Deep Reinforcement Learning Based Adaptive Duty Cycle Optimization of KY Converter for Intelligent Electric Vehicle Battery Management Systems

Abstract

Efficient and adaptive control of DC–DC converters is critical for improving charging performance and operational reliability in electric vehicle battery management systems. Conventional proportional–integral–derivative (PID) controllers exhibit limited adaptability under dynamic load variations and nonlinear battery characteristics. This study proposes a deep reinforcement learning (DRL)–based duty cycle optimization strategy for a KY converter used in EV charging applications. A custom simulation environment was developed, and a Proximal Policy Optimization (PPO) agent was trained using Monte Carlo multi-run validation to learn optimal control actions under varying operating conditions. The proposed AI-based controller was evaluated against a conventional PID baseline using statistical performance metrics. Results demonstrated consistent improvement in cumulative reward, enhanced voltage regulation stability, and reduced ripple influence across multiple independent runs. Statistical validation confirmed the robustness and repeatability of the DRL approach. The findings indicate that reinforcement learning–based adaptive control can improve energy efficiency and charging reliability, offering practical potential for intelligent industrial battery management systems and next-generation smart EV charging infrastructure.