Optimal Charging Method for Effective Li-ion Battery Life Extension Based on Reinforcement Learning

TL;DR

This paper introduces a reinforcement learning-based charging strategy using SAC that optimizes Li-ion battery life extension while allowing flexible charge times tailored to user needs, outperforming previous methods.

Contribution

It presents a novel SAC-based approach that adapts to varying battery parameters and user preferences, improving battery longevity without compromising convenience.

Findings

Effective battery life extension demonstrated

Flexible charging times accommodate user needs

Robustness to battery aging parameters shown

Abstract

A reinforcement learning-based optimal charging strategy is proposed for Li-ion batteries to extend the battery life and to ensure the end-user convenience. Unlike most previous studies that do not reflect real-world scenario well, in this work, end users can set the charge time flexibly according to their own situation rather than reducing the charge time as much as possible; this is possible by using soft actor-critic (SAC), which is one of the state-of-the-art reinforcement learning algorithms. In this way, the battery is more likely to extend its life without disturbing the end-users. The amount of aging is calculated quantitatively based on an accurate electrochemical battery model, which is directly minimized in the optimization procedure with SAC. SAC can deal with not only the flexible charge time but also varying parameters of the battery model caused by aging once the offline learning is completed, which is not the case for the previous studies; in the previous studies, time-consuming optimization has to be implemented for each battery model with a certain set of parameter values. The validation results show that the proposed method can both extend the battery life effectively and ensure the end-user convenience.