PSO-SVM算法在土体参数反演中的应用  

Application of PSO-SVM Algorithm in Soil Parameters Inversion

作　　者：沙勇华 陈志坚[1] Sha Yonghua Chen Zhijian(School of Earth Science and Engineering, Hohai Universit)

机构地区：[1]河海大学地球科学与工程学院,南京市211100

出　　处：《勘察科学技术》2017年第3期45-49,共5页Site Investigation Science and Technology

基　　金：江苏省政策引导类计划(产学研合作);编号:BY2015002-05

摘　　要：结合单桩静力载荷试验的桩传力机理,提出了PSO-SVM算法与有限元联合反演模型。运用PSO优化的SVM预测算法解决了在选择惩罚系数和核函数参数方面的难题。同时基于正交试验与有限元分析获得了神经网络训练样本,建立起土体力学参数与桩顶竖向位移之间的高度非线性映射关系。利用实测位移预测反演得到了土体参数,将其带入正演运算,并与实测位移曲线进行了相似度计算。研究表明PSO-SVM算法在确定土体参数上有一定的适用性,同时PSO-SVM算法与有限元联合反演模型的建立也为参数反演问题的解决提供了一种新的思路。Combined with the pile-transferring force mechanism of single pile static load test, a joint in- version model of PSO-SVM algorithm and finite element is established. The SVM prediction algorithm op- timized by PSO is applied to solve the difficult problems on choosing the penalty coefficient and the kernel function parameter. Meanwhile, the neural network training samples is obtained based on the orthogonal experiment and FEA. A high nonlinear mapping relation between the soil mechanic parameters and the vertical displacement of pile top is built. The soil parameters is predicted and inverted by measured dis- placement, and bring it in the forward operation, the similarity with the measured displacement curves is calculated. The results show that PSO-SVM algorithm has certain applicability in determining soil param- eters, meanwhile, the established joint inversion model of PSO-SVM algorithm and finite element provid- ed a new way for solving the problems of parameters inversion.

关 键 词：PSO-SVM算法 有限元 土体参数反演 单桩静力载荷试验 

分 类 号：TU43[建筑科学—岩土工程] TU473.1[建筑科学—土工工程]