Neural Networks-Based Adaptive Predictive Current Control of a Grid-Connected Three-Phase PWM Converter

This article presents a new technique for optimizing predictive current control of three-phase pulse-width modulation converters sensitive to variations in input filter parameters, namely resistance and inductance. These variations often lead to power ripples and current harmonics. An online estimation method is proposed based on adaptive linear neurons. This technique incorporates an adaptive neural network identifier and an adaptive neural network filter to refine the estimation even in the presence of harmonic distortions. Furthermore, to enhance the robustness of the estimator, a recursive least squares algorithm is used for weight updating. Simulation results demonstrate the efficiency and superiority of the proposed method compared to conventional approaches in terms of response time and ripple reduction.