神经网络方法在位移反分析中的应用研究

为解决传统位移反分析建模复杂、求解速度慢等问题，基于MATLAB的二次开发语言M语言，编写了用于位移反分析的BP神经网络源程序．针对传统BP网络收敛速度慢的缺点，采用L—M优化算法及归一化方法来加快网络的训练速度．对圆形巷道弹塑性问题的力学参数进行了反演分析．反演结果表明，MATLAB与神经网络相结合应用于位移反分析，具有建摸快、模型简捷、求解速度快、精度高等优点，可以在工程实际中推广应用．

Application Study of the Neural Network Approach in Back Analysis of Displacements

In order to resolve the problems of conventional back analysis of displacements for complex modeling and slow speed in problem-solving,the program of BP neural network compiled by the M language of MATLAB was used for back analysis of displacements.Aiming at the disadvantage of slow-footed convergence of the traditional BP neural network,Levenberg-Marquardt optimization arithmetic and normalization method were used to quicken the network training rate.They were successfully used in the inverse analysis of the mechanical parameters of a standard elastic-plastic problem about underground engineering.Contrasted with the traditional methods,the method of artificial neural network is fast in model building and calculation,brief in model structure,and high in precision etc.So it can be used for back analysis of displacements in engineering.