Battery State of Health Prediction Based on Telematics Data from Pure Electric Commercial Vehicles

With the rapid development of the electric vehicle industry, accurate prediction of the battery state of health (SOH) is crucial for reliable operation and extending service life. This study utilizes large-scale telematics data collected from pure electric commercial vehicles, starting with data cleansing and health indicator extraction. Three distinct prediction models are developed: a Long Short-Term Memory (LSTM) network, a Spiking Neural Network (SNN), and a hybrid LSTM-SNN architecture. Results demonstrate that the LSTM model achieves high prediction accuracy, while the SNN exhibits the lowest maximum absolute error (MaxAE). The hybrid LSTM-SNN model delivers optimal predictive performance, attaining a root mean square error (RMSE) of 0.293 and a coefficient of determination ($\mathbf{R}^{\mathbf{2}}$) of 0.986, improving over individual models. These findings provide data-driven support for battery health analysis as well as customer identification and retention in automotive enterprises.