This is the author s version of a work that was submitted/accepted for publication in the Electric Power Systems Research journal in the following source: Božidar Filipović-Grčić, Dalibor Filipović-Grčić, Petar Gabrić, Estimation of load capacitance and stray inductance in lightning impulse voltage test circuits, Electric Power Systems Research, Volume 9, February 05, Pages 439-446, ISSN 0378-7796, http://dx.doi.org/0.06/j.epsr.04..007. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in Electric Power Systems Research, [Vol. 9, 05]. Copyright 05 Elsevier S.A. Notice: Changes introduced as a result of publishing processes such as copy-editing, formatting and technical enhancement may not be reflected in this document. For a final version of this work, please refer to the published source: http://dx.doi.org/0.06/j.epsr.04..007 URL: http://www.sciencedirect.com/science/article/pii/s037877964004064
Estimation of Load Capacitance and Stray Inductance in Lightning Impulse Voltage Test Circuits Božidar Filipović-Grčić a, *, Dalibor Filipović-Grčić b, Petar Gabrić b a, * Corresponding author: B. Filipović-Grčić (e-mail: bozidar.filipovic-grcic@fer.hr) is with the Faculty of Electrical Engineering and Computing, University of Zagreb, 0000 Zagreb, Croatia, tel.: +385 69 74; fax: +385 69 890. b D. Filipović-Grčić (e-mail: dfilipovic@koncar-institut.hr) and P. Gabrić (e-mail: pgabric@koncar-institut.hr) are with the Končar Electrical Engineering Institute, 0000 Zagreb, Croatia. Abstract: In order to obtain the lightning impulse voltage waveshape regarding front time, time to half and relative overshoot magnitude within the limits prescribed by IEC 60060-, it is useful to accurately estimate the test circuit parameters, e.g. load capacitance and circuit inductance. A stray inductance consists of the inductance of impulse generator and the inductance of connecting leads. Load capacitance consists of voltage divider capacitance, test object and parasitic capacitances. In practice, the test object capacitance is often unknown. Capacitance measurement takes time and makes testing procedure more complex. Also, it is very difficult to estimate parasitic capacitances although their influence can sometimes be significant. This paper presents a new genetic algorithm (GA) based method for fast and accurate estimation of load capacitance and circuit inductance during lightning impulse voltage testing of a capacitive load. Computational and experimental verification of the method is successfully performed for standard and non-standard lightning impulse waveforms with various relative overshoot magnitudes. Keywords: lightning impulse voltage testing, genetic algorithms, load capacitance, stray inductance
. Introduction High voltage equipment has to be tested with lightning impulse (LI) voltage in order to prove the capability against such overvoltages. In order to simulate the effect of transient overvoltage on high voltage equipment the various national and international standards define the impulse voltages and their appliance to a test object. Time parameters of lightning impulse voltage are shown in Fig. according to IEC standard []. Fig. Lightning impulse voltage time parameters [] Tolerances of. µs ± 30 % for front time T and 50 µs ± 0 % for time to half-value T are permitted. The test circuit has an inductance which consists of the inductance of impulse generator, ground leads and the connecting leads. In some cases inductance causes overshoot and oscillation at the crest of the lightning impulse voltage waveform. Overshoot usually occurs when the connecting leads from impulse generator to test object are very long and the inductance is comparably high. In case of a test object with high capacitance, low values of the impulse generator front resistors are used which in some cases can lead to oscillations occurrence. Fig. shows the overshoot β which 3
represents the increase of amplitude of an impulse voltage due to a damped oscillation (frequency range usually 0. MHz to MHz) at the peak caused by the inductance of the test circuit and the load capacitance. Fig. Determination of overshoot β magnitude from the recorded lightning impulse voltage and base curve Overshoot magnitude β is the difference between the extreme value of the recorded impulse voltage curve and the maximum value of the base curve. The base curve is an estimation of a full lightning impulse voltage without a superimposed oscillation. The relative overshoot magnitude β represents the ratio of the overshoot magnitude to the extreme value and it is defined by expression (). U e U b β ' = 00 % () U e According to [], the relative overshoot magnitude shall not exceed 0 %. In high voltage laboratories, lightning impulse voltages are most commonly produced using the Marx lightning impulse generator []. Equivalent circuit of the impulse generator is shown in Fig. 3. 4
Fig. 3. Single-stage impulse generator circuit The generator capacitance C is slowly charged from a DC source until the spark gap G breaks down. Resistor R primarily damps the circuit and controls the front time T, while resistor R discharges the capacitors and controls the time to half T. C represents the capacitance of test object and all other capacitive elements which are in parallel to the test object (e.g. capacitor voltage divider used for measurement, additional load capacitor, sometimes used for avoiding large variations of T and T if the test objects are changed, and parasitic capacitances). L represents the inductance of impulse generator and the connecting leads. Available values of R and R are limited in practice and therefore the standardized nominal values of T and T are difficult to achieve. Changing these resistors on the generator usually requires a trial-and-error process or accumulated experience with previous impulse tests on similar equipment. For this reason it is obvious that the simple and easy-to-use method for generator parameter determination would make lightning impulse testing procedure less complicated and less time-consuming. Many published papers deal with the calculation of impulse generator parameters: Thomason [3] determined circuit formulas of the most commonly used impulse generators circuits; Feser [4], Kannan and Narayana [5] and Del Vecchio et al. [6] investigated circuit design for the lightning impulse testing of transformers; Khalil and Metwally [7] developed a computerized method to reconfigure the impulse generator 5
for testing different types of objects. Methods described in the previously mentioned papers use C as an input parameter which means that the load capacitance should be known or measured before testing. In practice, test object capacitance is often unknown and the measurement of it takes time and makes testing procedure more complex. Also, parasitic capacitances cannot be taken into account by this approach although their influence can be significant especially when testing low capacitance objects of large dimensions. Genetic algorithm [8,9] and other optimization methods [0,] have already been used for curve fitting based estimation of lightning impulse parameters such as peak value, front time and time-to-half-value. However, the aim of this paper is to introduce a new genetic algorithm [3] (GA) based method for obtaining test circuit parameters in case of a capacitive load testing. The main advantage of this method is a fast estimation of the load capacitance and test circuit inductance from the recorded lightning voltage impulse and from the known values of generator capacitance, front and discharge resistance. Once when all circuit parameters are known it is less complicated to determine circuit elements which will provide T, T and β that are within limits prescribed by []. Hence, the presented method saves time and makes the reconfiguration of impulse generator easier.. Analysis of the lightning impulse voltage test circuit The method presented in this paper estimates lumped stray inductance and load capacitance in a test circuit. In the real situation, a stray inductance consists of the inductance of impulse generator and the inductance of connecting leads while load capacitance consists of voltage divider capacitance, test object capacitance and their parasitic capacitances. By taking this into account a more realistic equivalent scheme of the test circuit would be obtained. However, this model is more complex and solving it 6
would take more time and the algorithm has to be fast in order to be applied in practice. The presented method cannot differentiate all individual inductances and capacitances mentioned above. However, estimation of lumped inductance and capacitance in a test circuit proved sufficient for practical application. In [], [4] and [3] it is demonstrated that the calculations can be made much more easily if certain approximations are used, and these are found not to introduce appreciable errors in practice. Even more complicated circuit representations have been examined, particularly by Thomason [3], but the resulting expressions are of little more than academic or mathematical interest, especially as the stray capacitances and inductances are distributed throughout the circuit and no precise numerical values can be assigned to them. Therefore in practice it is convenient to simplify the calculations and use equivalent circuit shown in Fig. 3. Since in this paper excellent results were obtained by using equivalent circuit shown in Fig. 3 it is not convenient to take into account a more realistic equivalent scheme of the test circuit because it would not significantly improve the results. For that reason, a simplified circuit represented in Fig. 3 was used. The capacitances of the test object and of the voltage divider (and their stray capacitances) were lumped together in C, while the total inductance within the generator load circuit is combined to a single inductance L. Laplace transform of the circuit for lightning impulse voltage testing form Fig. 3 is shown in Fig. 4. U 0 sc s sc Fig. 4. Laplace transform of the circuit for lightning impulse voltage testing 7
Voltage U is determined by using the expression (). where: U ( s) U 0 =, () s( Z + Z ) Z Z = sc ; Z = R R R + R + sl + sc + sl + sc. (3) The output voltage U is calculated in frequency domain using the expression (4). U ( s) U ( s) Z 4 =, (4) Z 3 + Z 4 where: Z 3 R + sl; Z 4 = sc =. (5) The output voltage is expressed in the time domain using the inverse Laplace transform: = L ( U ( )) (6) U ( t) s 3. Method for estimation of circuit inductance and load capacitance Exact values of all circuit parameters are unknown in practice. Nominal test circuit parameters and their tolerances are given by manufacturers or can be more or less accurately measured, but always with measurement uncertainty. The charging voltage U 0, also, is not exactly known. GA based method determines circuit inductance L and load capacitance C for standard and non-standard lightning impulse waveforms. GA selects L and C in a wide boundary range of values and R, R, C and U 0 in a narrow boundary range around nominal values. All these parameters form a vector V i, an individual solution, while a group of vectors forms a population in GA terminology. Fig. 5 shows the flowchart of the described method. 8
Fig. 5. Method flowchart The first step is to obtain the input data consisting of the recorded impulse voltage from the measuring system, R, R, C, U 0 and their tolerances. GA generates the initial population of vectors V i, i= n. Population size n specifies how many individuals there are in each generation. Initial population is created randomly and it satisfies the defined bounds of parameters R, R, C, U 0, L and C. After the creation of the initial population, the output voltage U is determined in time domain for each V i. Since the GA performs many calculations finding an acceptable solution, it is very important to minimize the execution time. To achieve this, it is helpful to reduce the number of calculations by comparing measured waveform with GA results in several representative points only. Therefore, characteristic points (t cp, U cp ) are selected on measured waveform front (at 0 %, 30 %, 50 %, 75 %, 90 % and 00 % of the 9
amplitude), tail (at 90 %, 75 %, 60 %, 50 %, 40 % and 30 % of the amplitude) and at the local extremes in case of oscillatory impulses. For each t cp, voltage U is calculated with circuit parameters selected by GA and compared to measured values. Describing the waveform by characteristic points is a great advantage because it significantly reduces calculation time and enables this method to be used during laboratory impulse testing. The fitness function is the objective function minimized by the GA [5]. In this case, the fitness function takes into account the voltage percentage error for each characteristic point. Fitness function is calculated using the expression (7). U calculated ( tcp ) U cp ( tcp ) ε = 00 (%) (7) U ( t ) cp cp The algorithm stops when the voltage percentage errors of all characteristic points are lower than the user defined limit value or when a certain time elapsed. All calculations are performed using Matlab software on PC Intel core i5 CPU,.53 GHz with 4 GB RAM and a selected time period of min was enough to reach the stopping criteria. If stopping criteria is not fulfilled, then selection, crossover and mutation are performed. The stochastic uniform selection function chooses the parents for the next generation based on fitness results. The elite count specifies the number of individuals that are guaranteed to survive to the next generation (in this case 50). Crossover fraction specifies the fraction of the next generation, other than elite individuals, that are produced by crossover (in this case 80 %). The remaining individuals are produced by mutation. The scattered crossover function creates a random binary vector. It then selects the genes for which the vector value is a from the first parent, and the genes for which the vector value is a 0 from the second parent, and combines the genes to form the child. Mutation functions create small random changes in individuals from a population, and they provide genetic diversity and enable the GA to search a broader 0
space. Adaptive feasible mutation was used which randomly generates directions that are adaptive with respect to the last successful or unsuccessful generation. A step length is chosen along each direction so that bounds are satisfied. If T, T or β are not acceptable, it is necessary to modify the test circuit parameters. This usually requires changing R and/or R and/or introduction of additional capacitor C ad in parallel with the test object. Now, with known L and C, one can easily calculate output voltage using the finite pool of R, R and C ad available in a high voltage laboratory instead of trial-and-error voltage applications in actual test circuit. In case of a purely capacitive test objects, maintaining the T is much more complicated than maintaining the T. Stray inductance L and front resistance R tend, in general, to retard the length of the wave front. The inductance also introduces oscillations in the wave [3]. It may be impossible to realize standard waveforms within the standard tolerances for certain test circuits and test objects. In such cases extension of front time T or overshoot may be necessary (guidance for such cases should be given by the relevant Technical Committee). Of all parameters used in the GA based method it is noticed that only the population size significantly affects the GA convergence rate and execution time, while the influence of other parameters is small. At first, a smaller population sizes were used but the convergence rate was slow and the execution time was long. Execution time of this algorithm should not be longer than a few minutes (in this paper a min criterion adopted) in order to be applied in practice. It is noticed that the increase of population size to a certain value improves the convergence rate and shortens the execution time to an acceptable value. Satisfactory convergence rate and acceptable execution time were
achieved with the population size of 500 in case of computational verification and with the population size of 000 in case of experimental verification. 4. Verification of the presented method The presented method is verified in the following sections: 4. Computational (numerical) test. 4. Tests with recurrent surge generator. 4.3 Tests in high voltage laboratory. 4. Computational verification All circuit parameters are exactly known in this example and the method s ability to determine L and C is tested. The following cases are examined: a) U without oscillations, L=0, GA selects only C ; b) U with oscillations L=6.5-950 µh, relative overshoot β =9.09-33.33 %, GA selects L and C. In both cases other parameters are C =50 nf, C = nf while R and R are varied to provide standard and non-standard waveforms with T =0.6 -.8 µs and T = 30-70 µs. GA input values are R, R, C, U 0 and characteristic points from U. The size of GA population in each generation is n=500. GA task is to choose L and C in a fairly wide range of values (boundaries 0. - 0 nf for C and 0-4000 µh for L) and to find the output voltage which best fits the inputted U. Table shows estimated C for case a). Tables -6 show estimated C and L for case b).
Table Estimated C, circuit without stray inductance T /T (µs) R (Ω) R (Ω) Estimated C (nf) Number of generations 0.6/30 03.9 8.7 0.9999998 7 0.6/50 98.7 376. 0.9999996 7 0.6/70 96. 940.6 0.9999999 7./30 43.4 78.5.0000003 7./50 43.0 343.7.0000000 7./70 404.3 906.8 0.9999998 8.8/30 683.8 75.5.0000000 7.8/50 64.5 33. 0.9999967 5.8/70 6.6 875. 0.999994 5 Table Estimated C and L (relative overshoot 9.09 %) T /T (µs) Exact L (µh) Estimated L (µh) Estimated C (nf) Number of generations 0.6/30 6.5 6.50 0.9999 0.6/50 6.0 5.977.000 4 0.6/70 5.8 5.793.000./30 0 09.99.0000./50 07 06.98.000 0./70 05 05.0 0.9999 9.8/30 55 55.09 0.9995 9.8/50 48 48.00.0000 9.8/70 40 39.97.000 0 Table 3 Estimated C and L (relative overshoot 6.67 %) T /T (µs) Exact L (µh) Estimated L (µh) Estimated C (nf) Number of generations 0.6/30 45.0 44.995.0000 0 0.6/50 43.5 43.50 0.9999 0 0.6/70 43.5 43.506.0000 0./30 87 86.93.000 8./50 8 80.96.000 9./70 78 77.96.000 9.8/30 447 446.85.000 9.8/50 45 44.90.0004 8.8/70 40 40.06 0.9998 9 3
Table 4 Estimated C and L (relative overshoot 3.08 %) T /T (µs) Exact L (µh) Estimated L (µh) Estimated C (nf) Number of generations 0.6/30 73.5 73.464.0005 0 0.6/50 7 7.05 0.9999 9 0.6/70 70.5 70.669 0.9984 9./30 30 30.7 0.999 9./50 98 98.06 0.9999 0./70 90 90.06 0.9998 8.8/30 745 744.09.0007 8.8/50 695 694.49.000 8.8/70 670 669.7.0006 8 Table 5 Estimated C and L (relative overshoot 8.57 %) T /T (µs) Exact L (µh) Estimated L (µh) Estimated C (nf) Number of generations 0.6/30.3 0.9994 8 0.6/50 6 6.00.0005 8 0.6/70 5.5 5.57 0.9996 0./30 55.5 55.8.0006 8./50 490.5 490.9.0005 8./70 475 475.00.000 8.8/30 70 69.8.000 8.8/50 60 59.7.0005 7.8/70 0 09.3.00 7 Table 6 Estimated C and L (relative overshoot 33.33 %) T /T (µs) Exact L (µh) Estimated L (µh) Estimated C (nf) Number of generations 0.6/30 0 0.00 0.9999 8 0.6/50 95 95.8 0.9988 0 0.6/70 95 94.9.0007 8./30 900 899.4.0003 7./50 850 849.9.000 8./70 80 80.64 0.9988 8.8/30 340 339.7.000 8.8/50 050 049.5.000 7.8/70 950 95.4 0.9995 7 Figs. 6-8 show comparison between input (original) waveform and calculated waveform obtained with estimated L and C. 4
U (kv) 60 40 0 00 80 60 40 0 0 0 3 4 5 6 7 8 9 0 t (µs) Fig. 6. Comparison of original 0.6/30 µs impulse voltage waveform (colored lines) and estimated waveform (black dotted lines) 60 U (kv) 40 0 00 80 60 40 0 0 0 3 4 5 6 7 8 9 0 t (µs) Fig. 7. Comparison of original./50 µs impulse voltage waveform (colored lines) and estimated waveform (black dotted lines) 60 U (kv) 40 0 00 80 60 40 0 β =33.33 % β =8.57 % β =3.08 % β =6.67 % β =9.09 % β =0 % Estimation 0 0 3 4 5 6 7 8 9 0 t (µs) Fig. 8. Comparison of original.8/70 µs impulse voltage waveform (colored lines) and estimated waveform (black dotted lines) 5
Fig. 9 shows the change of fitness value ε throughout generations for waveform./50 µs and relative overshoot β =9.09 %. 5 0 Fitness value (%) 5 0 5 Fig. 9. Change of fitness value ε throughout generations for waveform./50 µs and β =9.09 % In all cases algorithm successfully estimates C and L with high precision within the following stopping criteria: a time period of min elapsed or the fitness value of 0. % is achieved. The largest percentage differences between estimated and known C and L are 0.6 % and 0.4 %, respectively. 4. Tests with RSG RSG is a low voltage single stage equivalent of a high voltage impulse generator. It is usually used to study the voltage distribution in high voltage windings during impulse voltage stresses. 0 0 3 4 5 6 7 8 9 0 Generation Parameters of impulse circuit are in fact parameters of RSG so, in this case, the exact values of circuit parameters are unknown, but the nominal values are stated by RSG manufacturer. Fig. 0 shows the test setup. The impulse voltages are recorded using a digital oscilloscope (500 MHz, GS/s) connected to PC. 6
Recorded waveform Data acquisition Digital oscilloscope Recurrent surge generator Fig. 0. Test setup for generation of low voltage impulses with RSG RSG parameters are U 0 =50 V, R =6.8 Ω, R =00 Ω, C =470 nf, C =00 nf and L is varied (0 µh, 0 µh and 30 µh). GA selects L from 0. µh to mh and C from 0 nf to 00 nf. The size of GA population in each generation is n=000. Fig. shows the comparison between measured and calculated (GA obtained) waveforms. Measured and GA obtained waveforms show excellent agreement. 30 80 40 U (V) 00 60 0 80 40 0 0 5 0 5 0 5 30 t (µs) Fig.. Comparison of measured and estimated impulse voltage waveforms 7
Percentage differences between real and estimated C and L cannot be exactly determined because the exact values of circuit parameters are unknown, but only the nominal values are stated by RSG manufacturer. Nevertheless, comparison of estimated and nominal values shows a good agreement (Table 7). Table 7 Estimated C and L Estimated Estimated L (µh) C (nf) 0 0.8 07.7 0 0.6 03.7 7 30 30.7 03.3 7 Nominal L (µh) Number of generations 4.3 Tests in high voltage laboratory In Section 4. the exact values and in Section 4. the nominal values of all circuit parameters were known. In this case L and C are unknown. The tests are performed on oil-paper insulated inductive voltage transformer U m =45 kv and SF 6 insulated current transformer U m =765 kv. The measurement of impulse voltages was carried with measurement systems consisting of digital impulse analyzing system and capacitor voltage divider which fulfill the requirements of [], [7] and [8]. 4.3. Oil-paper insulated inductive voltage transformer U m =45 kv Fig. shows the test setup for lightning impulse voltage testing of an oil-paper insulated inductive voltage transformer in high voltage laboratory. Parameters of the test circuit are: U 0 (68 to 798 kv), R =600 Ω, C =50 nf and R is varied. GA selects L from 0. µh to mh and C from 0.5 nf to 0 nf. 8
Leads Impulse voltage generator Capacitor voltage divider Inductive voltage transformer Fig.. Lightning impulse voltage test of oil-paper insulated inductive voltage transformer (U m =45 kv) in high voltage laboratory C consists of voltage divider capacitance (nominal value 87 pf), test object capacitance (unknown) and parasitic capacitances (unknown). At first the impulse voltage is recorded with R =4 Ω (yellow curve in Fig. 3). U (kv) 800 700 600 500 400 300 00 00 Fig. 3. Comparison of measured and estimated impulse voltage waveforms in case of testing the oil-paper insulated inductive voltage transformer This waveform contains some oscillations with front time T =0.77 µs and tail time T =4.4 µs. According to [] T is too short. 0 0 3 4 5 6 7 8 9 0 t (µs) 9
Increasing the external front resistor R value helps to damp oscillations while the value of external leads inductance causes an oscillatory natural response of the system. The critical serial resistance R c for the circuit to be non-oscillatory is given by the following well-known equation: where: R c L C = (8) C C + C = (9) This equation is suitable for determining the limiting values for the front resistor R. In this case R c =64 Ω. Since high voltage laboratory has a finite pool of R, the closest value obtained was R =95 Ω. L increases with R due to the stray inductance of the resistors, so it is better to choose R slightly higher than calculated R c. Measurement and simulation results are shown in Fig. 3 and Table 8. Table 8 Estimated C and L Measured U 0 Estimated Estimated T /T (µs) (kv) L (µh) C (nf) 4 0.77/4.4 68 45.8.96 6 0.88/5.6 687 49.6.88 95.07/58.4 740 53.7.80 4 464.88/60. 770 88.7.9 0 550.0/6.8 798 05.3.94 0 Nominal R (Ω) Number of generations It can be seen that the oscillations are heavily damped when R =95 Ω which is close to calculated R c, while T and T are within the limits of []. Other examples in Table 8 only show how front resistors introduce stray inductances and that this can be taken into account. Estimated C is quite stabile as expected and varies from.80 nf to.96 nf. Increase of R damps the oscillation and reduces peak voltage without affecting wave tail and, as 0
a consequence, T increases. In Fig. 3, the charging voltage U 0 of the impulse generator was increased in order to compensate the influence of R on peak voltage reduction. 4.3. SF 6 insulated current transformer U m =765 kv Fig. 4 shows the test setup for lightning impulse voltage testing of SF 6 insulated current transformer in high voltage laboratory. Current transformer Leads Impulse voltage generator Capacitor voltage divider Fig. 4. Lightning impulse voltage test of SF 6 insulated current transformer (U m =765 kv) in high voltage laboratory Parameters of the test circuit are: U 0 =640 kv, R =36 Ω, C =53.6 nf and R =40 Ω. GA selects L from 0. µh to mh and C from 0.5 nf to 0 nf. C consists of voltage divider capacitance (nominal value 63 pf), test object capacitance (6 pf, measured with Schering bridge and standard capacitor) and parasitic capacitances (unknown value). GA simulation gives C =.69 nf and L=99.9 µh after generations. Fig. 5 shows measured and estimated impulse waveforms.
600 400 00 000 800 600 400 00 Fig. 5. Comparison of measured and estimated impulse voltage waveforms in case of testing the SF 6 insulated current transformer U m =765 kv There is a significant difference between estimated C and the sum of capacitances of the test object and voltage divider. This implies that parasitic capacitances have great influence on C and cannot be disregarded. 0 0 3 4 5 6 7 8 9 0 t (µs) Fig. 6 shows a comparison between measured and estimated waveforms when parasitic capacitances are excluded from the simulation. U (kv) U (kv) 000 750 500 50 000 750 500 50 0 0 3 4 5 6 7 8 9 0 t (µs) Fig. 6. Comparison of measured and estimated impulse voltage waveforms in case of testing the SF 6 insulated current transformer U m =765 kv; parasitic capacitances are not included in simulation
Load capacitance is set to a sum of voltage divider and test object capacitances while circuit inductance is varied from 0 to 00 µh. It is obvious that even when test object capacitance value is known it may not be enough for accurate estimation of impulse voltage waveform. According to the previous statement, it can be concluded that a test object capacitance measurement may give insufficient information about a total load capacitance. 5 Conclusion The paper describes a new GA based method for estimation of load capacitance and circuit inductance in case of lightning impulse testing of capacitive loads. The presented method enables a fast and accurate estimation of unknown circuit parameters: the total load capacitance, including parasitic capacitances which cannot be disregarded when testing low capacitance objects, and test circuit inductance. The method is successfully verified on several examples. Future work will be focused on expanding the capabilities of the method for estimating the test circuit parameters in case of impulse testing of low inductance loads, such as low voltage windings of power transformers. References [] IEC 60060-, High-voltage test techniques Part : General definitions and test requirements, Edition 3, September 00. [] E. Kuffel, W. S. Zaengl, J. Kuffel, High voltage engineering: fundamentals, Second Edition, Newnes, August 000. [3] J. L. Thomason, Impulse Generator Circuit Formulas, Transactions of the American Institute of Electrical Engineers, Vol. 53, Issue:, pp. 69 76, January 934. 3
[4] K. Feser, Circuit Design of Impulse Generators for the Lightning Impulse Voltage Testing of Transformers, Bulletin SEV/VSE Bd. 68, 977. [5] S. R. Kannan and Y. Narayana Rao, Generator loading limits for impulse testing low-inductance windings Proceedings of the Institution of Electrical Engineers, Vol., Issue 5, pp. 535 538, May 975. [6] Robert M. Del Vecchio, Rajendra Ahuja and Robert Dean Frenette Determining Ideal Impulse Generator Settings from a Generator Transformer Circuit Model, IEEE Transactions on Power Delivery, Vol. 7, No., January 00. [7] M. El-Adawy Khalil and I.A. Metwally, Reconfiguration of impulse-voltage generator for conducting standard lightning tests: A comparative investigation, The International Journal for Computation and Mathematics in Electrical and Electronic Engineering (COMPEL), Vol. 7, Issue, pp. 50 533, 008. [8] K. C. P. Wong, H. M. Ryan, J. Tindle, J. Blackett and M. W. Watts, Digital measurement of lightning impulse parameters using curving fitting algorithms, Eleventh International Symposium on High Voltage Engineering, 999. (Conf. Publ. No. 467). [9] W. M. Al-Hasawi, K. M. El-Naggar, A genetic based algorithm for digitally recorded impulse parameter estimation, Power Tech Conference Proceedings, IEEE Bologna, Vol., pp. 3-6, June 003. [0] T. R. McComb, J. E. Lagnese, Calculating the parameters of full lightning impulses using model-based curve fitting, IEEE Transactions on Power Delivery, Vol. 6, Issue 4, pp. 386 394, 99. 4
[] E. A. Feilat, A. Al-Hinai, HV impulse parameter estimation using ESPRITbased method, nd International Conference on Electric Power and Energy Conversion Systems (EPECS), 0. [] G. Ueta, T. Tsuboi and S. Okabe, Evaluation of Overshoot Rate of Lightning Impulse Withstand Voltage Test Waveform Based on New Base Curve Fitting Methods - Study on Overshoot Waveform in an Actual Test Circuit, IEEE Transactions on Dielectrics and Electrical Insulation, Vol. 8, No. 3, pp. 783-79, June 0. [3] R. L. Haupt, S. E. Haupt, Practical Genetic Algorithms, (Wiley & Sons, second edition, 004). [4] F. S. Edwards, A. S. Husbands, F. R. Perry, The development and design of high-voltage impulse generators, Proceedings of the IEE - Part I: General, Vol. 98, Issue:, pp. 55-68, May 95. [5] User s Guide, Genetic algorithm and direct search toolbox, The MathWorks, Matlab 009. [6] IEC 6083-: Instruments and software used for measurement in high-voltage impulse tests - Part : Requirements for instruments, Edition, June 00. [7] IEC 6083-: Instruments and software used for measurement in high-voltage and high-current tests - Part : Requirements for software for tests with impulse voltages and currents, Edition, July 996. 5