Soft Actor Critic Algorithm Based Reinforcement Learning Controller for Boost Converter

Abstract

ABSTRACT Boost converters are one of the most commonly used converters in power electronics due to their advantages, such as simple structure and low‐cost implementation. Due to these issues, many papers have been published about the control method of this topology. This paper proposes a reinforcement learning (RL) controller based on the soft actor‐critic (SAC) algorithm for boost converters. This approach aims at decreasing the limitation of conventional controllers, such as long settling times, overshoot, undershoot and instability. It also removes the need for a system model, which is often necessary in traditional control methods. Furthermore, the proposed method addresses the challenges of control under dynamic conditions, which is a problem that has not been sufficiently explored in prior reinforcement learning (RL) based controllers. The proposed approach has demonstrated promising performance in both steady‐state and dynamic regulation across a wide range of operational variations. Simulation and experimental results show the superiority of the SAC algorithm over the conventional ones without relying on the system model.