Enhancing Power Converter Efficiency with Reinforcement Learning-Based PWM Control

Authors:

Abstract
This research paper explores the use of reinforcement learning (RL) to automatically generate control policies for power converters. The goal is to improve electrical efficiency, reduce electromagnetic interference (EMI) emissions, and enhance the robustness of the converters. The RL agents are designed to learn a key task of power converters, specifically optimized pulse-width modulation (PWM), using interpretable policies. In practical learning scenarios, the agents can utilize batch data generated through interaction with a simulation model under different operating conditions. To evaluate the effectiveness of the learned control policies, they are deployed to a prototyping board for functional testing and feasibility analysis. The results indicate that the learned controllers can accomplish multiple objectives and require less deployment effort compared to model-predictive methods, which demand substantial calibration expertise. Furthermore, the learned controllers demonstrate competitive performance in terms of electrical losses, EMI emissions, and adaptation to changes in operating points when compared to traditional converter control methods.