基于稳健有序独立成分分析(ROICA)的矿产预测

摘要

Independent component analysis (ICA) has the advantage of handling the higher 梠rder statistics which are ignored by principle component analysis (PCA). However, decomposed signals by ICA have the character of indeterminacy for the sequences, signs and scales which are of great importance in mineral (i. e. geochemistry) data. Therefore, a new mineral resources prediction approach based on robust ordinal independent component analysis(ROICA) is proposed. Firstly, the mineral data are processed by robust algorithm. Then, a ICA method is applied to the data and factor analysis is introduced to eliminate the indeterminacy among the decomposed independent components. The proposed new algorithm is applied to mineral resources prediction in Shuangren ore district, Guangdong province. Compared with the traditional mean-variance method, the chemical elements (Au and Ag)rndecomposed by the ROICA accord well with the practical distribution of gold and silver ore bodies while combination of chemical elements (e. g, Cu, Pb and Zn) has no indication significance as they have no association with gold and silver ore bodies. This method can be used for prediction of mineral resources.%独立成分分析(ICA)充分考虑了以往主成份分析(PCA)没考虑到但又非常重要的数据高阶统计特性,但1CA分离后的信号具有顺序、符号、幅度的不确定性,而矿产资源预测又必须弄清分离后信号(地质变量)的地质意义；为此,提出了一种基于稳健有序独立成分分析(ROICA)的矿产预测方法.首先对地质数据进行稳健预处理,然后进行独立成分分析,借鉴因子分析的思想解决独立成分分析算法不确定性的局限,从而可以反映分离变量的有序性.将ROICA方法用于广东省肇庆双壬铜矿区进行矿产资源预测.实验结果表明:该区金银矿体的赋存空间位置与传统均值方差方法给出的异常位置明显不对应,没有指示意义；而经过稳健有序ICA处理后的Au、Ag独立成分异常与实际的Au、Ag矿体吻合度很高；Cu、Pb、Zn独立成分异常与Au、Ag矿关联性不大,反映研究区的Cu、Pb、Zn等其他元素或组合不具有Au、Ag矿的指示意义.ROICA方法可用于矿产资源预测.