Data-driven Predictive Control for Heating Demand in Buildings

Abstract

This research is part of IPIN project (Smart Grid Innovation Programme) which aims to minimize the supply temperature for the TU Delft buildings by predicting and managing the buildings’ heat requirement and the heat supply. The prediction of heating demand is currently performed by a physic-based simulation tools which gives good estimations of the thermal energy demand of the building but it requires a large number of unknown input parameters (building & system characteristics). The estimation of these parameters is a time- & budget-consuming task, in addition to reducing the accuracy of the heating demand prediction. This thesis was proposed in order to give an optimal solution to the inconvenience mentioned above. The goal of this research is to study the possibility of using simple and fast mathematical models to predict the heating demand of the building with enough accuracy and physical meaning. The final model resulted in a multivariate linear equation defined by weather data, indoor air temperatures and the internal heat gains of the building. The equation shows a high predictive potential and accuracy level. The data collected from the previous season (2.5 months) are able to predict the next month with an accuracy in the range of 90-99%. This study concludes that the multivariate linear regression model is a more suitable predictive tool than a physics-based model for large scale implementations.