Research On Prediction And Forecast Of Slopes Deformation

A slope is continuously deformed under the influence of external forces,and a landslide hazard will be occurred when the deformation value exceeds the limit of the bearing capacity of the slope.Therefore,prediction and forecasting the deformation value of slopes is one of the effective means to prevent landslide disasters.This thesis focuses on the prediction methods of slope deformation,which is based on the existing monitoring data of slope.The main research results and innovations of the thesis is as follows.(1)Three single prediction models,including grey GM(1,1)model,grey Verhulst model and BP neural network model,as well as two combined prediction models,including grey Markov model and grey neural network model,are studied from the aspects of principle,derivation process and modeling steps.Then the advantages and disadvantages of the above models are analyzed one by one.(2)Aiming at the problem that the existing grey prediction model cannot update the modeling data,which results in the influence of the old information on the accuracy of the model,two improved models,Grey GM(1,1)Dynamic model(GDM(1,1)model)and Grey Verhulst Dynamic model(GVD model),based on dynamic updating of the modeling data with variable length sliding window,are proposed.The two models are validated by the deformation monitoring data of the B3 monitoring point of Xintan landslide.The experimental results show that the two models are effective.The fitting accuracy and prediction accuracy of the improved model are better than those of the traditional grey model,and the improvement effect is obvious.In order to further improve the accuracy of GVD model and its adaptability to fluctuation data,a Grey Verhulst Markov Dynamic model(GVMD model)combining GVD model with Markov chain is proposed.The validity of the improved model is verified by the deformation monitoring data of Xintan landslide.(3)In view of the fact that the gray model in the traditional gray neural network model usually adopts the gray GM(1,1)model,while the research on the combination of the gray Verhulst model and the BP neural network is less.Therefore,a Residual error correction Grey Verhulst Neural Network(RGVNN)model is proposed,in which GVD model is used instead of grey GM(1,1)model,and BP neural network is used to correct the prediction residual of GVD model.The RGVNN model is established based on the monitoring data of open pit slope,and the prediction results are compared with those of grey Verhulst model and BP neural network.The results show that,compared with the grey Verhulst model and BP neural network,the mean relative error of RGVNN model is reduced by 96.3% and 76.7%,respectively.The prediction accuracy of RGVNN model is much better than that of grey Verhulst model and BP neural network.(4)An Auxiliary Grey Neural Network Optimizd by Mind Evolutionary Algorithm(MEA-AGVNN)model is proposed to solve the problem that the weights and thresholds of Grey Verhulst Neural Network(GVNN)model random initialization cause the prediction accuracy to decrease.MEA-AGVNN model introduces Mind Evolutionary Algorithm(MEA)to optimize the weights and thresholds of the network.The optimized weights and thresholds are taken as the initial weights and thresholds of Auxiliary Grey Neural Network(AGVNN).MEA-AGVNN model is established based on the deformation monitoring data of open pit slope,and the prediction results are compared with AGVNN model.The experimental results show that the mean relative error of MEA-AGVNN model is 46.9% lower than that of AGVNN model.The prediction accuracy of MEA-AGVNN model is better than that of GVNN model.