Data-driven analysis in building energy consumption

Abstract

In a global effort focusing on driving sustainability, the building sector has been widely recognised to have the most potential in reducing building energy consumption and greenhouse gas emission. Technology advancement in the building sector ensures that building data tracking is available, which allows people to study the relationship and characteristics of variables that would ultimately impact a building’s energy consumption. Therefore, it is essential to provide an accurate and efficient means to analyse such data to allow for an accurate representation of building energy consumption.

This study made use of Seattle’s building energy consumption data in 2019, published by Seattle’s government as a case study to develop a data-driven prediction and optimisation approach, to evaluate the city’s energy consumption, in the form of Site Energy Used (SiteEU), Site Energy Used Intensity (SiteEUI) and Total Greenhouse Gas Emissions (Total GHG Emissions). As there are multiple variables that contribute to energy consumption in a building and they are highly complex and have a non-linear relationship, an Artificial Neural Network (ANN) regression model is proposed to predict the building energy consumption. The prediction results from the ANN model show a coefficient of determination value, R2 , for SiteEU, SiteEUI and Total GHG Emissions to be 1.0, 0.87 and 0.3 respectively. These results show that ANN is a good model when it is used to predict SiteEU and SiteEUI, while there is room for improvement in the prediction of Total GHG Emissions. Furthermore, feature of importance of each model is analysed to determine the impact each variable has on the objectives, and it is found that the gross floor area of a building and electricity used in a building contributes to predicting SiteEU, SiteEUI and Total GHG Emissions to a greater extent than the other variables used.

A multi-objective optimisation approach using NSGA-II is then proposed to optimise both SiteEUI and Total GHG Emission, using the variables that are identified that have the most significant influence on them, from the feature of importance analysis. The implementation of the NSGA-II shows an overall improvement of 46.4% for both objectives after optimisation.