结构系统可靠度分析的SVC抽样迁移算法

为有效避免结构系统可靠度计算过程中复杂的约界分析处理,针对系统可靠度问题多失效模式的固有特点,引入了系统极限状态曲面的概念,并利用支持向量分类算法(support vector classification, SVC)对该失效曲面进行了直接重构.在此基础上,结合LHS(Latin hypercube sampling)抽样迁移策略,提出了计算结构系统可靠度的SVC抽样迁移算法.通过对比分析两个典型算例表明:本文算法具有较高的抽样效率和收敛性能,与传统Monte Carlo法相比,其抽样工作量减少87%,计算结果相对误差不超过1%,且可有效避免现有β约界算法中需要人为假定失效状态的缺陷,更适用于实际结构可靠度问题的分析求解. 

Support Vector Classification Algorithm by Migration Sampling for Structural System Reliability Evaluation

In order to avoid the cumbersome bounding operation during system reliability evaluation procedures, a concept of system limit state surface was introduced on account of the characteristic of system multi-failure modes. Employing the support vector classification (SVC) algorithm to divide the security domain and failure domain, reconstitution of the system limit state surface could be achieved. On this basis, a migration sampling strategy using Latin hypercube sampling (LHS) was combined together with the SVC algorithm, finally leading to a new system reliability evaluation method of SVC migration sampling. Two widely used illustrative examples were introduced and analyzed by different methods for comparison. The results show that the SVC migration sampling algorithm possesses good sampling efficiency as well as superior convergence property. Compared with traditional Monte Carlo sampling methods, the algorithm presented can reduce sampling times by 87%with a relative error of less than 1%. The gross assumption of component failure state by β-unzipping method can also be avoided, which is more suitable for practical application to actual structure assessment.