一种基于加权非负矩阵分解的矿产预测方法  

On mineral resources prognosis based on weighted nonnegative matrix factorization

作　　者：余先川[1] 任雅丽[1] 初晓凤[1] 徐金东[1] 刘石华 李鸿镇 张洁[1] 

机构地区：[1]北京师范大学信息科学与技术学院,北京100875 [2]广东省地勘局722地质大队,广东汕头440500

出　　处：《地质学刊》2013年第1期71-76,共6页Journal of Geology

基　　金：国家自然科学基金项目"面向矿产预测的分层混合模糊-神经网络敏感性分析"(41072245;40672195)

摘　　要：提出了一种新颖的基于加权非负矩阵分解的矿产预测方法,运用非负矩阵分解的非负性、降维性及稀疏性对多维矿产数据进行处理。通过R型聚类分析,按照变量相似度将变量聚合成群,对相关性高的元素的聚类结果进行加权非负矩阵分解得到基向量,进行回归分析验证基向量用于矿产预测的有效性。最后,以广东省新寮岽铜多金属矿区数据为例,通过基向量预测圈定异常,绘制矿产预测分布图,得到明显的异常区域,取得了好的预测结果。The authors presented a new method on mineral resources prognosis based on weighted nonnegative matrix factorization anti processed the muhidimensional mineral data by using non-negativity, dimension reduction and sparseness of nonnegative matrix factorization. R-cluster analysis was adopted to polymerize the variables in groups according to the similarities of the variables, and base vectors were obtained by weighted nonnegative matrix factorization on clustering result of the high correlation elements. Besides, regression analysis was used to verify the validity of using base vectors on mineral resources prediction. The experimental results on the data ahout Sinliaodong＇s copper polymetallie mine showed that the anomaly by the base vector prediction, and mineral forecast maps were delineated. The obvious abnormal areas were aequired and verified to be a good prediction result.

关 键 词：加权非负矩阵分解 矿产预测 聚类分析 空间数据挖掘 广东 

分 类 号：P392[天文地球—地球物理学]