Data-Driven State of Charge Estimation for Power Battery With Improved Extended Kalman Filter

Accurately monitoring battery state of charge (SOC) is essential for battery system safety. However, single and open-loop combination algorithms are mainly used for SOC estimation currently, which may have the problems of low accuracy and poor reliability. Here, a closed-loop combination algorithm with the variance-compensation extended Kalman filter (VCEKF) and back-propagation (BP) neural network is developed for SOC estimation. First, the second-order resistance-capacitance model is established, and the model's parameters are identified by the forgetting factor recursive least square (FFRLS). Second, the extended Kalman filter (EKF), the variance compensation algorithm, and the BP neural network are merged under a closed-loop structure. Specifically, the variance compensation algorithm updates the process noise covariance of the EKF algorithm in real-time, while the BP neural network simultaneously provides compensation value to obtain the finally estimated SOC. Afterward, the proposed algorithm is testified under different driving schedules. Experimental results show that the accurate SOC estimation under different driving schedules is realized using the proposed algorithm and also illustrate a better performance than the commonly used open-loop algorithm and EKF-based algorithms.