Investigation of machine learning techniques for building energy consumption estimation

Redundant consumption of energy in the buildings is an important issue that causes to the increasing problems of climate change and global warming in the world. Therefore, it is necessary to develop efficient energy management approaches in the buildings. Accurate prediction of energy consumption plays an important role in improving the energy efficiency of buildings. In recent years, data-driven methods, which provide more accurate prediction results than standard physical or statistical methods, have become increasingly important in estimating the energy consumption of buildings. In this thesis, machine learning methods are compared to perform hourly building energy consumption estimation to be integrated into the Building Energy Management System (BEMS). Random Forest (RF), Gradient Boosted Trees (GBT), Support Vector Machine (SVM), Artificial Neural Network (ANN) and Deep Neural Network (DNN) methods and one year of energy consumption data from a sample building have been used for the prediction process. Three categories of parameters have been utilized for accurate prediction of energy consumption: meteorological, time and air-conditioning demand information. Experimental results have shown that machine learning methods are successful in predicting the total energy consumption of a building with high performance.