Studies Of ANN Back Analysis In Rock Displacement Prediction Of LiuShan Tunnel And Its Application

The mechanical parameters are very important in the calculation of geotechnical engineering. During the construction, the design can be timely amended according to mechanical parameters from back analysis. Of course, this can reduce cost. However, the studies of back analysis in the traditional geotechnical engineering need tack into account the deformation mechanism of the surrounding rock, and need use elastic-plastic constitutive relation. Therefore, the relations between back-analysis factors are extremely complicated, and parameter solving has difficulties too much. So we had to leave out much relative factors. No knowing nonlinear -relationship between deformation and mechanical parameters, we can solve question of back analysis. But because of high nonlinear function, learning ability, associative memory, and strong capacity of error tolerance of artificial neural network, it is able to construct the mapping between mechanical parameters and displacements from practical samples.This paper take LiuShan tunnel in the Hang-Hui express highway for example, and fit the measured data of the deformation of the tunnel, and make a statistic analysis to its residual, and thus find the distributed regulation of the measured data of the rock. Firstly, the author selects Ansys8.1 finite elements as platform, and takes the mapping relationship between mechanical parameters and displacements of the tunnel as analytical samples of the neural net wok. Based on those samples, the nonlinear mathematical model can be constructed through the technical means of learning and training of the neural network. Secondly, take the deform value from LiuShan tunnel into the nonlinear model which have been constructed through ANN. So the values of the surrounding rock parameters have been obtained. Lastly, based on the results of back analysis in this thesis, the author predicted the deformation of another rock with the same type. The result indicated that predicted values fitted well actually measured values.The researches show that the function between surveying data and physical parameters is indirectly established, depending on ANN. Therefore, it is easy for technical staffs to master and use back analysis, because they needn't learn profound elastic-plastic mechanics and finite elements theory. This theory of back analysis provides a new means and a good method for further research.