自组织人工神经网络与聚类法在矿区沉积物分类中实用性对比

Application of self-organizing mapping artificial neural networks compared with hierarchical clustering analyses in sedimentary type analysis

为确切判定淮北矿区新第三纪沉积物成因类型,分别用自组织人工神经网络(SOM)和聚类分析方法对宿南等矿区的19组样本进行分类.对比发现SOM的分类结果与实际情况更吻合.从机理和应用方式上探讨了两种方法的功能差异,证明SOM方法分类操作过程简便易行,具有残缺自动识别能力,分类结果唯一,在沉积物无监督成因分类中,优于聚类分析方法.

Both SOM （Self-Organizing Mapping） network and hierarchical clustering （HC） methods were tried to classify 19 soil samples from Sunan mining district and near regions for the Neocene sediment type recognition in Huaibei coalfield. It is found that the results of SOM are fit for the known types closer than those of HC. The function differences between the two methods were discussed on their mechanism and applied ways. SOM is proved to be more convenient in classifying applications, and to be able to identify incomplete samples in a unique result without any prior knowledge.