中文摘要：

投影寻踪聚类模型在多因素聚类分析中被广泛应用并取得了满意的效果，然而，该模型还存在如密度窗宽参数取值由经验确定等不足，有待改进提高。本文针对投影寻踪聚类模型的不足，首次把投影寻踪的思想和动态聚类方法结合起来，构造投影指标，提出了投影寻踪动态聚类新模型。新模型在程个运算过程中不需人为给定参数，聚类结果客观、明确，而且它还具有稳定性好、操作简便等特点。天然草地分类的实际应用表明，投影寻踪动态聚类模型切实可行，取得r很好的效果，在多因素聚类分析领域具有广阔的应用前景。

英文摘要：

The present paper is aimed at introducing a projection pursuit dynamic cluster （PPDC） model initiated by the author. The said model has combined the dynamic cluster method with the projection pursuit principle based on the study of the existing problems with the projection pursuit dynamic cluster. As is known, PPC model is widely used in multifactor cluster analysis, though there still remain some problems to be solved in practice, one of which is the cutoff radius as an important parameter. In the model initiated by the authors, a new projection index is included based on dynamic cluster method, whose operation process can be divided into four steps. Thus, our new model enjoys the following advantages over the original one. It has avoided the problem of the parameter calibration successfully in PPC model, but also results in direct output. Therefore, it enables the cluster results to be more objective and definite, as well as robust and easy to operate in practice. As an application example, our model can be applied to the study of a multifactor natural grassland classification. Due to the above said advantages, the results of our research may lead to four major conclusions ： （ 1 ） By means of the linear projection technique, our model can give out six classification indexes on the grassland samples, thus able to convert them into one dimension projection value, indicating the environmental comprehensive quality of the sample. （2） The optimal projection direction can he easily found by using the genetic algorithm so as to classify the samples automatically into three types according to the projection value. Thus, it is possible to get rid of the subjectivity of the original model. （3） It is more appropriate to apply the new model to the study of natural grasslands, classify them so as to get optimum results. （4） The new model not only initiates a new approach to the study of multifactor natural grassland classification, but also provides a powerful tool to solve some similar problems, thus