基于对偶神经网络与直接积分法的结构可靠度计算方法

he structural Reliability analysis considered of random factors is paid widely attention by engineers and scholars due to reflecting the structural characteristics and the bearing actual situation. As an important part of the structural reliability analysis, reliability calculation exists the problem that the calculation amount and the calculation accuracy are difficult to be taken into account together, so it seriously impacts on the promotion and practical application of reliability theory. The direct integration method of structural reliability starting from the definition of reliability theory is easy to been understood by the engineering and technical personnel, but there are still mathematics difficulties in the calculation of multiple integrals. So a dual neural network method has been present for calculating multiple integrals by this paper. Dual neural network consists of two neural networks. The neural network A is used to learn to the integrand function and the neural network B is used to simulate the original function. Using the derivative relationship between the network output and the network input, the neural network B is derived from the neural network A. On this basis, this method combines with the performance function of integral regional rules to overcome the multiple integration difficulties and to achieve the precise calculation of the reliability. Compared with Monte Carlo simulation method (MCS), Hasofer-Lind method (HL) and first-order second moment method (FOSM), the proposed method is an efficient and high-precision method to solute the structural reliability of random variables.