Study Of Landslide Forecast Model Based On The Multiple Structured Data Mining

With the large-scale social and economic development, the scope of human activities gradually expanded, as well as the frequency and intensity of landslide events showed an increasing trend, resulting in the gradual increasing of casualties and economic losses. The geological conditions of Three Gorges Reservoir are complex. With the further expansion of the implementation of various projects, which lead to more frequent human activity, it has a huge impact on the surrounding geological environment in the reservoir area, as a consequence, it directly or indirectly led to the recurrence of existed landslides and the occurrence of new landslides, but the current domestic economic conditions and human resource cannot offer enough resource to treat all dangerous landslides. Therefore, the spatial and temporal study of landslide prediction model based on data mining technology have a more important practical significance, which contribute to the strategy of disaster prevention and the safe operation of the Gorges Reservoir.This paper summarized the research status on aspects such as the landslide prediction model characteristics of time and space, the multi-structured data mining and the distributed data system platform construction at home and abroad and the problems within them. On the basis of previous studies, with the Zigui-Badong district in the Three Gorges Reservoir Area where landslide disasters and human engineering activities are frequent as the research area, it establishes the distributed storage and sharing mechanism in the multi-structured database of landslide under the NoSQL framework. With temporal and Spatial Analysis of landslide prediction model and stability criterion extracted as a starting point, it takes advantage of text data, monitoring data and spatial data to establish a multi-factor evaluation structure respectively and define the corresponding formula and finally establish the time prediction model of landslide stability and the evaluation model of regional landslide disaster body based on multi structure data mining. With the data storage system based on NoSQL theory, a data mining platform is designed in the client, which could realize the output and visualization of landslide displacement predicted value. Taking into account the exponential growth trend of the landslide monitoring data in the future at the same time, it takes a parallel conversion to the association rule algorithms of landslide prediction model under the MapReduce programming framework, and improve the efficiency of processing the large-scale landslide data. Concrete results and conclusions are as follows:(1) Four data types in the study area were analyzed and finished to formulate a unified multi-structured data storage standards of landslide. By analyzing the data characteristics between the professional monitoring data of landslides and others in research area, it discussed the current problems of Landslide Database:the traditional relational database table structure is not uniform, and ineffective to manage multi-source heterogeneous data of landslide. All landslide historical data were divided into the initial monitoring data, spatial data, text data and image data to establish a multi-structured data storage standards. The data were converted to BSON format and stored in MongoDB which is a document database system under NoSQL storage structure. Finally, the monitoring data sets, text data sets and image data sets were obtained. The efficient key-value form provides a structural specification for a variety of landslide data types with different sources and loose relationships.(2) Combined with multi structure evaluation factor, it establishes the time and space prediction model of landslide and multidimensional criterion extraction processes to achieve single landslide stability evaluation and analysis of landslide-prone area. Based on multi-structured data mining theory, a comprehensive evaluation factor coefficient calculation formula is given by the analyzing of landslide time prediction model from the Baijiabao landslide monitoring data and text data. The results showed that rainfall is the leading cause of landslide deformation. The response relationship between the surface deformation displacement of the Baijiabao landslide and the influence factors is analyzed in detail, and the result shows that the landslide is influenced by the seasonal hydraulic power and determines the monthly cumulative rainfall as the evaluation index which guides the value of the secondary exponential smoothing model parameters. The results show that the optimization model is more accurate than the original model for the prediction of the cumulative displacement of the landslide, and it is better to predict the short-term deformation trend of the landslide. At the same time, under the consideration of the inducing factors, an optimized Arima model is establishes. The results show that the model of landslide displacement relative to the fitting and prediction ability well, and the average relative error higher than the original ARIMA model by 6.28%. Because the water cycle system of the research area has a great influence on slope stability of the weak, the occurring of landslides near to the river in large area will cause channel congestion and the loss of human lives and properties. Therefore, stability evaluation of landslide in the Three Gorges Reservoir area will appear particularly important. Among the mining process of vectors and discrete properties such as the landslide front elevation, the distances to river, the area of numerical indexes and the distribution of lithology, it reveals that the danger may occur with a gradient in the range of 15° ～45° and a distance to the banks in the 0.1 ～117.90m, which provides a priori rules of hazard discrimination in this type of newborn landslide. For example, the Baojiabao landslide belongs to higher risk cases. Based on the existing classification knowledge of landslide evolution stage, the association rules model is constructed with rainfall, reservoir water level, and ground water monitoring indicators. The results show that Baijiabao landslide is more easily effected by the sustained rainfall and reservoir water level fluctuation, it will also cause the rapid decline of groundwater that lead to the landslide accelerated quickly into the stage of deformation, as a result the accuracy of the criterion for landslide stability prediction reached 91.07%. On the side, starting from the object-oriented multiscale segmentation and the expert classification, the C5.0 decision tree model based on regional landslide susceptibility theory is constructed with remote sensing images, reservoir water, slope, slope structure, and engineering rock group data to achieve the susceptibility prediction of four units types in the research area. After multiscale segmentation, the study area is totally divided into 2279 object, it shows that the average correct rate of training samples and test samples is 91.64%, and the Kappa coefficients are 0.84 and 0.51 respectively. The model predicted results mainly appear in no susceptibility area and high susceptibility area, and the low susceptibility area and middle susceptibility area add up to only 141 in the forecast of space frequency, that accounts for 6.19% of the total number of objects. The experiments show that C5.0 decision tree algorithm has better classification, and can divide the regional spatial stability clearly. The landslide-prone area classification forecast map of the research area is established by the decision tree model, and it shows that the high susceptibility cell is more prone to occur on both sides of the Yangtze River and its tributaries, the engineering rock groups usually develop into soft intermediated-acidic and soft hard alternate with rock group. Summarizing the slope body structure and the development law of slope, the results show that the high susceptibility cell is prone to occur in the area of 15 degrees to 30 degrees with a consequent slope or oblique slope. This is consistent with historical vector data, proved that the model predicted results is reliable.(3)A multi-structure landslide data mining platform based on MongoDB is established to achieve the function of multi-structure landslide data mining. Based on the MongoDB database and Java language framework, it realizes the multi-structured data distributed storage and query, and the function of the quadratic exponential model optimized programmatically, and deploys the data platform in the server and client respectively. With Shuping landslide as the research area, the evaluation indexes are searched from the related documents and monitoring data in the data analysis phase to achieve the size of multi-structure evaluation factor coefficients, it reveals that the reservoir water level indicator coefficient reached a maximum of 0.65. The subsequent experiments also prove that the rapid decline of reservoir water level has played a relatively significant role in the rule of causing Shuping landslide instability, especially in the stage of destruction and shear deformation expansion, the rapid decline of the reservoir water level is the most important factor to induce landslide instability. Under the premise of the high correlation between reservoir water level fluctuation and the stage of the evolution of the Shuping landslide, the index optimization model is imported by programming, and the model parameter value is corrected by the dynamic correction of reservoir water level. Finally, the output and visualization of the model predictive value and the measured value of the cumulative displacement of the landslide are realized by using the Java form as well as the curve drawing component, and the average relative error of the model is 5.5%.(4)The Apriori parallelization algorithm based on cloud computing environment is designed to realize the frequent rapid extraction from the massive monitoring data. By analyzing the Aporiori algorithm process in the landslide prediction model, and combined with the MapReduce theory of parallel programming framework, it achieves the Map and Reduce designed for frequent extraction. Using Hadoop 1.2.1 stable version, it builds the server clusters that contains eight nodes. The two above parallel algorithm deployed in the cluster cloud computing platforms are compared the difference in efficiency of treating the landslide monitoring data sets between the stand-alone and cluster system. The results show that the speed-up ratio will improve while the size of data grows. For example, when the amount of data reaches 60822, the speed-up ratio is 1.56. The parallel algorithm based on graphs can solve the time bottleneck problem of mass landslide data mining in the stand-alone systems, because the task can be assigned to the work cycle of each processor. It can save the resources of the whole cost, and enhance the efficiency.