A hybrid framework for remaining driving range prediction of electric taxis

To reduce Electric vehicle (EV) users' mileage anxiety and optimize the EV battery energy management system, this study proposes a method for establishing a Remaining Driving Range (RDR) prediction with accuracy, computational efficiency, and interpretability using real EV driving data. This method integrates data -driven and model -based approaches and supports both offline training and online execution. Initially, the RDR is physically decomposed into Remaining Discharge Energy (RDE) and Energy Consumption Rate (ECR). Furthermore, to account for the degradation due to long-term battery operation and the uncertainty in driving energy consumption, RDE and ECR are transformed into predictions of the State of Health (SOH) and an ECR coefficient alpha. The data -driven model LightGBM and an improved TCN-GRU are used to predict these two key parameters. This study utilized real -world driving data from multiple electric taxis in Shanghai, China, spanning 2.5 years, to validate the effectiveness of this methodology and analyzed its prediction accuracy under various speed conditions, temperatures, seasons, and data collection frequencies (1 Hz and 0.1 Hz) through comparative experiments, and finally discussed its computational efficiency and interpretability. This methodology applies to EVs in urban road environments, particularly for the RDR prediction of electric taxis.