Study Of Prediction Models And Their Applications Of The Coal Thickness Change Based On Kernel Method

The thickness of coal seam is essential for the design and exploitation of coal mine. Accurate prediction of coal seam thickness provides a powerful geological guarantee for the mine production. For seismic exploration most coal seam belongs to the thin layer, while the thickness prediction of thin layers has long been recognized as one of the challenges. The traditional forecasting method is based on the comparison of interpolation results using drilling data, but with poor precision, due to the high cost and lower density of the drilling. Therefore its applied range has a very big limitation.3D seismic technology has the advantage of collecting massive information from a large area, so we can use abundant seismic information to detect the changes of coal seam thickness. At present, 3D seismic exploration has become one of the main methods to solve the geological problems in coal mining because of its advanced technology and high resolution.Taking the first mining area of Shilawusu mine as the study area and combining the geological with the 3D seismic exploration data, this paper presents a prediction method of coal seam thickness based on seismic attribute data and kernel methods. The simulation results have been verified in the case study. The paper consists mainly of the researches in following 5 aspects.(1)The 3D seismic attribute analysis technique is studied in three aspects: its extraction, classification, and optimization. The definition of seismic attribute and its geological implication are also studied in depth. Especially the grey correlation analysis is studied for optimizing seismic attributes. Before forecasting model training, several properties of the attributes closely related to the thickness of coal seam are extracted through seismic attribute optimization, and used as the characteristic value of training model. Such a process can not only improve the prediction accuracy, but also reduces the temporal and spatial complexity of the algorithm.(2)The least square support vector machine(LS-SVM) and the simple multiple kernel learning theory(SimpleMKL) are studied, the coupled simulated annealing(CSA) algorithm are used for the regularization parameter r and kernel parameter s of least square support vector machine(SVM), a k-cross validation method is proposed to optimize the regularization parameter C of simpleMKL.(3)Taking chaotic time series as a simulation example, which is created by a nonlinear function and Lorenz system, to simulate and test the predictive capability of LS-SVM and the SimpleMKL. By the simulation, regularization parameter r and kernel parameter s of least square support vector machine(LS-SVM) are optimized using the coupled simulated annealing algorithm(CSA) algorithm, and the regularization parameter C of simple MKLis optimized using k-cross validation method. And a fuzzy time series prediction method is proposed, based on phase space reconstruction and LS-SVM/SimpleMKL.To measure the accuracy of forecast model, the absolute error, relative error, square error and correlation coefficient are selected as the prediction effect index of the overall evaluation model, and the applicability of the prediction model is analyzed based on interpolation prediction, extrapolation prediction and anti-noise ability of the nonlinear function.(4) A prediction method of coal seam thickness is proposed in this paper based on LS-SVM model and Kriging interpretation. In the method, the variogram function model reconstructed using the least square support vector machine, so adaptive fitting of the variation function can be realised according to the different data characteristics, and the prediction accuracy of the thickness of the coal seam is improved. Finally, the coal seam thickness in the study area are predicted by using a spherical function, exponential function, gaussian function and least square support vector machine(LS-SVM) as variational function model respectively, and the prediction accuracy of each model mentioned above is compared through the cross validation method.(5) This paper presents a prediction method of coal seam thickness based on grey relational analysis and LS-SVM/Simple MKL. Based on the 3D seismic and borehole data in the study area, seismic attributes and coal thickness are extracted firstly, Grey Relational Analysis is used to optimize the seismic attributes and reduce the dimension of the input samples, which reduces the time and space complexity, and to optimize the regularization parameter r and kernel parameter s of least square support vector machine(LS-SVM) using the coupled simulated annealing algorithm(CSA) algorithm, and k-cross validation method is proposed to optimize the regularization parameter C of SimpleMKL. The optimized attributes and the corresponding thickness of coal are used as the input and output features of LS-SVM/SimpleMKL and prediction model of coal seam thickness is obtained by training. The accuracy analysis is made by using absolute error and relative error, and obtained good predictive results. Finally the prediction model is used for coal seam thickness prediction in the whole study area, and 3D visualization display of the thickness of coal seam based on VTK is realized, and show good results.