Modulation strategy design method for single-inductor multi-port converter based on reinforcement learning

Abstract

Single-inductor multi-port converters have the characteristics of "more silicon and less magnetism" and have great application potential in many fields. However, single-inductor multi-port converters have many switching modes and the modulation strategy design is complex. The current design method is to manually select the switching mode sequence and perform modal analysis. The design process requires power electronics expertise and experience. This paper proposes a modulation strategy design method for single-inductor multi-port converters based on reinforcement learning. This method uses a neural network to generate the modulation strategy and adopts a set of simple rules to provide rewards for training the neural network. Through reinforcement learning, the neural network can summarize experience in trial and error without human intervention and finally generate the optimal modulation strategy. This method has many advantages: (1) It only requires known conditions such as port voltage and converter structure as input, and some simple constraints as rewards, avoiding complicated manual design; (2) It can use a neural network to generate the optimal modulation strategy under different operating conditions, and has strong adaptability. Based on this method, a modulation strategy design for a single-inductor multi-port converter is carried out, and the effectiveness of the method is verified by experiments.