Deciphering Exterior: Building Energy Efficiency Prediction with Emerging Urban Big Data

In the UK, 28 million households consume 25% of the total energy and contribute to 25% of the carbon emissions. It is vital to focus on sustainability and energy efficiency within the building sector for decarbonizing purposes. However, traditional methods that involve detailed building simulation or on-site inspections are time-consuming, labor-intensive, and expensive. Recent advancements in high-resolution satellite/aerial thermal infrared images, street view images, widely available building attributes, and deep learning methodologies offer the potential to estimate building energy efficiency at a large scale. In this research, we propose a novel methodology framework for classifying energy efficiency in buildings using only external and widely existing building data. We have designed and trained an end-to-end multi-channel deep learning model utilizing high-resolution thermal infrared and optical remotely sensed images, street view images, socio-economic indicators, and building morphological data. Building energy performance datasets from Glasgow and Edinburgh were organized for this study as validation purposes. Applying the trained model, our workflow achieved F1 scores of 0.64 and 0.69 for the two cities, respectively. Further predictions and analyses reveal a surprising correlation between neighborhood-level building energy efficiency and socioeconomic deprivation, suggesting that more deprived neighborhoods tend to have better building energy efficiency. This research demonstrates how external building data and deep learning can be leveraged to assess building energy efficiency and has the potential to be applied globally to address the net-zero carbon agenda for sustainable living in the future.