engleski

Modelling energy efficiency of public buildings by neural networks and its economic implications

Machine learning methods, such as artificial neural networks, have shown their success over statistical methods in previous research. However, they have not been exploited enough for the purpose of efficient prediction of energy efficiency. In the domain of public buildings owned by state, improving energy efficiency could significantly save the state budget. Therefore it is important to estimate the influence of characteristics of buildings and their interdependence in order to decide how to allocate resources for the reconstruction of public buildings. In this paper, methodology of neural network is used on the real dataset of Croatian public buildings covering the input space of 130 building attributes. After data pre-processing, two approaches of variable selection were used, based on statistical methods and sensitivity analysis. The most accurate model was selected, and economic implications of suggested model are also discussed. The results show that neural network methodology has the potential in predicting energy efficiency and estimating important features for classifying buildings.

machine learning, artificial neural networks, high-dimensional data, energy efficiency, public buildings

This work has been fully supported by Croatian Science Foundation under Grant No. IP-2016-06- 8350 ; Methodological Framework for Efficient Energy Management by Intelligent Data Analytics" ; (MERIDA).