基于密度聚类自组织RBF神经网络的出水氨氮软测量研究

发布时间：2018-01-09 08:36

本文关键词：基于密度聚类自组织RBF神经网络的出水氨氮软测量研究　出处：《北京工业大学》2016年硕士论文　论文类型：学位论文

【摘要】：近年来,随着社会的快速发展,水污染问题愈发严峻,而水体富营养化是导致水污染的重要原因,水体富营养化会破坏生态环境,影响人类健康,因此避免水体富营养化是我国城市污水处理厂建设的主要目标之一。由于水体富营养化的机理过程复杂,影响因素众多,难以建立其精确的数学预测模型,致使水体富营养化难以预防,而水中氨氮的含量是水体富营养化的关键参数,其值大小可以用来评价水质,预防污染。因此,为了实现污水处理厂出水氨氮的实时测量,文中提出了一种基于密度聚类自组织RBF神经网络的出水氨氮软测量模型,实现了出水氨氮的及时、准确预测。本文主要的研究工作包括以下几点:1.基于污水氨氮处理过程的机理分析与数据处理,确定了出水氨氮软测量模型的辅助变量。辅助变量的选取是建立出水氨氮软测量模型的关键,本文通过对活性污泥法中生物脱氮过程的机理分析及实际污水处理厂可测量变量,总结出与出水氨氮相关的7种辅助变量,在对数据进行归一化处理后,利用主元分析法对初步的7种辅助变量进行了去相关性处理,最终将出水氨氮软测量模型的辅助变量维数由7维降为5维。2.针对RBF神经网络结构参数难以确定的问题,设计出一种密度聚类自组织RBF神经网络。该密度聚类算法以某个样本点密度值大小及样本间的欧氏距离为条件进行RBF神经网络结构的自组织调整,从而实现网络结构的确定,并利用梯度下降算法对网络参数进行训练,确定最终的RBF神经网络结构及参数。非线性系统逼近实验表明:所提出的自组织机制能够优化RBF神经网络结构,提高了网络预测精度。3.建立了一种基于密度聚类自组织RBF神经网络的出水氨氮软测量模型,解决了出水氨氮在线测量精度不高的问题。将提出的密度聚类自组织RBF神经网络应用于出水氨氮的软测量模型中,由于密度聚类自组织RBF神经网络能够根据样本数据的特点进行网络结构的自组织调整,使得建立的软测量模型更接近实际的污水处理过程,实验结果验证了所建立的出水氨氮软测量模型的有效性。4.开发了一种出水氨氮软测量平台。该平台主要包括用户注册及登录模块、数据处理模块、模型训练及预测模块。首先,利用LabVIEW 2012软件编写了平台操作界面,主要集合了模型选择、参数设置等功能;然后利用Access 2013数据库和Matlab R2012a软件编写了后台运行程序,以实现用户信息存储及软测量模型调用;最后,通过用户信息、数据处理、模型调用等模块间的信息传输,实现出水氨氮软测量过程的可视化。
[Abstract]:In recent years, with the rapid development of society, the problem of water pollution is becoming more and more serious, and eutrophication is an important cause of water pollution, water eutrophication will damage the ecological environment and affect human health. Therefore, avoiding eutrophication of water body is one of the main goals of the construction of municipal wastewater treatment plant in China. Because of the complex mechanism of eutrophication and many factors affecting the eutrophication, it is difficult to establish its accurate mathematical prediction model. It is difficult to prevent eutrophication, and the content of ammonia nitrogen in water is the key parameter of eutrophication, and its value can be used to evaluate water quality and prevent pollution. In order to realize the real-time measurement of ammonia nitrogen in effluent of wastewater treatment plant, a soft sensing model of ammonia nitrogen in effluent based on density clustering self-organizing RBF neural network is proposed in this paper. The main research work in this paper includes the following points: 1. Mechanism analysis and data processing based on the process of ammonia nitrogen treatment of sewage. The auxiliary variables of the effluent ammonia nitrogen soft sensing model are determined, and the selection of the auxiliary variables is the key to the establishment of the effluent ammonia nitrogen soft sensor model. Based on the mechanism analysis of biological denitrification process in activated sludge process and the measurable variables in actual wastewater treatment plant, seven auxiliary variables related to ammonia nitrogen in effluent were summarized, and the data were normalized. The primary 7 auxiliary variables were treated by principal component analysis (PCA). Finally, the dimension of the auxiliary variable of the effluent ammonia soft sensor model is reduced from 7 to 5. 2. Aiming at the problem that the parameters of RBF neural network structure are difficult to determine. A density clustering self-organizing RBF neural network is designed. The density clustering algorithm is based on the Euclidean distance between samples and the size of the density of a sample point to adjust the RBF neural network structure. In order to determine the network structure and use gradient descent algorithm to train the network parameters. The final RBF neural network structure and parameters are determined. The nonlinear system approximation experiments show that the proposed self-organization mechanism can optimize the RBF neural network structure. A soft sensing model of effluent ammonia nitrogen based on density clustering self-organizing RBF neural network was established. The problem of low precision of on-line measurement of ammonia nitrogen in effluent was solved. The density clustering self-organizing RBF neural network was applied to the soft sensing model of ammonia nitrogen in effluent. Because the density clustering self-organizing RBF neural network can adjust the network structure according to the characteristics of the sample data, the established soft-sensor model is closer to the actual sewage treatment process. The experimental results verify the effectiveness of the model. 4. A soft sensing platform of effluent ammonia nitrogen is developed. The platform mainly includes user registration and login module, data processing module. Model training and prediction module. Firstly, the platform operating interface is compiled by using LabVIEW 2012 software, which mainly integrates the functions of model selection, parameter setting and so on. Then, the background running program is written by using Access 2013 database and Matlab R2012a software to realize user information storage and soft sensor model call. Finally, through the information transmission between user information, data processing, model transfer and other modules, the visualization of effluent ammonia soft sensing process is realized.
【学位授予单位】：北京工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：X832;TP183

【相似文献】