Extending EV Battery Lifetime: Digital Phenotyping Approach for Departure Time Prediction

Abstract

Battery degradation, a gradual loss of usable capacity over time, is one of the major hurdles for widespread adoption of electric vehicles (EVs). We introduce delayed full-charging (DFC) algorithm to mitigate degradation and extend the lifetime of EV batteries in battery management systems (BMS). When the EV is plugged in, the DFC algorithm charges batteries up to approximately 80% state of charge (SOC) and delays full charging until the predicted unplug time (tunplug). This approach significantly reduces the time batteries remain fully charged (t100%), thereby mitigating degradation while ensuring charging time for EV users to utilize the full battery capacity. For predicting tunplug, we propose a novel methodology that uses digital phenotyping to predict departure times. This method leverages smartphone data to capture irregular but predictable departure patterns by reflecting relevant behavioral and environmental contexts. A case study with 48 participants was conducted to empirically evaluate the departure time prediction performance using tree-based ensemble models trained on smartphone data, compared to a baseline Long Short-Term Memory (LSTM) model trained on historical data. Results reveal that models utilizing mobile passive features achieved a Mean Absolute Error (MAE) as low as 2.18 hours on weekdays and 4.46 hours on weekends, demonstrating superior effectiveness in capturing irregular patterns compared to the baseline model trained only on historical temporal features.