Geostatistics And Neural Network-based Digital Metallogenic Prediction

The digitized mineralization forecast establishes the ore body spatial distribution shape, the structure as well as the ore deposit geology attribute (for example, grade, lithology etc.) the digitized mineralization model by the aid of to technical method and the theory and so on the computer, geological statistics, computation geometry, computer graphics, then carries on the mineralization quota forecast. The digitized ore deposit model establishment regarding provides the science the mineralization forecast rationale, the realization resources/reserves dynamic management has the vital significance.The BP neural network technology is the commonly used non-linearity fitting method currently, it can approach any assigned non-linear relations by the arbitrary precision, from inputs through various neurons between coupling function relations realization to the output highly the non-linear mapping. This paper is based on the BP neural network and the geological statistics theory, takes the western Hunan gold ore as the example, and uses the personal status modeling, the element correlation analysis, the mineralization tendency rule, the depth portion reserves forecast and other ore deposit modeling technical problems to do the development research.The paper first elaborates geological statistics fundamental concepts like region variable, variation function and its theoretical model fitting, structure analysis. It then studies the western Hunan gold mine (bed) the body variation function characteristic, and uses spherical, linear, and logarithm to do the fitting and the overlapping confirmation of the math modeling. V1,V2,V3,V4,V7 and V8 these 6 mineral lodes in the trend with the tendency in variation function are calculated, the regulation found in the research is:Upside variation function variable space bigger, the conversion coefficient are less, the depth portion variable space changes slightly, which means that the ore deposit formed matter originates from the depth portion mineralization mechanism. It also does the pattern recognition of the V7,V8 mineral lode variation function and V1,V2,V3,the V4 mineral lode variation function discovered recently. The recognition result indicates that V7 mineral lode variation function and V3 mineral lode are extremely similar. It's possible to have the same characteristic as V3 mineral lode, i. e. mineralization is stable, a scale bigger. However the V8 mineral lode variation function is different from other mineral lodes, it possibly belongs to kind of new mineralization characteristic type. Finally, the forecast using the BP neural network model to the depth portion mineral lode variation function characteristic is shown, based on which the author has used the variation function characteristic to do the analysis and the forecast of the mineral lode depth portion concentrated the rule.The paper has introduced the minute shape geology theory, through analyzing the western Hunan gold ore massive geology sampling data, indicated that the thickness of westem Hunan gold ore mineral lode, and that Au, Sb, the WO₃ these useful ingredient content all has many multiples fractal shapes the characteristic, also the mineral lode in the trend with the tendency, upside and the depth portion all has the aeolotropy. With the ore deposit mineralization geology condition analysis, the author has proved that the westem Hunan gold ore deposit the formation has experienced the multi-period multi-stage mineralization superimposition.The author has used the BP neural network technology to compare V7, the V8 arteries with V1,V2,V3,the V4 arteries in the minute shape characteristic, the result indicates that V7 and the V3 arteries existence similar minute shape characteristic. Therefore, these two mineral lodes' saved characteristic, scale, hardness index in the tax are probably similar. Then used the BP neural network was usedto carry on the analysis to the mineral lode deep partial shapes characteristic, thus has carried on the forecast to the mineral lode depth portion shape characteristic. Divided the shape geology theory and the BP neural network technology has alternately confirmed the previous depth portion mineralization matter origin mechanism conclusion mentioned in the variation function before.Many kinds of methods have been used to analyze the grade and the metal quantity overall relevance of western Hunan gold ore Au, Sb, WO₃, the result indicates that during 3 kinds of useful components the grade are partial relevant, but metal quantity are remarkable relevant. This rule promulgates that if several continual center-sections metal quantities the relevance is worse, indicating that this sector possibly does not exist the discovered ore body, which has the certain instruction significance to the prospecting.The paper has also studied the BP neural network technology with emphasis in the western Hunan gold ore mineralization forecast model establishment. The highly non-linearity have been studied characteristic of BP neural network to propose one kind of multiple network architecture mineralization forecast model. This model first takes the many kinds of samples data as input, calculates the results of distinction non-linearly the variation function, divided the shape theory, and the trend analysis. Then it outputs factor that obtains these independents.