The Prediction Model Of Maximum Subsidence In Mining Area Of Loess Based On Support Vector Machines

The ground subsidence deformation is caused by coal mining.The maximum subsidence value is the key indicator that measures the ground subsidence deformation.In the mining subsidence,mining depth,mining height,seam dip,rock hardness,length and width of working face and other factors have the important influence on the maximum subsidence value.The ground maximum subsidence value of the loess mining area needs to take into account the influence of the soil layer thickness and properties and so on.These factors have complex nonlinear relationship with the maximum subsidence value,the existing maximum subsidence prediction model involvesseveral parameters and can't reveal the complex nonlinear features.Therefore,the prediction accuracy and range of application are limited largely.Support Vector Machine can solve the problem by Kernel Function,it maps the sample space dimensions to a feature space dimension that is high enough and converts a complex nonlinear problem to a linear problem.Support Vector Machine has punishment mechanism that removes gross data automatically,and insures the accuracy and reliability of the model,it provides the effective method to establish the maximum subsidence prediction model of the loess mining area.This article relies on principles and features of the Support Vector Machine and then generalizes some steps of the maximum subsidence prediction model's structure.Then using the iterative method gets the best training sample according to the steps.And the model is trainedby the best one,whichacquires the number of support vectors and coefficients,and ultimately composes the Support Vector Machine model.According to the process of modeling,andbased on a scripting language of the MATLAB platform developes a visualization program.Thevisualization program packages the process of the model training and application,and provides a simple and pellucid interface form for users to operate.Support vector machine regression model is compared with existing models.we find the accuracy and reliability of itare better than other function models.Take the different values of the input variables separatelyto further reveal variation law between model variables and study their relationship with the maximum subsidence value.The results indicate that maximum subsidence has highsensitivity on rock hardness,mining height and seam dip than other variables.In addition,samples that has high prediction accuracy and error studied found width-deep ratio and soilthickness ratio of the sample predicted deviation is less than 0.3 and 0.35,instead,width-deep ratio and soil thickness ratio of the sample is at least one more than 0.4-0.5.Tn this paper,Support Vector Machine Regression Model builded can be used for quantitative estimates of maximum ground subsidence in mining area of loess,it has certain promotion application value.