基于投影寻踪的天然草地分类模型

提出了基于投影寻踪的天然草地分类模型(GQC-RAGAPP)，利用该模型可把各天然草地多维分类指标值综合成一维投影值，投影值越大表示该草地的环境综合质量越高，根据投影值的大小就可对草地样本集进行合理分类。建议用实码加速遗传算法进行GQC—RAGAPP的建模，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过程复杂、编程实现困难的缺点。实例计算的结果说明，直接由样本数据驱动的GQC—RAGAPP模型用于天然草地分类简便可行，适用性和可操作性较强，可应用于各种非线性、非正态高维数据分类、评价等区域可持续发展研究中。

Classification model of natural grassland quality based on projection pursuit

A new method based on projection pursuit for natural grassland quality classific ation, called GQC RAGAPP for short, is presented in this paper. The basic idea of GQC RAGAPP model is to project high dimension indexes of natural grassland q uality classification to projective values in only one dimension space, to descr ibe classification structure by using a projective index function, to search the optimal projective directions according to the projective index function, and t o analyze the classification structure characters of the high dimension data by the projective values. The problem to construct and optimize projective index fu nction is suggested to be optimized by using real coded accelerating genetic alg orithm developed by the authors, called RAGA for short. The modeling of GQC RAGAPP is the key in this paper, which includes four steps as following. Step 1 is to standardize each index of natural grassland quality classification according to the minimum and the maximum in the classification sample set. Step 2 is to construct the projective index function, which aim is to synthesize many dimensions sample data to one dimension z(i) named projective value wi th projective direction a, where a is an unit length vector. The demands of the scattering characters of the projective values of z(i) are that local projec tive dots should be denseness, that it is best to condense the dots to some grou ps, and that the dot groups should be dispersed. Based on the demands, a project ive index function Q(a) is designed as Q(a)=S zD z, where S z is the standard variance of z(i), and D z is the local density. Step 3 is to optimize the projective index function so that the optimal projecti ve direction a * can be estimated. As a general kind of optimization method s based on the mechanics of natural selection and natural genetics in biology, R AGA can be applied to deal with the optimal problem of maximizing the projective index function Q(a) both easily and effectively. Step 4 is to classify natural grassland samples according to the projective valu es z *(i) of the samples, which can be gained by substituting the optimal p rojective direction a * according to Step 3. The series of {z *(i)} ca n be sorted orders from big to small. The bigger the value of z *(i) is, th e better the natural grassland quality is, based on which the sample set can be classified. The computation results of the case study can include four terms as following. (1)In GQC RAGAPP model, many classification indexes values of grassland sample s can be synthesized projection value with only one dimension, which indicates e nvironmental comprehensive quality of grassland samples. (2)The grassland samples can be naturally classified according to the projectio n value of each grassland sample. (3)By using real coding based accelerating genetic algorithm developed by the a uthors, the modeling of GQC RAGAPP can be predigested the realized process of p rojection pursuit technique, and can overcome the shortcomings of large computat ion amount and difficulty of computer programming in traditional projection purs uit methods. (4) Applying GQC RAGAPP model driven directly by samples data to classifying na tural grassland samples is simple and feasible, its computed result is steady, a nd its applicability and maneuverability is good. (5) GQC RAGAPP model can be applied to classification and evaluation of nonline ar, non normal distribution and high dimensional index data in regional sustain able development study.