Demand forecasting model for load shifting strategy in building energy management system

Abstract

Among the sectors with the highest energy consumption are transport, industries, and buildings. Buildings are responsible for the third part of energy consumption and almost 40% of CO2 emissions worldwide. The search to improve the comfort of the occupants inside the buildings has brought a consequence that buildings are increasingly equipped with devices that help to improve the thermal comfort, visual comfort, and air quality inside the buildings, causing more energy demand regardless of the type of building making buildings an untapped efficiency potential. This doctoral thesis presents a model for forecasting electricity demand in buildings based on machine learning for load-shifting strategies, which can be implemented in building energy management systems. First, the state of the art of building energy management systems is analyzed, as well as the different management strategies used within these systems. Second, within the predictive control model management strategy, the forecast models of energy consumption in buildings are analyzed, as well as the methods, input variables, prediction horizon, and metrics. Finally, about the analysis carried out on the energy consumption forecasting models, a short-term energy consumption forecasting strategy based on machine learning is developed that allows forecasting the demand for the next 24 hours from any time of the previous day.