Sustainable energy management and control for Decarbonization of complex multi-zone buildings with renewable solar and geothermal energies using machine learning, robust optimization, and predictive control

Although predictive control is an effective approach leveraging weather forecast information to control indoor climate, forecast errors would lead to poor energy management decisions that may cause thermal discomfort for occupants. A machine learning-driven robust model predictive control framework is proposed for sustainable multi-zone buildings using renewable energies. This framework addresses uncertainties of weather forecast in energy management, reduces overall electricity expenses, and ensures thermal comfort for occupants. Sustainable technologies, including solar panels, geothermal heat pumps, and battery energy storage systems are considered. Optimization of the predicted mean vote index, considering factors such as humidity, temperature, and clothing insulation, is achieved to improve thermal comfort. Initial analysis and modeling of temperature, solar radiation, and humidity forecast errors are conducted using density-based spatial clustering of applications with noise and K-means clustering. Machine learning techniques then construct disjunctive data-driven uncertainty sets of forecast errors. Employing a data-driven optimization method, control inputs are produced at each interval to reduce overall electricity expenses while maintaining a comfortable environment for building inhabitants. This study presents a year-long simulation of a sustainable, multi-zone, two-story building in Ithaca, New York, focusing on managing humidity, temperature, and the predicted mean vote index, taking into account dynamic energy pricing. The proposed approach results in 6.9% less electricity cost with a higher stability of the energy system than the data-driven robust control approach without clustering methods.