A Neural Network Direct Control of STATCOM: Design and Experimental Validation

Neural Network-Assisted Model Predictive Control of a Modular Multilevel Converter

This paper describes an advanced algorithm for current control in a modular multilevel power converter. The proposed algorithm is based on deadbeat model predictive control, which is further enhanced by applying an artificial neural network. Rather than supplanting the existing predictive controller, the neural network is used to supplement it, thereby improving control performance. The reference current tracking capability is enhanced under all operating conditions, particularly in cases where the reactive elements of the converter exhibit a relatively high resistive component. By introducing the neural network, the inherent delay of model predictive control in such scenarios is virtually eliminated, which reduces the overall control error. The network is trained offline on a set of control inputs obtained by optimizing the current response in a wide range of operating modes. Algorithm verification is performed for a three-phase modular multilevel converter in different steady-state and transient conditions. A state-of-the-art high-fidelity real-time simulator is used in both the algorithm development and verification stages.

Design of Parallel Hybrid Active Power Filter with Adaptive DC-Link Voltage Control Based on Artificial Neural Network

In this study, an Artificial Neural Network (ANN)-based adaptive DC-link voltage (Vdc) controller is developed for a Parallel Hybrid Active Power Filter (PHAPF). The proposed controller aims to simultaneously determine the DC-link reference voltage (Vdc_ref) and the PI controller gains (Ki, Kp) as a function of the operating conditions. Training data for the ANN are obtained from simulations performed in the MATLAB/Simulink environment. Simulations performed using this training set show that adapting the DC-link reference voltage reduces total harmonic distortion (THD) compared to a PHAPF with a fixed Vdc_ref and reduces the DC-link voltage at low-power loads, which has the potential to lower switching losses, while adaptive PI gains improve transient behavior after large load changes. Therefore, the two adaptive quantities affect complementary aspects of performance. The trained ANN model is coded in the C programming language and implemented on a microcontroller-based control card. A 5 kVA PHAPF system is designed and fabricated for experimental verification. Experimental results demonstrate that the proposed ANN-based adaptive DC-link voltage control algorithm achieves lower total harmonic distortion (THD) than PHAPFs employing a constant Vdc_ref. In addition, reducing the DC-link voltage under low-power operating conditions has the potential to decrease voltage stress across the power switches and reduce switching losses. Furthermore, the proposed ANN-based adaptive DC-link voltage control algorithm exhibits better harmonic suppression performance despite the processing load and filtering delays.

Artificial Neural Network-Based Control Strategy Implemented on a Multiport DC-DC Converter in a DC Microgrid

This paper proposes an artificial neural network (ANN)-based control strategy for the bidirectional dual-input single-output (BDISO) DC-DC converter. The approach aims to combine the robustness of nonlinear control with the computational efficiency of data-driven implementation. A supertwisting sliding mode controller (STSMC) is first employed offline to generate representative state-action datasets that capture the converter's nonlinear dynamics under both buck and boost modes. The extracted state features and corresponding optimal switching actions are then used to train an ANN, enabling it to approximate the STSMC control law. During online operation, the trained ANN directly produces low-latency switching commands for voltage regulation, thereby eliminating the need for real-time STSMC computation. The proposed controller is evaluated in MATLAB/Simulink and benchmarked against conventional proportional-integral (PI) and fixed-frequency sliding mode control (FSMC) methods. Simulation results demonstrate that the ANN-based controller achieves superior transient response, precise voltage regulation, and reduced chattering, confirming its potential as a compact and efficient solution for multi-port DC-DC converter applications.

Neural network control and optimization for fuel cell hydrogen supply system

Simulation and Real‐Time Implementation for Power Quality Enhancement Systems Using Adaptive Neural Network Strategy

To address the requirements of current tracking control and power quality improvement for active power filters (APF), this paper proposes a nested terminal sliding mode control scheme based on hippocampal neural network to overcome the performance limitations of existing APF control methods. First, the circuit structure of the APF is elaborated, and the mathematical model including lumped system uncertainties is derived. Then, a nested terminal sliding mode surface is designed to ensure that the tracking error converges to zero in finite time, which achieves performance improvement compared with traditional linear sliding mode control that can only realize asymptotic convergence. Afterward, a hippocampal‐inspired neural network that mimics the human hippocampal information processing mechanism is introduced for the first time to learn the unknown nonlinear terms in the sliding mode controller, effectively weakening the adverse effects of system uncertainties on control performance. A novel feature selection mechanism is proposed to extract and process key information within the network, greatly reducing the network computational overhead. A dual‐loop structure is designed in the neural network to improve the processing efficiency of time‐varying harmonic signals, and the online adaptive update law of network parameters is derived based on Lyapunov theorem to guarantee system stability. Finally, simulation and hardware experimental results verify the effectiveness of the proposed algorithm. This method reduces the total harmonic distortion (THD) of the grid source current to 1.29% in simulation and 3.03% in experiment. Compared with mainstream methods, it exhibits excellent current tracking ability, strong robustness, and superior grid harmonic suppression performance.

Neural network-based energy-input shaping combined control for anti-sway and positioning of crane systems under non-zero initial conditions

Portal bridge cranes, as typical underactuated systems, have wide industrial applications. However, in practical operations, the initial swing angle and angular velocity of the crane inevitably exist, which degrades the performance of these methods in real-world applications. This paper proposes a control method based on the combination of energy control, input shaping, and neural network, in order to solve the problem of anti-swing and positioning of portal bridge cranes under non-zero initial conditions. Through the energy control method based on Lyapunov theory, the initial swing angle is quickly suppressed, combined with the optimized four pulse input shaping method to ensure the positioning accuracy of the trolley, and the neural network is used to adaptively adjust the control parameters, which overcomes the problems of sensitivity to initial conditions and insufficient adaptability of the traditional methods. Numerical simulations and physical experiments show the effectiveness of the hybrid method suggested.