Prediction of Heat Demand for Building Energy Managers: An IoT and Control Perspective

As the world becomes more urban, the focus on energy reduction in buildings grows with aims to tackle climate change. Being heat demand a representative slice of the total load of any non-tropical building, this paper develops a practical and less invasive forecast method for Building Energy Managers within a control framework. Methodologically, after performing a Graphical Data Analysis of heat demand profiles from two dwellings and an industrial facility, terms of a General Additive Model are developed, combining this powerful machine learning technique, adaptiveness, and the convenience of an available online meteorological forecast. The algorithm was able to predict thermal energy usage of 75% of the test days with an error below 25%, disregarding profile type and looking always at the next 48 hours. Also, results showed that the true added value of ambient temperature as exogenous and explanatory variable comes with its daily aggregation.