多项式基函数神经网络的结构可靠性分析

  目的  随着液化天然气（LNG）船舶结构研究和设计深度的提高，需要有能够快速和准确地评估不确定性因素的可靠性分析方法。为此，提出基于改进随机森林−蒙特卡罗（RF-MC）法来解决A型独立液舱支座结构失效概率的计算问题。  方法  首先，根据不确定性因素的概率分布，使用MC法生成样本集；然后，以局部离群因子为准则，筛选出失效面附近的样本点，再对筛选出的样本点进行有限元计算后添加至训练集，通过重复训练随机森林近似模型，直至满足精度要求；最后，使用近似模型判别样本点是否失效，结合MC法计算结构的失效概率。  结果  综合考虑算法的准确率、复杂度和效率并结合算例1和2，可以发现在分析可靠性问题时改进RF-MC法比MC和BP-MC等方法具有更大优势。算例3的应用结果表明了改进RF-MC法在A型独立液舱支座结构可靠性分析中的适用性。  结论  研究结果可为LNG船舶的优化设计提供可行的技术方案。

Structural reliability analysis based on polynomial basis function neural network

  Objectives  In response to the increasing depth of research and design on liquefied natural gas (LNG) ship structures, higher requirements are put forward for a reliability analysis method that can quickly and accurately evaluate uncertain factors. This paper proposes a method based on an improved random forest-Monte Carlo method (RF-MC) to solve the calculation of the failure probability of A-type independent liquid tank support structures.  Methods  First, the MC method is used to generate a sample set according to the probability distribution of uncertain factors, then take the local outlier factor (LOF) as the criterion for filtering out sample points near the failure surface. After selecting the sample points, they are calculated using finite element software and added to the training set to train the random forest (RF) model. The generation, filtering and training process is repeated until the approximate model meets the accuracy requirements. Finally, the approximate model is used to determine whether the sample points are invalid, then combined with the MC method to calculate the failure probability of the structure.  Results  Considering the accuracy, complexity and efficiency of the algorithm, and combined with Cases 1 and 2, it is found that the improved RF-MC method has better advantages than MC or biased probability (BP)-MC in analyzing reliability problems. The results of Case 3 show applicability of the method in reliability analysis of an A-type independent liquid tank support structure.  Conclusions  This study provides a feasible technical solution for future optimization design of liquefied gas carriers.