International Journal of Applied Power Engineering (IJAPE)

Vol. 13, No. 4, December 2024, pp. 973~981

ISSN: 2252-8792, DOI: 10.11591/ijape.v13.i4.pp973-981  973

Improved convergence speed using hybrid AI for TD EM

modeling in power electronics

Bessem Zitouna1, Mohamed Tlig1,2, Sassia Hedia1, Jaleleddine Ben Hadj Slama1
1
Laboratory of Advanced Technology and Intelligent Systems (LATIS), Université de Sousse,
Ecole Nationale d’Ingénieurs de Sousse, Sousse, Tunisia
2
IRSEEM, ROUEN-Normandie Univ, ESIGELEC, Rouen, France

Article Info ABSTRACT

Article history: This paper presents a time-domain (TD) approach based on hybrid artificial
intelligence (AI) to speed up convergence of radiating sources
Received Oct 24, 2023 characterization in power electronics. To obtain a representative equivalent
Revised Aug 6, 2024 model of device under test, a dedicated optimization framework has been
Accepted Aug 15, 2024 developed in TD using a particle swarm optimization (PSO) toolbox. In
addition, for elementary feature extraction, a pseudo-Zernike moment
invariant (PZMI) descriptor has been defined. Finally, with the aim of
Keywords: identifying remaining dipole parameters and classification problems,
artificial neural networks (ANN) have been implemented. A coupling of TD
Artificial intelligence electromagnetic (EM) inverse method based on a PSO algorithm along with
Electromagnetic compatibility PZMI and ANN application has been investigated and applied to a real test
Near-field case. Experimental measurements have been conducted using the near-field
Power electronics scanning technique above an alternating current (AC)/direct current (DC)
Time-domain converter. Obtained results are discussed based on a comparison between
measured and estimated EM field distributions using both the hybrid AI
method and a conventional TD inverse method based on genetic algorithms
(GA) only. This study confirms that, compared with those given by non-
hybrid method, the proposed algorithm further improves the convergence
speed while maintaining high accuracy. Hence, the present work offers an
impressive perspective for radiated emissions characterization using hybrid
AI algorithms.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Mohamed Tlig
Laboratory of Advanced Technology and Intelligent Systems (LATIS), Université de Sousse
Ecole Nationale d’Ingénieurs de Sousse
Sousse 4023, Tunisia
Email: [email protected]

1. INTRODUCTION
Technological changes speed up exponentially over time. This can be noticed through the highly
progressive miniaturization and embedding of electronic equipment, along with an impressive increase in the
operating frequency, integration rate, and required power. However, this development provokes risks of
frequent and undesired electromagnetic (EM) interferences in and around systems, which can cause its
malfunctioning. Thus, electromagnetic compatibility (EMC) studies have to be investigated before designing
and manufacturing of any electronic device. It has become necessary to develop a representative radiation
model to estimate EM emissions of each embedded system. In this context, the generation of an equivalent
model using an inverse method has been widely proposed [1]-[3]. To circumvent the limits of forward
measurement methods, the EM inverse problem resolution based on the near-field (NF) scanning has been

Journal homepage: http://ijape.iaescore.com

974  ISSN: 2252-8792

extensively developed [1]-[4]. In the literature, to estimate radiated disturbances of power electronic circuits,
the most popular method used is the EM inverse method based on elementary dipoles (electrical and/or
magnetic) [5], [6]. In particular, for energy control and power systems, due to an important switching
activity, high levels of EM interference can be generated. Consequently, a wide frequency band and an
important transient response are issued. Indeed, it is difficult to handle EM inverse methods in the frequency
domain [3]. Therefore, researchers have investigated the use of time-domain (TD) analysis to obtain an
equivalent model representing the radiation behavior of the device under test (DUT) for all the frequency
band [2]-[4], [7]. These studies have shown adequate results in particular when the number of variables is not
very important. However, in the case of a complex structure, some weaknesses can be noticed such as a
computational complexity and the lack of convergence warranties. As a solution, advanced artificial
intelligence (AI) has been introduced to provide more efficiency in characterizing radiating
structures [8]-[11]. These algorithms are adequate for the identification and optimization problems [12]. For
instance, Labiedh and Slama [13] have investigated the use of the particle swarm optimization (PSO)
technique for EMC application. However, these results suggest multiple possible improvements. Thus,
further extension studied can be made for intuitively obtaining the optimal patterns from the AI models in the
least computation time.
In this work, a hybrid AI algorithm is proposed to further improve the convergence speed of the TD
EM inverse method. The workflow is based on the automation of the pre-analysis processing of radiation
maps using TD analysis. This will help in deducing more significant information on the obtained elementary
dipole characteristics, such as the type (magnetic or electrical) as well as position coordinates parameters (xd,
yd, zd). By analogy with the FD–based method [14], the pre-analytical phase has been carried out using the
coupling of the pseudo-Zernike moment invariant (PZMI) and the artificial neural network (ANN). The latter
has been introduced as an image-processing technique for EMC applications. Indeed, PZMI has been
successfully used in the field of pattern recognition for feature extraction and it can provide information
about system redundancy [15], [16]. In addition, ANN is used for clustering and in supervised
classification [17]. For EMC studies of power electronics, these methods appear very interesting for radiating
dipoles identification. However, it is still crucial to involve a good performance criterion allowing finding an
optimum or a satisfactory solution for the inverse problem resolution. Preview study [18], it has been
demonstrated that for this purpose PSO technique is more efficient than genetic algorithms (GA) in terms of
complexity, iteration and accuracy. Besides, PSO has also ease of implementation. For this purpose, to
guarantee convergence and reduce the processing time, we suggest applying the PSO method using TD
analysis in a hybrid approach. The proposed method is a coupling of the EM TD inverse method based on a
PSO algorithm along with a PZMI descriptor and an ANN, which is an original use of AI algorithms in EMC
applications. For validation purposes, experimental measurements have been carried out using a magnetic
field probe. An automated NF test bench has been developed for TD measurement. The device under test
(DUT) is a flyback alternating current (AC)/direct current (DC) converter typically found in low and
medium-power applications as in automotive and battery charging. In these kind of circuits, multiple
transient disturbances with a short duration occur and affect the functioning of the system.
The organization of the paper is as follows: It starts by presenting the TD EM inverse problem with
respect to NF scanning. Then, in section 3, the proposed hybrid AI algorithm is described for equivalent
radiation model investigations. Section 4 includes implementation and a real application of the procedure,
followed by a detailed discussion. For comparison purposes, measured and obtained results have been
exposed to those found by temporal inverse method based on GA only. Finally, section 5 presents the main
conclusions.

2. TIME-DOMAIN ELECTROMAGNETIC INVERSE METHOD

To fulfill the demanding needs of recent technological advancements, such as in power systems
where undesired EM radiation has to be investigated at early design stages, in addition, to forward
characterization problem, inverse methods have been widely developed. It consists in retrieving various
unknown information of the model inputs starting from a given response (outputs). In the literature, this
process has been applied for different purposes, mainly to provide a framework for the characterization of
black-box systems behavior [19].
In EMC studies, inverse problem has been employed to identify potential emissive sources through
an equivalent radiation model of the DUT and hence point out harmful components. Authors in [9] and [20]
have proposed to use the EM inverse method along with TD analysis. In fact, either it is possible to provide
temporal signals of EM fields by taking measurements with high precision timing instruments in the NF
region of DUT or by simulations using numerical solutions or using temporal analytical equations that
describe EM radiation of elementary dipoles [3], [9]. The aim is to obtain a set of equivalent electric,

Int J Appl Power Eng, Vol. 13, No. 4, December 2024: 973-981
Int J Appl Power Eng ISSN: 2252-8792  975

magnetic or both dipoles type representing an accurate radiation model along time. Indeed, in NF, there are
strong capacitive and inductive effects from currents and charges flowing in the circuit and generating
considerable EM disturbances. The acquired equivalent model provides an approximate estimation that
represents at best the real DUT radiation behavior. Figure 1 shows a flowchart picturing the principle of the
studied method.

Figure 1. Principle of TD EM inverse method

3. PROPOSED HYBRID AI METHOD

Several research papers have investigated the application of artificial intelligence in power systems.
For instance, an interesting work has been presented in [21]. Other approaches using AI algorithms in the
study of EM interferences have been introduced in [11]-[13] and [22]. Preview study [14], great work has
been done in the frequency domain in order to introduce machine learning-based methods for EMC studies.
In time-domain, Zitouna and Slama [3] have proposed the use of an EM inverse method based on genetic
algorithms only. This method consists in identifying a single radiating dipole per local map (scan window)
until finding all equivalent sources while going through the total field map. Indeed, an optimization process is
carried out to search a magnetic dipole in a first iteration, and then if a convergence has not been achieved,
the search feature becomes an electric dipole and so on. Consequently, this procedure is time and memory
consuming. It is necessary to switch to a more suitable method that requires a minimum of information on the
type of dipole and its involved parameters. In this study, a hybrid artificial intelligence algorithm is
developed based on the coupling of three techniques; PSO as an optimization method, PZMI descriptor, and
artificial neural network (ANN) for the evaluation of the main dipole characteristics. On the one hand, PSO is
a population-based search algorithm where the system is initialized with a population of random solutions,
and the investigation for an optimal solution is carried out by updating generations. Hence, the potential
solutions (particles) move in the problem space by following the current optimum particles. In the literature,
PSO has been applied for several optimization studies such as explained in [13], [18], and [23]-[25]. Unlike
GA, PSO has more ease of implementation and no evolution operators, such as mutation and crossover.
Therefore, it is computationally more efficient in terms of memory and speed requirements. On the other
hand, the PZMI descriptors are employed for pattern recognition. It has been widely established that PZMI
allows capturing information from the map with a satisfactory robustness to noise [14], [16], [26], [27]. In
addition, ANN technique has been extremely successful in clustering and classification processes in
particular in the areas of computer vision [14], [20], [21]. In this work, coupling of PZMI and ANN is
employed as an image-processing tool with an aim to classify radiating sources using TD analysis. To
implement this algorithm, MATLAB toolboxes such as particle swarm optimization research toolbox and
neural network toolbox are employed.
The proposed TD search algorithm treats first the highest radiation level (potential source) and
continues until reaching a weak magnitude. To reconstruct the shape of each dipole excitation signal with a
good accuracy, the initially measured (or simulated) radiation map is introduced to a subprogram that seeks
Improved convergence speed using hybrid AI for TD EM modeling in power electronics (Bessem Zitouna)
976  ISSN: 2252-8792

the signal with the maximum amplitude at each iteration. Based on the measurement position that
corresponds to the most intense radiation area, we extract a scan window and propose an elementary dipole
for this zone. The extracted window scans the entire map. It is worth noting that the size of the scan window
relies on the measurement height (distance between the measuring probe and DUT) and the step of
displacement of the probe between two successive measuring points. In extreme cases where several
radiating sources are very close to each other and are located in the same extraction window, the algorithm
will gradually decrease the size of the window until the convergence is achieved. If the method fails to
converge, the initial sizes of the extraction window are taken over and the search frame is extended to more
than one equivalent dipole.
In the context of inverse problems, a critical step in the problem setup is the configuration of the
model, which means to characterize the solution by a set of parameters (working variables and desired
outputs). This choice is not evident because the solution may not be stable, exist or be unique and the model
may need to define some assumptions. Thus, to facilitate the convergence of the search algorithm and avoid
possible couplings, a reduced set of parameters is chosen. In this study, as in [3], parameters are the radiating
source center coordinates (𝑋d, 𝑌d, d), the orientation angles (θ, φ), and the moment waveform (Md). The
main steps of the proposed hybrid AI method are summarized in Figure 2 and explained below, this
procedure is repeated until the defined noise level is reached and all sources of radiation on the maps are
identified:
- In the extracted scan window, an image-processing algorithm utilizing PZMI descriptors and an artificial
neural network (ANN) is implemented to determine the position and type of the equivalent dipole. More
theoretical details are provided in [14].
- To accelerate the convergence of the proposed method, for optimization routines, a PSO algorithm is
carried out to search remaining parameters (Md, θ, φ), as in [13].
- Consequently, the identified parameters of the first dipole are used to calculate the magnetic field emitted
by identified dipole using analytical equations of the magnetic field. Then, the generated distribution is
subtracted from initial cartography.

Figure 2. Summary of the proposed hybrid AI approach

Int J Appl Power Eng, Vol. 13, No. 4, December 2024: 973-981
Int J Appl Power Eng ISSN: 2252-8792  977

4. RESULTS AND DISCUSSION

4.1. Application to an AC/DC
The most practical way to evaluate the proposed method is to apply it to a real test case. Hence, for
experimentation purpose, an AC/DC converter of a flyback topology has been studied, as show in Figure 3.
More details of the DUT and the test environment can be found in [9]. In order to perform TD measurements,
a NF test bench has been developed using a high precision oscilloscope with a wide bandwidth. Magnetic
field signals are measured using a magnetic probe of 1.6 mm radius that moves above the circuit board to
scan the whole surface. Since it is mandatory to take measurements under the same conditions, the motion of
the shielded probe is automated and can be carried out in the three axes (x, y, and z). Acquisition has to be
synchronized with a reference signal, which is chosen to be the control signal of the optocoupler of the
converter because of its receptivity, periodicity and electrical isolation. In order to obtain a distribution map
per time step, TD signals showing the system behavior are measured for a defined duration. An example of
EM field signals of 20 us duration is depicted in Figure 4. As we can see, radiated signals are of different
shapes along time. It is not evident to be treated using FD domain method (extraction radiation maps at all
spectrum frequencies). Furthermore, in this test case, we have focused on the following working variable
configurations:
- The height of measurement: in order to avoid troublesome components while using the probe, we have
preferred to carry out measurements on the bottom side of the board. Thus, for this particular practical
case, it is possible to define a good distance of scan. We have measured at 12 mm of height.
- The scanning area has to include the surface of interest of (38 mm × 54 mm). The probe moved into 560
different scanning points with a step of 2 mm.
Source identification is a supervised learning issue. ANN has been employed to classify radiating
dipoles using PZMI vectors. We have used a simple structure of multi-layer perceptron network (MLP) that
contains the PZMI vector as an input, along with hidden and output layers giving identified sources
information. A back-propagation learning phase has been carried out to make a training with a database of
1158 PZMI vectors (from different maps). Indeed, 30% of the database has been employed for testing and the
rest has been generated to create the NN. Table 1 presents the learning data set. To evaluate the efficiency
and accuracy of the proposed hybrid AI method, its results are compared with those obtained using a non-
hybrid approach. By applying a genetic algorithm (GA)-based method to the same measured time-domain
(TD) signals, we generate alternative maps to represent the radiation behavior of the studied converter.
Information’s about the chosen parameters for each optimization method are given in Table 2.

Figure 3. The studied structure

Figure 4. Measured radiation signal

Improved convergence speed using hybrid AI for TD EM modeling in power electronics (Bessem Zitouna)
978  ISSN: 2252-8792

4.2. Conventional method/hybrid AI method comparison

Tables 3 and 4 present the obtained results using a conventional method base on GA algorithm and
those obtained by using the proposed method. Based on obtained results in Tables 3 and 4, we can conclude
that methods based on the proposed hybrid method as well as the one based on the GA only give efficient and
accurate results that represents well the DUT radiation sources. Figure 5 illustrates an estimation of the
obtained positions using the proposed hybrid method. It is worth noting that this representation highlights
that during the operation of the converter, the transformer has been the most radiating component. Then we
can list the transistor, the inductor, and lastly the diodes. Estimated EM radiation maps at t = 9.95 µs and
t = 9.78 µs based on identified dipoles parameters using both methods are presented in Figure 6. The second
line represents the magnetic field distribution derived from the time-domain (TD) electromagnetic inverse
method using a genetic algorithm (GA). The third line displays the results obtained using the proposed hybrid
AI method, which integrates particle swarm optimization (PSO), PZMI descriptors, and artificial neural
network (ANN) algorithms.
It is crucial to highlight that employing both methods. In particular, the hybrid AI method using the
coupling of PZMI/ANN/PSO-based has given about 8% relative error, as shown in Table 5. Indeed, the
major advantage of implementing a hybrid method is the significant reduction in computing time by about 35
times less than the GA-based method, which is an interesting achievement. Therefore, using the proposed
hybrid AI method can be an original solution when dealing with real complex structures such as those
employed in power systems. It is worth noting that it is possible to perform measurements and apply the
proposed method also for Hx and Hy components.

Table 1. ANN database

The learning database Output
Test base 348 PZMI vectors
Magnetic dipole 370 with output 0
Electric dipole 370 with output 1
Indifined cases 70 with output 0.5

Table 2. Control parameters

PSO parameter Value GA Value
Error limit 10% Error limit 10%
Swarm size 48 Population size 120
Nombre of generations Inf Nombre of generations Inf
Number of parameters 6 Number of parameters 6

Table 3. Obtained results using a conventional GA-based method

Non-hybrid method
N Dipole Md (A/m²) × e-7 Xd Yd Zd (mm) θi(rad) φi(rad)
1 M 3.12 14.6, -13.3, -6.45 0.01 0.13
2 M 4.42 14.2, -3.2, -7.9 1.02 -2.32
3 M 11.1 1, -2.1, -12.1 0.16 0.17
4 M 2.96 -9.6, 2, -6.1 -0.2 -0.31
5 M 3.86 -13.4, 11.9, -6.5 -0.76 2.25
6 M 1.48 -3.5, -10.8, -3.7 1.27 3.13
7 M 0.18 -13.6, -7.4, -6.2 1.45 0.15

Table 4. Obtained results using the proposed method

Hybrid method
N Dipole Md (A/m²) × e-7 Xd Yd Zd (mm) θi(rad) φi(rad)
1 M 3.11 14.71, -13.25, -6.44 0.02 0.17
2 M 4.41 14.1, -3.3, -7.86 0.97 -2.39
3 M 11.07 1.1, -1.99, -12.05 0 3.13
4 M 2.99 -9.61, 2.08, -6.17 -0.18 0.1
5 M 3.83 -13.09, 12.1, -6.52 -0.77 2.24
6 M 1.51 -3.49, -11.1, -3.7 1.26 3.11
7 M 0.19 -14.2, -7.41, -6.03 1.43 0

Table 5. Performance of each method

PZMI/ANN/PSO-based method Value GA-based method Value
Relative error (%) 7.97 Relative error (%) 9.51
computing time (min) 38.26 computing time (min) 1355.45

Int J Appl Power Eng, Vol. 13, No. 4, December 2024: 973-981
Int J Appl Power Eng ISSN: 2252-8792  979

Figure 5. Approximate representation of equivalent sources positions using the hybrid method results

Figure 6. Measured and estimated maps using both non-hybrid and hybrid methods at two-time steps

5. CONCLUSION
This paper has outlined a detailed investigation of a hybrid AI algorithm for EMC applications. The
proposed method based on coupling of PZMI descriptor, ANN, and PSO technique has been implemented.
To assess and validate the proposed approach, we applied it to a measured NF radiation signals emitted by a
flyback AC/DC converter. This structure is complex enough to validate the performance of the proposed
method, especially when using experimental measurements. A dedicated TD measurement test bench has
been developped for the characterization of radiating sources. A comparison between TD inverse method
based on GA only and the proposed hybrid method has proven the robustness of the latter and shown a very
intresting improvement in the speed of convergence. The equivalent model parameters obtained are
comparable to those from the non-hybrid method. The proposed approach has demonstrated satisfactory
performance in both efficiency and accuracy. It is worth noting that obtained results are an example of the
high potential of using hybrid AI-based methods in EMC applications. It opens the way to more interesting
studies, particularly in the area of power electronic systems.

Improved convergence speed using hybrid AI for TD EM modeling in power electronics (Bessem Zitouna)
980  ISSN: 2252-8792

REFERENCES
[1] H. Endo, T. Takagi, and Y. Saito, “A new current dipole model satisfying current continuity for inverse magnetic field source
problems,” IEEE Transactions on Magnetics, vol. 41, no. 5, pp. 1748–1751, May 2005, doi: 10.1109/TMAG.2005.846037.
[2] L. Beghou, F. Costa, and L. Pichon, “Detection of electromagnetic radiations sources at the switching time scale using an inverse
problem-based resolution method—application to power electronic circuits,” IEEE Transactions on Electromagnetic
Compatibility, vol. 57, no. 1, pp. 52–60, Feb. 2015, doi: 10.1109/TEMC.2014.2363675.
[3] B. Zitouna and J. B. H. Slama, “Enhancement of time-domain electromagnetic inverse method for modeling circuits radiations,”
IEEE Transactions on Electromagnetic Compatibility, vol. 58, no. 2, pp. 534–542, Apr. 2016, doi: 10.1109/TEMC.2016.2520882.
[4] Y. Liu, B. Ravelo, and A. K. Jastrzebski, “Time-domain magnetic dipole model of PCB near-field emission,” IEEE Transactions
on Electromagnetic Compatibility, vol. 58, no. 5, pp. 1561–1569, Oct. 2016, doi: 10.1109/TEMC.2016.2578953.
[5] H. Song, J. Chen, D. Zhou, D. Hou, and J. Lin, “An equivalent model of magnetic dipole for the slot coupling of shielding
cavity,” in 2008 8th International Symposium on Antennas, Propagation and EM Theory, Nov. 2008, pp. 970–973, doi:
10.1109/ISAPE.2008.4735382.
[6] A. Junge and F. Marliani, “Prediction of DC magnetic fields for magnetic cleanliness on spacecraft,” in 2011 IEEE International
Symposium on Electromagnetic Compatibility, Aug. 2011, pp. 834–839, doi: 10.1109/ISEMC.2011.6038424.
[7] M. Gustafsson and S. He, “An optimization approach to two‐dimensional time domain electromagnetic inverse problems,” Radio
Science, vol. 35, no. 2, pp. 525–536, Mar. 2000, doi: 10.1029/1999RS900091.
[8] D. D. Micu, C. C. Georgios, and L. Czumbil, “Artificial intelligence techniques applied to electromagnetic interference problems
between power lines and metal pipelines,” in Recurrent Neural Networks and Soft Computing, 2012, pp. 253–274, doi:
10.5772/37637.
[9] B. Zitouna and J. B. H. Slama, “Time domain inverse method based on the near field technique to solve electromagnetic
interference problems: application to an AC/DC flyback converter,” IET Power Electronics, vol. 11, no. 13, pp. 2133–2139, Nov.
2018, doi: 10.1049/iet-pel.2018.5157.
[10] P. Eliardsson and P. Stenumgaard, “Artificial intelligence for automatic classification of unintentional electromagnetic
interference in air traffic control communications,” in 2019 International Symposium on Electromagnetic Compatibility - EMC
EUROPE, Sep. 2019, pp. 896–901, doi: 10.1109/EMCEurope.2019.8872082.
[11] L. Czumbil, D. D. Micu, and A. Ceclan, “Artificial intelligence techniques applied to electromagnetic interference problems,” in
International Conference on Advancements of Medicine and Health Care through Technology: IFMBE Proceedings, 2009, pp.
339–344, doi: 10.1007/978-3-642-04292-8_75.
[12] Y.-F. Shu, X.-C. Wei, J. Fan, R. Yang, and Y.-B. Yang, “An equivalent dipole model hybrid with artificial neural network for
electromagnetic interference prediction,” IEEE Transactions on Microwave Theory and Techniques, vol. 67, no. 5, pp. 1790–
1797, May 2019, doi: 10.1109/TMTT.2019.2905238.
[13] W. Labiedh and J. B. H. Slama, “Contribution of the PSO in the electromagnetic inverse method in terms of convergence and
simplicity of implementation,” Journal of Electromagnetic Waves and Applications, vol. 28, no. 18, pp. 2339–2349, Dec. 2014,
doi: 10.1080/09205071.2014.967366.
[14] S. Saidi and J. B. H. Slama, “A near-field technique based on PZMI, GA, and ANN: application to power electronics systems,”
IEEE Transactions on Electromagnetic Compatibility, vol. 56, no. 4, pp. 784–791, Aug. 2014, doi:
10.1109/TEMC.2013.2290271.
[15] G. Y. Chen and C. Li, “Ridgelet moment invariants for robust pattern recognition,” Pattern Analysis and Applications, vol. 24,
no. 3, pp. 1367–1376, Aug. 2021, doi: 10.1007/s10044-021-00996-8.
[16] D. An-Wen and G. Chih-Ying, “Fast and stable algorithms for high-order pseudo Zernike moments and image reconstruction,”
Applied Mathematics and Computation, vol. 334, pp. 239–253, Oct. 2018, doi: 10.1016/j.amc.2018.04.001.
[17] Z. Zhang, “A gentle introduction to artificial neural networks,” Annals of Translational Medicine, vol. 4, no. 19, p. 370, Oct.
2016, doi: 10.21037/atm.2016.06.20.
[18] S. M. Almufti, A. Y. Zebari, and H. K. Omer, “A comparative study of particle swarm optimization and genetic algorithm,”
Journal of Advanced Computer Science & Technology, vol. 8, no. 2, pp. 40–45, Oct. 2019, doi: 10.14419/jacst.v8i2.29401.
[19] X. Chen, Z. Wei, M. Li, and P. Rocca, “A review of deep learning approaches for inverse scattering problems (invited review),”
Progress In Electromagnetics Research, vol. 167, pp. 67–81, 2020, doi: 10.2528/PIER20030705.
[20] S. Hedia, B. Zitouna, J. B. H. Slama, and L. Pichon, “Electromagnetic time reversal in the near field: characterization of transient
disturbances in power electronics,” IEEE Transactions on Electromagnetic Compatibility, vol. 62, no. 5, pp. 1869–1878, Oct.
2020, doi: 10.1109/TEMC.2020.2965735.
[21] M. A. Laughton, “Artificial intelligence techniques in power systems,” in IEE Colloquium on Artificial Intelligence Techniques in
Power Systems, 1997, pp. 1/1-119, doi: 10.1049/ic:19971179.
[22] S. Poudel and U. R. Dhungana, “Artificial intelligence for energy fraud detection: a review,” International Journal of Applied
Power Engineering (IJAPE), vol. 11, no. 2, pp. 109–119, Jun. 2022, doi: 10.11591/ijape.v11.i2.pp109-119.
[23] V. A. Kusuma, A. A. Firdaus, S. S. Suprapto, D. F. U. Putra, Y. Prasetyo, and F. Filliana, “Leveraging PSO algorithms to achieve
optimal stand-alone microgrid performance with a focus on battery lifetime,” International Journal of Applied Power Engineering
(IJAPE), vol. 12, no. 3, pp. 293–299, Jul. 2023, doi: 10.11591/ijape.v12.i3.pp293-299.
[24] L. Meng, S. Yin, and X. Hu, “An improved Mamdani fuzzy neural networks based on PSO algorithm and new parameter
optimization,” Indonesian Journal of Electrical Engineering and Computer Science, vol. 1, no. 1, pp. 201–206, Jan. 2016, doi:
10.11591/ijeecs.v1.i1.pp201-206.
[25] D. Craciun, Y. Besanger, and N. Hadjsaid, “Dynamic equivalents of distribution power systems with multiple distributed
generations using a PSO based estimation method,” in IEEE, General Meeting 2010, Jul 2010, Minneapolis, United States, 2010.
[Online]. Available: https://hal.science/hal-00523400.
[26] C. L. Deepika, D. A. Kandaswamy, C. Vimal, and B. Sathish, “Invariant feature extraction from fingerprint biometric using
pseudo Zernike moments,” Singaporean Journal Scientific Research (SJSR), vol. 3, no. 2, pp. 150–154, 2010.
[27] H. Zhu, M. Liu, H. Ji, and Y. Li, “Combined invariants to blur and rotation using Zernike moment descriptors,” Pattern Analysis
and Applications, vol. 13, pp. 309–319, Aug. 2010, doi: 10.1007/s10044-009-0159-9.

Int J Appl Power Eng, Vol. 13, No. 4, December 2024: 973-981
Int J Appl Power Eng ISSN: 2252-8792  981

BIOGRAPHIES OF AUTHORS

Bessem Zitouna received the master of research and Ph.D. degrees, both in
electrical engineering, from the National Engineering School of Tunis (ENIT), Tunisia, in
2007 and 2015, respectively. He carries out research tasks in the Laboratory of Advanced
Technology and Intelligent Systems (LATIS). His research interests include electromagnetic
compatibility (EMC), near-field techniques, modeling, radiated emissions, and power
electronic systems. He can be contacted at email: [email protected].

Mohamed Tlig received the master of research and Ph.D. degrees, both in
electrical engineering, from the National Engineering School of Tunis (ENIT), Tunisia, in
2011 and 2015, respectively. He is a member of LATIS Laboratory. Currently, he is an
associate professor at the National Engineering School of Sousse (ENISo). His main research
interests include electromagnetic compatibility (EMC) of transportation systems, reliability of
power components and conducted EMI measurements. He can be contacted at email:
[email protected].

Sassia Hedia received the Ph.D. degree in electrical engineering from the
University of Paris-Saclay, France and University of Sousse, Tunisia. She also received the
M.Res. degree in intelligent and communicating systems and the engineering degree in the
industrial electronics engineering from the National Engineering School of Sousse, Tunisia, in
2016 and 2015, respectively. Her current research interests include embedded electronics
engineering, timing, EMC, modeling, radiated emissions, and near-field techniques. She can
be contacted at email: [email protected].

Jaleleddine Ben Hadj Slama received the engineer and Ph.D. degrees in
electrical engineering from the École Centrale de Lyon, France, in 1994 and 1997,
respectively. Since 2015, he has been a full professor of electrical engineering at ENISo,
Tunisia, where he is currently the Leader of the Power Quality Research Group at LATIS
Laboratory. His research interests include reliability of transportation systems, EMC, near-
field techniques, EMI issues of power systems, modeling, smart grid, green datacenters,
renewable energy, and development of remote engineering education. He can be contacted at
email: [email protected].

Improved convergence speed using hybrid AI for TD EM modeling in power electronics (Bessem Zitouna)