大数据平台下基于人工免疫系统的MBR膜污染研究

发布时间：2018-05-05 19:33

本文选题：膜生物反应器 + Elman神经网络　；参考：《天津工业大学》2017年硕士论文

【摘要】：膜生物反应器(MBR)作为污水处理领域中的新兴技术,具有良好的发展前景,并且逐渐受到相关领域专家与学者的重视。然而,膜污染严重影响了 MBR技术的发展,如何调节系统运行条件和参数,减缓膜污染现象,以及如何建立膜污染的仿真模型已经称为研究热点。本文旨在建立影响MBR膜污染因素与表征膜污染程度的膜通量间的非线性关系,从而完成对膜通量的预测。在建立仿真预测模型之前,需要对膜生物反应器系统运行过程中产生的数据进行处理,从中提取出建模所需要的数据。随着MBR的发展,系统产生的数据必然会越来越多,因此本文引入大数据分布式系统来完成对膜污染相关数据的存储和处理分析。利用Hadoop常用组件建立完整的数据处理模型,并通过shell脚本完成数据的自动处理。其中包括使用Flume组件完成数据的采集,使用MapReduce程序完成数据的清洗,使用Hive组件建立数据仓库并分析表中数据,以及使用Sqoop组件将分析后的结果数据导入到关系型数据库中。理论研究证明,神经网络能够任意精度逼近任意连续函数,能够建立输入与输出之间的复杂非线性关系模型,在预测领域应用比较广泛。本文使用Elman动态神经网络来建立仿真预测模型,它属于局部回归的反馈网络,具有关联记忆功能,在预测精度上优于传统的BP神经网络。但是,由于Elman神经网络仍采用前馈网络的梯度下降法进行网络训练,因此存在网络训练速度慢、容易陷入局部极小值等缺陷,且有网络权值敏感和网络结构不易确定等问题。针对这些问题,引入基于浓度的自适应免疫遗传算法来优化Elman网络模型,实现了动态递归神经网络的自动构造和设计。实验结果表明,使用免疫遗传算法优化的Elman网络优于未优化的Elman网络,并且在与使用传统GA优化后的模型对比后发现,免疫遗传算法优化的Elman网络效果更好,预测精度有很大提升,且优化后的网络预测比较稳定。
[Abstract]:Membrane bioreactor (MBR), as a new technology in wastewater treatment, has a good prospect of development, and has been gradually paid attention to by experts and scholars in related fields. However, membrane fouling has seriously affected the development of MBR technology. How to adjust the operating conditions and parameters of the system, slow down the membrane fouling phenomenon, and how to establish the simulation model of membrane fouling has become a research hotspot. The purpose of this paper is to establish a nonlinear relationship between the factors affecting the fouling of MBR membrane and the membrane flux which characterizes the fouling degree of the membrane so as to complete the prediction of the membrane flux. Before establishing the simulation prediction model, the data generated during the operation of the membrane bioreactor system should be processed, and the data needed for modeling should be extracted from it. With the development of MBR, more and more data will be generated by the system. Therefore, big data distributed system is introduced to store and process the data related to membrane fouling. A complete data processing model is established by using Hadoop components, and the data is processed automatically by shell script. It includes using Flume component to complete data collection, using MapReduce program to complete data cleaning, using Hive component to build data warehouse and analyzing data in table, and using Sqoop component to import the analyzed result data into relational database. The theoretical study shows that the neural network can approach any continuous function with arbitrary precision and can establish a complex nonlinear relation model between input and output. It is widely used in the field of prediction. In this paper, Elman dynamic neural network is used to establish the simulation prediction model, which belongs to the feedback network of local regression and has the function of associative memory, which is superior to the traditional BP neural network in prediction accuracy. However, the Elman neural network still uses the gradient descent method of feedforward neural network to train the network, so the network training speed is slow, it is easy to fall into the local minimum, and the network weight is sensitive and the network structure is difficult to determine. To solve these problems, an adaptive immune genetic algorithm based on concentration is introduced to optimize the Elman network model, and the automatic construction and design of dynamic recurrent neural network are realized. The experimental results show that the Elman network optimized by immune genetic algorithm is superior to the unoptimized Elman network, and compared with the model optimized by traditional GA, it is found that the Elman network optimized by immune genetic algorithm is more effective. The prediction accuracy is greatly improved, and the network prediction after optimization is relatively stable.
【学位授予单位】：天津工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：X703;TP311.13

【参考文献】