基于MLS-SVM的结构整体可靠度与全局灵敏度分析

As one of efficient surrogate models, the support vector machine (SVM), which is based on the principle of structural risk minimization in statistical learning, has been widely used in structural reliability analysis with implicit limit state functions. However, the traditional support vector machines still have many shortcomings, such as the selection of kernel function, global basic variable space modeling, computational efficiency, etc. In order to overcome these shortcomings, this paper proposes a new support vector machine model based on moving least squares (MLS) technology named MLS-SVM. With this model, the training sample sets can be adaptive in the global basic variable space. Then this model is applied to global reliability and sensitivity analysis of reinforced concrete (RC) frame structures, and then is compared with the support vector machines based on the regenerative kernel function (RK-SVM) and the least square technique (LS-SVM). It is shown by the numerical results that, compared with the two comparative models, the MLS-SVM model has higher accuracy and better computational efficiency.