Computers and Electrical Engineering 123 (2025) 110248

Contents lists available at ScienceDirect

Computers and Electrical Engineering

journal homepage: [Link]/locate/compeleceng

Revolutionizing power electronics design through large language

models: Applications and future directions
Khalifa Aliyu Ibrahim a ,∗, Patrick Chi-Kwong Luk a , Zhenhua Luo a , Seng Yim Ng b ,
Lee Harrison b
a Centre for Energy Engineering, Cranfield University, MK43 0AL, United Kingdom
b Energy Research Lab, Cranfield Technology Park, Unit 28/29 Innovation Centre, MK43 0AL, United Kingdom

ARTICLE INFO ABSTRACT

Keywords: The design of electronic circuits is critical for a wide range of applications, from the electri-
Artificial intelligence fication of transportation to the Internet of Things (IoT). It demands substantial resources, is
Power electronics design time-intensive, and can be highly intricate. Current design methods often lead to inefficiencies,
AI driven design
prolonged design cycles, and susceptibility to human error. Advancements in artificial intelli-
Large language model
gence (AI) play a crucial role in power electronics design by increasing efficiency, promoting
High frequency AC (HFAC)
Wireless power transfer
automation, and enhancing sustainability of electrical systems. Research has demonstrated the
Generative pre-train transformer (GPT) applications of AI in power electronics to enhance system performance, optimization, and
control strategy using machine learning, fuzzy logic, expert systems, and metaheuristic methods.
However, a review that includes the recent AI advancements and potential of large language
models (LLMs) like generative pre-train transformers (GPT) has not been reported. This paper
presents an overview of applications of AI in power electronics (PE) including the potential of
LLMs. The influence of LLMs-AI on the design process of PE and future research directions is
also highlighted. The development of advanced AI algorithms such as pre-train transformers,
real-time implementations, interdisciplinary collaboration, and data-driven approaches are also
discussed. The proposed LLMs-AI is used to design parameters of high-frequency wireless power
transfer (HFWPT) using MATLAB as a first case study, and high-frequency alternating current
(HFAC) inverter using PSIM as a second case study. The proposed LLM-AI driven design is
verified based on a similar design reported in the literature and Wilcoxon signed-rank test was
conducted to further validate the result. Results show that the LLM-AI driven design based
on the OpenAI foundation model has the potential to streamline the design process of power
electronics. These findings provide a good reference on the feasibility of LLMs-AI on power
electronic design.

1. Introduction

The demand for efficient power electronics (PE) is increasing due to rapid developments in data science, the Internet of Things
(IoT), renewable energy, automobiles, sensor technology, edge computing, household appliances and big data analytics [1–3].
PE faces challenges such as thermal management, electromagnetic interference due to high frequency operation, scalability, and
integration with renewable energy systems. Additionally, fault detection, recycling, compliance with international standards and
sustainability in manufacturing remain critical issues. Addressing these challenges requires advanced tools like and methodologies

∗ Corresponding author.
E-mail addresses: [Link]@[Link] (K.A. Ibrahim), [Link]@[Link] (P.C.-K. Luk).

[Link]
Received 5 October 2024; Received in revised form 18 February 2025; Accepted 4 March 2025
Available online 20 March 2025
0045-7906/© 2025 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license
([Link]
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

to enhance efficiency and innovation. Consequently, design innovation in PE is required for efficient sustainable power systems and
to meet future demands. Traditional design methodologies are time-consuming [4] and prone to mistakes [5]. Design automation
and integration of artificial intelligence can overcome this challenge [6,7]. Artificial intelligence (AI) is one of the most rapidly
changing fields of research in this 21st century. AI has been applied to several areas of study ranging from automobile [8],
health care [9], business and finance [10], computer vision and robotics, education [11–13] engineering [14], and other industries
as summaries in Fig. 1. Another specific area in which AI is applied is power electronics which includes, designs of power
modules [15], intellectualization of controllers [16,17], implementation of maximum power point tracking controls for renewable
energy systems [18–21], anomaly detection on inverters [22], prediction of the remaining useful life of converters [23,24],
and other electronic devices. Nevertheless, AI applied to PE is characterized by challenges due to the diverse nature of design
characteristics [25]. The challenges reported are complexity [15,26] and stages in the design process, high speed of tuning
requirement in control, and high sensitivity in condition monitoring for ageing detection therefore presenting a challenge in the
implementation of AI in PE. The recent advances in large language models AI can now be implemented to address design stage
complexity. The adaption of this advancement would improve design to increase accuracy and reduce design time. Large language
models (LLMs) have the potential to improve PE design workflow, automate repetitive tasks, and make decisions at a higher level
of intelligence. The use of LLMs in PE design presents both significant opportunities and considerable challenges. While LLMs have
the potential to streamline processes like design computation, modelling, and software development, they are heavily dependent on
large datasets for training and fine-tuning, which may not always be available or of sufficient quality in PE contexts. Furthermore,
the computational intensity of LLMs, coupled with high costs for implementation, maintenance, and ethical considerations poses a
significant challenge for widely adoption. By addressing these challenges, LLMs could unlock transformative advancements in PE,
enabling more efficient, effective, and ethical design processes. These will open new potentials in power electronics and advance
the technology to achieve future demand and sustainability.

Fig. 1. Applications of artificial intelligence.

There has been research on the applications of AI in power electronics (PE) around design optimization, control and maintenance
as reported in [6,27]. Various AI methods such as genetic algorithms [28,29], artificial neural networks [30,31], particle swarm
optimization, and stochastic optimization [32] have been reported. However, AI applications are often applied at a specific stage
of the PE design process stage. Similarly, reviews related to AI in power electronics do not include the recent advancement of AI
such as large language models (LLMs). For example, recent reviews as reported in [6,25,33–36] discuss various methods including
expert systems, neural networks, genetic algorithms, fuzzy logic, metaheuristic methods, and machine learning, but did not include
recent developments in LLM such as generative pretrain transformers (GPT). Therefore, there is a need to provide an overview of
AI applications in power electronics and include the feasibility of LLMs to advance PE research and applications. To our knowledge,
we are the first to present an LLM-driven design for a power electronics system. This paper aims to fill this gap to benefit the
research community by classifying the applications of AI in power electronics at design, maintenance, and control establishing the
feasibility and the impacts that LLMs can bring to PE design. PE design equipped with LLM capabilities reduced design time and
complexity while guaranteeing the accuracy and reliability of PE systems within the design process. This paper adopted narrative
literature review with elements of comparative analysis to assesses the current methodologies, presents the existing gaps, and
proposes future research directions on the feasibility of AI applications to power electronics design. The contributions of this paper

2
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

can be summarized as follows.

(i) Provided a good foundation in the implementation of LLM AI on power electronics design to streamline the design process
for beginners, improve researchers’ and engineers’ productivity.
(ii) Investigated the feasibility of LLMs AI applications to PE design using high-frequency wireless power transfer (HFWPT) as an
example.
(iii) Proposed a LLMs-AI driven framework for PE design.
(iv) Highlighted the potential impact of AI, challenges, and future research direction in AI applications to PE including LLMs.

The paper begins with a review of current AI applications to PE and proceeds with LLMs AI considering the power electronics
design process. The rest of the paper is organized as follows: Section 2 discusses milestones, types, and methods of AI applied to
PE including a research outlook landscape. Sections 3 to 4 present LLMs with integration to PE and the current design approach in
PE proposing an AI-based framework. Section 5 provided the first case study of LLM implementation for the design of PE focusing
on wireless power transfer with validation against reference paper. Section 6 discusses the second case study of high frequency AC
inverter design and validation against results reported in the reference paper. Section 7 gives an outlook on future AI applications
for PE, and Section 8 concludes the paper.

2. Function of artificial intelligence in power electronics

AI techniques applied to power electronics can be categorized into design, control, and maintenance. These applications involve
optimization [37], classification [38], regression, and exploration of data structures [6]. Optimization is about finding solutions that
maximize or minimize objective functions, whereas classification implies placing input data items into distinct classes. Classification
is applied in maintenance ranging from anomaly detection to fault diagnostics as reported in [39,40]. Regression presents a statistical
methodology that predicts continuous values of target variables, based on their relation to the input factors [41,42]. In a control
system application, an intelligent controller can use regression analysis to build mathematical models of dependencies between
the input signals and control variables as reported in [41]. Substantial reviews on AI applications to power electronics have been
reported by [6,14,25,35]. Fig. 2 shows the research trend in AI applications to power electronics from the year 2000 to November
2024 based on Scopus data base. The substantial rise in the number of publications means the area is developing rapidly, gaining
more attention from researchers. The complete workbook search strings, compile data with link to the Scopus search can be found
in the supplementary section.

Fig. 2. Research output trend based on number of publications research.

Fig. 3 shows a landscape of AI applications to power electronics. This map represents an analysis across countries based on the
number of publications, and funding sources behind those publications. 22% of the publications have defined the funding sources,
while 78% are undefined. The first 14 countries representing the defined 22% are analysed. The table lists the countries with their
respective number of publications corresponding to the funding percentage of that country represented in the pie chart. India is at
the top, with 711 publications with 1% of the funding as shown in the pie chart. China has 632 with the largest funding sponsorship
of 42%, followed by the United States had 360 publications with colour corresponding to 17% of the funding. Europe has 15% of
the funding investment, while South Korea has 10%. Other countries, from Germany and Canada, have 3% each, while Japan and
Taiwan have 2% each. United Kingdom, Italy and Qatar have 1% each of funding. This represents a global outlook of the funding
landscape and publications in AI applications to PE among different countries. The applications of AI can be classified into design,
control, and maintenance as discussed in the following subsections.

3
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Fig. 3. Global AI research on PE outlook and funding sponsorship.

2.0.1. Design
The use of AI methods in modelling and optimization is essential in finding the optimum set of parameters for which to obtain the
best performance with a given constraint in PE design. In [43], a heatsink design of a three-phase inverter has been optimized using
a genetic algorithm (GA) combined with a finite-element analysis technique, realizing a striking reduction in size compared with
conventional designs. Research by [44] reported that multi-objective optimization of a solar power-based microgrid system uses
a genetic algorithm to maximize power distribution. Research by [45] employed particle swarm optimization (PSO) to optimize
component values of electronic circuits to enhance performance. Other optimization techniques, like Ant Colony Optimization
(ACO) [46–48], Artificial Bee Colony (ABC), differential evolution [48–50], Non-dominated Sorting Genetic Algorithm-II (NSGA-
II) [51,52], and hybrid AI algorithms [53] have also been applied to different power electronics optimization problems to prove
their capabilities and effectiveness.
In [54], two feed-forward (FF) NN models map was used to automate design parameters and improve the design reliability of
power electronics systems. Their research reported a speed-up in the iterative design process. Data-driven design and automation of
design in power electronics have the potential to reduce iterative design time [55,56]. Another article studies the applications of a
nonlinear autoregressive network with exogenous inputs for thermal modelling of power electronics computational effort to reduce
without the loss of accuracy [57]. In [58] the role of AI-driven optimization techniques, in enhancing the efficiency, reliability, and
robustness of renewable energy-based microgrid systems is investigated. It highlights the potential of AI and parallel computing
to address uncertainties and improve design of renewable microgrid. However, none of the research studies the feasibility of the
applications of AI in the complete design process of power electronics and optimized the design cycles due to the lack of advanced
algorithms to oversee the complexity of the PE design process. With the recent development of advanced LLM algorithms, it is
possible to investigate the feasibility of improving the design process. Table 1 gives a summary of optimization algorithms used in
PE with their strengths and weaknesses.

2.0.2. Control
The applications of AI techniques in the control of power electronics are classified into fuzzy logic control, neural network
control, and reinforcement learning control. Various optimization tasks such as PID tuning, maximum power point tracking (MPPT)
methods, most of which use fuzzy logic, and metaheuristic methods like GA-ACO for obtaining optimal solutions have been
reported. A self-enhanced optimal control of converters has been proposed by [78]. In addition, regression is applied to control as
a nonlinear input–output mapping application of AI techniques like fuzzy logic [79–81], neural networks [82], and reinforcement
learning [83–85].
The applications of fuzzy logic controllers have been made in control tasks of power electronics, such as speed control [79,86]
and MPPT [87]. It expresses a control strategy by using fuzzy sets and rules under expert knowledge. Control of the generator
speed of a variable-speed wind generation system to maximize the extracted wind power output is implemented in a fuzzy logic
controller through the adaptive approach [87]. Research by [88] employed neural network tools for approximating nonlinear
functions showing 15% improvements. In addition, a hybrid approach, such as recurrent fuzzy neural network (RFNN), is used in
dynamic control tasks due to their ability to treat time-series data as reported in [89]. The reinforcement learning control is applied
to dynamic task optimization through learned strategies used to control interactions taken with the environment. Research by [18]
controllers for MPPT in renewable energy systems based on RL adapt to changing conditions without the use of any predefined
dataset, hence greatly improving the system’s autonomy and efficiency. The various applications of AI in control depend on the task
requirement in terms of dynamic performance, robustness, generalization, and convergence speed.

4
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Table 1
Optimizations AI algorithm applied in PE.
Optimization Applications in PE Strengths Weaknesses Ref.
algorithm
GA Optimization of component Global optimization capability, Can be computationally [43,44,59–
parameters, design of magnetic good for complex multi-objective expensive, and may converge to 62]
components problems local optima if not tuned
PSO Parameter optimization for power Simple implementation, fast Prone to premature convergence, [28,63–66]
converters, Magnetic component convergence, effective for can struggle with discrete or
design, Efficiency maximization, continuous optimization problems combinatorial problems
control optimization
ABC Optimization of power electronics Good for dynamic optimization, Can be less stable in convergence, [50,67,68]
circuits, design of high-frequency flexible and robust, fast and performance can degrade
magnetic components, loss convergence behaviour with large problem sizes.
reduction strategies
ACO Path optimization for power flow, Effective for pathfinding and Slower convergence requires [46,69]
design and optimization of routing, good at handling discrete careful parameter tuning
magnetic components, component problems, can adapt to dynamic
placement in circuits changes
Differential Optimization of DC inductors, Simple and flexible, good Premature convergence, limited [70,71]
Evolution geometric parameters & quality convergence, manages local search capability
(DE) factor non-linearities
NSGA-II Multi-objective optimization, filter Pareto front exploration, High computational costs, [25,72]
parameters design, magnetic adaptability to changes, improved premature convergence, limited
design, multi-objective design quality applicability
optimization
ANN-GA Geometric parameters, hybrid Time and cost minimization, Data dependency, limited [30,73–75]
optimization, multi-objective improved accuracy, improved generalization, complexity
optimization, design parameter design quality
synthesis of passive components
ANN-PSO Hybrid optimization, magnetic Cycle time reduction, improved Data dependency, limited training [74,76,77]
design, passive components accuracy. scenarios, implementation
parameter optimization complexity

2.0.3. Maintenance
Maintenance is a set of preventive activities that includes condition monitoring, anomaly detection, fault diagnosis, and remaining
useful life (RUL) prediction according to the IEEE standard [90]. The maintenance activities included offline training, condition
monitoring, and decision-making to improve electronic systems’ reliability and safety [91]. In [92,93], parameter identification, data
preprocessing, and component degradations to extract insightful knowledge for prognostic health management (PHM) applications
have been reported. Research by [94] reported a model-free method based on AI algorithms which set up the relationship between
the available signals and the parameter of interest; on the other hand, model-based methods use the system dynamics and models
to optimize the identification of parameters. Abnormality detection and fault diagnosis are also part of maintenance problems in
guaranteeing the reliability [95] and safety of the power electronic system [96]. AI techniques identify abnormal patterns and
classify faults using historical data or real-time monitoring [97]. Predictive maintenance tries to project the RUL of components
to allow initiative-action taking in maintenance. AI techniques such as supervised and reinforcement learning can also use model
degradation behaviour of components and predict their RUL.

2.1. Types of AI techniques in power electronics

This section discussed types of AI and applications in power electronics. A review paper by [6] reported that the most applied AI
method in PE is ML and reported that AI methods in power electronics can be divided into expert systems, fuzzy logic, metaheuristic
methods, and machine learning. The sections below discuss each method.

2.1.1. Expert system

Expert systems create a catalogue of Boolean logic, which includes specialized information using IF-THEN rules [98], and it
works by following human reasoning. ES is one of the earliest AI methods for applications in PE which uses a database with the
expert’s knowledge embedded in Boolean logic [99]. This technique mimics the human brain’s reasoning, implemented by a set of
IF-THEN rules. According to [6], the use of this technique has reduced and has 0.9% usage compared to other techniques due to
limitations in universality and adaptability relative to more progressive AI methods.

5
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Fig. 4. Classification of machine learning.

2.1.2. Fuzzy logic

The fuzzy logic technique refers to extended Boolean logic designed to process system uncertainty and measurements [100].
This approach is based on fuzzy sets and membership functions, creating a fuzzy space for the input signals, and supplying
nonlinear mapping between starts and stops. The methods of fuzzy logic are fuzzification, rule inference, knowledge base,
and defuzzification [101–103]. This method is efficient in system uncertainties-related tasks [104]. However, it required expert
experience in designing the membership functions and fuzzy rules. Fuzzy logic efficiently models uncertainties and imprecise
measurements using fuzzy sets and membership functions in the solution of otherwise-fussy optimization problems. Research
by [105], employed neuro-fuzzy logic system to design the navigation of a mobile robot which clearly shows the applicability
of the technique to uncertain model.

2.1.3. Metaheuristic methods

Metaheuristic techniques are methods of optimization inspired by biological evolution or natural phenomena that are turned into
optimization algorithms [106]. This method includes [32], ant colony algorithms [47] which mimic how ants effectively search for
food, particle swarm optimization [45], differential evolutions [48], etc. It can be trajectory-based methods, such as Tabu Search
and Simulated Annealing, and population-based methods. It is suitable for finding solutions for complex optimization tasks [107].
Table 1 summarizes the metaheuristic optimization methods. The Metaheuristics technique is suitable for global optimization and
one of the most widely used optimization methods in the field of power electronics.

2.1.4. Machine learning

Machine learning means the automated learning of principles and trends from data, or interactions with an environment by
a computer. It is classified into supervised (SL), unsupervised learning (UL), and reinforcement learning as illustrated in Fig. 4.
Regression uses an algorithm that predicts specific attributes of an object using available data and other attributes [41]. On the
other hand, classification is a common task in fault diagnosis [39]. It involves assigning an object to a specific class with input–
output variables. It is used as a classification algorithm to determine two or more classes from a data set. Tasks in supervised learning
include regression and classification in Fig. 4. This approach involves preparing data for further work with each object having a label
that distinguishes it from other objects as illustrated in Fig. 8. Fault diagnosis in power electronics is mostly realized by employing
supervised learning methods. The trained models that classify the types of faults with adequate accuracy require labelled data
Supervised learning (SL) uses an input–output pair provided during training so that general functional relations can be
established [37]. Working on power electronics in fault diagnosis and remaining useful life (RUL) prediction applies this kind of
approach. Supervised learning could further divide based on connectionism-based schemes like neural networks, and probabilistic
graphic schemes, into memory-based approaches like kernel methods [6]. Fig. 5 provides a summary of SL based on methods
including their variants.
Fig. 6 shows an example of SL and UL of ML applied to electronic topologies to illustrate differences in training approach.
These models can identify the features, which are extracted from the schematics. Unsupervised learning solves the clustering and
data compression problem. [108] reported NN-based model can learn the relation of operating conditions to degradation patterns
and applied to predict the RUL of IGBTs. As illustrated in Fig. 6, there are no predefined output labels, and applied in detecting
certain patterns in the data, reducing the dimensionality of the data before further processing. This type of learning allows the
model to learn independently by deriving patterns and making conclusions based on unlabelled data. Tasks in UL include clustering
and dimensionality reduction. Unsupervised learning-based anomaly detection approaches do not consider any labelled data [22].
Such approaches to anomaly detection may lead to the identification of data patterns, which in turn may indicate faults. In [109],
k-means algorithm and fitting a Gaussian mixture model to detect anomalies in power electronic systems by grouping similar data
points. These techniques are useful in detecting abnormal behaviour, which in most cases may lead to a fault. Clustering is an
unsupervised machine learning method that divides a data set into groups or clusters based on similar features and characteristics.

6
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Fig. 5. Types of supervised learning machine learning methods with their variants.

It helps understand the internal structure of a data set and allows for more efficient work. Ensemble methods are another type of
machine learning algorithm that is used to increase prediction accuracy. These methods are based on the Condorcet Jury theorem,
which states that if a jury has independent opinions and the probability of making the right decision is more than 50%, the
probability of making the right decision will approach 100% as the number of jurors increases. There are various perspectives
on ensemble methods, including majority voting averaging and weighted averaging. The ensemble method is further divided into
begging, boosting and random forest.

Fig. 6. Supervised and unsupervised learning of PE topologies.

Reinforcement learning (RL) employs a learned control strategy formulated by reward maximization through interaction with
the environment. This approach is suitable for dynamic optimization tasks like MPPT in renewable energy systems [18]. A research
by [83] proposed a PL-based controller. Their results show an advantage over complicated nonlinear processes. The learning from
interactions with the environment is a factor taken into consideration by the RL models as they update the prediction in view
based on new data and system states. In [85], applying a developed model using the RL methods for RUL prediction has been
reported. The results show that RL-assisted control is more adaptive to changing operating conditions and degradation patterns,
hence more accurate and timely predictions for maintenance planning. Table 2 shows a summary of AI techniques applied with PE
based on applications, functioning layer, advantages, and disadvantages, while a summary of ML algorithms applied to PE with their
applications, advantages, and limitations, including the prospect of LLMs AI is provided in Table 3. The following section discusses
the design of PE and the prospect of LLMs AI.

7
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Table 2
Summary of AI techniques applied to PE based on functioning layer, advantages, and disadvantages.
AI technique Applications Functioning layer/Method Advantages Disadvantages
Expert Design Regression Speeding up the design Limitation of understanding
systems maintenance process aids GA in component decision basis, dependency on
selection expert knowledge
Fuzzy logic Design control Classification, Regression, and Suitable for system The complexity of the design
maintenance can be uncertainties applications, process, challenging
Takagi–Sugeno–Kang-type or utilizes expert knowledge, knowledge acquisition
Mamdani method. multi-criteria optimization
Metaheuristic Design control Optimization tasks can be Efficient and scalable, less Intensive computational
methods maintenance trajectory-based or expert experience needed burden, no guarantee for
population-based methods. global optimum
Machine Design control Classification, Regression, data No training dataset is High computational cost,
learning maintenance structure exploration required, learns from system requires significant resources,
Optimization, and can be interaction, reduces the limited knowledge base,
supervised unsupervised or computational burden, limited applications, and
reinforced learning. improves analytics accuracy, knowledge base, requires
and establishes mapping with large, labelled datasets,
functional relationships generalization can be
between inputs and outputs. challenging, limited
knowledge.

Table 3
ML applied to PE and the prospect of LLMs.
ML algorithm Applications Advantages Limitations
RL Dynamic optimization No training dataset is required, learns High computational cost, requires significant
from system interactions resources, limited knowledge base
UL Data preprocessing, Learn Reduces computational burden, improves Limited applications and knowledge base,
independently analytics accuracy primarily used as preprocessing
SL Predictive modelling, Establishes mapping and functional Requires large, labelled datasets, generalization
Interpolation relationships between inputs and outputs can be challenging, limited knowledge.
LLMs Knowledge-based driven, Vast PE knowledge base, natural May attract cost, data privacy issues, potential
learn independently language understanding, contextual bias, security, hallucination
information

3. Power electronics design and the prospect of large language model

The demand for more energy-efficient electrical systems is increasing. However, the design process of electrical systems remains
the same. In [4], it is reported that more than 12 months are needed for efficient converter and inverter design. Table 4 provides
a more comprehensive time for each design process stage based on Fig. 7. This time frame would differ based on the designer’s
experience, project objectives and complexity. Power electronics research is now focusing more on sustainability, innovative designs
and electronics as a key goal. Technological advances need to include power electronics design process optimizations to improve
energy efficiency and sustainability [110]. The lack of an AI tool that integrates system design phases is a major issue. Most design
tools and processes use manual calculations and human expertise. Research by [54] adapted artificial neural network-based to
automate power electronic design. The process optimizes design using reliability metrics and links them to design parameters.
Another research by [111] studied machine learning models for electronic circuit design to reduce human error and generate layouts
automatically. Another study by [112] suggested an AI circuit design enhancer to speed up design and optimization. They employed
a genetic algorithm, NSGA-II algorithm, and artificial neural network optimized circuit design. In [3], design of framework for IoT
has been investigated using hybrid deep learning. In addition, [3] proposed a design framework for IoT enable for autonomous
vehicle application. Their result demonstrated a promising result in improving accuracy and speed up design. Another research
by [113] investigated combined AI-driven stage design with traditional Chinese opera stage design and reported improved design
performance. A study by [114] reported current approach to power electronics design lacks automation which limits production,
applications, and penetration. Other studies include [54,115–123] also follow the same pattern. However, their approach ignores the
entire power of electronics integrating every step of the design process such as conception, specification, mathematical modelling,
simulation, prototyping, testing, certification, and mass production. This limits the potential of how AI can improve power electronics
design. These will help in innovating and iterating power electronics more rapidly and productively. Therefore, researchers and
designers are working on novel strategies to meet the demands and advancement of the field.
Recent advancements in fine-tuning techniques for LLMs in power electronics have demonstrated their potential for optimization,
security, and design automation in PE. Research by [124] proposed an LLM-based agent using the Pandapower library to optimize
power system states, with a secondary agent verifying compliance with GOST 32144-2013 standards, resulting in improved

8
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Table 4
Power electronic design process stages and estimated time.
Electronic design process stage Estimated time (weeks)
Conceptual design 0–4
Design Specification 1–6
Computation/Mathematical modelling 1–8
Simulation 1–12
Design parameter optimization 0–8
Prototyping (PCB layout, Schematic design, Component selection) 4–12
Code/Software development 1–8
Hardware testing and debugging 2–12
Set up, integration and system testing 4–8
Certification and compliance testing 4–8
Final prototyping and validation 6–10
Mass production Ongoing (months)

optimization outcomes with GPT-4-Turbo, GPT-3.5, and Claude-2.1. A study by [125] investigated security threats in cyber–
physical power systems (CPPS), emphasizing domain-specific fine-tuning and rigorous validation to mitigate risks. Another research
by [126] introduced an LLM-based multi-agent framework leveraging zero-shot prompting for automated PE controller design,
reducing design cycle times. In [127], a human-in-the-loop (HITL) framework combining interactive learning and fine-tuning for
routine and innovative engineering tasks, enhancing success rates and adaptability for complex problems is proposed. A study
by [128] developed PE-GPT, a multimodal LLM integrating retrieval-augmented generation (RAG) and metaheuristic algorithms
for modulation design and circuit optimization, achieving a 35.6% accuracy improvement. In [129], a LP-COMDA is proposed using
physics-informed LLM integrating surrogate models and optimization algorithms, achieving a 63.2% error reduction in low-data
scenarios through physics-informed fine-tuning. In this research, we propose a hybrid approach integrating prompt engineering,
transfer learning, and RAG to enhance the contextual relevance and domain-specific accuracy for power electronics design. Our
method combines theoretical constraints with adaptive, data-driven learning for multi-objective design optimization, bridging the
gap between theoretical modelling and practical application in different power electronics scenarios. Table 5 provide a summary of
recent works on LLM fine-tune techniques for PE.

Table 5
Recent LLM fine-tuning techniques in power electronics.
Study Technique LLM models used Application focus Key outcome
[124] Prompt engineering GPT-4-Turbo, Power system Improved voltage deviation
GPT-3.5, Claude-2.1 optimization results
[125] Transfer learning Fine-tuned LLMs CPPS security Enhanced anomaly detection
threat detection accuracy
[126] Zero-shot prompting LLM Agents PE control design Reduced design cycle time
[127] Prompt engineering, ChatGPT, Claude, Routine & novel Improved task-solving flexibility
Iterative learning Bard PE tasks
[128] RAG, Metaheuristics PE-GPT Modulation & 35.6% accuracy improvement
circuit design
[129] Physics-informed LP-COMDA Modulation design 63.2% error reduction, 33x
fine-tuning automation faster design
This work Prompt engineering, Open-AI Custom LLM Multi-objective PE Enhanced adaptability and
Transfer learning, RAG design practical performance

In [130], AI design automation based on a graph neural network model to reduce circuit design time is proposed. Their results
show equal performance to manually optimized design system which save designers weeks of design time. However, this approach
did not consider the complete electronic design process. In addition, it is difficult to generalize the overall design time because each
power electronics design task depends on the type, aim, specification, and complexity of the system. Another research by [131] stated
that power electronics system productivity, reliability, and flexibility improve when the design process methodology is improved.
They proposed a design tool called ‘Power Electronics Design 4.0 (PED 4.0)’ which includes five power electronics design stages as
specifications, topologies selection, simulations, prototype, and production. However, the proposed method did not include design
computation, modelling, code or software development, certification, or compliance. Fig. 7 shows an extended version including
design computations, modelling, coding, certification, and compliance.
The flowchart in Fig. 7 shows a systematic approach to power electronics design from concept to mass production. Starting with
the objective and design concept (O&CD), the process develops the initial specification and requirements (SP/R). Computational, or
mathematical modelling (C/MM) creates a design specification that meets the goal. A computation or mathematical modelling that
is successful, followed by a simulation and optimization (S&O). The initial design requirement is compared to simulation results.
Successful simulation-driven design iteration leads to prototype creation, including component selection (CS), schematic creation
(SC), and printed circuit board (PCB) creation. Software development, hardware testing, debugging, and system test setup follow the

9
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Fig. 7. PE design framework.

prototype. After a successful prototype test, regulatory standards established certification and compliance (CC) and final prototype
validation (PV). CC and PV success indicate mass production readiness.
It is necessary to assess the design process of power electronic systems and propose novel approaches integrating AI to facilitate
an effective shift towards sustainable power systems design and development. The escalating demand for more energy-efficient
electrical systems affects design time. This is where LLMs such as GPT will be able to make the process easier. GPT models can
generate coherent, contextually appropriate text and conduct a wide variety of natural language processing (NLP) tasks due to the
advanced transformer architecture proposed by Google and vast training data. Their scalability and adaptability make them quite
suitable for applications both in research and the real world. This allows the update of design support systems for researchers
and engineers with LLMs, improving productivity, reducing errors, and saving large costs. The following section discussed LLMs. A
summary of existing AI design assistant tools including a High Frequency Power Electronics (HFPE) LLM proposed in this research
are provided in Table 6, indicating strength and capabilities at each design stage of PE design.

4. Large language model AI

Large language models have demonstrated capabilities recently in natural language processing (NLP) due to their ability to
understand and generate human-like text [142]. These models are trained over vast amounts of data to either process or generate text
with a great degree of accuracy due to the attention mechanism in the transformer architecture [143]. This transformer architecture
(TA) becomes more intelligent to be able to learn, generate, translate, summarize, and answer questions related to both text and
numerical data. This makes an interesting area of research for design performance analysis to investigate the feasibility of LLM in
the fast-tracking PE design process. In addition, LLMs can be developed for design automation and optimization of distinct phases
involved in power electronics design. Starting from interpreting the design requirements to documentation generation, parameter
optimization, and performance simulation, LLMs increase the productivity of design and reduce the risk of errors. With an increase
in information, computing power, and AI algorithms, LLMs have the potential to revolutionize the design of power electronics
processes. The TA was proposed by Google in 2017 [143], which uses traditional recurrent, and convolutional neural networks with
self-attention mechanisms as shown in Fig S4 in the supplementary information document. The various kinds of position recordings
include using decoder or encoder or encoder–decoder models. The attention calculates, at every word in a sequence, the relative
value and relevance of words concerning other words in that sequence based on Eq. (1) as reported in [143]. Where Q represents the
query matrix, K is the key matrix, and V is the value matrix, 𝐾 𝑇 transpose of the matrix K, and 𝑑𝑘 dimension of the key vectors. The

10
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Table 6
Collection of HFPE design assistant tools applied in PE including the prospect of LLMs.
Ref Tool AI technique Applications O&CD SP/R C/MM S&O CS SC PCB CC MP
[132] CADY AI Natural language Automatic ↓ ↓ ↓ ↓ ↓ ↓ ↑ ↑ ↓
with computer electrical
vision schematic
inspection
[133] MagNet Machine learning HF simulation, ↓ ↓ ↑ ↑ ↓ ↓ ↓ ↑ ↓
Princeton Optimization
[134] [Link] Machine learning PCB design ↓ ↑ ↓ ↓ ↑ ↓ ↑ ↑ ↓
assistant
[135] M2Spice Machine learning HF, SPICE Netlist ↓ ↓ ↓ ↑ ↓ ↓ ↑ ↓ ↓
Princeton generation
[136] CoupL Non-AI advanced Design ↓ ↓ ↑ ↑ ↑ ↓ ↓ ↓ ↓
Princeton modelling optimization
[137] [Link] Machine learning Magnetics ↓ ↓ ↓ ↑ ↑ ↑ ↓ ↑ ↓
optimization,
Simulation
[138,139] [Link] Machine learning HF magnetic ↓ ↓ ↑ ↑ ↓ ↓ ↓ ↑ ↓
with advanced components
OpenAI LLMs design, Core loss
analysis
[140] AI-Mag ANN, Finite HF inductor ↓ ↓ ↑ ↑ ↓ ↓ ↓ ↑ ↓
element method modelling
[141] Open Machine learning Magnetics design, ↓ ↓ ↑ ↑ ↓ ↓ ↑ ↑ ↓
magnetics Simulation
This High Machine learning Power electronics ↑ ↑ ↑ ↓ ↑ ↓ ↓ ↑ ↓
paper Frequency with advanced design assistant
Power OpenAI LLMs
Electronics
(HFPE)

Applicability & Strength: ↑ Full Capability ↓ Partial Capability.

Fig. 8. Trends of LLMs pre-trained AI models timeline.

functions of each part of the transformer architecture are summarized in Table S1 in the supplementary information document. LLMs
have developed more capabilities in response to the growing need for machines to perform complex language tasks, as language
models are essential to human communication and self-expression. These state-of-the-art artificial intelligence systems are capable
of processing, generating, and generalizing to multiple tasks with coherent communication. From statistical to neural language
modelling, and from pre-trained language models (PLMs) to LLMs, NLP has advanced historically. A substantial number of LLMs
are being developed by researchers that can be applied to various tasks. Figs. 8 and 9 show an increasing in the number of released
LLMs and LLMs proposed over the years collected from Databricks. These timelines show the developments and patterns of natural
language processing in different fields over the years.

( )
𝑄𝐾 𝑇
[𝐻]𝐴 (𝑄, 𝐾, 𝑉 ) = 𝑆𝑜𝑓 𝑡𝑚𝑎𝑥 √ (1)
𝑑𝑘

11
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Fig. 9. Trends of LLMs instructions tuned AI models timeline.

4.1. LLMs framework-based OpenAI for power electronics design

LLMs represent a new development in artificial intelligence [144]. Their evolution can be related to earlier neural network
architectures, which paved the way for more advanced models like transformers. Fast-paced development in artificial intelligence
is impacting different industrial activities, including power electronics. Generative pre-trained transformers (GPT) developed by
OpenAI can enhance design strategy to ensure reliability and improve productivity reported by [144,145]. The development trend
of the instructions-tuned foundation model of OpenAI is shown in Fig S5 in the supplementary information document. Fig. 10
presents the design framework integrated with LLN-AI Design Assistant. This reduced the design time as opposed to the traditional
framework of Fig. 7. In this process, the knowledge regarding PE is provided to the AI, and hence researchers and engineers can
efficiently respond to and make design changes, thus reducing the iteration process by a tremendous amount. It automates activities
connected to Objective and Conceptual Design (O&CD), Specification and Requirement (SP/R), Computation and Mathematical
Modelling (C/MM), Component Selection (CS), Coding plus Schematic Design (C/SD), and Certification and Compliance (CC) to
reduce human effort and the possibility of errors. MATLAB, being used for simulation and other AI implementation methods, makes
the workflow smooth from conceptual design to mass production without waits due to manual transitions. Real-time feedback from
the AI helps in shortening the decision cycle and brings down revision and validation times. Moreover, automating component
selection, schematic design, and PCB development accelerates the prototyping phase itself. This approach avoids the tedium of the
process, and much consumption of time as compared to the traditional framework, which relies exclusively on manual intervention.
Thus, it provides ways in which the overall design time can be reduced under the proposed sequential design processing framework.
This proposed sequential LLM design processing framework has the potential to shorten the total time spent on the overall design.
For a given defined objective and design concept (O&CD) with initial specification and requirements (SP/R), the mathematical
expression of the AI framework can be represented by the following (2) to (6).
{
𝑅𝑒𝑝𝑒𝑎𝑡, 𝑖𝑓 𝐴𝐼 𝑅 ≠ 𝑆𝑃 ∕𝑅
𝐴𝐼 𝑅 = 𝑀𝑜𝑑𝑒𝑙 (𝐾) → (2)
𝑃 𝑟𝑜𝑐𝑒𝑒𝑑, 𝑖𝑓 𝐴𝐼 𝑅 = 𝑆𝑃 ∕𝑅
{
𝑅𝑒𝑣𝑖𝑠𝑒, 𝑖𝑓 𝐶∕𝑀𝑀 ≠ 𝑆𝑃 ∕𝑅
𝐶∕𝑀𝑀 → (3)
𝑃 𝑟𝑜𝑐𝑒𝑒𝑑 𝑡𝑜 𝑆∕𝑂 , 𝑖𝑓 𝐶∕𝑀𝑀 = 𝑆𝑃 ∕𝑅

𝑆𝑡𝑎𝑟𝑡𝑃 𝑟𝑜𝑚𝑝𝑡 → (𝑆) → 𝑂𝑝𝑡𝑖𝑚𝑖𝑧𝑒(𝑆) → (𝑆 ≈ 𝑆𝑃 ∕𝑅) → 𝑃 → (𝐶𝑆 ≈ 𝑆) → (𝑆𝐷 + 𝐻𝑇 𝐷 + 𝑆𝑈 ∕𝑆𝑇 = 𝑆𝑃 ∕𝑅) (4)

𝐴𝐼 𝑟𝑒−𝑝𝑟𝑜𝑚𝑝𝑡 → 𝐿𝐿𝑀 − 𝐴𝐼 → 𝑂𝑝𝑒𝑛 𝐴𝐼 − 𝐵𝑎𝑠𝑒𝑑 𝑀𝑜𝑑𝑒𝑙 → 𝐴𝐼 𝑅 → 𝐼𝑓 𝐴𝐼 𝑅 = 𝑆𝑃 ∕𝑅 (5)

𝐸𝑛𝑑 𝑃 𝑟𝑜𝑚𝑝𝑡 → 𝑉 & 𝑀, 𝑖𝑓 𝐻𝑇 𝐷 + 𝑆𝑈 ∕𝑆𝑇 = 𝐶𝐶 (6)

where 𝑆 and 𝑃 are simulation and prototyping, respectively. 𝑆𝐶 schematic design, 𝑃 𝐶𝐵 printed circuit board development,
𝐶𝑆 combined schematic design, 𝐻𝑇 𝐷 hardware test & debugging, 𝑆𝑈 ∕𝑆𝑇 set up integration & system test, 𝑆𝑃 ∕𝑅 specification
and requirements to be met, 𝐶𝐶 certification standardization, 𝐴𝐼 𝑅 AI response, 𝐾 embedded knowledge, 𝑀 mass production, V
validation. This can further be expressed as (7) to (9), respectively.
{
𝑅𝑒𝑝𝑒𝑎𝑡, 𝑖𝑓 𝑆 ≠ 𝑆𝑃 ∕𝑅
𝑆 → 𝑂𝑝𝑡𝑖𝑚𝑖𝑧𝑒(𝑆) → (7)
𝑃, 𝑖𝑓 𝑆 ≈ 𝑆𝑃 ∕𝑅
{
𝑅𝑒𝑝𝑒𝑎𝑡, 𝑖𝑓 𝐶𝑆 ≠ 𝑆
𝑃 = 𝑆𝐶 + 𝑃 𝐶𝐵 → (8)
𝑃 𝑟𝑜𝑐𝑒𝑒𝑑𝑡𝑜𝑇 𝑒𝑠𝑡𝑖𝑛𝑔, 𝑖𝑓 𝐶𝑆 ≈ 𝑆

12
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Fig. 10. PE design automation framework based on LLM instruction tune AI model.

{
𝑅𝑒𝑝𝑒𝑎𝑡, 𝑖𝑓 𝑆𝐷 + 𝐻𝑇 𝐷 + 𝑆𝑈 ∕𝑆𝑇 ≠ 𝑆𝑃
𝐻𝑇 𝐷 + 𝑆𝑈 ∕𝑆𝑇 → (9)
𝑉 ∕𝑀, 𝑖𝑓 𝑆𝐷, 𝐻𝑇 𝐷, 𝑆𝑈 ∕𝑆𝑇 = 𝐶𝐶
Total design time T for a non-AI framework and AI framework can be expressed in (10) and (11), respectively.

𝑇𝑁𝑜𝑛−𝐴𝐼 = 𝑇𝑆𝑃 ∕𝑅 + 𝑇𝐶∕𝑀𝑀 + 𝑇𝑆∕𝑂 + 𝑇𝑃 + 𝑇𝐶𝑆𝐷 + 𝑇𝐻𝑇 𝐷 + 𝑇𝑆𝑈 ∕𝑆𝑇 + 𝑇𝑇 𝐶𝐶 + 𝑇𝑉 (10)

𝑇𝐴𝐼 = 𝑇𝑆𝑃 ∕𝑅,𝐴𝐼 + 𝑇𝐶∕𝑀𝑀,𝐴𝐼 + 𝑇𝑆∕𝑂,𝐴𝐼 + 𝑇𝑃 ,𝐴𝐼 + 𝑇𝐶𝑆𝐷,𝐴𝐼 + 𝑇𝐻𝑇 𝐷,𝐴𝐼 + 𝑇𝑆𝑈 ∕𝑆𝑇 ,𝐴𝐼 + 𝑇𝑇 𝐶𝐶,𝐴𝐼 + 𝑇𝑉 ,𝐴𝐼 (11)

The total design time savings can be expressed as (12).

𝛥𝑇 = 𝑇𝑁𝑜𝑛−𝐴𝐼 − 𝑇𝐴𝐼 (12)

Substituting Eqs. (10) and (11) into (12) would give (13), and introducing a saving factor 𝛼 𝑖 , the total AI design time reduction
can be expressed as (14), where each T𝑖 is the time for stage i in the non-AI framework.

𝛥𝑇 = (𝑇𝑆𝑃 ∕𝑅 − 𝑇𝑆𝑃 ∕𝑅,𝐴𝐼 ) + (𝑇𝐶∕𝑀𝑀 − 𝑇𝐶∕𝑀𝑀,𝐴𝐼 ) + ⋯ + (𝑇𝑉 − 𝑇𝑉 ,𝐴𝐼 ) (13)

∑
𝛥𝑇 = 𝛼 𝑖 ⋅ 𝑇𝑖 (14)

Fig. 11 shows a framework of how to create a customized model from an OpenAI base model; this ranges from data sources to the
end user. It involves a pathway of different stages: data collection, followed by base model selection, then embedding and fine-tuning
following certain rules and orders, and testing with refinements before deployment for practical use. The model’s knowledge base is
primarily based on reputable published sources from IEEE, Scopus, and Texas Instruments. In this research, 2000 pages of data were
used to fine-tune the model for high frequency AC inverter and wireless power application design based on the design automation
framework present in Fig. 10.
Fig. 12 shows the procedure used to create LLM-AI assistant called ‘High Frequency Power Electronic (HFPE)’ to design HFWPT
system. The based model is design to accept data into its knowledge section, process the uploaded data through embedding such as
retrieval-augmented generation and fine-tuning, until reaching the objective instructions and rules. If the objective is not reached,
there is a need to go back to refine the instructions, improve the data quality in the knowledge section. The model keeps learning
from the uploaded data based on its based training and provided instructions. It answers user query, gets further training as it
operates with the based generative large language model to keep advancing. The model can make changes over time by taking
feedback from users and performance reviews. There is a constant need for updated and further finetuning the models to improve
for best performance. The application programming interface (API) allows developers to interact with and utilize the AI model
programmatically. Table 7 provides a summary of the AI algorithm used in power electronics-based design applications stages

13
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Fig. 11. LLM framework development for fine-tuning and creating customized GPT for power electronics design.

Fig. 12. High Frequency Power Electronic (HFPE) customized LLM-AI creating procedure for power electronics design.

indicating strength at each stage including LLM. The next section discussed a case study with the practical demonstration of the
LLM AI assistant in the design and analysis of HFWPT systems. Simulations were run, and literature analysis was used to validate
the proposed LLM-driven design.

14
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Table 7
AI algorithm applications based on stages of PE design process.
Ref AI algorithm O&CD SP/R C/MM S&O CS SC PCB CC MP
[98] Expert systems ↓ ↓ ↓ ↑ ↓ ↓ ↓ ↓ ↓
[87,100,146,147] Fuzzy Logic ↓ ↓ ↓ ↑ ↓ ↓ ↓ ↓ ↓
[29,32,60] GA (Genetic ↓ ↓ ↓ ↑ ↓ ↓ ↓ ↓ ↓
Algorithm)
[51,52,72] NSGA-II ↓ ↓ ↓ ↑ ↓ ↓ ↓ ↓ ↓
[45,76,148] PSO ↓ ↓ ↓ ↑ ↓ ↓ ↓ ↓ ↓
[50,68] ABC ↓ ↓ ↓ ↑ ↓ ↓ ↓ ↓ ↓
[47] ACO ↓ ↓ ↓ ↑ ↓ ↓ ↓ ↓ ↓
[48,70] DE ↓ ↓ ↓ ↑ ↓ ↓ ↓ ↓ ↓
[30,53,149,150] ANN ↓ ↓ ↓ ↑ ↓ ↓ ↓ ↓ ↓
[73,75,151] ANN-GA ↓ ↓ ↑ ↑ ↓ ↓ ↓ ↓ ↓
[76,152] ANN-PSO ↓ ↓ ↑ ↑ ↓ ↓ ↓ ↓ ↓
[83–85] RL ↓ ↓ ↑ ↑ ↓ ↓ ↓ ↓ ↓
[22,153] UL ↓ ↓ ↑ ↑ ↓ ↓ ↓ ↓ ↓
[37,108] SL ↓ ↓ ↑ ↑ ↓ ↓ ↓ ↓ ↓
[107,154] Metaheuristic ↓ ↓ ↓ ↑ ↓ ↓ ↓ ↓ ↓
Methods
Proposed in this LLMs based on ↑ ↑ ↑ ↓ ↑ ↓ ↓ ↑ ↓
work GenAI

Applicability & Strength: ↑ Full Capability ↓ Partial Capability.

5. LLM-AI-driven design of HFWPT (Case study 1)

Wireless power transfer (WPT) has been studied since the 1880s by Nikola Tesla [155,156]. Research and development in
semiconductor materials like wide band gap devices, has made the development of high-frequency wireless power electronics
working at high power density, low size [157], and enhanced performance possible [158]. These materials operate at lower losses,
higher voltages and temperatures which enables system miniaturization and development of compact systems. Recent progress in
Silicon carbide (SiC) and gallium nitride (GaN) based switches has made wireless power systems operate at even higher frequencies
by designing an effective AC power inverter. Fig. 13 shows a simplified circuit diagram of the HFWPT system.

Fig. 13. Basic simplified circuit diagram of series–series HFWPT system.

High-frequency wireless power transfer (HFWPT) technology particularly in the tens of kilohertz to megahertz range, has
experienced major advancements. In a recent study performed by [159], a wireless power transfer system operating at a frequency of
6.78 MHz was presented. The system exhibited an efficiency of 95% for a power level greater than 1 kW. Another research by [160]
based on an inductive power transfer system having a nominal power of 2 kW and a frequency of 3.39 MHz demonstrated a peak
efficiency of 95% when delivering an output of 3 kW. These advances prove the possibility of high efficiency and high power in
systems that work at high frequencies [161]. High-frequency wireless power transmission yields higher power density and is more
compact, making it suitable for high-power applications with minimal coupling requirements [162]. High-frequency wireless power
transfer technology has applications in various fields such as electric vehicle charging [163], smart grid systems [164], wireless
charging of portable biomedical electronic devices [165], and high-power applications with low coupling requirements. However, it
faces challenges and limitations. A comprehensive review by [166] discussed challenges including design, prototypes, international
standards, and commercialization of both dynamic and static WPT. Research by [167–169] reported that series-series (SS) topology
is suitable for high-frequency magnetic resonance wireless power transfer. Eqs. (15) to (21) are used in modelling and designing

15
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

wireless power transmission systems, which are resonant at high frequencies based on SS topology. They provide the calculation of
the resonant frequency (𝑓𝑜 ) of an LC circuit, governed by the inductance and capacitance of the LC components as shown (15). A
coil inductance depends on the number of turns (N), coil circular radius( 𝑅), the permeability of space (𝜇𝑜 ), the relative permeability
(𝜇𝑟 ), and the G.M.R of the wire (𝑟) as shown in Eqs. (16).
1
𝑓𝑜 = √ (15)
2𝜋 𝐿𝑜 𝐶𝑜
) (
8𝑅
𝐿𝑜 = 𝑁𝑜2 𝑅𝜇𝑜 𝜇𝑟 [𝑙𝑛 − 2] (16)
𝑟
where 𝑓𝑜 represent primary (𝑓1 ) and secondary (𝑓2 ) side frequency, while (𝐿𝑜 ) represent primary side inductance (𝐿1 ) and secondary
side inductance (𝐿2 ) as represented in Fig. 14. The primary capacitance (𝐶𝑝 ) can be calculated considering the secondary inductance,
secondary capacitance (𝐶𝑠 ), and primary inductance as shown in (17). The coupling coefficient (𝑘) is a constant ranging from 0 ≤
k ≤ 0.5 for a loss coupling, and complete mathematical modelling of HFWPT is discussed in [167]. These provide an estimate of
the degree of magnetic coupling existing between two inductors in which (M) is the mutual inductance. The quality factor is the
measure of the efficiency of the resonant circuit and is calculated as the ratio of inductive reactance to resistance on the primary
and secondary sides where 𝜔 = 2𝜋f. 𝑄𝑝 and 𝑄𝑠 can be calculated for the primary and secondary sides using Eq. (19). Maximum
efficiency can be calculated using Eq. (20) considering the coupling coefficient and quality factors of the primary and secondary
sides, or currents through the primary and secondary sides with load resistance, primary and secondary resistances as shown in Eq.
(21). These equations provide the basis for designing wireless power transfer.
𝐿 2 𝐶𝑠
𝐶𝑝 = (17)
𝐿1
𝑀
𝑘= √ (18)
𝐿1 𝐿2
⎧ 𝜔𝐿1
⎪𝑅 , 𝐹 𝑜𝑟 𝑃 𝑟𝑖𝑚𝑎𝑟𝑦 𝑆𝑖𝑑𝑒
𝑄 = ⎨ 𝜔𝐿1 (19)
⎪ 𝑅2, 𝐹 𝑜𝑟 𝑆𝑒𝑐𝑜𝑛𝑑𝑎𝑟𝑦 𝑆𝑖𝑑𝑒
⎩ 2
𝑘2 𝑄 1 𝑄 2
𝜂= √ (20)
(1 + 1 + 𝑘2 𝑄1 𝑄2 )2
𝐼22 𝑅𝐿
𝜂= (21)
𝐼12 𝑅1 + 𝐼22 𝑅2 + 𝐼22 𝑅𝐿

5.1. WPT simulation

The HFWPT system is designed using LLM-AI-driven design assistance as proposed in Section 3. After the creation, finetuning
and successful deployment of the LLM-AI. It is prompted to design an HFWPT using the design specifications as reported in [167].
The input design specifications are shown in Table 8. The proposed design was simulated in MATLAB to analyse and compare with
the results reported in [167]. The LLM proposed design was simulated and the results output was used to design optimized the
filter using an iterative prompt approach between the LLM and simulation output. The HFPE often provide potential values of the
design using based on design equations provided in the knowledge section and instructions during creation and fine tuning. The
parameter values can be further refined or optimized to meet the design requirement. The MATLAB simulation block presented in
Fig. S6 in the supplementary information documents, which includes a gate drive, high-frequency inverter, high-frequency wireless
power system, rectification, and a resistive load. A ‘Chat sheet’ and video of HFPE-AI driven design showing all the prompts and
responses is provided in the supplementary sections of this paper.
Table 8
Input design specification.
Parameter [167] HFPE LLMs-AI (This work)
Frequency (Hz) 41 500 41 500
Circular coil radius (cm) 3.8 3.8
Primary & Secondary turns 25 25
G.M.R of the wire (mm) 3.5 3.5
Primary & Secondary resistance (Ω) 0.0714 0.0714
Resistive load (Ω) 150 150
Duty ratio 0.75 0.75

5.2. WPT LLM-driven design results

The results of the LLM-AI-driven design of the HFWPT and validation are shownin Figs. 14 to 15 and Table 9, respectively. The
sinusoidal nature of both the voltage and current waveforms shows that the system is conducting the necessary function within the
specified frequency range. This type of waveform analysis is crucial for designing and optimizing WPT systems to ensure efficient

16
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Fig. 14. HFPE-AI proposed HFWPT design at 1 cm spacing, 41.5 kHz, (a) and (b) primary currents, and voltage, (c) and (d) secondary voltage and currents.

power transfer and minimize losses. The graphs presented in Fig. 14 illustrate the waveforms of primary and secondary currents
and voltages for a wireless power transfer system under different coupling coefficient (k) values. The primary current and voltage
waveform at k = 0.2, 0.1, 0.075, and 0.05 significantly changes in amplitude corresponding to changes in the coupling coefficient,
with the highest amplitude observed for k = 0.2 and the lowest for k = 0.05 (Fig. 14a and b). A similar trend is on the secondary
side (Fig. 14c and d), where the amplitudes decrease as (k) decreases. These results provide a clear visual representation of how
the coupling coefficient impacts the performance of the wireless power transfer system, like the result reported in [3].

Fig. 15. HFPE-AI proposed HFWPT design system output DC (a) voltage, (b) current.
The overall DC output in Fig. 15, shows a steady state of current at approximately 1.4 A and voltage at approximately 200 V at k
= 0.2. A decrease in current and voltage indicated a decrease in power output as the value of k decreased from 0.2 to 0.05 indicating
a decrease in efficiency. This agrees with the results reported in other literature. The HFPE-AI was also used to optimize the output
fillers and provide a potential supplier of suitable components to proceed to the prototype. The design parameters computation
was completed in seconds. The complete design including filter parameter optimization was completed in minutes. The currents,
proposed by the LLM-AI driven design assistant may be due to potential issues such as incorrect quality factor assumption. The
HFPE-AI design assistance assumes a k = 0.2 to calculate the mutual inductance of the WPT system. Future work should focus on
investigating the trade-off and dependability of LLM AI WPT systems-driven design and adopt other advanced LLM AI techniques.
Further insight into the overall output DC performance shows that the AI-designed system works correctly under the expected

17
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

conditions, thus making the 250 V potentially applicable to electric vehicles (EV). In addition, the ability of the HFPE-AI design
assistance to assist with component selection based on specific design applications has demonstrated the potential of reducing design
time at the stage of prototyping. Appropriate component ratings selection reduces losses and improves system robustness. The overall
design is achieved by simply prompting an AI that is finetuned for a specific system design to show the feasibility of automation
and reducing design time of the O&CD, SP/R, C/MM, CS, C/SD, and CC as proposed in Fig. 7. Table 9 compares the design output
of the HFPE AI proposed design validation against a reference design reported in [167]. Both designs operate at 41.5 kHz, with
a circular coil radius of 3.8 cm, 25 primary and secondary turns each, and wire G.M.R of 3.5 mm. The HFPE AI was prompted
to design an HFWPT system based on the specifications reported in [167]. The reference paper uses 73.43 μH for primary and
secondary self-inductances, slightly differing from the LLMs-AI design calculated value of 73.6 μH with a standard deviation (SD)
of 0.0085. The LLMs-AI design also calculated the mutual inductance of 14.72 μH, 7.36 μH, 5.52 μH, and 3.6 μH for 𝑘1 , 𝑘2 , 𝑘3 , and
𝑘4 , respectively, which is not reported in [167]. The primary and secondary compensation was obtained to be 0.2 μF, the same as
the referenced paper (SD = 0). To ensure proper spacing, the efficiency decreases as k decreases, with an SD of 42.9, 20.1, 0.3, 18.3
compared to the simulation in [167] and 44.5, 21.8, 1.9, 16.6 compared to the experiment in [167]. The closest LLM-AI-driven
design efficiency is 36.4% with 𝑘3 = 0.075, SD of 0.3 and 1.9 compared to 36.0% and 33.6% in [167], respectively. These results
show the potential of LLMs AI to optimize the design process of power electronics systems. These improvements could increase
engineers’ and researchers’ productivity, enhance performance and efficiency, and maintain consistency in design parameters.
Table 9
Output comparison of HFPE (LLMs-AI) design with [167].
Parameter [167] LLM-AI (This work) Standard Deviation
Sim Exp 𝑘1 𝑘2 𝑘3 𝑘4 [167] Simulation [167] Experiment
(0.2) (0.1) (0.075) (0.05)
𝑘1 𝑘2 𝑘3 𝑘4 𝑘1 𝑘2 𝑘3 𝑘4
Primary & 73.43 73.43 73.6 73.6 73.6 73.6 0.085 0.085
Secondary
Self-inductance
(μH)
Mutual – – 14.72 7.36 5.52 3.68 – – – – – – – –
Inductance (μH)
Primary & 0.2 0.2 0.2 0.2 0.2 0.2 0 0 0 0 0 0 0 0
Secondary series
capacitance (μF)
Efficiency (%) 36 33.6 96.6 64.4 36.4 10.1 42.9 20.1 0.3 18.3 44.5 21.8 1.9 16.6

6. LLM-AI-driven design of HFAC inverter (Case study 2)

High-frequency alternating current (HFAC) systems first proposed by NASA for space equipment, have found applications in
a variety of fields due to their simpler structure, higher power density, and transmission efficiency compared to traditional DC
systems. Fig. 16 shows a simplified LCLC HFAC system circuit diagram. HFAC system eliminate the need for AC–DC and DC–AC
therefore fewer components are required in the system resulting in higher reliability [170]. The switching network generates a high-
frequency quasi-square-wave voltage to convert the DC voltage into HFAC voltage or current. Increasing the switching frequency
allows miniaturization of inverters and reduces cost. However, this creates several challenges, including switching and magnetic
losses. The challenge can be improved by proper design methods, topology, drive circuits, and magnetic components. This makes
HFAC suitable as a second case study to investigate the feasibility of LLM-driven design. Wide band gap (WBG) devices such as Silicon
Carbide (SiC) and Gallium Nitride (GaN) are at the front end of the HF system [171]. The WBG devices enable system to work at high
frequency switching employing soft switching technique zero voltage switching (ZVS) and zero current switching (ZCS) within the
system. HFAC systems have many applications and detailed studies have been reported in [172,173]. For example, [174] investigated
control strategy in HFAC. Their results show that proportional integral resonant (PIR) control shows better performance and lower
total harmonic attenuation capability. As such, the PIR control is adopted in the LLM-driven design. Full information about the
design including design correlations can the found in the supplementary documents. Preliminary studies in HFAC reported various
advantages, including flexibility in meeting different voltage requirements, ease of electrical isolation with compact high-frequency
transformers, fewer components, and reduced system integration complexity. Phase-Shift Modulation (PSM) is one of the essential
control strategies in HFAC systems and is utilized in high-efficiency applications with reduced switching losses. As such the power
flow in the LLM-driven design is controlled by PSM through an adjustment of the phase angle between the input and output voltage
waveforms. This technique substantially minimizes switching losses by permitting ZVS and ZCS soft switching [175–179]. The LCLC
HFAC simulation was conducted using PSIM. Design specification from [174] is adopted to allow the LLM-driven design validation
and assessment.
The PSIM simulation block as shown in Fig. S7 in the supplementary information document includes a Texas Instrument (TI)
second-order low pass filter (LPF) UAF42, PIR controller, phase shift modulation logic gate to drive the high-frequency AC inverter,
and a resistive load using the HFAC H-bridge topology. The complete PIR circuit diagram and all other necessary information can
be found in Fig S8, Fig S9 and Table S3 in the supplementary documents, respectively. A ‘Chat sheet’ and video of HFPE-AI driven
design showing all the prompts and responses are also provided in the supplementary sections of this paper. The design equations
for the HFAC and the complete specification of the inverter can also be found in Table S1 in the supplementary documents.

18
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Fig. 16. Circuit diagram of single-phase HFAC.

The LLM-AI is prompted to design a 130 W HFAC inverter using the design specifications as reported in [174]. The results of the
LLM-AI-driven design of the LCLC HFAC inverter and validation are shown in Figs. 17 to 18 and Table 10, respectively. The HFPE-AI
proposed HFAC design operating at 25 kHz demonstrates balance electrical dynamics. Fig. 17(a) illustrates the resonant current 𝐼𝑎𝑏 ,
maintaining a sinusoidal waveform with consistent amplitude which is common in optimal resonance circuit. The corresponding
H-bridge arm voltage 𝑉𝑎𝑏 exhibits a square waveform, aligning with the expected operational behaviour of a high-frequency inverter
topology working based on phase shift modulation. Fig. 17(b) shows the achieved zero voltage switching (ZVS) and zero current
switching (ZCS) within the system. ZVS is achieved in leg B as shown in the plot of voltage and current 𝐼𝐷𝑆 𝐷 across switch CD.
Similarly, ZCS is achieved in leg A as shown in the plot of voltage and current 𝐼𝐷𝑆 𝐵 across switch CD. Soft switching achieved in
the system’s design minimizing switching losses and ensuring high efficiency. These results closely align with those reported in the
reference paper [174], and further detailed comparisons can be found in Table 10. This comprehensive illustration underscores the
practical application of the LLM-driven design.

Fig. 17. HFPE-AI proposed HFAC design at 25 kHz (a) resonant current, and H-bridge arm voltage (b) ZVS and ZCS achieved, respectively.

The proposed AI design output performance is presented in Fig. 18 showing the output current 𝐼𝑜 and voltage 𝑉𝑜 . The simulation
results show a total harmonic distortion (THD) of 1.23% and a peak voltage of 41.6 V, compared to the experimental results from
the reference paper, which reported a THD of 1.65% and a peak voltage of 39.4 V [174]. While the simulation values differ slightly
from the experimental data, this is expected due to the inherent differences between simulations and experiments. Further detailed
comparisons can be found in Table 10 with the reference data to validate the LLM-AI-driven design.
Table 10 provides a detailed comparison of the HFPE (LLMs-AI) design results from this study with the reference work [174],
highlighting key electrical parameters and performance metrics. The load resistance in this work is 6.03 Ω as proposed by the

19
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Fig. 18. HFPE-AI proposed HFAC design output currents, and voltage with THD at 25 kHz.

LLMs-AI, closely matching the reference value of 6.00 Ω, with a standard deviation (SD) of 0.03 Ω. The series capacitance and
inductance also align well with the reference values, showing minimal deviations with SDs of 0.102 μF and 0.004 μH, respectively,
while both designs maintain a series resonant frequency of 25 kHz without any deviation. The parallel network shows moderate
deviations, particularly in inductance, with an SD of 11.34 μH, reflecting possible slight errors from the LLMs-AI design approach.
The output voltage of 41.6 V in this study is slightly higher than the reference value of 39.4 V, with an SD of 1.56 V, demonstrating
an improvement in voltage amplitude. Additionally, the total harmonic distortion (THD) achieved is 1.23%, lower than the reference
value of 1.65%, with an SD of 0.30%, as reported in [174]. The simulation efficiency for the proposed design is 92.91%, with a SD
of 1.59%, compared to the reference simulation efficiency of 94.50%. These results collectively demonstrate the potential LLMs-AI
design assistant approach in high-frequency AC applications, which can be extended to other areas. A Wilcoxon signed-rank test was
conducted to compare the Ref. [174] and LLMs-AI (this work). With a 𝑝-value of 0.50, the test revealed no significant difference
between the two groups. This indicates that we fail to reject the null hypothesis, suggesting that there is no evidence of a systematic
difference between the reference paper and the LLM-AI design. The complete Wilcoxon statistical test can be found in the data
spreadsheet in the supplementary information section.
Table 10
Output comparison of HFPE (LLMs-AI) design with [174].
Parameter [174] LLMs-AI (This work) SD
Method Experiment Simulation
Load resistance (Ω) 6.00 6.03 0.03
Reactance of series capacitor (Ω) – 17.28 –
Series capacitance (μF) 0.47 0.37 0.102
Series inductance (μH) 110.00 109.996 0.004
Series resonant frequency (kHz) 25.00 25.00 0.00
Reactance of parallel capacitor (Ω) – 3.54 –
Parallel capacitance (μF) 1.80 1.81 0.01
Parallel inductance (μH) 17.00 28.34 11.34
Parallel resonant frequency (kHz) – 37.50 –
Output voltage (V) 39.4 41.6 1.56
Total harmonic distortion (%) 1.65 1.23 0.30
Efficiency (%) - Experimental 93.33 – –
Efficiency (%) - Simulation 94.50 92.91 1.59

7. Discussion and future research direction

AI algorithms are used in various sectors to personalize and simplify interactions, and tasks, or solve problems. This creates an
interesting and novel opportunity for the integration of LLMs, such as GPT, within the processes of designing power electronics. The
approach has prospects that can make the design process much faster, with reduced potential for errors, and increased productivity
in current trends and future demands in the field of power electronics. ChatGPT, published by Open AI in 2022, is one of the
fastest-adopted AI technologies in history and marks a new AI beginning in 2023. It is based on natural language processing and is
used to understand and enhance predictions of algorithms in various sectors. Machine learning incorporates random forest methods
for clustering, regression, and classification. Regression presents a statistical methodology that predicts continuous values of target
variables, based on their relation to the input factors. In the applications of a control system, an intelligent controller can use
regression analysis to build mathematical models of dependencies between the input signals and control variables. The concept of

20
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Table 11
Prospect of LLMs AI applied to power electronics design.
Design phase Applications of Description
LLMs
Generating design Analyzes large datasets of existing designs to propose new configurations.
Objective & concepts
Conceptual
Refining Interprets user inputs and compares them with industry standards to refine design
specifications specifications.

Design Requirements Utilizes natural language processing to extract and analyse requirements from technical
specification & analysis documents and stakeholder inputs.
Requirement Specification Automatically checks the consistency and completeness of specifications against known
verification standards and prior projects.
Parameter Automates the optimization process by exploring potential solutions and leveraging
optimization historical data.
Computation/
Mathematical Performance Integrates with simulation tools to run and interpret simulations, suggesting modifications
modelling simulation for improvement.
Model-based design Uses mathematical models to simulate and validate system behaviour under various
operating conditions.
Sensitivity analysis Identifies critical design parameters and assesses their impact on overall system
Simulation & performance.
Optimization
Scenario testing Runs multiple simulations to evaluate system robustness and performance across different
scenarios.
Automated Analyzes design requirements to recommend suitable components based on availability,
Prototyping component selection cost, and performance.
Prototyping Generates detailed instructions and documentation for creating prototypes, ensuring
assistance accuracy and efficiency.
Code generation Automatically generates code for embedded systems based on design specifications and
Code/Software models.
development
Software validation Uses LLMs to validate software against design requirements and industry standards.
Automated test Creates test cases and procedures for hardware validation using patterns learned from
Hardware test and generation historical test data.
debugging
Fault diagnosis Analyzes test results to diagnose faults and suggest corrective actions.
System integration Provides integration guidelines and checks compatibility between different subsystems.
Set up, Integration support
and system test
End-to-End testing Executes comprehensive tests to ensure all system components work together seamlessly
and meet design specifications.
Generating technical Automatically generates comprehensive documentation, including design descriptions and
Certification and documentation testing procedures.
compliance
Compliance checks Analyzes design against industry standards and regulatory requirements, suggesting
modifications for compliance.
Final prototype Final system Ensures the final prototype meets all design and performance criteria through rigorous
validation validation testing and validation.
Production Provides recommendations for scaling up production efficiently while maintaining quality
Mass production optimization standards.
Continuous Analyes production data to suggest ongoing improvements and optimizations in the
improvement manufacturing process.

using LLMs for the optimization of design processes in power electronics has been relatively new. While applications of LLM cut
across quite a few domains, there is very little on record about its applications in power electronics design time reduction and
accuracy improvement. Future work would include impact assessment to identify trade-offs, where researchers have used LLMs
in the prediction of material properties, and parsing design documents, specifications, and simulation results for optimal power
electronics design. This research into exploring the integration of LLMs like GPT in the design process of power electronics bridges
an important gap in the current methodologies by leveraging advanced AI to drive design efficiency and accuracy. This work will
advance the field of power electronics, and this application of LLMs in the engineering design process will set a concrete precedent
for other such developments. Table 11 highlights how LLMs can enhance various stages of the power electronics design process,
from conceptual design to compliance checks, by automating and optimizing key tasks.
The integration of LLMs into power electronics design presents both opportunities and challenges. One significant challenge is
the dependency on large datasets. LLMs require extensive amounts of data for training and fine tuning to function effectively, which
in the context of PE design may not always be readily available. LLMs’ reliance on large datasets not only raises concerns about
data availability but also about the quality of the data. Potential biases can affect the generalization ability of the models, limiting
their effectiveness across different PE applications. Implementing LLMs in PE design is computationally intensive, often necessitating

21
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

Table 12
Future research directions for LLMs in power electronics design.
Research direction Potential research areas Description
Advanced LLMs AI algorithms
Hybrid models Combining LLMs with other Integrate LLMs with RL or graph neural networks (GNNs) for
AI techniques enhanced performance.
Generalization Mitigating hallucination and Explore possibility cross-reference model outputs with trusted
ability of LLMs improving accuracy databases and simulation tools to validate accuracy in critical
applications like PE design.
Meta-learning Models learning to learn Reduce the need for extensive fine-tuning by enabling models to
adapt more quickly to new tasks.
Explainability and Making LLMs more Develop techniques to help PE engineers understand model
transparency explainable and transparent decisions, increasing trust and facilitating adoption.
Real-time implementation of LLMs
Real-time design Developing real-time Improve the performance and efficiency of power electronics
optimization optimization algorithms systems with reduced computational latency within simulation
platforms such as MATLAB, PSIM, ALTIUM, etc.
Edge computing Deploying LLMs closer to the Reduce latency and enhance real time processing capabilities for
source of data applications like adaptive control in power converters.
Interdisciplinary collaboration
Collaborative Platforms for sharing data, Accelerate research and development through joint projects and a
platforms models, and insights across multidisciplinary approach.
disciplines using LLMs
Cross domain Methods for transferring Adapt techniques from other fields for power electronics design.
knowledge transfer knowledge across domains
using LLMs
Educational Promoting integration of AI Need for a new generation of engineers skilled in both AI and
initiatives and power electronics in power electronics leveraging the advancement in LLMs.
education
Data-driven design, optimization and sustainable
Big data analytics Developing more AI Reveal patterns and trends from large datasets to inform design
algorithms and analytics to decisions.
simplify power electronics Improve efficiency and stability of renewable energy generation,
design storage, and distribution by system troubleshooting during design,
or maintenance.
Predictive Developing predictive Enhance the reliability and lifespan of power electronic systems by
maintenance maintenance strategies using predicting failures and optimizing maintenance schedules.
LLMs Optimize charging/discharging cycles, predict energy demand, and
ensure efficient utilization.
Digital twins Creating digital twins of Enable real-time monitoring and optimization through simulations
power electronics systems of physical systems.
using LLMs Analyses and optimizes grid operations, detect faults and integrate
distributed energy resources.

high-performance hardware and substantial energy consumption. Proprietary constraints can limit access to model or level of fine
tuning, potentially leading to models that lack robustness or bias. Future research should focus on creating LLM not depending on
foundation model to shared datasets and exploring data augmentation techniques to mitigate these issues. Another challenge lies in
the potential bias and ethical considerations associated with AI models during creation or fine tuning. Bias in training data can lead
to inaccurate results, which is concerning in various applications of power electronics. Addressing this requires developing methods
for bias detection and correction, as well as establishing ethical guidelines for AI use in PE. This requirement can be a barrier,
especially for smaller organizations with limited budgets. For smaller organizations or projects with tight budgets, the practicality
of implementing LLMs is a critical issue. High costs associated with creating and fine tuning can make adoption challenging. The
financial investment required for deploying LLMs includes not only computational resources but also ongoing maintenance and
updates. The total cost for the OpenAI ‘Team workspace’ used is $25 per seat/month for the monthly subscription. This is crucial
for ensuring that the AI models remain effective as technologies and design requirements evolve. The ability of LLMs refers to their
capacity to apply learned knowledge to new and varied situations in electronics design based on continuous data update that can also
be a financial burden. Potential users and creators may find these costs as a barrier, potentially widening the gap between large and
small organizations in technological capabilities. Ethical implications are equally important, especially in sensitive applications like
power electric and sourcing data from data base like Scopus, IEEE, google scholar etc. The use such data may raises concerns about
accountability, transparency, and the potential for unintended consequences. More case study of LLM driven design of PE is need. In
addition, strategies to enhance real world applicability, such as incorporating stochastic elements and experimental validation are
ares for future research directions. To address these issues the following mitigations can be adopted as outline in Table 12. Tables

22
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

13 and 14 outline the challenges and future research directions for integrating LLMs into power electronics design, with potential
advancements and applications to PE.
Table 13
LLMs implementation limitations.
Type Mitigation strategies
Data diversification Incorporate a wide range of data sources to enhance model robustness.
Bias mitigation techniques Employ algorithms specifically designed to detect and reduce bias in AI models.
Data availability and Poor-quality data, Use reliable data sources for data gathering
quality including inaccuracies, Encourage industry collaboration to develop comprehensive datasets while respecting
biases, or inconsistencies, proprietary boundaries.
impacts model performance Implement rigorous data validation and cleaning processes to ensure accuracy and
consistency.
Use robust data preprocessing techniques to handle biases and inconsistencies.
Efficient model Explore the development of smaller, more efficient models requiring less data
development without compromising performance.
Model Optimization Research into model compression and optimization to reduce computational load.

Computational costs Cloud computing Utilize cloud services for scalable resources without significant upfront investment.
and resource demands Algorithm efficiency Develop more efficient algorithms tailored to PE applications to lower resource
demands.
Open-source platforms Leverage open-source AI tools and models to reduce costs.

Accessibility for smaller Shared resources Participate in consortiums or partnerships to share computational resources and
organizations expertise.
Training and education Invest in training programmes to build in-house capabilities, minimizing reliance on
external specialists.
Hallucination Design clear and specific prompts to guide the model towards accurate and relevant
outputs.
Ensure that datasets undergo rigorous validation to identify and remove inaccuracies
Generalization ability or inconsistencies. Data from standard data base should be use which include web of
of LLMs science, Scopus, IEEE, etc
Broad training data Use diverse datasets covering a wide range of scenarios and conditions.
Transfer learning Apply knowledge gained from one domain to another, reducing the need for
extensive retraining.
Continuous Learning Implement mechanisms for models to learn incrementally from new data over time.
Ethical frameworks Establish clear ethical guidelines for AI deployment in PE, focusing on transparency
Practical barriers and and responsibility.
ethical implications Regulatory compliance Ensure that AI applications meet all regulatory standards and certifications required
in the industry.
Stakeholder engagement Involve a broad range of stakeholders, including policymakers, industry experts, and
the public, in discussions about AI integration.

Table 14
LLMs challenges and potential solutions for implementation.
Challenge Description Potential solutions
Data availability and quality
Domain- Highly data sensitive as the functionality More additional case studies are required.
specific data depends on data used for development Encourage collaboration between academia, industry, and research
sensitive and fine tuning. organizations to create comprehensive datasets based on a reliable
database.
Integrate empirical data through simulations or other AI models.
Model adaptation and customization
Fine-tuning Requires expertise in both AI and power Use pretrained models on large, general datasets or fine-tune-based
complexity electronics to fine-tune models domain-specific datasets.
successfully. Implement techniques such as dropout or early stopping to prevent
overfitting.
Computational requirements
High training Significant computational resources are Use cloud platforms to access scalable resources on a pay-as-you-go
costs required for training and fine tuning basis.
LLMs

8. Conclusion

We have reviewed AI applied in power electronics and proposed the potential of LLMs in electronic design. This study provides
key information on implementing a large language model AI assistant tool including challenges, limitations and future research

23
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

directions. Two case studies – high-frequency wireless power transfer design and HFAC system design – are presented to investigate
the design approach. A ‘Chat sheet’ and video of the case studies have been provided in the supplementary section. The designs
were compared with reference designs available in the literature. Additionally, we proposed an LLMs-AI driven design framework to
improve the design process. Advancement in design methodologies and design tools such as LLM-AI driven design has the potential
to streamline the design process of power electronic. LLM-AI driven design optimization enables quicker iteration, significantly
reduces development cycles, and fosters innovation in power electronics design. Advancements in LLM-AI driven design, have
the potentials to revolutionize the design process of power electronics, including conceptualization, modelling, rapid prototyping,
component selection, parameter optimization, and validation. This can significantly decrease the time required to produce a circuit
board. Furthermore, LLM has the potential to unify the design framework for power electronics with a combined knowledge-based
assistance data base. The use of AI methodology has significant prospects in PE design, and LLMs including OpenAI GPT, have
the capabilities of understanding and processing technical specifications, design parameters, and optimization of power electronics
which can generate new design ideas. This paper highlights trends in methodology usage together with advantages, limitations,
and future research directions for AI algorithms in power electronics. In addition, LLM perspectives with interconnections among
relevant AI algorithms, core functions, and applications are also discussed.

Declaration of Generative AI and AI-assisted technologies in the writing process

A generative AI called High Frequency Power Electronic (HFPE) was specifically created and optimized to investigate the
feasibility of LLM as PE design assistant. After using this tool, the authors reviewed and edited the content as needed and take
full responsibility for the content of the publication.

Funding

The Energy Research Lab (ERL) and Cranfield University have sponsored this work.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing
interests: Khalifa Aliyu Ibrahim reports financial support was provided by Cranfield University - Through Energy Research Lab. If
there are other authors, they declare that they have no known competing financial interests or personal relationships that could
have appeared to influence the work reported in this paper.

Acknowledgements

The Energy Research Lab (ERL) has provided Cranfield University a PhD Studentship for the first author.

Appendix A. Supplementary material

A supplementary document, data, complete ‘Chat sheet’ and video of the LLM-AI driven designs of the case studies are provided.
Supplementary material related to this article can be found online at [Link]

Data availability

Data will be made available on request.

References

[1] Salomez F, Helbling H, Almanza M, Soupremanien U, Viné G, Voldoire A, et al. State of the art of research towards sustainable power electronics. Sustain
2024;16:2221. [Link]
[2] Zahnstecher B, Carobolante F, McCune E, Gerber D, Kirkpatrick D, Olsson M, et al. Energy efficiency. In: 2023 IEEE future networks world forum. FNWF,
IEEE; 2023, p. 1–148. [Link]
[3] Akhunzada A, Al-Shamayleh AS, Zeadally S, Almogren A, Abu-Shareha AA. Design and performance of an AI-enabled threat intelligence framework for
IoT-enabled autonomous vehicles. Comput Electr Eng 2024;119:109609. [Link]
[4] Lagier T, Dworakowski P, Chedot L, Wallart F, Lefebvre B, Maneiro J, et al. How good are the design tools in power electronics? In: 2020 22nd European
conference on power electronics and applications. EPE’20 ECCE Europe, IEEE; 2020, p. P.1–P.12. [Link]
2020.9215942.
[5] Lin Y-W, Tang TLE, Sangiovanni-Vincentelli AL, Schiavon S, Spanos CJ. From electronic design automation to building design automation: Challenges
and opportunities. 2023.
[6] Zhao S, Blaabjerg F, Wang H. An overview of artificial intelligence applications for power electronics. IEEE Trans Power Electron 2021;36:4633–58.
[Link]
[7] Das SR, Ray PK, Sahoo AK, Singh KK, Dhiman G, Singh A. Artificial intelligence based grid connected inverters for power quality improvement in smart
grid applications. Comput Electr Eng 2021;93. [Link]
[8] Ding L, Wei X-C, Tang Z-Y, Wen J, Gao L, Gao RX-K. Near-field scanning based shielding effectiveness analysis of system in package. IEEE Trans Compon
Packag Manuf Technol 2021;11:1235–42. [Link]

24
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

[9] Sadeghi Z, Alizadehsani R, Cifci MA, Kausar S, Rehman R, Mahanta P, et al. A review of explainable artificial intelligence in healthcare. Comput Electr
Eng 2024;118:109370. [Link]
[10] Han M, Yang T, Zhong J, Zhong Y. AI applications and supply chain concentration. Appl Econ Lett 2023;1–5. [Link]
2210813.
[11] Schleiss J, Laupichler MC, Raupach T, Stober S. AI course design planning framework: Developing domain-specific AI education courses. Educ Sci (Basel)
2023;13:954. [Link]
[12] Lee SJ, Kwon K. A systematic review of AI education in K-12 classrooms from 2018 to 2023 : Topics, strategies, and learning outcomes. Comput Educ:
Artif Intell 2024;6:100211. [Link]
[13] Zhang Y, Cao J. Design of english teaching system using artificial intelligence. Comput Electr Eng 2022;102:108115. [Link]
compeleceng.2022.108115.
[14] Rahman H, D’Cruze RS, Ahmed MU, Sohlberg R, Sakao T, Funk P. Artificial intelligence-based life cycle engineering in industrial production: A systematic
literature review. IEEE Access 2022;10:133001–15. [Link]
[15] Jackson DM. Multi-objective power electronics system design and optimization, a machine learning approach. Oregon State University; 2021.
[16] Zhan X, Wang W, Chung H. A neural-network-based color control method for multi-color LED systems. IEEE Trans Power Electron 2019;34:7900–13.
[Link]
[17] Ponce P, Mendez E, Molina A. Teaching fuzzy controllers through a V-model based methodology. Comput Electr Eng 2021;94:107267. [Link]
10.1016/[Link].2021.107267.
[18] Wei C, Zhang Z, Qiao W, Qu L. An adaptive network-based reinforcement learning method for MPPT control of PMSG wind energy conversion systems.
IEEE Trans Power Electron 2016;31. [Link]
[19] Ouyang Y, Zhao W, Wang H. Simulation and study of maximum power point tracking for rim-driven tidal current energy power generation systems.
Energy Rep 2023;9. [Link]
[20] Kumar A, Patel N, Gupta N, Gupta V. Design, analysis and implementation of electronically interfaced photovoltaic system using ARM Cortex-M4
microcontroller. Comput Electr Eng 2022;98:107701. [Link]
[21] Renjith PN, Alfurhood BS, Prashanth KV, Patil VS, Sharma N, Chaturvedi A. Coordination of modular nano grid energy management using multi-agent
AI architecture. Comput Electr Eng 2024;115:109112. [Link]
[22] Baker M, Fard AY, Althuwaini H, Shadmand MB. Real-time AI-based anomaly detection and classification in power electronics dominated grids. IEEE J
Emerg Sel Top Ind Electron 2023;4:549–59. [Link]
[23] Meng Z, Yang Y, Gao Y, Ai S, Zhang Y, Lv Y, et al. Prediction method of driving strategy of high-power IGBT module based on MEA-BP neural network.
IEEE Access 2020;8:94731–47. [Link]
[24] Rao PS, Yaqoob SI, Ahmed MA, Abdinabievna PS, Yaseen SM, Arumugam M. Integrated artificial intelligence and predictive maintenance of electric
vehicle components with optical and quantum enhancements. Opt Quantum Electron 2023;55:855. [Link]
[25] Shen X, Zuo Y, Kong J, Martinez W. Artificial intelligence applications in high-frequency magnetic components design for power electronics systems: An
overview. IEEE Trans Power Electron 2024;39:8478–96. [Link]
[26] Kim Y, Shin D, Petricca M, Park S, Pontino M, Chang N. Computer-aided design of electrical energy systems. In: IEEE/ACM international conference on
computer-aided design, digest of technical papers, ICCAD. 2013, [Link]
[27] Petkovski M, Kostov M. Model based design and hardware in the loop testing in power electronics courses. TEM J 2012;1:292–6.
[28] Diz S de L, López RM, Sánchez FJR, Llerena ED, Peña EJB. A real-time digital twin approach on three-phase power converters applied to condition
monitoring. Appl Energy 2023;334. [Link]
[29] Musiałek F, Szabra D, Wojtas J. Time-efficient SNR optimization of WMS-based gas sensor using a genetic algorithm. Sensors 2024;24:1842. http:
//[Link]/10.3390/s24061842.
[30] Santamargarita D, Salinas G, Molinero D, Bueno EJ, Vasić M. Tradeoff between accuracy and computational time for magnetics thermal model based on
artificial neural networks. IEEE J Emerg Sel Top Power Electron 2023;11:5658–74. [Link]
[31] Smahi A, Makhloufi S. Artificial neural network-based virtual synchronous generator for frequency stability improving of grid integrating distributed
generators. Comput Electr Eng 2024;120:109877. [Link]
[32] Liu M. Genetic algorithms in stochastic optimization and applications in genetic algorithms in stochastic optimization and applications in power electronics
power electronics. Uknowledge 2016.
[33] Blauth TF, Gstrein OJ, Zwitter A. Artificial intelligence crime: An overview of malicious use and abuse of AI. IEEE Access 2022;10:77110–22.
[Link]
[34] Gros I-C, Lü X, Oprea C, Lu T, Pintilie L. Artificial intelligence (AI)-based optimization of power electronic converters for improved power system stability
and performance. In: 2023 IEEE 14th international symposium on diagnostics for electrical machines, power electronics and drives. SDEMPED, IEEE;
2023, p. 204–10. [Link]
[35] Ozcanli AK, Yaprakdal F, Baysal M. Deep learning methods and applications for electrical power systems: A comprehensive review. Int J Energy Res
2020;44. [Link]
[36] Li Y, Xue C, Zargari F, Li YR. From graph theory to graph neural networks (GNNs): The opportunities of GNNs in power electronics. IEEE Access
2023;11:145067–84. [Link]
[37] Moukafih Y, Ghogho M, Smaili K. Supervised contrastive learning as multi-objective optimization for fine-tuning large pre-trained language models.
In: ICASSP, IEEE international conference on acoustics, speech and signal processing - proceedings. vol. 2023-June, 2023, [Link]
ICASSP49357.2023.10095108.
[38] Qin Z, Guo H, Cheng X, Dou Z, Lian H, Li X, et al. Recent advances of logic gate circuits based on metallo-organic compounds. ACS Mater Lett
2024;6:1504–21. [Link]
[39] Sivapriya A, Kalaiarasi N. A novel enhanced deep learning-based fault diagnosis approach for cascaded multilevel inverter. E- Prime - Adv Electr Eng
Electron Energy 2023;5:100253. [Link]
[40] Chen H, Hu C, Han B, Miao K. A method of diagnosing analog circuit soft faults using boruta features and LightGBM. Electron (Basel) 2024;13:1123.
[Link]
[41] Xiao S, Shen M, Yu L. Energy saving analysis of refrigeration room group control based on kernel ridge regression algorithm. Int J Refrig 2023;153.
[Link]
[42] Kulikov A, Loskutov A, Bezdushniy D. Relay protection and automation algorithms of electrical networks based on simulation and machine learning
methods. Energies (Basel) 2022;15:6525. [Link]
[43] Wu T, Wang Z, Ozpineci B, Chinthavali M, Campbell S. Automated heatsink optimization for air-cooled power semiconductor modules. IEEE Trans Power
Electron 2019;34. [Link]
[44] Dantonio M, Shi C, Wu B, Khaligh A. Design and optimization of a solar power conversion system for space applications. IEEE Trans Ind Appl 2019;55.
[Link]
[45] Tianxiang L, Hang Y, Chuanru W, Qingyuan W, Pengfei S, Xiaoyun F. Parameter identification of train basic resistance based on particle swarm
optimization. In: 2022 IEEE 17th conference on industrial electronics and applications. ICIEA, IEEE; 2022, p. 1572–7. [Link]
ICIEA54703.2022.10006030.

25
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

[46] Zhang Jun, Chung HS-H, Lo AW-L, Huang Tao. Extended ant colony optimization algorithm for power electronic circuit design. IEEE Trans Power Electron
2009;24:147–62. [Link]
[47] Feng X, Wang Y, Dong T, Liao Y, Zhang Y, Lin Y. A routing algorithm for node protection in wireless sensor network based on clustering ant colony
strategy. In: Lecture notes in computer science (including subseries lecture notes in artificial intelligence and lecture notes in bioinformatics), vol. 13340,
LNCS; 2022, p. 184–93. [Link]
[48] Liu X-F, Zhan Z-H, Zhang J. Resource-aware distributed differential evolution for training expensive neural-network-based controller in power electronic
circuit. IEEE Trans Neural Netw Learn Syst 2022;33:6286–96. [Link]
[49] Zghoul FN, Alteehi H, Abuelrub A. A mayfly-based approach for CMOS inverter design with symmetrical switching. Algorithms 2023;16:237. http:
//[Link]/10.3390/a16050237.
[50] Zhou Z, Ning Y, Zhou X, Zheng F. Improved artificial bee colony algorithm-based channel allocation scheme in low earth orbit satellite downlinks.
Comput Electr Eng 2023;110:108838. [Link]
[51] Lopez-Arredondo J, Tlelo-Cuautle E, Fernandez FV. Optimization of LDO voltage regulators by NSGA-II. In: 2016 13th international conference on
synthesis, modeling, analysis and simulation methods and applications to circuit design. SMACD, IEEE; 2016, p. 1–4. [Link]
2016.7520743.
[52] Yang X, Zhou D, Song W. A NSGA-II-based layout method for cable bundles with branches using machine learning. IEEE Access 2021;9. http:
//[Link]/10.1109/ACCESS.2021.3091623.
[53] Li Y, Wang Y, Li Y, Zhou R, Lin Z. An artificial neural network assisted optimization system for analog design space exploration. IEEE Trans Comput-Aided
Des Integr Circuits Syst 2020;39. [Link]
[54] Dragicevic T, Wheeler P, Blaabjerg F. Artificial intelligence aided automated design for reliability of power electronic systems. IEEE Trans Power Electron
2019;34:7161–71. [Link]
[55] Ferracuti F, Fonti A, Ciabattoni L, Pizzuti S, Arteconi A, Helsen L, et al. Data-driven models for short-term thermal behaviour prediction in real buildings.
Appl Energy 2017;204:1375–87. [Link]
[56] Muttillo V, Pomante L, Santic M, Valente G. Systemc-based co-simulation/analysis for system-level hardware/software co-design. Comput Electr Eng
2023;110:108803. [Link]
[57] Kleiner J, Stuckenberger M, Komsiyska L, Endisch C. Real-time core temperature prediction of prismatic automotive lithium-ion battery cells based on
artificial neural networks. J Energy Storage 2021;39. [Link]
[58] Adewuyi OB, Krishnamurthy S. Appraisal of viable optimization techniques for efficient design and control planning of renewable energy-based microgrid
systems. Comput Electr Eng 2024;120:109652. [Link]
[59] D’Antuono V, De Matteis G, Trotta D, Zavoli A. Optimization of UAV robust control using genetic algorithm. IEEE Access 2023;11. [Link]
1109/ACCESS.2023.3325845.
[60] Makhmudov T. Tuning parameters of PID controller for simplest electric power system using a genetic algorithm. AIP Conf Proc 2023;2612. http:
//[Link]/10.1063/5.0113019.
[61] Ji B, Song X, Sciberras E, Cao W, Hu Y, Pickert V. Multiobjective design optimization of IGBT power modules considering power cycling and thermal
cycling. IEEE Trans Power Electron 2015;30. [Link]
[62] Alfergani A, Alaaesh S, Shamekh A, Khalil A, Asheibi A. Improved power sharing in inverter based microgrid using multi-objective optimization. Comput
Electr Eng 2023;110:108902. [Link]
[63] Xiao Z, Guo Z, Balyan V. Fault diagnosis of power electronic circuits based on improved particle swarm optimization algorithm neural network. Electr
2023;22:365–72. [Link]
[64] Wang Q, Zhao P, Du X, Lin F, Li X. Electromagnetic vibration analysis and slot-pole structural optimization for a novel integrated permanent magnet
in-wheel motor. Energies (Basel) 2020;13. [Link]
[65] Azharuddin M, Jana PK. Particle swarm optimization for maximizing lifetime of wireless sensor networks. Comput Electr Eng 2016;51:26–42. http:
//[Link]/10.1016/[Link].2016.03.002.
[66] Alasali F, El-Naily N, Mustafa HY, Loukil H, Saad SM, Saidi AS, et al. Innovative protection schemes through hardware-in-the-loop dynamic testing.
Comput Electr Eng 2024;119:109559. [Link]
[67] Akkan N, Altun M, Sedef H. Modeling and parameter extraction of OFET compact models using metaheuristics-based approach. IEEE Access
2019;7:180438–50. [Link]
[68] Unlersen MF, Balci S, Sabanci K. Thermal behavior estimation of the power switches with an empirical formulation optimized by Artificial Bee Colony
algorithm. Microelectron Reliab 2021;127. [Link]
[69] Tiwary A. Advanced and innovative optimization techniques in controllers: A comprehensive review. Int J Recent Innov Trends Comput Commun
2021;9:23–7. [Link]
[70] Qing A. Differential evolution. Wiley; 2009, [Link]
[71] Biswas P, Suganthan P, Amaratunga G. Optimal power flow solutions using algorithm success history based adaptive differential evolution with linear
population reduction. In: 2018 IEEE international conference on systems, man, and cybernetics. SMC, IEEE; 2018, p. 249–54. [Link]
SMC.2018.00053.
[72] Ma Y, Chou F-I, Yang P-Y, Tsai J-T, Yang Z-Y, Chou J-H. Optimal parameter design by NSGA-II and Taguchi method for RCD Snubber circuit. IEEE
Access 2020;8:182146–58. [Link]
[73] Prasad LB, Sahu S, Gupta M, Srivastava R, Mozhui L, Asthana DN. An improved method for MPPT using ANN and GA with maximum power comparison
through perturb & observe technique. In: 2016 IEEE uttar pradesh section international conference on electrical, computer and electronics engineering,
UPCON 2016. 2017, [Link]
[74] Jarndal A, Hamdan S, Bettayeb M. On neural networks modeling based on GA, PSO and GW optimization techniques. In: 2019 8th international conference
on modeling simulation and applied optimization. ICMSAO, IEEE; 2019, p. 1–5. [Link]
[75] Ambekar M, Kittur J. An integrated WPCA-ANN-GA approach to solve multiresponse optimization problems. In: 2018 international conference on
computational techniques, electronics and mechanical systems. CTEMS, IEEE; 2018, p. 338–43. [Link]
[76] Djerboub K, Allaoui T, Champenois G, Denai M, Habib C. Particle swarm optimization trained artificial neural network to control shunt active power
filter based on multilevel flying capacitor inverter. Eur J Electr Eng 2020;22:199–207. [Link]
[77] Vak G, Reddy MD. PSO trained feed forward neural network based SAPF for power quality enhancement in distribution networks. Int J Electr Electron
Eng Telecommun 2023;12:279–87. [Link]
[78] Wan Y, Xu Q, Dragicevic T. Safety-enhanced self-learning for optimal power converter control. IEEE Trans Ind Electron 2024. [Link]
TIE.2024.3363759.
[79] George MA, Kamat DV, Kurian CP. Electronically tunable ACO based fuzzy FOPID controller for effective speed control of electric vehicle. IEEE Access
2021;9:73392–412. [Link]
[80] Yao X, Zhao W, Liu Y. Adaptive fuzzy-based fault-tolerant boundary control of flexible timoshenko manipulators under directed graph. IEEE Trans Fuzzy
Syst 2024;32:294–305. [Link]

26
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

[81] Prasad Sharma K, Baine N. Application of a fuzzy logic based controller for peak load shaving in a typical household. In: 2019 IEEE international
conference on fuzzy systems. FUZZ-IEEE, vol. 2019-June, IEEE; 2019, p. 1–6. [Link]
[82] Hossain I, Shafiullah Md, Abido M. Induction motor speed control employing LM-NN based adaptive PI controller. Renew Energy Power Qual J
2020;18:97–102. [Link]
[83] Li Y, Du J, Jiang W. Reinforcement learning for process control with application in semiconductor manufacturing. IISE Trans 2024;56:585–99.
[Link]
[84] Stanojev O, Kundacina O, Markovic U, Vrettos E, Aristidou P, Hug G. A reinforcement learning approach for fast frequency control in low-inertia power
systems. In: 2020 52nd North American power symposium. NAPS, IEEE; 2021, p. 1–6. [Link]
[85] Haque MS, Choi S. Deep Deterministic Gradient Policy (DDGP) reinforcement learning assisted degradation-aware control of solid-state transformer. In:
2021 IEEE applied power electronics conference and exposition. APEC, IEEE; 2021, p. 2090–6. [Link]
[86] Barrero F, Gonzalez A, Torralba A, Galvan E, Franquelo LG. Speed control of induction motors using a novel fuzzy sliding-mode structure. IEEE Trans
Fuzzy Syst 2002;10:375–83. [Link]
[87] Salem AA, Aldin NAN, Azmy AM, Abdellatif WSE. Implementation and validation of an adaptive fuzzy logic controller for MPPT of PMSG-based wind
turbines. IEEE Access 2021;9. [Link]
[88] Hosen MA, Khosravi A, Kabir HMD, Johnstone M, Creighton D, Nahavandi S, et al. NN-based prediction interval for nonlinear processes controller. Int
J Control Autom Syst 2021;19:3239–52. [Link]
[89] Fei J, Chen Y, Liu L, Fang Y. Fuzzy multiple hidden layer recurrent neural control of nonlinear system using terminal sliding-mode controller. IEEE Trans
Cybern 2021. [Link]
[90] Van Der Broeck CH, Kalker S, De Doncker RW. Intelligent monitoring and maintenance technology for next-generation power electronic systems. IEEE J
Emerg Sel Top Power Electron 2023;11. [Link]
[91] Yang Y, Sangwongwanich A, Blaabjerg F. Design for reliability of power electronics for grid-connected photovoltaic systems. CPSS Trans Power Electron
Appl 2016;1. [Link]
[92] Boulzazen H, Douzi C, Joubert E, Dherbécourt P, Kadi M, Fouquet F. PHM method for detecting degradation of GaN HEMT ON resistance, application
to power converter. E- Prime - Adv Electr Eng Electron Energy 2022;2:100060. [Link]
[93] Abuelnaga A, Narimani M, Bahman AS. Power electronic converter reliability and prognosis review focusing on power switch module failures. J Power
Electron 2021;21. [Link]
[94] Laadjal K, Bento F, Jlassi I, Cardoso AJM. Online condition monitoring of electrolytic capacitors in DC-DC interleaved boost converters, adopting a model-
free predictive controller. In: 2021 IEEE 15th international conference on compatibility, power electronics and power engineering. CPE-POWERENG, IEEE;
2021, p. 1–6. [Link]
[95] Song Y, Sahoo S, Yang Y, Blaabjerg F. Stability constraints on reliability-oriented control of AC microgrids - theoretical margin and solutions. IEEE Trans
Power Electron 2023;38. [Link]
[96] Cui Y, Hu J, Tallam R, Miklosovic R, Zargari N. Reliability monitoring and predictive maintenance of power electronics with physics and data driven
approach based on machine learning. In: Conference proceedings - IEEE applied power electronics conference and exposition - APEC. vol. 2023-March,
2023, [Link]
[97] Li D, Channa IA, Chen X, Dou L, Khokhar S, Azar NA. A new deep learning method for classification of power quality disturbances using DWT-MRA in
utility smart grid. Comput Electr Eng 2024;117:109290. [Link]
[98] Li W, Ying J. Design and analysis artificial intelligence (AI) research for power supply - Power electronics expert system (PEES). In: Conference proceedings
- IEEE applied power electronics conference and exposition - APEC. 2008, [Link]
[99] Bose BK. Artificial intelligence techniques in smart grid and renewable energy systems - Some example applications. Proc IEEE 2017;105. http:
//[Link]/10.1109/JPROC.2017.2756596.
[100] de Souza LB, Prieto Soares P, Mendonca M, Mourhir A, Papageorgiou EI. Fuzzy cognitive maps and fuzzy logic applied in industrial processes control. In:
2018 IEEE international conference on fuzzy systems. FUZZ-IEEE, vol. 2018-July, IEEE; 2018, p. 1–8. [Link]
[101] Chen Y, Liu Z, Chen CLP, Zhang Y. Fuzzy control of switched systems with unknown backlash and nonconstant control gain: A parameterized smooth
inverse. IEEE Trans Fuzzy Syst 2022;30. [Link]
[102] Yang C, Jiang Y, Na J, Li Z, Cheng L, Su C-Y. Finite-time convergence adaptive fuzzy control for dual-arm robot with unknown kinematics and dynamics.
IEEE Trans Fuzzy Syst 2019;27:574–88. [Link]
[103] Xia X, Song H, Zhang Y, Gui L, Xu X, Li K, et al. A particle swarm optimization with adaptive learning weights tuned by a multiple-input multiple-output
fuzzy logic controller. IEEE Trans Fuzzy Syst 2023;31:2464–78. [Link]
[104] Li Y, Lu G, Li K. Adaptive fuzzy secured control for multiagent systems under DoS attacks and intermittent actuator failures. IEEE Trans Fuzzy Syst
2024;32. [Link]
[105] Mishra DK, Thomas A, Kuruvilla J, Kalyanasundaram P, Prasad KR, Haldorai A. Design of mobile robot navigation controller using neuro-fuzzy logic
system. Comput Electr Eng 2022;101:108044. [Link]
[106] De Leon-Aldaco SE, Calleja H, Aguayo Alquicira J. Metaheuristic optimization methods applied to power converters: A review. IEEE Trans Power Electron
2015;30:6791–803. [Link]
[107] Ravi S, Premkumar M, Abualigah L. Comparative analysis of recent metaheuristic algorithms for maximum power point tracking of solar photovoltaic
systems under partial shading conditions. Int J Appl Power Eng (IJAPE) 2023;12:196. [Link]
[108] Ahsan M, Hon ST, Batunlu C, Albarbar A. Reliability assessment of IGBT through modelling and experimental testing. IEEE Access 2020;8:39561–73.
[Link]
[109] Moosavi SS, Kazemi A, Akbari H. A comparison of various open-circuit fault detection methods in the IGBT-based DC/AC inverter used in electric vehicle.
Eng Fail Anal 2019;96:223–35. [Link]
[110] Robledo Edwin. The impact of AI on the electronic design industry. Autodesk; 2024, [Link]
of-ai-on-the-electronic-design-industry/ (Accessed 29 May 2024).
[111] Khamis AK, Agamy M. Converter circuits to machine learning: Optimal feature selection. In: 2022 IEEE energy conversion congress and exposition. ECCE,
IEEE; 2022, p. 1–7. [Link]
[112] Jose J, Therattil JP, Jarin T, Deepanraj B, Sumbwanyambe M. Intelligent circuit optimization for a wireless power transfer system based on L2CL-
LCL compensation. In: 2024 international conference on distributed computing and optimization techniques. ICDCOT, IEEE; 2024, p. 1–6. http:
//[Link]/10.1109/ICDCOT61034.2024.10515425.
[113] Yang Z. Scientific and technological creative stage design using artificial intelligence. Comput Electr Eng 2022;103:108395. [Link]
compeleceng.2022.108395.
[114] Marques Cardoso AJ. Power electronics design methods and automation in the digital era: Evolution of design automation tools. IEEE Power Electron
Mag 2020;7. [Link]
[115] Liu C, Ma R, Bai H, Gechter F, Gao F. A new approach for FPGA-based real-time simulation of power electronic system with no simulation latency in
subsystem partitioning. Int J Electr Power Energy Syst 2018;99. [Link]

27
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

[116] Shekhar A, Rituraj G, Van Der Sande R, Ahmadi M, Deshmukh R, Bauer P, et al. Development of reliable power electronic systems using real time digital
twin based power hardware-in-the-loop testbed. In: 2023 IEEE belgrade powerTech. IEEE; 2023, p. 1–6. [Link]
2023.10202818.
[117] Wang H, Ma K, Blaabjerg F. Design for reliability of power electronic systems. In: IECON 2012-38th annual conference on IEEE industrial electronics
society. IEEE; 2012, p. 33–44. [Link]
[118] Ma Ke, Vernica I, Blaabjerg F. Advanced design tools for the lifetime of power electronics - study case on motor drive application. In: 2016 IEEE
8th international power electronics and motion control conference. IPEMC-ECCE Asia, IEEE; 2016, p. 3255–61. [Link]
7512816.
[119] Blaabjerg F, Wang H, Vernica I, Liu B, Davari P. Reliability of power electronic systems for EV/HEV applications. Proc IEEE 2021;109:1060–76.
[Link]
[120] Alcaide AM, Buticchi G, Chub A, Dalessandro L. Design and control for high-reliability power electronics: State-of-the-art and future trends. IEEE J Emerg
Sel Top Ind Electron 2024;5:50–61. [Link]
[121] Saddriwala MA, Alam M. Artificial neural network based decentralized current-sharing control for parallel connected DC-DC converters in DC microgrid
application. Comput Electr Eng 2024;120:109731. [Link]
[122] Kamal M, Bostani A, Webber JL, Mehbodniya A, Mishra R, Arumugam M. Total harmonic distortion reduction based energy harvesting using grid-based
three phase system and integral-derivative. Comput Electr Eng 2023;109:108744. [Link]
[123] Adnan M, Ahmed I, Iqbal S, Fazal MR, Siddiqi SJ, Tariq M. Exploring the convergence of Metaverse, Blockchain, Artificial Intelligence, and digital twin for
pioneering the digitization in the envision smart grid 3.0. Comput Electr Eng 2024;120:109709. [Link]
[124] Arnautov KV, Akimov DA. Application of large language models for optimization of electric power system states. In: 2024 conference of Young researchers
in electrical and electronic engineering. ElCon, IEEE; 2024, p. 314–7. [Link]
[125] Ruan J, Liang G, Zhao H, Liu G, Sun X, Qiu J, et al. Applying large language models to power systems: Potential security threats. 2023, http:
//[Link]/10.48550/arXiv.2305.11202.
[126] Cui C, Liu J, Feng J, Hui P, Ghias AMYM, Zhang C. Large language models based multi-agent framework for objective oriented control design in power
electronics. 2024.
[127] Li R, Pu C, Tao J, Li C, Fan F, Xiang Y, et al. LLM-based frameworks for power engineering from routine to novel tasks n.d. [Link]
arXiv.2305.11202.
[128] Lin F, Li X, Lei W, Rodriguez-Andina JJ, Guerrero JM, Wen C, et al. PE-GPT: A new paradigm for power electronics design. IEEE Trans Ind Electron
2024;1–14. [Link]
[129] Liu J, Lin F, Li X, Lim KH, Zhao S. Physics-informed LLM-agent for automated modulation design in power electronics systems. 2024.
[130] Zhang G, He H, Katabi D. Circuit-GNN: Graph neural networks for distributed circuit design. In: 36th international conference on machine learning. ICML
2019, vol. 2019-June, 2019.
[131] Rafin SSH, Hussein H, Mohammed OA. An introduction to power electronics design methodology. In: 2023 IEEE design methodologies conference. DMC,
IEEE; 2023, p. 1–6. [Link]
[132] Altair. [Link]. 2023, [Link] (Accessed 6 July 2024).
[133] Magnetism and magnetic materials. Princeton University; 2024, [Link] (Accessed 6 July 2024).
[134] Flux. [Link]. 2023, [Link] (Accessed 6 July 2024).
[135] Princeton Power Electronics Research Lab. M2Spice: Advanced tools for planar magnetics modeling. 2015, [Link]
html (Accessed 6 July 2024).
[136] Princiton University. CoupL. 2020, [Link] (Accessed 6 July 2024).
[137] FreneticAI. [Link]. 2024.
[138] KU Leuven. [Link]. 2024, [Link] (Accessed 6 July 2024).
[139] Tian F, Cobaleda DB, Martinez W. Automatic data extraction based on semiconductor datasheet for design automation of power converters. In:
2022 international power electronics conference. IPEC-Himeji 2022- ECCE Asia, IEEE; 2022, p. 922–7. [Link]
ECCE53331.2022.9806859.
[140] ETH Zurichat. AI-mag. Power Electronic Systems Laboratory; 2021, [Link] (Accessed 6 July 2024).
[141] Open Magnetics. Open magnetics. 2024, [Link] (Accessed 6 July 2024).
[142] Yang J, Jin H, Tang R, Han X, Feng Q, Jiang H, et al. Harnessing the power of LLMs in practice: A survey on ChatGPT and beyond. ACM Trans Knowl
Discov Data 2024;18:1–32. [Link]
[143] Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN, et al. Attention is all you need. In: Adv neural inf process syst. vol. 2017-December,
2017.
[144] Al-Janabi OM, Alyasiri OM, Jebur EA. GPT-4 versus bard and bing: LLMs for fake image detection. In: 2023 3rd international conference on intelligent
cybernetics technology & applications. ICICyTA, IEEE; 2023, p. 249–54. [Link]
[145] Yuliang L, Hongliang L, Xiang B, Lianwen J. A brief analysis of ChatGPT:historical evolution, current applications, and future prospects. J Image Graph
2023;28:893–902. [Link]
[146] Xia J, Li B, Su S-F, Sun W, Shen H. Finite-time command filtered event-triggered adaptive fuzzy tracking control for stochastic nonlinear systems. IEEE
Trans Fuzzy Syst 2021;29:1815–25. [Link]
[147] Odry A, Tadic VL, Odry P. A stochastic logic-based fuzzy logic controller: First experimental results of a novel architecture. IEEE Access 2021;9:29895–920.
[Link]
[148] Singh S, Singh B. Optimized passive filter design using modified particle swarm optimization algorithm for a 12-pulse converter-fed LCI-synchronous
motor drive. IEEE Trans Ind Appl 2014;50. [Link]
[149] Agrawal S, Palwalia DK, Kumar M. Performance analysis of ANN based three-phase four-wire shunt active power filter for harmonic mitigation under
distorted supply voltage conditions. IETE J Res 2022;68:566–74. [Link]
[150] Wang H, Yang C, Liao X, Wang J, Zhou W, Ji X. Artificial neural network-based virtual synchronous generator dual droop control for microgrid systems.
Comput Electr Eng 2023;111:108930. [Link]
[151] Khorani V, Forouzideh N, Nasrabadi AM. Artificial neural network weights optimization using ICA, GA, ICA-GA and R-ICA-GA: Comparing performances.
In: 2011 IEEE workshop on hybrid intelligent models and applications. IEEE; 2011, p. 61–7. [Link]
[152] Choudhury S, AKr Luhach, Alnumay W, Pradhan B, Roy DS. A neuro evolutionary scheme for improved IoT energy efficiency in smart cities. Comput
Electr Eng 2022;104:108443. [Link]
[153] Abegaz BW. Unsupervised learning methods for voltage regulation in smart grids. In: 21st IEEE international conference on environment and
electrical engineering and 2021 5th IEEE industrial and commercial power system europe, EEEIC/I and CPS europe 2021 - proceedings. 2021,
[Link]
[154] Mahata S, Kubanek D, Herencsar N. Design of fractional-order transitional filters of the Butterworth-Sync-Tuned, Butterworth-Chebyshev, and Chebyshev-
Sync-Tuned types: optimization, simulation, and experimental verification. Comput Electr Eng 2024;116:109200. [Link]
2024.109200.

28
K.A. Ibrahim et al. Computers and Electrical Engineering 123 (2025) 110248

[155] Porto RW, Brusamarello VJ, Müller I, Cabrera Riaño FL, Rangel De Sousa F. Wireless power transfer for contactless instrumentation and measurement.
IEEE Instrum Meas Mag 2017;20:49–54. [Link]
[156] Quirós JC, Guerrero EV, Sangeno JK, Triviño A. Magnetic integration of circular pads and LCC-LCC for EV wireless charging tolerant to misalignment.
IEEE Access 2023;11:98558–65. [Link]
[157] Sagar A, Kashyap A, Nasab MA, Padmanaban S, Bertoluzzo M, Kumar A, et al. A comprehensive review of the recent development of wireless power
transfer technologies for electric vehicle charging systems. IEEE Access 2023;11:83703–51. [Link]
[158] Pengyang Z, Junping H, Sheng L, Lingling C. Design of 6.78 MHz magnetic coupling resonant omnidirectional wireless power transfer based on class
E2 conversion. In: 2022 IEEE 5th international electrical and energy conference. CIEEC, IEEE; 2022, p. 2092–7. [Link]
2022.9846627.
[159] Gu L, Zulauf G, Stein A, Kyaw PA, Chen T, Davila JMR. 6.78-MHz wireless power transfer with self-resonant coils at 95% DC–DC efficiency. IEEE Trans
Power Electron 2021;36:2456–60. [Link]
[160] Nikiforidis I, Kwan CH, Yates DC, Bampouras K, Gawith J, Pucci N, et al. A 3 kW 3.39 MHz DC/DC inductive power transfer system with power combining
converters. In: 2023 IEEE wireless power technology conference and expo, WPTCE 2023 - proceedings. 2023, [Link]
2023.10215793.
[161] Liu Y, He J, Ge B, Li X, Xue Y, Blaabjerg F. A simple space vector modulation of high-frequency AC linked three-phase-to-single-phase/DC converter.
IEEE Access 2020;8:59278–89. [Link]
[162] Detka K, Górecki K. Wireless power transfer—A review. Energies (Basel) 2022;15. [Link]
[163] Subudhi PS, Krithiga S. Wireless power transfer topologies used for static and dynamic charging of EV battery: A review. Int J Emerg Electr Power Syst
2020;21. [Link]
[164] Djokic SZ, Hirst D, Bonatto BD, Arango H, Liu J, Pullins SW, et al. Dealing with a complex smart grid: An integrated perspective. In: 2015 IEEE power
& energy society general meeting. vol. 2015-September, IEEE; 2015, p. 1–5. [Link]
[165] Khan SR, Pavuluri SK, Cummins G, Desmulliez MPY. Wireless power transfer techniques for implantable medical devices: A review. Sensors (Switzerland)
2020;20. [Link]
[166] Mohamed AAS, Shaier AA, Metwally H, Selem SI. A comprehensive overview of inductive pad in electric vehicles stationary charging. Appl Energy
2020;262. [Link]
[167] Kavitha M, Bobba PB, Prasad D. Comprehensive mathematical modelling and experimental analysis of a wireless power transfer system for neighborhood
electric vehicles. In: 2015 IEEE international WIE conference on electrical and computer engineering. WIECON-ECE, IEEE; 2015, p. 275–9. http:
//[Link]/10.1109/WIECON-ECE.2015.7443916.
[168] Corti F, Bertolini V, Faba A, Cardelli E, Reatti A. Design procedures for series–series wpt systems: A comparative analysis. Comput Electr Eng
2024;119:109511. [Link]
[169] Jadhav K, Urlam V, Harer R, Naik S, Deshpande A, Jadhav H. Design and implementation of resonance based wireless power transfer system. In: 2020
IEEE international conference for innovation in technology. INOCON, IEEE; 2020, p. 1–5. [Link]
[170] Chawda GS, Su W, Wang M. A comprehensive review of high-frequency AC microgrids for distribution systems. IEEE Trans Smart Grid 2024.
[Link] 1–1.
[171] Bouchour AM, Oualkadi AEl, Latry O, Dherbécourt P, Echeverri A. Estimation of losses of GaN HEMT in power switching applications based on
experimental characterization. Comput Electr Eng 2020;84:106622. [Link]
[172] Jain P, Pahlevaninezhad M, Pan S, Drobnik J. A review of high-frequency power distribution systems: For space, telecommunication, and computer
applications. IEEE Trans Power Electron 2014;29:3852–63. [Link]
[173] Luk PC-K, Ng ASY. High frequency AC power distribution platforms. In: Power electronics in smart electrical energy networks. vol. 34, London: Springer
London; 2008, p. 175–201. [Link]
[174] Zhou H, Liu J, Fang Z, Zhang P, Zhang B, Ma M, et al. Control strategy for resonant inverter in high frequency AC power distribution system with
harmonic suppression and transient performance improvement. Energies (Basel) 2022;15:8992. [Link]
[175] Tsai C-T, Liang T-C, Kuo Y-C, Luo Y-C. An improved forward converter with PFC and ZVS features for split-phase charger applications. Comput Electr
Eng 2016;51:291–303. [Link]
[176] Dong C, Wei H, Xu J. Design and analysis of soft-switching and small-signal model grid-connected micro-inverter based on staggered parallel connection.
Comput Electr Eng 2024;117:109294. [Link]
[177] Periyathiruvadi MT, Jothimani G. Power factor preregulation in power supplies using modified single stage non-isolated interleaved sepic converter with
minimal part count. Comput Electr Eng 2024;119:109476. [Link]
[178] Shalbaf AA, Shahidi N, Hemati M. A high-gain interleaved DC-DC converter with reduced components for EV charging application. Comput Electr Eng
2024;118:109316. [Link]
[179] Zhao Q, Li Z, Wang Z, Ying Y, Wang D. Push-pull forward LLC resonant converter for low input voltage applications. Comput Electr Eng 2024;118:109399.
[Link]

29