基于回归支持向量机的指标规范值的水质评价模型  

Model of water quality evaluation with normalized indexes values based on regression support vector machines

作　　者：李祚泳[1] 张正健[2] 

机构地区：[1]成都信息工程学院,四川成都610041 [2]中国科学院水利部成都山地灾害与环境研究所,四川成都610041

出　　处：《中国环境科学》2013年第8期1502-1508,共7页China Environmental Science

基　　金：国家自然科学基金资助项目(51179110);科技基础性工作专项项目(2011IM011000)

摘　　要：传统的回归支持向量机的水质评价模型不具有普适性和通用性,当指标较多时,模型的学习效率和求解精度均会受到影响.若适当设定3类水体(地表水、地下水和富营养化水体)各项指标的参照值及指标值的规范变换式,使不同指标的同级标准的规范值差异不大,从而可以认为用规范值表示的不同指标皆'等效'于某个规范指标.因此,可建立用规范值表示的任意m项指标组合皆适用的水质评价的回归支持向量机模型.通过实例对模型的实用性进行了效果检验,结果表明:用基于回归支持向量机的指标规范值的水质评价模型对河桥地表水、黑龙洞泉域地下水和山仔水库富营养化水体的水质评价结果与用BP神经网络评价法、模糊综合评价法和属性识别评价法的评价结果基本一致.Generally, the traditional method to evaluate the water quality using regression support vector machine is not an universal way. The learning efficiency and the accuracy will be significantly influenced due to the increase of the number of the index. For three different water bodies （surface water, groundwater and eutrophic water body）, the present study sets the proper reference values and normalized transformation forms for all the indexes of them. The differences among all the indexes will be decreased for the same quality level water after the normal transformation, and the different indexes represented with normalized values can be equivalent to a certain value. Therefore, it is possible to build the water quality evaluation model using regression support vector machines （SVR） with any combination of m（2〈m〈72） normalized indexes. The practicality of the model was verified by samples from HeQiao surface water, HeiLongDong spring groundwater and eutrophication of SanZi reservoir. The evaluation gets the similar results with the evaluations using BP neural network, fuzzy sets and attribute recognition evaluationary methods.

关 键 词：指标 规范变换 水质评价 回归支持向量机 模型 

分 类 号：X824[环境科学与工程—环境工程]