Comparison of regression models for estimation of carbon emissions during building's lifecycle using designing factors: a case study of residential buildings in Tianjin, China

Many studies have been conducted on life cycle assessment and control measures for carbon emissions of buildings. Methods proposed by these studies usually require not only specific accounting model, but also detailed inventory data, which is not available at early design stage. Seeing that the importance of design phase to carbon emissions during building's lifecycle, a study on regression model of carbon emissions using designing factors was done. Firstly, based on process analysis method, the carbon emissions of 207 residential buildings in Tianjin were calculated. The results show that annual carbon emissions per floor area are between 30 and 60 kgCO(2)/(m(2).year), with manufacture phase and operation phase accounting for 11%-25% and 75%-87%, respectively. Then, correlation analysis and elastic net were used to determine 12 designing factors for predictive model; At last, four regression techniques, PCR, RF, MLP and SVR were used to develop regression models, respectively; comparison and process analysis of model development were given later. The results show that SVR has the optimal predictive accuracy among four models, its corresponding coefficient of determination can reach to 0.800. This regression model can be utilized to estimate carbon emissions based on designing factors, which can help designers make a strategic decision at early stage. (C) 2019 Elsevier B.V. All rights reserved.