Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

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

net/publication/343666674

PSS/E Based Power System Stabilizer Tuning Tool

Conference Paper · June 2020


DOI: 10.1109/SIELA49118.2020.9167137

CITATION READS

1 172

4 authors, including:

Nikolay Nikolaev Yulian Rangelov


Technical University of Varna Technical University of Varna
54 PUBLICATIONS   64 CITATIONS    82 PUBLICATIONS   68 CITATIONS   

SEE PROFILE SEE PROFILE

Ara Panosyan
Siemens
5 PUBLICATIONS   5 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

GARPUR: Generally Accepted Reliability Principle with Uncertainty modelling and through probabilistic Risk assessment View project

Създаване на лабораторен комплекс за научни изследвания в областта на релейната защита и автоматизация на


електроенергийните системи View project

All content following this page was uploaded by Nikolay Nikolaev on 21 August 2020.

The user has requested enhancement of the downloaded file.


PSS/E Based Power System Stabilizer Tuning Tool
Nikolay Nikolaev, Yulian Rangelov Ara Panosyan, Ngoc Tuan Trinh
Dept. Electric Power Engineering System Dynamics
Technical University - Varna SIEMENS AG - PTI
Varna, Bulgaria Erlangen, Germany
n.nikolaev@tu-varna.bg, y.rangelov@tu-varna.bg ara.panosyan@siemens.com, ngoc.trinh@siemens.com

Abstract— Since their introduction, Power System third-order synchronous machine and a first-order automatic
Stabilizers have become widely accepted as an efficient voltage regulator (AVR) model [4, 12].
supplementary excitation controller for extending the power
transfer capability limits of the system by adding damping to the The Heffron-Phillips model is amenable to simple analysis
low-frequency electromechanical modes of oscillation to and thus it is possible to derive simple, closed-form solutions
improve the overall system stability. The purpose of the tool for certain transfer functions and torque-related expressions.
presented in this paper is to enable an easy to use, flexible and This simple model of the SMIB system, however, ignores the
accurate tuning of various types of Power System Stabilizers, higher-order dynamics of the generator and excitation system,
both for a SMIB, as well as multi-machine system which are used in the simulation of the dynamic performance
configurations. The tool is built in Python and relies on the of power systems, and is also not easily extendable for multi-
dynamic models of the PSS/E library, and the PSS/E steady- machine systems [5].
state and dynamic simulation modules. The key component of
this new tool is PSS/E’s linearization module, which generates II. PSS/E BASED PSS TUNING TOOL
the linear state-space model around a steady-state operating The purpose of the tool presented in this paper is to enable
point of the full non-linear dynamic model of the system. It
an easy to use, flexible and accurate tuning of various types of
provides valuable tools necessary for the analysis of the stability
phenomena and tackling with the optimization of the stabilizer
Power System Stabilizers, both for a SMIB, as well as multi-
parameters and validation of the results. The core features of machine system configurations. The tool is built in Python and
the tool were demonstrated with the well-established Kundur’s relies on the dynamic models of the PSS/E model library, and
four-machine test system. Utilizing interpreted language, such the PSS/E steady-state and dynamic simulation modules. The
as Python, makes it very convenient for potential users to key component of this new tool is PSS/E’s linearization
flexibly create custom scripts to perform the analysis and module, which generates the linear state-space model around
optimization according to their needs and preferences. a steady-state operating point of the full non-linear dynamic
model of the system.
Keywords—stabilizer, PSS/E, Python, tuning, phase
compensation PSS/E uses a numerical linearization procedure, which
computes the linearized models using the perturbation method
I. INTRODUCTION [6]. The linearization routine implemented in PSS/E can be
found in the literature named as forward-difference
As the integration of renewable generation rapidly
approximation [7]. The main advantage of such numerical
progresses, challenges to heavily loaded transmission systems
linearization procedure, compared to an analytical
arise. These challenges are further amplified by a weak grid
linearization, is that it does not require a pre-defined linear
infrastructure due to the lack of required upgrades for the
model of each non-linear model in the dynamic model library.
transmission infrastructure. Under such a weak system and
The numerical linearization procedure can, therefore, be
heavy load conditions, small-signal stability in the form of
applied directly to any dynamic model representation in
low-frequency electromechanical oscillations in the 0.1 to 3.0
PSS/E, including user-defined models, for which an analytical
Hz frequency range is often a key limiting factor for the power
linearization may not be available.
transfer.
The linearization with PSS/E also has a clear advantage
Since their introduction more than five decades ago,
over the linear Heffron-Phillips model, as it can be easily
Power System Stabilizers (PSS) have become widely accepted
applied to any multi-machine system configuration with any
as an efficient and cost-effective supplementary excitation
number of higher-order dynamics models, including different
controller for extending the power transfer capability limits of
controllers, such as AVRs and speed governors, as well as
the system by adding damping to the low-frequency
HVDC and FACTS [13].
electromechanical modes of oscillation to improve the overall
system stability. Several methods have been proposed and The steady-state and dynamic models in PSS/E are hence
applied over the years for the tuning of PSS. It is, however, the basis for obtaining the linear models required for the PSS
important to realize that the correct tuning of a PSS is crucial tuning. Although the Stabilizer Tuning techniques themselves
for its performance, while a poorly tuned PSS may deteriorate are implemented in Python, the network model, and the
the overall system stability even more [1-3]. steady-state and dynamic simulation modules in PSS/E are
frequently utilized by the tool during the different stages of the
The most well-proven and widely-used basic techniques,
tuning procedure, as well as during the tuning validation stage.
which have been successfully utilized for power system
stabilizer tuning, are the phase compensation and root locus Hence, the tool combines the unique capabilities of PSS/E
methods. These PSS tuning techniques are traditionally based in dynamic modeling, linearization and analysis, with the
on the well-known Heffron-Phillips (K-Constant) model of large range of functions and modules available in Python for
Single Machine connected to an Infinite Bus (SMIB) system, pre- and post-processing, as well as for automation and
which is a simplified fourth-order linear model including a optimization.
Fig. 1. Structure of the tool

III. STRUCTURE OF THE TOOL the tuning procedures and asses the performance and quality
of the stabilizer. This module has both Time-domain and
The structure of the tool is depicted in fig. 1. This diagram Frequency-domain features. The first relies on the Dynamic
presents the high-level modules of the tool as well as the simulation solver in PSS/E, since it is easily accessible
required PSS/E modules and files. The Python tool through the API and reliably simulates the nonlinear power
communicates with the PSS/E through the existing extensive system model. The following options are provided to the user
API the latter provides. The API is denoted with dashed blue for an arbitrary machine in the model:
arrows in the diagram. In few cases the API is insufficient and
the interaction is done through file exchange. The tool is • Turbine reference GREF step response;
entirely coded with the object-oriented paradigm.
• Voltage reference VREF step response.
Initially, the workflow begins with creating a Tuning
Project. One project may contain multiple Tuning Cases. The The simulation results are returned to Python through an OUT
instantiation of a case requires a pair of SAV (power flow file which is imported and made available to the user to readily
data) and DYR (dynamic data) files to be loaded. Generally, display them with the tool’s embedded plotting or use the
the user can start with a single pair of data files and use the advanced features of the Matplotlib package [8]. All results
Population of Cases module to generate a set of multiple from the tool can also be exported as CSV files for processing
operating points for the considered synchronous machine. with external software.
Besides the operating point, the module allows one to vary the The set of frequency-domain tools work on the linearized
impedance and the voltage of the swing generator, thus model and include:
allowing one to select the most appropriate case for tuning.
• Uncompensated phase response, also known as P-Vr
The Data classes store the parameters of the machines and characteristic;
their regulators. The tool allows the user to make changes to
those parameters directly in Python. • Compensated phase response;
The tool takes advantage of the PSS/E ASTR activity, • Eigenvalue analysis;
which can linearize the power system model around the
current operating point. The ASTR produces a binary LSA file. • Root locus;
Then the tool runs the LSYSAN program (which is part of the • Closed-loop bode response from Vref to SPEED for a
PSS/E package of additional executables) to convert the LSA specified machine.
file to readable ASCII text file. The latter holds information
about the variables and the state-space matrices. The ASCII IV. PHASE RESPONSE TUNING
file is imported by the tool and stored in the memory as a One of the most widely applied techniques for tuning
linearized model for the current tuning case. power system stabilizers (PSS) is the phase compensation
Once the linear model is obtained, the Optimization approach [5]. It consists of the following steps:
procedure can begin. First is the Phase tuning of the stabilizer. 1. Evaluation of the inherent P-Vr characteristic GP-Vr of
Usually, then follows the Gain tuning. Particular details about the synchronous generator, as defined with the following
the optimization are provided in the sections that follow. transfer function:
The Analysis/Verification module is a significant part of
the tool. It allows one to study the machine before and after
ΔTe Although a zero-phase would theoretically result in a
GP −Vr ( s) = (1) maximum damping torque, such a full phase compensation is
ΔVref
usually not recommended in practice. This is due to the
where Te is the electromagnetic torque and Vref is the inevitable uncertainties associated with the models and their
reference voltage of the excitation control. parameters. Hence, it is always desirable to have a lightly
undercompensated system with a certain phase lag (e.g. 15°-
2. Tuning of the PSS parameters to make its transfer
30°) at the frequencies of interest, to avoid a negative
function equal to:
synchronizing torque contribution by the PSS. Therefore, the
KS user-defined phase target T can be set at a certain phase lag
GPSS ( s ) = (2) (e.g. 20°), while the upper limit UL would be set to a value
GP −Vr ( s ) close to 0°, to minimize possible phase lead resulting in a
where KS is the stabilizer gain. negative synchronizing.
The phase response of GP-Vr is called the uncompensated The objective (5) formulates a constrained nonlinear
phase. In practice, condition (2) may not be fulfilled entirely, programming problem. In the tool, it is solved with the
since the stabilizers have fixed structures. The ultimate goal optimization module of SciPy [9]. The selection of initial
of the phase compensation approach is to make the phase parameter values is crucial for successful optimization. The
response of the compensated transfer function GC as close as tool allows the user to set those initial parameters, or allow the
possible to zero about the frequency range of tool to generate random sets and execute the optimization for
electromechanical oscillations: each of them, finally providing the solution whose objective
is the best. This approach is called multi-start and it is a
ΔTe
GC ( s ) = GPSS ( s ) ⋅ GP −Vr ( s ) = (3) pseudo-global optimization. Future work will be dedicated to
Δω implementing true global solver since the problem is highly
nonconvex and it could be more efficient than the multi-start
In the ideal case, the transfer function (3) should have a procedure.
zero phase response and represent a pure gain. In such a case,
the rotor speed deviations will induce an in-phase
electromagnetic torque. The latter will increase the damping 4.0

of the electromechanical oscillations, i.e. the damping torque.


weights

3.0

In essence, the phase tuning procedure is a curve-fitting 2.0

problem. The decision variables are the PSS lead-lag block W


1.0
parameters. The objective is to minimize:
fmin=0.1 0.5 1.0 1.5 fmax=2.0
2
  1  
f, Hz

min  Phase {GPSS ( s )} − Phase    (4) 10.0


compensated phase

PSS params  
 GP −Vr ( s )  
5.0
 0.0
UL
upperlimits

-5.0
In the practical implementation of the tuning procedure in T
-10.0
targets

the PSS tool, the objective (4) is modified to accommodate the -15.0
-20.0
option for user preferences on the desired phase
compensation: fmin=0.1 0.5 1.0 1.5 fmax=2.0
f, Hz
2
  1   Fig. 2. Compensated phase response – limits and targets
min  Phase {GPSS ( s )} − Phase   + T  W (5)
PSS params  
  GP −Vr ( s )   V. GAIN TUNING
given that: After the phase compensation is completed, one can
proceed with the tuning of the stabilizer gain. Besides tuning
LB ≤ PSS params ≤ UB (6)
the gain of only one stabilizer, the tool also allows coordinated
tuning of gains upon a set of stabilizers selected by the user.
Phase {GC ( s )} ≤ UL (7)
The objective of the coordinated tuning of gains is to
where LB and UB are the lower and upper bounds of the PSS minimize the H∞ norm of the power system, computed from
parameters, T is the user-defined target for the compensated all ΔVref inputs to all Δω and ΔPe outputs. This objective
phase, W is the weight the user puts onto the different parts of guarantees robust performance and good damping.
the considered frequency range, and UL is the upper limits
imposed on the compensated phase, generally to avoid The problem has two types of constraints:
overcompensation. T, W, and UL are explained in fig. 2. As • Lower and upper limits for the gain values;
depicted on the figure, the frequency range considered for
tuning can be divided into several bands (four in the example). • The value of any gain should be less than 1/3 of the
For every band, the user can specify different targets and thus value where the system becomes unstable.
achieve exact, over or under-compensation. The weights W The first constraint makes the solution compliant with the
control the significance of the different bands in the curve- capabilities of the stabilizer hardware controller. The second
fitting process. The greater the weight is, the more the one guarantees that the uncertainties of the computer model
particular point will be penalized in the objective and thus will not lead to instabilities in the real power plant. This is also
better phase compensation is expected for the respective band. the recommendation of WECC during the commissioning of
the stabilizer [10].
The root-locus feature of the tool can be used as an TABLE I. COMPARISON OF THE ELECTROMECHANICAL MODES
alternative approach for PSS gain selection. One can trace Thyristor
how the modes shift with increasing gain and check the excitation Area 1 Area 2
boundary where the system becomes unstable. An example is with high Interarea mode local mode local mode
presented in fig. 3. gain, PSS is G1 and G2 G3 and G4
OFF
The power system stabilizer gain can also be determined
f = 0.61 Hz f = 1.14 Hz f = 1.17 Hz
using the root-locus technique, which represents the migration Kundur [14]
ξ = -0.008 ξ = 0.07 ξ = 0.07
of the modes as the gain is gradually increased from zero
f = 0.61 Hz f = 1.13 Hz f = 1.17 Hz
towards infinity. Firstly, the root-locus technique enables PSS/E tool
ξ = 0.009 ξ = 0.09 ξ = 0.09
obtaining the critical gain, also known as instability gain,
The frequency and damping of the electromechanical
beyond which the system becomes unstable. To ensure an
modes obtained with the tool are almost identical with [14].
adequate gain margin of 6 to 10 dB, the maximum acceptable
There are insignificant discrepancies, mainly in the damping,
gain is accordingly assumed to be 1/2 to 1/3 of the critical gain
which are due to differences in the linearization techniques,
[11]. A similar procedure is recommended by the WECC
the modeling of the machine (the stator resistance is omitted
Power System Stabilizer Tuning Guidelines and is usually
in PSS/E) and the inclusion of governor control. The results
carried out by the gain margin tests during commissioning.
that follow will demonstrate the tuning of generator G1, while
The root-locus plots also facilitate the selection of the optimal
the rest of the stabilizers remain out-of-service. The tuning of
gain values, for which the maximum damping ratio of the
G2, G3, and G4 is performed with the same procedure and for
most relevant modes are achievable. It is important to note,
the sake of brevity is not presented here.
that the gain tuning should be aimed at providing an adequate
amount of damping to local modes and a high contribution to The uncompensated phase response GP-Vr(s) of G1 is
interarea modes. Excess of local mode damping is often depicted in fig. 4. Considering the results for the
unnecessary, and could cause the stabilizer to saturate in electromechanical modes, the tuning range for the phase
response to oscillations following large disturbances, and compensation is set between 0.5 and 1.25 Hz.
thereby deteriorating the system performance, by allowing
interarea swings to increase in the second swing [2]. -20

Ks=0.0
3.0 Ks=3.0 -30
Ks=6.0
2.5 Ks=9.0
Phase [deg]
Frequency [Hz]

Ks=12.0 -40
2.0 Ks=15.0
Damping=0.05
1.5 -50

1.0
-60
0.5
0.5 1.0 1.5 2.0 2.5 3.0
0.0 Frequency [Hz]

-20.0 -17.5 -15.0 -12.5 -10.0 -7.5 -5.0 -2.5 0.0 Fig. 4. Uncompensate phase response of G1
Real
Fig. 3. Example for gain selection based on the root locus The weights, upper limits and the targets for the phase
compensation are presented in fig. 5. The tuning frequency
VI. DEMONSTRATION OF THE TOOL range is subdivided into three bands. The first and the last ones
are where our interarea and local modes are located.
To demonstrate the tool we utilize the well established Therefore, high weight 5 was set for them. The middle band
Kundur’s four-machine, two-area power system [14]. The has unity weight since it is not that important. The optimized
high gain thyristor excitation is represented with the ESAC4A compensated phase response is also depicted in fig. 5. The
model (the transient gain reduction is omitted), the generators optimization routine has converged in 15 seconds and the
models are GENROU, the stabilizer model is PSS1A. Unlike compensation is within ±1° of the desired target.
the original implementation, we included a simple governor
TGOV1 to support the frequency regulation.
The model validation was made by modal analysis, with
the stabilizers omitted. The comparison of the modes is
presented in Table I.
5 (a) Interarea
4
0.70
Weights

3
0.69
2
0.68 45
1 0.67 30

Frequency [Hz]
5 0.66 20
4 15
Phase [deg]

Compensated phase 0.65


3
Target compensation 10
2 0.64
Upper limits
1 0.63
0 5 0
0.62
0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2
Frequency [Hz] 0.61
Fig. 5. Compensated phase response of G1
0.60
-0.4 -0.3 -0.2 -0.1 0.0
The next step is tuning the gain of G1’s stabilizer. The gain Real
is varied from 0 to 45. The mode shift is presented in Table II
and fig. 6. Note that the mode of area 2 is not presented since (b) Local
it is not affected by the PSS of G1. As the gain approach 20,
3.5
the relative damping of the interarea mode increases to 0.08
but the further increase of the gain does not improve the 45
3.0
damping. The local mode relative damping improves from
0.093 to 0.242. However, as the gain becomes greater than 15, 30
the frequency of the local mode increases substantially. 2.5
Frequency [Hz]

Moreover, after gain 30 the mode shift turns its direction to


2.0 20
the right, which is not desirable. Based on these observations
a conservative value of 15 pu was selected for the stabilizer 15
gain. If the gain is increased three-fold (to 45) the modal 1.5 10
5 0
analysis shows that there are no unstable eigenvalues.
1.0
TABLE II. ROOT LOCUS FOR G1 GAIN TUNING
0.5
Gain Interarea mode Area 1 local mode
Ks Real f, Hz ξ Real f, Hz ξ 0.0
-6.0 -5.0 -4.0 -3.0 -2.0 -1.0 0.0
0 -0.038 0.619 0.010 -0.667 1.133 0.093 Real
5 -0.079 0.619 0.020 -0.966 1.140 0.134 Fig. 6. Root locus for G1 gain tuning
10 -0.226 0.634 0.057 -2.181 1.311 0.256
The optimized stabilizer parameters for all four machines
15 -0.302 0.648 0.074 -3.609 1.556 0.346 in the test system are presented in Table III. Apart from the
20 -0.338 0.658 0.082 -4.780 1.891 0.373 modal analysis, validation for the proper tuning was also made
in the frequency and the time-domain. The closed-loop Bode
30 -0.364 0.672 0.086 -5.454 2.603 0.316 diagram of G1 for the response of the rotor speed from
45 -0.370 0.679 0.087 -4.918 3.140 0.242 perturbations from Vref is presented in fig. 7. The figure
depicts two cases – the PSS is OFF and ON respectively. The
peaks for the case without PSS clearly indicate the interarea
and local modes (see Table I). The effect of properly tuned
stabilizer can be seen as the peaks are effectively shaved.
The transient response of the system, obtained with the
nonlinear model, is depicted in fig. 8. One can see the
considerable improvement in the oscillations damping when
the stabilizers are put into service.
TABLE III. PSS OPTIMIZED PARAMETERS solving this problem that is so critical for the stability of the
PSS1A system. It provides valuable tools necessary for the analysis of
G1 G2 G3 G4 the stability phenomena and tackling with the optimization of
Param.
Tw 10.000 10.000 10.000 10.000 the PSS parameters and validation of the results. The core
T1 0.129 0.135 0.127 0.130 features of the tool were demonstrated with the well-
T2 0.021 0.020 0.021 0.020 established Kundur’s four-machine test system. Utilizing
T3 3.798 2.993 4.254 2.953 interpreted language, such as Python, makes it very
T4 5.160 5.420 5.014 5.432 convenient for potential users to flexibly create custom scripts
Ks 15.000 15.000 15.000 15.000 to perform the analysis and optimization according to their
needs and preferences.
Future updates of the tool could include the
PSS ON implementation of global optimization routines that would
0.25 PSS OFF allow higher quality of the tuning, which is less dependent on
the selection of the initial parameter values.
0.20
REFERENCES
0.15 [1] E. V. Larsen and D. A. Swann, “Applying power system stabilizers
Magnitude [pu]

Part I: General concepts,” IEEE Trans. Power Appar. Syst., vol. PAS-
100, no. 6, pp. 3017–3024, 1981.
0.10 [2] E. V. Larsen and D. A. Swann, “Applying power system stabilizers
Part II: Performance objectives and tuning concepts,” IEEE Trans.
Power Appar. Syst., vol. PAS-100, no. 6, pp. 3025–3033, 1981.
0.05
[3] E. V. Larsen and D. A. Swann, “Applying power system stabilizers
Part III: Practical considerations,” IEEE Trans. Power Appar. Syst.,
0.00 vol. PAS-100, no. 6, pp. 3034–3046, 1981.
0.0 0.5 1.0 1.5 2.0 2.5 [4] “IEEE Tutorial Course Power System Stabilization via Excitation
Frequency [Hz] Control.” [Online]. Available: https://resourcecenter.ieee-
pes.org/technical-publications/tutorial-papers/PES09TP250.html.
Fig. 7. Bode response Vref-to-Speed for G1
[Accessed: 27-Jan-2020].
[5] M. J. Gibbard, P. Pourbeik, and D. J. Vowles, Small-signal stability,
0.0000
control and dynamic performance of power systems. University of
Speed [pu]

Adelaide Press, 2015.


-0.0002 [6] J. Persson, J. G. Slootweg, L. Rouco, L. Söder, and W. L. Kling, “A
comparison of eigenvalues obtained with two dynamic simulation
software packages,” in 2003 IEEE Bologna PowerTech - Conference
Proceedings, 2003, vol. 2, pp. 246–252.
Pelec [pu]

7.1 [7] P. E. Gill, W. Murray, and M. H. Wright, Practical optimization.


7.0
Academic Press, 1981.
[8] “Matplotlib: Python plotting — Matplotlib 3.1.2 documentation.”
[Online]. Available: https://matplotlib.org/. [Accessed: 27-Jan-2020].
1.040 [9] “SciPy.org — SciPy.org.” [Online]. Available: https://www.scipy.org/.
Eterm [pu]

[Accessed: 27-Jan-2020].
1.035 PSS ALL ON [10] “WECC Power System Stabilizer Tuning Guidelines.” [Online].
PSS ALL OFF Available: http://wecc.biz. [Accessed: 27-Jan-2020].
1.030 [11] J. Chow and J. Sanchez-Gasca, Power system modeling, computation,
0 2 4 6 8 10 12 14
Time [s] and control. Wiley-IEEE Press, 2019.
[12] W. G. Heffron and R. A. Phillips, “Effect of a Modern Amplidyne
Fig. 8. Step response in Vref on G1
Voltage Regulator on Underexcited Operation of Large Turbine
Generators,” Trans. Am. Inst. Electr. Eng. Part III Power Appar. Syst.,
vol. 71, no. 1, pp. 692–697, 1952.
VII. CONCLUSIONS AND FUTURE WORK
[13] PSS/E 34 Program Operation Manual (POM) [Accessed: 27-Jan-2020]
The problem for tuning power system stabilizers is far [14] P. Kundur, Power system stability and control. McGraw-Hill, 1994.
from being straightforward. The developed PSS/E-Python
tool presented in this paper demonstrates its potency for

View publication stats

You might also like