土体残余强度的PSO-LSSVM模型

利用最小二乘支持向量机(LSSVM)建立土体残余强度模型,以液限、塑性指数、粘粒含量和偏差等为输入变量,通过改变输入变量的结构建立2个LSSVM模型,并采用粒子群优化(PSO)算法设定模型参数,分别预测残余摩擦角值,并与实验值、人工神经网络(ANN)模型作比较,得出LSSVM模型具有较好的效果,另外对LSSVM的输入变量进行敏感性分析,得出偏差对模型的影响最大,印证文献中结论并说明模型的合理性。

PSO-LSSVM Model for Residual Strength of Soil

This paper shows a least squares support vector machine (LSSVM) method for predicting the residual strength of soil.The inputs of the LSSVM model consist of four variables,including liquid limit (LL),plasticity index (PI),the clay fraction (CF) and the deviation.By changing the structure of the input variables, two LSSVM models are developed. By using particle swarm optimization (PSO) algorithm for setting the parameters of the LSSVM model,the LSSVM model can predict the residual friction angle.Compared with the experimental values and predicted values of artificial neural network (ANN) models,the LSSVM models work better.Sensitivity analyses have also been performed on the input variables.It concludes that deviation has the biggest effect,which verifies the conlusions in previous articles,and explains the validity of the model.