基于机器学习的SIP数据砂岩型铀矿氧化-还原带识别研究

Machine learning of identifying the redox transitional zone by SIP data in sandstone-type uranium deposits

砂岩型铀矿成矿过程中,氧化-还原作用是实现铀富集成矿的重要作用之一,有效圈定深部氧化-还原过渡带范围能够对指示深部找矿提供重要线索。频谱激电法(SIP)利用氧化-还原带岩石电化学特性差异,基于柯尔-柯尔(Cole-Cole)模型可反演获得地下介质多个复电阻率参数。采用该参数特征组合开展改进的Kmeans聚类,通过无监督机器学习算法实现数据结果自动分类,从而达到划分砂岩型铀矿氧化-还原分带的目的。经验证地层分带识别结果与钻探剖面验证情况一致性较好,表明Kmeans++法能够有效识别SIP结果数据中蕴含的砂岩型铀矿氧化-还原分带信息,为铀矿勘探缩小找矿范围,指示下一步找矿方向提供参考。

In the metallogenic process of sandstone-type uranium deposits,redox reaction is one of the significant factors to enrich uranium.Accurately delimiting the range of the redox transition zone can provide crucial clues for deep ore prospecting.Based on the difference of electrochemical properties between rocks in oxidation-reduction zones,Spectral Induced Polarization(SIP)method can invert and obtain multiple complex resistivity parameters of underground lithologies with Cole-Cole model.By applying the combination of these parameters to an enhanced Kmeans clustering,SIP data can automatically cluster the litholgy of different complex resistivity parameters through unsupervised machine learning algorithms,and reach the goal of distinguishing the oxidation-reduction layers in sandstone-type uranium deposits.With the dilling verification,the SIP data identified stratigraphic zones were of good consistency with the geological outcome,indicating that the Kmeans++method can effectively identify the redox zoning information from SIP result data.The study provided reference for cutting down the prospecting area and orientating the next prospecting direction in uranium exploration.