基于均匀设计及遗传神经网络的大坝力学参数反分析方法

Back analysis on mechanical parameters of dams based on uniform design and genetic neural network

将均匀设计理论、BP神经网络和遗传算法三者结合起来,应用于大坝力学参数反分析中。首先对基本遗传算法进行改进,使得改进后的遗传算法具有很好的全局和局部寻优能力,将它作为BP神经网络的学习算法,形成遗传神经网络。然后利用均匀设计方法设计大坝坝体和坝基的材料力学参数样本,通过有限元正分析得到坝体的计算位移样本,训练遗传神经网络映射坝体计算位移值与材料力学参数之间的复杂非线性关系。最后将实测位移值输入训练好的遗传神经网络,即可得到各参数的反演值。本文以清江隔河岩水电站重力拱坝为例,反演分析了坝体混凝土的弹性模量、线膨胀系数以及坝基主要岩体的弹性模量等参数。经检验、评价与对比验证,结果表明该方法可以大大地缩短反分析时间,提高反分析效率和准确性。

A new approach for back analysis of mechanical parameters of dams combining BP neural network with uniform design and genetic algorithm was established. Firstly, basic genetic algorithm was improved for very good global and local searching capability. The genetic neural network using the improved genetic algorithm as its learning algorithm was established to overcome the shortcoming of BP algorithm. Secondly, the sample of material parameters was designed by the uniform design method, and the sample of the calculated displacement of dams was obtained by use of the finite element method. Through these samples, the above genetic neural network was trained to describe the sophisticated nonlinear relationship between displacement and material parameters of dams. Finally, the actual dam displacement was input into the trained genetic neural network to obtain the real material parameters. As an example, the elastic moduli and the linear dilatable coefficient of concrete of the dam body and the elastic moduli of rockmass under the base of Geheyan Dam were back-analyzed by the above method. It was shown that the method could shorten the time of back analysis and improve the efficiency and accuracy of back analysis.