HVAC System Energy Demand Prediction in Electric Vehicles: Integration of Physics-based Decomposition with XGBoost Machine Learning Using Multi-Seasonal Driving Data

Auxiliary systems (HVAC, thermal management) significantly impact the range of electric vehicles under varying weather conditions. This study developed an XGBoost machine learning framework to predict auxiliary power using 95,028 real-world measurements of electric vehicles in Poland, Italy, and Germany (−8°C to +33.5°C). A physics-based energy decomposition combined with feature engineering achieved R² = 0.9986 with a mean ab-solute error of 35 W. The feature importance analysis revealed the position of the accelera-tor pedal (0.4153) as the strongest predictor, alongside the heating efficiency per tempera-ture differential (0.2716), indicating the coupling between traction demand and auxiliary loads. Heating dominated cooling by a 7:1 ratio with a 44-fold power variation in the temperature range. The contribution of the additional power ranged from 75% during idle to 12% during highway driving, contradicting static overhead assumptions. Results demonstrate that auxiliary loads require context-aware prediction rather than fixed as-sumptions, enabling improved range forecasting for electric vehicles.