基于神经网络的可靠性分析新方法

对于大部分工程可靠性分析中的隐式极限状态方程，建立一种基于样本筛选的神经网络可靠性分析方法。该方法利用具有强大的非线性映射能力的神经网络，来近似隐式极限状态方程中的输入变量与输出变量的关系，从而使得隐式可靠性分析转化为显示可靠性分析问题，大大减少了计算失效概率的工作量。与已有的神经网络可靠性分析方法相比，所提方法选择近似极限状态方程而不是近似极限状态函数的策略，并通过筛选训练样本实现这一策略，从而提高可靠性分析的精度，算例结果充分显示所提方法的优越性。

NEW RELIABILITY ANALYSIS METHOD BASED ON ARTIFICIAL NEURAL NETWORK

For implicit limit state equations in most engineering reliability analysis,a new method is presented on the basis of artificial neural network(ANN),where the training samples are appropriately selected.Due to the powerful tool of function approximation,ANN is employed to obtain the relationship of the input parameters and the output parameters in the implicit limit state equation.The reliability analysis for the implicit limit state is then transformed to that for the explicit limit state,and the computational efforts are greatly decreased.Comparing with available reliability analysis based on ANN, the presented method has a different strategy on selection of the training samples,in which the implicit state equation can be more appropriately approximated.The precision of the presented method is higher than that of the available method,and this advantage is illustrated by examples.