Hybrid XGBoost-PINN Multi-Stage Framework for Scalable Building Temperature Prediction

Building temperature prediction is crucial for energy optimization and control in smart cities. We present a hybrid framework combining XGBoost with physics-informed neural networks (PINN) in a multi-stage sequential scaling approach. Starting from single-zone, single-day predictions, we progressively scale to multi-zone, multi-year forecasts using real-world data from Google's Smart Building Simulator. Our method incorporates physics-enhanced features, temporal encodings, and inter-zone interactions, achieving mean absolute errors (MAE) as low as 0.169°F for weekly multi-zone predictions. For longer horizons, we employ ensemble strategies, demonstrating robust performance up to 2.5 years. This work advances urban AI by enabling accurate long-term building dynamics modeling for downstream control tasks. Link to the code for the NeurIPS \textbf{UrbanAI-2025 Workshop Contest}: \url{https://colab.research.google.com/drive/1ulQzwq-awYVQYI7yLk_jW93oZE5aIt?usp=sharing}