AI-Driven Predictive Energy Management for Sustainable Smart Buildings: A Machine Learning Framework for Optimizing Building Performance in Hot-Arid Climates

Abstract

Buildings in hot-arid regions have one of the highest demands for cooling due to high temperatures, solar gains, large temperature variations, and prolonged air-conditioning operating hours. Most of the time, conventional building management systems are not able to proactively manage the indoor built environment due to the inability to predict the peak load cooling demands in advance. An AI-based predictive energy management system for smart buildings in hot-arid regions is proposed and evaluated in this research. An 8,400 m² office building was taken as an example for smart building energy performance assessment. The study prepared a two-year hourly physics-informed operational data set for model training and testing, which included outdoor air temperature, relative humidity, solar radiation, building occupancy, past electricity consumption data, indoor temperature, and HVAC electricity consumption. Five machine learning algorithms were trained and evaluated to predict the hourly electricity consumption for one-hour-ahead forecasting. Extra Trees regressor showed the best performance with R², RMSE, MAE, and MAPE equal to 0.975, 14.11, 10.85, and 7.23. A forecast-driven supervisory energy management system for smart buildings was designed to manage the building energy performance by considering occupancy-related HVAC operation scheduling, start-up, pre-cooling, re-set points, and demand response. Compared to conventional approaches, the proposed method decreased the whole-building electricity consumption by 14.26% (from 1.515 to 1.299 GWh) and reduced the HVAC electricity consumption by 21.09%. The peak demand was lowered by 19.59% for the same test year. The energy use intensity was reduced from 180.37 to 154.65 kWh/m² per year or 14.26% for the same building. Meanwhile, the occupant comfort was provided for 98.94% of occupied hours. The research shows that demand-side energy management in smart buildings can transform traditional building management from reactive to proactive mode. The proposed method can be applied in various aspects of sustainable construction, buildings commissioning, facilities management, project delivery, and digital handover in hot-arid regions with a fast-growing construction industry.