| dc.description.abstract | Whole-life carbon emissions (WLCE) studies play a crucial role in understanding the environmental impact of buildings and promoting sustainable building design practices. This thesis addresses significant research gaps in the existing studies of WLCE in the building sector, such as the lack of understanding of variations in WLCE of similar properties, the absence of statistical methods to analyse key factors contributing to variations, and the need for interdisciplinary research on machine learning algorithms for WLCE prediction. These gaps underscore the necessity of this research and are filled through a comprehensive set of three objectives in this thesis. First, the WLCE of 145 similar residential properties in proximity in the UK was calculated using the process-based life cycle analysis method and the available high-resolution electricity consumption data. Second, statistical analysis was conducted to identify key factors contributing to variations in the WLCE of these similar properties in proximity, using data readily available from previous research on building attributes and occupant characteristics collected from questionnaires and surveys. Carbon reduction strategies were also proposed based on the results of the analysis. Lastly, WLCE prediction models were built using various machine learning algorithms and building descriptor factors, with performance evaluation carried out.
The key findings include: (1) WLCE of similar residential properties in proximity varies significantly from 21 to 193 t CO2eq, with WLCE intensity ranging from 0.5 to 2.6 t CO2eq/m2; (2) Statistical analysis indicates that factors related to building attributes and occupant characteristics significantly influence WLCE variations, with energy-efficient building design and low-carbon occupant lifestyles emerging as useful reduction strategies; and (3) Machine learning algorithms demonstrated accurate and efficient WLCE prediction capabilities using limited building descriptor factors, with Random Forest performing the best for both WLCE and WLCE intensity prediction.
These findings illustrate the importance of recognising significant WLCE variations among similar buildings in proximity, as individual property attributes and occupant characteristics are crucial for accurate life cycle analysis. They also underscore the need for data-driven approaches to predicting buildings’ WLCE. The successful implementation of machine learning algorithms demonstrates that it is feasible to use them to estimate building WLCE in the early design stages. Overall, this research provides solid evidence to support data-driven decision-making in sustainable building design, promotes a life cycle perspective, and contributes to carbon reduction goals in the building sector through innovative artificial intelligence methods. | en_GB |
