IMPLEMENTATION OF MACHINE LEARNING METHODS FOR PREDICTING ENERGY DEMAND OF PCM-INTEGRATED BUILDINGS

Abstract

Numerous machine learning methods have been employed to predict the energy consumption of PCM-integrated buildings. However, the following research gaps have not been addressed yet. 1) Most researchers developed prediction models by only considering building parameters; 2) Only one research team developed prediction models by considering environmental parameters. However, they did not consider some of the important environmental parameters including precipitation and air pressure; 3) No research team have proposed the prediction model by considering the future climate scenario and evaluated the impact of hyperparameters especially for decision tree-based algorithms. This research aims to evaluate the efficacy of different decision tree subcategories (fine, medium, and coarse trees) for predicting energy consumption in PCM-integrated buildings for future climate scenario, considering extensive building and environmental parameters. A database for the energy consumption of PCM-integrated buildings for 11 cities in the hot semi-arid climate zone was created through energy simulations. The results showed that the Fine Decision Tree-based prediction model (FDT3) was the most reliable and accurate prediction model, having R2 values greater than 94% for both training and testing phases. Overall, the developed model will be useful in providing valuable insights into the possibilities for sustainable building design and climate-responsive energy management.