Author(s): Li Jingbo; Tang Daochu; Jiang Yazhou; Bai Runbo
Linked Author(s):
Keywords: Rock parameter; Wall rock; Deformation prediction; Finite difference method; Neural networks
Abstract: Rock parameters are the fundamental components of the constitutive law of the wall rock in the diversion tunnel. The determination of rock parameters based on real observations of wall rock deformation plays a crucial role for effective deformation prediction of diversion tunnel. In this study, a neural computing method is proposed to determine the rock parameters. First, the real observations of wall rock deformation are acquired by using modern measure devices. Second, an empirical finite difference model is established in an attempt to predict these real observations. The difference between the outputs and the real observations of wall rock deformation indicates the accuracy of the finite difference model. Though adjusting the rock parameters, a group of finite difference models can be created, leading to a group of deformation predictions. Thus, the rock parameters and the corresponding predictions of deformation form a nonlinear mapping. Third, a neural network model is proposed to characterize the nonlinear mapping by using pairs of rock parameter and deformation prediction, given by the finite difference model. The well trained neural network model extracts the underlying relation between the rock parameter and deformation. Finally, the real observations of the wall rock deformation are fed into the neural network model to give rise to the rock parameters. The obtained rock parameters can support a reasonable constitutive model that can be used for deformation prediction. The effectiveness of the proposed method is validated in the deformation prediction of the diversion tunnel of the Wudongde hydropower station.
Year: 2013