A novel state of charge (SOC) estimation method for lithium-ion batteries is proposed. The method is made by combining a model-based method and a data-driven method. Model-based methods can show acceptable estimation error without large data. However there is a limit in reducing the error because inaccuracy of model still exists. A data-driven method can solve this problem by learning data. The method proposed in this paper optimizes parameters of extended Kalman filter (EKF) with reinforcement learning (RL) when estimating SOC using EKF. This method utilizes both the advantage of model-based method and the advantage of data-driven method. Even if the RL is slightly trained, an acceptable SOC estimation error can be obtained through model-based estimation. When RL is trained more with data, the error decreases. The proposed method are validated by simulation with battery charge/discharge data.
展开▼
