A Predictive Control Method Based on Neural Predictor and Soft Actor–Critic for Power Converters

Abstract

This article focuses on introducing soft reinforcement learning (RL) techniques into the finite control-set model predictive control (FCS-MPC) framework to enhance robust performance. More precisely, building upon a neural predictor, an intelligent agent trained using the soft actor–critic algorithm is developed to explore the optimal control input embedded within the MPC framework. Meanwhile, during training, a constraint based on a Lyapunov function is introduced, and a corresponding update law for the weights is provided. Furthermore, this proposed method guarantees the stability of the system integrated with the RL intelligent agent. Finally, both simulation and experimental results validate the superiority of this approach compared to the existing FCS-MPC methods.