Font Size: a A A

Research On Prediction Of Dissolved Oxygen Concentration And Distribution Equilibrium In Artificial Water Based On GF-LSTM And GAN Network

Posted on:2021-03-25Degree:MasterType:Thesis
Country:ChinaCandidate:J W ZhengFull Text:PDF
GTID:2531306110472894Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
This research proposes a prediction strategy of dissolved oxygen(DO) concentration in artificial water which is based on long-short-term memory(LSTM)neural network and generative adversarial network(GAN).This strategy aims to improve the efficiency of water treatment method such as artificial flowing which is designed to balance DO distribution in artificial water to prevent water eutrophication.Firstly,taking a 35m~2 water body area of Jinghu which located in the school as the research object,using a DC pump with different voltages to push the flow,using an intelligent water quality monitoring system to periodically and regularly collect dissolved oxygen concentration data in the water body as a raw data sample.After that,we enlarged the data sets by training the original data sets with generative adversarial networks(GAN)to get the adequate,similar and reliable new data.The genetic algorithm and the improved first-order filtering algorithm are used to process the noise data of dissolved oxygen,and the LSTM network is used to construct a dissolved oxygen concentration prediction model:GF-LSTM(Genetic And Filtering Algorithm-Long Short Term Memory Network).The experimental results show that:compared with the commonly used BP network,the average error of GF-LSTM network prediction is reduced by 62%and the mean square error is reduced by 75%;compared with the M-SVR(Multioutput-Support Vactor Regression)network,the average error of GF-LSTM network prediction is reduced by 68%and the mean square error is reduced by 82%;compared with the traditional LSTM network,the average error of GF-LSTM network prediction is reduced by 22%,the mean square error is reduced by 50%.Accordingly in this case,GF-LSTM network can work with higher accuracy and better generalization ability.Based on the DO concentration data,use the RBF spatial interpolation method to reconstruct the dissolved oxygen concentration distribution map,and create the evaluation standard of dissolve oxygen distribution equilibrium:discrete coefficient,can be used to instruct water treatment more efficiently.
Keywords/Search Tags:dissolve oxygen concentration prediction, genetic and filtering algorithm- long short term memory network, generative adversarial network, discrete coefficient
PDF Full Text Request
Related items