基于ν-SVR和GA的初始地应力场位移反分析方法研究

Study on ν-SVR and modified GA in back analysis of initial stress fields from displacements

基于一种改进的支持向量机(ν-SVR)和改进的遗传算法(GA),提出一种初始地应力场位移反分析方法。该方法通过正交设计方法安排较少次数的正分析方案,用ν-SVR对正分析中测点位移值与参数构成的样本集进行学习,建立参数取值与观测点测值的非线性隐式方程;采用确定性或随机反分析的思路确定反分析的优化目标函数,并利用GA搜索最优参数取值。采用ν-SVR方法建立的参数取值与位移量测点测值的非线性隐式方程,能够以很高的精度拟合和预测不同参数取值时的观测点测值,因此,可以用该隐式方程代替正分析,减少计算量;采用改进的GA方法能够准确搜索到最优参数;搜索到的最优参数值与理论值相当吻合。算例表明,基于ν-SVR和GA的初始地应力场位移反分析方法是一种行之有效的初始地应力场位移反分析方法,可以广泛用于初始地应力场确定性反分析和随机反分析。

A method of back analysis of initial stress fields from displacement is presented based on a kind of modified support vector machine named v-SVR and a kind of modified genetic algorithms. By using orthogonal experiment method, only several positive analysis are done. v-SVR used the measuring points＇ displacement and parameters of positive analysis as samples to build the nonlinear implicit equations between them. Then modified genetic algorithms are used to find the optimal parameters on deterministic or stochastic optimal object function. The equations built by v-SVR can fit and forecast the measuring points＇ displacements under different parameters with high precision. So these equations can substitute positive analysis, and the amount of computation will reduce greatly. With modified genetic algorithms, the optimal parameters can be found accurately. Example shows the method presented is an effective back analysis method for initial stress fields from displacements. It can be widely used in deterministic or stochastic back analysis of initial stress fields from displacements.