Research On Carbon Emission Trend Prediction In The Construction Industry Based On Spatiotemporal Graph Neural Networ

The global carbon emission crisis is becoming increasingly dire,and China has taken steps to encourage the green and low-carbon growth of its various industries.In particular,the construction sector has been responsible for more than one-third of the total carbon emissions,and it is becoming a trend to reduce these emissions.Therefore,it is essential to analyze the carbon emissions of buildings.This study endeavors to explore the accounting and forecasting of carbon emissions in the building life cycle,despite the intricate and costly nature of the current approach to calculating construction carbon emissions,which necessitates a vast amount of databases for backing.This thesis commences with a comparison of pertinent research outcomes both domestically and internationally,based on literature reading.It then examines the technique for constructing life cycle carbon emission assessment based on life cycle evaluation theory,which encompasses accounting scope,inventory model,activity analysis and data sources.Moreover,it examines the significance and accounting scope of life cycle carbon emission from construction materials,as well as establishing the carbon emission accounting model.Screening the relevant literature for influencing factors related to the study of carbon emissions in the construction industry,gray correlation analysis was employed to identify the most important variables.This was based on the research above.The carbon emission prediction model of the construction industry takes into account GDP,gross output value,labor productivity,and the number of employees in construction enterprises as its input variables.This thesis then builds a carbon network from the perspective of non-Euclidean spatial data set based on the results of better carbon emission accounting and by analyzing the structure of data between provinces and cities.In response to the disregard of spatial features in prior studies,this thesis constructs a prediction study of carbon emissions in the construction industry from two temporal and spatial perspectives,utilizing a spatio-temporal graph neural network model.Three models-ARIMA,the classic one-are compared and examined by three metrics: average relative error,average absolute error,and R2.The traditional prediction model’s three error indicators are more pronounced than those of the spatio-temporal graph neural network model,and its prediction results are more significantly divergent from actual values,implying that the spatio-temporal graph neural network prediction model is superior.This thesis’ s research results offer a potential for gauging carbon emissions in the construction sector,as well as a reference point and practical direction for further investigations into other structures’ carbon emissions.