A Dual-Driven SOC Estimation Framework: Fusion of Multiscale Temporal Encoding Network and EKF Based on Feature Dimensionality Reduction

Accurate state of charge (SOC) estimation is crucial for battery safety. Although the mechanism and data fusion estimation methods are relatively accurate and interpretable, the existing fusion strategies mostly rely on a single feature, ignoring the influence of multiple features. Directly designing fusion strategies based on multiple features will undoubtedly increase the complexity. To address these issues, a novel multifeature dimensionality reduction fusion framework is proposed. The battery is characterized by the Thevenin model, and its parameters are identified by the forgetting factor recursive least squares FFRLS) method. With these parameters, the SOC and open-circuit voltage are then estimated by the extended Kalman filter (EKF) algorithm. A multiscale temporal encoding network (MSTEN) is proposed to mine temporal information at different scales to estimate the SOC. The input features of the MSTEN are subjected to feature dimensionality reduction by kernel principal component analysis (KPCA), and the fusion strategies are designed according to the results of dimensionality reduction. The final SOC estimation results are integrated based on these fusion strategies. The effectiveness of the proposed method is validated by multiple driving cycle experiments on the LG 18650-HG2 dataset. These experiments demonstrate that the root mean square error (RMSE) of the proposed method is less than 0.44%, and the mean absolute error (MAE) is less than 0.32%, under different operating conditions.