Development and Selection of Load Forecasting Techniques for Non-Residential Buildings

Non-residential buildings account for a significant portion of energy consumption and greenhouse gas emissions among all structures. A key element of their Building Management System (BMS) is the electrical load forecasting functionality. As numerous techniques have been developed for load forecasting, adaptability has emerged as a fundamental feature of a robust forecasting system, given that no single technique outperforms all others. In this paper, we evaluate a diverse pool of twelve load forecasting techniques for a hourly 24-hour-ahead load forecasting task across four real non-residential building datasets using two Key Performance Indicators (KPIs). The results demonstrate that the diversity of the pool is essential for producing the most accurate forecasts.