Optimal fast charging of lithium-ion batteries through continual hybrid model learning

Towards the goal of addressing the critical challenge of extended charging times for lithium-ion batteries (LiBs), this study introduces a novel learning-based fast charging control framework that optimizes charging schedules throughout the LiB's lifespan. This is achieved by first continually learning a virtual hybrid model, which is then utilized to generate data via latent imagination for fast charging policy training with deep reinforcement learning (DRL). Unlike traditional heuristic methods, which are often conservative, or purely physics model-based approaches that struggle to capture the complex dynamics of LiB operation and degradation, our hybrid model continuously adapts with operational data, enabling the generation of customized fast charging policies as the battery degrades. Through high-fidelity simulations and comparisons with standard CCCV charging protocols, we find that the proposed framework achieves a significant charging speed improvement at different ambient temperatures and cooling efforts while ensuring battery health. Copyright (C) 2024 The Authors. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/)