Study On Stability Evaluation Of Rock Slope Based On BP Neural Network

In this thesis,the stability of highway rock slope is studied.Based on the BP artificial neural network,the rock slope stability prediction model of rock slope is established,and the stable slope angle is predicted.Meanwhile,the stability of rock mass is graded with RMR rock mass grading system.Through investigations,the engineering data from a large number of highway slope projects are collected and then used as training samples for the established BP artificial neural network.The influence factor of the input layer of the model is based on the parameters of the RMR rock mass grading system,and the slope engineering survey data are quantified according to the parameter scoring standard.For the rock slope with multiple sets of joints,the values of the initial parameters are determined respectively by the methods of arithmetic average adjustment,weighted average adjustment,and weighted average adjustment with wedges.Then the adjusted parameters are input to the BP neural network model to predict the stable slope angle.The predicted values are compared with the actual values of existing slopes.When the prediction value of a slope angle is larger than the actual value,the slope stability cannot be directly judged by comparison.In such cases,the finite element method is adopted to analyze the slope stability and verify the reliability of the predicted value.The main research results are as follows:(1)When the values of the parameters according to the RMR rock mass classification system are determined by the methods of weighted average adjustment with wedges,the results of rock mass rating are the closest to the engineering reality.(2)The predicted values are compared with the actual values of existing slopes.The results show that it is feasible to predict the slope angle of highway rock slope by BP artificial neural network(3)The finite element method analysis shows that the results obtained by BP neural network are correct,which verifies the reliability of the BP neural network method.(4)When the values of the parameters in the RMR rock mass classification system are determined by the methods of weighted average adjustment with wedges,the predicted value given by BP neural network is closest to the actual value.