滑坡位移预测的支持向量机模型参数选择研究

支持向量机(Support Vector Machine, SVM)已被广泛应用到滑坡位移预测,但在具体使用时,SVM的惩罚系数C、核函数参数δ及松弛系数ζ这三个重要参数的取值选择成为影响预测精度的关键。为有效分析SVM三参数取值对滑坡位移预测精度的影响规律,以三峡库区浮托减重和动水压力型两类典型水库滑坡为代表的连续6年地表位移、降雨及库水位监测数据为研究对象,首先,采用移动平均法将位移数据分解为趋势项和周期波动项,并区分训练集和检验集;再结合对滑坡变形机理及影响因素的分析,选择相应预测变量分别建立趋势项和波动项位移预测SVM模型;然后,在固定两参数情形下,通过改变另一参数的取值大小以获得SVM训练集与检验集的预测精度变化规律;最后,建立起典型水库滑坡SVM位移分解预测的参数取值推荐范围。该取值范围可以作为滑坡位移预测SVM模型的参数寻优初始搜索范围,可以在保证预测精度的前提下大大提高搜索效率。

Study on Parameter Value Selection of Support Vector Machine for Displacement Prediction of Landslides

Support Vector Machine (SVM) has been widely used in the prediction of landslide displacement, and the value selection of three import parameters (penalty factor C, kernel function parameter δ and relaxation factor ζ) was the key point affecting the prediction accuracy. In order to effectively analyze the law of the influence of value selection of above three parameters on landslide displacement prediction accuracy by using SVM model, six consecutive years of monitoring data, which include surface displacement, rainfall capacity and reservoir water level of two typical reservoir landslides in Three Gorges reservoir area are used. Firstly, the moving average method is adopted to decompose the cumulative displacement into trend term and periodic term of displacement, at the same time, the training and test data set are built; then based on the analysis of landslide deformation mechanism and influential factors, the SVM prediction models of trend term and periodic term of displacement are established according to corresponding forecast variables; Next, the relationship between parameter C,δ,ζvalue with prediction accuracy of the training and test data is analyzed; Finally, the three parameters value range are recommended, and which can be used as the initial search range of parameters optimization for the SVM model of landslide displacement prediction to enhance the search efficiency with guaranteed prediction accuracy.