25. Substituting the values of Cm, C2,
and Rdiv, we obtain the time constant of the measurement system :

 C 
τl = Cm Rdiv 1 + 2  ≅ 51
. [µs], (5.18)
 Cm 

where Cm = 94.6 [pF], C2 = 160.4 [pF], and Rdiv = 20 [kΩ ].

Thus, the induced voltage v2(t) is decreased to 80 % within 1.0 [µs]. If Cm were
constant, i.e. no corona, then the influence of Rdiv could be removed by the deconvolution
of eq. (5.16) using such as an FFT (Fast Fourier Transform) method (for example, see
page 538-540 of ref. [11]). In reality, Cm increases as space charges are generated − this

is what we are measuring − , and thus we cannot remove the influence of Rdiv. Therefore,
we have to remind that coupling-factor curves shown as follows are affected by the
resistance of the PD connected to the induced phase. But those data can still be used for
the validation of a simulation model such as one proposed in Sec. 5.5, if a simulation is
performed considering Rdiv. The data also give us the time variation of an instant
coupling factor, which cannot be obtained from conventional coupling-factor data defined
as the ratio of crest values. Furthermore, if a simulation model is validated with the
following data, then we can reproduce the true coupling-factor curves using the model.
In this section, we attempt only qualitative analysis of the data, and a further investigation
is provided in Sec. 5.6.2 with the reproduced true coupling-factor curves by the proposed
model.

Fig. 5.26 shows applied and induced phase waveformes in case of separation =
1.2 [m]. The conventional coupling factors K = v2CREST /v1CREST calculated from the
waveforms clearly increases as the applied voltage increases due to the generation of
space charges. It is also observed that the peak value of the applied voltage is
suppressed by the increase of the capacitance when space charges are largely generated.
Fig. 5.27 shows the coupling-factor curves produced from the data of Fig. 5.26 based on

5-49
eq. (5.14). The coupling factor calculated by eq. (5.13) is plotted using a broken line in
the figure. Around the very beginning of the wave front, a coupling factor calculated by
eq. (5.14) is inaccurate, because both the applied and induced voltages are almost zero.
Thus, we start from a certain point where the instant coupling factor becomes stable.
The coupling-factor curve in the positive 100 [kV] case, where no corona is observed,
begins around the geometrical coupling-factor value and monotonously decreases as the
applied voltage increases. (It should be reminded that if there were no leakage current,
the curve should follow the geometrical coupling-factor value.) This agrees with eqs.
(5.16) and (5.17), and the decrease of the instant coupling factor is due to a leakage
current through the induced-phase PD. Therefore, when the gradient of a coupling-
factor curve is greater than that in the positive 100 [kV] case indicated by a dotted line in
Fig. 5.27, the instant coupling factor is increased by generated space charges due to
corona. It is observed that a higher applied voltage generates more space charges,
because a higher voltage generates more space charges. In the cases above positive 400
[kV], the increase of coupling factor due to the generation of space charges is more
noticeable than the decrease due to a leakage current through the induced-phase PD, and
the gradient of the curves even becomes positive. Also, positive polarity corona
generates more space charges than negative corona. These characteristics of the space-
charge generation are in accordance with the characteristics obtained from q - v curves in
Sec. 5.2.5. When the applied voltage begins decreasing at the wave tail, the wave
steepness of the voltage becomes much slower than the time constant of the measurement
system. Thus, the leakage current becomes dominant, and the coupling-factor curves
approach zero.

The following conclusions are obtained based on the observations in this


section :

1) The use of a coupling-factor curve defined by the ratio of instant applied and induced
voltages enables to observe the time variation of electrostatic coupling between
wires. Field data for this have been obtained.

2) A leakage current through an induced-phase PD affects the coupling-factor curves.

5-50
But the data obtained with the leakage current can still be used for the validation of a
simulation model.

3) A higher applied voltage generates more space charges, and positive-polarity corona
is more noticeable than negative one.

40
positive, 100 kV, sep = 1.2 m applied
induced
30
voltage, kV

20

10

0
0 2 4 6
time, µs

100 -100
positive, 200 kV, sep = 1.2 m applied negative, 200 kV, sep = 1.2 m applied
80 induced -80 induced
voltage, kV

voltage, kV

60 -60

40 -40

20 -20

0 0
0 2 4 6 0 2 4 6
time, µs time, µs

Fig. 5.26 Applied and induced voltages (conductor separation = 1.2 [m], HDCC8sq,
line length = 22.1 [m], fair weather) (continues to the next page)

5-51
200 -200
positive, 400 kV, sep = 1.2 m applied negative, 400 kV, sep = 1.2 m applied
induced induced
voltage, kV

voltage, kV
100 -100

0 0
0 2 4 6 8 10 0 2 4 6 8 10
time, µs time, µs

300 -300
positive, 600 kV, sep = 1.2 m applied negative, 600 kV, sep = 1.2 m applied
induced induced

200 -200
voltage, kV

voltage, kV

100 -100

0 0
0 2 4 6 0 2 4 6
time, µs time, µs

400 -400
positive, 800 kV, sep = 1.2 m applied negative, 800 kV, sep = 1.2 m applied
induced induced
300 -300
voltage, kV

voltage, kV

200 -200

100 -100

0 0
0 2 4 6 0 2 4 6
time, µs time, µs

Fig. 5.26 Applied and induced voltages (conductor separation = 1.2 [m], HDCC8sq,
line length = 22.1 [m], fair weather) (continued from the previous page)

5-52
1

(a) positive
0.8
coupling factor separation = 1.2 m
0.6
geometrical coupling factor

0.4

0.2

0
0 100 200 300
applied voltage, kV

(b) negative
0.8
separation = 1.2 m
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 -100 -200 -300
applied voltage, kV

Fig. 5.27 Measured coupling-factor curves (conductor separation = 1.2 [m], HDCC8sq,
line length = 22.1 [m], fair weather)

5-53
5.3.2 Effects of conductor separation and radius on coupling-factor curves

To observe the effects of conductor separation on a coupling-factor curve, field


tests have been carried out with varying the conductor separation, using the same
experimental setup used in the previous section. The separation of two HDCC8sq wires
is modified to 0.9 and 0.6 [m], and the test conditions except the conductor separation
are not modified. Fig. 5.28 shows the waveforms of applied and induced phase voltages,
in case that the conductor separation is 0.9 [m], and Fig. 5.29 shows produced coupling-
factor curves. Also, Figs. 5.30 and 5.31 show the applied and induced voltages and the
produced coupling-factor curves respectively, in case that the conductor separation is 0.6
[m]. In comparison with the coupling-factor curves including the 1.2 [m] case in Fig.
5.27, the generation of space charges increases as the conductor separation becomes
shorter. This is because a shorter separation gives a more concentrated electric field at
the surface of the conductors. In all the cases, positive corona generates more space
charges than negative one as a polarity effect. Especially, in the 0.6 [m] case, the instant
coupling factor of the positive 800 [kV] case is increased to more than 0.3 by corona.
Fig. 5.32 shows coupling-factor curves for OE60sq wires with the separation 1.2 [m].
Compared with Fig. 5.27, we can extract the effects of conductor radius. It should be
noted that the decrease of the instant coupling factor due to a leakage current through the
induced-phase PD is much smaller in Fig. 5.32. This is because Cm and C2 of the
OE60sq wires are larger, and the time constant of the measurement system becomes
longer. Therefore, all the curves first follow the geometrical coupling factor. If space
charges are generated, the instant coupling factor becomes larger than the geometrical
value, although slight decrease due to the leakage current is observed for each curve. At
the wave tail, the instant coupling factor approaches zero, because the wave steepness at
the wave tail of the applied voltages is far longer than the time constant of the
measurement system.

The characteristics observed from all those curves agree well with our common
knowledge of corona discharge, and the data could be used for the validation of a corona
model.

5-54
40
positive, 100 kV, sep = 0.9 m applied
induced
30
voltage, kV

20

10

0
0 2 4 6
time, µs

100 -100
positive, 200 kV, sep = 0.9 m applied negative, 200 kV, sep = 0.9 m applied
80 induced -80 induced
voltage, kV

voltage, kV

60 -60

40 -40

20 -20

0 0
0 2 4 6 0 2 4 6
time, µs time, µs

200 -200
positive, 400 kV, sep = 0.9 m applied negative, 400 kV, sep = 0.9 m applied
induced induced
voltage, kV

voltage, kV

100 -100

0 0
0 2 4 6 8 10 0 2 4 6 8 10
time, µs time, µs

Fig. 5.28 Applied and induced voltages (conductor separation = 0.9 [m], HDCC8sq,
line length = 22.1 [m], fair weather) (continues to the next page)

5-55
300 -300
positive, 600 kV, sep = 0.9 m applied negative, 600 kV, sep = 0.9 m applied
induced induced

200 -200
voltage, kV

voltage, kV
100 -100

0 0
0 2 4 6 0 2 4 6
time, µs time, µs

400 -400
positive, 800 kV, sep = 0.9 m applied negative, 800 kV, sep = 0.9 m applied
induced induced
300 -300
voltage, kV

voltage, kV

200 -200

100 -100

0 0
0 2 4 6 0 2 4 6
time, µs time, µs

Fig. 5.28 Applied and induced voltages (conductor separation = 0.9 [m], HDCC8sq,
line length = 22.1 [m], fair weather)

5-56
1

(a) positive
0.8
coupling factor separation = 0.9 [m]

0.6
geometrical coupling factor

0.4

0.2

0
0 100 200 300
applied voltage, kV

(a) negative
0.8
separation = 0.9 [m]
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 -100 -200 -300
applied voltage, kV

Fig. 5.29 Measured coupling-factor curves (conductor separation = 0.9 [m], HDCC8sq,
line length = 22.1 [m], fair weather)

5-57
40
positive, 100 kV, sep = 0.6 m applied
induced
30
voltage, kV

20

10

0
0 2 4 6
time, µs

100 -100
positive, 200 kV, sep = 0.6 m applied negative, 200 kV, sep = 0.6 m applied
80 induced -80 induced
voltage, kV

voltage, kV

60 -60

40 -40

20 -20

0 0
0 2 4 6 0 2 4 6
time, µs time, µs

200 -200
positive, 400 kV, sep = 0.6 m applied negative, 400 kV, sep = 0.6 m applied
induced induced
voltage, kV

voltage, kV

100 -100

0 0
0 2 4 6 8 10 0 2 4 6 8 10
time, µs time, µs

Fig. 5.30 Applied and induced voltages (conductor separation = 0.6 [m], HDCC8sq,
line length = 22.1 [m], fair weather) (continues to the next page)

5-58
300 -300
positive, 600 kV, sep = 0.6 m applied negative, 600 kV, sep = 0.6 m applied
induced induced

200 -200
voltage, kV

voltage, kV
100 -100

0 0
0 2 4 6 0 2 4 6
time, µs time, µs

400 -400
positive, 800 kV, sep = 0.6 m applied negative, 800 kV, sep = 0.6 m applied
induced induced
300 -300
voltage, kV

voltage, kV

200 -200

100 -100

0 0
0 2 4 6 0 2 4 6
time, µs time, µs

Fig. 5.30 Applied and induced voltages (conductor separation = 0.6 [m], HDCC8sq,
line length = 22.1 [m], fair weather)

5-59
1

(a) positive
0.8
coupling factor separation = 0.6 m

0.6
geometrical coupling factor

0.4

0.2

0
0 100 200 300
applied voltage, kV

(b) negative
0.8
separation = 0.6 m
coupling factor

0.6

0.4

0.2

0
0 -100 -200 -300
applied voltage, kV

Fig. 5.31 Measured coupling-factor curves (conductor separation = 0.6 [m], HDCC8sq,
line length = 22.1 [m], fair weather)

5-60
1

(a) positive
0.8
separation = 1.2 m
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 100 200 300 400 500
applied voltage, kV

(b) negative
0.8
separation = 1.2 m
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 -100 -200 -300 -400 -500
applied voltage, kV

Fig. 5.32 Measured coupling-factor curves (conductor separation = 1.2 [m], OE60sq,
line length = 22.1 [m], fair weather)

5-61
5-62
5.4 Historical Review of Corona Models

This section gives a historical review of corona models to make clear the
development strategy of a wave-steepness-dependent corona model which is proposed in
the following sections.

In the early days of transmission-line corona studies, corona models were


proposed based on one of the following theories :

1) Increase of Capacitance: The generation of space charges due to corona increases


the equivalent capacitance of a conductor to the ground. Thus, corona discharge
can be simulated using a nonlinear capacitance of which the value is given as the
slope of a q - v curve obtained by a field test.

2) Peek’s quadratic law: Peek’s quadratic law, which is determined by experiments


under AC voltages, can be applied to surge corona. Then, the equation of Peek’s
quadratic law is solved simultaneously with the wave equations of a transmission line
using numerical methods.

Theory 1) was first proposed by C.F. Wagner and B.L. Lloyd [1]. In the
reference, they introduced the concept of q - v curve for the first time and performed a
number of q - v curve measurement using a corona cage or a conductor plate. They also
simulated wave propagation characteristics including corona using a digital computer,
where corona discharge was represented as nonlinear capacitances of which the value was
given as the slope of a measured q - v curve as illustrated in Fig. 5.33 (a). A piece-wise
linear approximation was used. In the simulations, a transmission line was represented
as pi sections, and the shunt capacitance of each pi section was replaced with the
nonlinear capacitance as in Fig. 5.33 (b). From the simulation result, they explained that
transmission-line corona discharge caused wave delay above the onset voltage at the
wave front of a surge. Fig. 5.34 shows a schematic explanation of the wave delay.

5-62
The following formula which gives the velocity of a traveling wave approximately
explains the wave delay, because if C is increased by corona then the velocity c becomes
small above a corona-onset voltage.

1
c= , ( L, C : inductance and capacitance per - unit - length) (5.19)
LC

measure
dapproximated

v
(a) q - v curve representation (b) simulation circuit

Fig. 5.33 Transmission-line corona simulation by Wagner and Lloyd

∆T : wave delay

vCR
applied

Fig. 5.34 Wave delay due to corona pointed out by Wagner and Lloyd

5-63
Theory 2) was first proposed by J. Umoto and T. Hara, and they derived a pair
of nonlinear wave equations including corona as follows [12]. The wave equations of a
transmission line in the absence of corona are expressed as :

∂v ∂i
− =L
∂x ∂t
(5.20)
∂i ∂v
− =C ,
∂x ∂t

where v, i : voltage and current on conductor at distance x from the sending end,
L and C : per-unit-length inductance and capacitance.

Peek’s quadratic law is now introduced as the second term of the above second equation
as :

∂v ∂i
− =L
∂x ∂t
(5.21)
∂i ∂v ( v − v CR )2
− =C + K ,
∂x ∂t v

and K is given as :

K = σC r / 2h ×10 − 11 ( f + 25), (5.22)

where ? C : corona-loss constant, r, h : radius and height of conductor, and f : frequency.


Although eq. (5.21) is a set of time-domain partial-differential equations, eq. (5.22) is
expressed in the frequency domain. Using the following approximation, eq. (5.22) is
transformed into the time-domain :


f + 25 ≅ f ≅ . (5.23)
∂t

Thus, eq. (5.21) becomes

5-64
∂v ∂i
− =L
∂x ∂t
(5.24)
∂i ∂v σC r / 2 h ×10 − 11 ∂
− =C + ( v − v CR )2 .
∂x ∂t v ∂t

Umoto and Hara further introduced the third term to obtain a better agreement with field
tests. The final form of the wave equations including corona discharge is given by :

∂v ∂i
− =L
∂x ∂t
(5.25)
∂i ∂v σC r / 2 h ×10 − 11 ∂ ( v − v CR )2
− =C + ( v − v CR ) + K
2
.
∂x ∂t v ∂t v

Physically, the second term of the above second equation corresponds to a nonlinear
capacitance, and the third term a nonlinear resistance. Umoto and Hara proposed three
numerical methods to solve the above nonlinear partial-differential equations, and wave-
propagation simulations were carried out in ref. [12]. They also determined the corona-
loss constant ? C by comparing the calculated results with field tests.

Based on one of the above two theories, the following corona models have been
proposed.

•Piece-Wise Linear Approximation Model by H.M. Kudyan and C.H. Shih [13]

This model is based on theory 1), and a piece-wise linear approximation of a q -


v curve is essentially the same as one proposed in ref. [1] and also illustrated in Fig. 5.33
(a). Fig. 5.35 shows the equivalent circuit for the practical implementation of the model
in an EMTP-type program. The main feature of the model is that the equivalent circuit
automatically takes into account the dynamics of corona using diodes. In the figure, C0
represents the geometrical capacitance, and Cp and Cn represent the increase of the slope
of a q - v curve for positive and negative polarities respectively. DC voltage sources Ep
and En are set to positive and negative corona-onset voltages. Thus, if a positive-

5-65
polarity surge exceeding the corona-onset voltage Ep is applied, then diode Dp turns on
and Cp is connected to C0 in parallel to increase the total capacitance. In case of a
negative-polarity surge, Cn is connected to C0 in parallel in the same manner. One point
that is contrary to theory 1) is that Rp and Rn represent corona losses for positive and
negative corona respectively. Nowadays, it has been made clear that corona losses are
caused by the histeresis of a q - v curve, and the resistance representation of the losses
could be doubtful.

Dp Dn

C0 Cp Cn

Rp Rn
Ep En

Fig. 5.35 Equivalent circuit of Kudyan and Shih’s corona model

•Mathematical Model by M.M. Suliciu and I. Suliciu [14]

M.M. Suliciu and I. Suliciu created a rate-type constitutive equation which


produces a family of q -v curves. The model can reproduce the wave-steepness
dependence of corona. But the model is too mathematical to obtain the constants of the
equation from physical data. This model could be classified into the group of theory 1).

5-66
•Trial-and-Error Model by A. Inoue [15]

This model is classified into the group of theory 1), because Inoue proposed the
following equations to express a q - v curve.

F (v ) = C0 (0 ≤v ≤vCR and ∂v / ∂t ≤0 ) (5.26a)

F (v ) = C0 + m1 k1 (v − v CR ) m1 − 1 / v (vCR ≤v ≤vx and ∂v / ∂t ≥ 0 ) (5.26b)

F (v ) = C0 + m2 k 2 (v − v CR ) m2 − 1 / v (vx ≤v ≤vM and ∂v / ∂t ≥ 0 ), (5.26c)

where C0 : geometrical capacitance, and vCR : corona-onset voltage.

He carried out a number of field tests regarding the propagation characteristics of surge
corona, and parameters vx, m1, k1, m2, k2 were determined by a trial-and-error procedure
by comparing calculated and field-test waveforms. Transient calculations were
performed by representing a transmission line with lumped T sections. Because a
number of T sections are cascaded, the simulation circuit becomes essentially identical to
the pi-section representation by Wagner and Lloyd shown in Fig. 5.33 (b). Although the
model shows a good accuracy, field tests are always required to determine the parameters
vx, m1, k1, m2, k2.

•Nonlinear Circuit Model by K.C. Lee [16]

This model is a nonlinear-circuit representation of theory 2). To efficiently


implement theory 2) in an EMTP-type program, Lee proposed a nonlinear circuit
illustrated in Fig.5.36. The values of conductance and capacitance in the figure were
determined by the following equations :

5-67
2
− 11  v CR 
G = σG r / 2h ×10 1 −  (5.27a)
 v 

 v 
C = 2σC r / 2h ×10 − 11 1 − CR  (5.27b)
 v 

Transients were calculated using the compensation method [17] with the trapezoidal rule
of integration.

C G

Fig. 5.36 Nonlinear representation of corona by Lee

•Linear Circuit Model by H. Motoyama and A. Ametani [18]

In the same manner as the Kudyan model, Motoyama and Ametani translated the
above nonlinear-circuit into a linear circuit with diodes illustrated in Fig. 5.37. While
the Kudyan model based on theory 1), this model on theory 2). That is, the model is a
piece-wise linear approximation of the above nonlinear circuit. An advantage of the
model is that the practical implementation is quite convenient, because the model consists
of linear lumped elements and diodes which have already been prepared in the EMTP.

5-68
D1 D2 D3

C1 C2 C3

R1 R2 R3
E1 E2 E3

Fig. 5.37 Linear corona model by Motoama and Ametani

Other models [19,20, etc] have also been proposed based on theories 1) or 2).
But the serious disadvantages of the theories may be summarized as follows. As for
theory 1), field tests of a q - v curve or surge propagation are always required to
construct a model, although the q - v curve representation properly translates the physical
phenomena of corona. One of the purposes of simulation is that we can obtain a result
without field tests. Thus, the requirement of such the field-test data is
counterproductive as a simulation model. As for theory 2), the approximation
introduced in eq. (5.23) is quite doubtful, because such the frequency-to-time
transformation has to be performed using the theory of Fourier or Laplace transforms.
For example, a numerical convolution should be used. Also, Peek’s quadratic law was
determined by experiments under AC voltages. It is doubtful whether the law is
applicable to surge corona or not.

To reproduce q - v curves without field tests and Peek’s quadratic law, physical
models have recently been proposed. Those models attempt to reproduce a q - v curve
from the geometrical configuration of a line and physical constants. The physical
models may be classified into the following two categories.

5-69
•Finite-Difference Model by M.T. Correia de Barros [6]

The methodology of this model is to solve the macroscopic physical laws that
describe the complete discharge phenomena of transmission-line corona using the finite-
difference method. The method first discretizes the space around a conductor in coaxial
regions. Then, equations which describe electric-field distribution, particle conservation
with recombination, streamer development, and ionization are solved simultaneously at
the boundaries of the coaxial regions. Inception delay is also considered as a part of the
streamer development calculation. If the thickness of each coaxial region is selected to
be thin enough, the method should reproduce the real physical phenomena, and in fact the
model shows a good agreement with field tests. Furthermore, the model can express the
wave-steepness dependence of surge corona, that is, it reproduces different q - v curves
for different wave steepness. But it should be noted that the model is quite time-
consuming. A wave-steepness-dependent corona model which is proposed in the
following sections borrows some calculation procedures from this model for the
realization of the dependence.

•Corona Shell Models [21-24]

The concept of “corona shell” takes notice of a fact that space charges generated
by corona form a shell around the conductor. This concept is quite rational, because the
simplification considerably makes easy the calculation of electric field. J.J. Claudé et.
al. for the first time proposed the concept for the calculation of steady-state corona [21],
and recently it was applied to surge corona by different authors [22-24]. Although the
methodology is quite simple, the models do not reproduce the wave-steepness
dependence of corona. A wave-steepness-dependent corona model that is proposed in
the following sections is based on one proposed by J.F. Guillier et. al. [24]. Because the
Guillier method is described in detail in the following sections, the explanation of the
corona shell is omitted here.

5-70
5.5 Wave-Steepness-Dependent Corona Model [25]

The previous section has reviewed the history of corona models. As a result,
models based on a q - v curve should be recommended, because models based on Peek’s
quadratic law are doubtful due to its applicability to a surge corona. But the
requirement of measured data reduces the value of the q - v-curve-based models as a
simulation model, and thus models based on physical laws which attempt to simulate q - v
curves from the geometrical configuration of a transmission line and physical constants
have recently been proposed. Those physical models may be quite reasonable in terms
of the development of a simulation model. The wave-steepness dependence of corona is
an important consideration, because the q - v curve for a lightning surge is different from
that of a switching surge. So far, only the finite-difference model, which is one of the
most rigorous physical models, can reproduce the dependence, but the model is quite
time-consuming due to its space discretization [6]. To implement a corona model in a
line model, the corona model is to be inserted in many points of the line model by
discretization, and thus the calculation procedure should be simple. Although the
calculation procedures of corona-shell models are based on quite simple physical laws,
the models cannot reproduce the dependence. Therefore, a model that can reproduce
the wave-steepness dependence of q - v curves by a simple calculation procedure is highly
expected.

A wave-steepness-dependent corona model, that is proposed in this section, is


based on Guillier’s corona-shell model [24], and the process of inception delay, streamer
development, and ionization is introduced in the model to represent the wave-steepness
dependence. Because the dependence is realized by simple calculation procedures, and
because the electric-field calculation procedure is based on simple Guillier’s corona-shell
model, the total calculation procedure is still simple enough to gain computational speed,
compared with the finite-difference model. The simplicity enhances the implementation
of the proposed corona model into a line model. The model can be applied to both
lightning- and switching-surge studies because of the ability to reproduce the wave-

5-71
steepness dependence.

To develop a wave-steepness-dependent corona model, the following discharge


phenomena should be considered [6] :

(a) statistical inception delay


(b) streamer development process
(c) ionization process
(d) space-charge drift
(e) space-charge recombination

Phenomena (a), (b), and (c) are fast and of the order of a few microseconds. On the
other hand, (d) and (e) are slow enough to be neglected in a surge region. The surge
region may correspond to a time window of the order of one microsecond to some
milliseconds. Using numerical examples, it is now shown that phenomena (d) and (e)
can be neglected as follows. In case of positive corona, the mobility of positive ions in

the atmosphere is approximately µ + ≅15


. ×10 − 4 [m2/V⋅s] [21], and the electric-field

strength at the streamer tip of corona is almost constant to be Ec+ = 4 ∼ 5 × 105 [V/m]
[24]. The drift velocity of positive ions is then evaluated as :

vd+ = µ+ Ec+ = 60 ∼ 75 × 10− 6 [m/µs]. (5.28)

The above value indicates that the drift of positive ions is negligible in the surge region.
In case of negative corona, the drift velocity of electrons in the atmosphere is evaluated

from values µ − ≅18


. ×10 − 4 [m2/V⋅s] and Ec− = 11 ∼ 18 × 105 [V/m] as :

vd− = µ− Ec− = 200 ∼ 320 × 10− 6 [m/µs]. (5.29)

The above value, which is a little larger than vd+ however, also indicates that the drift of
electrons is negligible in the surge region. Because the drift of both positive ions and
electrons is negligible, phenomenon (d) can be neglected in a simulation regarding surge

5-72
propagation. The recombination of positive ions and electrons takes place as a result of
the drift of both the particles. Thus, phenomenon (e) can also be neglected, if (d) is
neglected. As a result, both phenomena (d) and (e) can be neglected in the surge region.

The above investigation implies that we could simulate corona discharge by


solving equations that describe the physical phenomena of (a), (b), and (c) simultaneously,
but the calculation procedure would be complicated. To avoid such the complicated
calculation procedure, the concept of the quasi-static state of corona is introduced in the
proposed model. The quasi-static state is a state in which phenomena (a), (b), and (c)
have already taken place and been in their steady states, but phenomena (d) and (e) can
still be neglected (have not taken place yet). The quasi-static state may correspond to a
time region about from 10 [µs] to some [ms]. In a q - v curve simulation using the
proposed model, we first calculate the quasi-static q - v curve, and then phenomena (a),
(b), and (c) are taken into account by simple calculation procedures. The quasi-static q -
v curve can be calculated by a corona-shell method which has recently been proposed by
Guillier et. al. [24] and is briefly described in the next section.

5.5.1 Calculation procedure of quasi-static q - v curve

To calculate a quasi-static q - v curve using Guillier’s method, consider an


instant when corona discharge has just developed to radius r = rc around a conductor of
which the radius and height are r0 and h, as illustrated in Fig. 5.38. The applied voltage
is assumed to be positive and rising with respect to time at the instant. Inside the corona
region r ≤rc, molecules of the atmosphere are ionized by the applied electric field. Then,
positive ions generated by the ionization are attracted to the outer direction, and electrons
to the inner direction also by the applied electric field. The electrons reach the
conductor surface and disappear, and the positive ions form a charge shell at radius r = rc.
It should be noted that the formation of the charge shell and the disappearance of the
electrons are assumed to take place instantaneously, and that the formation time of this

5-73
charge configuration is neglected. This assumption enables to calculate the total charge
amount q as a function of only v, but (a) statistical inception delay, (b) streamer
development process, and (c) ionization process are ignored in the calculation procedure.
Therefore, it can be concluded that Guillier’s method calculates a quasi-static q - v curve,
although Guillier et. al. do not mention in ref. [24]. Let the amount of charges on the
conductor surface be q0, on the charge shell qc, and the electric-field strength at the
charge shell Ec. The total amount of charges q = q0 + qc is given by Gauss’theorem as :

1 1 
q = 2πε0 E c /  + . (5.30)
rc 2h − rc 

The conductor voltage v is calculated as :

 r (2h − rc )  rc rc (2h − rc ) 2h − r0
v = r0 E 0 − c E c  ln + E c ln , (5.31)
 h(2h − r0 )  r0 2h rc

+ +
+
conductor + +

q0 + r0 +

+ +

+ rc +
space-charge shell + +
+ + +
qc
h
corona region

Fig. 5.38 Schematic explanation of "charge shell" of transmission-line corona

5-74
where E0 : electric-field strength at the surface of conductor.

If it is assumed that the electric-field strength at the conductor surface can be evaluated
by the following formula proposed by G. Hartmann [8] :

 0.1269 
E 0 = m ⋅2.594 ×10 6 1 + 0.4346  [V/m], (5.32)
 r0 

where m : surface-state coefficient of conductor.

Assuming that the electric-filed strength at the tip of corona is Ec+ = 4 ∼ 5 × 105 [V/m] as
mentioned above, then the corona radius rc can be calculated by eq. (5.31) with respect to
a given applied voltage v, because the values of E0 and Ec+ are now known.
Unfortunately, eq (5.31) cannot be solved with respect to rc analytically, and thus a
Newton-Raphson iteration scheme may be used for the numerical calculation. If rc
determined at the previous time step is used as the initial guess of the Newton-Raphson
scheme (r0 is used at time zero), the iteration procedure rapidly converges. By
substituting the value of rc in eq. (5.30), we can finally calculate the total amount of
charges q as a function of an applied voltage v. In case of negative corona, q can also be
calculated as a function of v in the same manner by replacing Ec+ with Ec− = 11 ∼ 18 × 105
[V/m]. When an applied voltage becomes decreasing, corona discharge rapidly
disappears, and the amount of charges follows q = C0v, where C0 is the geometrical
capacitance. Of course, if an applied voltage is lower than a corona-onset voltage given
by eq. (5.32), the amount of charges is calculated by q = C0v. By the way, eq. (5.32) is
in the same form as experimental one proposed by Peek. In the Peek formula,
coefficients are determined by a number of experimental data. Later, Hartmenn
theoretically evaluated the values of the coefficients, and eq. (5.32) shows the resulting
formula.

5-75
5.5.2 Introduction of wave-steepness dependence

The wave-steepness dependence of corona cannot be reproduced by the quasi-


static q - v curve calculation procedure described in the previous section, because the
amount of total charges q is calculated as a function of only an applied voltage v by
assuming that charges instantaneously move to a certain configuration determined by the
applied voltage. In reality, phenomena (a), (b), and (c) first take place in the order of
some microseconds, and then a quasi-static state results from those phenomena. Thus,
the phenomena have to be taken into account, when the wave-front time of an applied
voltage is some microseconds. The wave-steepness-dependent phenomena can be
included as follows. As discussed in detail in refs. [26,27], corona does not start
growing immediately after an applied voltage exceeds the corona-onset voltage given by
eq. (5.32) because of the detachment time of seed electrons. Thus, there is a time delay
before the actual inception, and this time lag is called a statistical inception delay τd,
phenomenon (a). After the delay time, charges do not immediately move to a certain
configuration but grow with the velocity of corona streamers vst, phenomenon (b). In
stead of using rc given by eq. (5.31), the following formula gives a corona radius taking
into account phenomena (a) and (b) [6] :

rc ' = r0 + v st ⋅( t − τd ) ⋅u(t − τd ), (5.33)

where t is a time starting when an applied voltage exceeds the onset voltage of
eq.(5.32), and u(t) : unit step function.

The corona streamers cannot develop beyond r = rc determined by eq. (5.31), because an
applied voltage does not supply energy for further development. That is, when the
streamers reach r = rc, the configuration of charges is in a quasi-static state. There is
one more important consideration. Due to the presence of the statistical inception delay
τd, the electric-field strength at the surface of a conductor is far greater than that given by
eq. (5.32), because an applied voltage continues rising during the time lag. Thus, at the

5-76
moment of an actual inception corresponding to a time when the streamers begin growing,
we assume that corona can immediately develop to radius

hq0
rc′= h − h2 − , q0 = C0 v CR
′, (5.34)
πε0 Ec

where v’CR : applied voltage at the moment of actual inception (the derivation of the

above equation is given in the following footnote (*1). This assumption is quite rational,
because the streamer development from r = r0 to r = r’c determined by the above equation

is so fast due to the excessive electric field. As a result, the actual corona radius r’c is

determined as : 1) at the moment of an actual inception, r’c is determined by eq. (5.34)

and starts growing as determined by eq. (5.33); 2) when r’c reaches rc determined by eq.

(5.31), then corona stops growing. By substituting the actual corona radius r’c

determined by the above calculation procedure into eq. (5.30), the total charge amount q
can be obtained. We now need to modify q for introducing the ionization process of
corona (phenomenon (c)). The relaxation time of the ionization τi can be taken into

_______________________________________________________________

(*1) The following equation is obtained by solving eq. (5.30) with respect to rc :

hq0
rc′= h ± h 2 − .
πε0 Ec

In the above equation, the plus sign of ± is not appropriate regarding a transmission-line
configuration, and thus we chose the minus sign. Substituting q = q0 that is the total
charge amount at the moment of an actual corona inception, eq. (5.34) is obtained.

5-77
account by applying a first-order low-pass filter with time constant τi to the amount of

generated charges ∆q = q − q0. The “filtered” generated charge amount ∆q’is given in
the Laplace s-domain as :

1
∆Q ′
( s) = ∆Q( s), (5.35)
1 + sτ i

where Laplace s-domain counterparts are denoted in uppercase letters. Using the Bi-
Linear Transform (BLT) :

2 1− z− 1
s= (5.36)
∆t 1 + z − 1

that is the frequency-domain representation of the trapezoidal rule of integration, eq.


(5.35) can be transformed into the Z-domain as :

2τ i
1−
a + a1 z − 1 1 ∆t ,
∆Q ′
(z) = 0 ∆Q( z ), a 0 = a1 = , b1 = (5.37)
1 + b1 z − 1 2τ i 2τ i
1+ 1+
∆t ∆t

and the inverse Z-transform of the above equation is :

∆q ′
(t ) = a 0 ∆q (t ) + a1 ∆q (t − ∆t ) − b1 ∆q ′
(t − ∆t ). (5.38)

The above equation can directly be implemented in computer code. Finally, the actual
total charge amount is obtained in the following form :

(t ) = q 0 (t ) + ∆q ′
q′ (t ). (5.39)

As described above, the total amount of charges q’(t) is calculated as a function of an


applied voltage v(t) which varies with respect to time, taking into account wave-
steepness-dependent phenomena (a), (b), and (c). The calculation procedure can be

5-78
summarized as shown in Fig. 5.39. It is clear from the figure that the calculation
procedure of the proposed corona model is quite simple, and that the model is suitable for
implementing in a line model. In the figure, a quasi-static corona radius is calculated in
each iteration step of each time step of a transient simulation using a Newton-Raphson
iteration scheme. But the quasi-static q - v curve may be calculated in advance of the
transient simulation, because it can be calculated as a function of only an applied voltage.
And the curve may be fitted with an analytical formula such as a polynomial to be stored
in memory, in order to gain computational speed of the transient simulation.

5-79
START :
input time t and voltage v

electric-field
NO strength at conductor surface
E0 exceeds corona-onset
voltage by eq. (5.32)?

YES

calculation of corona radius of


quasi-static state rc by eq. (5.31)

t > τd ?
NO
τd : (a) statistical inception
delay

YES

t = τd ? YES
moment of actual inception?

NO

Calculation of actual corona Calculation of actual corona


radius r'c by eq. (5.33) considering radius r'c by eq. (5.34)
(a) inception time delay τd
(b) streamer velocity v st
(if r'c > rc , then r'c = rc )

Charge calculation by q' = Cv Calculation of actual amount of


charges q' considering (c) relaxation
time of ionization τi by eqs. (5.30,39)

END : output total


charge amount q' ( t )

Fig. 5.39 Calculation procedure of wave-steepness dependent corona model

5-80
5.6 Simulaion of q - v and Coupling-Factor Curves

This section presents q - v and coupling-factor curve simulations carried out by


the proposed wave-steepness-dependent corona model. Measured q - v and coupling-
factor curves presented in Secs. 5.2 and 5.3 and also in ref. [5] are reproduced by the
proposed model. Pure coupling-factor curves which could not be obtained from
measurements due to the lack of experimental facilities in Sec. 5.3 are now given by
simulation. In all the simulations, the simulation time step is set to 0.01 [µs], and 1.0 ×
10− 6 is used as the convergence constant of the Newton-Raphson iteration scheme in eq.
(5.31). The digitally recorded waveforms of applied voltages are used in the
simulations.

5.6.1 Simulation of q - v curves

In order to reproduce measured q - v curves of an OE60sq wire shown in Fig.


5.12, simulations using the proposed wave-steepness dependent corona model are carried
out. The conductor radius and height are set to 5.0 [mm] and 1.83 [m] respectively, as
shown in Fig. 5.5. The parameters of the corona model used in the simulations, m:
surface-state coefficient of conductor, Ec+, Ec− : electric-field strength at the tip of corona

for positive and negative polarities, τd: statistical inception delay, vst: velocity of corona

streamers, and τi: relaxation time of ionization, are given in Table 5.4. The calculated
results are shown in Fig. 5.40, and agree well with measured results shown in Fig. 5.12.
Each q - v curve has its own path, especially just before the decrease of the applied
voltage. This wave-steepness dependence is also reproduced accurately by the model.
Fig. 5.41 shows corona radius which varies with respect to applied voltage. It is
observed that the corona radius of the positive cases is much larger than that of the
negative cases, and that it reaches almost 25 [cm] in the 600-kV case.

5-81
12

10
(a) positive
charge, µC/m
8

2
geometrical capacitance
0
0 200 400 600
voltage, kV

12

10
(b) negative
charge, µC/m

2
geometrical capacitance
0
0 200 400 600
voltage, kV

Fig. 5.40 Calculated q - v curves (OE60sq, fair weather)

5-82
(a) positive
radius, cm 20

10
conductor radius

0
0 200 400 600
voltage, kV

(b) negative
20
radius, cm

10
conductor radius

0
0 200 400 600
voltage, kV

Fig. 5.41 Calculated corona radius (OE60sq, fair weather)

5-83
Table 5.4 Parameters of corona model (OE60sq wire)

Ec+ Ec− τd vst τi


m
[kV/m] [kV/m] [µs] [m/µs] [µs]
0.5 800 2,000 0.16 1.0 0.22

It should be noted that Ec+ and Ec− used in the simulations are larger than the
standard values mentioned in Sec. 5.5.1 (Ec+ = 400 ~ 500 [kV/m], Ec− = 1,100 ~ 1,800
[kV/m]). The proposed corona model assumes that Ec+ and Ec− are almost constant in
the calculation procedure of a quasi-static q - v curve, as stated in refs. [24,28]. But E.J.
Los pointed out that Ec+ and Ec− were actually not constant when a corona radius rc was
small [29]. He gathered experimental data of Ec+ and Ec− with respect to rc from
different papers: Wagner and Lloyd’s [1], McCann’s [2], Maruvada’s [5], and his data
[29], and plotted together as shown in Fig 5.42. The figure shows that Ec+ and Ec− are
large when rc is small. Only Wagner and Lloyd’s data (vertical ground plane) indicate

that Ec+ and Ec− are significantly large when rc is small, and the others indicate slightly
large on the other hand. A subsequent study by T. Minakawa and M. Ishii proposed the
following approximate formulas to determine Ec+ and Ec− from rc [30] :

80
E c+ = + 550 [kV / m], (5.40)
rc + 0.07

180
E c− = + 650 [kV / m]. (5.41)
rc + 0.012

The above formulas satisfy the Wagner and Lloyd data as shown in Fig. 5.42, and thus
give significantly large values when rc is small. The author tried to implement the above
formulas in the proposed corona model, but calculated charge amount was always too

5-84
small compared with field data due to the too large values of Ec+ and Ec− . Therefore, the
formulas have been removed from the model, and large constant values are simply used
instead as shown in Table 5.4. The constant values can be regarded as the mean values,
and give a good agreement with the field-test results. It should be noted that this sort of
consideration regarding the variation of the electric-field strength at the corona tip is
important when a single thin conductor is of interest such as ones used in the
measurements of this thesis, because the electric-field strength is almost constant when
the corona radius is larger than 30 [cm] as shown in Fig. 5.42. In a real transmission line,
the equivalent radius of a conductor bundle used as a phase wire is quite large, e.g. the
equivalent radius of a typical 500-kV phase wire is 38 [cm], and thus the standard values
of Ec+ and Ec− can be used.

40

(positive), (negative): McCann ( h = 15.2m),

, : McCann ( h = 9.1m),
at the corona tip [kV/cm]

30 : Wagner & Lloyd (cylindrical ground plane),


electric-field strength

, : Wagner & Lloyd (vertical ground plane),

: Maruvada et. al.,

, : G.E. (case 1), , : G.E. (case 2)


20
: eq. (5.40), : eq (5.41)

10

0
0 100 200 300
corona radius [cm]

Fig. 5.42 Electric-field strength at the corona tip versus corona radius

5-85
Figs. 5.43 and 5.44 show calculated q - v curves and corona radius of an
HDCC8sq wire. The conductor radius and height are set to 1.8 [mm] and 1.85 [m]
respectively as illustrated in Fig. 5.6 and Table 5.3. Table 5.5 shows the parameters of
the corona model used in the simulations. Because the conductor radius is also small,
Ec+ and Ec− are chosen to be the same large constant values as the previous OE60sq wire.
By adjusting the statistical inception delay τd, and the velocity of corona streamers vst for
each charging voltage, the calculated results agree well with measured q - v curves shown
in Fig. 5.19. A higher charging voltage gives a smaller statistical inception delay τd and
a greater velocity of corona streamers vst. The variation of the statistical inception delay
can be explained by a critical volume theory proposed in ref. [26]. The variation of the
velocity of corona streamers agrees with experiments carried out by H. Rather [9, page
73]. Therefore, it may be concluded that the parameters of the wave-steepness
dependence satisfy our physical knowledge, and the theoretical evaluation of the
parameters should be an important future work. Fig. 5.45 illustrates how the wave-
steepness dependent parameters τd, vst, and τi contribute to a q - v curve. The statistical
inception delay τd increases an actual inception voltage, and the velocity of corona
streamers vst determines the slope just after the inception. Then, the relaxation time of
ionization τi suppresses the increase of the charge amount.

To confirm the accuracy of the proposed corona model for different wave-front
times, measured q - v curves presented in ref. [5] are used for the validation, because the
author and his sponsor could not prepare an impulse generator which can generate a long
wave-front-time surge such as a switching surge. In this comparison, the proposed
model can be validated in terms of the wave-steepness dependence of corona. The
conductor radius is 1.52 [cm], but the conductor height cannot be found in ref. [5].
Thus, the height was calculated from the radius and the geometrical capacitance value
obtained from the measured q - v curves. It should be noted that the q - v curves are
measured using a corona cage, and that the geometrical capacitance must be different
(presumably larger) from the real value without the cage. Also, the electric-field
distribution around the conductor may be different. Ref. [5] mentions that the q - v
curves are measured under a fair weather. The q - v curves are simulated by the

5-86
6

5 (a) positive

4
charge, µC

1
geometrical capacitance
0
0 100 200 300
voltage, kV

-6

-5 (b) negative

-4
charge, µC

-3

-2

-1
geometrical capacitance
0
0 -100 -200 -300
voltage, kV

Fig. 5.43 Calculated q - v curves (HDCC8sq, line length = 22.1 [m], fair weather)

5-87
(a) positive
20
radius, cm

10
conductor radius

0
0 100 200 300
voltage, kV

(b) negative
20
radius, cm

10
conductor radius

0
0 200
voltage, kV

Fig. 5.44 Calculated corona radius (HDCC8sq, line length = 22.1 [m], fair weather)

5-88
Table 5.5 Parameters of corona model for HDCC8sq wire

Ec+ Ec−
m
[kV/m] [kV/m]
0.3 800 -2,000

v τd vst τi
[kV] [µs] [m/µs] [µs]
+200 1.2 0.005 0.02
+300 0.45 0.015 0.02

+400 0.33 0.026 0.02

+500 0.27 0.038 0.02

+600 0.25 0.045 0.02

+700 0.25 0.060 0.02

+800 0.25 0.060 0.02

-300 0.40 0.060 0.02


-400 0.31 0.060 0.02

-500 0.25 0.060 0.02

-600 0.25 0.060 0.02

-700 0.25 0.060 0.02

-800 0.25 0.060 0.02

5-89
q τi

quasi-static q - v
curve

vst

vCR τd v

Fig. 5.45 Effects of wave-steepness-dependent parameters on q - v curve

proposed corona model with parameters given in Table 5.6. Because the conductor
radius is greater than the previous OE60sq and HDCC8sq wires, the standard values
mentioned in Sec. 5.5.1 are used as the electric-field strength at the corona tip Ec+ and Ec− .

A switching-surge voltage shown in Fig. 5.46 (a), of which the wave-front time is 260
[µs] and of the wave-tail time is 2700 [µs] (260/2700 µs), is used as the input to the
corona model to simulate the q - v curves. The simulations are carried out by setting the
maximum voltage of the applied voltage to +285, ±340, +400, and +450 [kV]. The
calculated q - v curves are shown in Fig. 5.47 (a) and corresponding measured q - v
curves replotted from ref. [5] are given in Fig. 5.47 (b). Because the wave-front time of
the applied voltages is much longer than the wave-steepness-dependent parameters
shown in Table 5.6, the dependence is not observed in both the calculated and measured
results, and the calculated curves are identical to corresponding quasi-static curves. At
the turning points (when the applied voltage starts decreasing) the calculated curves are

5-90
Table 5.6 Parameters of corona model for conductor with radius 1.52 [cm]

Ec+ Ec−
m
[kV/m] [kV/m]
0.6 500 -1,100

v τd vst τI
[kV] [µs] [m/µs] [µs]
+285 0.7 0.3 0.5
±340 0.6 0.3 0.5

+400 0.5 0.9 0.5

+450 0.5 0.9 0.5

acute. On the other hand, the measured curves are round. In Sec. 5.2.8, an adjacent
conductor modifies a q - v curve (see Fig. 5.22), and the measured curves are acquired
using a corona cage. Thus, we cannot discuss on the accuracy at the turning points.
Considering the above, the calculated results can be said to agree well with the measured
results.

Fig. 5.46 (b) shows a lightning-surge voltage (2.5/60 µs), and q - v curves are
simulated by the proposed model when the voltage is applied. The maximum value of
the applied voltage is set to the same values as the previous switching-surge case. The
calculated and measured results are shown in Fig. 5.48 (a) and (b) respectively, and show
a good agreement. The wave-front time is of the same order of the wave-steepness-
dependent parameters shown in Table 5.6, significant dependence is observed. The
same simulations are carried out by Correia de Barros in ref. [6] using a rigorous finite-
difference model, and the parameters shown in Table 5.6 are taken from the reference.

5-91
voltage, p.u.
1

0.5

0
0 500 1000 1500 2000 2500
time, µs
(a) switching surge
voltage, p.u.

0.5

0
0 10 20 30 40 50
time, µs
(b) lightning surge

Fig. 5.46 Applied-voltage waveforms

5-92
10
r0 = 1.52 cm positive
8 260/2700 µs negative
fair weather
charge, µC/m

2
(a) calculated
0
0 100 200 300 400 500
voltage, kV

Fig. 5.47 Calculated and measured q - v curves (r0 = 1.52 [cm], switching surge)

5-93
10
r0 = 1.52cm positive
8 2.5/60 µs negative
fair weather
charge, µC/m

2
(a) calculated
0
0 100 200 300 400 500
voltage, kV

Fig. 5.48 Calculated and measured q - v curves (r0 = 1.52 [cm], lightning surge)

5-94
Because the proposed corona model gives almost the same results as the rigorous model
from the same parameters, it can be concluded that the proposed model is physically valid.
Furthermore, the proposed model reproduces the wave-steepness dependence of corona
by a simpler calculation procedure compared with the rigorous model.

5.6.2 Simulation of coupling-factor curves

In order to simulate a coupling-factor curve, the theoretical background is first


given. Consider an n-phase transmission line. In the absence of corona, voltage and
charge vectors v = (v1, v2, ..., vn)T, q = (q1, q2, ..., qn)T are related by the potential-
coefficient matrix P0 :

v = P0 q. (5.42)

In the presence of corona, eq. (5.42) is extended to the following incremental relation by
the reason mentioned in Sec. 5.1 (see eq. (5.4)).

∂v ∂q
= P (v ) (5.43)
∂t ∂t

Compared with eq. (5.4), the incremental potential-coefficient matrix P(v) is the inverse
of the incremental capacitance matrix C(v).

C (v ) = P − 1(v ) (5.44)

Thus, we first calculate P(v), and then C(v) can be obtained by the above relation. The
i-th row of eq. (5.43) is

∂v i ∂q ∂q ∂q
= Pi1 1 + L + Pii i + L + Pin n . (5.45)
∂t ∂t ∂t ∂t

5-95
Assume that all the wires except the i-th wire are grounded, and that the i-th wire is in a
corona condition. Then, eq. (5.45) becomes :

∂v i ∂q
= Pii ( v i ) i , (5.46)
∂t ∂t

and thus the diagonal elements of P(v) is given as :

∂v i
Pii ( v i ) = (i = 1, 2, ..., n) (5.47)
∂qi

In order to obtain the off-diagonal elements, consider the i-th and j-th wires as illustrated
in Fig. 5.49. The i-th wire is in a corona condition, and q0i and qci are charges on and
around the i-th wire respectively. Applying Gauss’ theorem to q0i and qci, we can
assume that the total amount of charges qi = q0i + qci are equivalently on the i-th wire.
In the figure, a broken line encloses the region to which Gauss’ theorem is applied.
Therefore, the off-diagonal elements of P(v) are the same as those of P0 :

1 ( hi + h j )2 + d 2
Pij = ln (i = 1, ..., n, j = 1, ..., n, i ≠ j). (5.48)
2πε0 ( hi − h j ) 2 + d 2

where hi: height of i-th wire, hj: height of j-th wire, dij: horizontal separation
between i-the and j-th wires.

Because we are now able to calculate P(v), C(v) can be obtained by the matrix inversion
of eq. (5.44).

5-96
i-th conductor j-th conductor

q0i
hj
qci
corona
hi vj

dij

Fig. 5.49 Gauss’theorem applied to determine off-diagonal elements of incremental


potential coefficient matrix P(v)

To reproduce measured coupling-factor curves presented in Secs. 5.3.1 and


5.3.2, the equivalent resistance of a potential divider (PD) connected to the induced phase
Rdiv has to be considered as mentioned in Sec. 5.3.1 (see Fig. 5.24). The author’s
simulation experience indicates that the capacitance of measurement system Cs is also
required to obtain a good accuracy. Thus the equivalent circuit of the coupling-factor
measurement system becomes Fig. 5.50. Using the symbols in the figure, a set of state
equations describing the dynamics of the system is obtained as :

5-97
i-th conductor j-th conductor
i1 i2

Rdiv

e(t) I.G. v1 v2
C(v,t)
Cs

Fig. 5.50 Equivalent circuit of coupling-factor measurement system


v = e( t ),
1
i1  C11 C12  ∂  v1 
  =    , (5.49)
i2  C21 C22  ∂t v 2 
 v ∂v
− i2 = 2 + Cs 2 ,
 Rdiv ∂t

and is modified as :

∂v 2
 ∂t = f ( v 2 , t ),

  v2
(5.50)
 f ( v2 , t ) = − 1 ∂e( t ) 
 + C12 .

 C 22 + C s  Rdiv ∂t 

Solving the above state equation, a coupling-factor curve is calculated by

5-98
v2 ( t )
K ( v1 ) = K ( v1 ( t )) = . (5.51)
v1 ( t )

A predictor-corrector scheme is used to solve eq. (5.50).

Coupling-factor curves of two HDCC8sq wires are simulated by the


combination of the proposed wave-steepness-dependent corona model and the above
calculation method. The calculated results with conductor separation 1.2, 0.9, and 0.6
[m] are shown in Figs. 5.51, 5.52, and 5.53 respectively. Those figures correspond to
measured results shown in Figs. 5.27, 5.29, and 5.31, and show a good agreement. In
the simulations, the parameters shown in Table 5.5 are used in the corona model. The
capacitance of the PD is calculated to be Cdiv = 5 [pF] by its time response, and the
capacitance of the lead wire is calculated to be Clead = 10 [pF]. Also, from Fig. 5.18,
Ccorrect = 18.5 [pF] may be required to correct the geometrical capacitance. All together,
the capacitance value of the measurement system is assumed to be Cs = Cdiv + Clead +
Ccorrect = 33.5 [pF]. Then, the pure coupling-factor curves of the three cases are
reproduced by simulation. A significantly large and small values are set to Rdiv and Cs
respectively to neglect their effects. The calculated results are shown in Figs. 5.54, 5.55,
and 5.56. The results show a significant increase of the coupling factors due to the
geometrical space charges.

In the same manner, coupling-factor curves of two OE60sq wires are calculated
and shown in Fig. 5.57, which corresponds to measured curves of Fig. 5.32. The corona
model parameters given in Table 5.4 are used, and the capacitance of the measurement
system is assumed to be Cs = Cdiv + Clead + Ccorrect = 23 [pF]. Fig. 5.58 shows calculated
pure coupling-factor curves. Each curve has its own path, indicating the wave-
steepness dependence.

5-99
1

(a) positive
0.8
separation = 1.2 m
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 100 200 300
applied voltage, kV

(b) negative
0.8
separation = 1.2 m
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 -100 -200 -300
applied voltage, kV

Fig. 5.51 Calculated coupling-factor curves (sep. = 1.2 [m], HDCC8sq, fair weather)

5-100
1

(a) positive
0.8
separation = 0.9 [m]
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 100 200 300
applied voltage, kV

(a) negative
0.8
separation = 0.9 [m]
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 -100 -200 -300
applied voltage, kV

Fig. 5.52 Calculated coupling-factor curves (sep. = 0.9 [m], HDCC8sq, fair weather)

5-101
1

(a) positive
0.8
separation = 0.6 m
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 100 200 300
applied voltage, kV

(b) negative
0.8
separation = 0.6 m
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 -100 -200 -300
applied voltage, kV

Fig. 5.53 Calculated coupling-factor curves (sep. = 0.6 [m], HDCC8sq, fair weather)

5-102
1

(a) positive
0.8
separation = 1.2 m
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 100 200 300
applied voltage, kV

(b) negative
0.8
separation = 1.2 m
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 -100 -200 -300
applied voltage, kV

Fig. 5.54 Pure coupling-factor curves (sep. = 1.2 [m], HDCC8sq, fair weather)

5-103
1

(a) positive
0.8
separation = 0.9 [m]
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 100 200 300
applied voltage, kV

(b) negative
0.8
separation = 0.9 [m]
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 -100 -200 -300
applied voltage, kV

Fig. 5.55 Pure coupling-factor curves (sep. = 0.9 [m], HDCC8sq, fair weather)

5-104
1

(a) positive
0.8
separation = 0.6 m
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 100 200 300
applied voltage, kV

(b) negative
0.8
separation = 0.6 m
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 -100 -200 -300
applied voltage, kV

Fig. 5.56 Pure coupling-factor curves (sep. = 0.6 [m], HDCC8sq, fair weather)

5-105
1

(a) positive
0.8
separation = 1.2 m
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 100 200 300 400 500
applied voltage, kV

(b) negative
0.8
separation = 1.2 m
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 -100 -200 -300 -400 -500
applied voltage, kV

Fig. 5.57 Calculated coupling-factor curves (sep. = 1.2 [m], OE60sq, fair weather)

5-106
1

(a) positive
0.8
separation = 1.2 m
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 100 200 300 400 500
applied voltage, kV

(b) negative
0.8
separation = 1.2 m
coupling factor

0.6
geometrical coupling factor

0.4

0.2

0
0 -100 -200 -300 -400 -500
applied voltage, kV

Fig. 5.58 Pure coupling-factor curves (sep. = 1.2 [m], OE60sq, fair weather)

5-107
As a result of this section :

1) The proposed corona model has accurately reproduced q - v and coupling-factor


curves including the wave-steepness dependence of corona.

2) The accuracy of the proposed corona model is almost the same as the rigorous
finite-difference model, although the proposed calculation procedure is much
simpler.

3) The theoretical and qualitative evaluation of the wave-steepness-dependent


parameters: statistical inception delay τd, velocity of corona streamers vst, relaxation

time of ionization τi, which varies with respect to time, voltage, temperature,
humidity, and etc., may be an important future work.

5-108
References of Chapter 5

[1] C.F. Wagner and B.L.Lloyd, “Effects of Corona on Traveling Waves”, AIEE Trans. on Power
Apparatus and Systems, Vol.74, part Ⅲ, 1955.
[2] G.D. McCann, “The Effect of Corona on Coupling Factors between Ground Wires and Phase
Conductors,” AIEE Trans., Vol. 62, 1943.
[3] M.T. Correia de Barros, “Identification of the Capacitance Coefficients of Multiphase
Transmission Lines Exhibiting Corona under Transient Condition”, IEEE Trans. on Power
Delivery, Vol. 10, No. 3, 1995.
[4] T. Noda, H. Fukuzono, S. Sekioka, N. Nagaoka, and A. Ametani, “Charge - Voltage Measurement
of Transmission-Line Corona without a Corona Cage,” Proc. of the Seventh Annual Conference of
Power & Energy Society, IEE of Japan, Session II, Paper No. 427, p. 451, 1996. (in Japanese: 野
田,福園,関岡,長岡,雨谷,「コロナケージを用いない送電線コロナ電荷−電圧曲線の
測定」,平成 8 年電気学会電力・エネルギー部門大会,論文 II, 427, p. 451, 1996.)
[5] P.S. Maruvada, H. Menemenlis, and R. Malewski, “Corona Characteristics of Conductor Bundles
under Impulse Voltages”, IEEE Trans. on Power Apparatus and Systems, Vol. PAS-96, No. 1,
1977.
[6] M.T. Correia de Barros, “Wide Bandwidth Modeling of Corona on High Voltage Transmission
Lines”, IEEE Dielectries and Electrical Insulation Society Conference, Electrical Insulation and
Partial Discharges, Paper No. 78, 1994. Or the same theory can be found from :
Célia de Jesus and M.T. Correia de Barros, “Modelling of Corona Dynamics for Surge
Propagation Studies,” IEEE Trans. on Power Delivery, Vol. 9, No. 3, 1994.
[7] M. Ohki, High Voltage Engineering, Maki Publishing, 1982. (in Japanese: 大木,「高電圧工学」,
槙書店,1982.)
[8] G. Hartmann, “Theoretical Evaluation of Peek’s Law,” IEEE Trans. on Industry Applications, Vol.
IA-20, No. 6, 1984.
[9] T. Ohshige and M. Hara, High Voltage Phenomena, Morikita Publishing, 1973. (in Japanese: 大
重,原,「高電圧現象」,森北出版,1973.)
[10] L.V. Bewley, “Traveling Waves on Transmission Systems,” Dover Publications Inc., New York,
1963.
[11] W.H. Press, S.A. Teukolsky, W.T. Vetterling, and B.P. Flannery, “Numerical Recipes in C − The
Art of Scientific Computing − ”, Second Edition, Cambridge University Press, 1988.
[12] J. Umoto and T. Hara, “Numerical Analysis of Line Equations Considering Corona Loss on
Single-Conductor System,” Vol. 89, No. 968, 1969. (in Japanese: 卯本,原,「コロナ損を考慮
した単導線系伝搬方程式の数値解析」,電学誌,Vol. 89, No. 968, 1969.)
[13] H.M. Kudyan and C.H. Shiu, “A Nonlinear Circuit Model for Transmission Line in Corona”,
IEEE Trans. on Power Apparatus and Systems, Vol. PAS-100, 1981.
[14] M.M. Suliciu and I. Suliciu, “A Rate Type Constitutive Equation for the Description of the Corona
Effect”, IEEE Trans. on Power Apparatus and Systems, Vol. PAS-100, 1981.
[15] A. Inoue, “Propagation Analysis of Overvoltage Surges with Corona Based upon Charge Versus
Voltage Curve”, IEEE Trans. on Power Apparatus and Systems, Vol. PAS-104, No. 3, 1985.

5-109
[16] K.C. Lee, “Non-linear corona models in an electromagnetic transients program (EMTP)”, IEEE
Trans. on Power Apparatus and Systems, Vol. PAS-102, No. 9, 1983.
[17] H.W. Dommel and W.S. Meyer, “Computation of electromagnetic transients,” Proc. IEEE, Vol.
62, 1974.
[18] H. Motoyama and A. Ametani, “Development of a Linear Model for Corona Wave Deformation
and its Effect on Lightning Surges,” Trans. IEE Japan, Vol. 107-B, No. 3, 1987. (in Japanese : 本
山,雨谷,「コロナ波形変わいの線形モデルの開発と雷サージへの影響」,電学誌,Vol.
107-B, No. 3, 1987.)
[19] R.J. Harrington, “Implementation of Computer Model to Include the Effects of Corona in
Transient Overvoltage Calculations,” IEEE Trans. on Power Apparatus and Systems, Vol. PAS-
102, No. 4, 1983.
[20] P.S. Maruvada, D.H. Nguyen, and H. Hamadani-Zadeh, “Studies on Modeling Corona
Attenuation of Dynamic Overvoltages,” IEEE Trans. on Power Delivery, Vol.4, No. 2, 1989.
[21] J.J. Claudé, C.H. Gary, and C.A. Lefévre, “Calculation of Corona Losses Beyond the Critical
Gradient in Alternating Voltage,” IEEE Trans. on Power Apparatus and Systems, Vol. PAS-88,
No. 5, 1969.
[22] A. Semlyen and H. Wei-Gang, “Corona Modelling for the Calculation of Transmission Lines,”
IEEE Trans. on Power Delivery, Vol. PWRD-1, No. 3, 1986.
[23] M.A. Al-Tai, H.S.B. Elayyan, D.M. German, A. Haddad, N. Harid, and R.T. Waters, “The
Simulation of Surge Corona on Transmission Lines,” IEEE Trans. on Power Delivery,Vol.4, No.
2, 1989.
[24] J.F. Guillier, M. Poloujadoff, and M. Rioul, “Damping Model of Traveling Waves by Corona
Effect along Extra High Voltage Three Phase Lines,” IEEE Trans. on Power Deliverry, Vol. 10,
No. 4, 1995.
[25] T. Noda, N. Nose, N. Nagaoka, and A. Ametani, “A Wave-Front-Time Dependent Corona Model
for Transmission-Line Surge Calculations,” Proc. of the Seventh Annual Conference of Power &
Energy Society, IEE of Japan, Session I-J, Paper No. 36, pp. 209-214, 1996. (to be published in
Trans. IEE of Japan, in Japanese: 野田,能勢,長岡,雨谷,「線路サージ計算に用いる波頭
長依存コロナモデル」,平成 8 年電気学会電力・エネルギー部門大会,論文 I-J, 36, pp.
209-214, 1996.)
[26] N. Harid and R.T. Waters, “Statistical Study of Impulse Corona Inception Parameters on Line
Conductors,” IEE Proc.-A, Vol. 138, No. 3, 1991.
[27] C. Menemenlis and H. Anis, “Influence of the Delay of the First Corona Pulse on the Switching
Impulse Breakdown Probability,” IEEE Trans. on Power Apparatus and Systems, Vol. PAS-94,
No. 2, 1975.
[28] R.T. Waters, D.M. German, A.E. Davis, N. Harid, and H.S.B. Elayyan, “Twin Conductor Surge
Corona,” Fifth International Symposium on High Voltage Engineering, Braunschweig, Federal
Republic of Germany, 24-28, August, 1987.
[29] E.J. Los, “New Studies of the Transient Glow Discharge in Parallel Electrode Gaps,” IEEE Trans.
on Power Apparatus and Systems, Vol. PAS-99, No. 2, 1980.
[30] T. Minakawa and M. Ishii, “Lightning Surge Analysis Taking Account of the Dynamic
Characteristics of Surge Corona,” Proc. of the Seventh Annual Conference of Power & Energy
Society, IEE of Japan, Session I-J, Paper No. 35, pp. 203-208, 1996. (in Japanese: 皆川,石井,
「コロナのダイナミックな特性を考慮した雷サージ波形解析」,平成 8 年電気学会電力・
エネルギー部門大会,論文 I-J, 35, pp. 203-208, 1996.)

5-110
6 REALIZATION OF NONLINEARITY
DUE TO CORONA DISCHARGE

Summary

This chapter presents an efficient method to deal with nonlinear corona branches in
the proposed line model. First, the line model is discretized in space, and the wave-steepness-
dependent corona model described in the preceding chapter is inserted as a corona branch at
each of the discretization points. The Norton equivalent circuit of two linear portions of the
proposed line model seen from the corona branch at a discretization point is then derived, and a
nonlinear state-space equation is established together with the corona branch. This nonlinear
state-space equation is solved by the trapezoidal rule of integration and the predictor-corrector
method. The use of the trapezoidal rule enables to solve the nonlinear corona branches with
the same degree of accuracy as the other electrical parts in an EMTP-type program. Also, the
use of the predictor-corrector method avoids the Jacobian calculation of iteration, which is
difficult to be evaluated by a physical corona model. The calculated results of wave
propagation including the corona and skin effects are compared with field tests, in order to
validate the proposed line model.

6-1
6.1 State-Space Approach To Deal with Nonlinear Corona Branches

6.1.1 Space discretization

J.L. Naredo et al. proposed an interesting method to calculate a transmission-


line transient with corona. The method deals with only the voltages and currents at the
sending and receiving ends by use of a modified lattice diagram which takes into account
the wave delay due to corona [1]. But the method is not general enough to take into
account the frequency dependence of the line, although no requirement of space
discretization considerably saves a computation time. When a transmission-line
transient calculation is performed considering both the corona nonlinearity and the
frequency dependence, a space-discretization approach is normally used to insert corona
branches [2-4]. The present thesis also follows the space-discretization approach, and
an efficient method to solve the nonlinear corona branches is presented. Fig. 6.1
illustrates the schematic explanation of the space discretization, and a line model is
discretized into N sections in the figure. Each discretized portion of the line model is
also represented by a phase-domain IARMA line model proposed in Chap. 4. Thus, the

line
線路モデルmodel
#1 #2 #N

コロナ
corona
ブランチ
branch

Fig. 6.1 Discretization of line model and insertion of corona branches

6-2
complete frequency dependence, including the frequency dependence of its modal-
transformation matrices, is taken into account. In Chap. 2, frequency-dependent and
nonlinear wave equations describing the complete dynamics of a transmission line are
derived in eq. (2.12). Because eqs. (2.12) (a) and (b) are already represented by the
phase-domain IARMA line model which represents each line section here, the rest of the
equations (c) and (d) are now represented by a nonlinear capacitance of which the value is
calculated by a wave-steepness-dependent corona model proposed in the previous
chapter. This is much easier to be understood by comparing Figs. 2.7 and 6.1. In Fig.
2.7, the linear part, enclosed by a broken line, is represented by the phase-domain
IARMA line model, and the nonlinear part, enclosed by another broken line, is
represented by the wave-steepness-dependent corona model. It should be noted that the
proposed discretization shown Fig. 6.1 does not have corona branches at both the ends so
that the equivalent circuit seen from an EMTP-type program, to which the proposed line
model is implemented, becomes linear. This considerably helps the implementation.
The corona branches are internally solved in the line model separated from the EMTP-
type program as follows.

6.1.2 Norton equivalent circuit of linear portions of line model

Consider an n-phase line. Fig. 6.2 illustrates k-th discretization point to which
a corona branch ∆C(v) is connected, where ∆C(v) represents the increase of capacitance
due to corona per section length. Note that this ∆C(v) is equal to one in eq. (2.12)
multiplied by section length ∆l = l / N. Each element of ∆C(v) can be calculated as
described in Sec. 5.6.2. Because the equivalent circuit of the phase-domain IARMA line
model illustrated in Fig. 4.2 (b) consists of a conductance matrix Y00 and a past-history
current-source vector J’, the equivalent circuit of the corona branch including the
lefthand and righthand line models becomes Fig. 6.3 (a). The equivalent circuit can
further be simplified as shown in Fig. 6.3 (b), where Y00t = 2Y00 and J’t = J’2L + J’1R
respectively represent the total conductance matrix and the total past-history current-

6-3
node k
lefthand righthand
line model … line model

corona
branch
∆C(v)

Fig. 6.2 Corona branch connected to a discretization point

lefthand line model node k righthand line model node k


… … … … … … … …

Y00 ∆C(v) Y00 Y00t ∆C(v) vk

J’2L J’1R J’t

(a) (b)

Fig. 6.3 Equivalent circuits at a discretization point

6-4
source vector of the lefthand and righthand line models.

6.1.3 State-space equation of corona branch

The circuit equation of Fig. 6.3 (b) is

∂v k
∆C ( v k ) + Y00t v k = J t′
, (6.1)
∂t

where vk is the node-voltage vector at node k.

In the absence of corona, i.e. ∆C(vk) = 0, the above equation can immediately be solved
as :

v k = Y00− 1t J t′
. (6.2)

In the presence of corona, eq. (6.1) is modified into the following form :

∂v k
 = f ( t , v k ),
 ∂t (6.3)
 f ( t , v k ) = ∆C − 1 ( v k )( J t′− Y00t v k )

The above equation is a vector, nonlinear, first-order, and partial-differential equation,


and thus in the nonlinear state-space equation form. In order to solve the state-space
equation, a number of methods have been proposed (for example, see ref. [5]). In this
thesis, the above equation is solved by the trapezoidal rule of integration together with
the predictor-corrector method as described in the following section.

6-5
6.1.4 Numerical solution by trapezoidal rule of integration and predictor-corrector
scheme

In order to solve the nonlinear state-space equation, the trapezoidal rule of


integration is used together with the predictor-corrector method. Applying the
trapezoidal rule of integration to eq. (6.3), the following implicit formula is obtained :

∆t
v ( t ) = v ( t − ∆t ) + { f ( t − ∆t , v ( t − ∆t )) + f ( t , v ( t ))}, (6.4)
2

where subscript k is omitted for simplicity.

Because the value of f(t,v(t)) cannot be obtained at time t (implicit), the numerical
integration is impossible, and a further modification is required. The predictor-corrector
method modifies the above into the following recursive formula :

∆t
v ( l ) ( t ) = v ( t − ∆t ) + { f ( t − ∆t , v ( t − ∆t )) + f ( t , v ( l − 1) ( t ))}, (6.5)
2

where l = 1, 2, … is the number of iteration.

Because the above formula cannot start by itself, the forward Euler rule :

v ( 0 ) ( t ) = v ( t − ∆t ) + ∆tf (t − ∆t , v ( t − ∆t )) (6.6)

is usually used to obtain the initial guess of the iteration scheme of eq. (6.5). Eq. (6.6)
predicts the present solution only from the information available at the previous time step,
and thus called a predictor. Then, eq. (6.5) recursively improves the predicted solution,
and thus a corrector. In the above iteration procedure, when the maximum difference of
an improved solution from the previous iteration step becomes smaller than a user
specified error constant ε, namely,

6-6
max ∆v i( l ) = max v i( l ) − v i( l − 1) < ε ( i : element index), (6.7)
i i

then v(l)(t) is regarded as the solution, and we now proceed to the next time step.

In an EMTP-type program, the trapezoidal rule of integration is normally used in


order to solve the linear part (R-L-C) of a circuit [6]. Thus, the use of the trapezoidal
rule of integration enables to solve the nonlinear corona branches with the same degree of
accuracy as the linear part. The advantage of the use of the predictor-corrector method
is avoiding the Jacobian calculation of iteration. The most common Newton-Raphson
iteration scheme requires the Jacobian calculation in spite of its high rate of convergence.
If the Newton-Raphson method is applied to eq. (6.4), then the following Jacobian matrix
J has to be calculated for the iteration :

∂Fi
J ij = ,
∂v j
(6.8)
∆t
F = v( t − ∆t ) + { f ( t − ∆t , v ( t − ∆t )) + f ( t , v ( t ))} − v ( t ).
2

Because the proposed corona model is not only nonlinear but also wave-steepness-
dependent, the calculation of the above partial derivative is difficult and numerically
unstable. If the above Jacobian matrix is inaccurate, the iteration does not converge.
Therefore, the predictor-corrector method, which requires no Jacobian calculation, is far
more numerically stable, and thus more advantageous than the Newton-Raphson method
in this particular case.

A more generalized predictor-corrector approach to solve an arbitrary nonlinear


circuit has been developed by the author [7], and the theory is briefly described in
Appendix B.

6-7
6.2 Effects of Corona on Wave Propagation

Fig. 6.4 shows the configuration of a single-phase line. By means of wave-


propagation simulations of this simple line, the effects of corona on wave propagation are
investigated. In the simulations, the frequency dependence and the nonlinearity due to
the skin and corona effects are taken into account using the proposed line model. The
line length is 10 [km], and the line model is discretized into 10 sections to insert 9 corona
branches. Fig. 6.5 shows q - v curves of the line for positive and negative polarities.
The curves calculated by the proposed wave-steepness-dependent corona model are
plotted using solid lines, and those without the wave-steepness parameters are plotted
using broken lines. The wave-steepness-dependent parameters, given in Table 6.1,
modify the curves, and the statistical inception delay τd especially increases the actual
inception voltage. Each of the corona branches inserted shows this q - v characteristic.

1.5 cm
conductor
resistivity
ρc = 0.02 µΩ m
5m

earth resistivity ρe = 100 Ω m

Fig. 6.4 Conductor configuration of a single-phase line

6-8
6

5
(a) positive
charge, µC/m
4

1 with wave-steepness dependence

without wave-steepness dependence


0
0 200 400 600
voltage, kV

5
(b) negative
charge, µC/m

1 with wave-steepness dependence

without wave-steepness dependence


0
0 200 400 600
voltage, kV

Fig. 6.5 q - v curves of single-phase line

6-9
Table 6.1 Corona model parameters

m Ec+ Ec− τd vst τi

0.6 500 kV/m -1,100 kV/m 0.5 µs 0.9 m/µs 0.5 µs

Fig. 6.6 shows a traveling wave long the line, when a 500-kV, 2.5/60 µs surge voltage is
applied at the sending end. Time step used in the simulation is 0.05 [µs]. A wave-
delay is observed above the corona onset voltage as described in Sec. 5.4 (see Fig. 5.34),
and the traveling wave at 10 [km] shows more delay than at 5 [km]. This indicates that
the wave delay becomes larger as propagating. Because the number of the discretization
is only 10, small oscillations are observed in the calculated waveforms due to reflections
at discretization points. The sudden change of the capacitance from a large value to the
geometrical value, when the applied voltage begins decreasing, may also cause the
oscillations. In comparison of Fig. 6.6 (a) and (b), it is clear that the effects of corona is
smaller in the negative polarity case, and the wave delay is observed to be also small.
Fig. 6.7 shows a traveling wave calculated without the corona branches, but the
frequency dependence is considered. It is clear that the waveforms without corona do
not show the wave delay, and thus the inclusion of corona is quite important when an
applied voltage exceeds the corona-onset voltage.

In order to show the effects of the wave-steepness-dependent parameters on


wave propagation, Fig. 6.8 shows calculated results of wave propagation with and
without the wave-steepness dependence. When an applied voltage is relatively small
with respect to the corona-onset voltage, not significant difference is observed as shown

6-10
600
applied
5 km
10 km
400
[kV]
voltage

200

0 20 40 60 80 100
time [µs]
(a) positive polarity

-600
applied
5 km
10 km
-400
[kV]
voltage

-200

0 20 40 60 80 100
time [µs]
(b) negative polarity

Fig. 6.6 Wave propagation considering skin and corona effects

6-11
600
applied
5 km 10 km
400
[kV]
voltage

200

0 20 40 60 80 100
time [µs]

Fig. 6.7 Wave propagation with skin effects (without corona)

in Fig. 6.8 (a). But when the applied voltage is large enough, then a significant
difference is observed as shown in Fig. 6.8 (b). With the dependence, the actual
inception voltage becomes higher due to the statistical inception delay τd. The streamer

velocity vst and the relaxation time of ionization τi suppress the space-charge generation,
i.e. suppress the corona effects. This agrees to the q - v curves shown in Fig. 6.5. In
ref. [8], A. Inoue pointed out that the wave delay calculation by C.F. Wagner et al. [9]
becomes too large, compared with his field-test results. This is quite interesting,
because the calculated results by the present wave-steepness-dependent corona model
show a shorter wave delay than those without the dependence. The wave-steepness
dependence may explain the wave delay quantitatively.

6-12
600
applied
5 km
10 km
400
[kV]
voltage

200

0 with wave-steepness dependence


without wave-steepness dependence

0 20 40 60 80 100
time [µs]
(a) applied voltage = 500 kV

applied
1000 5 km
10 km
[kV]
voltage

500

with wave-steepness dependence


0 without wave-steepness dependence

0 20 40 60 80 100
time [µs]
(b) applied voltage = 1,000 kV

Fig. 6.8 Effects of wave-steepness-dependent parameters

6-13
6.3 Wave-Propagation Simulations

6.3.1 Single-phase case

Fig. 6.9 shows the conductor configuration of the 450-kV Shiobara test line of
the Central Research Institute of Electric Power Industry (CRIEPI). Inoue carried out a
number of field tests using this line [8]. Because he resulted that the increase of
electrostatic coupling due to corona was less than 5 % and negligible in this line, a
single-phase test using conductor ⑦ is simulated by the proposed line model. The
conductor is an ACSR of which the radius is 12.65 [mm], and the height is 22.2 [m].
The earth resistivity is assumed to be a standard value 100 [Ω m], and the corona model
parameters are given in Table 6.2. The length of the line is 1,410 [m], and voltages were
recorded at 352.5, 705, and 1057.5 [m] from the sending end. In the simulation, the line
model is discretized into 12 sections to insert corona branches. Figs. 6.10 (a) and (b)

④ ⑤

③ ⑥

② ⑦

① ⑧ 22.2 m

Fig. 6.9 Conductor configuration of Shiobara test line of CRIEPI

6-14
Table 6.2 Corona model parameters

m Ec+ Ec− τd vst τi

0.6 400 kV/m -1,100 kV/m 0.01 µs 0.9 m/µs 0.01 µs

show the calculated and the measured voltage waveforms respectively. The calculated
results agree well with the measured results.

6.3.2 Three-phase case

Although not many field data of wave propagation with corona on a multiphase
line are available, M. Ouyang and P.G. Kendall presented field data obtained on a 33-kV
wood-pole line [10]. Fig. 6.11 shows the conductor configuration of the three-phase
line. The conductors (7/.166 in. strand, hard-drawn copper cable) are arranged in
horizontal configuration with 1.07 [m] spacing between adjacent conductors and have an
average height 8.84 [m] above the ground. The earth resistivity was measured to be 50
[Ω m]. Impulse voltages were applied to phase a using an impulse generator, and
waveforms of traveling waves were measured. The line model is discretized into 10
sections to insert three-phase corona branches, and waveforms at 5.55 [km] from the
sending end are calculated. Fig. 6.12 (a) shows the calculated results, and Fig. 6.12 (b)
the measured results. The parameters of the corona model is given in Table 6.3. It
should be noted that the proposed line model reproduces the waveforms not only on the

6-15
2000

1500 applied
voltage, kV

352.5 m 705 m
1000
1057.5 m

500

0
0 1 2 3
time, microseconds
(a) calculated result

(b) field-test results


Fig. 6.10 Wave propagation on single-phase line

6-16
1.07 m

a b c

hard-drawn copper
cable
r0 = 2.1 mm
8.84 m

ρe = 50 Ω m

Fig. 6.11 Conductor configuration of 33-kV wood pole line

Table 6.3 Corona model parameters

m Ec+ Ec− τd vst τi

0.6 800 kV/m − 1,100 kV/m 1.0 µs 0.3 m/µs 0.5 µs

6-17
applied phase but also on the induced phases. Therefore, this simulation validates that
the proposed line model can be used in the general multiphase case.

6-18
300
phase a sending end
5.55 km
kV

200
voltage,

100

0
0 10 20 30 40
time, microseconds
120
100
kV

80
60
voltage,

40
20 phase b
0
-20
-40
20 30 40
time, microseconds
100
kV

80
60
40
voltage,

20
0 phase c
-20
-40
20 30 40
time, microseconds

Fig. 6.12 (a) Calculated result of wave propagation on 33-kV wood pole line

6-19
Fig. 6.13 (a) Measured result of wave propagation on 33-kV wood pole line

6-20
References of Chapter 6

[1] J.L.Naredo, A.C.Soudack, J.R.Marti, "Simulation of transients on transmission lines with corona
via the method of characteristics," IEE Proc.-Gener. Transm. Distrib., Vol. 142, No. 1, pp. 81-
87, 1995.
[2] H. Wei-Gang and A. Semlyen, “Computation of Electromagnetic Transients on Three-Phase
Transmission Lines with Corona and Frequency Dependent Parameters, “ IEEE Trans. on Power
Delivery, Vol. PWRD-2, No. 3, pp. 887-898, 1987.
[3] S. Carneiro Jr., H.W. Dommel, J.R. Marti, and H.M. Barros, “An Efficient Procedure for the
Implementation of Corona Models in Electromagnetic Transients Programs,” IEEE Trans. om
Power Delivery, Vol. 9, No. 2, pp. 849-855, 1994.
[4] H.M. Barros, S. Carneiro Jr., and R.M. Azevedo, “An Efficient Recursive Scheme for the
Simulation of Overvoltages on Mutiphase Systems under Corona,” IEEE PES Summer Meeting,
94 SM 468-9 PWRD, 1994.
[5] R.A. Rohrer, Circuit Theory : An Introduction to the State Variable Approach, McGraw-Hill,
1971.
[6] H.W. Dommel, “Digital Computer Solution of Electromagnetic Transients in Single- and Multi-
phase Networks,” IEEE Trans., Power Apparatus and Systems, Vol. PAS-88 (4), pp.388-398,
1969.
[7] T. Noda, K. Yamamoto, N. Nagaoka, and A. Ametani, “A Predictor-Corrector Scheme for Solving
a Nonlinear Circuit,” Proc. of International Conference on Power Systems Transients ‘97 (IPST),
Seattle, Washington USA, 1997. (provisionally accepted)
[8] A. Inoue, “Study on Propagation Characteristics of High-Voltage Traveling Waves with Corona
Discharge,” CRIEPI Report, No. 114. (in Japanese: 井上,「コロナ放電を伴う高電圧進行波
伝搬特性に関する研究」,電力中央研究所総合報告,114.)
[9] C.F. Wagner and B.L.Lloyd, “Effects of Corona on Traveling Waves”, AIEE Trans. on Power
Apparatus and Systems, Vol.74, part Ⅲ, 1955.
[10] M. Ouyang and P.G. Kendall, “Tests on Distortion and Attenuation of Waves on an Overhead
Line,” IEEE Trans. on Power Apparatus and Systems, Vol. PAS-94, No. 2, pp. 498-507, 1975.

6-21
7 CONCLUSIONS

The present thesis has proposed an accurate transmission line model for power-
system transient analysis. The proposed model takes into account the complete
frequency dependence due the skin effects of conductors and earth soil, and also the
nonlinearity due to corona discharge along the conductors.

In Chapter 2, a set of frequency-dependent and nonlinear wave equations


describing the dynamics of a transmission line has been derived as the basis of the
subsequent detailed modeling of the frequency dependence and the nonlinearilty. In the
proposed wave equations, the frequency dependence and the nonlinearity are separated
from each other, and this enables to model both the characteristics independently in the
general multiphase case.

In Chapter 3, a simple and accurate closed-form formula for calculating the


frequency-dependent series impedance of an overhead line has been derived using a
double-exponential approximation. Compared with the rigorous Carson-Pollaczek’s
infinite integral evaluated by a numerical integration with computation expense, the
proposed formula shows a good accuracy with maximum error 1% in practical conductor
configurations. Because the proposed formula is in a closed-form (logarithmic function
form), it generates no numerical discontinuities unlike Carson-Galloway’s series and
asymptotic expansions connected at a certain frequency. The numerical tests of the
proposed formula have been carried out in various conditions, and the results show that
the proposed formula is far more accurate than Dubantan-Deri’s approximate formula.
As a physical representation, it has been made clear that the proposed double-exponential
approximation represents the earth-return current distribution by two complex ground-
return planes to gain more accuracy than Deri’s approximation which uses one complex
ground-return plane.

7-1
Chapter 4 has described the time-domain realization of the frequency
dependence in a transient calculation, and a phase-domain IARMA line model has been
developed. Because the proposed line model is directly formulated in the phase domain,
the modal transformation, of which the inclusion of the frequency dependence has caused
numerical instability due to mode crossing in the practical implementation of a modal-
domain model, itself is avoided. Thus, the proposed line model takes into account the
complete frequency dependence in the phase domain. In order to accelerate the line
model, an IARMA (Interpolated AutoRegressive Moving-Average) convolution has been
developed in terms of fitting, time-step determination, modeling of modal traveling-time
differences, stability, approximation order, and fitting reduction by conductor-
configuration symmetry. The IARMA convolution allows each frequency-dependent
element in the line model to use its own economical time step, and thus an efficient
computation is achieved. A steady-state initialization method of the proposed line
model has also been developed to make feasible a transient calculation starting from a
steady-state, e.g. a fault-surge calculation. A set of matrix criteria, which assures that
the line model behaves as a passive circuit and thus a numerically stable model, has been
derived for a stable transient calculation. In order to validate the proposed line model,
various switching and fault-surge calculations have been carried out, and the results agree
well with rigorous solutions and field-test results. It has been shown that the proposed
line model is computationally efficient compared with the most widely used modal-
domain model (J.Marti model) which assumes constant real transformation matrices.
Also, it has been shown that the proposed line model has an improved accuracy compared
with the J.Marti model in a fault-surge calculation which requires a good reproduction of
the frequency-dependent modal-transformation matrices from the steady-state frequency
to a surge frequency region.

In the first half of Chapter 5, a novel charge-voltage (q - v) curve measurement


method has been developed to make clear the mechanism of corona discharge along
transmission wires, and various field data have been accumulated. Because the
proposed method requires no charge-measuring conductor such as a corona cage, a q - v
curve is measured under the real electric-field distribution of a transmission-line

7-2
configuration. The effects of line length on the accuracy of the q - v curve measurement
has been investigated, and a length which gives a good condition of the measurement has
been determined. Also, the effects of conductor radius and the presence of an adjacent
conductor have experimentally been investigated. The results show that the q - v curve
of a thin conductor is different from a thicker one, and the wave-steepness dependence of
the thin conductor is smaller. It has been demonstrated that an adjacent conductor, such
as a corona cage, causes an error in a q - v curve. Coupling-factor curves, representing
the increase of electrostatic coupling between two conductors due to corona, have also
been measured and the effects of conductor separation and radius have been investigated.

In the latter half of Chapter 5, a wave-steepness-dependent corona model has


been developed. The proposed model is based on the concept of a quasi-static state of
corona which is being proposed in this thesis, and statistical inception delay, streamer
development process, and ionization process are introduced to express the wave-
steepness dependence. The measured q - v and coupling-factor curves have successfully
been reproduced by the proposed model. Also, q - v curves measured by Maruvada et
al., which show heavy wave-steepness dependence, have accurately been reproduced.
Therefore, it has been validated that the proposed corona model can be used for both
lighting and switching-surge studies. As a future work, the theoretical identification of
the wave-steepness parameters: statistical inception delay, streamer velocity, and
relaxation time of ionization, may be important, in order to make the proposed line model
a fully theoretical simulation model.

Chapter 6 has described the realization of the nonlinearity due to corona in a


time-domain simulation. In the proposed method, the line model is discretized in space,
and nonlinear corona branches, each of which is the wave-steepness-dependent corona
model, are inserted at the discretization points. A nonlinear state-space equation
describing the dynamics of the corona branches in the discretized line model has been
derived by use of Norton equivalent circuit of the linear portions of each linear line
section. The trapezoidal rule of integration combined with the predictor-corrector
iteration scheme is used to solve the state-space equation efficiently and stably. Using

7-3
the proposed method, wave-propagation transients have been calculated including both
the skin and corona effects, and the results show a good agreement with field tests.
Also, the effect of the wave-steepness dependence on wave propagation has been made
clear by the simulation.

The proposed line model has been implemented in the ATP version of EMTP,
which is widely used in the world, and is expected to be used in practical insulation-
design studies. As a future work, a number of practical transient calculations may be
required to make the present line model a robust simulation model which can be used in
real industrial simulations.

7-4
APPENDIX A

LINEARIZED LEAST-SQUARES METHOD

This appendix gives a detailed description of the Linearized Least-Squares


(LLS) Method. The matrix form of eq. (4.38) for k = 1, 2, · · ·, K is

e = Ax − b (A.1)

where

e = (err'1 , err' 2 , L , err' K )T (A.2)


− j

z k (1 ≤ j ≤ M + 1)
Akj =  (A.3)
− j+ M + 1

− z k Gk ( M + 2 ≤ j ≤ M + N + 1)

x = ( a 0 , a1 , L , a M , b1 , L , bN )T (A.4)

b = ( H1 , H 2 , L , H K )T (A.5)
e : error vector ; k-th order complex vector
A : design matrix; ( K , M + N + 1) complex matrix
x : parameter vector ; ( M + N + 1) -th order real vector
b : response vector; K-th order complex vector

Because the error vector e is complex, equation (A.1) is rewritten to be real.

A-1
Re{err'1 }
 
 M 
Re{err' K } Ar  br 
r =  =  x −   ⇔ r = Cx − d (A.6)
Im{err'1 } Ai  bi 
 M 
 
Im{err' K } 

where r : residual vector ; 2K-th order real vector

A = A + jA , b = br + jbi
r i

A 
C =  r  : modified design matrix ; (2K, M + N + 1) real matrix
Ai 

br 
d =   : modified response matrix ; 2K-th order real vector
bi 

Performance index which indicates a sum of square errors is expressed as

K
J ' ( x) = r T r = ∑ | err ' k |2 (A.7)
k =1

When J’(x) is minimized, the least-squares solution of eq. (A.6) is obtained. The well-
known normal equation is obtained under the condition ∂J' ( x ) ∂x = 0 .

C T Cx = C T d (A.8)

This is an equation to be solved. But the condition number of C T C has a tendency to


be large, in other words, bad. Thus, there are many numerical difficulties to solve eq.
(A.8). To avoid these numerical difficulties, the modified design matrix C is
decomposed.

A-2
C = QR (A.9)

where Q : orthogonal matrix, R : upper triangle matrix

Substituting eq. (A.9) into eq. (A.8) gives the following modified normal equation.

Rx = Q T b (A.10)

For R is an upper triangle matrix, this equation can be solved only by backward
substitution. The condition number of R is smaller than that of CTC, thus the solution x
of equation (A.10) is much more accurate than that of equation (A.8). For
decomposition from C into QR, famous Householder's transformation is used. It is very
accurate and fast. Modified Gram-Schmidt's transformation is also famous, but it is
proved that their accuracy and computational speed are equal theoretically. As a result,
the linearization of the error function and the application of Householder's transformation
make calculations accurate and fast.

From eq. (4.37), a performance index of the nonlinear LS method is


expressed as

2
K
a + a1 z k− 1 + ⋅⋅⋅+ a M z k− M
K
J ( x ) = ∑ |errk | = ∑ 0 2
−1 −N
− Gk (A.11)
k =1 k =1 1 + b1 z k + ⋅⋅⋅+ bN z k

The performance index J(x) generates many local minimum poles as illustrated in Fig. A.1
(a), because denominator parameters b1,⋅⋅⋅, bN are included as nonlinear parameters.
Thus, assurance of convergence into the optimum value highly depends on an initial value.
While, from equation (4.38), a performance index of the linearized LS method is
expressed as

K K
J ' ( s) = ∑ | err'|2 = ∑ {(Re{err' k })2 + (Im{err' k })2 } (A.12)
k =1 k =1

A-3
For the parameters are real numbers, the modified performance index J'(x) is a second
order polynominal of all the parameters, and coefficients of the second order terms are
positive. Consequently, the modified performance index J'(x) generates only one
minimum pole as shown in Fig. A.1 (b), and a true LS solution is always identified. this is
why the calculation process of the linearized LS method is stable and fast, and it requires
no iterative calculations.

Depending on characteristic of a pulse transfer function approximated by the


linearized LS method, there are better cases using another error function instead of eq.
(4.38). In a case that amplitude of given frequency characteristic (ω k , Gk ) is highly
different between pass region and cut region correctly. To evaluate relative error, eq.
(A.1) is solved with replacing A with A’= WA and b’= Wb. The solution is

R 'x = Q ' T b (A.13)

where C ' = Q ' R ' : modified design matrix of A’, W = diag(wk) : weighting value
matrix ; diagonal matrix, wk : weighting value for the k-th datum. For the matrix W’is a
upper triangle matrix, this equation can be solved with the same manner of equation
(A.10).

A-4
Fig. A – 1 Comparison of performance index

A-5
APPENDIX B

A PREDICTOR-CORRECTOR SCHEME
FOR SOLVING A NONLINEAR CIRCUIT

Summary

This appendix presents a predictor-corrector iteration scheme which extends EMTP’s


nodal-conductance approach to include arbitrary number and configuration of nonlinear ele-
ments in a subnetwork. Because the proposed method avoids the Jacobian calculation of
iteration by use of the predictor-corrector scheme, any nonlinearity can be included in a
simulation by its own suitable representation. Also, the clear logic of the trivial automatic
formulation of circuit equations is available in the method. Calculated results show that the
method is stable for a large time step and also for simultaneous abrupt changes of the
characteristics of two or more nonlinear elements.

B-1
B.1 Introduction

For transient calculations of a power system, two major formulation


approaches : 1) nodal-conductance approach (NCA) and 2) state-variable approach
(SVA) were proposed. The NCA was originally proposed by H.W. Dommel [1], and
the Electro-Magnetic Transients Program (EMTP) has been developed based on this
approach. On the other hand, the SVA has been used in the field of electronic circuit
analysis [2] and was first applied to power system transient studies by S.N. Talukdar [3].

Transient calculations of a nonlinear circuit have become more important,


because modern power systems such as an HVDC (high voltage dc) system and a FACTS
(flexible ac transmission system) include a number of nonlinear devices. To deal with
nonlinear elements in a circuit, a compensation method was added to the NCA [4].
Although the compensation method is computationally efficient, it cannot deal with
certain circuit topology such as a case that nonlinear elements create a floating
subnetwork in the Thévenin equivalent calculation procedure. The compensation
method usually uses the Newton-Raphson iteration to obtain the solution of a set of
nonlinear circuit equations, when given nonlinear characteristics are expressed analyti-
cally, or when two or more nonlinear elements to be solved individually, even if each of
them is expressed as piece-wise linear, are included in a subnetwork (disconnected by
distributed-parameter lines). But the Newton-Raphson iteration requires the Jacobian of
the nonlinear elements, which is sometimes difficult to be evaluated. On the other hand,
the SVA is inherently able to deal with nonlinear elements, but the non-trivial automatic
formulation of state equations is remained unsolved. Thus, the trivial automatic
formulation of nodal equations is one of the advantages of the NCA, compared with the
SVA. Sophisticated hybrid approaches combining NCA and SVA have recently been
proposed [5,6].

Rather than purposing such the sophisticated approaches, the proposed scheme

B-2
extends the NCA to allow an arbitrary number of nonlinear elements in a subnetwork
based on a predictor-corrector iteration scheme. By extending the NCA rather than
combining NCA and SVA, the clear logic of the trivial automatic formulation is available
as well as the original NCA, and nonlinear elements can be connected to any branches
unlike the compensation method. The present method avoids the Jacobian calculation of
iteration by use of the predictor-corrector scheme. Thus, any nonlinearity can be
included without evaluating its Jacobian, and each nonlinear element can be described
separately in its own suitable form. Also, the method can easily be implemented in an
existing EMTP-type program. Transient calculations of a single-phase diode-bridge
rectifier circuit, a transistor switching circuit, and a nonlinear transformer model are
demonstrated in this appendix. According to the calculated results, the method is stable
even for a large time step, and properly converges even when two or more nonlinear
elements rapidly change their characteristics simultaneously (e.g. simultaneous switching
of power devices).

B-3
B.2 Formulation of Nonlinear Circuit

A. Optimum ordering of nodes

As described above, the clear logic of the trivial automatic formulation of nodal
equations is one of the advantages of the NCA, and the equations of any circuit topology
are expressed in the following matrix form, when all the elements in the circuit are linear :

G v ( t ) = J (t ), (B.1)

where G : nodal-conductance matrix, v(t) : node voltage vector, and J(t) : current-
injection vector. Each element in the circuit is discretized by the trapezoidal rule of
integration. When all the elements are linear, G remains constant and J(t) is time
varying. Therefore, to save computation time, the triangular factorization of G is
performed only once preceding the time-step loop, and v(t) is calculated by the backward
substitution at each time step. When a circuit includes nonlinear elements, then G
becomes dependent on an instantaneous solution v(t), that is, the triangular factorization
of G is required during an iteration procedure at each time step. Even in this case, the
computation of the triangular factorization can be minimized by ordering nodes as
illustrated in Fig. B.1 : from inner to outer, linear nodes without switches, linear nodes
with switches, nonlinear nodes without switches, and nonlinear nodes with switches.
(linear node : only linear elements are connected, nonlinear node : one or more nonlinear
elements are connected.) At each time step, only the portion of the nonlinear nodes is
factorized during an iteration procedure. When one or more switches operate, then the
portion of the linear nodes with switches is also factorized. The portion of the linear
nodes without switches remains unchanged through out a simulation. This ordering
approach is closely related to one proposed for the compensation method in [4].

B-4
linear nodes nonlinear nodes

linear nodes
without switches
triangular factorization
when switch operation
linear nodes triangular factorization
with switches at each time step
nonlinear nodes
without switches
nonlinear nodes
with switches

Fig. B.1 Nodal-conductance matrix G

B. Predictor-corrector iteration

When a target circuit includes nonlinear elements, G becomes dependent on an


instantaneous solution v(t), and an iteration procedure is required to find the solution of
the following nodal-conductance equation :

G ( t , v ( t )) v ( t ) = J ( t ). (B.2)

The present method uses a predictor-corrector scheme for the iteration, because it does
not require the Jacobian, which is sometimes difficult to be evaluated. The solution of
the following equation v(0)(t) gives the first estimation of the iteration (= predictor) :

G ( t , v ( t − ∆t )) v ( 0 ) ( t ) = J ( t ). (B.3)

It should be noted that v(0)(t) is different from the solution at the previous time step,

B-5
because J(t) has already been updated in (B.3). Then, improved solutions are
recursively obtained by the following iteration scheme (= corrector) :

G ( t , v ( k − 1) ( t )) v ( k ) ( t ) = J ( t ), (B.4)

where k = 1, 2, … is the number of iteration. When the maximum difference of an


improved solution from the previous iteration step becomes smaller than a user specified
error constant ε, namely,

max ∆v i( k ) = max v i( k ) − v i( k − 1) < ε (i : node index), (B.5)


i i

then v(k)(t) is regarded as the solution, and we now proceed to the next time step. It
should be noted that the proposed scheme requires no restriction on the number and
configuration of nonlinear nodes, and thus arbitrary number of nonlinear elements can be
connected to any branches. Also, the proposed scheme does not modify the basic
equation of the NCA, the trivial automatic formulation of (B.2) is still available as well as
the original NCA. Those features facilitate the implementation of the proposed scheme
in an existing EMTP-type program. Any integration rule can be used to discretize each
element, but the trapezoidal rule may be preferred in terms of accuracy, speed, and
stability.

C. Relation with ODE formulation

The predictor-corrector scheme described above starts an iteration with a


solution predicted from the information available at the previous time step, and then
simply improves the predicted solution recursively. The scheme seems different from
the standard predictor-corrector scheme of ordinary differential equations (ODEs) [2].
But it can be shown that the present scheme is mathematically the same as the standard
scheme as follows. Consider a set of nonlinear ODEs (state equations) describing the

B-6
dynamics of a circuit :

dx ( t )
= f ( x ( t ), u( t ), t ) (B.6)
dt

where x(t) : state-variable vector and u(t) : input vector. Applying the trapezoidal rule
of integration, we obtain the following implicit formula :

∆t
x (t ) = x( t − ∆t ) + { f ( x( t − ∆t ), u(t − ∆t ), t − ∆t ) + f ( x ( t ), u( t ), t )}.
2
(B.7)

The standard predictor-corrector scheme modifies the above into the following recursive
formula :

∆t
x ( k ) ( t ) = x( t − ∆t ) + { f ( x( t − ∆t ), u(t − ∆t ), t − ∆t ) + f ( x ( k − 1) ( t ), u( t ), t )},
2
(B.8)

where k = 1, 2, … is the number of iteration. Because the above formula can not start
by itself, the forward Euler rule :

x ( 0 ) ( t ) = x ( t − ∆t ) + ∆tf ( x( t − ∆t ), u( t − ∆t ), t − ∆t ) (B.9)

is usually used to obtain the initial guess of the iteration. Equation (B.9) predicts the
present solution only from the information available at the previous time step, and thus
called a predictor. Then, (B.8) recursively improves the predicted solution, and thus a
corrector. On the other hand, (B.3) predicts the present solution, and (B.4) recursively
improves the predicted solution, in the same manner as (B.9) and (B.8) respectively.
Therefore, the proposed scheme can be regarded as an NCA version of the predictor-
corrector method.

B-7
D. Comparison with compensation method

In the compensation method, we first have to build the Thévenin equivalent of


the linear portion of a circuit seen from nonlinear branches. The voltage source of the
Thévenin equivalent is calculated by replacing the nonlinear branches with open-circuit
branches. Consider an example circuit illustrated in Fig. B.2(a). When the Thévenin
equivalent of the circuit is calculated, the nonlinear elements are disconnected as shown in
Fig. B.2(b), and a floating subnetwork is created. The voltages of node N1 and N2
cannot be calculated. Fig. B.2(c) illustrates another example. When both the nonlinear
inductance and resistance are disconnected, the voltage of node N2 becomes unknown as
shown in Fig. B.2(d). In the former case, a large resistance may be put in parallel with
each nonlinear element, but this slightly modifies the solution. In the latter case, if both
the nonlinear elements are modeled as one element, then the voltage of node N2 is not
necessary to be calculated. But this considerably reduces the maintainability of each
model in terms of the modularity of a model. On the other hand, the proposed method

SRC N2 SRC ? N2

N1 N1 ?
∼ ∼

(a) (b)
N1 N3 N1 ? N3

N2 N2

(c) (d)

Fig. B.2 Problems of the compensation method

B-8
can deal with those cases without any modification. Each model can be developed
independently, and a model can easily be used for other purposes. Also, the avoidance
of the Jacobian calculation enhances the general describability of a model. Therefore,
the proposed method may show a good performance together with a model description
language such as the MODELS language [7].

E. Modeling of nonlinear component

The characteristics of a nonlinear element may be described using a language


such as Fortran, C, or MODELS, or may be given as a point list representing its piece-
wise linear characteristics. Whatever description method is used, it is quite important to
make clear the types of information to be exchanged between a host program which
performs the present predictor-corrector scheme and a model describing a nonlinear
element. The host program first gives a predicted branch voltage vb(0) obtained by (B.3)
to the model. The model calculates the equivalent conductance Geq(1) and current source
Jeq(1) from the predicted branch voltage, and gives them back to the host program. Then,
the host program starts an iteration based on (B.4). During the iteration, the host
program gives an improved branch voltage vb(k), and the model gives back the conduc-
tance Geq( k + 1 ) and current source Jeq( k + 1 ) for the next iteration step in the same manner.
A certain nonlinearity requires to be expressed as a function of its branch current instead
of voltage. In this case, the branch current ib(k) is calculated from the given branch
voltage vb(k) as :

ib( k ) = Geq v b + J eq
(k ) (k ) (k)
, (B.10)

and then used in order to calculate Geq( k + 1 ) and Jeq( k + 1 ) . Especially as for the v - i
characteristic of a switching device, the equivalent conductance should be calculated as a
function of v where ∂i/∂v is gentle, and a function of i where ∂i/∂v is steep, for the stable
convergence of iteration.

B-9
B.3 EXAMPLES

A. Single-phase diode-bridge rectifier circuit

A single-phase diode-bridge rectifier circuit illustrated in Fig. B.3 is analyzed


using the proposed method. Each diode in the circuit is modeled as a nonlinear
resistance of which the v - i characteristic is defined in Fig. B.4. Measured v - i char-
−3 6
acteristic is approximated by v(i) = 81.65×10 ln (51.03 × 10 i), and its equivalent
resistance RD is calculated as :

RD = v ( i ) / i (V > 0), RD = 400 kΩ (V ≤0). (B.11)

Because ∂i/∂v is considerably steep when the diode is on, the equivalent
resistance is calculated as a function of i. A startup transient is calculated, and node
voltage waveforms are shown in Fig. B.5. For this simulation, ∆t = 0.1 ms (166.7 points
/ period) may be reasonable and the calculated waveforms are plotted using solid lines.
Even when ∆t is set to 2.0 ms (8.3 points / period), the calculated results are still stable
and plotted using broken lines. It is clear that the present method is stable even for a
large time step. The calculated waveforms of the input voltage and output current are
replotted in Fig. B.6(a), and agree well with measured ones shown in Fig. B.6(b).

B-10
node 3

Ls
99.2 µ H L
D1 D2 88.9 mH
node 2
E node 1
10 V
∼ C
96.2 µ F I
R
99.4 ohm
D3 D4 node 4

Fig. B.3 Single-phase diode-bridge rectifier circuit

0.2
current (mA)

measured
0.1 approximated
0
-1 -0.5 0 0.5 1
voltage (V)
Fig. B.4 Voltage-current (v - i) characteristic of diode

B-11
10 node 1

0 10 20 30 40 50
-10
10 node 2

0 10 20 30 40 50
voltage (V)

-10
10 node 3

0 10 20 30 40 50
-10 : ∆t = 0.1 ms
10 node 4 : ∆t = 2.0 ms

0 10 20 30 40 50
-10
time (ms)

Fig. B.5 Calculated waveforms of node voltages

B-12
10 current 100

current [mA]
voltage (V)

0 0

-10 voltage -100


0 10 20 30 40 50
time (ms)

(a) calculated (startup)

10 100

current [mA]
voltage (V)

current
0 0

voltage
-10 -100
0 10 20 30 40 50
time (ms)

(b) measured (steady state)

Fig. B.6 Measured waveforms of input voltage and output current

B-13
B. Transistor switching circuit

An inverter-type transistor switching circuit is illustrated in Fig. B.7. The


circuit includes four transistors and four diodes, thus eight nonlinear elements. Each
transistor is modeled as a nonlinear resistance, and the v - i characteristic (VCE :
collector-emitter voltage versus IC : collector current, with parameter Ib : base current) is
given in Fig. B.8. On and off delays, and rising and falling time constants are considered
in the transistor model [8]. The diodes are modeled in the same manner as in the
previous example. Fig. B.9(a) shows base-current waveforms controlling the transistors.
Ib1 controls Tr1 and Tr4, and Ib2 Tr2 and Tr3. The voltage waveform across the load is
calculated by the proposed method and shown in Fig. B.9(b). The calculated result
agrees well with measured one shown in Fig. B.9(c). Although eight nonlinear elements
abruptly change their characteristics simultaneously, the proposed predictor-corrector
scheme properly converges.

Tr1 D1 Tr3 D3
98.7 µH 285 Ω
50 V

Tr2 D2 Tr4 D4

Fig. B.7 Inverter-type transistor switching circuit

B-14
8

current (A) 6 Ib = 1.0 mA


Ib = 0.8 mA
4 Ib = 0.6 mA

Ib = 0.4 mA
2
Ib = 0.2 mA

0
0 1 2 3 4 5
voltage (V)

Fig. B.8 VCE - IC characteristic of a transistor

1.5
current (mA)

Ib1, Ib2
1

0.5

0
0 0.1 0.2 0.3
time (ms)

(a) base current

Fig. B.9 Simulation of transistor switching circuit (continues to the next page)

B-15
voltage (V) 50

-50
0 0.1 0.2 0.3
time (ms)

(b) calculated voltage waveform across load

(c) measured voltage waveform across load

Fig. B.9 Simulation of transistor switching circuit

B-16
C. Transformer model

The iron-core nonlinearity of a transformer has to be included in a simulation


predicting harmonics and inrush currents. A transformer model proposed by L.O. Chua
et. al. [9] is illustrated in Fig. B.10. The iron-core nonlinearity is modeled by a nonlinear
inductance and resistance connected in parallel. Fig. B.11(a) shows a measured
hysteresis loop of an 100-V transformer. From the hysteresis loop, the nonlinearities of
the inductance and resistance are determined as shown in Fig. B.11(b) and (c)
respectively. Each of the determined nonlinearities is then approximated by a
polynomial, and used in the subsequent simulations. The winding resistance and the
leakage inductance are measured to be 22.9 Ω and 5.79 mH in the primary side. Fig.
B.12(a) shows a measured applied-voltage waveform, and Fig. B.12(b) shows measured
and calculated current waveforms. Those current waveforms show a good agreement.
The calculated result is obtained by the proposed method running a simulation until the
calculated waveform is settled in a steady-state. A startup transient is also calculated and
shown in Fig. B.13. Inrush currents are reproduced by the simulation.

Rs
r l

E sinωt ∼ L R

transformer model

Fig. B.10 Transformer model

B-17
0.4

magnetic flux (wb)


0.2

-0.2

-0.4
-40 -20 0 20 40
current (mA)

(a) hysteresis loop

0.5
measured
magnetic flux (wb)

0.4
fitted
0.3

0.2

0.1
(b) i - φcharacteristic
0
0 10 20 30
current (mA)
15
measured
fitted
current (mA)

10

(c) v - i characteristic
0
0 50 100
voltage (V)

Fig. B.11 Iron-core nonlinearity of transformer

B-18
100

voltage (V)
0

-100

0 10 20 30 40 50 60
time (ms)

(a) voltage

40
current (mA)

20

-20
measured, calculated
-40
0 10 20 30 40 50 60
time (ms)

(b) current

Fig. B.12 Voltage and current waveforms of transformer

100
current (mA)

0 100 200
time (ms)

Fig. B.13 Startup transient of transformer

B-19
B.4 Conclusions

An extension to the NCA has been presented to allow arbitrary number and
configuration of nonlinear elements in a subnetwork by use of the predictor-corrector
scheme. The trivial automatic formulation is available in the method as well as the
original NCA, and nonlinear elements can be connected to any branches unlike the
compensation method. The method avoids the Jacobian calculation of iteration which is
sometimes difficult to be evaluated. The present method has been applied to a single-
phase diode-bridge rectifier circuit, a transistor switching circuit, and a nonlinear
transformer model, and the calculated results agree well with actual measurements. The
proposed scheme has shown to be robust even for a large time step, and also for
simultaneous and abrupt changes of the characteristics of nonlinear elements.

* This chapter was reproduced from ref. [10] presented by the author.

B-20
References of Appendix B

[1] H.W. Dommel, “Digital computer solution of electromagnetic transients in single- and multi-
phase networks,” IEEE Trans., Power App. and Syst., Vol. PAS-88 (4), pp. 388-398, 1969
[2] R.A. Rohrer, “Circuit Theory: An introduction to the state-variable approach,” McGraw-Hill,
1971.
[3] S.N. Talukdar, “METAP - A modular and expandable program for simulating power system
transients,” IEEE Trans., Power Apparatus and Systems, Vol. PAS-95 (6), pp. 1882-1891, 1976.
[4] H.W. Dommel, “Nonlinear and time-varying elements in digital simulation of electromagnetic
transients,” IEEE Trans., Power Apparatus and Systems, Vol. PAS-90, pp. 2561-2567, 1971.
[5] J.M. Zavahir, J. Arrillaga, and N.R. Watson, “Hybrid electromagnetic transient simulation with
the state variable representation of HVDC converter plant,” IEEE Trans., Power Delivery, Vol.
PWRD-8 (3), pp. 1591-1598, 1993.
[6] Y. Kang and J.D. Lavers, “Transient analysis of electronic power systems: reformulation and
theoretical basis,” IEEE PES Summer Meeting, 95 SM 518-1 PWRS, 1995.
[7] L. Dubé and I. Bonfanti, "MODELS: A new simulation tool in the EMTP", European Trans. on
Electrical Power Engineering (ETEP), Vol.2, no.1, pp.45-50, January/February 1992.
[8] N. Nagaoka, S. Fujii, and A. Ametani, “An analysis of operational characteristics of high
frequency inverters by EMTP,” PESC '88 Record, pp. 731-738, 1988.
[9] L.O. Chua and K.A. Stromsmoe, “Lumped-circuit models for nonlinear inductors exhibiting
hysteresis loop,” IEEE Trans., Circuit Theory, Vol. CT-17, No. 4, 1970.
[10] T. Noda, K. Yamamoto, N. Nagaoka, and A. Ametani, “A Predictor-Corrector Scheme for
Solving a Nonlinear Circuit,” Proc. of International Conference on Power Systems Transients ‘97
(IPST), Seattle, Washington USA, 1997. (provisionally accepted)

B-21
APPENDIX C

INVESTIGATION OF LIGHTNING-INDUCED
VOLTAGES TO ELECTRONIC APPLIANCE

Summary

This appendix proposes a method for calculating a voltage induced to an electronic


appliance from a lightning stroke. A printed circuit in the electronic appliance is modeled as a
loop circuit, and the induced voltage is calculated from the interlinked magnetic flux generated
by a lightning-stroke current. An experiment is carried out to confirm the accuracy of the
proposed method, and a good agreement is obtained.

C-1
C.1 Introduction

Malfunctions of an electronic appliance caused by electromagnetic waves


invading from the environment have become significant as an ElectroMagnetic
Compatibility (EMC) problem [1]. A lightning stroke is one of the largest sources of the
invading electromagnetic waves, and is quite important to determine the induced voltages
to an electronic appliance for avoiding its malfunctions.

This appendix proposes an analytical method for determining the induced


voltage to an electronic appliance nearby a lightning stroke. A printed circuit in the
electronic appliance is modeled as a closed loop, and an induced voltage to the loop is
calculated as a differentiation of an interlinked magnetic flux generated by the lightning-
stroke current using a vector-potential approach. At first, a formula which gives the
induced voltage caused by the lightning current with a step waveform is derived, and then
the voltage caused by the lightning current with an arbitrary waveform is calculated using
a numerical convolution technique.

An experiment is carried out using a scaled model to confirm the accuracy of the
proposed method. The lightning-stroke current is modeled under the following two
assumptions. 1) The currents of a stepped leader and a final-jump streamer are
negligible, because they are much smaller than that of a return stroke. 2) The velocity of
the return-stroke current is less than the light velocity in free space because of the
discharge-path development mechanism. A special wave guide consisting of two wires
folded around a polyvinyl-chloride pipe in the opposite direction to each other is used as a
return stroke model for reducing its propagation velocity and longitudinal magnetic flux.
An induced voltage to a closed loop nearby the waveguide is measured.

C-2
C.2 Calculation Method

A. Lightning-stroke model

In a process of a lightning stroke, a stepped leader is firstly developed from the


bottom of a thundercloud to the earth, and then a return stroke with a large current and a
strong flash is triggered by a final-jump streamer from the earth to the top of the
oncoming stepped leader. Dart leaders and accompanied return strokes may follow in
the case of a multiple lightning flash. For an investigation of an induced voltage, the
return-stroke current is important, because the upward current of the return stroke is
much greater than those of the stepped leader and the final-jump streamer [2,3]. A
lightning stroke model proposed in ref. [4] is applied to the theoretical analysis of this
appendix. The return-stroke current of the model propagates along a vertical line from
the earth to the thundercloud at a velocity less than the light velocity and without
deformation. Fig. C.1 shows the model with its image current. The present analysis is
carried out under the following assumptions.

a) Only a return-stroke current is taken into account.

b) Discharge path is vertical to the earth surface.

c) Earth is perfect conductor.

d) A current waveform is a step function.

The assumption d) is not a serious limitation, because it can be generalized to an


arbitrary waveform by use of a numerical convolution technique which will be described
in Section 2.4. According to the above assumptions, the lightning current and its image
are expressed as :

C-3
return-stroke current: i1 = I 0 u( t − z '/ c' ) (C.1)

image current: i2 = I 0 u ( t + z '/ c' ) (C.2)

where I0: amplitude of discharge current, u(t): unit-step function, c': velocity of
discharge-current development.

z' = ζ i1

0 z y
r
x

z' = - ζ i2 image

Fig. C.1 Lightning-stroke model

C-4
B. Equivalent circuit of printed circuit

An electronic signal inside an electronic appliance propagates along a pattern on


a printed circuit board. When voltage differences between terminals on the circuit are
disturbed by invading electromagnetic waves, the signals are interfered. As illustrated in
Fig. C.2, a pattern on the printed circuit board usually forms a loop circuit of which area
is S. The induced voltage e due to a lightning stroke should be calculated for the signal
interference. According to the electromagnetic theory, the induced voltage e can be
decomposed into electrostatic and electromagnetic induction. The former is represented
by a scalar potential φ, and the latter by a vector potential A, respectively. The scalar
potential φ does not contribute to the induced voltage e, because both terminals are
electrostatically short-circuited by the loop. Therefore, only the vector potential A
generated by the lightning stroke current is necessary to determine the induced voltage,
and the scalar potential φ can be ignored. In other word, the effect of the absolute
voltage of the appliance with respect to the earth on the signal interference is small
because the absolute voltage just causes the rise of the ground level of the appliance.
On the contrary, a rise of a voltage difference between terminals on the printed circuit
board causes a significant effect.

The magnetic flux density B at the point of the appliance can be obtained from
the vector potential A using B = ∇ × A. The induced voltage e can be deter-mined
from the following Faraday's law.

∂φ ∂
e = − = − ( B ⋅S )
∂t ∂t
(C.3)
∂ ∂A
= − {(∇ × A)⋅S} = − ∇ ×  ⋅S
∂t ∂t

where φ: magnetic flux, S: area vector of the closed loop (|S| = S). The overvoltage
estimation should be performed in the severest condition for protecting an electronic
appliance. When the surface of the closed loop is perpendicular to the earth surface and
is parallel to the lightning path, the intensity of interlinked magnetic flux becomes the

C-5
maximum, i.e. the highest voltage condition. According to the assumption (b) in the
previous section, the vector potential A consists only of the z component.

A = (0, 0, Az)

∂2 Az
e =−S (C.4)
∂r∂t

The derivation of ∂2Az/∂r∂t will be described in the following section. Because the scale
of the electronic appliance is much smaller than that of the lightning stroke path, the term
∂2Az/∂r∂t can be assumed to be constant anywhere inside the appliance.

When the loop is terminated with an impedance Z as shown in Fig. C.3 (a), a
current flows into the loop conductor. In that case, the voltage drop due to the
inductance of the loop conductor should be considered. Fig. C.3 (b) illustrates
Thevenin's equivalent circuit of the closed loop, in which L is the inductance of the loop
conductor. The source e is given in eq. (C.4) and expresses the induced voltage when
the closed loop is open-circuited. The induced voltage v across the load Z can be
calculated by the equivalent circuit.

Z
v= e (C.5)
Z + jω L

C-6
S
B θ e

S
e

Fig. C.2 Modeling electronic appliance by a closed loop

v Z = e v Z

Fig. C.3 Thevenin equivalent of closed loop terminated with impedance Z

C-7
C. Vector potential generated by lightning stroke

As described in the previous section, ∂2Az/∂r∂t should be calculated to obtain


the induced voltage v. In this section, the vector potential A generated by a lightning
stroke is determined. According to the assumption (c) described in Section 2.1, the
effect of the earth can be taken into account by considering the image current of the
actual lightning stroke. The contribution of the actual current i1, to the vector potential
is first introduced, and then the contribution of the image current i2 is taken into account.
Because the vector potential A1z generated by the actual current i1 consists only of z
component, it is given in the following form :

µ0 ∞ i1 ( z ' , t − R / c)
4π ∫0
A1z ( P , t ) = dz ' (C.5)
R

where, c: velocity of light, P(r, z): position of the closed loop, and R = (r2 + (z −
z')2)1/2: distance between i1 and P. Considering the development velocity of the return
stroke c', the following equation is obtained by substituting eq. (C.1) into the above
equation.

µ0 I 0 ζ u( t − τ 0 )
A1z ( P , t ) =
4π ∫0 dz ' (C.6)
r 2 + (z − z' ) 2

where τ 0 = r 2 + z 2 / c , and ζ is given as the solution of t - R/c - ζ /v = 0 in the


following form:

ct − zc' c − ( z − c' t ) 2 + (1 − c' 2 c 2 )r 2


ζ = (C.7)
c / c'(1 − v 2 c 2 )

The integral of eq. (C.6) can be carried out analytically, and the substitution of eq. (C.7)
into eq. (C.6) gives:

C-8
ζ
µ0 I 0 
A1z = ln r 2 + ( z − z ' ) 2 + z '− z  u(t − τ 0 )
4π   
0
µ I r 2 + (z − ζ ) 2 + ζ − z
= 0 0 u(t − τ 0 )
4π r 2 + z2 − z

µ I  c' t − z + ( z − c' t ) 2 + (1 − c' 2 c 2 )r 2



= 0 0 ln
4π  r 2 + z2 − z
 (C.8)
c' 
− ln 1 + u(t − τ 0 )
c

The above equation expresses the contribution from the actual current i1. The term A2z
caused by the image current i2 can be calculated in the same manner. Finally, the total
vector potential Az is given in the following form :

µ I  c' t − z + ( z − c' t ) 2 + (1 − c' 2 c 2 )r 2


 c'
Az = A1z + A2 z = 0 0 ln − ln 1 +
4π  r2 + z2 − z c

(C.9)
− c' t − z + ( z + c' t ) + (1 − c' c )r
2 2 2 2 
c' 
− ln + ln 1 − u(t − τ 0 )
r 2 + z2 − z c

The derivative of the above equation with respect to r and t gives the induced voltage e
due to a step-function discharge current.

∂2 Az
e = − S
∂r∂t
− 3/2
µ I c'  c' 2  
  c' 2   
= S 0 0 1 − 2 r ( z − c' t ) 2 + 1 − 2 r 2  (C.10)
4π  c   
  c   
− 3/2
  c' 2    

+ ( z + c' t ) 2 + 1 − 2 r 2   + S µ0 I 0 c' r
δ(t − τ 0 )

  c     2πc z (1 − c' / c 2 ) + r 2
2 2

C-9
where δ(t) : Dirac’s delta function, and e = 0 when t < τ0.

D. Induced voltage for arbitrary waveform current

Eq. (C.10) gives the induced voltage under the assumption that the waveform of
the return-stroke current is a step function. The induced voltage e for an arbitrary
current waveform can be evaluated in the following convolution integral.

t
e(t ) = e S ( t )i( 0 ) + ∫0 e S ( t − τ)i' ( τ)dτ (C.11)

where eS(t) is induced voltage for unit step current. The above integral can be calculated

using a digital computer by an discretization with time step ∆t.

n
e( n∆t ) = e S ( n∆t )i( 0 ) + ∆t ∑ eS (( n − k )∆t )i'( k∆t ) (C.12)
k =0

C-10
C.3 Experimental Setup

A. Modeling of lightning-stroke current [5]

It is well-known that the development velocity of a return-stroke is less than the


light velocity. A solenoid reduces the longitudinal propagation velocity of
electromagnetic waves, but it generates unnecessary longitudinal magnetic flux.
Therefore, a waveguide illustrated in Fig. C.4 is used to model the return stroke. The
waveguide consists of two polyvinyl-chloride-insulated wires folded around a polyvinyl-
chloride pipe in the opposite direction to each other. Each magnetic flux generated by
the wires cancels the longitudinal component. The wire is 6.5 [m] long and its core
diameter is 1.6 [mm]. The length of the pipe is 2.5 [m] and its radius 30 [mm]. The
waveguide is hung at the height of 9 [cm] from the earth surface, and a current is injected
into the bottom of the waveguide.

Fig. C.5 shows the experimental setup. An inductance (L = 0.2 [µH]) is


connected in series to reduce the spike currents of the pulse generator. Fig. C.6 shows
the injected current waveforms without and with a capacitor C in Fig. C.5. The rise time
is observed as 3.2 [ns] from Fig. C.6 (a). The waveform shown in (a) can be regarded as
a step function up to t = 45 [ns]. A reflected wave is observed at t = 28 [ns] in Fig. C.6
(a). The arrival time and its amplitude indicates that 25 % of the current propagates
with light velocity. The current with light velocity would be due to the electrostatic
coupling between the insulated wires. The major reflected wave is observed at t = 43
[ns]. From the result, the 75 % of the current propagates along the folded wires with
56 % of the light velocity. Fig. C.6 (b) shows that the rise time of waveform is 12.5 [ns]
and is longer by 4 times than that in the case of no-capacitor. The waveguide
approximately represents a lightning-stroke current of which the velocity is less than the
light velocity.

C-11
6 cm

2.5 m

Fig. C.4 Waveguide for modeling discharge path

discharge-path model
10 cm
0.2 μH
coaxial cable L 51 Ω V oscilloscope
2.5 m

pulse C h
51 Ω
generator 147 pF
9 cm

alminum plates

Fig. C.5 Experimental Setup

C-12
current [ A ] 0.3

0.2

0.1

-0.1
0 20 40 60
time [ ns ]

(a) Step-function current (without C )

0.6
current [ A ]

0.4

0.2

0 20 40 60
time [ ns ]

(b) Slow front current waveform ( C = 147 pF )

Fig. C.6 Model-lightning current

C-13
B. Modeling of electronic appliance by loop circuit

To model an electronic appliance, a closed loop is used and is placed as its


surface is perpendicular to the earth and parallel to the lightning path. The arrangement
gives the severest condition in terms of interlinked magnetic flux. The induced voltage
to the loop is led by two coaxial cables (3D-2V) to an oscilloscope for canceling common
mode noise. In this case, the terminating impedance of the closed loop is 100 [Ω ] which
is given as the sum of the cable surge impedances. The inductance L of the closed loop
used for the test is measured to be 0.366 [µH] by an electronic bridge.

C-14
C.4 Experimental and Calculated Results

A. Induced voltage by step-function current

Induced voltages are shown in Fig. C.7 when a return-stroke current waveform
is assumed to be step function, and the loop is placed at the height of 0.8 [m] and the
distance 1 [m] from the waveguide. The solid line shows the calculated induced voltage
under the assumption that the return-stroke current is a step function with the amplitude
of 0.366 [A] which is the maximum value of the applied current in Fig. C.6 (a), and its
velocity is 56 % of the light velocity. The experimental result averaged during 2 [ns] to
reduce the noise is shown by the dotted line in the figure.

20
calculated
result
voltage [ mV ]

10
experiment
result

0
0 10 20
time [ ns ]

Fig. C.7 Induced voltage by step-function current

C-15
The peak voltage of the experimental result is smaller by 22% than that of the
calculated result. As described in Section 3.1, 75 % of the return-stroke current
propagates with 0.56 times of the light velocity. If the 75 % of the current component
contributes to the induced voltage, the calculated result agrees well with the experimental
result. A difference observed at the wavefront may be caused by the current with the
light velocity and an error after 15 [ns] by the reflection from the top of the waveguide.

B. Induced voltage by slow-front current

In this section, an induced voltage is investigated when the return-stroke current


is given as shown in Fig. C.6 (b). Calculated results in Fig. C.8 are obtained by a
numerical convolution in eq. (C.12). Experimental results averaged for 2 [ns] to reduce
high frequency oscillation are given by the broken line. The calculated peak voltages are
higher than experimental one because of the current with the light volocity. Those
waveforms agree with the experimental results regardless of the position of the loop in all
the cases. The results make clear that the voltage generated by the lightning current for
an arbitrary waveform can be evaluated by use of a numerical convolution technique.

C-16
40

30 calculated
result
voltage [ mV ]

20

10
experimental
result
0

-10
0 10 20 30
time [ ns ]

(a) distance r = 1.0 m, height h = 0.8 m

40

30
voltage [ mV ]

20

10

-10
0 10 20 30
time [ ns ]

(b) distance r = 1.2 m, height h = 0.8 m

Fig. C.8 Induced voltages by arbitrary waveform currents (continues to the next page)

C-17
40

30
voltage [ mV ]
20

10

-10
0 10 20 30
time [ ns ]

(c) distance r = 1.5 m, height h = 0.8 m

40

30
voltage [ mV ]

20

10

-10
0 10 20 30
time [ ns ]

(d) distance r = 1.0 m, height h = 1.21 m

Fig. C.8 Induced voltages by arbitrary waveform currents


(continued from the previous page)

C-18
C.5 Conclusion

In this appendix, an analytical formula is derived to evaluate the induced voltage


to a loop circuit from a lightning stroke. Its accuracy has been confirmed in comparison
with experimental results. The proposed method estimates an induced voltage to a
printed-circuit board in an electronic appliance and is useful to lightning protection
studies of an electronic appliance.

* This chapter is a revised version of ref. [6] presented by the author. The delta
function of eq. (C.10) is missing in ref. [6], and it was pointed out in the conference.

References of Appendix C

[1] A.Ametani: "Electro-Magnetic Interference in An Intelligent Building", Proc. ISCE, Vol. 1,


pp.24-29, 1994.
[2] S.Yokoyama: "A Study of Induced Lightning Surge Phenomena by Means Numerical Analysis
Method", T.IEE Japan, Vol.98-B, No.11, pp.561-586, 1976.
[3] R.H.Golde: "Lighting", Vol.2 Lighting Protection, Ch. 23 Protection of Distribution Lines,
Academic Press, London, 1977
[4] L. V. Bewlay: "Traveling waves on transmission systems", John Wiley & Sons, 1933.
[5] Y.Watanabe, S.Kato, M.Azuma and E.Zaima: "Surge Characteristics of A Miniature
Transmission Tower Model Simulated Upward Lightning Current Path", T. IEE Japan, Vol. 114-
B, No. 10, pp.1066-1072, 1994.
[6] T. Noda, J. Takami, N. Nagaoka, and A. Ametani, "Basic Investigation of a Lightning-Induced
Voltage to an Electronic Appliance," Proc. 23rd International Conference on Lightning
Protection (ICLP), Volume II, pp. 690-695, Firenze, Italy, 1996.

C-19
APPENDIX D

LIST OF PAPERS PRESENTED BY THE AUTHOR

BOOK

1 T. Noda, L. Dubé, “Simulation Model Description Language MODELS, Tutorial


Examples,” Japanese EMTP Committee, 1995. (in Japanese: 野田 琢, L. Dubé,
「Simulation Model Description Language MODELS 例題集」,日本 EMTP 委員
会, 1995.)

IEEE TRANSACTIONS

2 T. Noda, N. Nagaoka, A. Ametani, "Phase Domain Modeling of Frequency-


Dependent Transmission Lines by Means of an ARMA Model," IEEE Trans. on
Power Delivery, Vol. 11, No. 1, pp. 401-411, January, 1996.

3 T. Noda, N. Nagaoka, A. Ametani, "Further Improvements to a Phase-Domain


ARMA Line Model in Terms of Convolution, Steady-State Initialization, and
Stability," IEEE Power Engineering Society Summer Meeting, Denver, Colorado,
USA, 1996. (to be published in IEEE Trans.)

TRANS. IEE OF JAPAN

4 T. Noda and N. Nagaoka, “Development of ARMA Models for a Transient

D-1
Calculation using Linearized Least-Squares Method,” Trans. IEE of Japan, Vol.
114-B, No. 4, pp. 396-402, 1994. (in Japanese: 野田,長岡,「線形化最小 2 乗
法による ARMA 過渡現象計算モデルの開発」,電気学会論文誌 B, Vol.
114-B, No. 4, pp. 396-402, 1994.)

5 T. Noda, N. Nagaoka, and A. Ametani, “Fault-Surge Calculations using the Phase-


Domain ARMA Line Model,” Trans. IEE of Japan, Vol. 116-B, No. 11, pp. 1409-
1414, 1996. (in Japanese: 野田,長岡,雨谷,「相領域 ARMA 線路モデルを
用いた地絡サージ計算」,電気学会論文誌 B, Vol. 116-B, No. 11, pp. 1409-
1414, 1996.)

6 T. Noda, N. Nose, N. Nagaoka, and A. Ametani, “A Wave-Front-Time Dependent


Corona Model for Transmission-Line Surge Calculations,” Proc. of the Seventh
Annual Conference of Power & Energy Society, IEE of Japan, Session I-J, Paper
No. 36, pp. 209-214, 1996. (to be published in Trans. IEE of Japan, in Japanese:
野田,能勢,長岡,雨谷,「線路サージ計算に用いる波頭長依存コロナモデ
ル」,平成 8 年電気学会電力・エネルギー部門大会,論文 I-J, 36, pp. 209-214,
1996.)

7 T. Matsuura, T. Noda, N. Nagaoka, and A. Ametani, “A Simplified Calculation


Method of a Line Surge Considering the Frequency-Dependent Effect,” Trans. IEE
of Japan, Vol. 116-B, No. 6, pp. 706-711, 1996. (in Japanese: 松浦,野田,長岡,
雨谷,「周波数依存効果を考慮した線路サージ簡易計算」,電気学会論文誌
B, Vol. 116-B, No. 6, pp. 706-711, 1996.)

INTERNATIONAL CONFERENCE & JOURNAL PAPERS

8 T. Noda, J. Takami, N. Nagaoka, and A. Ametani, "Basic Investigation of a


Lightning-Induced Voltage to an Electronic Appliance," Proc. 23rd International
Conference on Lightning Protection (ICLP), Volume II, pp. 690-695, Firenze,

D-2
Italy, 1996.

9 T. Noda, K. Yamamoto, N. Nagaoka, and A. Ametani, “A Predictor-Corrector


Scheme for Solving a Nonlinear Circuit,” Proc. of International Conference on
Power Systems Transients ‘97 (IPST), Seattle, Washington USA, 1997.
(provisionally accepted)

10 A. Ametani, N. Nagaoka, T. Noda, and T. Matsuura, "A Simple and Efficient


Method for Including a Frequency-Dependent Effect in a Transmission Line
Transient Analysis," Proc. International Conference on Power Systems Transients
‘95 (IPST), pp. 11-16, , Lisbon, Portugal, 1995.

11 A. Ametani, N. Nagaoka, T. Noda, and T. Matsuura, "A Simple and Efficient


Method for Including Frequency-Dependent Effects in a Transmission Line
Transient Analysis," International Journal of Electrical Power & Energy Systems,
October/November, 1996.

DOMESTIC CONFERENCE PAPERS

12 T. Noda, T. Masakawa, and N. Nagaoka, “Fast Transient Analysis using the Z


Transform,” Proc. Kansai-Branch Meeting, IEE of Japan, Paper No. G9-12, 1991.
(in Japanese: 野田,政川,長岡,「Z 変換による高速過渡現象解析−高速コ
ンボリューション−」,平成 3 年電気関係学会関西支部連合大会,G9-12,
1991.)

13 T. Noda, T. Masakawa, and N. Nagaoka, “A Transmission-Line Surge Calculation


Method using the Z Transform,” Proc. Annual Meeting of IEE of Japan, Paper No.
1151, 1992. (in Japanese: 野田,政川,長岡,「Z 変換による線路サージ計算
法」,平成 4 年電気学会全国大会,1151, 1992.)

14 T. Noda and N. Nagaoka, “An Estimation Method of Linear Network's Transfer


Function,” Conference of Electric Power System Technology, IEE of Japan, PE-

D-3
92-150, 1992. (in Japanese: 野田,長岡,「線形回路網伝達関数推定の一手法」,
電気学会電力技術研究会資料,PE-92-150, 1992.)

15 T. Noda and N. Nagaoka, “An Estimation Method of Linear Network's Transfer


Function,” Proc. Kansai-Branch Meeting, IEE of Japan, Paper No. G4-8, 1992. (in
Japanese: 野田,長岡,「線形回路網伝達関数推定の一手法」,平成 4 年電気
関係学会関西支部連合大会,G4-8, 1992.)

16 T.Noda, M.Kubota, and N.Nagaoka, “A Phase Domain Surge Calculation


Method,” Proc. Annual Meeting of IEE of Japan, Paper No. 1387, 1993. (in
Japanese: 野田,窪田,長岡,「実領域多相線路サージ計算法」,平成 5 年電
気学会全国大会,1387, 1993.)

17 T. Noda and N. Nagaoka, “A Linearized Least-Squares Method for Estimating


Linear Network's Transfer Function - Relative Error Evaluation -,” Proc. Annual
Meeting of IEE of Japan, Paper No. 1397, 1993. (in Japanese: 野田,長岡,「線
形回路網伝達関数推定に適した線形化最小 2 乗法―相対誤差評価法―」,平
成 5 年電気学会全国大会,1397, 1993.)

18 T. Noda and N. Nagaoka, “A Phase Domain Surge Calculation Method (Part II) -
A Surge Calculation on a Cable -,” Proc. Kansai-Branch Meeting, IEE of Japan,
Paper No. G4-31, 1993. (in Japanese: 野田,長岡,「実領域多相線路サージ計
算法 第 2 報−ケーブルサージ計算への応用」,平成 5 年電気関係学会関西
支部連合大会,G4-31, 1993.)

19 K. Yamabuki, T. Noda, and N. Nagaoka, “Contactless Measurement of Surge


Waveforms (Part 1),” Proc. Kansai-Branch Meeting, IEE of Japan, Paper No. G4-
30, 1993. (in Japanese: 山吹,野田,長岡,「周波数変換法を用いた非接触形
サージ測定装置 第1報」,平成5年電気関係学会関西支部連合大会,G4-30,
1993.)

20 T. Noda and N. Nagaoka, “A Time-Domain Surge Calculation Method with

D-4
Frequency-Dependent Modal-Transformation Matrices,” Conference of Electric
Power System Technology, IEE of Japan, PE-93-153, 1993. (in Japanese: 野田,
長岡,「変換行列の周波数依存性を考慮した時間領域線路サージ計算法」,
電気学会電力技術研究会資料,PE-93-153, 1993.)

21 K. Yamabuki, T. Noda, and N. Nagaoka, “Basic Investigation of Contactless Surge


Measurement,” Conference of Electric Power System Technology, IEE of Japan,
PE-93-152, 1993, (in Japanese: 山吹,野田,長岡,「非接触形サージ測定装置
の開発に関する基礎的研究」,電気学会電力技術研究会資料,PE-93-152,
1993.)

22 T. Noda, T. Sawada, K. Fujii, and N. Nagaoka, “ARMA Models for Transient


Calculations and its Identification - Identification of Coefficients and Orders -,”
Proc. Annual Meeting of IEE of Japan, Paper No. 1373, 1994. (in Japanese: 野田,
澤田,藤井,長岡,「ARMA 過渡現象計算モデルとその作成法」,平成 6 年
電気学会全国大会,1373, 1994.)

23 K. Yamabuki, T. Noda, and N. Nagaoka, “Contactless Measurement of Surge


Waveforms on a Mutiphase Line,” Proc. Annual Meeting of IEE of Japan, Paper
No. 1381, 1994. (in Japanese: 山吹,野田,長岡,「多相線路における非接
触サージ測定装置」,平成6年電気学会全国大会,1381, 1994.)

24 T. Matsuura, T. Noda, and N. Nagaoka, “A Simple Method for Including


Frequency-Dependent Effects in a Transmission-Line Surge Calculation (Part 1), ”
Proc. Annual Meeting of IEE of Japan, Paper No. 1367, 1994. (in Japanese: 松浦,
野田,長岡,「線路サージ計算における周波数依存効果の簡略導入法 第 1
報」,平成 6 年電気学会全国大会,1367, 1994.)

25 T. Noda, N. Nagaoka, and A. Ametani, “A Phase-Frequency Characteristic


Synthesis of an Impedance from the Magnitude-Frequency-Characteristic,” Proc.
Annual Meeting of IEE of Japan, Paper No. 1710, 1995. (in Japanese: 野田,長岡,

D-5
雨谷,「インピーダンス絶対値周波数特性からの位相特性の合成」,平成 7
年電気学会全国大会,1710, 1995.)

26 T. Matsuura, T. Noda, N. Nagaoka, and A. Ametani, “A Simple Method for


Including Frequency-Dependent Effects in a Transmission-Line Surge Calculation
(Part 2), ” Proc. Annual Meeting of IEE of Japan, Paper No. 1699, 1995. (in
Japanese: 松浦,野田,長岡,雨谷,「線路サージ計算における周波数依存
効果の簡略導入法 第2報」,平成7年電気学会全国大会,1699, 1995.)

27 T. Noda, N. Nagaoka, and A. Ametani, “Analysis of Vertical-Conductor Surge


Impedance using the Concept of Retarded Inductance,” Conference of Discharge
and High Voltage Technology, IEE of Japan, ED-95-234/HV-95-105, 1995, (in
Japanese: 野田,長岡,雨谷,「遅延インダクタンスを用いた垂直導体サー
ジインピーダンスの解析」,電気学会放電高電圧合同研究会, ED-95-234/
HV-95-105, 1995.)

28 T. Noda, J. Takami, N. Nagaoka, and A. Ametani, “Basic Investigation of a


Lightning Induced Voltage to an Electronic Appliance,” Conference of Discharge
and High Voltage Technology, IEE of Japan, ED-95-231/HV-95-102, 1995, (in
Japanese: 野田,高見,長岡,雨谷,「落雷によって電子機器に生じる誘導
電圧の基礎的検討」,電気学会放電高電圧合同研究会, ED-95-231/HV-95-102,
1995.)

29 T. Matsuura, T. Noda, N. Nagaoka, and A. Ametani, “A Simplified Transmission-


Line Surge Calculation Method Considering Frequency-Dependent Effects,”
Conference of Discharge and High Voltage Technology, IEE of Japan, ED-95-
188/HV-95-59, 1995, (in Japanese: 松浦,野田,長岡,雨谷,「周波数依存効
果を考慮した線路サージ簡略計算法」,電気学会放電高電圧合同研究会,
ED-95-188/ HV-95-59, 1995.)

30 T. Noda, N. Nose, N. Nagaoka, and A. Ametani, “An EMTP Compatible Corona


Model (Part 1),” Proc. Kansai-Branch Meeting, IEE of Japan, Paper No. G5-31,

D-6
1995. (in Japanese: 野田,能勢,長岡,雨谷,「EMTP 線路モデルに導入容
易なコロナモデル 第 1 報」,平成 7 年電気関係学会関西支部連合大会, G5-
31, 1995.)

31 J. Takami, T. Noda, N. Nagaoka, and A. Ametani, “Investigation of Induced


Voltages to a Loop Circuit,” Proc. Kansai-Branch Meeting, IEE of Japan, Paper
No. G5-43, 1995. (in Japanese: 高見,野田,長岡,雨谷,「ループ回路に誘導
される雷過電圧の一検討」,平成 7 年電気関係学会関西支部連合大会,
G5-43, 1995.)

32 T. Noda, H. Fukuzono, S. Sekioka, N. Nagaoka, and A. Ametani, “Charge -


Voltage Measurement of Transmission-Line Corona without a Corona Cage,” Proc.
of the Seventh Annual Conference of Power & Energy Society, IEE of Japan,
Session II, Paper No. 427, p. 451, 1996. (in Japanese: 野田,福園,関岡,長岡,
雨谷,「コロナケージを用いない送電線コロナ電荷−電圧曲線の測定」,平
成 8 年電気学会電力・エネルギー部門大会,論文 II, 427, p. 451, 1996.)

33 T. Noda, N. Nagaoka, A. Ametani, “A General Formulation Method of a Nonlinear


Circuit for Transient Calculations,” Proc. of the Seventh Annual Conference of
Power & Energy Society, IEE of Japan, Session II, Paper No. 429, p. 455, 1996.
(in Japanese: 同英文表題,平成 8 年電気学会電力・エネルギー部門大会,論
文 II, 429, p. 455, 1996.)

34 N. Nagaoka, T. Noda, K. Yamabuki, H. Kajino, J. Takami, and A. Ametani,


“Contactless Measurement of Surge Current and Voltage on Overhead Line using
Closed Loops,” Proc. of the Seventh Annual Conference of Power & Energy
Society, IEE of Japan, Session I-J, Paper No. 38, pp. 221-226, 1996. (in Japanese:
長岡,野田,山吹,梶野,高見,雨谷,「閉回路を用いた架空送電線サージ
電流・電圧非接触測定」,平成 8 年電気学会電力・エネルギー部門大会,論
文 I-J, 38, pp. 221-226, 1996.)

D-7
35 K. Yamamoto, T. Noda, N. Nagaoka, and A. Ametani, “Transient Calculation of a
Switching Circuit using a Predictor-Corrector Based Nonlinear Circuit Analysis
Method,” Proc. Kansai-Branch Meeting, IEE of Japan, Paper No. G5-12, 1996.
(in Japanese: 山本,野田,長岡,雨谷,「予測子−修正子法を用いた非線形
回路解析法によるスイッチング回路の解析」,平成 8 年電気関係学会関西支
部連合大会, G5-12, 1996.)

36 K. Yoshida, T. Noda, N. Nagaoka, and A. Ametani, “An Immitance Synthesis


Method of a Network using the Linearized Least-Squares Method,” Proc. Kansai-
Branch Meeting, IEE of Japan, Paper No. G5-11, 1996. (in Japanese: 吉田,野田,
長岡,雨谷,「線形化最小 2 乗法を用いた回路網イミタンス合成法」,平成
8 年電気関係学会関西支部連合大会, G5-11, 1996.)

LECTURES

37 T. Noda, “An Introduction to the MODELS language,” Workshop of Workgroup


C, Japanese EMTP Committee Meeting, 1995. (in Japanese: 野田,「MODELS 例
題集の紹介」,日本 EMTP 委員会 C 部門ワークショップ, 於 日立製作所本
社, 1995.)

38 T. Noda, “An Introduction to Taku Noda Frequency Dependence (NODA


SETUP)”, Workshop of Workgroup A, Japanese EMTP Committee Meeting, 1996.
(in Japanese: 野田,「Taku Noda Frequency Dependence (NODA SETUP)の概
要」,日本 EMTP 委員会 A 部門ワークショップ,於 明電舎, 1995.)

D-8

View publication stats

You might also like