T-S模糊神经网络模型训练样本构建及其在鸣翠湖水质评价中的应用  

Construction of training samples for T-S fuzzy neural network model and its application in water quality evaluation of Mingcui Lake

作　　者：杨程 郭亚昆[2] 郑兰香[3] 李春光 景何仿 YANG Cheng;GUO Ya-kun;ZHENG Lan-xiang;LI Chun-guang;JING He-fang(College of Civil Engineering,North Minzu University,Yinchuan 750021,China;Faculty of Engineering&Informatics,University of Bradford,Bradford BD71DP,UK;School of Resources and Environment,Ningxia University,Yinchuan 750021,China)

机构地区：[1]北方民族大学土木工程学院,银川750021 [2]工程与信息学院布拉德福德大学,布拉德福德BD71DP [3]宁夏大学资源环境学院,银川750021

出　　处：《水动力学研究与进展（A辑）》2020年第3期356-366,共11页Chinese Journal of Hydrodynamics

基　　金：国家自然科学基金项目(11761005,11861003);宁夏自然科学基金项目(2019AAC03136);宁夏科学技术厅重点研发计划项目(2019BEG03048);宁夏教育厅高等学校科学研究项目(NGY2018040);国家留学基金委公派访问学者项目(201808645049)。

摘　　要：该文利用T-S模糊神经网络模型评价鸣翠湖水质。根据不同的训练样本数量和训练次数设置多种工况,基于T-S模糊神经网络探讨了训练样本的三种构成方法,利用MATLAB软件进行编程计算。结果表明:训练样本的构成方式和数量明显影响神经网络的训练效果;由标准样本或监测样本训练的T-S模糊神经网络模型认知能力和泛化能力不足,对检验样本评价的准确率在80%以内。当混合样本中有足够多的监测样本时,可训练神经网络模型完全准确评价检验样本。将混合样本训练的模型应用于银川鸣翠湖的水质评价,结果显示从2014年到2019年水质类别从Ⅳ类逐渐提高到Ⅲ类,评价结果与当地实际情况相符。说明采用混合样本进行水质预测评价是合理的,以混合样本作为训练样本是一种简单有效的数据处理方法。In this paper,the T-S fuzzy neural network model is used to evaluate the water quality of Mingcui lake.The various working conditions are set according to the number of training samples and training times,and three different samples are chosen to train T-S fuzzy neural network model.MATLAB software is used for computation.Results show that the composition and number of training samples significantly affect the training effect of the neural network,the T-S fuzzy neural network trained by standard samples or monitor samples is insufficient in cognitive ability and generalization,and the accuracy rate of the evaluation of inspection samples is less than 80%.Furthermore,in order to improve the prediction,mixed samples are also used.Results show that as long as there are enough monitoring samples in the mixed samples,the T-S fuzzy neural network model trained by mixed samples can completely and accurately evaluate the inspection samples.Finally,the model is applied to evaluate the water quality of Mingcui lake in Yinchuan,China.The evaluation results show that,the category of water quality gradually increased from IV in 2014 toⅢin 2019.The results are consistent with the local actual situation.Therefore,it is reasonable to use mixed samples for water quality prediction and evaluation and it is a simple and effective data processing method to use mixed samples as training samples.

关 键 词：模糊神经网络 训练样本 水质评价 混合样本 

分 类 号：X824[环境科学与工程—环境工程]