Using Genetic Algorithm to Optimize Parameters of Support Vector Machine and Its Application in Material Fatigue Life Prediction

Abstract

Support vector machine is a new kind of learning method based on solid theoretical foundation, but this method has the characteristic of sensitivity to parameter. According to this characteristic, this paper use genetic algorithm to optimize the parameters of SVM and cross validation is introduced to reduce the dependence of the parameters on the training samples. Through the analysis of fatigue data for the relevant literature, take the parameters of the best generalization ability as the final parameters and apply the obtained model (GA-SVR) in material fatigue life prediction. Compared with the conventional SVR model and PSO-SVR model, the mean square error and the square of correlation coefficient are used to verify the reliability and accuracy of the three models. The results show that, the GA-SVR model can predict the fatigue life of materials with high accuracy.