基于曲线投影动态聚类模型的大气质量监测布点研究

聚类数和聚类方法是聚类分析研究的核心,具有重要的实用价值。针对各单项大气质量监测指标的不相容问题,在投影寻踪动态聚类模型的基础上,通过引入S型曲线作为投影方式,提出了基于曲线投影动态聚类的大气质量监测布点研究新方法。从理论上讲,作为一种完全由数据驱动的方法,该模型充分体现了投影寻踪聚类的设计思想,能把最佳聚类数和对应的聚类结果的分析有机地融合在一起,具有物理意义清晰,可操作性强等特点。实例应用结果表明,通过把待研究地区大气监测点多维指标综合成一维投影指标,并利用免疫进化算法优化目标函数,模型能提供明确、客观及稳健的监测布点最佳方案,实现大气质量监测经济性和代表性的平衡。上述研究可为其它非线性、非正态高维数据分类问题开辟一条新途径。

Study on the Atmosphere Qulity Monitoring Sites Based on Curve Prjection Dynamic Cluster Model

The cluster number and the cluster method are key points in clustering analysis,and thus has great value in practical application. Aimed at the non-uniformity of each atmosphere qulity monitoring index and the possible existing problems resulted from adopting linear projection in projection pursuit dynamic cluster(PPDC) model,through introducing S curve to be as projection way,a new curve projection dynamic cluster model,which can be used to study the atmosphere qulity monitoring sites,is proposed. From the point of theoretic analysis,as a method tottly driven by data,the model fully menifests the essence of projection pursuit cluster,and can determine the best cluster number and the corresponding cluster results simultanously,besides that,it is also easy to operate and has clear physical foundation. Taken an actual observation as examples,the study results show that,by synthesizing the multi-dimension indexes of the studied monitoring spots into one dimension of projection values and by using the immune evolution algorithm to optimizing the object function,the model can provide definite,objective and robust scheme for selecting the best atmosphere qulity monitoring sites,and thus realizes the banlance between cost and representiveness in this aspect. the abolve study can offer a new approach for the nonlinear,non-normal,and high dimensional data.