基于支持向量机的隧道变形预测方法

Method of Predicting Tunnel Deformation Based on Support Vector Machines

提出一种基于支持向量机的隧道变形预测新方法。支持向量机基于结构风险最小化,具有更强的泛化能力,是一个凸二次优化问题,能够保证所得解就是全局最优解。采用RBF和Bspline核函数学习某隧道前30天的收敛监测数据,用学习得到的最佳支持向量机网络预测30天后隧道的收敛。结果表明,支持向量机回归和预测的最大相对误差不超过6 5%。通过对比发现,Bspline核函数比RBF核函数效果更好。

A new method for predicting tunnel deformation based on supporting vector machines (SVM) is presented. It has stronger generalization ability because the SVM theory is based on the principle of minimizing structure risks. The algorithm of SVM is a convex quadratic optimization problem. Therefore it is guaranteed that the solution is a global optimum solution. The Radical Basic Function and Bspline kernel function are adopted to study the convergence data of the first 30 days of a tunnel. The best SVM network obtained is used to predict the convergence of the remaining days. The results show that the maximum regression and prediction relative errors are not greater than 6.5%. Through this comparison it is found that the Bspline kernel function has better effects than the RBF kernel function.