least squares heidler
TRANSCRIPT
EUROPEAN TRANSACTIONS ON ELECTRICAL POWEREuro. Trans. Electr. Power 2011; 21:329–344Published online 28 April 2010 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/etep.445
Least squares estimation of Heidler function parameters
*C
NayE-
Co
Slavko Vujevic*, Dino Lovricy and Ivica Juric-Grgic
University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture,
Croatia
SUMMARY
The aim of the proposed paper is to present an effective numerical algorithm for the computation of Heidlerfunction parameters. The basic six channel-base current quantities can be prescribed: current peak value,front duration, time to half value, current steepness factor, charge transfer at the striking point and specificenergy. The approximation of the unknown three lightning current parameters for Heidler function isachieved using the least squares method. For the purpose of better convergence, the Marquardt least squaresmethod has been applied. The proposed algorithm can be successfully applied for lightning currentmodelling in power engineering as well as in research on electromagnetic compatibility. Copyright #
2010 John Wiley & Sons, Ltd.
key words: lightning current; Heidler function parameters; least squares method; Marquardt method
1. INTRODUCTION
In recent years, a number of experiments have been conducted with the purpose of directly measuring
and recording the current waveform produced by a lightning strike. The most important data were
accumulated and analysed by Berger, who recorded on top of a telecommunications tower situated on
the mountain San Salvatore in Switzerland [1,2]. Based on this data, a new analytical function has been
proposed by Heidler in Reference [3]. A variant of this new analytical function, for a current steepness
factor n¼ 10, has been incorporated in IEC 62305–1 as the power function recommended for lightning
current modelling in both lightning research and engineering applications [4]. This new function
permits a good separation of the characteristic lightning current quantities.
The computation of all Heidler function parameters simultaneously requires solving a system of
nonlinear equations. In Reference [5], the authors claim that the system of nonlinear equations often
has no real solution, and they instead use a set of approximated functions that represent the current rise
time and current decay time. In this way, they avoid solving the nonlinear integral equations. The
authors then use a graphical algorithm in combination with these approximations to obtain Heidler
function parameters.
In this paper, Heidler function parameters are computed by simultaneously solving a system of two,
three or four nonlinear equations. The numerical algorithm developed is a robust and reliable tool,
which can be used to compute Heidler function parameters for any combination of the input
requirements. Furthermore, with minor modifications, this universal numerical algorithm can be used
to compute the parameters of any other lightning current approximation function, for example the often
orrespondence to: Slavko Vujevic, University of Split, Faculty of Electrical Engineering, Mechanical Engineering and
val Architecture, Croatia
mail: [email protected]
pyright # 2010 John Wiley & Sons, Ltd.
330 S. VUJEVIC, D. LOVRIC AND I. JURIC-GRGIC
used double exponential approximation. In relation to the prescribed set of lightning current quantities,
four different cases are possible.
2. HEIDLER REPRESENTATION OF THE LIGHTNING CURRENT
According to Reference [4], the lightning current approximation of the first stroke can be represented
by the plot shown in Figure 1, where I0 is the current peak value, t0 is the virtual starting time, t1 is the
time to 10% of 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 the peak value, T1 is the front duration and T2 is the (virtual) time to half
value.
Figure 1. Lightning current approximation by Heidler function.
The current function most frequently used in the lightning research is Heidler function, which can be
expressed by the following equations [3–5]:
i tð Þ ¼ I0
h� xðtÞ � yðtÞ ¼ I0
h� x � y (1)
x ¼ x tð Þ ¼tt1
� �n
1 þ tt1
� �n (2)
y ¼ y tð Þ ¼ e� t
t2 (3)
where h is the correction coefficient of the current peak value, x(t) is the rise function, y(t) is the decay
function, t1 is the time constant determining current rise-time, t2 is the time constant determining
current decay-time, and n is the current steepness factor.
Copyright # 2010 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power 2011; 21:329–344
DOI: 10.1002/etep
HEIDLER FUNCTION PARAMETERS 331
3. SET OF NONLINEAR EQUATIONS
According to Figure 1, the two basic requirements for the estimation of lightning current parameters h,
t1 and t2 can be written as:
i ¼ 0:9 � I0 for t ¼ t2 (4)
i ¼ 0:5 � I0 for t ¼ th (5)
Two additional requirements are deduced from the value of the charge transfer at the striking point
Q0 and the specific energy W0:
Z10
i � dt ¼ Q0 (6)
Z10
i2 � dt ¼ W0 (7)
The specific energy W0 represents the energy dissipated by the lightning current in a unit resistance
[4].
From Equations (1) and (4)–(7), the following four normalized nonlinear equations are obtained:
R1 ¼ 1
0:9 � h � x t2ð Þ � y t2ð Þ � 1 ¼ 0 (8)
R2 ¼ 1
0:5 � h � x thð Þ � y thð Þ � 1 ¼ 0 (9)
R3 ¼ I0
Q0 � h�Z10
xðtÞ � yðtÞ � dt � 1 ¼ 0 (10)
R4 ¼ I20
W0 � h2�Z10
x2ðtÞ � y2ðtÞ � dt � 1 ¼ 0 (11)
Integrals in nonlinear Equations (10) and (11) are solved using a combination of the Simpson’s rule
for interval t2 [0, tg] and analytical integration for t> tg:
R3 ¼ I0
Q0 � h�Ztg0
xðtÞ � yðtÞ � dt
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}Simpson0s rule
þ I0 � t2
Q0 � h� e�
tgt2 � 1 (12)
R4 ¼ I20
W0 � h2�Ztg0
x2ðtÞ � y2ðtÞ � dt
|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}Simpson0s rule
þ I20 � t2
W0 � h2 � 2� e�
2�tgt2 � 1 (13)
where:
tg ¼ t1 �ffiffiffiffiffiffiffi104n
p(14)
Copyright # 2010 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power 2011; 21:329–344
DOI: 10.1002/etep
332 S. VUJEVIC, D. LOVRIC AND I. JURIC-GRGIC
is chosen in such a way that, for t> tg, the rise function x(t)� 1 and Heidler function (1) can be quite
accurately approximated by:
i tð Þ � I0
h� yðtÞ ¼ I0
h� e�
tt2 (15)
According to Equations (2) and (14), the rise function x(tg)¼ 0.9999 and Heidler function (1), for
t> tg, is approximated by Equation (15) with a percentage error less than 0.01%.
4. LEAST SQUARES ESTIMATION OF t1 AND t2
In Figure 2, a flow chart is presented that describes the estimation of the lightning current parameters h,
t1 and t2 introduced by Equations (1)–(3). Parameters t1 and t2 are computed by the Marquardt least
squares method from the corresponding system of nonlinear equations, while the parameter h is
computed from a linear equation at the beginning of the iterative procedure as well as at the end of each
iteration.
The current steepness factor n, current peak value I0, front duration T1 and time to half value T2 are
prescribed lightning current quantities for all cases. However, in some cases the charge transfer at the
striking point Q0 and/or the specific energy W0 can be additional input data. The estimation of Heidler
function parameters t1 and t2 is based on the Marquardt least squares method (also known as
Levenberg–Marquardt method) [6,7].
In each rth iteration, the parameters tmax, h and t1 have to be computed, where the auxiliary
parameters tmax and t1 are computed by solving the corresponding nonlinear equations, while the
coefficient h can be computed using Equation (B.2) derived in Appendix B:
h ¼ rh ¼tmaxrt1
� �n
1 þ tmaxrt1
� �n � e�tmax
rt2 (16)
The lightning current parameters t1 and t2 can be estimated in four cases depending on the available
input requirements. The requirements described by Equations (4) and (5) are known in all cases, while
the value of the charge transfer at the striking point Q0 and the specific energy W0 described by
Equations (6) and (7) are known in some cases. This gives a total of four possible combinations of input
requirements. In all cases, the unknown parameters t1 and t2 can be estimated by solving the set of m
nonlinear equations using the least squares method, where one needs to minimize the objective
function representing the sum of squares of several nonlinear functions depending on the input set of
requirements:
S ¼Xmi¼1
E2i ¼ minimum (17)
where:
Case 1: m¼ 2; E1¼R1 and E2¼R2
Case 2: m¼ 3; E1¼R1, E2¼R2 and E3¼R3
Case 3: m¼ 3; E1¼R1, E2¼R2 and E3¼R4
Case 4: m¼ 4; E1¼R1, E2¼R2, E3¼R3 and E4¼R4
The system of nonlinear Equations (8), (9), (12) and (13) is solved simultaneously using the
Marquardt least squares method:
Aþ l � D½ � � Dt1
Dt2
� �¼ Aþ 3 � D½ � � Dt1
Dt2
� �¼ B (18)
where l is an adjustable positive parameter which is used to control the iteration.
Copyright # 2010 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power 2011; 21:329–344
DOI: 10.1002/etep
Figure 2. Estimation of the lightning current parameters.
Copyright # 2010 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power 2011; 21:329–344
DOI: 10.1002/etep
HEIDLER FUNCTION PARAMETERS 333
334 S. VUJEVIC, D. LOVRIC AND I. JURIC-GRGIC
Observing the left hand side, as l!1, matrix A becomes insignificant with respect to the dominant
elements of the matrix l�D. Consequently, the solution of the current iteration is influenced only by the
elements of the matrix l�D, and this reduces the Marquardt least squares method to the steepest
descents method. On the other hand, if the parameter l! 0, then the solution of the current iteration is
influenced only by the elements of the matrix A and this reduces the Marquardt least squares method to
the Gauss–Newton or generalized least squares method [7]. The steepest descent method is known to
be convergent but slow, while the Gauss–Newton method is rapid but the convergence is less reliable.
The main advantage of the Marquardt least squares method [6] is the possibility of choosing the
parameter l as to follow the Gauss–Newton method as much as possible to ensure rapid convergence,
while at the same time retaining certain characteristics of the steepest descent method to prevent
divergence. After a number of numerical experiments, the parameter l was chosen to have the value 3,
which enables rapid and reliable convergence in all cases.
The matrix D is a diagonal matrix whose diagonal elements are identical to the diagonal elements of
matrix A [8,9], which is defined by the following equation:
A ¼ JT � J (19)
where the Jacobian matrix J can be computed approximately:
J ¼
DE1
Dt1
DE1
Dt2
..
. ...
DEm
Dt1
DEm
Dt2
2664
3775 (20)
Approximations of partial derivatives are used, and they are computed as follows:
DEi
Dt1
¼ Ei t1 þ 1; t2ð Þ � Ei t1; t2ð Þ; i ¼ 1; 2; :::;m (21)
DEi
Dt2
¼ Ei t1; t2 þ 1ð Þ � Ei t1; t2ð Þ; i ¼ 1; 2; :::;m (22)
where m is the index number of Equations (8), (9), (12) and (13) as stated before.
According to the Marquardt least squares method, the matrix on the right hand side in the Equation
(18) is:
B ¼ �JT �
E1
E2
..
.
Em
8>><>>:
9>>=>>; (23)
The iterations will continue until the conditions for the convergence of parameters t1 and t2 are
achieved. In all cases the condition for the convergence of the values of t1 and t2 is:
S ¼Xmi¼1
E2i < 10�14 (24)
where m¼ 2 for the first case, m¼ 3 for the second and third cases and m¼ 4 for the fourth case. When
this condition is achieved the iteration process is over, and the final values of parameters t1, t2 and h are
acquired.
5. COMPUTATION OF PARAMETER tMAX
As presented in Figure 3, the input data for computation of the unknown auxiliary parameter tmax are
the current steepness factor n and the time constants obtained in the rth iteration rt1 and rt2.
Copyright # 2010 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power 2011; 21:329–344
DOI: 10.1002/etep
Figure 3. Computation of tmax.
HEIDLER FUNCTION PARAMETERS 335
In each rth and kth iteration, the parameter tmax can be computed using nonlinear Equation (C.7)
derived in Appendix C:
Fmax ¼ktmax
rt1
� �nþ1
þktmax
rt1
� �� n � rt2
rt1
¼ 0 (25)
The nonlinear Equation (25) is solved using a special case of the Marquardt least squares method for
a single nonlinear equation. Analogously, as in Equation (18), it can be stated:
J2max � 1 þ lð Þ � Dtmax ¼ B (26)
where:
B ¼ �Jmax � Fmax (27)
Copyright # 2010 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power 2011; 21:329–344
DOI: 10.1002/etep
336 S. VUJEVIC, D. LOVRIC AND I. JURIC-GRGIC
In each rth and kth iteration, the Jacobian Jmax can be computed using Equation (C.9) derived in
Appendix C:
Jmax ¼ @Fmax
@ðktmaxÞ¼ 1
rt1
� 1 þ nþ 1ð Þ �ktmax
rt1
� �n (28)
where Fmax is described by Equation (25).
Then the growth of the parameter Dtmax can be computed by:
Dtmax ¼ � Fmax
Jmax � 1 þ lð Þ ¼ � Fmax
4 � Jmax
(29)
where l ¼ 3.
The value of tmax for the kth iteration is computed from the following expression:
ktmax ¼ k�1tmax þ Dtmax (30)
As stated in Equation (30), the value of parameter tmax will grow in every iteration by the value
of Dtmax. The iterations will continue until the condition for the convergence of the parameter tmax is
achieved:
S ¼ F2max < 10�14 (31)
6. COMPUTATION OF PARAMETER t1
The input data for computation of the unknown auxiliary parameter t1 are the current steepness factor n,
coefficient h and the time constants obtained in the rth iteration rt1 and rt2. The procedure for
estimating the parameter t1 is presented in Figure 4.
In each rth and kth iteration, the parameter t1 can be computed using nonlinear Equation (D.3)
derived in Appendix D:
F1 ¼ 0:1 � hþkt1rt1
� �n
� 0:1 � h� e�
k t1rt2
� �¼ 0 (32)
The nonlinear Equation (32) can be solved in the same way as described for the
parameter tmax. Again, in the first iteration (k¼ 0), the first step is to assume the value of the
variable t1. Analogously, as in Equations (26)–(29), the growth of the iteration for the parameter t1 can
be found from:
Dt1 ¼ � F1
J1 � 1 þ lð Þ ¼ � F1
4 � J1
(33)
where l ¼ 3.
In each rth and kth iteration, the Jacobian J1 can be computed using Equation (D.5) derived in
Appendix D:
J1 ¼ @F1
@ðkt1Þ¼ n
rt1
�kt1rt1
� �n�1
� 0:1 � h� e�
k t1rt2
� �þ 1
rt2
�kt1rt1
� �n
�e�k t1rt2 (34)
where F1 is described by Equation (32).
The value of t1 for the kth iteration is then computed from the following expression:
kt1 ¼ k�1t1 þ Dt1 (35)
As presented in Equation (35), the value of parameter t1 will grow in each iteration by the
value of Dt1. The iterations will continue until the condition for the convergence of parameter t1 is
achieved:
S ¼ F21 < 10�14 (36)
Copyright # 2010 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power 2011; 21:329–344
DOI: 10.1002/etep
Figure 4. Computation of t1.
HEIDLER FUNCTION PARAMETERS 337
7. COMPUTING THE MAXIMUM OF THE CURRENT STEEPNESS
The maximum of the current steepness didt
� �max
can be computed from the requirement:
d2i
dt2
� �t¼ts
¼ 0 (37)
where ts is the time when the current derivative attains its maximum. The flow chart of the procedure
for computing the maximum of the current steepness is presented in Figure 5. In the first step, the
unknown value ts is computed using the Marquardt least squares method, while, in the second step, the
maximum of the current steepness can be computed using a simple linear equation.
Copyright # 2010 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power 2011; 21:329–344
DOI: 10.1002/etep
Figure 5. Computation of the maximum of the current steepness.
338 S. VUJEVIC, D. LOVRIC AND I. JURIC-GRGIC
In each kth iteration of the Marquardt least squares method, according to Equations (1)–(3) and
Equation (37), the following nonlinear equation Fs and its Jacobian Js can be obtained:
Fs ¼ t22 � f 00 � 2 � t2 � f 0 þ f ¼ 0 (38)
Js ¼@Fs
@ðktsÞ¼ t2
2 � f 000 � 2 � t2 � f 00 þ f 0 (39)
where the function f and its derivatives are given by:
f ¼ktst1
� �n
1 þ ktst1
� �n (40)
f 0 ¼ n
t1
�ktst1
� �n�1
1 þ ktst1
� �nh i2(41)
Copyright # 2010 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power 2011; 21:329–344
DOI: 10.1002/etep
HEIDLER FUNCTION PARAMETERS 339
k� �n�2 n� 1 � nþ 1ð Þ � kts
� �n
f 00 ¼ n
t21
� ts
t1
� t1
1 þ ktst1
� �nh i3(42)
f 000 ¼ n
t31
�kts
t1
� �n�3
�n� 2ð Þ � n� 1ð Þ � 4 � n2 � 1ð Þ � kts
t1
� �n
� nþ 1ð Þ � nþ 2ð Þ � ktst1
� �2�n
1 þ ktst1
� �nh i4(43)
The value of parameter ts in each kth iteration can be computed using the following relations:
kts ¼ k�1ts þ Dts (44)
Dts ¼ � Fs
Js � 1 þ lð Þ ¼ � Fs
4 � Js(45)
where l ¼ 3. The iterations will continue until the condition for the convergence of the parameter ts is
achieved:S ¼ F2
s < 10�14 (46)
Using the computed ts, the maximum of the current steepness is given by:
di
dt
� �t¼ts
¼ di
dt
� �max
¼ I0
h� f 0 � f
t2
� �� e�
tst2 (47)
In Heidler grapho-analytical algorithm [5], the maximum of the current steepness is computed using
the approximation y(t) � 1, where y(t) is decay function defined by Equation (3). However, in this
paper, decay function approximation is not used.
8. NUMERICAL EXAMPLES
The following three test cases are shown to illustrate the application of the proposed procedure.
8.1. Example 1
In the following example, the unknown parameters of Heidler function are computed by
simultaneously solving the nonlinear equations E1, E2 , E3 and E4 (Case 4). The input data are
taken from Reference [4], and they represent the maximum values of lightning current quantities for
Table I. Computed parameters of Heidler function for 10/350ms for I0¼ 200 kA, Q0¼ 100 C andW0¼ 10 MJ/V.
n t1 (ms) t2 (ms) h didt
� �max
ðkA=msÞ
3 5.824 463.498 0.936 30.3494 7.731 468.095 0.940 28.7235 9.630 470.705 0.941 28.0266 11.516 472.391 0.940 27.6827 13.403 473.546 0.938 27.4798 15.286 474.391 0.935 27.3569 17.168 475.036 0.933 27.275
10 19.048 475.544 0.930 27.22211 20.928 475.954 0.927 27.18512 22.806 476.291 0.924 27.15913 24.684 476.574 0.921 27.14114 26.561 476.815 0.917 27.12715 28.438 477.021 0.914 27.118
Copyright # 2010 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power 2011; 21:329–344
DOI: 10.1002/etep
Figure 6. Heidler representation of the first short stroke 10/350ms for I0¼ 200 kA, Q0¼ 100 C andW0¼ 10 MJ/V.
340 S. VUJEVIC, D. LOVRIC AND I. JURIC-GRGIC
the first short stroke for Lightning Protection Level I: I0¼ 200 kA, T1/T2¼ 10/350ms, Q0¼ 100 C and
W0¼ 10 MJ/V. Parameters t1, t2, h and (di/dt)max are computed using a computer program written in
MATLAB that implements the Marquardt least squares method described previously. The results of the
computations are shown in Table I for the current steepness factor n 2 3; 4; :::; 15f g, while the Heidler
lightning function for n¼ 5, 10, 15 is shown in Figure 6.
8.2. Example 2
In this example, the unknown parameters of Heidler function are computed by simultaneously solving
the nonlinear equations E1 and E2 (Case 1). The input data are taken from Reference [4], and they
represent the maximum values of lightning current quantities for the subsequent short stroke for
Lightning Protection Level I: I0¼ 50 kA, T1/T2¼ 0.25/100ms. Parameters t1, t2, h and (di/dt)max are
computed using the Marquardt least squares method, as before, and are presented in Table II for the
current steepness factor n 2 3; 4; :::; 15f g, while the Heidler lightning function for n¼ 5, 10, 15 is
shown in Figure 7.
Table II. Computed parameters of Heidler function of the subsequent stroke 0.25/100ms for I0¼ 50 kA.
n t1 (ms) t2 (ms) h didt
� �max
ðkA=msÞ
3 0.129 142.442 0.991 328.6164 0.177 142.777 0.992 303.3595 0.224 142.955 0.993 292.3366 0.270 143.064 0.993 286.5297 0.317 143.137 0.993 283.0948 0.363 143.189 0.993 280.8959 0.409 143.228 0.993 279.404
10 0.455 143.258 0.993 278.34711 0.501 143.282 0.993 277.57212 0.547 143.302 0.992 276.98713 0.593 143.318 0.992 276.53514 0.639 143.332 0.992 276.17915 0.685 143.343 0.992 275.894
Copyright # 2010 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power 2011; 21:329–344
DOI: 10.1002/etep
Figure 7. Heidler representation of the subsequent stroke 0.25/100ms for I0¼ 50 kA.
Table III. Comparison of the computed Heidler function parameters for typical subsequent return stroke.
h t1 (ms) t2 (ms)
Marquardt least squares method 0.837 0.265 3.607Heidler grapho-analytical algorithm [5] 0.84 0.27 3.5Percentage difference (%) �0.358 �1.887 2.966
HEIDLER FUNCTION PARAMETERS 341
8.3. Example 3
As a final example, in Table III, a comparison is made between Heidler function parameters computed
by the Heidler grapho-analytical algorithm [5] and those computed by the Marquardt least squares
method for typical subsequent return stroke with input data: I0¼ 13 kA, n¼ 5 and Q0¼ 50 mC.
In this case, the results computed by the proposed numerical algorithm and the Heidler grapho-
analytical algorithm [5] are relatively close. Although the relatively close results in this case, Heidler
grapho-analytical algorithm does not enable the separation of the input parameters, while the proposed
numerical algorithm can find the best estimation of lightning current parameters depending on the used
set of input parameters.
9. CONCLUSION
In this paper, the lightning current parameters of Heidler function are computed using the Marquardt
least squares method to solve several different systems of nonlinear equations. Four cases for obtaining
the lighting current parameters are presented for the chosen set of input parameters. In the first case,
two nonlinear equations require simultaneous solving, in the second and third cases, there are three
nonlinear equations, and in the fourth case, there are four nonlinear equations to solve simultaneously.
In all cases, four basic input parameters are prescribed, while an additional two input parameters—
charge transfer at the striking point and the specific energy—may be prescribed. The iterative process
of computing the lightning current parameters is illustrated in detail in four flow charts, which describe
Copyright # 2010 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power 2011; 21:329–344
DOI: 10.1002/etep
342 S. VUJEVIC, D. LOVRIC AND I. JURIC-GRGIC
the various segments of the solution process and thus provide general guidance for implementing these
algorithms in a computer program.
The proposed numerical algorithm represents an advancement in relation to the Heidler grapho-
analytical algorithm. The proposed algorithm is much more robust and suitable for software
implementation than the Heidler grapho-analytical algorithm. Besides these advantages, the proposed
numerical algorithm can take into account the specific energy of the lightning strike as an additional
input parameter. This parameter is important in determining the mechanical effects of the lightning
stroke, more specifically the thermal effects. Since each of the lightning current parameters tends to
dominate each failure mechanism, introduction of the specific energy as the input parameter is an
advancement in analysing the mechanical effects of lightning strike. Furthermore, disregarding the fact
that this numerical algorithm improves the lightning current parameter computation, it also enables the
separation of the input parameters, while the Heidler grapho-analytical algorithm does not allow for the
separation of the input parameters.
Unlike the Heidler grapho-analytical algorithm, the proposed numerical algorithm can be easily
modified to accommodate the need of computing the parameters of any other lightning current function
approximation used in electric power engineering as well as in research on electromagnetic
compatibility.
REFERENCES
1. Berger K, Anderson RB, Kroninger H. Parameters of lightning flashes. Electra 1975; 41:23–37.
2. Anderson RB, Erikson AJ. Lightning parameters for engineering application. Electra 1980; 69:65–101.
3. Heidler F, Cvetic J. A class of analytical functions to study the lightning effects associated with the current front.
European Transaction on Electric Power (ETEP) 2002; 12(2):141–150.
4. IEC 62305-1: Protection Against Lightning - Part 1: General Principles, 2006.
5. Heidler F, Cvetic J, Stanic BV. Calculation of lightning current parameters. IEEE Transactions on Power Delivery
1999; 14(2):399–404.
6. Marquardt DW. An algorithm for least-squares estimation of nonlinear parameters. Journal of the Society for
Industrial and Applied Mathematics 1963; 11(2):431–441.
7. Lill SA. Optimization in Action. Academic Press: London, 1976.
8. Vujevic S, Kurtovic M. Direct and iterative automatic interpretation of resistivity sounding data. International Journal
for Engineering Modelling 1992; 5(3–4):65–118.
9. Vujevic S. A combined analysis of earthing grids in karstic soil. PhD Thesis, University of Split, Faculty of Electrical
Engineering, Mechanical Engineering and Naval Architecture: Split 1994. (in Croatian).
APPENDIX A: NOMENCLATURE
(di/dt)max m
Copyright # 2
aximum of the current steepness
F1 s
tarting nonlinear equation for computing the parameter t1 Fmax s tarting nonlinear equation for computing the parameter tmaxFs s
tarting nonlinear equation for computing the parameter ts I0 c urrent peak valuen c
urrent steepness factorQ0 c
harge transfer at the striking pointT1 f
ront durationT2 (
virtual) time to half valuet0 v
irtual starting timet1 t
ime to 10% of peak valuet2 t
ime to 90% of peak valuetg t
ime parameter used for approximation of Heidler functionth t
otal time to half value of the peak valuetmax t
ime to the peak valuets t
ime when the current derivative attains its maximumW0 s
pecific energy010 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power 2011; 21:329–344
DOI: 10.1002/etep
HEIDLER FUNCTION PARAMETERS 343
x(t) r
Copyright # 2
ise function
y(t) d
ecay functionh c
orrection coefficient of the current peak valuel a
rbitrarily chosen constant with a positive valuet1 t
ime constant determining current rise-timet2 t
ime constant determining current decay-timeAPPENDIX B
The current i(t) attains its peak value I0 at the time tmax. After substituting I¼ I0 and t¼ tmax in
Equations (1)–(3), the following equation is obtained:
I0 ¼I0h�
tmax
t1
� �n
1þ tmaxt1
� �n � e�tmax
t2 (B.1)
After dividing both sides of Equation (B.1) by I0/h, the equation for computing correction
coefficient h is obtained:
h ¼tmaxt1
� �n
1þ tmaxt1
� �n � e�tmax
t2 (B.2)
APPENDIX C
The time derivative of Heidler function described by Equations (1)–(3) equals zero for t¼ tmax when
the lightning current attains its peak value:
di
dt
t¼tmax
¼ I0h� dx
dt� yþ x � dy
dt
� � t¼tmax
¼ 0 (C.1)
where the time derivative of the rise function x described by Equation (2) is:
dx
dt¼
nt1� t
t1
� �n�1� 1þ t
t1
� �nh i� n
t1� t
t1
� �n�1� t
t1
� �n
1þ tt1
� �nh i2 ¼nt1� t
t1
� �n�1
1þ tt1
� �nh i2 (C.2)
while the time derivative of the decay function y described by Equation (3) is:
dy
dt¼ � 1
t2� e�
tt2 (C.3)
After substituting Equations (2), (3), (C.2) and (C.3) into Equation (C.1), the following equation for
t¼ tmax can be obtained:
nt1� tmax
t1
� �n�1�e�
tmaxt2 � 1þ tmax
t1
� �nh i� tmax
t1
� �n� 1t2� e�
tmaxt2
1þ tmax
t1
� �nh i2 ¼ 0 (C.4)
The numerator of Equation (C.4), which is equal to zero, can be written in the next form:
e�tmax
t2 � n
t1� tmax
t1
� �n�1
� tmax
t1
� �n
� 1t2
� tmax
t1
� �2�n� 1t2
" #¼ 0 (C.5)
010 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power 2011; 21:329–344
DOI: 10.1002/etep
344 S. VUJEVIC, D. LOVRIC AND I. JURIC-GRGIC
From Equation (C.5) follows equation:
tmax
t1
� �2�nþ tmax
t1
� �n
� n � t2t1
� tmax
t1
� �n�1
¼ 0 (C.6)
which can be transformed into the next nonlinear equation for computing tmax:
Fmax ¼tmax
t1
� �nþ1
þ tmax
t1
� �� n � t2
t1¼ 0 (C.7)
The Jacobian Jmax of Fmax is defined by:
Jmax ¼@Fmax
@tmax(C.8)
Substitution of Equation (C.7) into Equation (C.8) and partial differentiation with respect to tmax
yields:
Jmax ¼@Fmax
@tmax¼ nþ 1
t1� tmax
t1
� �n
þ 1
t1¼ 1
t1� 1þ nþ 1ð Þ � tmax
t1
� �n (C.9)
APPENDIX D
The current i(t) attains 10% of the peak current value I0 at the time t1. After substituting I¼ 0.1�I0 and
t¼ t1 in Equations (1)–(3), the following equation is obtained:
0:1 � I0 ¼I0h�
t1t1
� �n
1þ t1t1
� �n � e�t1
t2 (D.1)
After dividing both sides of Equation (D.1) by I0/h, follows the next equation:
0:1 � h ¼t1t1
� �n�e�
t1t2
1þ t1t1
� �n (D.2)
Rearranging the previous expression the next nonlinear equation for computing t1 is obtained:
F1 ¼ 0:1 � hþ t1t1
� �n
� 0:1 � h� e�t1
t2
� �¼ 0 (D.3)
The Jacobian J1 of F1 is defined by:
J1 ¼@F1@t1
(D.4)
Substitution of Equation (D.3) into Equation (D.4) and partial differentiation with respect to t1 yields:
J1 ¼@F1@t1
¼ n
t1� t1
t1
� �n�1
� 0:1 � h� e�t1
t2
� �þ 1
t2� t1
t1
� �n
�e�t1t2 (D.5)
Copyright # 2010 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power 2011; 21:329–344
DOI: 10.1002/etep