Edge-cloud collaborative method for state of charge estimation of lithium-ion batteries by combining Kalman filter and deep learning

Accurate state of charge (SOC) estimation is crucial for ensuring the safety and reliability of lithium-ion batteries (LIBs). However, achieving high-precision real-time SOC estimation remains challenging due to complex operational conditions and limited computational resources in onboard battery management systems . This paper presents an efficient edge-cloud collaborative method for online SOC estimation of LIBs that achieves both high accuracy and real-time performance. At the edge side, an adaptive cubature Kalman filter (ACKF) is implemented based on a simplified electrochemical model to capture battery dynamics. The cloud-side integrates the feature extraction capabilities of convolutional neural networks (CNN), the sequential modeling enhancements of long short-term memory networks (LSTM), and the dynamic focusing attention mechanism (AM). This CNN-LSTM with attention network (CLAN) framework performs post-processing and fusion of edge-side SOC estimation results with regularization techniques to achieve more reliable outcomes. Experimental validation under typical driving cycles demonstrates that the proposed method achieves a mean absolute error (MAE) of 0.41 % and root-mean-square error (RMSE) of 0.49 %, significantly enhancing accuracy by over 35 % compared to individual methods. Furthermore, the proposed method demonstrates robust generalization capabilities across various operating conditions while maintaining an optimal balance between computational efficiency and estimation accuracy through strategic resource allocation. • Novel edge-cloud collaborative SOC estimation combines KF-based and DL methods. • Edge-side ACKF with the electrochemical model ensures real-time interpretability. • Cloud-side CLAN model integrates CNN-LSTM with attention mechanism. • Regularized fusion strategy achieves MAE of 0.41 %, improving by over 35 %. • Optimized resource allocation enables robust estimation across various conditions.