Abstract:

Ore prospecting relies on accurate target prediction. The traditional information-based qualitative metallogenic prognostic method, however, has not been able to perform geodata deep mining. In this paper, Big Data deep mining methodology and leveling technique were applied to 1∶200000 scale stream sediment geochemical survey data collected from the Qilian and Longshou Mountains region, Gansu Province to eliminate systematic errors in the raw data. Through regression analysis an information repair model was established to improve the correlation between survey results and copper ore deposits. Using discriminant analysis algorithm, a series of quantitatively optimized prediction models for copper ore prospecting were developed. These prediction models quantitatively predicted copper prospecting target areas. According to the statistical analysis, the proportion of known copper ore deposits exceeded 22% of the class 1 or class 2 predicted target areas, while the target areas only covered 1.72% of the total studied area. Thus, quantitative prediction of prospecting target area using Big Data has greatly increased prediction accuracy while markedly reduced prediction acreage. By characteristic analysis of model elements, the predictive power of the prediction model series was quantitatively evaluated, which provided a basis for the quantitative evaluation of ore genesis and regional ore control conditions.

Key words: geochemical survey, data mining, series of models, prospecting target areas, quantitative predicting