High-Accuracy Battery State of Charge Estimation Strategy Based on Deep Belief Network Cascaded With Extended Kalman Filter

Battery state of charge (SOC) estimation is one of the main functions of the battery management system in electric vehicles. If the actual SOC of the battery differs significantly from the estimated value, it can lead to improper battery usage, resulting in unexpected rapid voltage drops or increases, which can affect driving safety. Therefore, high-accuracy SOC estimation is of great importance for battery management and usage. Currently used SOC estimation methods suffer from issues such as strong dependence on model parameters, error propagation from measurements, and sensitivity to initial values. In this study, we propose a high-precision SOC estimation strategy based on deep belief network (DBN) feature extraction and extended Kalman filter (EKF) for smooth output. The proposed strategy has been rigorously tested under different temperature conditions using the dynamic stress test (DST) and urban dynamometer driving schedule (US06) driving cycles. The mean absolute error (MAE) and root-mean-square error (RMSE) of the proposed strategy are controlled within 1.1% and 1.2%, respectively. This demonstrates the high-precision estimation achieved. To further validate the generality of this strategy, we also apply it to graphene batteries and conduct tests under US06 and highway fuel economy test (HWFET) driving cycles at temperatures of 25 °C and −10 °C. The test results show MAE of 0.47% and 2.01%, respectively.