| 基于XGBoost-SHAP方法的建设项目碳排放空间异质性分析 |
| 摘要点击 160 全文点击 11 投稿时间：2024-06-04 修订日期：2024-08-22 |
| 查看HTML全文
查看全文 查看/发表评论 下载PDF阅读器 |
| 中文关键词 碳排放 空间异质性 XGBoost算法 SHAP算法 可解释性 |
| 英文关键词 carbon emission spatial heterogeneity XGBoost algorithm SHAP algorithm interpretability |
|
| 中文摘要 |
| 为使公路建设碳减排更有效，聚焦高速公路建设过程中的碳排放空间异质性，基于广东省A高速公路项目40个分段样本筛选出的构造物类型、桥隧比、设计坡度、路线长度、填方量、挖方量和水泥消耗量这7个碳排放影响关键指标，训练与验证了XGBoost碳排放预测模型，构建了解释这40个路段碳排放空间异质性的SHAP算法，研究了路段特征对碳排放的影响、总特征贡献和特征交互效应. 结果表明，水泥消耗量的增加对碳排放的非线性增长贡献最大，路线长度、挖方量和桥隧比对碳排放的贡献度也较为显著；冷热点分析发现坡度高于2.5%且地形复杂的路段碳排放趋高，存在聚集效应；XGBoost-SHAP模型较地理加权回归模型GWR能更清晰解释碳排放的空间分布特征及其影响因素，在捕捉关键碳源和理解碳排放空间分布特征方面表现更佳. 基于以上发现，提出了公路建养碳减排的针对性综合策略，以推动公路建设的可持续发展. |
| 英文摘要 |
| To make the carbon emission reduction of highway construction more effective， this study focuses on the spatial heterogeneity of carbon emission in the process of highway construction. Based on the seven key indicators of carbon emission impacts of the type of structure， bridge-to-tunnel ratio， design gradient， route length， fill volume， excavation volume， and cement consumption screened out from the 40 segment samples of highway project A in Guangdong Province， we trained and validated the XGBoost carbon emission prediction model and constructed the SHAP algorithm to explain the impacts， total feature contributions， and feature interaction effects of the features of these 40 road segments. The SHAP algorithm was constructed to explain the spatial heterogeneity of carbon emissions of these 40 road sections， and the influence of road section features on carbon emissions， total feature contribution， and feature interaction effect were investigated. The results showed that the increase in cement consumption contributed the most to the nonlinear increase of carbon emission， and the route length， excavation volume， and bridge-to-tunnel ratio also contributed significantly to the carbon emission. The analysis of hot and cold spots revealed that the carbon emission tended to be higher in the road sections with a gradient higher than 2.5% and with complex topography， and there exists an agglomeration effect. The XGBoost-SHAP model explained the spatial distribution of carbon emission more clearly than the geographically weighted regression model （GWR）； the model captured the characteristics and their influencing factors and also the spatial distribution and effects on carbon emission. The XGBoost-SHAP model could explain the spatial distribution of carbon emissions and its influencing factors more clearly than GWR and performed better in capturing the key carbon sources and understanding the spatial distribution of carbon emissions. Based on the above findings， this study proposes a comprehensive strategy for carbon emission reduction in highway construction and maintenance to promote the sustainable development of highway construction. |
