MSCC-DRL: Multi-Stage constant current based on deep reinforcement learning for fast charging of lithium ion battery

Abstract

The growing electric vehicle market has made extreme fast charging (XFC) a critical challenge for the lithium-ion battery community. With a goal of achieving a 15-minute recharge time or less, the research community has developed several protocols for fast charging, relying on accurate electrochemical models and mathematical optimization algorithms. However, the effectiveness of such techniques is severely limited by their high computational costs as well as the uncertainty and complexity of the mathematical algorithms used. To address these limitations, this paper proposes a model-free multistage constant current technique based on deep reinforcement learning (DRL) while ensuring battery safety and reducing the battery degradation rate with the multi-stage charging profile. The proposed technique uses a Proximal Policy Optimization (PPO) algorithm to train the DRL agent to learn a reliable control policy through interaction with the battery, aided by a comprehensive reward function. The proposed model is validated through a comparison with a 6CC-CV profile for fast charging. Furthermore, the robustness of the proposed DRL model against different electrode thicknesses and porosity is also evaluated. Simulation results demonstrate that the proposed DRL framework can charge the battery in 14 min for the thickest cathode and 6 min for the thinnest cathode while maintaining battery safety.