Reducing Electric Vehicle Range Anxiety with Trip-Specific Charge Usage Predictions

Range anxiety remains a major concern for electric vehicle (EV) drivers due to unpredictable charge usage influenced by terrain and user behavior variations. To address this issue, we propose a data-driven approach to provide accurate trip-specific battery consumption for EV drivers. First, we present a new multi-fidelity battery performance dataset with 117 laboratory verified current profiles from 23 velocity-based drive cycles, capturing diverse road conditions, driving and charging patterns. We also introduce charge usage score, a metric comparing battery consumption to a reference trip at constant maximum velocity for same duration. Secondly, we demonstrate that incorporating additional trip-specific features, such as terrain, driving and charging patterns improves charge usage score prediction. The MAPE (mean absolute percentage error) of machine learning prediction reduced from 5.2% to 4.3% for current-based models and 7.5% to 5.3% in velocity-based models. Finally, we show that charge usage scores can be predicted using only 10% of initial trip data, achieving MAPE of 7.4% (velocity-based models) and 4.4% (current-based models), respectively. Using this early prediction capability, we demonstrate that switching from an assertive to a defensive driving style can save 3% more charge over an 800 second trip. Essentially, this closed-loop feedback system allows drivers with sufficient time to adjust their route, charging strategy and driving style in real time, reducing range anxiety. We hope our work offers a framework for advancing battery management systems and enhancing EV adoption.