Understanding Building Energy Efficiency in Glasgow by Multi-source Deep Learning Framework

Keywords: Building Energy Efficiency, EPC, Deep Learning, Google Street View
Session Type: Virtual Paper Abstract
Day: Tuesday
Session Start / End Time: 3/1/2022 02:00 PM (Eastern Time (US & Canada)) - 3/1/2022 03:20 PM (Eastern Time (US & Canada))
Room: Virtual 20

Abstract

With buildings consuming nearly 40% of energy in developed countries, it is important to estimate the building energy efficiency and understand what elements contribute to it. In this study, we propose a data-fusion framework to accurately predict building energy efficiency. We take into account richer influencing factors and study the contribution of images reflecting the appearance of buildings, as a complement to traditional building characteristics, to predicting their energy efficiency. The data used includes property structural attributes, morphological attributes and building facade images. With these data, we would like to build a comprehensive understanding of what affects building energy usage. We choose Glasgow, UK as a study site with more than 160, 000 properties and 30,000 buildings. We get the energy efficiency, morphological and structural attributes from the Scottish Energy Performance Certificates (EPC) data set, and facade images of buildings corresponding to each property are retrieved to provide another layer of information. The experiment is conducted on the property level and achieved a validation accuracy of 88.2%. We also compared the performance improvements between our data-fusion framework with traditional morphological attributes. The result shows that with building facade images, validation accuracy increases from 79.6% to 88.2%. Our research demonstrates the potential of using multi-source data in building energy efficiency prediction and helps understand building energy efficiency for future studies in Net Zero Carbon policy-making