A vehicle-cloud collaborative framework for state of health estimation of lithium-ion batteries via multi-feature fusion and hybrid data-driven–empirical modeling

With the rapid development of battery technology, lithium-ion batteries (LIBs) have become essential to electric vehicles due to their superior performance. Accurate and real-time estimation of state of health (SOH) is critical for ensuring battery reliability and safety. This paper proposes a novel and practical vehicle-cloud collaborative framework for SOH estimation of LIBs, explicitly designed to balance estimation accuracy and real-time performance. A data-driven model based on CNN-LSTM-Self-attention is deployed on the cloud platform to capture complex temporal dependencies and assign adaptive weights to key health features, thereby improving estimation accuracy. On the vehicle side, a lightweight double exponential decay empirical model is employed to provide real-time SOH estimates with low computational cost. A Kalman filter-based fusion algorithm is then introduced to integrate both cloud and vehicle estimates, balancing accuracy and real-time performance. Experimental results on the NASA battery dataset demonstrate that the proposed method achieves a mean absolute error of 0.019 and root mean square error of 0.024. Validation using real-world vehicle data further validates the method's effectiveness. Compared with classical machine learning models, such as Support Vector Regression, Random Forest, and Extreme Gradient Boosting, as well as the advanced deep learning model CNN-Transformer, the proposed approach achieves superior estimation accuracy. Overall, this study provides a comprehensive insight into vehicle-cloud collaborative SOH estimation and highlights its strong potential for enabling accurate, reliable, and real-time battery health monitoring in practical electric vehicle applications.