Evaluation of electric vehicle consumption models based on real-world driving data

Abstract

As the adoption of Electric Vehicles (EVs) accelerates, understanding and accurately predicting EV energy consumption is essential for optimizing EV range, refining charging infrastructure, and improving overall efficiency. With advances in real-world data collection, this study evaluates seven distinct EV consumption models using driving data from a BMW i3 (60 Ah). The models were assessed based on assumptions, computational methods, and their ability to incorporate factors such as EV dynamics and environmental conditions. Performance metrics, including Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Symmetric Mean Absolute Percentage Error (SMAPE), and R-squared (R${}^2$) were employed to compare model accuracy. The model incorporating detailed resistance forces and constant efficiency factors emerged as the most accurate across all metrics, demonstrating superior performance in predicting energy consumption. Conversely, a model incorporating key resistive forces and efficiency parameters while considering a minimum velocity threshold for power consumption showed the highest error rates. Furthermore, integrating machine learning techniques with physical models proved beneficial, enhancing predictive accuracy with the lowest SMAPE and capturing complex patterns in the data, achieving a balance between accuracy and interpretability. This research offers actionable insights into optimizing EV range, designing charging infrastructure, and improving energy efficiency modeling, providing a valuable reference for stakeholders in electric mobility and energy management. A key strength of this study is its systematic comparison of seven energy consumption models, using real-world driving data from a BMW i3 (60 Ah) to establish a new benchmark for EV energy modeling.