Assessment of predictive models for the estimation of heat consumption in kindergartens

The service sector remains the only economic sector that has recorded an increase (3.8%) in energy consumption during the last decade, and it is projected to grow more than 50% in the following decades. Among the public buildings, educational are especially important since they have high abundance, great retrofit potential in terms of energy savings and impact in promoting a culture of energy efficiency. Since predictive models have shown high potential in optimizing usage of energy in buildings, this study aimed to assess their application for both finding the most influential factors on heat consumption in public kindergarten and heat consumption prediction. Two linear (Simple and Multiple Linear Regression) and two non-linear (Decision Tree and Artificial Neural Network) predictive models were utilized to estimate monthly heat consumption in 11 public kindergartens in the city of Kragujevac, Serbia. Top-performing and most complex to develop was the Artificial Neural Network predictive model. Contrary to that, Simple Linear Regression was the least precise but the most simple to develop. It was found that Multiple Linear Regression and Decision Tree were relatively simple to develop and interpret, where in particular the Multiple Linear Regression provided relatively satisfying results with a good balance of precision and usability. It was concluded that the selection of proper predictive methods depends on data availability, and technical abilities of those who utilize and create them, often offering the choice between simplicity and precision.