In order to better solve granular flow program PFC2D, meso-mechanical parameters was select in the process of simulation calculation, based on PFC2D programs, in order to improve the traditional method of BP neural network, established in granular flow simulation analysis of rock mass macro mechanical parameter and meso-mechanical parameters of nonlinear models, entering a macro mechanical parameter, can be fast, accurate retrieval of meso-mechanical parameters. This study concluded that improved BP neural network model for feature extraction of samples apply to instances, access to knowledge, limestone can be realized between the macro-and meso-mechanical parameters of mechanical parameters of linear maps. By improved BP neural network model to enter macro mechanical parameter of limestone, rapid inversion of meso-mechanical parameters of limestone; the inversion results as calculation parameters, macro mechanics parameters of computational value and experimental value obtained compared to accuracy in general, higher than 90%; if the RES is equal to 10, when the BP network model has a 6 neurons in the hidden layer, inversion of its best performance. It also shows that the improvement of BP neural network model for determination of meso-mechanical parameters of limestone has a good retrieval performance, popularization and application of the method for particle flow theory provides a new technical means.
Rock areas of research and forecasting disasters depend on the kinematic behavior of the rock, usually by conventional macro analysis method for the study of rock as a whole, on the microscopic changes in the components of the rock particles, and the composition of cracks, etc. few studies. Particle flow method of non-continuous media theory, is the macroscopic mechanical parameters of the particles and micromechanics parameters linked to simulate the interaction among the particles (Miao et al. 2012).
P.A Cundall (Wang et al. 2013) and O.D.L Strack using discrete element theory developed microscopic particle flow program PFC, PFC is that based on the media to describe the basic mechanical properties, to from the contact between the state and the variation of particle research, therefore, PFC is very suitable for the mechanical properties of heterogeneous materials in rock and soil mass study (Sekins et al. 2004). Currently in the field of numerical simulation of geotechnical, mining and materials have been widely used.