IEEE TRANSACTIONS ON MAGNETICS, VOL. 46, NO.

1, JANUARY 2010

141

Transformer Core Parameter Identication Using Frequency Response Analysis

A. Shintemirov, W. H. Tang, and Q. H. Wu
Department of Electrical Engineering and Electronics, The University of Liverpool, Liverpool L69 3GJ, U.K.
We present a novel model-based approach for parameter identication of a laminated core, such as magnetic permeability and electrical conductivity, of power transformers on the basis of frequency response analysis (FRA) measurements. The method establishes a transformer core model using the duality principle between magnetic and electrical circuits for parameter identication with genetic algorithms. We use reference input impedance frequency responses, calculated by a well-known lumped parameter model of a three-phase transformer and nite-element computations, to analyze identication accuracy of the method. The results verify the ability of the approach to accurately identify the core lamination parameters with respect to the reference values. The approach can be used for parameter identication of a demagnetized core with known geometrical parameters when the core lamination samples are unavailable for experimental tests. The approach can also be employed for transformer core modeling and FRA result interpretation at low frequencies. Index TermsFrequency response analysis, genetic algorithm, magnetic-electric duality principle, parameter identication, transformer core.

I. INTRODUCTION

URING recent years, frequency response analysis (FRA) has been recognized as the most reliable condition monitoring technique for transformer winding displacement and deformation assessment. It is established upon the fact that the shape of a winding frequency response at high frequencies is associated with winding geometry. The appearance of clear shifts in resonance frequencies or new resonant points on a response may characterize faulty conditions of windings [1]. A range of research activities have been undertaken to utilize FRA in the development of suitable lumped parameter mathematical models of transformer windings [2][6]. In [2][4] analytical expressions were used to estimate parameters of winding models based on the geometry of windings. The nite-element method (FEM) was applied in [5] and [6] for accurate parameter estimation. It was shown that the geometry of the laminated magnetic core and its frequency-dependent characteristics, i.e., permeability and eddy-current losses, have to be taken into account during the calculation of inductive and resistive parameters, which clearly affect the frequency responses of a transformer at low frequencies [4], [5], [7]. One of the techniques to model low-frequency responses associated with a transformer core is to apply the duality principle between magnetic and electrical circuits [8], [9]. This approach has been widely used for time domain transient studies of power transformers [10][12] and subsequently was applied for transformer low-frequency response analysis [7], [13], [14]. In the above core simulation studies the parameters of a core lamination were assumed to be constant and known. However, it is necessary to take into account frequency dependent behaviors of the lamination in order to accurately model frequency responses of a transformer core.

Manuscript received November 07, 2008; revised April 14, 2009. Current version published December 23, 2009. Corresponding author: Q. H. Wu (e-mail: qhwu@liv.ac.uk). Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TMAG.2009.2026423

In practice, magnetic core parameters are determined using additional experimental tests on lamination samples. For instance, a lamination sample taken out from a transformer was used for precise measurements of complex permeability with a single sheet tester in [15], [16]. However, in industry conditions it is not always possible to obtain a core lamination sample from a stacked core of a transformer to perform additional tests. For the case when the magnetic core geometry is known, an indirect method of estimation of small signal permeability was proposed in [15]. However, this method utilizes inductance measurements of a transformer winding section at low frequencies and a 3-D FEM model of a power transformer presented in [17][19], that introduces complex computations and, thereby, reduces the method practicability. Recently, evolutionary algorithms such as genetic algorithms (GAs) [20], particle swarm optimizer (PSO) [21], etc., were utilized to identify parameters of transformer winding models using FRA measurements. During a learning process, an evolutionary algorithm optimizes model parameters in order to reduce the difference between real FRA measurements and corresponding simulations of winding models [22], [23]. One of the advantages of this model-based approach is that evolutionary algorithms require only approximate range of possible values for each parameter as initial estimates for learning. This disregards additional tests on transformer insulation samples, which are not always possible to obtain. This paper presents a novel approach for parameter identication of a laminated core of power transformers based on reference FRA measurements with GA. The magnetic-electric duality principle is applied to establish the equivalent electrical circuit and mathematical model of a core, which is used for model-based parameter identication. Reference frequency responses, being calculated with a well-known lumped parameter model of a three-phase transformer and FEM computations, are utilized to explore the potential of the proposed identication approach. The proposed approach is applied only for a demagnetized core due to the strong effect of remanent core magnetization on FRA measurements at low frequencies, associated with the

0018-9464/$26.00 2009 British Crown Copyright

142

IEEE TRANSACTIONS ON MAGNETICS, VOL. 46, NO. 1, JANUARY 2010

Fig. 1. Cross section of a three-limb transformer.

Fig. 2. Equivalent magnetic circuit of a transformer core.

magnetic core of power transformers. In fact, the amount of remanent magnetization determines the small signal permeability of the core and, hence, changes of both the magnitudes and frequencies of the rst resonances, which affects identication results of core parameters [24]. II. MATHEMATICAL MODEL OF A TRANSFORMER CORE A. Equivalent Magnetic Circuit Consider a cross section of a typical three-limb wound core transformer as shown in Fig. 1, where the phase A winding is excited for frequency response measurements. Assume that all the magnetic uxes are conned to the magnetic material of the core and the leakage uxes in the air paths around the windings are considered to be negligible. Thus, the magnetic core of the transformer can be divided into three sections with respect to and of each transformer phase [9]. uniform uxes To establish the equivalent magnetic circuit of the core, each section of the magnetic core is represented by its reluctance , which provides a relation between the corresponding ux and the magnetomotive force required to establish that ux along the length of the section. The reluctance of a section is determined by the magnetic parameters of lamination and the core geometry [8], [9]:

The skin depth depends on the angular frequency magnetic eld and is dened as follows:

of the

(3) where is the electrical conductivity of the core lamination. The equivalent magnetic circuit of the core is presented in and represent the reluctances of the Fig. 2, where three paths in the magnetic core carrying magnetic uxes and respectively. The magnetomotive force is dened and the current of the excited by the number of turns winding on phase A as below [9]: (4) In general, the magnetic paths for the lateral phase A and , phase C sections can be assumed to be equal, i.e., which are different from the path of central phase B. Therefore, the relations between the reluctances are expressed as follows [13]:

(5) where is the symmetry coefcient of the transformer and . B. Equivalent Electrical Circuit According to the topological principle of duality, the equivalent electrical circuit of a transformer core is derived directly from its magnetic circuit. Consider the equivalent magnetic circuit of the core in Fig. 2. The circuit contains two internal meshes (loops) in which two nodes 1 and 2 are pointed, and a reference node 0 is marked outside the circuit. These nodes are then joined by branches, one of which passes through each element of the magnetic circuit (dotted lines in Fig. 2) [9]. The corresponding equivalent electrical circuit of the core is shown in Fig. 3, where the corresponding nodes are marked. of each core section is Since the complex reluctance not only nonlinear and frequency dependent but also takes into account eddy currents effect due to complex magnetic (1), [17], it is represented by a corresponding permeability

(1) where is the length of the magnetic ux path along each secdenotes the tion, is the cross-sectional area of the core, complex effective relative permeability of the core lamination is the free space permeability. in the rolling direction, and is dened The complex effective relative permeability using the following expression [5], [17]:

(2) represents the where denotes the skin depth, stacking factor, i.e., the fraction of steel in the laminated core, and are the thicknesses of a single lamination sheet of the core with and without insulation layer included respectively, and denotes the local magnetic permeability.

SHINTEMIROV et al.: TRANSFORMER CORE PARAMETER IDENTIFICATION USING FREQUENCY RESPONSE ANALYSIS

143

Fig. 3. Equivalent electrical circuit of a transformer core.

impedance in the equivalent electrical circuit in the form of an inductance to include a nonlinear relationship and a to reect eddy-current losses in the core. The resistance impedance is given as follows [15], [16]:

Fig. 4. Typical input impedance frequency response of a core model (a lateral phase winding is excited).

(6) is the reciprocal of reluctance. where permeance , the corresponding inTaking into account that ductance and resistance of each section of the equivalent circuit are [16]:

A similar procedure is used to derive the expressions for the input impedance of the core taken from phases B and C of the transformer, assuming the same number of turns of the corre: sponding excited windings, i.e.,

(10) (11)

(7) In addition to inductive and resistive elements, the parallel capacitances are added into the equivalent electrical circuit to represent the total stray capacitances of windings on each limb of the transformer core [13], [14], [25]. These capacitances depend on the geometry of windings, and are generally the combinations of series intersection or interturn capacitances of windings and shunt capacitances between windings and between a winding and a core [26], [27]. The dielectric losses of windings are not considered due to their negligible contribution at low frequencies comparing to the eddy-current losses of the core. Having obtained the equivalent electrical circuit of the core, it is possible to derive the expression for the input impedance of the core measured from the terminals of the phase A winding of the transformer: III. ESTIMATION OF THE EQUIVALENT CIRCUIT PARAMETERS FROM FRA MEASUREMENTS It is known that the very low-frequency region of the open circuit FRA measurements of a transformer is mainly associated with the properties of a magnetic core [1], [3]. Consider a typical input impedance frequency response of the core model when the lateral phase A is excited. The response is calculated using (8) and is divided into the three frequency ranges as shown in Fig. 4. In fact, different combinations of the core model parameters are mostly responsible for certain frequency ranges in Fig. 4, [13], [14]. With regard to the equivalent electrical circuit of a core (Fig. 3), the FRA responses in frequency range 1 are mostly determined by the combination of the section impedances , whereas the responses in range 3 are mainly dened by the combinations of the section capacitances . These features can be used for the estimation of the equivalent electrical circuit parameters from FRA measurements as discussed below. A. Impedance Estimation The frequency response below the rst resonance frequency (range 1 in Fig. 4) is mostly determined by the combination of since they are much lower the phase section impedances than the parallel capacitor impedances of each phase

(8) where the equivalent admittances of each phase section of the circuit are dened as below:

(9)

144

IEEE TRANSACTIONS ON MAGNETICS, VOL. 46, NO. 1, JANUARY 2010

Thus, the expression for the input impedance taken from the phase A winding in frequency range 3 (Fig. 4) is derived as follows:

(17) Thus, having obtained from FRA measurements the capacitances can be estimated as follows:

(18)
Fig. 5. Equivalent electrical circuits for impedance and capacitance estimation.

section [14]. In this case, the capacitor impedances can be neglected due to its sufciently higher values. The equivalent electrical circuit is then simplied as shown in Fig. 5(a). Taking into account (5) and (6), the relations between the phase section impedances are given as below [13]: (12) Thus, with aid of (12) the expression for the input impedance taken from the phase A winding at low frequencies (range 1 in Fig. 4) can be obtained as follows:

IV. MODEL-BASED IDENTIFICATION OF TRANSFORMER CORE PARAMETERS In practice, the parameters of core lamination, such as local magnetic permeability and the conductivity of the core lamination , are usually determined by a number of experimental tests using a lamination sample [15], [16]. However, as mentioned in Section I, in many occasions such tests can not be conducted due to the unavailability of the core lamination samples. In this study, GA is employed for the identication of the core lamination parameters utilizing reference FRA measurements of transformer windings. A. Foundation of GAs GA is a powerful evolutionary optimization technique, which is established based upon the principles of genetics and natural selections [20]. The fundamental differences between GAs and traditional optimization techniques, such as the least square optimization, etc., is that the GAs can perform highly parallel search of the solution space and do not demand accurate initial estimates [28]. GAs preserve the biological terminology and operate with a set (population) of optimized parameters (analogous to genes) being coded as a nite-length strings (analogous to chromosomes), representing potential solutions of an optimization problem. The population of the individuals (chromosomes) is undergone by the procedure of tness evaluation, which represents the survivability of individuals during a selection procedure. Then the ttest individuals are chosen as parents for performing crossover and/or mutation, thereby producing offsprings, which constitute a new generation of a population. The process continues until a given termination criterion is met or simply a certain number of generations is reached. The nally survived individual is treated as a variant of a desired solution [29]. A GA process can be briey expressed in the form of a sequence of operations: 1) Random generation of the initial population. 2) Evaluation of individuals in the population and its tness values calculation. 3) Performing selection procedure. 4) Performing genetic operations on selected individuals. 5) Replacing the previous generation by the offspring population after the genetic operations are performed.

(13) Hence, knowing from FRA measurements the following parameters can be estimated:

(14) and using (12):

(15)

B. Capacitance Estimation A similar procedure is applied for capacitance estimation. In frequency range 3 beyond the main resonance peaks as shown of each phase in Fig. 4, the parallel capacitor impedance section is much lower than the corresponding section impedance [14]. As a result, the phase section impedances can be neglected and the equivalent electrical circuit is simplied as shown in Fig. 5(b), [13]. In practice, transformer windings on different core limbs have the same design. Therefore, it is reasonable to assume that the parallel capacitances related to different phase windings are equal [13]: (16)

SHINTEMIROV et al.: TRANSFORMER CORE PARAMETER IDENTIFICATION USING FREQUENCY RESPONSE ANALYSIS

145

6) Repeating steps 25 until a termination criterion is met. 7) Presentation of the best individual in the nal population as the GA output. B. Parameter Identication With GA The model-based parameter identication with GA is based on searching for the optimal model parameters by minimizing the difference, i.e., tness, between reference and simulated model frequency responses. Therefore, for each individual of a population in GA, its total tness value is given as follows:

(19) where and are the magnitude and phase frequency responses respectively of the reference and the simulated with the GA identied parameters , where input impedances at frequency is the number of frequency points involved in a GA learning process. The following steps are performed for the core parameter identication in this study. Experimental FRA data or simulation data derived from a transformer winding model with predened parameters are used as reference frequency responses. Reference response points in a frequency range of interest are selected to create a reference dataset, being employed as training targets for GA learning. The initial search space for the identied parameters is established based on approximate estimations. GA learning is performed, in each step of which the predened training dataset is compared with the corresponding values of the simulated frequency responses at the same frequency points. The simulated frequency responses are generated using the established transformer core model with the parameters obtained during the GA learning process. V. SIMULATION RESULTS AND COMPARISON A. Reference Response Simulation In order to analyze the identication accuracy of the proposed approach, the lumped parameter model, presented in [4], [5], of a three-phase experimental transformer (400 kVA, 15/0.4 kV) having a -Y winding conguration is employed to calculate reference input impedance frequency responses due to its high degree of simulation accuracy in comparison with experimental measurements. The geometrical dimensions of the transformer are presented in [18] and used for FEM computation of inductive and resistive parameters of the lumped parameter model [17][19]. Capacitance and conductances of the model are estimated using the analytical expressions and measurements of dielectric material properties presented in [6].

Fig. 6. Effective complex permeability  ^ nated core (real and imaginary parts).

in the rolling direction of a lami-

The magnetic core effect is represented in the lumped paramof core eter model by the complex effective permeability lamination, calculated according to (2) and (3) using the folmm, lowing reference parameters: S/m, and [17]. As a result, the real and imaginary parts of the effective permeability in the rolling direction of the lamination are presented in Fig. 6 as a function of frequency according to (2). B. Identication of Local Magnetic Permeability The identication of the local magnetic permeability is implemented using the procedure for impedance estimation discused in Section III-A. for GA learning is created using the referThe dataset in the frequency range ence input impedance responses of 100300 Hz with aid of (14). Thus, GA learning is conducted using a set of randomly generated values of to create an initial population of GA individuals as explained in Section IV. The tness evaluation of each individual in the population is per, being analytically calculated using a formed by comparing using (19). value of associated with the individual, with In this study, the limits for parameter identication are estabfor the widely lished using a reasonable initial estimate of used core lamination material. In order to show the potential of GA learning the limits of the search space are assumed to be % from the value , which is 60% more than within the value , being used to obtain the reference responses in Section V-A. A reasonable estimate of electrical conductivity S/m is initially selected to calculate the skin depth using (3), which is a typical value for grain-oriented silicon steels used for transformer core laminations [30], [31]. In fact, this parameter does not greatly affect the identication results as illustrated below. The GA learning parameters are selected on the basis of numerous trials with various parameter combinations. The parameters are listed in Table I. Fig. 7 illustrates the comparisons of the reference and the identied with GA magnitude frequency responses of the transformer input impedance. The identied local

146

IEEE TRANSACTIONS ON MAGNETICS, VOL. 46, NO. 1, JANUARY 2010

TABLE III COMPARISON OF THE REFERENCE AND GA IDENTIFIED VALUES OF LOCAL MAGNETIC PERMEABILITY (100200 HZ FREQUENCY RANGE)

Fig. 7. Comparison of the input impedance magnitude frequency responses: reference and identied with GA.

TABLE I GA PARAMETERS

observed in Fig. 7, where the deviations between the impedance responses become larger at frequencies above 200 Hz. To reduce the effect of on the identication accuracy, it is suggested to narrow the frequency range being used for GA learning to lower frequencies. The results of identication of using impedance responses in the 100200 Hz frequency range for GA learning are listed in Table III. As seen from the table, rapidly improves having only the identication accuracy of at different values of about 0.01%0.3% deviations from , while the estimation range of is limited by 10%200% from . the reference value C. Identication of Conductivity and Capacitance The identication of conductivity is carried out in frequency range 2 containing main resonance frequencies as shown in Fig. 4. The reason is that the damping of the resonance peaks mostly depends on eddy-current losses characterized by the skin depth (3) and, hence, by the conductivity of the core lamination. Similar to the previous case, GA learning is conducted using a set of randomly generated values of and for the calculation with aid of (8). The tness evaluation is performed of with the reference dataset in the by comparing frequency range of 8001600 Hz (19). As previously mentioned, the search space limits for parameter identication are established using the reasonable initial estimate of for the known lamination material. In order to show the potential of GA learning the limits of the search space are % from the value S/m, assumed to be within , being used to obtain the which is 60% more than the value reference responses in Section V-A. The initial estimate of the section capacitances F is derived using the reference dataset in the frequency range of 23002500 Hz (range 3 in Fig. 4) with aid of , which (18) as discussed in Section III-B. The value of is within the range of the values identied in Section V-B, is utiduring GA learning process. lized for calculation of The results of the GA parameter identication at different are listed in Table IV. As clear from the table, the values of GA identied values of are very close to the reference with less than 5% deviations across the range of % from . At the same time, the identication error is only 0.2% in , being used in calculations. the case of the reference value

TABLE II COMPARISON OF THE REFERENCE AND GA IDENTIFIED VALUES OF LOCAL MAGNETIC PERMEABILITY (100300 HZ FREQUENCY RANGE)

magnetic permeability , which is 1.53% more than . the reference value Table II summarizes the reference and the parameters identied with GA using different values of in order to analyze the effect of its variation on identication results. The analysis of the table shows that the value of does not signicantly affect the identication results in a wide range of 20%200% from the , where deviations of do not exceed 2% reference value from . The deterioration of identication accuracy with the increased values of can be explained by the eddy current effect strengthening with the increase of frequencies. This can be

SHINTEMIROV et al.: TRANSFORMER CORE PARAMETER IDENTIFICATION USING FREQUENCY RESPONSE ANALYSIS

147

TABLE IV COMPARISON OF THE REFERENCE AND GA IDENTIFIED VALUES OF CONDUCTIVITY AND CAPACITANCE

TABLE V INITIAL SEARCH SPACE LIMITS AND GA IDENTIFIED VALUES OF THE CORE MODEL PARAMETERS Fig. 8. Comparison of the input impedance magnitude frequency responses of phase A (LV winding) of a three-phase transformer using the core model and the lumped parameter model.

However, there is no reference value for , which represents the total stray capacitances of windings of each phase section in the transformer core model. On the contrary, the reference lumped parameter model divides windings into a group of winding sections, i.e., discs and turns, which are represented by a number of lumped electrical elements such as disc inductance and resistance, capacitance between discs of a winding, capacitance between a disc and a core, etc. Thus, the reference model does not consider total winding capacitances as opposed to the established transformer core model used for the parameter identication. D. Frequency Response Simulation of Transformer Core On the basis of the estimates of and being individually identied in Sections V-B and V-C using FRA references in different frequency ranges, more narrow search space limits can be established for nal identication of the core parameters with GA. These search space limits are listed in Table V. In this case GA learning is conducted using a set of randomly generand for the calculation of with aid ated values of of (8) in the frequency range of 1002500 Hz. As seen from Table V the nal GA identied values of and are very close to the corresponding reference values having negligible deviations in a practical sense. The identied parameters are then used to simulate the input impedance magnitude frequency response of phase A and B in Figs. 8 and 9 with the established transformer core model, where the corresponding reference responses of the three-phase lumped parameter model are also given. The responses in Fig. 8 show the rst two resonance peaks, appearing on the frequency responses taken from the lateral phase A winding of the transformer. This can be explained by the different ux paths through the central and the lateral limbs

Fig. 9. Comparison of the input impedance magnitude frequency responses of phase B (LV winding) of a three-phase transformer using the core model and the lumped parameter model.

of the transformer core as shown in Fig. 1. This results to the two different reluctances of the core equivalent magnetic circuit as expressed by (5). On the other hand, only one resonance peak appears at low frequencies on the impedance responses of the central phase B winding in Fig. 9 due to same ux paths through the lateral limbs of the core [4]. The comparison of the responses in the two gures shows a good resemblance both in resonance frequencies and magnitudes between the responses at low frequencies associated with the transformer core. This shows the accurate identication of lamination parameters of transformer core using FRA measurements. As a result, an accurate model of transformer core is developed, which can be applied to FRA result interpretation at low frequencies.

148

IEEE TRANSACTIONS ON MAGNETICS, VOL. 46, NO. 1, JANUARY 2010

VI. CONCLUSION In this paper a novel model-based identication approach has been developed to determine parameters of a demagnetized transformer core on the basis of FRA measurements, without requirements to carry out additional experimental tests on lamination samples. Based upon the duality principle between magnetic and electrical circuits, a transformer core model is established and employed for model-based parameter identication with GA. For numerical study, reference input impedance frequency responses are obtained using a well-known lumped parameter model of a three-phase transformer including core effect. The initial search space for GA learning is dened based on the estimated values and the properties of widely used lamination material. The simulation results show that the presented identication approach allows to determine accurate parameter values with respect to the reference ones, where deviations between the identied core parameters and the reference ones are negligible in a practical sense. The core model is further applied to simulate frequency responses of different phase windings, which show the potential for accurate transformer core modeling. Comparing with the previously developed methods for magnetic parameters estimation [15], [16], the proposed approach establishes and utilizes the frequency-dependent core model, which has a simple form and a clear physical meaning. As a result, it possesses great practicability for FRA result interpretation at low frequencies. It should be mentioned that in the current study only the linear case without hysteresis is considered. Therefore, further research should be directed to employ a more complex model of frequency dependent behavior of magnetic materials taking into account the hysteresis effect [33], [34]. Furthermore, since only simulation studies are conducted in this study, a further investigation should needs to be undertaken to verify the proposed identication approach with experimental measurements. ACKNOWLEDGMENT The rst author would like to thank the Center for International Programs for awarding Kazakhstan Presidential Bolashak Scholarship and JSC Science Fund for a nancial grant within the frame of the Sharyktau competition to support his Ph.D. research in the University of Liverpool, U.K. REFERENCES
[1] S. A. Ryder, Diagnosing transformer faults using frequency responce analysis, IEEE Electr. Insul. Mag., vol. 19, no. 2, pp. 1622, 2003. [2] E. Rahimpour, J. Christian, and K. Feser, Transfer function method to diagnose axial displacement and radial deformation of transformer windings, IEEE Trans. Power Del., vol. 18, no. 2, pp. 493505, 2003. [3] M. Florkowski and J. Furgal, Detection of transformer winding deformations based on the transfer function-measurements and simulations, Meas. Sci. Technol., vol. 14, no. 11, pp. 19861992, 2003. [4] N. Abeywickrama, Y. V. Serdyuk, and S. M. Gubanski, Exploring possibilities for characterisation of power transfromer insulation by frequency response analysis (FRA), IEEE Trans. Power Del., vol. 21, no. 3, pp. 13751382, 2006. [5] N. Abeywickrama, Y. V. Serdyuk, and S. M. Gubanski, High-frequency modeling of power transfromers for use in frequency response analysis (FRA), IEEE Trans. Power Del., vol. 23, no. 4, pp. 20422049, 2008.

[6] E. Bjerkan, High Frequency Modelling of Power Transformers: Stresses and Diagnostics, Ph.D. dissertation, Norwegian Univ. Sci. Technol., Norway, 2005. [7] S. P. Ang, J. Li, Z. Wang, and P. Jarman, FRA low frequency characteristics study using duality transformer core modeling, in Proc. 2008 IEEE Int. Conf. Condition Monitoring and Diagnostics, Beijing, China, 2008. [8] E. C. Cherry, The duality between interlinked electric and magnetic circuits and the formation of transformer equivalent circuits, Proc. Phys. Soc., vol. 62B, pp. 101111, 1949. [9] G. R. Slemon, Magnetoelectric Devices: Transducers, Transformers, and Machines. New York: Wiley, 1966. [10] B. A. Mork, Five-legged wound-core transformer model: Derivation, parameters, implementation and evaluation, IEEE Trans. Power Del., vol. 14, no. 4, pp. 15191526, 1999. [11] M. Elleuch, New transformer model including joint air gaps and lamination anisotropy, IEEE Trans. Magn., vol. 34, no. 5, pp. 37013711, Sep. 1998. [12] F. Zhalefar and M. Sanaye-Pasand, Transformer core modeling as applied to slow transients and protective studies, in Proc. 2006 IEEE Power India Conf., 2006. [13] J. Pleite, C. Gonzles, J. Vzquez, and A. Lzaro, Power transformer core fault diagnosis using frequency response analysis, in Proc. IEEE MELECOM 2006 Conf., Mlaga, Spain, 2006. [14] J. Pleite, R. Prieto, R. Asensi, J. Cobos, and E. Olras, Obtaining a frequency-dependent and distributed-effect model of magnetic components from actual measurements, IEEE Trans. Magn., vol. 35, no. 6, pp. 44904502, Nov. 1999. [15] N. Abeywickrama, T. Daszczynski, Y. V. Serdyuk, and S. M. Gubanski, Determination of complex permeability of silicon steel for use in high frequency modeling of power transformers, IEEE Trans. Magn., vol. 44, no. 4, pp. 438444, Apr. 2008. [16] D. Roger, E. Napieralska-Juszczak, and A. Henneton, High frequency extension of non-linear models of laminated cores, Int. J. Comput. Math. Electr. Electron. Eng., vol. 25, no. 1, pp. 140156, 2006. [17] N. Abeywickrama, A. D. Podoltsev, Y. V. Serdyuk, and S. M. Gubanski, Computation of parameters of power transfromer windings for use in frequency response analysis (FRA), IEEE Trans. Magn., vol. 43, no. 5, pp. 19831990, May 2007. [18] A. D. Podoltsev, N. Abeywickrama, Y. V. Serdyuk, and S. M. Gubanski, Multiscale computation of parameters of power transfromer windings at high frequencies. Part I: Small-scale level, IEEE Trans. Magn., vol. 43, no. 11, pp. 39913996, Nov. 2007. [19] A. D. Podoltsev, N. Abeywickrama, Y. V. Serdyuk, and S. M. Gubanski, Multiscale computation of parameters of power transfromer windings at high frequencies. Part II: Large-scale level, IEEE Trans. Magn., vol. 43, no. 12, pp. 40764082, Dec. 2007. [20] D. E. Goldberg, Genetic Algorithms in Search, Optimization, and Machine Learning. Reading, MA: Addison-Wesley Longman, 1989. [21] J. Kennedy and R. Eberhart, Swarm Intelligence.. San Mateo, CA: Morgan Kaufmann, 2001. [22] W. H. Tang, S. He, Q. H. Wu, and Z. J. Richardson, Winding deformation identication using a particle swarm optimiser with passive congregation for power transformers, Int. J. Innov. Energy Syst. Power, vol. 1, no. 1, 2006. [23] V. Rashtchi, E. Rahimpour, and E. Rezapour, Using a genetic algorithm for parameter identication of transformer R-L-C-M model, Electr. Eng., vol. 88, no. 5, pp. 417422, 2006. [24] N. Abeywickrama, Y. V. Serdyuk, and S. M. Gubanski, Effect of core magnetization on frequency response analysis (FRA) of power transformers, IEEE Trans. Power Del., vol. 23, no. 3, pp. 14321438, 2008. [25] E. Dallago, G. Sassone, and G. Venchi, High-frequency power transformer model for circuit simulation, IEEE Trans. Power Electron., vol. 12, no. 4, pp. 664670, 1997. [26] H. Y. Lu, J. G. Zhu, and S. Y. R. Hui, Experimental determination of stray capacitances in high frequency transformers, IEEE Trans. Power Electron., vol. 18, no. 5, pp. 11051112, 2003. [27] A. Massarini and M. K. Kazimierczuk, Self-capacitance of inductors, IEEE Trans. Power Electron., vol. 12, no. 4, pp. 671676, 1997. [28] R. L. Haupt and S. E. Haupt, Practical Genetic Algorithms, 2nd ed. New York: Wiley, 2004. [29] W. Banzhaf, P. Nordin, R. E. Keller, and F. D. Francone, Genetic ProgrammingAn Introduction: On the Automatic Evolution of Computer Program and Its Applications. San Francisco, CA: Morgan Kaufmann, 1998. [30] A. J. Moses, Electrical steels: Past, present and future developments, IEE Proc., vol. 137(A), no. 5, pp. 233245, 1990.

SHINTEMIROV et al.: TRANSFORMER CORE PARAMETER IDENTIFICATION USING FREQUENCY RESPONSE ANALYSIS

149

[31] S. E. Zirka, Y. I. Moroz, P. Marketos, A. J. Moses, D. C. Jiles, and T. Matsuo, Generalization of the classical method for calculating dynamic hysteresis loops in grain-oriented electrical steels, IEEE Trans. Magn., vol. 44, no. 9, pp. 21132126, Sep. 2008. [32] Genetic Algorithm and Direct Search Toolbox Users Guide for Use With MATLAB, 2nd ed. Natick, MA: The MathWorks, Inc., 2008. [33] P. Nakmahachalasint, K. D. T. Ngo, and L. Vu-Quoc, A behavioral model for frequency-dependent hysteresis in power ferrites, IEEE Trans. Magn., vol. 40, no. 4, pp. 17841790, Jul. 2004. [34] Y. Zhai and L. Vu-Quoc, Analysis of power magnetic components with nonlinear static hysteresis: Proper orthogonal decomposition and model reduction, IEEE Trans. Magn., vol. 43, no. 5, pp. 18881897, May 2007.

Wenhu Tang (M05) received the B.Eng. and M.Eng. degrees in electrical engineering from Huazhong University of Science and Technology, Wuhan, China, in 1996 and 2000, respectively, and the Ph.D. degree in electrical engineering from The University of Liverpool, Liverpool, U.K., in 2004. He was a Postdoctoral Research Assistant at The University of Liverpool from 2004 to 2006. Since 2006, he has held a Lectureship in Power Engineering in the Department of Electrical Engineering and Electronics, The University of Liverpool, U.K. His research interests are transformer condition monitoring, power system operation, evolutionary computation, multiple criteria decision analysis, and intelligent decision support systems.

Almas Shintemirov was born in 1979. He studied electrical engineering at Pavlodar State University named after S. Toraygirov, Kazakhstan, and received the M.Eng. and Cand. Tech. Sci. (Ph.D.) degrees in 2001 and 2004, respectively. He received the Ph.D. degree in electrical engineering and electronics from the University of Liverpool, Liverpool, U.K., in 2009. His research interests include power transformer winding modeling and condition assessment, power system modeling, and evolutionary computation.

Henry Wu (M91, SM97) received the M.Sc.(Eng.) degree in electrical engineering from Huazhong University of Science and Technology (HUST), China, in 1981 and the Ph.D. degree from The Queens University of Belfast (QUB), U.K., in 1987. From 1981 to 1984, he was appointed Lecturer in Electrical Engineering at HUST. He worked as a Research Fellow and Senior Research Fellow at QUB from 1987 to 1991 and Lecturer and Senior Lecturer in the Department of Mathematical Sciences, Loughborough University, U.K., from 1991 to 1995. Since 1995, he has held the Chair of Electrical Engineering in the Department of Electrical Engineering and Electronics, The University of Liverpool, U.K., acting as Head of the Intelligence Engineering and Automation group. His research interests include adaptive control, mathematical morphology, evolutionary computation, condition monitoring and assessment, and power system control and operation. Prof. Wu is a Chartered Engineer. He is a Fellow of IEE.