Scalable Building Temperature Prediction for Smart HVAC Control: A Multi-Stage Learning Framework

Rohan Saha, Tushar Shinde

Keywords: Building Optimization, Temperature Forecasting, XGBoost, Multi-Stage Learning, Smart Buildings, HVAC Control

Abstract: Commercial buildings contribute 17\% to U.S. carbon emissions, with Heating, Ventilation, and Air Conditioning (HVAC) systems accounting for most energy consumption. We propose a novel multi-stage learning framework for accurate and scalable temperature prediction in smart buildings. Our approach systematically scales from single-zone, single-day forecasts to 123-zone, multi-week predictions, achieving a mean absolute error (MAE) of 0.195°F for single-zone tasks using XGBoost. Our framework advances energy-efficient HVAC control, reducing carbon footprints in commercial buildings.

Submission Number: 42