Algorithm of BPNN‐UKF based on a fusion model for SOC estimation in lithium‐ion batteries

Lithium‐ion batteries (LIBs) are widely used as energy storage devices. Accurately estimating the state of charge (SOC) is critical for the safe operation of LIBs, but the complex internal characteristics of LIBs make this a difficult task. In this study, an algorithm called BPNN‐UKF, which combines the backpropagation neural network (BPNN) and unscented Kalman filtering (UKF), is proposed. First, a fusion model construction method incorporating AdaBoost and recursive least squares (RLS) is developed to enhance the accuracy and generalization of the battery model. Next, the fusion model is applied to create BPNN‐UKF, which reduces the model dependence of UKF and the training requirements of BPNN. In this method, UKF estimates the SOC, and then BPNN learns the nonlinear relationship between SOC estimation error and process variables in UKF preliminary estimation. The BPNN output is used to update the state vector in real time during the new UKF estimation process. Experimental results reveal that the proposed method can drastically improve estimation accuracy compared with the UKF and the simple combined BPNN‐UKF, and it exhibits considerable generalization to common disturbances in SOC estimation and various battery working conditions. This study proposed a novel BPNN‐UKF algorithm based on a fusion model for SOC estimation in lithium‐ion battery in which the high dependence of KFs on battery model accuracy and the training effect requirements of NN can be optimized simultaneously.