Prof. dr. Marijana Zekić-Sušac
J. J. Strossmayer University of Osijek, Faculty of Economics in Osijek
Title: Machine Learning in Energy Consumption Management
Abstract: Energy management is one of the hot topics among researchers. How to consume less energy from non-renewable resources is among the biggest challenges and could be approached by different methods. Previous research in the area of energy management shows that various deterministic and stochastic methods have been used for predicting energy consumption. Statistical methods such as autoregressive moving averages (ARMA), cycle analysis or multiple regression are among the most frequent ones, while recent papers reveal that machine learning methods such as neural networks, support vector machines and others show more accuracy in prediction. The advantages of machine learning over standard statistical methods are in the fact that they do not have strong requirement regards stationarity and interdependence of input data, they are robust, and show more success in short-term prognoses. In this paper, several machine learning methods will be tested to analyze the influence of specific characteristics of buildings and implemented measures on energy consumption. The extracted knowledge could serve in the process of decision making on investing in measures for improving energy efficiency, as well as in future research for developing additional models.