用投影寻踪分类模型进行环境监测优化布点

Optimizing environmental monitoring spot distribution with PPC-RAGA Model

为处理各单项环境监测指标优选结果的不相容问题 ,提出用投影寻踪分类模型 ( PPC-RAGA )进行环境监测优化布点的新方法。利用该模型可把监测点多维环境指标值综合成一维投影值 ,投影值越大表示该监测点的环境综合质量越差 ,根据投影值的大小就可对环境监测点样本集进行合理分类。采用实码加速遗传算法进行 PPC-RAGA建模 ,简化了投影寻踪技术的实现过程 ,克服了目前投影寻踪技术计算复杂、编程实现困难的缺点 ,为投影寻踪技术在各种系统工程中的广泛应用提供了新的工具。应用实例表明 ,直接由样本数据驱动的 PPC-RAGA模型用于环境监测优化布点 ,简便可行 ,计算结果稳健 ,适用性和可操作性强 ,在非线性。

The present paper aims at introducing a new method known as the projection pursuing classification model (short for PPC-RAGA) in hoping to optimize the environmental monitoring spot distribution. In order to solve the non-uniformity problem when optimizing the environmental monitoring site distribution in correspondence with each index of the monitoring spots, it is necessary to obtain sufficiently diversified information from the indexes of the monitoring spots and increase the calculation precision. With the PPC-RAGA model developed by the present authors, the projection value can be synthesized with only one dimension from the many indexes of the monitoring spots, which helps to indicate the comprehensive environmental quality of the monitoring spots as well as the natural classification of the sampled monitoring spots in relation to the projection value of each spot. The so-called PPC-RAGA model developed by the authors makes it possible to predigest the realized process of the traditional projection pursuit methods with the real coding based on the accelerating genetic algorithm, thus overcoming the defects of large size of computation and computer programming complication in traditional projection methods. The above advantages are enough to make the method a new trenchancy tool for wide application of the projection pursuit technique to such applied systems engineering, as the environment systems engineering, resources systems engineering and city systems engineering. The actual applied results prove that the model is simple and feasible in actual applications driven by the samples data directly to optimize the monitoring spots distribution with high applicability and maneuverability, especially in classifying the nonlinear, non-normal distribution and high dimensional data.