Data Mining Study On Prediction Criterion Of Landslide Based On Decision Tree

Since 2003, the Three Gorges Reservoir began impoundment, which made the geology environment worsen based on the cycle action of water level, and induced a lot of ancient landslide resurrection, such as Baishuihe landslide, etc. Based on the influence of the factors of rainfall and reservoir water, these landslides deformation often increased of cycle volatility. Because this landslide before instability often experiences many accelerating deformation, using traditional engineering prevention ways will exhaust wealth and resources. In order to avoid disaster or ease of the influences of landslide disaster, it is particularly important to forecast, judge its deformation and take effective prevention measures.Prediction model and the prediction criterion is core of the landslide problem. Since the 20th century 60's, after several decades of domestic and foreign experts carefully study, the landslide prediction model and the theory has made considerable progress, made a variety of different parameters of the landslide prediction criterion, and also accumulated the experience of successful prediction in practice. However, due to the randomness, complexity and uncertainty of the evolution of slope deformation, the existing landslide prediction model and criterion still have some obvious shortcomings or deficiencies, did not really reveal the nature of landslide deformation and overall consideration of various kinds of factors.Facing massive landslide data, this paper introducing data mining methods and technology into the study, comprehensive analysis the spatial and temporal evolution of Baishuihe landslide deformation, finds that the landslide is not currently in the accelerating deformation stage. It is difficult to forecast accurately the destruction time of Baishuihe landslide because of the time scale. So this paper uses data mining technology to carry out long-term trends in qualitative and quantitative forecast, digs out comprehensive forecast criterion and makes the concept of comprehensive prediction index. Through analysis and study of these issues, the paper has achieved the following results:(1) The spatial and temporal evolution of Baishuihe landslide deformation is analysed.Through the analysis of landslide cumulative displacement-time curve, found that there is obvious characteristics of the ladder-shaped of the evolution of Baishuihe landslide deformation, and the the deformation in front of the landslide is obvious, while slow at the trailing edge, which proved the landslide is towed to determine. Studying the cracks distribution of the landslide, found that the landslide currently does not formate cracks traps system, the overall slip boundary does not form, and makes a comprehensive judgment of the landslide which is not yet accelerated deformation stage.(2) Middle and long-term forecast criterion of predisposing factors of Baishuihe landslide is established such as rainfall and reservoir.Firstly, using K-Means algorithm to cluster the displacement of months, according to the evolution of landslide deformation, the deformation of each phase of the landslide is divided into three categories, namely the slow deformation stage, progressive deformation stage and rapid deformation stage; And then using C5.0 decision tree algorithm to construct landslide trend forecasting model about predisposing factors, such as rainfall (daily rainfall, monthly rainfall), reservoir water (going up and down, the average water level change rate), and summarize prediction criterion of predisposing factors.(3) Combined Macroscopic deformation and other factors to establish a comprehensive prediction criterion of landslide, and proposed the concept of comprehensive prediction index.Using predisposing factor criterion, macroscopic deformation (cracks, local deformation, etc.), deep displacement change rate by C5.0 decision tree algorithm to build an integrated forecasting model, Kappa coefficient of forecasting results reached 0.9. Analyzing the comprehensive prediction criterion, found that the criterion is the more common superimposed factors to affect the landslide deformation, and too precise and absolute to fully show the factors stacking combinations. So this paper proposed the concept of comprehensive prediction index (â… ). Each factor is divided into four categories by the optimal cut point criteria digged out, and each category is assigned the same weight. Then this paper gets the comprehensive prediction index, through stacking each factor weight, which is the importance degree of each factor calculated at the time of constructing decision tree model. And the value range is 1.0-4.0.(4) Used time series methods to quantitatively predict cumulative displacement of landslide.Analysing Displacement monitoring data in January 2004 to November 2010, displacement data of 83 segments was gained in accordance with the time interval by month. Using the former 60 segments data to construct the model of accumulated displacement data-comprehensive prediction index with ARIMA algorithm, and to predict the latter 23 segments data.The average absolute error (MAE) is 25.014mm, which proved that using the comprehensive prediction index (â… ) to construct the prediction model of displacement can fully reflect the landslide deformation by itself, predisposing factors, changes in the external environment and other factors and the landslide displacement amount of prediction accuracy is higher.