Battery State of Charge Estimation Using Long Short-Term Memory Network and Extended Kalman Filter

In this paper, a long short-term memory network structure is developed to estimate state of charge for lithium-ion batteries owing to its time series characteristic. It is further followed by the extended Kalman filter to alleviate the process noise. The proposed algorithm shows reduced root mean squared error as low as 0.48%, compared with traditional algorithms like linear regression, support vector regression and general shallow neural network. Our work provides a feasible way to estimate state of charge of batteries for general dynamic loading conditions.