Research On The Stability Analysis And Landslide Prediction Of Soil Slope

Landslide is a common type of the abrupt geological hazards.And it seriously damages the safety of people's lives and property.With the unreasonable development of mineral resources,the number that landslide happened and the loss that landslide disasters caused become more and more serious.Therefore,it became an urgent problem to predict landslide hazard and reduce all kinds of losses caused by landslide hazard by putting forward more effective methods.Modern landslide prediction research mainly includes the combined application of multiple forecasting methods and the application of mathematical.In this paper,the improved GM(1,1)model and the optimized BP neural network model are used together to predict the displacement.And the obtained displacements are compared with the displacements obtained from the slope stability analysis to get the state range of the slope,and then to predict the safety state of the slope.In the process of landslide early warning,The arrangement of monitoring points and the treatment and utilization of monitoring data obtained by monitoring points are the decisive factors of early warning.In order to make the displacement monitoring data obtained by monitoring points reflects the sliding characteristics of the slope more appropriately,to improve the effectiveness of monitoring,this paper summarizes the monitoring method and monitoring points layout principle.It can not be explained that when the gray GM(1,1)prediction model to take 0.5,the prediction of slope displacement has the highest accuracy.This paper puts forward two improved methods of optimal weight selection and real-time data correction,which can effectively improve the prediction accuracy of the prediction model with the residual model.In order to solve the low convergence rate of BP artificial neural network and easy to fall into local minimum,this paper optimized the BP neural network model with the additional momentum method,the gradient descent method for adaptive lr and the elastic gradient descent method.Combined the improved GM(1,1)model and optimization of the BP neural network model into Grey BP neural network residual error model.Based on the initial prediction using the GM(1,1)model,the BP neural network model is used to predict the residual sequence,and then generate the new predicted sequence.To separately test the accuracy of the forecast data obtained from the original model and the improved combined model.The prediction accuracy of the model is improved obviously,and it is proved that the grey BP neural network residual error model is effective to predict the displacement.In the previous research on landslide prediction,after the prediction of slope displacement,only to predict the displacement of the value and change the rate of qualitative analysis,so as to determine the stability of the slope.Such predictions are too subjective to consider the actual internal strength state of the slope.In this paper,we simulate the different reduction of strength parameters of slope,to obtain the combination form of strength parameters by analyzing the stability of the slope using the finite element software when the slope just destroy.And comparing the obtained displacements with the displacements obtained from the slope stability analysis,The condition of the slope at the next time point is the condition whose displacement most close to the value.The concept of "state line" and "safety line" is put forward to judge the stability of slope in order to achieve the function of early warning.