An interpretable machine learning framework for residential building hourly operational carbon emission prediction with indoor and outdoor environment

Operational carbon emission prediction is the essential component of building carbon emission management. Hence, an interpretable machine learning framework is proposed for residential building hourly operational carbon emission prediction with indoor and outdoor environment parameters. In the proposed framework, a dynamic correlation graph is introduced to structure building carbon emission and environment parameter time series, which used to represent the relation among environment parameters and building carbon emission. The TCN-GCN algorithm and SHAP method are utilized to construct the hourly operational carbon emission prediction model for residential buildings. To validate the feasibility and effectiveness of the proposed method, case study are implemented with the experiment dataset from Smartline project in Cornwall, UK. The experiment results demonstrate the proposed DCG-TCN-GCN method can accurately predict residential building operational carbon emission with 0.7578∼0.8416 Corr, and 0.1288∼0.1827 RSE. The research results can provide benefits to more sustainable energy-efficient and low-carbon residential buildings. • An interpretable building carbon emission prediction framework is proposed. • DCG is structured to represent the correlation among environment and carbon emission. • TCN-GCN is utilized to construct hourly operational carbon emission prediction model. • SHAP is applied to explain the established machine learning model.