An improved multi-innovation error compensation-long-short-term memory network modeling method for high-precision state of charge estimation of lithium-ion batteries

Abstract

Accurately estimating lithium-ion batteries’ state of charge (SOC) is a vital decision-making technique in battery management systems (BMS), essential to ensuring operational safety and prolonging battery lifespan. The multi-innovation error compensation-long-short-term memory (MEC-LSTM) network modeling method is proposed in this paper to enhance SOC estimation’s accuracy. The extended Kalman filter’s (EKF) limitations are addressed through the sliding window multi-innovation theory, which improves the ability to capture the dynamic relationships of nonlinear systems. To reduce the EKF’s error, the LSTM network is introduced for modeling, and the SOC error in the training results is used for error compensation, which solves the problems of slow convergence speed and erratic output of the LSTM network, leading to a notable enhancement in SOC estimation performance. The algorithm’s feasibility is confirmed through data analysis across complex working scenarios. Findings reveal that under the Hybrid Pulse Power Characterization Test (HPPC), Dynamic Stress Test (DST), and Beijing Bus Dynamic Stress Test (BBDST) working conditions, the average absolute error and the root-mean-square error are all within 2%. Validation results underscore the method’s high precision regarding estimating SOC for lithium-ion batteries, offering new ideas for SOC estimation techniques.