RBF神经网络在MBR膜污染仿真预测中的应用研究

发布时间：2018-07-27 19:25

【摘要】：水污染问题随着世界经济的快速发展变得日益严重,如何预防和治理污水的问题成为许多学者的研究焦点。膜生物反应器(MBR)是一种新型高效的污水处理技术,它很好的结合了膜分离技术和生物反应技术,现在已广泛应用于生活污水,有机废水和工业废水等污水治理领域。膜污染会影响MBR膜分离装置的性能,严重的膜污染会导致膜通量的降低和膜使用寿命的缩短,这将增加MBR系统的运行成本。因此,对膜污染机理和预测控制方法的研究就是要确保膜生物反应器能够在稳定的低能耗条件下实现较大通量的运行。本文对膜生物反应器的工作机理、膜污染形成过程、膜污染影响及控制因素等基本问题进行了分析讨论。针对膜污染涉及的影响因素较为复杂的问题,首先通过主成分分析(PCA)方法选取出三种贡献率在95%以上的膜污染因子:混合液悬浮固体(MLSS)、操作压力(△P)和温度T。设定这三个影响因子为MBR膜污染仿真的三大输入参数,以膜通量作为输出参数,建立基于RBF神经网络的的膜污染智能仿真预测模型,实现对膜污染的预测控制。引入遗传算法,利用其突出的寻优特点对RBF预测模型的误差、训练速度、最大神经元数目以及神经元的间隔四个参数进行优化,提高RBF网络的仿真性能。整个实验过程采用MATLAB编程实现,结果表明,通过遗传算法优化的RBF神经网络对预测精度有明显提升,达到了预期效果。整个实验过程具有一定的理论价值和实践意义,对MBR实际工程应该可以起到积极的指导作用。由于膜污染的过程会产生大量数据,对污染数据进行存储和处理,能够更加便利的了解MBR系统不同时间段的膜污染状态,再结合膜污染预测模型的高精度预测功能,可以整体提高对膜污染的应对能力。Hadoop是目前应用最广泛的大数据处理平台,具有分布式处理信息的能力,将其引入MBR工艺中,利用Hive的批处理特性以及Impala的实时查询功能存储并处理庞杂的海量数据,构建分布式MBR膜污染数据存储分析平台,为控制膜污染进程提供了便利的信息化处理平台。
[Abstract]:With the rapid development of the world economy, the problem of water pollution becomes more and more serious. Membrane bioreactor (MBR) is a new and efficient wastewater treatment technology, which combines membrane separation technology and biological reaction technology. It has been widely used in sewage treatment fields such as domestic sewage, organic wastewater and industrial wastewater. Membrane fouling will affect the performance of MBR membrane separation device, and serious membrane fouling will lead to the decrease of membrane flux and the shortening of membrane service life, which will increase the running cost of MBR system. Therefore, the study of membrane fouling mechanism and predictive control method is to ensure that the membrane bioreactor can achieve large flux operation under stable and low energy consumption conditions. In this paper, the working mechanism of membrane bioreactor, the formation process of membrane fouling, the influence of membrane fouling and the control factors are analyzed and discussed. In view of the complex factors involved in membrane fouling, three kinds of membrane fouling factors, which contribute more than 95%, were selected by principal component analysis (PCA) method: the operating pressure (P) and temperature (T) of suspension solid (MLSS),. The three influence factors are set up as the three input parameters of MBR membrane fouling simulation. The membrane flux is taken as the output parameter, and the intelligent simulation prediction model of membrane fouling based on RBF neural network is established to realize the predictive control of membrane fouling. Genetic algorithm is introduced to optimize the four parameters of RBF prediction model, such as error, training speed, maximum number of neurons and the interval of neurons, so as to improve the simulation performance of RBF network. The whole experiment process is realized by MATLAB programming. The results show that the RBF neural network optimized by genetic algorithm can obviously improve the prediction accuracy and achieve the expected effect. The whole experiment process has certain theoretical value and practical significance, and should play a positive guiding role in the practical project of MBR. Because the process of membrane fouling will produce a large amount of data, the fouling data can be stored and processed, which can be more convenient to understand the membrane fouling status in different time periods of MBR system, and then combine the high precision prediction function of membrane fouling prediction model. Hadoop is the most widely used big data processing platform at present, which has the ability of distributed processing information. It is introduced into MBR process. Using the batch processing characteristic of Hive and the real-time query function of Impala to store and process massive data, a distributed MBR membrane fouling data storage and analysis platform is constructed, which provides a convenient information processing platform for controlling membrane fouling process.
【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;TP183

【相似文献】