基于小波支持向量机的围岩力学参数辨识  

Mechanical parameters identification of surrounding rock based on wavelet SVM

作　　者：王复明[1] 李晓龙[1] 苗丽[1] 徐平[1] 

机构地区：[1]郑州大学水利与环境学院,郑州450002

出　　处：《水力发电学报》2010年第3期184-190,共7页Journal of Hydroelectric Engineering

基　　金：河南省杰出人材创新基金(074200510006);河南省重大科技攻关项目(092101510100)

摘　　要：核函数形式对支持向量机(SVM)推广能力有重要影响,目前基于SVM的位移反分析中常用的径向基核函数(RBF)由于不能通过平移形成平方可积空间上的完备正交基,使得相应的支持向量机难以逼近该空间上的复杂多维函数,进而影响参数辨识的精度。小波函数恰具备这种特点,为此引入一种多维允许支持向量核函数——Littlewood-Paley小波函数,该核函数能够以其平移正交性逼近平方可积空间上的任意函数,从而提升支持向量机的推广性能;将Littlewood-Paley小波核函数与最小二乘支持向量机(LS-SVM)相结合形成一种新的岩体力学参数辨识方法,并应用于某抽水蓄能电站洞室围岩参数识别。分析结果表明,与RBF核函数相比,采用小波核函数的LS-SVM具有更高的识别精度,而且利用实测变形数据得到的辨识结果与试验值也较为接近,证明了反演结果的可靠性和所建议方法的有效性。The form of kernel function influences much on the generalization ability of a support vector machine（SVM）.By using RBF kernel function in common use for displacement back analysis at present,however,it is impossible through translation to construct a complete orthogonal basis for the square integrable space,so it is difficult to approximate a complicated function in this space and to achieve a good precision of parameter identification.Fortunately,wavelet functions have many excellent features,particularly the multidimensional admissive SVM kernel functions such as Littlewood-Paley wavelets that can approximate through expansion and translation an arbitray square integrable function with high precision and enhance generalization ability of the resulted SVM.This paper proposes a new back analysis method of rock mechanical parameters based on Littlewood-Paley wavelet kernel functions and least squares SVM（LS-SVM）.Application of this LS-SVM model to identification of the rock parameters for a pumped storage power station demonstrates a better accuracy than that of RBF kernels.The LS-SVM back analysis using the measured data of rock displacements shows the inversion parameters close to the experimental values,which verifies the reliability and the validity of the proposed method.

关 键 词：岩土力学 力学参数识别 反分析 围岩 小波核函数 

分 类 号：TU45[建筑科学—岩土工程]