投影寻踪动态聚类模型及其在地下水分类中的应用

针对投影寻踪聚类模型的不足，结合动态聚类方法对投影寻踪聚类模型进行了改进，建立了投影寻踪动态聚类模型。首先，利用投影技术将多因素（高维）问题投影到一维线性空间，以达到在一维空间研究高维数据的目标；其次，以动态聚类方法构造新的投影指标，对投影到线性空间的反映高维数据结构或特征的投影特征值序列进行聚类分析，进而完成多因素样本聚类分析。投影寻踪动态聚类模型是高维数据样本聚类分析的一种有效的统计方法，模型在整个运算过程中毋需人为给定参数，聚类结果合理、客观。投影寻踪动态聚类模型在地下水分类中的成功应用表明，投影寻踪动态聚类模型具有稳定性好、分类结果明确、操作简便等特点，为多因素聚类分析提供了一种新方法，有着广阔的应用前景。

Projection Pursuit Dynamic Cluster Model and Its Application in Groundwater Classification

A projection pursuit dynamic cluster (PPDC) model is proposed, which combines projection pursuit principle with dynamic cluster rule. Firstly, multifactor cluster problem is converted into single-factor cluster problem according to linear projection technique. Secondly, a new projection index based on dynamic cluster rule is constructed in the PPDC model, which would finish the sample clustering based on the projected characteristic value. In comparison with the existing projection pursuit cluster (PPC) model, we construct a new projection index based on dynamic cluster method in the PPDC model, which successfully avoids the problem of parameter calibration in the PPC model and makes the cluster results more objective. On the other hand, the cluster results can be outputted directly according to the PPDC model, but in the PPC model the cluster results can be got using other method to re-analyze projected characteristic values. A case study of groundwater classification is given at last, and the results show that the PPDC model is reasonable and easy to operate in practice. The PPDC model is a new method for multifactor cluster analysis and has a bright future in cluster analysis field.