Study On Landslide Displacement Prediction With Markov Chain Model

The landslides occur frequently in China, especiallyintheThreeGorgesReservior, which is the landslide-prone area.With the rise and fall of the reservoir water level each year after impoundment, unstable slopes on both sides of the river and its tributariesface new uncertainties. In order to respond to this change, relevant departments have decorated a large number of landslide monitoring point in the ThreeGorgesReservior and accumulated a mass of monitoring data after nearly a decade of monitoring.It isa hot research aspect in the field of landslide hazard that how to excavatepotential and useful information from these datascientifically, rationally and effectively, and thentake the decision-making support for the monitoring and predicting of landslide hazard.Markov chain model (referred to as the MC model) is a discrete stochastic process with the"stability ineffectiveness", is widely used in commercial, economic, environmental, military and other fields. In the fields of time series prediction, the Markov chain model occupies an important position, with a high scientific, adaptability and accuracy. Because of the Markov property in landslide displacement process, Markov chain model have a great application potential in landslide displacement prediction. Surroundingfitting, optimized fitting and time forecasting, the thesis studies the applications of Markov chain modelin the prediction of landslide displacement, with taking contrast analysis on multiple landslide displacement as the cut-in point, the association rules mining for the relationship between landslides displacement and impact factors as the analysis basis, and the Shuping landslide as the typical experimentsubject. Firstly, pretreatment methods for landslide monitoring data are studied, includingselection and building on variables, segmentation and grouping on data, judgment and treatment on singular value. Secondly, average displacement rates of100GPS monitoring points of13experiment landslides from May to September each year are obtainedrespectively. Thirdly, association rules between the displacement of these landslides and inducing factors are mined by using Apriori algorithm. Fourthly, ARIMA model is used to fit the accumulated displacement time series of Shuping landslide and MC model is used to optimize the fitting. At last, ARIMA-MC model is used to forecast the displacement time series of Shuping landslide.Through analysis and study of these issues, the paper has achieved the following results(1) Landslide deformation lawLandslideswhich are relatively stable have the characteristic of deformationwith the randomness of the statistical significance, while potential unstable or unstable landslidesmovealong the main slip direction.The deformationson different parts of the same landslide shows the same increase or decrease trend.(2)ARIMA-MC modelThe MAE (mean absolute error) of the optimization fitting with MC model improved14.037than the ARIMA model.Then the ARIMA-MC model is established for time series fitting and forecasting on landslide based on MC and ARIMA model, namelyARIMA-MC model. By comparing the measured value with the estimatedvalue, it is showed that the ARIMA model is only suitable for short-term forecast (4periods) and the prediction accuracy is improved generally with an order of magnitude. The prediction result would becomebad dramatically for medium-term and long-term forecasts (longer than4periods).