Abstract: A method of back analysis of initial stress fields from displacement is presented based on a kind of modified support vector machine named ?-SVR and a kind of modified genetic algorithms. By using orthogonal experiment method, only several positive analysis are done. ?-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 ?-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.

Key words: support vector machine(SVM), modified genetic algorithms, initial stress fields, back analysis, orthogonal experiment, ?－SVR, stochastic back analysis