基于KPCA-GPC的地震砂土液化预测  

Prediction model of seismic-induced sand liquefaction based on KPCA-GPC principle

作　　者：赵国彦[1] 彭俊 刘建 ZHAO Guoyan;PENG Jun;LIU Jian(School of Resources and Safety Engineering, Central South University, Changsha, Hunan 410083, China)

机构地区：[1]中南大学资源与安全工程学院,湖南长沙410083

出　　处：《中国地质灾害与防治学报》2017年第4期130-136,共7页The Chinese Journal of Geological Hazard and Control

基　　金：国家自然科学基金面上项目(51374244)

摘　　要：在砂土液化多指标综合判别法中,采用的判别因子数量一般为5~12,为防止特征冗余,引入核主成分分析(KPCA)对原始样本进行非线性特征提取,同时基于高斯过程分类(GPC)原理,构建了砂土液化预测的KPCA-GPC模型。以唐山地震砂土液化的25个案例为样本,选取平均粒径D50、地下水位dw、标准贯入击数N63.5、砂层埋深ds、地震烈度I、震中距离L、不均匀系数Cu、剪应力与有效上覆应力比τ_d/σ'v共8个指标作为判别因子,对该模型进行验证。研究结果表明:距离判别分析(DDA)、高斯过程分类(GPC)、Seed法的判别准确率分别为83%、83%、67%,而KPCA-DDA和KPCA-GPC的判别准确率均为100%,由此说明消除特征冗余的必要性及KPCA-GPC模型的适用性;同时,与DDA、SVM、BP等确定性判别方法相比,GPC可获得具有概率意义的砂土液化预测结果。In multi-index comprehensive discrimination method for sand liquefaction, the number of discriminant factors are generally 5 to 12. In order to prevent feature redundancy resulted by too many discriminant factors, kernel principal component analysis （KPCA） was introduced to extract the non-linear feature of raw data, then based on Gaussian process classification （GPC） principle, the prediction model of KPCA-GPC for sand liquefaction was constructed. Combined with 25 cases of sand liquefaction in Tangshan earthquake, eight factors were selected for model verification listed as follows： average particle size Ds0, groundwater table dw, standard penetration blow count N63.5, sand depth ds, magnitude I, epicentral distance L, coefficient of non-uniformity Cu, ratio of shearing stress to effective overburden stress ~＇a/o＂v. The results show that the accuracy of DDA （distance discriminant analysis） method, GPC method and Seed method are 83% , 83% and 67% , respectively. The accuracy of KPCA-DDA method and KPCA-GPC method are all 100% , it indicates the necessity of reducing feature redundancy of raw data and the applicability of the KPCA- GPC method. Compared to deterministic discrimination method such as DDA, SVM（support vector machine） and BP（neural network）, the GPC model can produce the prediction results of sandy soil liquefaction with probability significance which helps people very much in practical engineering.

关 键 词：砂土液化 特征冗余 核主成分分析 高斯过程分类 预测模型 

分 类 号：P642.116[天文地球—工程地质学]