Study On Prediction Criteria Of Accumulated Layer Landslide In Three Gorges Reservoir Using Dynamic Data Mining

With the growing of the world’s population, the gradual expansion of human activities, the engineering activities base on the technical and economic conditions continue to increase the extent of the disturbances of the geological environment. Landslide hazard, especially the frequency of large-scale landslide disasters is getting higher and higher, resulted in economic losses and casualties increase. The development of the accumulated layer landslide in reservoir is widely, and the water level and rainfall will likely induce the occurrence of landslide disasters. The landslide forecast criterion will reduce losses.With the emergence of the reservoir data of multi-source monitoring and disaster survey, mining useful criterion knowledge from the information is the key problem in landslide forecast study. Dynamic data mining is a new data processing tool, its purpose is extracting corresponding laws from the dynamically changing data, and the introduction of this method for landslide criterion provides a fast, real-time, accurate reference model.In order to get the valid landslide forecast criterion, this paper use dynamic data mining methods to fully get the relevant monitoring indicators from the dynamic changes in the data stream, forming landslide characteristic factor. The main contents and conclusions are as follows:(1) Data collection:This study analyses typical landslides’evolution of the spatial and time, getting the characteristics of landslides’size and distribution in reservoir area. Finish monitoring the displacement of landslide, reservoir water level and rainfall, the macro-geological deformation characteristics, landslide maps and a series of values, including images, data, text. For example, Shuping landslide, this paper introducing monitoring data table, including field properties and binding rules, and eventual establish landslide monitoring information database. (2) The partition of the landslide evolution:Through access to domestic and foreign landslide evolution characteristics described, with special geological background of the study area, according to the accumulated displacement-time curve and relative monthly-displacement characteristics, using k-means algorithm, to divide evolution stage, forming the initial8divided results, and merged high-related categories, then get the formation about the optimal solution of the four categories.(3) Data binning process:Operating continuous monitoring indicators binning process by MDLP algorithm method, using deformation stage as competent field, converting continuous data to discrete data type, so it is convenient for subsequent association rule mining as the former term.(4) Common deformation predisposing factor extraction:Base on large information, we found that the role of water-influence in the reservoir area is more significant to the deformation of landslide, this paper proposes the idea of rainfall and reservoir-water level as common factors, using rainfall, reservoir-water fluctuations as a predisposing factor to the landslide deformation stage for the initial rules. The high precision of the resulting rules to judge the sample data prove that the public predisposing factors as criterion model has some applicability.(5) Feature factor dynamic mining:If use the association rules only by common predisposing factors, there still exists errors that regarding unstable stages as the security phase, so relying solely on the common factors can’t fully reflect all landslide characteristics in study area. So this paper use insert data of landslide for study, with reference to the dynamic data mining process idea, use principles of triggers to forming dynamic analysis between new data and landslide evolution, to extract the potential characteristic factors. Merging higher correlation characteristic factors and common factors into constitute an integrated criterion.(6) Application of criterion results:With the use of comprehensive criteria by dynamic data mining, we can get repeatedly, fast and early warnings from the future monitoring data. Building trigger main function code on the criterion results, so there is a new way to get real-time deduction about the landslide evolution, providing a priori knowledge for accumulated layer landslide forecasting.