Verification of Energy Efficiency Measures in Three Apartment Buildings Using Gaussian Process

Abstract

The aim of this work was to examine whether the Gaussian process as a machine learning method is suitable for modelling time series data collected from buildings and whether it can be used to verify the effects of energy efficiency measures on three apartment buildings. A Gaussian process regression model was created using outdoor temperature and time information as inputs including information about the day of the week and the hour of the current day. Correspondingly, the output of the model was to estimate the hourly heating power demand corresponding to these inputs. The results provided by the created model were used as a reference point to verify the effects of energy efficiency measures taken on these residential buildings. The model was trained with 2016 hourly data. The 2017 data was used as test data to evaluate the functionality of the model. The impact assessment of the energy efficiency measures was performed with the measured data of 2019, which was compared with the results given by the model. Based on the performed modelling, it can be stated that using the Gaussian process, the need for hourly power of buildings was reasonably well modelled with even small amount of input variables. It can be assumed that the biggest uncertainty factor in the modelling is related to the domestic hot water consumption and the resulting power requirement. By measuring hot water consumption, modelling accuracy could probably be significantly improved. Based on the reviews, it could also be verified that the energy efficiency measures taken have had an impact on the peak power needs of residential buildings as well as on total energy consumption. For all three buildings, peak power needs appear to have decreased and overall energy consumption is lower than it would have been without the actions taken.