基于混合核函数PSO-LSSVM的边坡变形预测

支持向量机（SVM）的核函数类型和超参数对边坡位移时序预测的精度有重要影响。鉴于局部核函数学习能力强、泛化性能弱，而全局核函数泛化性能强、学习能力弱的矛盾，通过综合两类核函数各自优点构造了基于全局多项式核和高斯核的混合核函数，并引入粒子群算法（PSO）对最小二乘支持向量机（LSSVM）超参数进行全局寻优，提出了边坡位移时序预测的混合核函数PSO-LSSVM模型。将模型应用于锦屏一级水电站左岸岩石高边坡变形预测分析，并与传统核函数支持向量机预测结果进行对比分析。结果表明，该模型较传统方法在预测精度上有了明显提高，预测结果科学可靠，在边坡位移时序预测中具有良好的实际应用价值。

Forecasting of slope displacement based on PSO-LSSVM with mixed kernel

The kernel and the parameters of support vector machine(SVM) have a significant impact on precision of the time series prediction to slope displacement.In view of better learning capability of local kernels and better generalization capability of global kernels,the mixed kernel is constructed by a typical local kernel-radial basis function(RBF) and a typical global kernel-polynomial kernel.By use of particle swarm optimization(PSO),a new PSO-LSSVM model regression with mixed kernels is set up in this paper and applied to the left bank slope in Jinping Ⅰ Hydropower Station.Through comparing with the forecasting results of the existing SVM based on RBF,results demonstrate that the new model has great accuracy than the existing SVM with RBF only and has real application value in predicting deformations of slope.