An optimized relevant long short-term memory-squared gain extended Kalman filter for the state of charge estimation of lithium-ion batteries

Accurate state of charge (SOC) estimation of lithium-ion batteries by the battery management system (BMS) plays a prominent role in ensuring their reliability, safe operation, and acceptable durability in smart devices, electric vehicles, etc. In this paper, the effect of the training and testing working conditions on the accuracy of the SOC using a long short-term memory (LSTM) network is studied through transfer learning. Then, a relevant attention mechanism is introduced as a data optimizer for faster training of the LSTM network to establish a relevant LSTM (RLSTM). Finally, the SOCs estimated by the RLSTM are independently input with the working current to an extended Kalman filter (EKF) and a proposed squared gain EKF (SGEKF) method to iteratively denoise and optimize the accuracy of the final SOC under the three complex working conditions. The results show that the SOC estimation accuracy is influenced by the training and testing working conditions using the LSTM network, which provides a technique for accurate SOC estimation. Also, the established RLSTM network is computationally efficient for accurate SOC estimation. Moreover, the proposed hybrid RLSTM-SGEKF model has an overall maximum mean absolute error, mean squared error, root mean squared error, and mean absolute percentage error values of 0.35299%, 0.0017448%, 0.41765%, and 2.34403%, respectively, under the three complex working conditions. The proposed hybrid RLSTM-SGEKF model is optimal, robust, and computationally efficient for accurate SOC estimation of lithium-ion batteries for real-time BMS applications.