支持向量机近似模型的参数选取及其在结构优化中的应用

针对非线性结构响应预测的支持向量机(SVM)近似模型的参数选取问题,提出了应用粒子群算法进行参数优化,建立了具有最优参数的SVM近似模型,并与以训练集数据建立常规的SVM、二阶响应面(RSM)和径向基神经网络(RBFNN)近似模型进行对比.结果表明:以优化参数建立的SVM近似模型比常规的SVM近似模型有更好的预测能力;可以避免RSM和RBFNN近似模型中的过拟合现象,具有更优的推广能力.最后,将最优参数的SVM近似模型用于船舶结构优化中,取得了具有良好工程实用性的优化结果.

Parameter Selection for Support Vector Machine and Its Application in Structural Optimization

To select proper parameters for the support vector machine (SVM) regression model used for the prediction of non linear structural response, the particle swarm optimizer was introduced into parameter optimization. To make comparisons, the SVM regression model with regular parameters, the RSM and RBFNN regression models were also developed based on the training data set. The results show that the SVM regression model based on optimized parameters has a better prediction ability than the regular SVM and can solve the overfitting problem in the regression model developed by the response surface method and radial based function neural network, thus possessing better generalization ability. The application of the SVM with optimimized parameters in structural optimizations proves that it has good engineering practicability.