Battery State of Health Estimation via Reinforcement Learning

The state of health of a battery characterizes its performance in terms of loss of capacity compared to the beginning of its life. This paper proposes a reinforcement learning algorithm for identifying the capacity of lithium-ion batteries. The training phase of the algorithm is based on data derived from constant current and constant voltage charging operations. The technique exploits a state observer based on a dynamic model of the battery and on the capacity estimation obtained with the reinforcement learning technique. The reward is defined as the error between the estimated and measured battery voltage. The effectiveness of the proposed solution is validated by considering different C-rates battery charging.