PWM rectifier AI control for wind energy conversion system based on a dual star synchronous generator.

Abstract

This work investigates the integration of reinforcement learning into the control architecture of a Dual-Star Synchronous Generator coupled to a PWM rectifier and connected to the electrical grid. Two complementary approaches are studied and implemented. The first approach consists of directly replacing the conventional proportional-integral controller in the inner quadrature-axis current loop of the PWM rectifier with a Deep Deterministic Policy Gradient agent, which learns to generate the control voltage directly from system observations through interaction with a Simulink environment. The second approach preserves the PI controller structure while employing a Twin Delayed Deep Deterministic Policy Gradient agent as a supervisory layer that automatically tunes the proportional and integral gains according to the current operating conditions. This adaptive tuning strategy is applied to both the PWM rectifier current controller and the grid-side inverter current controllers. For the grid integration stage, the conventional control strategy based on Clarke and Park transformations, Phase Locked Loop synchronization, and dq-frame PI current control is first implemented and validated. The TD3 agent is then introduced to replace manual gain selection, and a time-weighted reward function is specifically designed to encourage fast and accurate regulation of the rectifier output voltage toward a 300 V reference. The results demonstrate that both reinforcement learning approaches achieve performance comparable to well-tuned conventional controllers, confirming the feasibility of integrating AI-based adaptive control into classical power electronics architectures. This work paves the way for significant future research. Future work should focus on experimental validation through real test bench implementation, exploration of advanced control strategies using machine learning approaches, and investigating the use of these multilevel inverter topologies for grid-connected renewable energy applications, addressing the key challenges associated with grid synchronization, power quality, and compliance with grid codes.