ANALYZING METHOD OF DEFORMATION INSTABiILLY MODE OF DANGEROUS ROCKS AT CLIFF FACES BASED ON L-M ARTIFICIAL NEURAL NETWORK

Clearing deformation instability mode of dangerous rocks is the premise of prevention and treatment research, but traditional forecasting methods have shortcomings of high-cost and weak practicality, especially under strong earthquake. in this paper, Adopting methods of neural network structure based on Levenberg-Marquardt and three dimensional discrete element numerical simulation, also considering deformation instability mode of dangerous rocks under four main varying conditions, including joint dip angle, length of rock, depth-width ratio and stacking layers of dangerous rocks, and taking deformation instability mode of dangerous rocks for research objects and influencing factors as the breakthrough point, the BP neural network model to predict deformation instability mode of dangerous rocks is established. Then the neural network is trained with samples of deformation instability mode of dangerous rocks calculated by numerical simulation. Finally, the paper analyzes the accuracy of the prediction model. the results show that this model is provided with good learning and generalization ability, and the accuracy of prediction is 86.7%. All these indexes validate that neural network based on Levenberg-Marquardt method to predict deformation instability mode of dangerous rocks is effective and feasible.