Reducing electric vehicle range anxiety with machine learning models incorporating human behavior

Range anxiety remains a significant concern for electric vehicle drivers due to uncertainties in accurately estimating battery charge consumption or state of charge during a trip. Factors such as terrain profiles, driving behaviors, and charging strategies further amplify this uncertainty. Most current predictive models are trained on lab data, which may not be fully relevant to real-world conditions. To bridge this gap, we developed a comprehensive dataset of 117 profiles consisting of velocity and current data capturing diverse terrain, driving, and charging patterns. We develop a quantitative metric called the charge saving score to evaluate battery consumption savings relative to a same-duration trip at constant maximum velocity. Our predictive models, incorporating terrain and human behaviors demonstrate reduced prediction errors on test dataset for both velocity (5.4%) and current models (4.3%). We further investigated the early prediction performance of our models at two time intervals. The best performing current model achieved comparable results, with errors of 4.3% and 2.7% at intervals of 300–350 and 300–750 seconds, respectively. The early prediction of the charge saving score enables personalized recommendations during the trip, allowing drivers to optimize their routes or driving styles, or charging strategies in real time, and ultimately helping to reduce range anxiety.