滑坡变形的支持向量机非线性组合预测

在分析支持向量机(SVM)用于时间序列预测和非线性组合原理基础上,提出基于支持向量机的非线性组合预测方法。利用4种单项预测方法,包括SVM、径向基函数前向型神经网络(RBF)、反馈型神经网络(El-man)及3层神经网络(ANN),分别进行滑坡变形时序的建模与预测。对4种方法的预测结果再采用线性组合方法(简单平均、方差倒数、改进最优加权系数)和非线性组合方法(SVM、BP神经网络)进行组合预测及方法性能的比较。结果表明,非线性组合的平均相对误差明显低于线性组合方法,且对滑坡这种具有一定程度不确定性的非线性系统,SVM的非线性组合方法有着更理想的预测效果,7步外推预测准确度控制在89.3%以上。而与BP神经网络非线性组合相比,SVM也具有更好的稳健性和泛化性。

Nonlinear Combination Predicting Based on Support Vector Machines for Landslide Deformation

A predicting method of nonlinear combination based on support vector machines(SVM) is presented.In this paper,a time series of landslide deformation modeling and prediction are preformed by four signal predicting methods,including the SVM,RBF network,Elman recurrent neural network and ANN,and these predicting results are combined to predict landslide displacement again by applying linear combination methods and nonlinear combination methods.A comparison of these methods is made for their predicting ability.The results show that nonlinear combination outperforms linear combination.And nonlinear combination of the support vector machine is of higher precision,and the accuracy of extrapolated prediction is up to 89.3% within 7 steps.Also,this method is of greater stability and generalization as compared with the BP network nonlinear combination method.