基于支持向量机回归的黄土湿陷性预测研究

打开文本图片集

关键词：黄土湿陷性；物性指标；相关性；支持向量机；预测模型

中图分类号：TU444 文献标志码：A 文章编号：1005-8249（2025）03-0073-05

DOI：10.19860/j.cnki.issn1005-8249.2025.03.014

Abstract：Toadressthechallenges inrapidandacurate predictionof loess collapsibilityin Central Gansuusingconventional methods，asupportvectormachineregression-based predictive modelwas developed.Leveragingdata fromarailwayprojectin thisregion，soil samplescollctedfrom manuall excavated testpitsunderwentcomprehensivelaboratorygeotechnicaltestingto obtainkey physical propertyidices.Correlations between losscollpsecoeffcientsandphysicalindicesweresystematicaly investigated through literatureanalysisandmathematical statistics.AGaussankemelfunction-basedSVMpredictionmodel wassubsequentlyestablished.The modelwastrainedwith45experimentaldatasets，while15datasetswereutilized forvalidationand eroranalysis.Resultsdemonstratethatthe modelachievesameanabsoluteeror（MAE）ofO.OO3，meanbias eror（MBE）ofO.Ol，androt meansquare error（RMSE）of O.O159onthetraining set.Forthetesting set，performance metrics indicate MAE=0.0132 ， MBE=0.0018 ，and RMSE=0.015 .These findings confirm the model’s high predictive accuracyfortimatinglosscollapsecoefcientsinCentralGansu，providingareliableovelmethodologyforpracticalegining applications.

Key words：loess colapsibility；physical property index；correlation；support vector machine；prediction model

0 引言

湿陷性土壤在受到地下水或降雨作用后，会出现显著的变形，这对基础设施、房屋及其他工程建设构成潜在威胁[1]。（剩余4952字）