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

net/publication/256461369

Least squares estimation of double-exponential function parameters

Conference Paper · September 2013

CITATIONS READS

6 1,080

3 authors, including:

Dino Lovric Slavko Vujevic

University of Split University of Split
45 PUBLICATIONS   226 CITATIONS    78 PUBLICATIONS   453 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Dino Lovric on 31 May 2014.

The user has requested enhancement of the downloaded file.

 11th International Conference on Applied Electromagnetics - ΠEC 2013
September 01 – 04, 2013, Niš, Serbia

LEAST SQUARES ESTIMATION OF DOUBLE-EXPONENTIAL FUNCTION

PARAMETERS
Dino LOVRIĆ1, Slavko VUJEVIĆ2 and Tonći MODRIĆ3
Abstract: In this paper an effective numerical algorithm for Double-exponential function can be used to approxi-
computation of double-exponential function parameters based on mate the lightning return stroke current and is described by
the available input data is presented. The parameter estimation is the following expression:
achieved using the Marquardt least squares method.
Keywords: Double-exponential function, lightning, Mar- I 0 − α⋅ t
quardt least squares method i(t ) =
η
⋅e (
− e −β⋅t ) (1)

INTRODUCTION where η is the correction coefficient of the current peak

In the field of electromagnetic analysis of lightning value, α and β are the parameters of the double-exponential
phenomena a number of mathematical functions are availa- function.
ble for modelling the channel-base lightning current [1]. According to Fig. 1, two basic requirements for the es-
The simplest of these functions is the double-exponential timation of double-exponential function parameters η, α
function [2], which, despite its numerous drawbacks, con- and β can be written as:
tinues to be in use mainly due to its simplicity.
In this paper, an algorithm for estimation of double- i = 0.9 ⋅ I 0 for t = t 2 (2)
exponential function parameters is presented, which is
based on a similar algorithm applied on the Heidler func- i = 0.5 ⋅ I 0 for t = t h (3)
tion [3]. The algorithm enables simultaneous solving of a
system of two, three or four nonlinear equations depending Two additional requirements can be deduced from the
on the available input data: current peak value, front dura- charge transfer at the striking point Q0 and the specific en-
tion, time to half value, charge transfer at the striking point ergy W0:
and specific energy.
∞
(4)
LIGHTNING RETURN STROKE CURRENT –
DOUBLE-EXPONENTIAL FUNCTION
∫ i ⋅ dt = Q
0
0

The lightning current approximation of the first return ∞

stroke is depicted in Fig. 1, where I0 is the current peak 2
(5)
value, t0 is the virtual starting time, t1 is the time to 10 % of ∫i 0
⋅ dt = W0
peak value, t2 is the time to 90 % of peak value, th is the
total time to half value of the peak value, tmax is the time to From equations (2-5), the following four normalized
the peak value, T1 is the front duration and T2 is the time to nonlinear equations can be obtained:
half value [4].
1
R1 =
0.9 ⋅ η
(
⋅ e −α⋅t 2 − e −β⋅t 2 − 1 ) (6)

1
R2 =
0.5 ⋅ η
(
⋅ e −α⋅t h − e −β⋅t h − 1) (7)

I 0 ⎛ 1 1 ⎞
R3 = ⋅ ⎜ − ⎟ − 1 (8)
Q 0 ⋅ η ⎝ α β ⎠

I 02
⎛ 1 2 1 ⎞
R4 = 2
⋅ ⎜ − + ⎟ − 1 (9)
η ⋅ W0 ⎝ 2 ⋅ α α + β 2 ⋅ β ⎠

Fig.1 – Lightning current approximation of the first return stroke.

1
Univ. of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, 21000 Split, Croatia, e-mail: dlovric@fesb.hr
2
Univ. of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, 21000 Split, Croatia, e-mail: vujevic@fesb.hr
3
Univ. of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, 21000 Split, Croatia, e-mail: tmodric@fesb.hr
LEAST SQUARES ESTIMATION OF DOUBLE- The parameter η is computed from the linear equation
EXPONENTIAL FUNCTION PARAMETERS given in Fig. 2. Then the parameters α and β are computed
Double-exponential function parameters can be estimat- by the Marquardt method from the corresponding set of
ed in four cases by solving a set of m nonlinear equations nonlinear equations in the rth iteration.
using the Marquardt method [3], [5], depending on the The nonlinear equation for computing the parameter tmax
available input requirements (Fig. 2): can be obtained from the following requirement:
Case 1: m = 2, E1 = R1 and E2 = R2 di
=0 (10)
Case 2: m = 3, E1 = R1, E2 = R2 and E3 = R3 dt t = t max
Case 3: m = 3, E1 = R1, E2 = R2 and E3 = R4
The following normalized nonlinear equation for com-
Case 4: m = 4, E1 = R1, E2 = R2, E3 = R3 and E4 = R4 putation of tmax can be obtained from (10):
In Fig. 2, a flowchart is presented that describes the es- k r
α⋅k −1 t max r
β⋅k −1 t max
timation of the double-exponential function parameters η, α Fmax = − r α ⋅ e − + r β ⋅ e− (11)
and β introduced in (1). In each rth iteration, parameters
tmax, η and t1 have to be computed, where the auxiliary pa- When estimating tmax, the initial value is taken to be
rameters tmax and t1 are computed by solving the corre- tmax = 0.9·T1.
sponding nonlinear equation. The abbreviation MSE in The nonlinear equation of computing the parameter t1
Fig. 2 stands for Marquardt method for a Single nonlinear can be obtained from the following requirement:
Equation. It is used for estimation of values of t1 or tmax.
i = 0.1⋅ I0 for t = t1 (12)

Fig.2 – Least squares estimation of double-exponential function parameters.

 Using (2), the following normalized nonlinear equation the method described in the previous section. The following
for computation of t1 can be obtained: parameters have been obtained: η = 0.9511, α = 2121.76 s-1
and β = 245303.6 s-1. Double-exponential function with
1 ⎛ − r α⋅k −1 t1 r k −1 these parameters is depicted in Fig. 3.
k
F1 = ⋅ ⎜ e − e − β⋅ t1 ⎞⎟ − 1 (13)
0.1 ⋅ η ⎝ ⎠

When estimating t1, the initial value is taken to be

t1 = 0.1·T1.
In Fig. 2, the matrix [D] is a diagonal matrix whose di-
agonal elements are identical to the diagonal elements of
matrix [A] defined by the following equation:

[A] = [J]T ⋅ [J] (14)

where the Jacobian matrix [J] can be computed by:

Fig.3 – Double-exponential function approximation of the first posi-
(
⎡ ∂E1 r −1 α, r −1 β
⎢
) ∂E1 (r −1 α, r −1 β) ⎤⎥ tive impulse.
⎢ ∂ r −1α ∂ r −1β ⎥ Other often used lightning current waveshapes some-
⎢ ⎥ times used for designing low-voltage power lines within
[J ] = ⎢   ⎥ (15)
⎢ ⎥ structures are the T1/T2 = 0.2/5 µs waveshape, the
(
⎢ ∂E m r −1 α, r −1 β ) (
∂E m r −1 α, r −1 β ⎥ ) T1/T2 = 4/16 µs waveshape and T1/T2 = 1.2/50 µs
⎢ ⎥ waveshape [1, 6]. Since these waveshapes are only defined
⎣⎢ ∂ r −1α ∂ r −1β ⎦⎥ by T1 and T2 values, only two nonlinear equations are sim-
ultaneously solved. Results of the estimation for these three
Vector {B} can be computed using following expression:
waveshapes are presented in Table I. Corresponding dou-
ble-exponential functions approximating these waveshapes
⎧ E
⎪ 1
(r −1 α, r −1 β) ⎫⎪ are depicted in Fig. 4. All functions have a current peak
{B} = ⎪⎨ 
⎪
⎬ (16) value of I0 = 1 A for plotting clarity purposes. Peak values
⎪ ⎪ can be changed accordingly since the parameters in Table I
⎪E m
⎩
( r −1 r −1
α, β ⎪
⎭
) are independent of the peak current value.
Table I
Partial derivatives of nonlinear equations (6-9) which
Double-exponential function parameters for fast-decaying light-
are required for computation of matrix [J] in (15) can be ning current waveshapes.
computed analytically using the following expressions:
Double-exponential function
Waveshapes
∂R 1 t ⋅ e − α⋅ t 2 ∂R 1 t 2 ⋅ e −β⋅t 2 T1/T2
parameters
=− 2 ; = (17) η α (s-1) β (s-1)
∂α 0.9 ⋅ η ∂β 0.9 ⋅ η
0.2/5 µs 0.93269 152921.46 11887358.7
∂R 2 t ⋅ e − α⋅ t h ∂R 2 t h ⋅ e −β⋅t h 4/16 µs 0.27475 117598.38 252722.5
=− h ; = (18) 1.2/50 µs 0.95847 14732.18 2080312.7
∂α 0.5 ⋅ η ∂β 0.5 ⋅ η
∂R 3 I0 ∂R 3 I0
=− 2
; = (19)
∂α Q0 ⋅ η ⋅ α ∂β Q0 ⋅ η ⋅ β2

∂R 4 I 02 ⎡ 2 1 ⎤
= ⋅
2 ⎢ 2
− ⎥
∂α W0 ⋅ η ⎣ (α + β) 2 ⋅ α 2 ⎦
(20)
∂R 4 I 02 ⎡ 2 1 ⎤
= ⋅
2 ⎢ 2
− ⎥
∂β W0 ⋅ η ⎣ (α + β) 2 ⋅ β 2 ⎦

NUMERICAL EXAMPLES
The first numerical example features Case 4, i.e. the
simultaneous solving of four nonlinear equations. The input
data taken from IEC 62305-1 Ed. 2 [4] represent the maxi-
mum values of lightning current quantities of the first posi-
tive impulse for Lightning Protection Level III-IV:
I0 = 100 kA, T1/T2 = 10/350 µs, Q0 = 50 C and
W0 = 2.5 MJ/Ω. Double-exponential function parameters
are estimated using a computer program that implements
Fig.4 – Double-exponential function approximation of fast-decaying
lightning current waveshapes.
However, in the case of communication lines which
have a different exposure to lightning than power lines,
different waveshapes are used for designing lightning pro-
tection system. These waveshapes are characterized by a
relatively sharp rise followed by a very slow current decay:
 the T1/T2 = 10/700 µs waveshape and the exponential function has the steepest rise in the time t = 0
T1/T2 = 10/1000 µs waveshape [6]. Again, these and can not approximate the recorded waveshape accurate-
waveshapes are only defined by T1 and T2 values, so only ly. Much better results can be obtained using the Heidler
two nonlinear equations are simultaneously solved. function [3, 9] or Javor function [10]. The input data of the
Results of the estimation for these two waveshapes are recorded waveshape current taken from [8] is: I0 = -1 A, T1
presented in Table II. Double-exponential functions approx- = 8 µs and T2 = 100 µs. The resulting double-exponential
imating these waveshapes are depicted in Fig. 5. function parameters are: η = 0.86481, α = 8421.53 s-1 and
β = 265585.9 s-1.
Table II
Double-exponential function parameters for slow-decaying light-
ning current waveshapes.
Double-exponential function
Waveshapes
parameters
T1/T2
η α (s-1) β (s-1)
10/700 µs 0.97423 1028.39 257923.7
10/1000 µs 0.98135 712.41 262026.6

Fig.7 – Recorded lightning current and its double-exponential func-

tion approximation.

Fig.5 – Double-exponential function approximation of slow-
decaying lightning current waveshapes.
In addition to standardized lightning current, the pre-
sented least squares method can easily be used to approxi-
mate the waveshapes of various recorded impulse stroke
currents. The recorded waveshape taken from [7] is depict-
ed on Fig. 6, along with the double-exponential function
approximation. The input data of the recorded waveshape
current taken from [7] is: I0 = 5 A, T1 = 9.3 µs and T2 = 90
µs. The resulting double-exponential function parameters
are: η = 0.80917, α = 10054.37 s-1 and β = 197765.3 s-1.
Fig.6 – Recorded impulse current and its double-exponential func- [8] K. Berger, R. B. Anderson, H. Kroninger: “Parameters of
tion approximation.
In the following example a typical negative first stroke [9] D. Lovrić, D. Vujević, T. Modrić: “On the estimation of
current waveshape is considered. The recorded waveshape
taken from [8] is depicted on Fig. 7, along with the double-
exponential function approximation. One can observe from
this figure the inadequacy of the double-exponential func-
tion approximation. The recorded lightning current is char-
acterized by a slow rise at the very beginning followed by a
much steeper rise. On the other hand, the double-
CONCLUSION
In this paper, a robust and effective algorithm for the
least squares estimation of double-exponential function
parameters is presented. Using this algorithm various
standardized and recorded lightning current waveshapes
can be approximated by the double-exponential function.
This algorithm can be easily modified to estimate the pa-
rameters of an arbitrary lightning current function.
REFERENCES
[1] V. Cooray: “Lightning protection”, V. Cooray, 2007, London,
pp. 67-72.
[2] C. E. R. Bruce, R. H. Golde: “The lightning discharge”, J. Inst.
Elect. Eng, Vol. 88, Part 2, 1941, pp. 487-520.
[3] S. Vujević, D. Lovrić, I. Jurić-Grgić: “Least squares estimation
of Heidler function parameters”, European Transactions on
Electrical Power, Vol. 21, 2011, pp. 329-344.
[4] IEC 62305-1 Ed. 2, Protection against lightning – Part 1: Gen-
eral principles, 2010.
[5] D.W. Marquardt: “An algorithm for least-squares estimation of
nonlinear parameters”, Journal of the Society for Industrial
and Applied Mathematics, Vol. 11, No. 2, 1963, pp. 431-441.
[6] M.A. Uman: “The art and science of lightning protection”,
Cambridge University Press, 2008, New York.
[7] Z. Stojković et al.: “Sensitivity analysis of experimentally
determined grounding grid impulse characteristics”, IEEE
Transactions on Power Delivery, Vol. 13, No. 4, 1998, pp.
1136-1142.
lightning flashes”, ELECTRA, No. 41, 1975, pp. 23-38.
Heidler function parameters for reproduction of various
standardized and recorded lightning current waveshapes”, In-
ternational Transactions on Electrical Energy Systems, Vol.
23, 2013, pp. 290-300.
[10] V. Javor: “New function for representing IEC 62305 Standard
and other typical lightning stroke currents”, Journal of Light-
ning Research, Vol. 4, 2012, pp. 50-59.

View publication stats