Abstract: Multi-agent Genetic Algorithm（MGA） is applied to effectively and successfully optimize the optimal projection vector in the Projection Pursuit Clustering（PPC） model. Two different normalization methods, without changing the evolution process, for projection vector get the same results for three cases with various amounts of samples. The two different Maximum and Minimum Normalization Methods（MMNMs） for evaluation indexes yield opposite number of projection vector coefficients. The difference between the projected values of the same sample with two different MMNMs is constant. The PPC model is thus suitable to exploratory research and confirmatory analysis. PPC model is mainly applied to large sample situation and gets properly reliable and effective results. The correlation between variables will influence PPC model’s effectiveness and rational.

Key words: Multi-agent Genetic Algorithm（MGA）, Projection Pursuit Clustering（PPC） model, projection vector, evaluation indexes, exploratory research, confirmatory analysis

摘要： 采用多智能体遗传算法（MGA）进行投影寻踪聚类（PPC）建模，对投影向量约束条件采用两种不改变迭代进化过程的归一化处理方法，经三种不同类型的数据分别进行建模，得到了相同的建模结果，有效地解决了求解最佳投影向量的最优化问题。对评价指标数据采用极大化或极小化（不同的归一化）处理方式，得到的投影向量系数互为相反数，同一样本的投影值之间只相差一个常数，说明PPC建模技术既可用于探索性研究，也可用于验证性分析。PPC技术主要用于大样本情况，稳健性和可靠性均较好；指标之间存在明显的相关性，会影响建模结果的有效性和合理性。

关键词: 多智能体遗传优化算法, 投影寻踪聚类模型, 投影向量, 评价指标, 探索性研究, 验证性分析