基于网格搜索的支持向量机砂土液化预测模型

Support vector machine model for predicting sand liquefaction based on Grid-Search method

在使用支持向量机对砂土液化进行预测时,不可避免地会遇到惩罚因子和核函数参数如何选取的问题,目前还没有确定这两个参数的选择模式,解决这一问题比较常用的办法有经验公式法、遗传算法和网格搜索法。对此本文基于网格搜索方法,使用LIBSVM软件包,寻找砂土液化训练样本的结构风险最小值所对应的支持向量机最优参数;使用最优参数,建立了支持向量机砂土液化预测模型。研究结果表明:预测样本的正确率可达87.5%,而且预测结果稳定;同时通过比较BP神经网络的砂土液化预测情况,可知支持向量机有更好的泛化能力,收敛速度也更快。

When the support vector machine is used for predicting sand liquefaction,the problem how to choose penalty factor and parameter of kernel functionis inevitably encountered,and the usual method for solving this problem is the empirical formula method,the genetic algorithm method and the grid-search method.This paper uses software called LIBSVM and the grid-search method to search parameters of the minimum structure risk of SVM through training the sample of sand liquefaction.Then,a predicting model of SVM with parameters,which have been obtained above,is set up to predict the sample of prediction,and the results show that the correct rate is 87.5%.Through comparing with the prediction of sand liquefaction by BP Network,the support vector machine shows better generalization ability and faster convergence speed.