Bayesian optimization + XGBoost based life cycle carbon emission prediction for residential buildings—An example from Chengdu, China: Fid-Bau Portal

Bayesian optimization + XGBoost based life cycle carbon emission prediction for residential buildings—An example from Chengdu, China

Pan, Haize / Wu, Chengjin

Abstract

The large amount of carbon emissions generated by buildings during their life cycle greatly impacts the environment and poses a considerable challenge to China’s carbon reduction efforts. The building design phase has the most significant potential to reduce building life-cycle carbon emissions (LCCO₂). However, the lack of detailed inventory data at the design stage makes calculating a building’s LCCO₂ very difficult and complex. Therefore, accurate prediction of building LCCO₂ at the design stage using relevant design factors is essential to reduce carbon emissions. This paper proposes an ensemble learning algorithm combining Bayesian optimization and extreme gradient boosting (BO-XGBoost) to predict LCCO₂ accurately in residential buildings. First, this study collected and calculated the LCCO₂ of 121 residential buildings in Chengdu, China. Second, a carbon emission prediction model was developed using XGBoost based on 15 design factors, and hyperparameter optimization was performed using the BO algorithm. Finally, the model performance was evaluated using two evaluation metrics, coefficient of determination (R²) and root mean square error (RMSE), and the prediction performance of other models was compared with that of the BO-XGBoost model. The results show that the RMSE of the proposed BO-XGBoost for predicting LCCO₂ in residential buildings is at least 40% lower compared to other models. The method adopted in this study can help designers accurately predict building LCCO₂ at the early design stage and provide methodological support for similar studies in the future.