基于遗传算法-神经网络混合训练技术的结构近似可靠分析方法

ATUDY ON THE QPPROXIMATE STRUCTURAL RELIABILITY ANALYSIS METHOD BASED ON GA-NN COMBINED TECHNOLOGY

根据多层神经网络映射存在定理，提出了基于遗传算法－神经网络混合训练技术的结构近似可靠性分析方法，数值试验表明，该方法建立的神经网络模型可以很好地拟合真实的极限状态函数。

The approximate structural reliability analysis method based on the genetic algorithms (GA) -neutral network (NN) combined technology is Proposed. The method includes three Parts. First, the structure is analyzed to get a study sample. Than we optimize the weight of NN by using GA to get the best weight for the study sample. Last, the structural reliability index by using GA is calculated. Since the NN may forms an arbitrary curve, the method may fit the real limit state curve than the response-surface method. Besides, since the method bases on the GA, the study efficiency of the NN is raised and the calculation of structural reliability index not needs the function constraint condition such as the continuity, the limit value, and the derivative existence The numerical results show that the neutral network model founded by this method fit with the real limit state curve.