An overview of reinforcement learning for power electronic converters: Topology derivation, parameter design, and control implementation

Power electronic converters, as core devices in modern smart grids, face three major technical challenges in topology derivation, parameter design, and control implementation. The discrepancy between traditional manual solutions and the demands for intelligent operation is increasingly evident, primarily manifested in topology optimization relying on trial and error, time-consuming parameter tuning, and a lack of adaptability in control strategies under complex operating conditions. Reinforcement learning (RL)–based artificial intelligence methods, with their strong self-adaptability, multi-objective cooperative optimization potential, and intelligent control capabilities for complex nonlinear systems, offer a new pathway to address these issues. However, there remains a lack of systematic literature reviews on the application of RL algorithms in power electronic converters, which poses significant challenges for related research. To address this gap, this paper classifies and organizes the mainstream RL algorithms applied in power electronic converters, summarizes their specific applications in topology derivation, parameter design, and control implementation, and finally provides an outlook on the prospects of RL in this field.