Research On The Optimization Method Of Support Vector Machine With Rough Set And Its Application In Resource Evaluation

Nowadays, blind mineral resources has gradually become the focus of mineral explora tion, for its deeply buried, the only way to find the target is to analysis a variety of geophysical, geochemical and remote sensing data indirectly. About these multidimensional data,the rough set theory is the better method that have better selection of culling the redundant attribute which can determined the attributes and blind mineral resources.Through the establish of the classification model, the property of attributes of blind mineral resources should be determined,and then the new targert should be prodicted. And so these field have been the hot topic.Support vector machine models of statistical learning theory have strong theoretical basis and the ability of classification based on VC dimension and minimize structural risk,it can solve the problem of small sample, nonlinear, over learning, dimension disaster and part minimum and so on in good condition; With the pretreatment function of rough set reducing the dimensionality of the multidimensional data, it can achieve classification performance and generalization ability. The prediction of prospecting target area get more accurate. In this study the research as follows:1)In this study, geophysical and chemical data was used by inverse distance interpolation and the raster data was generated based on ArcGIS in the Gejiu tin copper deposit test area in Yunnan province. And then 500 random samples in test area and 100 random sample points in tin copper mine area were extracted, and thus the attribute value of geophysical and geochemical exploration were given to these random sample points. Next the buffer area was set in some certain range of the tin mineral area, and different decision attribute values were added to the sample points, then a complete decision attribute system was built.2)Based on the continuous condition of attribute values, a appropriate neighborhood radius was set in MATLAB programming by using of the neighborhood rough set, the training samples were handled in norm and the 41 condition attributes were reduced by building the optimize methods of the fuzzy factor method. After that the weight was gave to each attribute reduction based on the importance way, and weight was analyzed by trying to pick up the order of attributes and to add attributes. In the preparation of the KNN algorithm, the appropriate parameters were designed, the noise data were eliminatecomprehensively by combining with the probability histogram, the semi variance covariance cloud analysis of ArcGIS statistics analysis. and the boundary samples were chose as the final training sample of SVM model, preprocessing neighborhood rough set were to be done.3)In MATLAB, the Gauss kernel function was selected to construct the SVM Model for the training samples, and through ten means to optimize the parameters of the model test of the cross validation method, then the optimal classification model was obtained.Finally the whole Gejiu tin mine area was did the tin mineral resources evaluation, and different attribute reduction and the prediction and evaluation system was comprised.