Estimation of battery SOC by Kalman fusion learning method based on expansive force and coupling characteristics

Accurate estimation of the state of charge (SOC) of lithium-ion batteries is crucial for battery management systems. In recent years, studies have shown that there is a mapping relationship between the expansion force and SOC. Using expansion force will improve the precision of the algorithm. Therefore, we propose a Kalman fusion learning method based on expansion force and coupling characteristics, which combines the traditional physical model and deep learning model to change the feature dimension. Considering the effect of expansion force in traditional model parameter identification, IAEKF is used to discard the fixed time window, compute more accurate a priori battery states, and obtain the forward and backward correlated information transfer equations while eliminating the approximation error of the expansion equation. A CNN-GA-LSTM network is used to extract the coupled characteristics, reconstruct the input vectors on the basis of the expansion force, and realize the loopback control of the time series prediction through the multi-parametric feature convergence state. The Kalman fusion learning method calculates the covariance matrix according to the error vectors of the current sensor, battery physical model, and deep learning model to obtain more precise results. Our method solves the problems of inaccurate traditional model parameters and difficult resolution of feature variables, while it makes up for the shortcomings of deep learning in terms of lack of mechanism explanation and insufficient dimension of data features. Results show that the RMS and MAE of KF-(IAEKF)-(CNN-GA-LSTM) algorithm are less than 0.6 % and 0.4 %, respectively, under various working conditions.