Predicting Deformations of Tunnel Surrounding Rock by Using Least Squares Support Vector Machine

Abstract

In order to overcome the disadvantages of high computational complexity and inconvenience when forecasting deformations of surrounding rock by using support vector machine of standard form (Vapnik SVM), a new deformation prediction method based on least squares support vector machine (LS-SVM) was presented. By using this method, the excavated rock mass was regarded as a time-dependent system with high uncertainty and a sliding time window was employed at first to select learning examples, then the examples obtained was used for training the corresponding LS-SVM prediction model. Finally the proposed method was applied to forecast the surrounding rock deformations of Xuejiazhuang Tunnel. The result shows that the method has relatively high prediction accuracy and therefore it is a feasible deformation prediction method with low computational complexity.