结合专家知识的住宅建筑物化阶段碳排放机器学习预测研究  

Prediction of carbon emissions by machine learning in the materialization stage of residential buildings incorporating with expert knowledge

作　　者：王志强 曹永祺 李硕 任金哥 WANG Zhiqiang;CAO Yongqi;LI Shuo;REN Jinge(School of Management Engineering,Qingdao University of Technology,Qingdao 266520;Smart City Construction Management Research Center(New Think Tank),Qingdao 266520;Shanghai Electric Power Design Institute Limited Company,Shanghai 200000;Xi'an University of Architecture and Technology,Xi'an 710055,China)

机构地区：[1]青岛理工大学管理工程学院,青岛266520 [2]智慧城市建设管理研究中心(新型智库),青岛266520 [3]上海电力设计院有限公司,上海200000 [4]西安建筑科技大学,西安710055

出　　处：《干旱区资源与环境》2025年第1期38-48,共11页Journal of Arid Land Resources and Environment

基　　金：国家自然科学基金项目(71471094)资助。

摘　　要：“双碳”目标下,针对设计阶段有效预测并控制住宅建筑物化阶段碳排放的问题,根据93个住宅建筑的特征数据,采用AHP-DEMATEL和机器学习算法,构建了主客观相结合的碳排放预测模型,并利用SHAP算法从全局和单个特征角度剖析其作用程度和交互效果。研究表明:ET模型的预测效果最好;除主客观排序差异明显的两个特征外,总建筑面积、建筑高度和建筑高宽比在综合排序中影响最大;多特征交互分析较好地给出了各特征值域变动范围与碳排放总量的关系,为住宅建筑碳排放预测及碳减排工作提供指导。Under the"dual-carbon"goal,the problem of how to effectively predict and control carbon emissions in the materialization stage of residential buildings at the early design stage is addressed.Based on the feature data of 93 residential buildings,AHP-DEMATEL and machine learning algorithms are used to construct a subjective-objective prediction model.The SHAP algorithm is used to analyze the degree of its role and interaction effects from the perspective of global and individual features.The study shows that the ET model gets the best prediction effect.Except for the two features with obvious differences in subjective-objective ranking,the total building area,building height and building aspect ratio have the greatest influence according to the combined ranking.The multi-feature interaction analysis gives a better relationship between the range of changes in the value domain of each feature and the total carbon emissions,which provides guidance and helpfulness for carbon emission reduction in residential buildings.

关 键 词：住宅建筑物化阶段 碳排放 AHP-DEMATEL 机器学习 SHAP 

分 类 号：F407.9[经济管理—产业经济]