Estimation of Remaining Driving Range of Electric Vehicles Based on Segment-Wise LSTM and Feature Hierarchical Classification

Abstract Accurate estimation of the remaining driving range (RDR) in electric vehicles is crucial for safety and user experience. This study proposes a data-driven RDR estimation method using a segment-wise LSTM framework, supporting both offline training and online execution. First, to address the impact of battery temperature and energy consumption, a hierarchical feature classification is introduced, improving reliability under complex driving conditions. Second, to preserve temporal dependencies, the LSTM training strategy is modified into a segment-wise LSTM, enabling independent segment training and eliminating ineffective inter-segment learning. Additionally, Bayesian optimization (BO) is applied to optimize model hyperparameters, enhancing accuracy and robustness. The method is validated using real-world driving data from electric vehicles in northern and southern China. Comparative experiments show that the proposed approach achieves higher accuracy, better adaptability, and lower computational costs than conventional methods.