The Research Of Landslide Forecast

Landslide was effected by both nonlinear factors like internal stratigraphic distribution characteristics and lithology, but also by random factors such as external rains, earthquakes and vegetation, the deformation complex process of landslide evolution was full of uncertainty and the randomness, gradual change and mutagenicity. The frequent occurrence of landslides caused heavy losses of life and property of the people, as possible to avoid or mitigate loss of life and property, landslide forecast highlighted a crucial role. In different types of landslide deformation process, its displacement-time curve was different during the various stages of landslide bred and sliding; Therefore, the development and changes of displacement of each different type of landslide deformation stage restricted the selection of landslide prediction model.This paper, based on the "Western Transportation Construction Technology Project-"Yunnan Highway Flood Hazard dynamic monitoring and early warning technology research"",by selecting Ding Jiafen typical landslide as research work sites, applied the established Ding Jiafen landslide deformation monitoring data to the demonstrated gray GM (1,1) model, collaborative forecasting model for slope instability, BP neural network model, the dynamic fractal dimension tracking prediction model, time series R/S model, utilized the fuzzy evaluation model deformation forecasting model to conduct a comprehensive assessment of different forecasting models by preferred assessment model, put forward the optimal use of the various areas of the model. In this paper, the main results are as follows:(1) Gray GM (1,1) modelIn view of the fact that the development coefficient of GM (1,1) model, the amount of gray action, albino differential equations and time response equation had great impact on the model, by changing the sequences of model background value z(1)(k) and boundary conditions of models to do the selection, used the residuals model to do the residual modification of optimization model, indicated that when model input equal interval function(displacement sequence of non-isochronous distance should be transformed to displacement sequence of isochronous distance), monotonic increasing function(decreased under special circumstances)and exponential function was similar to the trend of the time series of displacement, landslide prediction model got the maximum accuracy. Gray GM (1,1) model was suitable for the prediction of landslide deformation of late stage.(2) BP neural network modelBP neural network used three-tier network architecture, its input layer used a linear transfer function, hidden layer used a S-type transfer function (logsig), output layer used a linear transfer function. Structure of the network and the size of training vectors conducted network training by different training function., until got a network structure that met the slope deformation trend, and then used the trained network structure to slope deformation simulation prediction. The innovation lied in that, in this paper, we used adaptive learning rate adjustment method and increasing momentum method to improve general BP neural network model, and synchronous input landslide displacement and rainfall to improve forecasting model and the accuracy of model predictions. BP neural network model prevailed in the late stage of landslide deformation forecasting.(3) Collaborative forecasting modelIn view of the deformation prediction accuracy of collaborative models was not high and the residual deviation was large, this paper, utilized residuals series to build the models and optimized the original models, further clarified that when input the isochronous acceleration deformation displacement sequences, landslide forecasting model got a higher accuracy. Collaborative forecasting model was mainly applied to forecast the late stage of landslide deformation (deformation transformation stage forecast).(4) Dynamic fractal dimension tracking prediction model of fractalDynamic tracking prediction model based on fractal theory and a hypothesis that the birth process of landslides had the characteristics of chaos, without special mechanical assumptions, so dynamic fractal dimension could be used for all kinds of slope. Basis of fractal dimension calculation was based on the displacement to establish phase space, and the time displacement sequence we input should not be cumulative displacement sequence, it should only be the space that could characterize the whole process of slope deformation and has the characteristics of chaos, based on the fractal dimension D to determine the dynamic of landslide and conduct the forecast. Dynamic fractal dimension tracking prediction model was mainly used for long-term forecasts.(5) Time Series Model of fractalTime Series Model of fractal was to supplement and improve the the dynamic fractal dimension tracking prediction model, used to determine the steady state of landslide(the initial creep stage, uniform creep stage, acceleration creep stage). With the progressive instability of slope, H curve and C (t) curve showed a downward trend, indicated that the overall persistence of the landslide stability was at a reduce state; that hurst exponent H of landslide was greater than 0.5 and time function C (t) was greater than 0 showed that landslide was in a steady state as a whole; if Hi, C (t) value was larger, slope stability was more persistent, then explained slope got a higher stability when Hi=0.5, C(t)=0, the fractal dimension D fluctuated in the vicinity of 1, landslide was in a critical state, slight touches would caused overall landslide. Time series analysis method could reflect the dynamic deformation of slope excellently, the results coincided with the actual situation of landslide, then explained that time series analysis could not only be used for landslide prediction, but also to evaluate the stability of landslide and divide landslide hazard class.(6) The fuzzy deformation prediction model conducted a comprehensive evaluation of different model forecasts by preferred assessment model, comprehensively considered the landslide sliding state, the optimal use stage of the model and the data to select the prediction model. Utilized MATLAB software to integrate five prediction models into a landslide prediction software system, achieved comprehensive landslide forecasting and improved forecast accuracy.(7) Based on 2010 January to July Ding Jiafen Shuangbai County, Yunnan landslide workers point real-time monitoring data, according to the model prediction analysis, BP neural network model, gray GM (1,1) model, cooperative model could reflect the trend of slope deformation. Compared the displacement prediction accuracy of these three models, BP neural network model had the highest predictive accuracy, Gray GM (1,1) model followed by, and collaborative prediction model got a large error, this was mainly due to a fact that different models applied to different landslide development stage. The results of dynamic fractal dimension tracking prediction model and time series prediction model consisted with the present situation of landslide(uniform creep stage)basically, showed that both models could reflect the dynamic slide. Comparing these two models, the fractal dimension D of the dynamic fractal dimension tracking prediction model was significantly increased first and then decreased, the fractal dimension values did not fluctuate within 1, indicated that landslide was at uniform creep stage; time series models quantitatively described the decline index of the overall stability of landslide, showed that it was the supplement and improve model of the dynamic fractal dimension tracking prediction model.