Data Mining Approaches for Calculating the Energy Consumption of Buildings

Zero Budget Sustainability is a high potential service offered by Ploos Energieverlening. In this service, external financers invest in energy saving measures when the owner of the building lacks budget. The resulting savings are used to repay the investor. Currently, both financers and building owners lack confidence in the forecasted savings. It has been indicated that an independent scientific approach to estimate energy consumption of buildings can help to improve confidence. Based on a scientific literature study and expert knowledge, an artificial neural network has been built to calculate the energy consumption of commercial buildings on 15 minute time intervals. These estimations are based on climate variables, behavioral data and building characteristics including 20 specified low-level energy saving measures. The dataset is further improved by a fuzzy cluster analysis. The model has also been tested on a dataset containing residential buildings. The obtained results show that electricity consumption can be estimated with no more than 8.23 percent symmetric errors. Gas consumption can be estimated with 26 percent symmetric errors. Over a longer period the difference between estimated and observed consumption is 0.3 percent for electricity and 4.3 percent for gas. For residential buildings, absense of behavioral data about the residents resulted in less accurate predictions although adding re-supplied electricity from solar panels made an improvement. During the research project, meaningful recommendations and practical implications for the model were given to the company.