Application Of Support Vector Machine To Predict Pollymetallic Deposits In Geophysical And Geochemical Anomalies

A description of the ore body attributes (taste, shape, etc.) in mineral resourcesprospecting and evaluation is essential,it directly reflects the previously invested, even it isthe direct embodiment of a Prospecting. At present, people mainly depend on the drillingworks to understand the wealth of deposits, but the cost is expensive and time-consuming。With the continuous development of geological theory, people have been trying to carry outcomprehensive information prospecting theory, a variety of mineral prediction model whichbuilt on top of the statistical theory not only requires vast amounts of training samples, butthe prediction accuracy is not enough. To address the above issues, the subject combinedthe geological prospecting methods and machine learning new technologies to put the newstatistical learning theory (support vector machine) to apply the mineral into the forecastmodel, its unique and solid theoretical foundation (promotionthe nature of the sector, thestructural risk minimization, kernel function, the optimal regression hyperplane)，cansolve effectively previous prediction models which require large samples, overfitting, localminima, the problem of high dimension. Currently, Support vector machines in the weather,the aquifer water forecast, traffic accidents and other fields has achieved a successfulapplication, and has been recognized by the majority of scholars.This research project relies on the Inner Mongolia Autonomous Region Geology andMineral Exploration Fund: Baogedewula region of silver metal mineral ready exploration(test), New Barag flag, Inner Mongolia Autonomous Region. We let mine as a predictor ofobject and carried out multi-metal mineral quantitative prediction which based ongeophysical and geochemical anomalies of support vector machines, set up the predictionmodel. First collect examples of the application of a large number of domestic andinternational metallogenic prediction research results and support vector machine applied inthe relevant fields, analysis the premise of metallogenic prediction of geology-geophysics-geochemistry, elaborated on the content and framework of statistical learning theory andsupport vector machine.Second, completed correlation analysis of geochemistry parameters(amount of surface metal) or orebody grade and geophysical parameters (resistivity,polarizability, magnetic, etc.), and selected with strong correlation metal elements such asW parameters as a predictive model input feature vectors, establish a predictive model forMetal mineral that the output is the content of W at different depths. Through in-depthstudy experience and skills which related to the application of support vector machine, identified a number of technical methods for the prediction model required. Selected RBFas kernel function of the prediction model through cyclic authentication； throughcross-validation and grid search to obtain the best model parameters and establish a set ofmetallogenic prediction model which based on geophysical and geochemical anomalies,using the Visual C++6.0language programming, and completed the preparation of theforecast system. Finally, the model is applied to District of BaoGedewula molybdenum andtungsten mine for instance validation, and achieved good results；Full study of thegeological conditions in the area, summed up Geology-Geophysics-Geochemicalprospecting model.The main results of this research project:①The first time, support vector machineused for mineralization forecast which based on geophysical and geochemical anomalies ofground，achieved without the prophet conditions quantitative prediction of the undergroundore body②On the basis of summaring Shanagen mineralization law，set up the area’sgeological-geophysical-geochemical prospecting model.