engleski

Machine Learning in Energy Consumption Management

Energy efficiency is an important topic in the context of climate change and has gained much attention in EU directives and national strategic and action plans. Researchers strive to build models that could efficiently predict or explain the main factors that influence energy efficiency. Since buildings are the largest individual energy consumers [20] and building sector itself contains 40% of total primary energy consumption, the efficient models that decision makers could use to allocate resources in reconstructions of buildings are desirable. The methodology for calculating energy efficiency category of a building is determined by professionals. However, there is a lack of analyses that investigate relationships among various buildings’ attributes describing construction, geospatial data, climate data, heating data, cooling data and usage, as well as their connection to energy consumption. Since it is a complex issue which includes uncertainty and nonlinearity, it requires advanced methodology. In this paper, the aim is to investigate possibilities of several machine learning methods in extracting important predictors of yearly energy consumption of electricity and natural gas. The methods of artificial neural networks, CART decision trees, conditional inference trees, random forest, and support vector machines are tested on a real dataset of Croatian public buildings. The similarities and differences among the three tested methods in modelling energy consumption are discussed.

energy management, machine learning, support vector machines, artificial neural networks

nije evidentirano