基于地理探测器的自重黄土湿陷性预测模型研究  

Research on Collapsibility Prediction Model of Self-weight Loess Based on Geographic Detector

作　　者：李国华 周爱红 曹聪 袁颖 张淼 LI Guohua;ZHOU Aihong;CAO Cong;YUAN Ying;ZHANG Miao(Hebei GEO University,Shijiazhuang 050031,China;Hebei Technology Innovation Center for Intelligent Development and Control of Underground Built Environment,Shijiazhuang 050031,China;Shanxi Institute of Geological Survey Company Limited,Taiyuan 030006,China)

机构地区：[1]河北地质大学城市地质与工程学院,河北石家庄050031 [2]河北省地下人工环境智慧开发与管控技术创新中心,河北石家庄050031 [3]山西省地质调查院有限公司,山西太原030006

出　　处：《河北地质大学学报》2024年第6期84-92,共9页Journal of Hebei Geo University

基　　金：山西省国土资源厅省级地质勘查项目(SXZDF20170820);河北地质大学科技创新团队项目(KJCXTD-2021-08)。

摘　　要：为精确地利用物理力学性指标预测黄土的自重湿陷性系数,以山西省转型综合改革示范区——中部产业整合区自重湿陷性黄土为例,利用地理探测器分析了研究区13项物理力学性质指标与自重湿陷系数之间的关系,并通过相关性验证从而筛选出主要影响指标,以此建立机器学习预测模型。结果表明:研究区自重黄土湿陷性主要影响指标从高到低依次为孔隙比、干密度、天然密度、饱和度、含水率、液性指数和取样深度;通过建立的机器学习模型对比,其模型准确性从高到低依次为极致梯度提升模型、随机森林模型、梯度提升树模型和决策树模型。极致梯度提升模型真实有效性为83.24%,能够满足实际工程需要,对黄土湿陷性研究及相关工程实践具有一定的借鉴意义。In order to predict the self-weight collapse coefficient of loess using physical and mechanical indicators rapidly and effectively,this paper takes the loess of the Central Industrial Integration Zone in the Transformation and Comprehensive Reform Demonstration Area of Shanxi Province as an example.The study utilized geographic detector to analyze the relationship between 13 physical and mechanical property indicators in the research area and the self-weight collapsibility coefficient.Through correlation verification,the main influencing indicators were identified,and a machine learning prediction model was established.By comparing the established machine learning models,their accuracy ranges from high to low as follows:extreme gradient boosting model,random forest model,gradient boosting decision tree model,and decision Tree model.The real effectiveness of the extreme gradient boosting model is 83.24%.Therefore,the extreme gradient boosting model established can meet the practical engineering requirements and has certain reference significance for the study of loess collapsibility and related engineering practices.

关 键 词：自重湿陷性黄土 物理力学性质指标 地理探测器 交叉验证 

分 类 号：TU444[建筑科学—岩土工程]