山西转型综改示范区黄土湿陷性指标分析及其预测模型研究  

Analysis of Loess Collapsibility Indices and the Prediction Models in Shanxi Transformation and Comprehensive Reform Demonstration Zone

作　　者：李国华 周爱红 袁颖 黄虎城[3] 曹聪 LI Guohua;ZHOU Aihong;YUAN Ying;HUANG Hucheng;CAO Cong(School of Urban Geology and Engineering,Hebei GEO University,Shijiazhuang,Hebei 050031;Hebei Technology Innovation Center for Intelligent Development and Control of Underground Built Environment,Shijiazhuang,Hebei 050031;Shanxi Institute of Geological Survey Company Limited,Taiyuan,Shanxi 030006)

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：《地质与勘探》2024年第6期1246-1256,共11页Geology and Exploration

基　　金：山西省国土资源厅省级地质勘查项目(编号:SXZDF20170820);河北省教育厅科学研究项目(编号:ZD2022094);河北地质大学科技创新团队项目(编号:KJCXTD-2021-08)联合资助。

摘　　要：黄土湿陷会引发地面沉降、路基失稳、建筑物坍塌等,严重影响城市经济发展与工程建设适宜性。因此,查明黄土湿陷性影响因素,分析黄土湿陷性指标,进而建立黄土湿陷性预测模型,可为饱受黄土湿陷性困扰地区的地质灾害防治及工程建设提供依据。本文以山西省转型综改示范区—中部产业整合区黄土为研究对象,开展研究区原状土室内土工试验,得到71组黄土物理力学性质指标,根据黄土湿陷系数与13项物理力学性质指标间皮尔逊相关系数,找出该地区黄土湿陷性显著相关指标。在此基础上,分别选取因子分析方法消除显著相关指标间相关性前后两组参数,建立该地区黄土湿陷性预测的线性回归和机器学习模型。结果表明:研究区易溶盐和中溶盐离子含量较大,存在黄土湿陷隐患;黄土湿陷系数与孔隙比、干密度和天然密度相关系数在0.701~0.707之间,具有强相关性。通过对研究区建立的线性回归模型和机器学习模型综合对比发现,消除相关性前后的线性回归模型其有效性分别为80.95%、85.71%,BP神经网络模型为80.95%、71.43%,随机森林模型为90.48%、90.48%,说明随机森林模型不受指标间相关性影响,且在黄土湿陷性预测方面具有更高的显著性、准确度和适用性。因此,本次研究建立的随机森林预测模型有效性能够满足实际工程需要,可用于该场地黄土湿陷性预测,对黄土湿陷性研究及相关工程实践具有借鉴意义。Loess collapsibility can lead to ground subsidence,roadbed instability and building collapse,which will severely impact the suitability of urban economic development and engineering construction.It is therefore necessary to identify the influencing factors,indices and prediction models of loess collapsibility,in order to provide scientific reference for geohazard prevention and engineering construction in areas plagued by loess collapsibility.This work took the loess in the Central Industrial Integration Zone of Shanxi transformation and comprehensive reform demonstration zone as a reseach object.Indoor geotechnical tests were conducted on undisturbed soil within the research area,and 71 sets of physical and mechanical property indices for loess were yielded.According to the Pearson correlation coefficients between the collapsibility indices of loess and 13 physicomechanical indices,this study identified significant parameters notably correlated with loess collapsibility in the region.On that basis,the factor analysis method was employed to eliminate primary influencing factors derived from the correlation among these indices.Subsequently,linear regression and machine learning methods were employed to establish predictive models for loess collapsibility in the region.Results indicate that this area contains a large concentration of easily soluble salts and moderately soluble salt ions,posing a risk of loess collapsibility.The correlation coefficients between the loess collapsibility coefficient and the void ratio,dry density,and wet density range between 0.701 and 0.707,indicating a strong correlation.A comprehensive comparison of the established linear regression model and machine learning model suggests that,the effectiveness of the linear regression model before and after eliminating correlations is 80.95%and 85.71%,respectively,that of the BP neural network model is 80.95%and 71.43%,respectively,and that of the random forest model is 90.48%and 90.48%,respectively.The random forest model is unaffected by inter

关 键 词：黄土湿陷性 物理力学性质指标 相关性 因子分析 预测模型 转型综改示范区 山西省 

分 类 号：TU444[建筑科学—岩土工程]