基于极值统计学的青海拉陵灶火地区矿产资源评价研究

The singular data in multi - source geoscience - observation data can provide more information than the normal data, which is closely related to the geological events such as mineralization and metallogeny. The research on the change law and statistical modeling method of the multi - source geoscience observation data from the normal value to the singular value can provide the technical support for the automatic extraction of the geological anomaly and the statistical evaluation of the metallogenic potential.In view of this, based on the latest achievements of extreme value statistics theory and applied research, the project team aim to develop a new method of extreme value statistics for mineral resources evaluation, and use the application of extreme value statistics in the fields of Finance and earthquake prediction for reference, chose Lalingzaohuo region in Qinghai province as the study area, study establishment of the statistical model and the correlation of singularity between the different observation indexes. The model is about the change law of multi - source geoscience observation data from the normal value to the singular value. On the basis of, a new method for automatic extraction of ore forming geological anomalies and a new model for ore forming potential evaluation is deprived, delineation mineral target area and predict the experimentation area, therefore, A typical example of creating a new method for mineral resources evaluation based on extreme value statistics is established. The research can provide the new technology and the new method for the present mineral resources evaluation, and has the positive promoter action to the development of extreme value statistical theory, it has a great significance both in the theory and the practical.

多源地学观测数据中的奇异数据能够比正常数据提供更多与奇异地质事件（如矿化和成矿作用等）密切相关的信息。研究多源地学观测数据从正常值到奇异值的变化规律和统计建模方法，能够为自动提取致矿地质异常并对其成矿潜力进行统计评价提供技术支撑。鉴于此，项目组拟以极值统计学理论与应用研究的最新成果为基础，以研制极值统计学矿产资源评价新方法为目标，借鉴极值统计学在金融和地震预报等领域的研究思路，选定青海拉陵灶火地区为实验区，研究建立地物化遥多源地学观测数据从正常值到奇异值变化规律的统计模型及不同观测指标间奇异化的相关性，在此基础上，研制致矿地质异常自动提取的新方法和成矿潜力评价的新模型，圈定和预测实验区的矿产靶区，创建一个极值统计学矿产资源评价新方法研究的典型范例。研究工作能够为当代矿产资源评价提供新技术与新方法，对极值统计学理论的发展具有积极的促进作用，理论与实际意义重大。

项目研究将极值理论建模思想引入当代矿产资源评价研究领域，研究致矿地质异常自动提取技术和致矿地质异常成矿潜力统计评价方法，开发了极值统计学成矿潜力预测算法，开展了成矿潜力预测效果评价研究。项目研究以青海拉陵灶火地区为研究区，将POT阈值模型与GPD分布函数相结合，构建地球化学极值数据统计分布估计模型，提取了研究区多种地球化学元素的异常信息；选用Gumbel-Copula结构形式并结合Logistic函数，组建多元地质异常信息的超阈值模型，描述了多源地质异常极值信息间的相关程度；利用Copula-GARCH方法构建了研究区空间异常组合模式的成矿潜力统计模型，并分别利用ROC曲线和P-A曲线对模型成矿潜力预测效果开展了统计评价研究。项目研究结果表明：POT-GPD模型在一维致矿异常信息提取方面具有一定的有效性和可靠性；Gumbel-Copula结构形式能够有效刻画异常变量间的极值结构；利用Copula-GARCH模型预测的研究区成矿潜力效果优势明显，基于ROC曲线的评价指标AUC值和基于P-A曲线的评价指标P/A值分别为0.9163和7.058，均优于传统数据驱动型模型预测效果。项目研究成果推动了极值统计学在矿产资源评价领域的应用探索，对矿产资源评价方法体系的发展具有积极的影响。