基于因子分析的Hopfield神经网络在水质评价的应用  

Application of Hopfield Neural Network Based on Factor Analysis to Water Quality Evaluation

作　　者：卢文喜[1] 初海波[1] 王喜华[1] 龚磊[1] 

机构地区：[1]吉林大学环境与资源学院,吉林长春130021

出　　处：《水土保持通报》2012年第1期197-200,237,共5页Bulletin of Soil and Water Conservation

基　　金：国家水体污染控制与治理科技重大专项"辽河流域控制单元水质目标管理技术"(2009ZX07526-006-04-01);国家自然科学基金项目"基于数值模拟的表面活性剂强化的DNAPLs污染含水层修复过程优化问题研究"(41072171)

摘　　要：针对Hopfield神经网络的过度拟合问题,在因子分析的基础上,结合Hopfield神经网络模型提出了因子分析—Hopfield神经网络模型。以东辽河为例,采用因子分析法确定7个水质评价因子,再建立5×7的Hopfield神经网络进行水质综合评价,并与单一的Hopfield网络和传统的内梅罗指数法的结果进行了比较。结果表明,因子分析—Hopfield神经网络明显优于单一的Hopfield神经网络,不仅在一定程度上弥补了因子分析在实际应用中没有实现水质分级的缺陷,而且有效地降低了Hopfield神经网络的过度拟和的程度,评价结果更为科学合理,为水质综合评价提供了一种新的方法,具有极好的应用前景。To solve the over-fitting problem of Hopfield neural network,Hopfield neural network model with factor analysis was proposed based on factor analysis combined with Hopfield neural network model.Taking Dongliao River for an example,the model determined seven water quality evaluation factors using factor analysis method,created a 5×7 Hopfield neural network to evaluate water quality comprehensively,and compared the results from single Hopfield neural network and traditional Nemero Index method.Results showed that the factor analysis Hopfield neural network is much better than single Hopfield neural network.It not only makes up for the defect that factor analysis does not achieve the classification of water quality in practical applications,but also effectively reduces the extent of over-fitting of Hopfield neural network.The evaluation results are more reasonable and the model provides a new approach to comprehensive water quality evaluation with some excellent prospects.

关 键 词：因子分析 HOPFIELD神经网络 水质评价 东辽河 

分 类 号：X824[环境科学与工程—环境工程]