文章摘要
张士红.基于随机森林的四川省会理地区 “拉拉式”铜矿成矿预测[J].地质与勘探,2020,56(2):239-252
基于随机森林的四川省会理地区 “拉拉式”铜矿成矿预测
Random forest-based mineralization prediction of the Lala-type Cu deposit in the Huili area, Sichuan Province
投稿时间:2019-09-14  修订日期:2019-11-28
DOI:10.12134/j.dzykt.2020.02.001
中文关键词: 拉拉铜矿 随机森林 成矿预测 成分数据 受试者工作特征曲线(ROC)
英文关键词: Lala-type copper deposit, Random Forest, mineral prospect mapping, compositional data, Receiver Operating Characteristic (ROC)
基金项目:国家重点研发计划项目(编号:2017YFC0601500, 2017YFC0601501)资助
作者单位E-mail
张士红 中国地质科学院矿产资源研究所国土资源部成矿作用与资源评价重点实验室北京中国地质大学(北京)北京 kyanxiao@sohu.com 
摘要点击次数: 1926
全文下载次数: 852
中文摘要:
      四川省会理-会东矿集区是我国著名的铜资源基地。近年来,随着找矿勘查工作的深入,又提交了多处大中型铜矿床,表明该地区仍具有较大的找矿潜力。本文基于获取的地质、化探和物探数据,应用随机森林(Random Forest,RF)方法,在研究区开展“拉拉式”铜矿成矿潜力预测,取得了较好的效果,随机森林模型预测的平均袋外误差率为6.25%,受试者工作特征曲线(Receiver Operating Characteristic Curve,ROC)的AUC值为0.938。利用偏依赖图(Partial Dependence Plot,PDP)分析解释了预测变量与已知矿床(点)的响应关系,按平均精度下降法排序预测要素的重要性,对“拉拉式”矿床的找矿预测工作具有重要意义的前4个变量依次为:Cu元素含量,中-晚元古代(超)基性岩体临近度,Ni元素含量和PC2因子得分。从成矿地质条件角度而言,河口群地层无疑是“拉拉式”铜矿的重要找矿预测要素,但在随机森林模型中的重要性排序相对靠后。究其原因,一方面是与其它连续数值型预测要素不同,河口群地层是二值(0-1)变量;另外,河口群地层的分布范围受覆盖层的影响较大。根据随机森林模型生成的拉拉地区成矿有利度信息,圈定了6处找矿远景区。红铜山-落凼-红泥坡-姜驿高成矿有利度区带呈北北东向展布,主体上与研究区内重要的地质-地球化学预测要素异常空间分布一致;其中蒿枝坝-落凼-红泥坡-姑鲁迷找矿远景区是本区已探明铜矿的主要分布区,其内还有进一步勘探的潜力;同时,该异常区呈半环形态,结合地质勘探揭示的变质火山岩厚度分布,及其西部受一组后期NNE向走滑断裂限制的特点,显示出古火山活动中心在西部,并可能存在被切割分离的另一半环异常,这为该地区后续地质研究和铜矿勘查指明了方向。
英文摘要:
      The Huili-Huidong district of Sichuan Province, China is well known for its rich copper resources. In recent years, many large and medium-sized copper deposits have been explored, indicating that the district still has a good prospecting potential. Based on geologic, geophysical and geochemical data available, using the random forest (RF) method, this work made prediction of mineralization of the Lala-type copper deposits in this area. Sixteen copper deposits (occurrences) and 9 digital maps were used as independent variables associated with the Cu deposits in RF modeling. Global accuracy of Out of Bag, OOB, derived from the RF modeling is 93.75%. The ACU value under the ROC (Receiver Operating Characteristic) curve is 0.938. The RF modeling was used to rank the importance of predictive variables. The response relationship between predictive variables and known Cu deposits is illustrated and explained by a partial dependence plot (PDP). The logarithms of Cu, Ni content in stream sediments, distance to mafic intrusion, second principal component scores, PC2, of geochemical data based on isometric log-ratio transformed were the top 4 predictive variables selected according to the importance inferred with respect to the presence of Lala-type Cu deposits. Mineral potential maps were prepared by the RF model and were reclassified into three likelihood levels or zones according to a "accumulative frequency of deposits " function criterion, and 6 mineral prospects were delineated. The main NNE trending anomalous zone with high likelihood of discovery values in the predictive map appears consistent with the spatial distribution of important predictive variables, and the minor approximate EW-trending high anomalous level zones that are consistent with the occurrence of ore-forming geological structures. The superimposed zones of the NE-trending and EW-trending anomalies happen to be the main locations of known copper deposit sites. Their extensions to the east and west are covered by strata older than the Hekou Group and are considered as favorable copper prospects for further exploration. The results have also indicated that the combination of RF modeling is an effective tool for mineral potential mapping when only data of a few known mineral occurrences are available.
查看全文   查看/发表评论  下载PDF阅读器
关闭