Enhancing Battery Electric Vehicle Range Predictions Using Advanced Machine Learning Techniques

In this research, a comprehensive methodology is presented for predicting the driving range of Battery Electric Vehicles (BEVs) by implementing advanced machine learning techniques. The high levels of electro-mobility adoption mean accurate range predictions are needed to reduce consumer concerns about charging infrastructure and how far BEVs can travel between stops. The paper presents a comprehensive literature review highlighting the limitations of classical range prediction techniques and demonstrating the need for data-driven solutions. The information was compiled from manufacturer data, telemetry driving in real-world conditions, and ambient properties to create a complete dataset. Before the model can be trained, the data must undergo a series of rigorous preparation steps to ensure quality missing values will need to be imputed, outliers identified and handled, and categorical variables encoded. Model performance is improved by creating interaction features and centering data using feature engineering techniques. A total of five machine-learning models were evaluated: LightGBM, CatBoost, XGBoost, Gradient Boosting, and AdaBoost. These models are chosen for their ability to process large data and identify subtle trends in driving patterns and battery behavior. The performance of each model is evaluated by indicators MAE, MSE, and R². The performance of all models is good, with very high R² values above 0.85 indicating a strong correlation between predicted and actual values as shown below in the results. LightGBM achieves the lowest mean absolute error of 12.7654 km and explains nearly all the variance in the independent variables, with an R² value of 0.9156. The study then examines the implications of these findings for researchers and industry stakeholders, highlighting that machine learning can significantly enhance BEV range predictions. However, to build customer confidence in electric vehicles, researchers must focus on interpreting these algorithms to improve their reliability.