Hybrid Global and Sub-domain Approach for Accurate Hourly Cooling Load Forecasting inShort, Medium, and Long-term Horizons

Abstract Accurate hourly cooling load forecasting is essential for optimizing HVAC system management, reducing energy consumption, and achieving sustainability goals. This paper presents the Hybrid Global and Sub-domain Approach (HGSA), an innovative methodology that integrates advanced feature engineering, domain-based data segmentation, and ensemble modeling to achieve precise cooling load predictions across short-term, medium-term, and long-term horizons. HGSA effectively addresses the challenges of dynamic cooling load patterns through model fusion and periodic pattern recognition. Validated on real-world datasets, the proposed approach demonstrates superior performance over state-of-the-art methods, including Long Short-Term Memory (LSTM), Light Gradient Boosting Machine (LightGBM), Autoregressive Model (AR), and Facebook Prophet. Its high accuracy, adaptability to existing models, and efficient deployment make HGSA a robust and practical solution for applied scenarios, such as building energy saving management, HVAC optimization, and energy strategy development. The results establish HGSA as a reliable method for precise cooling load forecasting and strategic energy planning.