Abstract

The use of intermittent energy sources, such as sun and wind, is rapidly increasing. Since the power production originating from these sources is erratic and hard to control, there is a growing need for ways of balancing the electricity grid. This has lead to an increased demand for flexibility on the production side. During the last few years, the user side has been brought in to the discussion as well. Even though studies have shown that some household customers are motivated by the mere challenge of lowering their energy consumption, this presumes that customers are provided with comprehensible information on how and to what extent certain behavior affects electricity consumption. There currently are multiple services that provide electricity users with this type of information through real time analysis of their consumption and connected costs, and the market is on the rise. The purpose of this thesis is to contribute further to this field by providing customers with an additional incentive; predictions of future energy savings. Four different machine learning models have been used to predict how much energy a given household can save by decreasing their indoor temperature during the winter. The highest prediction accuracy is achieved with Support Vector Regression (SVR)