Reinforcement Learning-Based Optimization of Fast Charging Strategy for Electric Vehicle Batteries to Extend Cycle Life

Abstract

To address the problems of intensified polarization and excessively rapid cycle life degradation of electric vehicle batteries caused by the traditional constant current-constant voltage (CC-CV) fast charging strategy, an optimized reinforcement learning fast charging strategy based on Deep Q-Network (DQN) is proposed. Firstly, a second-order RC equivalent circuit model and an SEI film growth aging model are constructed to accurately characterize the electrochemical properties and life degradation mechanism of batteries during the charging process. Secondly, a reinforcement learning framework is designed, with the battery state-of-charge (SOC), voltage, temperature, and polarization voltage as the state space, the charging current step as the action space, and a multi-objective reward function considering both fast charging efficiency and life protection is established. Finally, experimental verification is carried out on 18650 ternary lithium batteries, comparing with the traditional CC-CV strategy and PID adaptive strategy. The results show that the proposed strategy controls the fast charging time at 42.8 minutes (SOC from 20% to 80%), which is 17.2% shorter than the CC-CV strategy; the capacity retention rate reaches 86.3% after 1000 cycles, which is 21.5% higher than the CC-CV strategy and 10.8% higher than the PID strategy; the peak polarization voltage is reduced by 34.7%, effectively suppressing the battery aging rate. This method realizes the dynamic balance between fast charging speed and cycle life, providing a new path for the engineering application of electric vehicle battery fast charging technology.