Auxiliary energy consumption of electric vehicles: Modeling and prediction using real-world vehicle data

This paper solves the problem of modeling and predicting the auxiliary energy consumption in battery electric vehicles (BEVs). Auxiliary energy in BEVs can contribute to the total energy consumption, affecting their driving range significantly. However, the current literature mostly focuses only on total or traction energy consumption and neglects modeling of the auxiliary energy consumption. Therefore, this study proposes and validates both statistical and machine learning models for trip-based auxiliary energy consumption, as well as machine learning models for seconds-based auxiliary energy consumption. The models are developed and tested using comprehensive datasets collected from 42 identical commercial BEVs operating under real-world driving conditions, ensuring robust performance and practical applicability. Through real-world dataset analysis, we find that auxiliary energy consumption can contribute up to 45% of the total energy usage, emphasizing its substantial impact. Using statistical modeling, we investigate key parameters influencing auxiliary energy consumption and achieve an R2 of 0.893 in prediction accuracy for trip-based auxiliary energy consumption. This is accomplished by leveraging the Multi-Layer Perceptron (MLP) model with the identified parameters, demonstrating the effectiveness of our approach. Furthermore, we identify that the trip duration t, the thermal management system parameters, including the heat pump (lHP), A/C compressor (wAC) and PTC (rPTC) are the most significant variables on trip-based auxiliary energy consumption prediction. Also, we observe prediction accuracy of 0.883 in R2 at a 20-s interval, identified as a knee point, using the XGBoost-based algorithm, with accuracy further improving to 0.906 in R2 at a 120-s interval.