A Fully Automated and Scalable Approach for Indoor Temperature Forecasting in Buildings Using Artificial Neural Networks

Improving the performance of buildings is a core pillar to attaining future energy and environmental goals in different countries, considering that the building sector is a major contributor in terms of both energy consumption and carbon emissions. These ambitious goals and the call for smarter, energy-efficient, and flexible buildings have called for innovative and scalable energy and indoor thermal comfort modeling and prediction approaches. This work presents a fully automated and scalable solution using Artificial Neural Networks to forecast indoor room temperatures in buildings. A case study of an 8500 m2 university building in Denmark was considered for testing and evaluating the proposed approach. An extensive dataset was constructed with sensor data from 76 rooms that contain both readings on indoor temperature, CO2 concentrations, and actuating signals on radiator valves and dampers, as well as outdoor ambient conditions. Using this dataset, a well-performing architecture is identified, which provides accurate temperature predictions in the various rooms of the building for prediction horizons of 24 hours.