Research On Data-driven Method For Diagnosis And Prediction Of Membrane Fouling In Membrane Water Treatment

How to reduce membrane fouling in the process of membrane water treatment and improve the quality of membrane water treatment effluent while reducing energy consumption has become a major issue in the development of modern wastewater treatment field that cannot be avoided.Membrane fouling diagnosis and prediction technology is an important method and necessary means to improve the reliability of membrane water treatment system operation and reduce the system operation cost.Membrane water treatment process is mostly a non-linear system with strong interference,complex structure,uncertain parameters,dynamic changes,strong coupling of membrane fouling factors,etc.It is difficult to accurately diagnose and predict membrane fouling.Therefore,it is a very meaningful research topic to explore how to use advanced science and technology to diagnose and predict membrane fouling in a purposeful,methodical and targeted way.In this dissertation,we take typical membrane modules of membrane water treatment system as the research object,and investigate the diagnosis and prediction method of membrane fouling of membrane water treatment based on data-driven approach.The main innovative research results of the thesis are as follows:(1)A multi-objective jellyfish search adaptive deep belief network(MOJS-ADBN)based membrane fouling diagnosis method is proposed for the strong nonlinearity and large time-varying characteristics of membrane fouling in membrane water treatment process.MOJS-ADBN is an improved diagnostic model based on deep belief network.First,the adaptive learning rate is introduced into the unsupervised pre-training phase of DBN to improve the convergence speed of the network.Second,the MOJS method is used instead of the gradient-based layer-by-layer weight fine-tuning method in the traditional DBN to improve the network feature extraction capability.Meanwhile,the Liapunov function is constructed to prove the convergence of the MOJS-ADBN learning process.Finally,the MOJS-ADBN is used in membrane module membrane fouling diagnosis for model diagnosis performance validation.The experimental results show that the proposed MOJS-ADBN diagnostic model has fast convergence and high diagnostic accuracy,which can provide a theoretical basis for membrane fouling diagnosis during the actual operation of membrane water treatment.(2)Aiming at the problem of difficult extraction of fault features of membrane modules,which leads to complicated model computation and insufficient diagnostic accuracy,a method combining attention mechanism and convolutional neural network(ECA-CNN)is proposed to solve the difficulty of fault feature extraction,reduce model complexity,and improve the diagnostic accuracy of membrane fouling of membrane modules.First,the picture features of the input layer are extracted using convolutional kernels,while a corrected linear unit is connected behind each convolutional layer,and a batch normalization layer is added to solve the problem of internal covariate shifts to improve the nonlinear model representation.Secondly,the ECA attention mechanism module is added after the batch normalization layer to extract important features and then connect the pooling layer to reduce the computational complexity of the network and improve the accuracy and efficiency of the network.Finally,membrane fouling diagnosis experiments are carried out to validate the study with membrane module operation data.The experimental results show that the proposed method can effectively improve the diagnosis accuracy of membrane fouling,and can realize the difficult classification and localization of all membrane fouling.(3)To address the problem of insufficient extraction capability of membrane fouling features from membrane bioreactor modules,which leads to complex structure of membrane fouling data and thus cannot achieve efficient localization and classification of membrane fouling in membrane bioreactors,a convolutional block attention module multiple input convolutional neural network(CBAM-MUL-CNN)model based on attention mechanism is proposed.First,the time domain and frequency domain information of the membrane contaminated data are used as the input of the CNN,and the features are extracted by the convolutional layer.Then,a fully-connected layer is used to splice the time-domain and frequency-domain features and input to the classifier for classification.The batch normalization layer in the model can effectively prevent gradient disappearance,the Re LU layer can improve the nonlinear model expression,the CBAM module can simplify the model complexity and improve the feature expression of the network,and the pooling layer can improve the fault tolerance of the model.The comparison results show that the model has excellent comprehensive performance in membrane fouling diagnosis experiments of parallel hollow fiber membrane devices,and can effectively achieve efficient classification and localization of all membrane fouling,which enables membrane water treatment to improve effluent quality and reduce energy consumption at the same time,providing a theoretical basis for practical production.(4)A membrane fouling prediction method based on improved multiple objective sparrow search algorithm adaptive deep belief network(IMOSSA-ADBN)is proposed to address the uncertainty,strong nonlinearity,strong coupling and non-smoothness of the membrane fouling prediction problem in membrane water treatment process.Firstly,the fixed learning rate in the unsupervised pre-training phase of DBN network is adaptively designed to improve the convergence speed of the network and enhance the training efficiency of the network.Secondly,the IMOSSA method is used to optimize the layer-by-layer weight fine-tuning method in the traditional DBN to improve the model prediction accuracy.Meanwhile,the convergence of the IMOSSA-ADBN model learning process is demonstrated.Finally,IMOSSA-ADBN is used in the prediction of membrane fouling.The experimental results show that the proposed IMOSSA-ADBN prediction model has fast convergence and high prediction accuracy,which can meet the requirements of membrane fouling monitoring accuracy and operation efficiency of the actual membrane water treatment process.(5)The membrane water treatment system has many parameters with large amount of data and high dimensionality,which makes the selection of feature variables difficult and the prediction model difficult to establish accurately.Therefore,a membrane pollution prediction model based on improved convolutional neural network and multistrategy fusion long-and short-term memory network is proposed with the deep learning model convolutional neural network as the main technical tool.Firstly,the improved long short-term memory(ILSTM)network is obtained by optimizing the structural parameters of the long short-term memory network with an improved multistrategy fusion bottle sheath search algorithm,which improves the convergence speed of the network and reduces the complexity and non-smoothness of membrane fouling.Secondly,the spatial attention channel and noise reduction module are fused to design a cross-channel soft threshold noise reduction attention mechanism to process membrane fouling information with high quality,and combine CNN and ILSTM to build a membrane fouling prediction model to achieve efficient prediction of membrane fouling in membrane water treatment.The prediction accuracy of the proposed model is greatly improved compared with CNN or LSTM alone,and the prediction is more accurate and faster with the cross-channel soft-threshold noise reduction attention mechanism.This dissertation combines the nonlinear characteristics of membrane water treatment system,in the context of strong noise and multiple disturbances,and conducts an in-depth study on data-driven membrane fouling diagnosis and prediction methods,and proposes a variety of new methods for membrane fouling diagnosis and prediction,which to a certain extent solves the problems of diagnosis and prediction caused by system complexity,coupling and uncertainty of membrane fouling factors under various complex situations such as sudden changes and multiple working conditions of membrane water treatment system.The lack of methods and low prediction accuracy,poor real-time,robustness and other problems.The research results of this dissertation have important reference value and practical significance for the development of intelligent diagnosis and prediction technology of membrane water treatment system and guarantee the safety and reliability of membrane water treatment system operation.