神经网络在电力电子领域的应用，比如控制的应用，拓扑的发现等等

提出了一种用高阶神经网络算法来诊断电力变换器故障的方法。以Buck变换器的故障诊断为例，设计高阶神经网络的诊断结构。采取Buck变换器在连续电流状态条件下不同工况 ...

本文针对微网孤岛运行中采用的U/f控制方法的不足，提出了一种基于BP神经网络算法的电压电流双闭环控制策略，设计了可自适应调整参数的PID控制器。采用具有自学习能力的BP神经 ...

文献[10] 将循环神经网络前馈控制器与单神经元比例–积分–微分控制器相结合对压电驱动器进行跟踪控制，有效提升了神经网络的建模精度以及控制器的跟踪性能。

本文提出了一种基于数字孪生技术的牵引变流器冷却系统PHM的构建方法。首先构建了牵引变流冷却系统关键子系统物理实体的数字孪生体，包括机理模型和数字模型；然后通过虚实之 ...

许多研究已证明，深度学习技术，尤其是卷积神经网络和长短期记忆网络(LSTM)，能够从设备运行数据中自动提取潜在的非线性特征，从而显著提升故障检测的准确性。某些研究已采用 ...

在微控制器上部署基于机器学习的算法大多需要进行离线训练，即在高性能计算平台(如个人PC、服务器或云)上预先完成机器学习模型的训练，随后将训练好的模型部署至目标设备(如 ...

该网络采用三层分层结构，每一层先通过质数组合卷积核实现多尺度特征提取，再结合Transformer编码器实现跨模态的多参量特征融合与复杂信号的捕获。实验结果表明，在各项评价 ...

将基于总体误差的自适应RBF神经网络应用于光伏电站阵列的故障识别，借助Simulink软件平台搭建光伏阵列模型并获取不同故障状态下的运行数据，开展了针对性的故障检测实验。

在逆变器中，功率器件开路故障是一种典型故障形式，会对输出性能产生显著影响 ... 文献[12]使用卷积神经网络进行故障诊断，虽然诊断准确率较高，但模型训练所需的 ...

Component parameters determine the modeling and control accuracy, and affect the performance and reliability of a power electronic system. Therefore, online identification of component parameters is very important. The existing identification methods mostly adopt the integer-order model, which is low in precision and inconsistent with practice. To solve these issues, the fractional-order model (FOM) of the buck converter is established by fractional calculus in this paper. Comparing the similarities in structure between the FOM and neural network, an identification method with the neural network is proposed. The method can realize the online identification of capacitance and order with high accuracy and fast convergence speed. Additionally, the identification results are influenced by the memory length of the FOM and the learning rate of the neural network. Finally, the experimental platform for parameter identification of buck converters is built, and the simulation and experimental results both verify the effectiveness of the proposed method.

The article deals with the control structure and implementation of a fully integrated two-phase cooling (TPC) system for power converters. A suitable testbed has been properly designed and built to perform the experimental campaign for the performance evaluation of the neural network-based control structure applied to an advanced multivariable TPC system. With respect to traditional cooling approaches, the proposed arrangement allows a greater extraction of the heat at a very low flow rate of the cooling fluid, even with standard industrial-grade heat-sinks. The technology could be implemented for the next generation of power electronics converters. The controllability of such a cooling system is still an open issue in many cases, and the proposed approach tries to suggest a feasible approach, which could be implemented on an industrial grade control board.

Heat sinks are a fundamental component of power electronics converters, so it is important to have a reliable method to study and optimize their size. Thermal analysis of heat sinks can be a complex problem as it involves different heat transfer mechanisms, and it is often necessary to use finite element simulations to obtain accurate results. However, these simulations, being very slow, are relegated to the validation process. This paper proposes a thermal model of heat sinks based on artificial neural networks. The model, unlike previous state-of-the-art models that only obtain the average temperature of the heat sink, is able to obtain a thermal map of the heat sink surface, as if it were an image, by using convolutional layers. The main advantage of this approach is that using these convolutional layers, the model is able to efficiently process how the elements are distributed on the heat sink. This model, valid for heat sinks of very different sizes in both laminar and turbulent flow, has an error of less than 1.5% and is 1500 times faster than finite element simulations, so it can be easily used in brute-force optimization processes, where many different designs need to be analyzed.

Finite control-set model predictive control (FCS-MPC) has been found as a promising alternative in the control of power converters and motor drives, albeit with model dependence issues. This inherent defect of the FCS-MPC controller triggered the widespread of model-free or data-driven control schemes in recent decades. This article, at hand, presents a data-enabled finite set predictive control solution subject to model dependence issues from the dynamic modeling point of view. In this regard, a dynamic-linearization data model is utilized to equivalently reformulate the governed power converter at each operation point. In pursuit of the accurate modeling of the plant, the time-varying parameters of the data model are updated online by an adaptive linear neural network, rendering a favorable influence on implementation. Additionally, an improved capacitance-less voltage balancing method is proposed to regulate the neutral point potential. Since the parameterless prediction process for both load currents and capacitor voltage relies solely on measured and historical input–output data of the plant, the destructive effect of parameter variations can be circumvented. To evaluate the correctness of the proposed solution, the comparative simulation and experimentation with the conventional method and state-of-the-art solutions are examined on a classic three-level neutral-point-clamped inverter.

This paper introduces physics-informed neural network (PINN) for control of grid connected converter by fusing its underlying equations into the training process, thereby reducing the requirement of qualitative training data. In comparison to the traditional data-driven methods, which either incur a significant computational burden, or use overly conservative surrogate models, it is explored that PINN can be easily optimized as per the performance requirements and is significantly superior in terms of computation time, data requirements (trained using only 3000 datapoints), and prediction accuracy (an average of 98.76%). As a result, PINN unravels new modeling orientation for power electronics, and is well-suited for commercial applications. Finally, its robustness under various grid disturbances has been validated under experimental conditions.

Existing time-series data-driven approaches for converter modeling are data-intensive, uninterpretable, and lack out-of-domain extrapolation capability. Recent physics-informed modeling methods combine physics into data-driven models using loss functions, but they inherently suffer from physical inconsistency, lower modeling accuracy, and require resource-intensive retraining for new case predictions. Consequently, catering for the challenges in current data-driven and physics-informed models, this article proposes a physics-in-architecture recurrent neural network (PA-RNN) for the time-domain modeling of power converters. The proposed PA-RNN consists of a physics-in-architecture core and a data-driven core in parallel. The physics-in-architecture core rigorously integrates circuit physical laws into its customized recurrent neural architecture by leveraging numerical differentiation, while a gated recurrent unit with layer normalization serves as the data-driven core to compensate for converter behaviors not characterized by physics. The PA-RNN modeling process is explained in detail with a design case. As 1-kW hardware and comprehensive algorithm experiments have verified the superiority of PA-RNN. Overall, PA-RNN is explainable and data-light as well as possesses good domain transfer capability to assess out-of-domain scenarios without training. This article envisions to democratize artificial intelligence for the modeling of power electronics systems.

The escalating global energy requirement, driven by population expansion and industrial development, has been met through traditional energy resources till now, many of which are now impending depletion. Renewable energy sources, specifically photovoltaic (PV) and wind power, have emerged as viable and sustainable options to fossil fuels. These systems are praised for their reliability, scalability, and cost-effectiveness, making them integral to modern energy frameworks. However, the integration of PV and wind power systems and power electronics-based loads introduces harmonic distortions, posing critical challenges to power quality and system stability. Addressing these concerns is imperative for realizing the full potential of renewable energy systems in sustainable energy development. To meet these concerns, this research proposes an ANN based DSTATCOM to mitigate power quality concerns in PV-wind power systems. Traditional DSTATCOM control appraches like “synchronous reference frame and instantaneous reactive power” often create challenges in parameter valuation and eficacy under uneven load scenarios. The model designed using XANN approach mitigates harmonics perfectly and showcase better performance even while operating under uneven non-linear loading scenarios. The model simulated using MATLAB and the results are validated using the realtime setup. The outcomes reflects the satisfactory performance interms of enhancing the power quality of the solar-wind systems.

GaN HEMTs are gaining widespread use in power electronics due to their excellent power and frequency characteristics. Accurate and efficient device models are essential for power circuit design and simulation. This article presents a novel approach to model construction, based on current components by their time derivatives. It incorporates advanced static and recurrent neural network concepts to design a precise model architecture. The proposed model integrates the effects of self-heating and couples three distinct current components: a zeroth-order time-derivative intrinsic current module, a first-order time-derivative capacitive current module, and a high-order time-derivative nonideal current module, resulting in a comprehensive GaN HEMT nonlinear dynamic switching output current model. Each model module is trained and validated using experimental measurements. The overall coupled model is further validated through experimental data, and its application is demonstrated in a circuit simulation of a standard double-pulse test with an inductive load. Finally, the proposed model is comprehensively evaluated against conventional models across various metrics, including time-frequency domain accuracy, time-step precision, computational efficiency, and generalization, highlighting the significant advantages of our approach.

The emulation of permanent magnet synchronous machines (PMSMs) is critical for the advancement of power electronics and drive converter testing, particularly within power-hardware-in-the-loop (PHIL) platform. Despite significant progress, and developing accurate machine models, the amount of resources and memory used by these accurate models are not ideal for real-time applications due to added latency. Hence, a research gap exists in developing models that while accurately and efficiently replicate the dynamic behaviors of the machine model under various operating conditions, are light in resource usage. This paper addresses this gap by introducing an artificial neural network (ANN)-based machine modeling approach and combines it with a neuro-fuzzy-based control strategy to ensure robust and precise performance of the system, that is to minimize the error between the electric machine emulators (EME) and physical PMSM test results. The ANN model requires only 0.68 KB of memory compared to the 4 MB needed for traditional 1,000 × 1,000 LUT-based models, which incur greater latency due to cache limitations and interpolation demands despite lower floating-point operation (FLOP) requirements. By using this optimized ANN model with an adaptive ANFIS controller, the proposed system So, the main objective is to enhance the performance and accuracy of EMEs in PHIL testing environments. The ANN model provides a resource-efficient yet precise representation of the PMSM, while the adaptive neuro-fuzzy inference system (ANFIS)-based controller dynamically adjusts its membership functions to adapt to changing system dynamics and loading conditions and provide proper control command.

Micro grids (MGs), comprising diverse micro sources like solar PV, wind generating system, Battery energy storage  system  power electronics, and constant power loads (CPLs), faces stability challenges. Stability analysis is vital during both design and operation. This paper explores a modelling method for CPLs in AC systems. A small-signal state-space model (SSM) of a micro grid is developed, including sub-models for the power controller and network/loads. The model is simulated to investigate PID, artificial neural network (ANN) and P-Q ANN controllers. The study examines the impact of factors such as P-Q gains, CPLs, and line impedance on micro grid stability. A model of an islanded micro grid in MATLAB/SIMULINK is developed and simulated for different time spans. Different case studies like power quality issues, constant power load and irradiance variations are investigated. Finally the Eigen value analysis is done in this work.

Insulated gate bipolar transistor (IGBT), as the most widely used power semiconductor device, its performance will directly determine the performance and stability of the power electronic system. With the use of power electronics in power systems, power semiconductors are more and more widely used in this field. With the growth of the use of time, power semiconductor devices will appear different degrees of decline characteristics, when the aging degree is more serious, the module internal material fatigue accumulation leads to the final failure of the package structure aging, a great threat to the stability of the system. Since material fatigue is difficult to observe directly, and the working environment of power semiconductor devices is complex, there is still a lack of more perfect life assessment strategy. The issues mentioned above were addressed in this paper by adopting a feed-forward neural network algorithm. Multiple characteristic parameters were utilized as the basis for determination, and a life assessment strategy for IGBT modules was designed, excluding the influence of environmental factors such as temperature and working conditions. Experimental verification was conducted for the Infineon FF450R17ME4 module, and a dataset was constructed by collecting feature parameters at different temperatures through offline experiments. This was completed to demonstrate the effectiveness of the evaluation strategy designed in the study.

This paper presents a fault diagnosis method for wireless power transfer (WPT) systems using an enhanced ConvNeXt neural network to achieve accurate and efficient fault identification. Focusing on the LCC-LCC type WPT systems, the proposed approach analyzes module components and their potential faults to construct a comprehensive fault set. Furthermore, both voltage and current data under various fault conditions are generated using Monte Carlo method, accounting for normal component value fluctuations. Then an improved ConvNeXt neural network is employed for fault classification, which offers advantages over traditional power electronics fault diagnosis methods by eliminating the need for additional circuit structures, thus reducing costs and circuit complexity. Compared to conventional approaches like backpropagation neural networks and standard convolutional neural network, the proposed method achieves superior fault identification accuracy. Experimental results demonstrate that the enhanced ConvNeXt model attains a fault diagnosis accuracy of 98%, outperforming some existing techniques.

Harmonic distortion in power systems can lead to inefficiencies, equipment failures and operational risks. Generally, the high order harmonics are introduced in a system when the electricity is controlled by electronics. Traditional detection methods, such as Fourier and wavelet analysis, typically face challenges with real-time detection due to high computational demands. The harmonic measurements are conducted on Power Quality Analyzers, Harmonic Analyzers and numeric meters. This paper proposes a novel neural network-based approach for harmonic distortion detection, utilizing the pattern recognition capabilities of Artificial Neural Networks (ANNs). The proposed method is simple, cost effective and feasible.

The precise modeling of power electronic switches is essential for accurate real-time simulation of power converter systems, particularly under high-frequency and dynamic operating conditions. Existing simulation techniques often rely on ideal switch models or simplified behavioral models, which cannot accurately capture transient behaviors like voltage and current spikes, electromagnetic interference, and switching losses. To address these limitations, this article proposes a novel connectionist neural network-based transient switch modeling approach that integrates time-node coupling mechanisms for enhanced dynamic behavior prediction. A high-quality dataset is generated through extensive offline simulations using a verified physics-based insulated gate bipolar transistor (IGBT) model. The connectionist neural network model leverages feedback connections across time steps, significantly improving its ability to describe transient waveforms under various operating conditions. Additionally, a dynamic neuron adjustment strategy is introduced to reduce hardware resource usage by dynamically allocating neurons based on the complexity of different time nodes. This ensures high modeling accuracy while alleviating computational burden on field programmable gate array platforms.

With the development of electronic technology and artificial intelligence technology, the power consumption of consumer electronics is increasing, such as sweeping robot, dining robot, electric vehicle and so on. And most of these consumer electronics using DC power from lithium batteries, which are organized together in series and parallel for a high power supply. The DC arc occured between connectors and wires is a potential threat to human safety. Lots of researchers and companies study and develop DC arc sensors to detect DC arc faults. Due to the limitations of the DC sensors, the detection range is concentrated in the low-frequency spectrum band(20 kHz–500 kHz). To address this constraint, we propose a neural network-based arc fault detection sensor for DC arc detection. Firstly, we design an arc signal acquisition module based on electromagnetic induction, which automatically matches the sampling frequency and achieves signal amplification. The sampling frequency can reach 4 MHz. Secondly, we propose a Leftover Gated Recurrent Neural Network that extracts sufficient features from the context of current information and performs classification. The test results demonstrate that the model has outstanding accuracy performance, with an improvement of 1.5% over existing models.

This paper proposes a general learning framework to derive topology of power electronics converters. To increase flexibility, a circuit is represented by a graph. A Graph Neural Network extract features of the circuit graph, which is further used in the RL framework. The topology derivation process is regarded as a Markov Decision Process. In each step, the RL agent selects and connects a new block to the initial block until a complete topology is made. To ensure that the derived circuits are feasible, basic circuit constraints are taken into consideration in the reward function. By using this framework, many new six-port, eight-port and ten-port converters are derived. Simulation results show that the derived circuits satisfy given constraints well.

No abstract available

This paper considers learning online (implicit) nonlinear model predictive control (MPC) laws using neural networks and Laguerre functions. Firstly, we parameterize the control sequence of nonlinear MPC using Laguerre functions, which typically yields a smoother control law compared to the original nonlinear MPC law. Secondly, we employ neural networks to learn the coefficients of the Laguerre nonlinear MPC solution, which comes with several benefits, namely the dimension of the learning space is dictated by the number of Laguerre functions and the complete predicted input sequence can be used to learn the coefficients. To mitigate constraints violation for neural approximations of nonlinear MPC, we develop a constraints-informed loss function that penalizes the violation of polytopic state constraints during learning. Box input constraints are handled by using a clamp function in the output layer of the neural network. We demonstrate the effectiveness of the developed framework on a nonlinear buck-boost converter model with sampling rates in the sub-millisecond range, where online nonlinear MPC would not be able to run in real time. The developed constraints-informed neuralLaguerre approximation yields similar performance with longhorizon online nonlinear MPC, but with execution times of a few microseconds, as validated on a field-programmable gate array (FPGA) platform.

The power conversion system based on the modular connection has widespread applications in various power electronic systems. To accurately estimate the state of health without recognizing the systematic mathematical model and to extend the lifetime, this article proposes a lifetime extension approach based on the Levenberg–Marquardt back propagation neural network (LM-BPNN) and power routing of interleaved dc–dc boost conversion systems. The LM-BPNN model is constructed based on the voltage, current, and temperature data generated by the system. On the basis of the trained LM-BPNN, the real-time cumulated damage estimation of each power cell in the conversion system can be accomplished. Applying the power routing concept, the dc–dc boost conversion system allocates different power to the cells according to the cumulated damage of each cell, thereby delaying the failure of cells with higher cumulated damage. Numerical simulation results show that the proposed lifetime extension approach can extend the overall system lifetime. Furthermore, an experimental setup of the interleaved dc–dc boost conversion is constructed to verify the proposed approach, which is of great significance for predictive maintenance in the industrial system.

This article presents a new intelligent sliding-mode control approach to effectively achieve harmonic suppression of an active power filter (APF). An intelligent complementary terminal sliding-mode controller is proposed to improve the control accuracy of current loop and deal with the lumped disturbances in the system. Complementary terminal sliding-mode control combines the features of complementary sliding-mode control and terminal sliding-mode control and does not depend on the accurate dynamic model. In addition, in order to reduce chattering, a multiloop neural network (MLNN), whose parameter learning laws of MLNN are derived based on the Lyapunov laws, is proposed to approximate the unknown nonlinear function term in the APF dynamic model, thus reducing the burden of sliding-mode control. Finally, within the computing power of control board, the detailed simulation and hardware experiments are carried out to demonstrate better harmonic suppression, and steady-state and dynamic performance compared with the existing methods.

Estimation of Remaining Useful Lifetime (RUL) of discrete power electronics is important to enable predictive maintenance and ensure system safety. Conventional data-driven approaches using neural networks have been applied to address this challenge. However, due to ignoring the physical properties of the target RUL function, neural networks can result in unreasonable RUL estimates such as going upwards and wrong endings. In the paper, we apply the fundamental principle of Physics-Informed Neural Network (PINN) to enhance Recurrent Neural Network (RNN) based RUL estimation methods. Through formulating proper constraints into the loss function of neural networks, we demonstrate in our experiments with the NASA IGBT dataset that PINN can make the neural networks trained more realistically and thus achieve performance improvements in estimation error and coefficient of determination. Compared to the baseline vanilla RNN, our physics-informed RNN can improve Mean Squared Error (MSE) of out-of-sample estimation on average by 24.7% in training and by 51.3% in testing; Compared to the baseline Long Short Term Memory (LSTM, a variant of RNN), our physics-informed LSTM can improve MSE of out-of-sample estimation on average by 15.3% in training and 13.9% in testing.

Data-driven approaches are promising to address the modeling issues of modern power electronics-based power systems, due to the black-box feature. Frequency-domain analysis has been applied to address the emerging small-signal oscillation issues caused by converter control interactions. However, the frequency-domain model of a power electronic system is linearized around a specific operating condition. It thus requires measurement or identification of frequency-domain models repeatedly at many operating points (OPs) due to the wide operation range of the power systems, which brings significant computation and data burden. This article addresses this challenge by developing a deep learning approach using multilayer feedforward neural networks (FNNs) to train the frequency-domain impedance model of power electronic systems that is continuous of OP. Distinguished from the prior neural network designs relying on trial-and-error and sufficient data size, this article proposes to design the FNN based on latent features of power electronic systems, i.e., the number of system poles and zeros. To further investigate the impacts of data quantity and quality, learning procedures from a small dataset are developed, and K-medoids clustering based on dynamic time warping is used to reveal insights into multivariable sensitivity, which helps improve the data quality. The proposed approaches for the FNN design and learning have been proven simple, effective, and optimal based on case studies on a power electronic converter, and future prospects in its industrial applications are also discussed.

Power distribution infrastructure is being harmed by the advent of nonlinear devices, which cause harmonics to enter into the power system networks and distort the voltage and current signals. Shunt Active Power Filter (SAPF) is a new power electronics-based technology that can reduce harmonics and improve the power quality in distribution networks. This research provides an efficient and inexpensive strategy to minimizing harmonics and improving the power quality in power distribution networks by employing Shunt Active Power Filter’s (SAPF), which uses the Particle Swarm Optimized Artificial Neural Network Controller (PSO-ANN). The goal of the PSO-ANN algorithms have been developed for SAPF is to improve system performance by lowering the amount of Total Harmonic Distortion (THD). In this work, the standard PI controller is initially tuned using the PSO algorithm to obtain the optimal gain values (Ki, Kp) for the PI controller. After that, these values of the PSO-PI controller's input and output will serve as a dataset for the ANN controller. Now, the PSO algorithm is being used to tune this ANN controller in order to acquire the optimal values for the weight and bias. Using the MATLAB/SIMULINK tool, the proposed algorithm's performance is evaluated and compared to that of a PSO-PI based SAPF and the conventional PI based SAPF. The results of the simulation demonstrate that a SAPF which is based on a PSO-ANN controller is capable of achieving superior THD in the drawing source current while maintaining minimum levels and which are acceptable in accordance with the IEEE-519 standard for harmonics.

The increasing deployment of solar photovoltaic (PV) systems in the electric grid, aimed at addressing the energy crisis and surging power demands, has expanded the potential vulnerability to cyberattacks due to the inter-networking of the grid-connected power electronics converters. In this paper, we propose a dynamic spatiotemporal graph neural network (DST-GNN) for cyberattack detection in grid-tied PV systems. Specifically, to exploit the inherent graph topology of the grid-tied PV system, we start by employing a GNN with a dynamic weighted adjacency matrix to capture the latent spatial correlations within signal data. Then, a one-dimensional convolution neural network (1D-CNN) is utilized to extract the underlying temporal patterns. Notably, we leverage the system dynamics to determine the dynamic graph weights and the number of graph convolution layers, while the hyper-parameters of 1D-CNN are designed based on the periodicity of input signals. Finally, the integration of the priori physical system knowledge further enhances the interpretability and improves the detection performance of DST-GNN. To the best of our knowledge, this is the first work that embeds the grid-tied PV system into a graph structure for cyberattack detection. The effectiveness of DST-GNN is evaluated through comprehensive case studies on a hardware-in-the-loop (HIL) grid-tied PV testbed, and numerical results demonstrate its superiority over baseline methods.

A real-time nonlinear model predictive control using a self-feedback recurrent fuzzy neural network (SFRFNN) estimator for an active power filter is developed to improve the performance of harmonic compensation. First, an SFRFNN with a recurrent structure and fuzzy rules is proposed as a prediction model for nonlinear systems. The SFRFNN merges the advantages of the fuzzy system and the recurrent neural network with a self-feedback structure, which can significantly improve the dynamic performance. Second, the optimization method based on gradient descent is employed to solve the optimal control problem. In addition, the convergence of the proposed SFRFNN and the stability of RT-NMPC are guaranteed using Lyapunov stability theory. Finally, the hardware experiment demonstrated that the proposed method has better performance in both steady and dynamic states compared with existing methods and RT-NMPC using a radial basis function neural network.

This paper proposes a novel Neural Lyapunov method-based transient stability assessment framework for power electronics-interfaced networked microgrids. The assessment framework aims to determine the large-signal stability of the networked microgrids and to characterize the disturbances that can be tolerated by the networked microgrids. The challenge of such assessment is how to construct a behavior-summary function for the nonlinear networked microgrids. By leveraging strong representation power of neural network, the behavior-summary function, i.e., a Neural Lyapunov function, is learned in the state space. A stability region is estimated based on the learned Neural Lyapunov function, and it is used for characterizing disturbances that the networked microgrids can tolerate. The proposed method is tested and validated in a grid-connected microgrid, three networked microgrids with mixed interface dynamics, and the IEEE 123-node feeder. Case studies suggest that the proposed method can address networked microgrids with heterogeneous interface dynamics, and in comparison with conventional methods that are based on quadratic Lyapunov functions, it can characterize the stability regions with much less conservativeness.

There has been an increasing interest in using model predictive control (MPC) for power electronic applications. However, the exponential increase in computational complexity and demand of computing resources hinders the practical adoption of this highly promising control technique. In this article, a new MPC approach using an artificial neural network (termed ANN-MPC) is proposed to overcome these barriers. A power converter with a virtual MPC controller is first designed and operated under a circuit simulation or power hardware-in-the-loop simulation environment. An artificial neural network (ANN) is then trained offline with the input and output data of the virtual MPC controller. Next, an actual FPGA-based MPC controller is designed using the trained ANN instead of relying on heavy-duty mathematical computation to control the actual operation of the power converter in real time. The ANN-MPC approach can significantly reduce the computing need and allow the use of more accurate high-order system models due to the simple mathematical expression of ANN. Furthermore, the ANN-MPC approach can retain the robustness for system parameter uncertainties by flexibly setting the input elements. The basic concept, ANN structure, offline training method, and online operation of ANN-MPC are described in detail. The computing resource requirement of the ANN-MPC and conventional MPC are analyzed and compared. The ANN-MPC concept is validated by both simulation and experimental results on two kW-class flying capacitor multilevel converters. It is demonstrated that the FPGA-based ANN-MPC controller can significantly reduce the FPGA resource requirement (e.g., 2.11 times fewer slice LUTs and 2.06 times fewer DSPs) while offering a control performance same as the conventional MPC.

The most attractive renewable energy resource that provides clean electricity via solar PV panels is solar irradiation received from the Sun. Solar energy is available mainly during the daytime, but solar photovoltaic (PV) panels can produce maximum power due to low efficiency. Hence, maximum power point tracking (MPPT) methods are used with solar PV systems. The interface required between solar PV panels and the load is a DC-DC converter, a power electronics device. This paper proposes a neural network-based PID controller for the boost converter. The well-known back prorogation neural network algorithm is used with PID structure to design a controller for the boost converter. The study is carried out with the help of MATLAB/Simulink software to show the test results. The simulations’ outcome shows the proposed controller’s efficacy when used with solar PV systems.

Abstract The rapid integration of renewable and decentralized energy sources is transforming the operation of power networks, leading to challenges in efficient electricity utilization, optimal harnessing of renewable energy, and maintaining stable voltage levels. To tackle these issues, clusters of multi-renewable energy systems are interconnected through electrical converters employing fuzzy controllers. Power quality (PQ) controllers have been developed for various renewable energy sources to meet the growing energy demands. Enhancing power electronics separation based on power storage and distributed generation offers significant opportunities for managing PQ. This research focuses on improving the performance of smart hybrid multi-renewable energy systems using microgrids through the utilization of artificial neural networks (ANNs). The proposed approach incorporates a fuzzy controller-based pulse generation technique for a DC-to-AC inverter, ensuring efficient power conversion while maintaining PQ factors. The fuzzy controller effectively handles stringent constraints on process parameters by leveraging adjustable variables. Hardware evaluation and performance analysis were conducted using a peripheral interface controller (PIC) controller, providing an accurate representation of the results.

This study develops an intelligent global sliding mode control using recurrent feature selection neural network for active power filter (APF). First, the dynamic model of an APF is constructed. Second, a conventional global sliding mode control (GSMC) is introduced to achieve the aim to track the quick changing reference signal for an APF current control strategy. Since uncertain parameters of APF are unavailable in advance, high performance current control cannot be assured in practical applications. In this article, to improve conventional GSMC for APF, the recurrent feature selection neural network (RFSNN) is proposed to learn uncertain function. Unlike the classical neural network, RFSNN can select useful network parameters and delete unfavorable network parameters to adjust the structure and parameters of the neural networks. Based on Lyapunov stability analysis, the online learning laws for network parameters are derived to satisfy the control objectives. Finally, the superiority and robustness of the proposed GSMC using RFSNN are verified by detailed experimental results.

In partially shaded photovoltaic (PV) arrays, the power-voltage curve displays multiple maxima, one of which is the global maximum power point (GMPP). One of the major concerns related to PV systems is to locate and track the GMPP in all circumstances to boost efficiency. Combining conventional hill climbing (HC) algorithm and artificial neural networks (ANNs), a new two-stage GMPP tracking method is introduced in this article that aims to be fast and accurate. Moreover, it does not require irradiance or temperature sensors. In the first stage, the current-voltage (I-V) curve is sampled at specific points determined based on the array I-V curve analysis with the objective to be the minimum samples possible that can reflect the changes of irradiance and temperature. Then a simple feedforward ANN is employed to estimate the neighborhood of the GMPP using these samples. In the second stage, the HC algorithm is adopted to ensure the GMPP is tracked precisely. The proposed method is validated by simulations in MATLAB/Simulink environment and experimental tests under uniform irradiance condition, partial shading conditions, and a wide range of temperatures.

The rapid development of wind energy conversion systems (WECS) brings new trends and challenges in the wind energy market. Recently, a new paradigm in control engineering known as the networked control system has emerged as a promising alternative to conventional control systems, where deep learning techniques can be effectively utilized to enhance system performance and adaptability. This paper proposes a multi-agent deep reinforcement learning (DRL) based approach to self-tune the rotor side and grid side converter controllers of PMSG-based WECS, where traditional proportional-integral-derivative controllers are replaced with twin delayed deep deterministic policy gradient-based agents. The proposed decentralised online DRL approach does not require tuning expertise to control the system effectively. The performance of the multi-agent DRL technique is investigated under different operating conditions and results are compared with the PID controller to see the effectiveness of the proposed technique.

In recent years, there has been a growing interest in using model-free deep reinforcement learning (DRL)-based controllers as an alternative approach to improve the dynamic behavior, efficiency, and other aspects of DC–DC power electronic converters, which are traditionally controlled based on small signal models. These conventional controllers often fail to self-adapt to various uncertainties and disturbances. This paper presents a design methodology using proximal policy optimization (PPO), a widely recognized and efficient DRL algorithm, to make near-optimal decisions for real buck converters operating in both continuous conduction mode (CCM) and discontinuous conduction mode (DCM) while handling resistive and inductive loads. Challenges associated with delays in real-time systems are identified. Key innovations include a chattering-reduction reward function, engineering of input features, and optimization of neural network architecture, which improve voltage regulation, ensure smoother operation, and optimize the computational cost of the neural network. The experimental and simulation results demonstrate the robustness and efficiency of the controller in real scenarios. The findings are believed to make significant contributions to the application of DRL controllers in real-time scenarios, providing guidelines and a starting point for designing controllers using the same method in this or other power electronic converter topologies.

Deep Learning (DL) networks have demonstrated exceptional capability in handling complex system dynamics but have traditionally been constrained by the computational limitations of DSPs lacking parallel processing capabilities. With the emergence of next-generation DSPs equipped with Neural Processing Units (NPUs), DL networks can now complement traditional linear controllers by addressing the nonlinear characteristics of power electronic systems in real time. This study proposes an online-learning DL-assisted 2 Poles − 2 Zeros (2P2Z) controller for a boost converter, aiming to enhance transient and steady-state performance across varying operating conditions. In particular, the proposed approach mitigates the adverse effects of the Right-Hand-Plane (RHP) zero, improving system stability and dynamic response. The effectiveness of the design is validated through continuous learning simulations after loading the pretrained model, demonstrating its capability to adapt and optimize control parameters dynamically.

Converter-based power plants will become the primary source of electrical energy in the future. These types of power plants need to be enhanced in terms of their resilience as well as increasing their share. This article proposes an improved combined control method aimed at enhancing resilience. The method focuses on solving grid-connected converter challenges under unbalanced disturbed grid voltage conditions. These conditions are more commonly observed in converter-based microgrids due to structural and operational issues, such as weak grids, un-transposed lines, single-phase loads, etc. The current references in the proposed method are calculated to minimize their dependency on measured grid voltage and system parameters as much as possible. Also, the proposed integrated method is designed to handle severe load changes and symmetrical voltage sags. Moreover, a resilient control system should be capable of interacting with the system and learning new information from events in accordance with the resilience definitions. Thus, the proposed method incorporates reinforcement learning as an added value to enhance the level of resilience in a more comprehensive and effective manner. Simulation and experimental results demonstrate that the proposed method reduces current THD by about 35%, decreases switching losses by 50%, and eliminates overshoots under dynamic conditions compared to conventional methods. These findings confirm the effectiveness and superiority of the proposed method in improving the resilience and performance of GCCs.

Considering the evolution of future microgrids (MGs) towards zero-inertia level due to the penetrations of distributed converter-based resources (DCRs), a large number of data produced by these generations will lead the control decisions to be more complicated than conventional power systems. This paper presents a control strategy for a zero-inertia MG with DCRs using a robust deep learning neural network (RDeNN). In a training phase, a sub-space state-based identification method is employed to monitor and analyze the data regarding stability indices, i.e., damping and frequency of dominant modes, and robustness against uncertainties. In addition, a mixed $H_{2}/H_{\infty }$ control strategy is applied to enhance the training efficacy in the frequency and voltage control loops of DCRs. The trained RDeNN is activated to make quick and effective control decisions by using only measured signals from the MG. Simulation results are verified in the zero-inertia MG (or the grid with 100% DCRs) and compared with several existing control techniques. The study results demonstrate the advantages of the proposed RDeNN in many aspects such as low computational time, require-less physical controller models, fast and flexible stabilizing responses, and high robustness against various time delays, data quality issues, and MG uncertainties.

A DC microgrid’s tightly regulated DC/DC converter encounters significant challenges in voltage stability, primarily due to the negative incremental resistance of constant power loads (CPLs). Conventional controllers often struggle with load variations and changes in system parameters. Therefore, there has been growing interest in adaptive machine learning algorithms, such as Deep Reinforcement Learning (DRL), to improve voltage regulation. This paper presents an end-to-end DRL framework based on a modified Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm. The framework is designed to directly control power switches for regulating the voltage of a DC/DC buck converter that supplies power to CPLs. Real-time experiments were conducted using OPAL-RT to validate the approach under diverse load cycles and converter parameter changes.Comparative analysis against other DRL-based control strategies, including Deep Q-learning (DQN) and Deep Deterministic Policy Gradient (DDPG), demonstrated the superior static and dynamic voltage response of the proposed modified TD3 DRL controller, particularly in scenarios involving load and parametric variations.

This paper presents a deep reinforcement learning (DRL) agent for duty ratio control of the dual active bridge (DAB) converter using triple phase shift (TPS) modulation. The operation of a dual active bridge under the TPS modulation technique was briefly discussed, stating its advantage over the other phase shift techniques, thus creating the requirement of a model-free control approach for the converter. The reinforcement learning method has evolved as the deep reinforcement learning method that can solve complex problems with rising computational accuracy. This paper focuses on a deep deterministic policy gradient (DDPG) agent founded on actor-critic policy, i.e. implemented as the DRL controller. A detailed discussion about the control strategy using the DRL controller was presented. The model was simulated using MATLAB/Simulink software version 2024b to verify the voltage regulation of the dual active bridge using the DDPG agent. Finally, the conclusion presents the advantages of using a DRL controller for the TPS modulation of the DAB converter.

DC microgrids are emerging as a promising solution for easy and efficient energy distribution in various industrial and residential applications. However, they experience voltage perturbations due to the variability of renewable energy sources, load fluctuations, and instability problems caused by constant power loads (CPL). DC microgrids may encounter voltage instability without effective voltage regulation, potentially resulting in facility damage or even power outages. Therefore, it is essential to develop a controller that copes with requirements of robustness, steady-state, and dynamic performance. This study presents a comparative analysis of three deep reinforcement learning agents for voltage control of buck converter feeding CPL. The three agents we compare include double deep Q-network (DDQN), our improved multilayered proximal policy optimization (PPO), and twin delay deep deterministic policy gradient (TD3) agents.

This paper proposes a novel Deep Reinforcement Learning (DRL) method for controlling a 23-level Single DC Source Hybrid Packed U-Cell (HPUC) converter. The HPUC topology generates a high number of voltage levels with minimal components but presents control challenges due to its numerous switching states and dynamic charging behavior. Unlike conventional control methods, which require accurate models and are sensitive to noise and parameter mismatches, DRL offers a model-free and resilient approach to the non-linear control of such complex systems. A Deep Q-Network (DQN) agent which is inherently model-free and suited for high-dimensional state spaces and discrete action spaces, is employed to address these issues. To validate the proposed method, simulations were conducted in the MATLAB/Simulink environment. The obtained results demonstrated the satisfactory performance of the proposed DRL method, achieving a Total Harmonic Distortion (THD) of 2.71% in the output current under steady-state, maintaining stable capacitor voltage balancing, and exhibiting rapid dynamic response (e.g., settling within approximately 40 ms for current step changes). Furthermore, its resilience was highlighted by its ability to maintain control despite a 25dB SNR noise condition and up to 15% variations in capacitor values.

In order to improve dynamic performance and robustness, this paper suggests a novel control framework for a modified Y-source DC–DC converter using a Real-Time Deep Reinforcement Learning (RTDRL) controller. High voltage gain, low switching stress, and a broad duty-cycle control range make the modified Y-source topology ideal for renewable energy applications that need effective power conversion under a range of load conditions. In nonlinear systems with inherent delays, traditional controllers like Model Predictive Control (MPC) and Proportional-Integral (PI) have trouble adapting and performing well. We overcome these constraints by implementing an entropy-regularized RTDRL algorithm that is based on the Soft Actor-Critic (SAC) method but has been adjusted to take into consideration time delays that are common in hardware that is based on digital signal processors (DSPs). Real-time stability is made possible by this delay-aware reinforcement learning technique, which incorporates historical action-state correlations into the policy.In order to effectively compensate for control delays in real-time environments, the augmented state incorporates not only the current observations but also previous system states and control actions.

For frequency and voltage stability control of grid-forming converters in high-power electronic scenarios, this paper proposes a grid-forming converter grid-connection stability control strategy based on passive control and deep reinforcement learning. Firstly, the virtual synchronous generator (VSG) is written in the port-Hamiltonian form to clarify the interconnection and dissipative structure, and the achievable passive control law is obtained by energy shaping and damping injection. Then, DDPG is introduced to adjust the damping parameters online, so that the control has adaptive ability under multiple working conditions, and the closed-loop system is proved to be asymptotically stable based on Lyapunov function. Finally, the simulation example analysis is carried out. In the simulation of power mutation, voltage imbalance, short-circuit fault and load change, this method significantly reduces the overshoot and adjustment time compared with VSG-PI and fixed parameter PBC, and improves the steady-state error and energy dissipation rate. The simulation results verify the effectiveness of the combination of physical consistency and strategy adaptation.

To address the insufficient modeling accuracy and slow dynamic response of the Capacitor-Inductor-Inductor-Capacitor (CLLC) resonant converter in Vehicle-to-Grid (V2G) automotive onboard charger, a deep learning-based model predictive control method for the post-stage V2G onboard charger is proposed. This approach utilizes deep learning techniques to perform steady-state implicit modeling of the CLLC resonant converter, enabling accurate calculation of voltage gain across a diverse array of switching frequencies and varying load conditions. Based on this model, a deep learning model predictive control (DL-MPC) strategy is developed. The DL-MPC method leverages deep learning-based voltage gain predictions for direct multi-step control sequence prediction. Simulations demonstrate that the proposed DL-MPC strategy outperforms conventional proportional-integral (PI) and model predictive control (MPC) control, effectively suppressing voltage overshoot and undershoot, achieving reductions of up to 60.9%, and reducing recovery time by up to 55.6% under load transients and reference voltage step changes, thereby enhancing dynamic response and stability. Experimental validation that DL-MPC achieves a 65.21% reduction in voltage dips and a 31.25% faster recovery time compared to PI control under load transients. Simulation and experimental results consistently show that the deep-learning-based modeling attains high fidelity, while the proposed DL-MPC strategy delivers superior control performance for the CLLC resonant converter, yielding a markedly improved dynamic response.

DC-DC boost converters play a crucial role in modern power electronics applications, particularly in renewable energy systems, electric vehicles, and portable electronic devices. These applications require efficient and reliable control strategies to ensure optimal performance. This paper explores the application of the Twin Delayed Deep Deterministic Policy Gradient (TD3) reinforcement learning algorithm to enhance system performance under nonlinearities, dynamic load changes, and input voltage fluctuations. The proposed TD3 - based controller is compared with the conventional proportional-integral (PI) controller. Simulation results demonstrate the superiority of TD3 in achieving a faster transient response and reduced overshoot. This work highlights the potential of reinforcement learning in power electronics, which, due to its adaptive nature, can provide improved performance over time.

No abstract available

No abstract available

Machine learning and deep learning tools are gaining popularity in electrical engineering applications. These are now being used to improve the machine's efficiency, accuracy, control, and monitoring. A deep learning-based single open switch fault diagnosis and fault tolerant strategy for matrix converter-based induction motor drive is proposed in this paper. This paper proposes the detection of open switch faults and identifies the faulty switch in real-time based on a neural network-based curve fitting technique called Bayesian regularization and furthermore replaces the faulty switch using a spare switch. To achieve this, the three-phase output currents and their rms values are measured. Then neural network is used to map this input data with a set of numeric targets. Based on its output, the faulty switch is determined in real-time. Simulation of Space Vector Modulation (SVM) based Matrix Converter fed induction motor with a defective bidirectional switch is carried out in MATLAB-Simulink environment, and the neural network-based fault detection and fault tolerant system detects the open switch fault and identifies the faulty switch in 20 ms, and the fault has been successfully cleared by fault-tolerant strategy after 50 ms of fault occurrence.

The imminent depletion of oil resources and increasing environmental pollution have driven the use of clean energy, particularly wind energy. However, wind turbines (WTs) face significant challenges, such as critical component failures, which can cause unexpected shutdowns and affect energy production. To address this challenge, we analyzed the Supervisory Control and Data Acquisition (SCADA) data to identify significant differences between the relationship of variables based on data reconstruction errors between actual and predicted values. This study proposes a hybrid short- and long-term memory autoencoder model with multihead self-attention (LSTM-MA-AE) for WT converter fault detection. The proposed model identifies anomalies in the data by comparing the reconstruction errors of the variables involved. However, more is needed. To address this model limitation, we developed a fault prediction system that employs an adaptive threshold with an Exponentially Weighted Moving Average (EWMA) and a fixed threshold. This system analyzes the anomalies of several variables and generates fault warnings in advance time. Thus, we propose an outlier detection method through data preprocessing and unsupervised learning, using SCADA data collected from a wind farm located in complex terrain, including real faults in the converter. The LSTM-MA-AE is shown to be able to predict the converter failure 3.3 months in advance, and with an F1 greater than 90% in the tests performed. The results provide evidence of the potential of the proposed model to improve converter fault diagnosis with SCADA data in complex environments, highlighting its ability to increase the reliability and efficiency of WTs.

This paper presents a model-free control approach using deep reinforcement learning(DRL) which aims at optimizing the performance of wave energy converter(WEC). Several researchers commonly utilize a linear WEC model, neglecting the inherent non-linear static friction of the WEC system. This oversight may result in a decline in control performance. Consequently, static friction is deliberately included as a modeling error in the control framework to compare the resilience between model-based reactive control (RC) and DRL method against disturbances. To further address the challenge of real-time wave information being difficult to acquire in practical applications of the DRL method, a novel DRL controller is proposed with no wave information. The simulations are conducted to compare the proposed DRL controller with the RC approach based on a nonlinear WEC system in MATLAB/Simulink. It is shown that when considering modeling errors of static friction force, the DRL-based controller can outperform reactive control in both regular and irregular wave conditions - by 20.0% and 6.2%. Preliminary experimental validation of the DRL controller has been obtained through wave tank testing in regular wave conditions, showing satisfactory performance and consistency with simulation results. This is the first successful practice of running the RL algorithm on a practical controller and proves the feasibility of the proposed approach.

This paper presents a power flow control strategy based on deep reinforcement learning (DRL) for triple active bridge (TAB) converter. The DRL-based agent is designed to control TAB converter instead of conventional decoupling controller. A wide range of voltage and power operation conditions are used to train the agent offline, where deep deterministic policy gradient (DDPG) algorithm is employed to find appropriate shift phase angle with the constraint of minimizing transmission power errors. After training process, the well-trained agent is adopted to achieve decoupled power control online for TAB converter. Simulation and experiment results are presented to verify the effectiveness of the proposed DRL-based power control strategy. This approach enables multidirectional power flow control without decoupled matrix and look-up table, while it exhibits robust performance in response to input voltage variations.

The recently proposed three-phase dual-purpose dc-ac/dc power converter (PC) as the new member of the universal converter family shows high potential for facilitating the integration of renewable energy sources (RESs) and energy storage systems (ESS) in any type of grid (dc or ac), aiming at high performance and versatility with minimal redundancy or modification. The proposed dual-purpose PC requires flexible, adaptive, reliable and scalable control strategies that align with the goal of providing a standardized solution for the power conversion stages in the coexisting ac and dc microgrids (MGs). In this paper, a model-free deep reinforcement learning (DRL)-based current control methodology is proposed for that dual-purpose PC. The non-linear and model-free nature of the proposed approach reduces the system sensitivity to parameter variations, increasing the system robustness and reliability; and can potentially reduce the development and operational costs since a detailed modeling and an extensive parameter tuning are avoided. Two DRL agents were trained by means of the twin delayed deep deterministic (TD3) policy gradient algorithm for both dc-ac and dc-dc modes of operation. Simulation results are provided to demonstrate the performance of the proposed control strategy.

This paper investigates the control structure without a current sensor in a boost DC-DC converter used in a DC Nano-grid, including a photovoltaic (PV) array. Instead of the current sensor, a PV array output current prediction system based on a deep neural network is used. There are three main factors in determining the accuracy of the discussed neural network: the number and type of data required for neural network training, neural network architecture, and its training algorithm. The input data will be processed and standardized in the discussed system before being applied to the neural network. Multi-layer feedforward architecture has been used to build the neural network model, consisting of one input layer including nine important features, six hidden layers, and one output layer. Adam's optimization algorithm is also used to minimize the Loss function. All forecasting steps, including data processing, neural network model building, and its training, have been done in the Jupyter lab environment of Python programming language. The obtained results indicate the system's successful performance, and the r2_score evaluation criterion, which is used to measure the accuracy of the prediction, has been calculated and provides acceptable results.

To improve the conversion efficiency of the dual-active-bridge converter, this article demonstrates a variable-frequency triple-phase-shift (TPS) control strategy with the help of the deep reinforcement learning method. More specifically, the twin delayed deep deterministic policy gradient (TD3) algorithm is adopted to train the agent offline with the aim of minimum power losses, under the TPS modulation with varying switching frequency. Moreover, the zero-voltage-switching performance has been considered during the training of the TD3 algorithm. Based on these, the trained TD3 agent acts as a fast surrogate predictor, which can produce appropriate control strategies in real-time for whole continuous operating conditions with soft switching and maximum conversion efficiency. The effectiveness and correctness of the proposed scheme is validated through the experimental results in a laboratory prototype.

The reinforcement learning (RL) control approach with application to power electronics systems has become an emerging topic, while the sim-to-real issue remains a challenging problem as very few results can be referred to in the literature. Indeed, due to the inevitable mismatch between simulation models and real-life systems, offline-trained RL control strategies may sustain unexpected hurdles in practical implementation during the transfer procedure. In this article, a transfer methodology via a delicately designed duty ratio mapping is proposed for a dc–dc buck converter. Then, a detailed sim-to-real process is presented to enable the implementation of a model-free deep reinforcement learning controller. As the main contribution of this article, the proposed methodology is able to endow the control system to achieve: 1) voltage regulation and 2) adaptability and optimization abilities in the presence of uncertain circuit parameters and various working conditions. The feasibility and efficacy of the proposed methodology are demonstrated by comparative experimental studies.

No abstract available

The CLLC DC-DC converter offers highly efficient DC voltage conversion for the highly renewable penetrated system. The modeling significantly influences the performance of CLLC DC-DC converters. However, both the current FHA-based and time-domain analysis methods are incapable to incorporate different switching frequencies and load conditions. Moreover, due to the inevitable non-monotonic voltage gain, the PI controller with these models leads to the uncontrolled steady-state voltage deviation. As a novel modeling method, deep learning implicitly builds extensively accurate and arbitrary mappings, which can solve the aforementioned modeling problems. This paper proposes an adaptive steady-state modeling method for the CLLC converter based on deep learning. Precise voltage gain can be provided over a wide range of switching frequencies and load conditions. Besides, a fast control strategy based on the proposed modeling method is developed. This strategy searches the optimal operating point using particle swarm optimization, which rapidly adjusts voltage gain and suppresses the steady-state voltage deviation even in non-monotonic situations. Finally, a 400V/300V, 2.4kW SiC-based CLLC converter prototype with distributed heterogeneous controllers is implemented to verify the proposed methods. The experimental results show the accuracy of the proposed adaptive modeling and the effectiveness of the fast control strategy. The proposed method improves the stability and extends the operating range of the CLLC converter, which benefits the development of highly renewable penetrated systems.

This paper presents a novel model-free switching and control method for three-level neutral point clamped (NPC) converter using deep reinforcement learning (DRL). Our approach targets two primary control objectives: voltage balancing and current control. In this method, voltage balancing, current control and selection of optimal switches are achieved using a reward function which is calculated based on various signals measured as observations of the DRL agent. Since the action space is discrete, a deep Q-network (DQN) agent is utilized. DQN is used due to its capability of handling high-dimensional state spaces. In order to highlight its pros and cons, the proposed method is compared with model predictive control (MPC), which is another popular non-linear control method for power electronic converters. The proposed method is evaluated and compared with the MPC method in grid-connected mode using simulations in Matlab/Simulink. To evaluate the practical performance of the DRL method, various experimental results are obtained. The simulation and experimental results demonstrate that the proposed method effectively achieves accurate voltage balancing and ensures steady operation even in the presence of various dynamic changes, including variations in the reference currents and grid voltage. Additionally, the method successfully handles uncertainties, such as sensor measurement noise, and accommodates parameter variations, such as changes in the capacity of the DC-link capacitors and line impedance. The results demonstrate that this method exhibits superior adaptability to real-time changes and uncertainties, delivering more robust performance compared to similar conventional methods like MPC. Thus, this method can be considered a promising approach for intelligent control of power electronic converters, especially when conventional methods such as MPC face challenges in performance and accuracy under severe parameter variations and uncertainties.

No abstract available

The CLLC DC-DC resonant converter offers improved efficiency and power density. Advanced control techniques are required to utilize the converter fully. However, the existing model based on the fundamental harmonic approximation, accurate only around the resonant frequency, hinders their realization. Moreover, due to the high system order of the CLLC converter, many of the successful modeling and control experiences for lower-order resonant converters cannot be reproduced. Driven by recent theoretical advances in deep learning, this article proposes a steady-state modeling method based on the multilayer neural network. It is accurate over a wide operating range. This article further proposes a model predictive control strategy for the CLLC converter to improve the dynamic response under step reference variation based on the new model. Finally, comparisons between the proposed control method and the PID controller are carried out through simulation. The results verify the effectiveness of the model predictive control strategy and the accuracy of the new modeling approach for the CLLC converter.

Growth in renewable energy systems, direct current (DC) microgrids, and the adoption of electric vehicles (EVs) will substantially increase the demand for bi-directional converters. Precise control mechanisms are essential to ensure optimal performance and better efficiency of these converters. This paper proposes a deep neural network (DNN)-based controller designed to precisely control bi-directional converters for vehicle-to-vehicle (V2V) and vehicle-to-grid (V2G) applications. This control technique allows the converter to quickly attain new reference values, enhancing performance and efficiency by significantly reducing the overshoot duration. To train the DNN controller, large synthetic data are used by performing simulations for various sets of conditions, and the results are validated with a hardware setup. The real-time performance of the DNN controller is compared with a conventional proportional–integral (PI)-based controller through simulated results using MATLAB Simulink (version 2023a) and with a real-time setup. The converter attains a new reference of about 975 μs with the proposed control technique. In contrast, the PI controller takes about 220 ms, which shows that the proposed control technique is far better than the PI controller.

No abstract available

Deep reinforcement learning-based algorithms exhibit significant potential in developing robust model-free control systems for the next power converter generation. This work presents a control strategy based on a deep reinforcement learning (DRL) framework to operate an Active Front-End (AFE). The research's originality lies in finding an optimal control policy that leverages DRL's capabilities to enhance the AFE control performance, all without prior information regarding power converter dynamics and parameters. Moreover, the control strategy is designed to ensure the adaptability of the converter across diverse operational scenarios. To this end, multiple intelligent agents are developed, trained, tested, and validated using the AFE converter dynamics. Simulated results demonstrated that the proposed control methodology exhibits robustness, effectively handling uncertainties associated with the converter. Also, the empirical findings reveal that the proposed control strategy presents a solid performance in the current control and DC-link voltage control tasks, with a maximum Total Harmonic Distortion of 4.25% for 10 kHz sampling frequency.

In the DC microgrid, a DC-DC converter has now been widely used as a conventional switching power supply. A suitable control system needs to be developed for a DC-DC converter to obtain the best dynamic response which is very much essential theoretically and in real-world applications since the variability of renewable energy supply is uncertain. A unique intelligent control method is proposed for buck converters to constant power loads (CPLs). In this work, the bus voltage stability issues arising in DC microgrids are addressed using deep reinforcement learning (DRL) by employing a Markov decision process (MDP). To adapt the agent-environment interaction via reward/penalties mechanism for the convergence of nominal bus voltage, a model-free DRL control method is suggested. The high-dimensional characteristic is collected by agent from complex systems which does not require any previous knowledge in order to create approximations. Lastly, the outcomes of the simulation analysis demonstrate that under various conditions, the recommended controller has a higher ability for self-improvement and learning.

A data-driven integrated active fault-tolerant control (IAFT) strategy for controlling the solid oxide fuel cell (SOFC) output voltage is proposed, which maintains satisfactory dynamic performance and eliminates constraint violations in the event of system failure. In addition, this article introduces an efficient replay deep meta-deterministic policy gradient (ER-DMDPG) for IAFTs, which combines priority experience replay and meta-learning techniques to improve the robustness and multitask cooperative learning capability of the IAFTs. The algorithm combines the controllers of the fuel reformer and direct current-direct current (dc-dc) converter into a single independent agent, which is trained by a cooperative meta-learner and a base learner to achieve multiobjective optimal active fault-tolerant control (FTC). It is experimentally demonstrated that the proposed method can maintain better dynamic performance and prevent constraint violations of fuel utilization across a wide range of working conditions.

No abstract available

For the 5G base transceiver stations (BTSs), the effective stabilization of full-bridge (FB) converters is necessary to supply the connected loads without any interruption. The stability challenges of such technologies are more intensified when the 5G BTS supplies constant power loads (CPL) with negative impedance instabilities. To meet this need, this brief presents an adaptive interval type-3 fuzzy logic system (IT3-FLS) employing deep reinforcement learning (DRL) for the efficient voltage stabilization of 5G-telecom power system (5G-TPS) supplying CPL. The Hardware-in-the-Loop (HiL) examinations are accomplished using an OPAL-RT platform to test the usefulness of the adaptive IT3-FLS from a systematic perspective.

No abstract available

This paper proposes an IGBT open-circuit fault diagnosis method that can maintain high accuracy under diverse operation conditions and circuit parameter variances. Different aspects of uncertainties are analyzed in the component parameters, operation conditions, and measurement errors of a three-phase inverter case study. A lightweight Convolutional Neural Network (CNN) is applied based on an obtained dataset covering a wide range of inverter operation scenarios and uncertainties. The comparisons with benchmarked conventional fault diagnosis method and with different machine learning methods are presented. The results verify the improved accuracy in open-circuit diagnosis considering complex operation conditions and meanwhile with reduced detection time in certain scenarios.

While data-driven methods start to be applied to fault diagnosis of power converters, there are still some limitations: (1) feature extraction relies on expert experience, (2) the model trained in one system cannot be applied to another different system, and (3) abundant fault data is difficult to obtain in practical applications. To address them, a transferable deep learning network for insulated bipolar gate transistor (IGBT) open-circuit fault diagnosis is proposed in three-phase inverters. First, the lightweight convolutional neural network (CNN) is constructed to automatically extract features from the original current signals and complete the operation condition identification. Then, the designed network is pre-trained with data from the source domain (simulation model). After that, a transfer learning strategy is designed to fine-tune the network by using a few data samples in the target domain using real-time hardware in the loop. Both simulation and hardware-in-the-loop results demonstrate the effectiveness of the proposed method with 99.52% and 98.30% diagnostic accuracy, respectively.

In this paper, a fault diagnosis method based on multi-branch one-dimensional convolutional neural network (1D-CNN) is proposed for the open-circuit fault of the inverter IGBT and current sensor in permanent magnet synchronous motor (PMSM) drive system. Firstly, to avoid adding additional sensors, a three-phase current acts as a highly responsive signal for fault detection. Secondly, a multi-branch 1D-CNN diagnosis model is built to solve the complexity and uncertainty problems caused by traditional artificial feature extraction and selection. Then, the residual connection is introduced into the diagnostic model to alleviate the disappearance of the network gradient and accelerate the convergence speed. Finally, the three-phase current is directly input into the model for training and classification. The test results show that the method can identify the open-circuit fault of the IGBT and current sensor within the sample length of a fundamental cycle, and the accuracy rate is 99.84%. Compared with other traditional machine learning methods, it has been proven to have higher diagnostic accuracy and good anti-noise ability.

This research introduces an enhanced fault detection approach, variational mode decomposition (VMD), for identifying open-circuit IGBT faults in the grid-side converter (GSC) of a doubly fed induction generator (DFIG) wind turbine system. VMD has many advantages over other decomposition methods, notably for non-stationary signals and noise. VMD’s robustness stems from its ability to decompose a signal into intrinsic mode functions (IMFs) with well-defined centre frequencies and bandwidths. The proposed methodology integrates VMD with a hybrid convolutional neural network–long short-term memory (CNN-LSTM) architecture to efficiently extract and learn distinctive temporal and spectral properties from three-phase current sources. Ten operational scenarios with a wind speed range of 5–16 m/s were simulated using a comprehensive MATLAB/Simulink version R2022b model, including one healthy condition and nine unique IGBT failure conditions. The obtained current signals were decomposed via VMD to extract essential frequency components, which were normalised and utilised as input sequences for deep learning models. A comparative comparison of CNN-LSTM and CNN-only classifiers revealed that the CNN-LSTM model attained the greatest classification accuracy of 88.00%, exhibiting enhanced precision and resilience in noisy and dynamic environments. These findings emphasise the efficiency of integrating advanced signal decomposition with deep sequential learning for real-time, high-precision fault identification in wind turbine power electronic converters.

The inverter is the most vulnerable module of photovoltaic (PV) systems. The insulated gate bipolar transistor (IGBT) is the core part of inverters and the root source of PV inverter failures. How to effectively diagnose the IGBT faults is critical for reliability, high efficiency, and safety of PV systems. Recently, deep learning (DL) methods are widely used for fault detection and diagnosis. Different from traditional diagnosis methods, DL methods use deep neural networks which can automatically extract the useful representative features from raw data. However, DL methods require large amounts of data, which leads to the high cost of communication, storage, and computation. To tackle these issues, a data-driven fault detection and diagnosis method for IGBT open-circuit faults based on compressed sensing (CS) and convolutional neural networks (CNN) is proposed in this paper. CS is adopted to compress raw signals, and the optimal value of compression ratio (CR) is determined by considering the trade-off between classification accuracy and model training time. The overlap sampling method is adopted for data segmentation. Meanwhile, overlap sampling can also increase the number of training samples and improve the sample correlation. The compressed signals are segmented and reconstructed into two-dimensional feature maps for model training. Finally, compared with CNN of the same structure, the developed CS-CNN model can compress 85% of data without accuracy loss. The performance comparison with the state-of-the-art networks demonstrates that the test accuracy is 98.68% and the model training time is much shorter than other methods.

In photovoltaic (PV) systems, machine learning-based methods have been used for fault detection and diagnosis in the past years, which require large amounts of data. However, fault types in a single PV station are usually insufficient in practice. Due to insufficient and non-identically distributed data, packet loss and privacy concerns, it is difficult to train a model for diagnosing all fault types. To address these issues, in this paper, we propose a decentralized federated learning (FL)-based fault diagnosis method for insulated gate bipolar transistor (IGBT) open-circuits in PV inverters. All PV stations use the convolutional neural network (CNN) to train local diagnosis models. By aggregating neighboring model parameters, each PV station benefits from the fault diagnosis knowledge learned from neighbors and achieves diagnosing all fault types without sharing original data. Extensive experiments are conducted in terms of non-identical data distributions, various transmission channel conditions and whether to use the FL framework. The results are as follows: 1) Using data with non-identical distributions, the collaboratively trained model diagnoses faults accurately and robustly; 2) The continuous transmission and aggregation of model parameters in multiple rounds make it possible to obtain ideal training results even in the presence of packet loss; 3) The proposed method allows each PV station to diagnose all fault types without original data sharing, which protects data privacy.

In various applications, the reliable and efficient detection of faults in electric machines is crucial, particularly in environments with high noise levels. To this end, the current study introduces an effective fault detection model utilizing the differential of Short-Time Fourier Transform (STFT) and a channel-wise regulated Convolutional Neural Network (CNN). The novel use of the differential of STFT is presented to enhance the diagnostic model's performance in noisy conditions compared with the conventional STFT. According to the inherent time-frequency domain information within the differential of STFT, a regulated CNN-based model is proposed to integrate spatio-temporal information into the feature map, thereby enhancing accuracy and reducing the computational demand. The method is evaluated on three datasets: the widely used Case Western Reserve University (CWRU) benchmark featuring bearing fault and vibration measurements, a dataset involving Permanent Magnet Synchronous Motor (PMSM) data with varying levels of Inter-Turn Short-Circuit (ITSC) fault and current measurements, and a dataset consisting of a mixture of mechanical and electrical faults. Comparative analysis highlights the superior performance of the proposed model over existing robust methods in the literature under both normal and noisy conditions.

No abstract available

No abstract available

No abstract available

Scheduled maintenance and condition monitoring of power transformers in smart grids is mandatory to reduce their downtimes and maintain economic benefits. However, to minimize energy losses during inspection, non-invasive fault diagnosis techniques such as thermogram imaging can enable continuous monitoring of transformer health with minimal out-of-service time. Deep learning (DL) has proven to be a fast and efficient intelligent diagnostic tool. In this paper, a DL-based thermography method is proposed called Trans-Light for transformers’ interturn faults detection and short-circuit severity identification. Trans-light extracts deep features from two deep layers of a convolutional neural network (CNN) rather than depending on one layer, thus obtaining more intricate patterns. Moreover, a Dual-tree Complex Wavelet Transform method is adopted which offers two enhancements. First, it acquires time–frequency knowledge besides the already obtained spatial information and second, it reduces the huge deep features dimensionality. Trans-light combines extracted deep features, then a feature selection process is applied to further reduce features’ size, thus decreasing computation burden and reducing classification and training time. To validate the proposed scheme’s diagnosis performance and robustness, different combinations of two CNN models, two feature selection methods, and six classifiers were tested, applying the proposed Trans-light framework, under noise-free and noise-existing conditions. Experimental results indicated that the combination of the LDA classifier, applied with the ResNet-18 CNN model and trained with merged deep features undergoing the chi-square (χ2) selection approach, attained superior performance under noise-free conditions. Compared to its counterparts in previous work, this configuration outperforms their performance since it uses the fewest features’ number yet maintains 100% classification accuracy. Besides, it attained robust performance under two different noise natures again with minimal features’ dimension, thus minimizing computational load and implementation complexity.

The global increase in the adoption of photovoltaic (PV) energy accentuates the imperative of maintaining system efficiency amidst environmental variabilities and faults. The processes of identifying, classifying, and rectifying defects are critical for ensuring the long-term sustainability and performance integrity of PV installations. This article introduces an innovative ensemble convolutional neural network (CNN) model that employs weighted feature fusion to enhance accuracy beyond what is achievable with a singular CNN architecture. By utilizing three proficient CNNs—VGG16, ResNet, and MobileNet—the fusion of deep features extracted from the last layers of these networks’ augments performance, while also capitalizing on the integration of data from multiple CNNs with distinct configurations. This methodology was applied to a publicly available infrared thermography imaging dataset, which includes 12 distinct defects. The proposed models have been subsequently trained, validated, and tested on this dataset. The outcomes indicate a substantial enhancement in the accuracy of defect classification compared to individual CNN models, with an average accuracy of 96%. This approach underscores its utility in defect identification, particularly demonstrating the capacity of the ensemble CNN to classify defects with high precision

From the perspectives of theoretical design and practical application, the existing fault diagnosis methods with the complex identification process owing to manual feature extraction and the insufficient feature extraction for time series data and weak fault signal is not suitable for AC/DC microgrids. Thus, this paper proposes a fault diagnosis method that integrates a convolutional neural network (CNN) with a long short-term memory (LSTM) network and attention mechanisms. The method employs a multi-scale convolution-based weight layer (Weight Layer 1) to extract features of faults from different dimensions, performing feature fusion to enrich the fault characteristics of the AC/DC microgrid. Additionally, a hybrid attention block-based weight layer (Weight Layer 2) is designed to enable the model to adaptively focus on the most significant features, thereby improving the extraction and utilization of critical information, which enhances both classification accuracy and model generalization. By cascading LSTM layers, the model effectively captures temporal dependencies within the features, allowing the model to extract critical information from the temporal evolution of electrical signals, thus enhancing both classification accuracy and robustness. Simulation results indicate that the proposed method achieves a classification accuracy of up to 99.5%, with fault identification accuracy for noisy signals under 10 dB noise interference reaching 92.5%, demonstrating strong noise immunity.

No abstract available

Developing novel fluoroether electrolytes with high-voltage stability is an effective strategy to improve the performance of lithium metal batteries (LMB). However, the vast chemical space of fluoroether is underexplored due to the absence of effective tools to evaluate the potential used in high-voltage LMB. Herein, a framework was developed in combination of Voting ensemble algorithms and graph convolution neural network (GCNN), allowing the fast assessment of oxidative stability of non-aqueous liquid electrolytes, synthesizability of solvents as well as the solvation ability of them to dissolve lithium salts. Potential fluoroether solvent candidates for high-voltage LMB were screened out from a virtual library comprising 5576 electrolytes constructed by a combination of 1510 solvents and 4 salts. Among them, two fluorinated ethers, 1,1,1,3,3,3-hexafluoro-2-(2-methoxyethoxy) propane and 7,7,8,8-tetrafluoro-3,12-dimethoxy-2,5,10,13-tetraoxatetradecane, were successfully synthesized and showed satisfactory high-voltage stability, sufficient solvation ability and satisfactory cycling with almost 99.5% coulombic efficiency in Li||NMC811 full cell. This work provided an efficient framework for the discovery of solvents with high-voltage tolerance in a vast structural space prior to experimental synthesis, accelerating the development of advanced electrolyte for high-energy-density rechargeable batteries.

Developing reliable emerging photovoltaic (e‐PV) technologies requires high‐throughput material discovery, device design, and processing optimization. However, the effective process of the resulting high‐dimensional, multivariate datasets remains a significant challenge. Integrating feature selection methods and machine learning (ML) provides a robust solution to reduce data dimensionality, improve predictive accuracy, and uncover material performance mechanisms. This review summarizes the advancements in synergizing feature selection methods, particularly the maximum relevance minimum redundancy (mRMR) method embedded, with Gaussian process regression (GPR) to advance e‐PVs research. It highlights the importance of integrating feature selection with ML and high‐throughput experimentation (HTE) frameworks to accelerate material screening, optimize manufacturing processes, and predict stability. Additionally, the review discusses key challenges such as data quality and model scalability and offers promising strategies to address these limitations. This data‐driven approach offers a systematic pathway toward the accelerated discovery and optimization of e‐PV technologies.

Low-frequency disturbances of power quality are one of the most common disturbances in the power grid. These disturbances are most often the result of the impact of power electronic and energy-saving devices, the number of which is increasing significantly in the power grid. Due to the simultaneous operation of various types of loads in the power grid, various types of simultaneous disturbances of power quality occur, such as voltage fluctuations and distortions. Therefore, there is a need to analyze this type of simultaneous interaction. For this purpose, a special and complementary laboratory setup has been prepared, which allows for the examination of actual states occurring in modern power networks. Selected research results are presented for this laboratory setup, which determine its basic properties. Possible applications and possibilities of the laboratory setup are presented from the point of view of current challenges.

A three-phase isolated AC-DC-DC power supply is widely used in the industrial field due to its attractive features such as high-power density, modularity for easy expansion and electrical isolation. In high-power application scenarios, it can be realized by multiple AC-DC-DC modules with Input-Parallel Output-Parallel (IPOP) mode. However, it has the problems of slow output voltage response and large ripple in some special applications, such as electrophoresis and electroplating. This paper investigates an improved Adaptive Linear Active Disturbance Rejection Control (A-LADRC) with flexible adjustment capability of the bandwidth parameter value for the high-power DC supply to improve the output voltage response speed. To reduce the DC supply ripple, a control strategy is designed for a single module to adaptively adjust the duty cycle compensation according to the output feedback value. When multiple modules are connected in parallel, a Hierarchical Delay Current Sharing Control (HDCSC) strategy for centralized controllers is proposed to make the peaks and valleys of different modules offset each other. Finally, the proposed method is verified by designing a 42V/12000A high-power DC supply, and the results demonstrate that the proposed method is effective in improving the system output voltage response speed and reducing the voltage ripple, which has significant practical engineering application value.

Linear Quadratic Regulator (LQR) is often combined with feedback linearization (FBL) for nonlinear systems that have the nonlinearity additive to the input. Conventional approaches estimate and cancel the nonlinearity based on the first principle or data-driven methods such as Gaussian Processes (GPs). However, the former needs an elaborate modeling process, and the latter provides a fixed learned model, which may be suffering when the model dynamics are changing. In this letter, we take a Deep Neural Network (DNN) using a real-time-updated dataset to approximate the unknown nonlinearity while the controller is running. Spectrally normalizing the weights in each time-step, we stably incorporate the DNN prediction to an LQR controller and compensate for the nonlinear term. Leveraging the property of the bounded Lipschitz constant of the DNN, we provide theoretical analysis and locally exponential stability of the proposed controller. Simulation results show that our controller significantly outperforms Baseline controllers in trajectory tracking cases.

Controlling nonlinear dynamics is a long-standing problem in engineering. Harnessing known physical information to accelerate or constrain stochastic learning pursues a new paradigm of scientific machine learning. By linearizing nonlinear systems, traditional control methods cannot learn nonlinear features from chaotic data for use in control. Here, we introduce Physics-Informed Deep Operator Control (PIDOC), and by encoding the control signal and initial position into the losses of a physics-informed neural network (PINN), the nonlinear system is forced to exhibit the desired trajectory given the control signal. PIDOC receives signals as physics commands and learns from the chaotic data output from the nonlinear van der Pol system, where the output of the PINN is the control. Applied to a benchmark problem, PIDOC successfully implements control with higher stochasticity for higher-order terms. PIDOC has also been proven to be capable of converging to different desired trajectories based on case studies. Initial positions slightly affect the control accuracy at the beginning stage yet do not change the overall control quality. For highly nonlinear systems, PIDOC is not able to execute control with high accuracy compared with the benchmark problem. The depth and width of the neural network structure do not greatly change the convergence of PIDOC based on case studies of van der Pol systems with low and high nonlinearities. Surprisingly, enlarging the control signal does not help to improve the control quality. The proposed framework can potentially be applied to many nonlinear systems for nonlinear controls.

Reinforcement Learning and the Evolutionary Strategy are two major approaches in addressing complicated control problems. Both are strong contenders and have their own devotee communities. Both groups have been very active in developing new advances in their own domain and devising, in recent years, leading-edge techniques to address complex continuous control tasks. Here, in the context of Deep Reinforcement Learning, we formulate a parallelized version of the Proximal Policy Optimization method and a Deep Deterministic Policy Gradient method. Moreover, we conduct a thorough comparison between the state-of-the-art techniques in both camps fro continuous control; evolutionary methods and Deep Reinforcement Learning methods. The results show there is no consistent winner.

In this study, the open-circuit faults diagnosis and location issue of the neutral-point-clamped (NPC) inverters are analysed. A novel fault diagnosis approach based on knowledge driven and data driven was presented for the open-circuit faults in insulated-gate bipolar transistors (IGBTs) of NPC inverter, and Concordia transform (knowledge driven) and random forests (RFs) technique (data driven) are employed to improve the robustness performance of the fault diagnosis classifier. First, the fault feature data of AC in either normal state or open-circuit faults states of NPC inverter are analysed and extracted. Second, the Concordia transform is used to process the fault samples, and it has been verified that the slopes of current trajectories are not affected by different loads in this study, which can help the proposed method to reduce overdependence on fault data. Moreover, then the transformed fault samples are adopted to train the RFs fault diagnosis classifier, and the fault diagnosis results show that the classification accuracy and robustness performance of the fault diagnosis classifier are improved. Finally, the diagnosis results of online fault diagnosis experiments show that the proposed classifier can locate the open-circuit fault of IGBTs in NPC inverter under the conditions of different loads.

Three-phase PWM rectifiers are adopted extensively in industry because of their excellent properties and potential advantages. However, while the IGBT has an open-circuit fault, the system does not crash suddenly, the performance will be reduced for instance voltages fluctuation and current harmonics. A fault diagnosis method based on deep feedforward network with transient synthetic features is proposed to reduce the dependence on the fault mathematical models in this paper, which mainly uses the transient phase current to train the deep feedforward network classifier. Firstly, the features of fault phase current are analyzed in this paper. Secondly, the historical fault data after feature synthesis is employed to train the deep feedforward network classifier, and the average fault diagnosis accuracy can reach 97.85% for transient synthetic fault data, the classifier trained by the transient synthetic features obtained more than 1% gain in performance compared with original transient features. Finally, the online fault diagnosis experiments show that the method can accurately locate the fault IGBTs, and the final diagnosis result is determined by multiple groups results, which has the ability to increase the accuracy and reliability of the diagnosis results. (c) 2020 ISA. Published by Elsevier Ltd. All rights reserved.

A three-phase pulse-width modulation (PWM) rectifier can usually maintain operation when open-circuit faults occur in insulated-gate bipolar transistors (IGBTs), which will lead the system to be unstable and unsafe. Aiming at this problem, based on random forests with transient synthetic features, a data-driven online fault diagnosis method is proposed to locate the open-circuit faults of IGBTs timely and effectively in this study. Firstly, by analysing the open-circuit fault features of IGBTs in the three-phase PWM rectifier, it is found that the occurrence of the fault features is related to the fault location and time, and the fault features do not always appear immediately with the occurrence of the fault. Secondly, different data-driven fault diagnosis methods are compared and evaluated, the performance of random forests algorithm is better than that of support vector machine or artificial neural networks. Meanwhile, the accuracy of fault diagnosis classifier trained by transient synthetic features is higher than that trained by original features. Also, the random forests fault diagnosis classifier trained by multiplicative features is the best with fault diagnosis accuracy can reach 98.32%. Finally, the online fault diagnosis experiments are carried out and the results demonstrate the effectiveness of the proposed method, which can accurately locate the open-circuit faults in IGBTs while ensuring system safety.

A fault diagnosis method for power electronics converters based on deep feedforward network and wavelet compression is proposed in this paper. The transient historical data after wavelet compression are used to realize the training of fault diagnosis classifier. Firstly, the correlation analysis of the voltage or current data running in various fault states is performed to remove the redundant features and the sampling point. Secondly, the wavelet transform is used to remove the redundant data of the features, and then the training sample data is greatly compressed. The deep feedforward network is trained by the low frequency component of the features, while the training speed is greatly accelerated. The average accuracy of fault diagnosis classifier can reach over 97%. Finally, the fault diagnosis classifier is tested, and final diagnosis result is determined by multiple-groups transient data, by which the reliability of diagnosis results is improved. The experimental result proves that the classifier has strong generalization ability and can accurately locate the open-circuit faults in IGBTs.

Considering the models that apply the contextual information of time-series data could improve the fault diagnosis performance, some neural network structures such as RNN, LSTM, and GRU were proposed to model the fault diagnosis effectively. However, these models are restricted by their serial computation and hence cannot achieve high diagnostic efficiency. Also the parallel CNN is difficult to implement fault diagnosis in an efficient way because it requires larger convolution kernels or deep structure to achieve long-term feature extraction capabilities. Besides, BERT model applies absolute position embedding to introduce contextual information to the model, which would bring noise to the raw data and therefore cannot be applied to fault diagnosis directly. In order to address the above problems, a fault diagnosis model named deep parallel time-series relation network(DPTRN) has been proposed in this paper. There are mainly three advantages for DPTRN: (1) Our proposed time relationship unit is based on full multilayer perceptron(MLP) structure, therefore, DPTRN performs fault diagnosis in a parallel way and improves computing efficiency significantly. (2) By improving the absolute position embedding, our novel decoupling position embedding unit could be applied on the fault diagnosis directly and learn contextual information. (3) Our proposed DPTRN has obvious advantage in feature interpretability. We confirm the effect of the proposed method on four datasets, and the results show the effectiveness, efficiency and interpretability of the proposed DPTRN model.

The worm gearbox is a high-speed transmission system that plays a vital role in various industries. Therefore it becomes necessary to develop a robust fault diagnosis scheme for worm gearbox. Due to advancements in sensor technology, researchers from academia and industries prefer deep learning models for fault diagnosis purposes. The optimal selection of hyperparameters (HPs) of deep learning models plays a significant role in stable performance. Existing methods mainly focused on manual tunning of these parameters, which is a troublesome process and sometimes leads to inaccurate results. Thus, exploring more sophisticated methods to optimize the HPs automatically is important. In this work, a novel optimization, i.e. amended gorilla troop optimization (AGTO), has been proposed to make the convolutional neural network (CNN) adaptive for extracting the features to identify the worm gearbox defects. Initially, the vibration and acoustic signals are converted into 2D images by the Morlet wavelet function. Then, the initial model of CNN is developed by setting hyperparameters. Further, the search space of each Hp is identified and optimized by the developed AGTO algorithm. The classification accuracy has been evaluated by AGTO-CNN, which is further validated by the confusion matrix. The performance of the developed model has also been compared with other models. The AGTO algorithm is examined on twenty-three classical benchmark functions and the Wilcoxon test which demonstrates the effectiveness and dominance of the developed optimization algorithm. The results obtained suggested that the AGTO-CNN has the highest diagnostic accuracy and is stable and robust while diagnosing the worm gearbox.

Fault diagnosis of rotating machinery is an important engineering problem. In recent years, fault diagnosis methods based on the Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) have been mature, but Transformer has not been widely used in the field of fault diagnosis. To address these deficiencies, a new method based on the Time Series Transformer (TST) is proposed to recognize the fault mode of bearings. In this paper, our contributions include: Firstly, we designed a tokens sequences generation method which can handle data in 1D format, namely time series tokenizer. Then, the TST combining time series tokenizer and Transformer was introduced. Furthermore, the test results on the given dataset show that the proposed method has better fault identification capability than the traditional CNN and RNN models. Secondly, through the experiments, the effect of structural hyperparameters such as subsequence length and embedding dimension on fault diagnosis performance, computational complexity and parameters number of the TST is analyzed in detail. The influence laws of some hyperparameters are obtained. Finally, via t-Distributed Stochastic Neighbor Embedding (t-SNE) dimensionality reduction method, the feature vectors in the embedding space are visualized. On this basis, the working pattern of TST has been explained to a certain extent. Moreover, by analyzing the distribution form of the feature vectors, we find that compared with the traditional CNN and RNN models, the feature vectors extracted by the method in this paper show the best intra-class compactness and inter-class separability. These results further demonstrate the effectiveness of the proposed method.

We introduce the Fourier Learning Machine (FLM), a neural network (NN) architecture designed to represent a multidimensional nonharmonic Fourier series. The FLM uses a simple feedforward structure with cosine activation functions to learn the frequencies, amplitudes, and phase shifts of the series as trainable parameters. This design allows the model to create a problem-specific spectral basis adaptable to both periodic and nonperiodic functions. Unlike previous Fourier-inspired NN models, the FLM is the first architecture able to represent a multidimensional Fourier series with a complete set of basis functions in separable form, doing so by using a standard Multilayer Perceptron-like architecture. A one-to-one correspondence between the Fourier coefficients and amplitudes and phase-shifts is demonstrated, allowing for the translation between a full, separable basis form and the cosine phase-shifted one. Additionally, we evaluate the performance of FLMs on several scientific computing problems, including benchmark Partial Differential Equations (PDEs) and a family of Optimal Control Problems (OCPs). Computational experiments show that the performance of FLMs is comparable, and often superior, to that of established architectures like SIREN and vanilla feedforward NNs.