基于神经网络和遗传算法的岩体参数辨识研究

本构模型和初始参数给不准已成为岩土力学理论分析和数值模拟的“瓶颈”。目前神经网络可能会避开复杂的本构模型而成为解决岩土工程的设计和施工问题的主要途径。然而,不同的神经网络结构对模型的学习和推广预测能力有直接的影响,因此构造一个令人满意的神经网络模型至关重要。本文结合遗传算法对神经网络结构进行优化,并用后验差检验方法对搜索到的神经网络结构进行测试;结合遗传算法,用测试好的神经网络结构代替常用的有限元数值方法进行岩体参数的辨识;最后,用实例较为详细地证明了这种方法的可行性。

The Research of Parameter Identification of Rock Mass Based on Artificial Neural Network and Genetic Algorithm

In the solution of Geotechnical,it is intractable to ascertain constitution model and parameter of rock and soil exactly. Nowdays, Artificial Neural Networks enable us to avoid the complex constitution model to resolve the problem in Geotechnical. The different Artificial Neural Networks,however,are influential to the ability of model in study and extension. Therefore, it is very important to constitute a satisfactory Artificial Neural Network. Genetic Algorithm is used to search most appropriate BP neural network structure. Back test is used to test the result. Finite Element Method substituted with BP Neural Networks to back-analyze parameter of rock and soil. Finally,the feasibility is proved by an example.