基于人工神经网络的结构随机响应分析

Structure random response analysis based on artificial neural net

结构的随机响应分析由于涉及到大量的样本,而使得有限元计算过程大量重复,尽管通过正交设计可减少分析样本数目,效率仍十分低下.采用BP人工神经网络这一映射迅速的运算方法,通过计算少量样本对应的有限元结果,进行训练,将训练好的网络用于结构的随机响应分析.算例计算和工程应用表明,在整个计算过程中,误差均在|E|=0.000 1范围内,神经网络的训练时间约占一次有限元计算时间的15%,因此对于三维问题利用神经网络可节省大量时间.

The process of finite element computing has to be repeated as many times as the number of samples in random response analysis of structure. Usually, the number of samples is very large even by adopting weightily sample technique, so the computing efficiency is slow. The back propagation artificial neural net is applied in random response analysis of structure. Thus, the process of finite element computing repeat several times and train BP net between the random input variables and the corresponding response parameters. The BP net trained can compute the response of other samples. The process of computing based on BP net shows that the error is within the scopes of ｜ E ｜ =0. 000 1. In the whole computing process, and training time of the neural net is abou 15% of the finite element computing time. For three-dimensional problem, neural net can save a lot of computing time.