基于序列代理模型的结构可靠性分析方法

工程中部件或系统的性能函数通常是隐函数形式，该文提出基于高效代理模型的结构可靠性分析方法，构造新增样本点学习函数，并用于指导在序列迭代过程中的样本点选择。所提出的学习函数考虑了变量权重并保证所选的样本点相互之间有一定的距离且分布在极限状态方程周围。算例分析表明该方法有较好的精度和鲁棒性，不仅适用于性能函数为隐函数时的结构可靠性分析，而且也适用于现有的各种代理模型（神经网络、支持向量机、响应面等），为结构可靠性分析提供新途径与新方法。

Structural Reliability Analysis Using Sequential Surrogate Models

Performance functions of components and structural systems are often given using implicit functions for many practical engineering. An efficient surrogate-models-based reliability analysis method is proposed in this paper, a new sample selection learning function is constructed as a guideline to adaptively select new sample point at each iteration. The proposed learning function considers the weights of variables and ensures that the selected sample points reside not only around the limit-state functions, but also far away each other. The numerical examples show that the proposed method is accuracy and robustness, thus it can be used for structural systems with implicit performance function and various existing surrogate models (e.g., neural networks, support vector machine, response surface model). The proposed method provides a novel method for structural reliability analysis.