Font Size: a A A

Research On The Application Of SA-RNN Network Improved By Bayesian Classifier And CFD In MBR

Posted on:2020-07-23Degree:MasterType:Thesis
Country:ChinaCandidate:M ZhangFull Text:PDF
GTID:2431330626464278Subject:Computer technology
Abstract/Summary:PDF Full Text Request
When MBR(membrane bio-reactor)system treats sewage,membrane pollution will not only shorten the service life of the membrane module,but also increase the operating cost of the system.The sludge production and system water production are important parameters for measuring MBR membrane pollution.This article calculates these two parameters by establishing an intelligent model to measure the membrane pollution degree of MBR system.MBR system membrane pollution prevention and control has a certain reference value.In the calculation of sludge production,this paper establishes a MBR membrane pollution calculation model based on the RNN network.In order to further improve the calculation accuracy of the sludge output by the RNN network,this paper uses SA algorithm and Bayesian classifier to make two Improvement.1.When training an RNN network,because the learning rate may be constant,the gradient descent method may not reach the global minimum.This paper uses the SA algorithm to dynamically control the size of the learning rate so that the gradient descent algorithm can reach the global minimum to improve training.effect.2.Due to the large difference in the scope of different types of sewage parameters,this article uses the industrial sewage data and domestic sewage data to train two RNN models,and simultaneously trains a Bayesian classifier(to enable it to perform domestic sewage and industrial sewage Classification);When calculating the sludge output,the Bayesian classifier is used to classify the input data,and then the corresponding RNN model is selected to calculate the sludge output according to the classification result,so as to improve the calculation accuracy of the sludge output by the RNN network.Based on this,this article uses Python programming to conduct experimental analysis.The analysis results show that after the improvement of the SA algorithm and the Bayesian classifier,the calculation accuracy of the sludge output by the RNN network has increased by an average of 19%,and the relative error is lower than that of the ordinary RNN network.It solves the problem of measuring the membrane pollution degree by calculating the sludge output,and has certain reference value for MBR engineering design and research.In the calculation of MBR system water production,this paper takes the tubular MBR as a sample,and uses CFD simulation software ANSYS 16.0 to carry out simulation simulation and calculation research.In the research process,the geometricmodel of the tubular MBR was first established with the ICEM CFD preprocessor in ANSYS 16.0,and the structured mesh was divided;then the mesh file was solved with the FLUENT solver in ANSYS 16.0 Calculate and set a flow monitoring window at the model outlet to calculate the water production of the tubular MBR to measure the membrane pollution degree of the MBR system.Finally,this paper compares the calculation results of the solver with the actual water output of the tubular MBR system.The comparison results show that the calculation results of the solver have high calculation accuracy and relatively small errors,which achieves accurate calculation of the MBR system production The purpose of water volume solves the problem of measuring the degree of membrane pollution by calculating the amount of water produced.It has certain reference value for future research on MBR membrane pollution field.
Keywords/Search Tags:SA calculation, Bayesian classifier, RNN network, membrane fouling, computational fluid dynamics
PDF Full Text Request
Related items