A data-driven framework for estimating remaining driving range in cargo electric vehicles

Abstract The increasing shift to sustainable transportation has fueled growing interest in electric vehicles, including the key issue of estimating the remaining driving range with high precision. The study develops a data-driven approach to predict the cargo electric vehicle’s remaining driving range that incorporates machine learning based estimation. Real-world operational data: a Musoshi Pop-Up Mini electric cargo vehicle was tested under various load and speed characteristics on a 2 km campus route, providing a possibility of high-resolution modeling of energy consumption patterns. After systematic preprocessing with feature engineering and segment-wise aggregation, seven regression algorithms: ElasticNet, Support Vector Regression, Random Forest, LightGBM, XGBoost, CatBoost, and ExtraTrees were optimized with Optuna-based Bayesian hyperparameter tuning and exhaustively compared in terms of RMSE, MAE, and R². Amongst these, the SVR model RMSE equal to 2.37, MAE equal to 1.75, and R² equal to 0.892 demonstrated the best performance and outperformed other ensemble and gradient boosting models. The obtained results prove that data-driven models, can reliably assess energy consumption and range for cargo EVs, which would ensure the safer and more reliable deployment of electric mobility systems.