基于支持向量机替代模型的可靠性分析

摘要

This paper establishes the surrogate-based reliability analysis method for the implicit performance functions using the support vector regression algorithm and classification algorithm, in which Monte Carlo simulation method is integrated with the Latin hypercube sampling technique. The effects of the related parameters on the SVM performance are discussed and the genetic algorithm is used to optimize the parameters to provide a rational selection of SVM model. The efficiency of the SVM model with different sampling size is discussed, which testifies the good performance of SVM model with small sampling size. Finally, the numerical examples indicate the feasibility and efficiency of the proposed method.%建立了基于支持向量机回归算法和分类算法的替代模型可靠性分析方法，与蒙特卡罗法结合，采用拉丁超立方抽样技术，进行隐式极限状态函数的可靠度计算。讨论了相关参数对支持向量机模型性能的影响，并通过遗传算法进行参数优化，为支持向量机模型的参数选择提供了依据。研究了不同训练样本数量对支持向量机模型预测值精度的影响，进一步证实了支持向量机的小样本特性。算例结果表明了本文方法的有效性和可行性。