Design Of Self-organizing Recurrent Fuzzy Neural Network And Its Application In Wastewater Treatment

Fuzzy neural network(FNN)is a network model that combines the nonlinear analysis capabilities of artificial neural network(ANN)and the fuzzy inference capabilities of fuzzy systems.Aiming at the ambiguity of human brain thinking,FNN uses numerical information to construct specific non-linear mapping to realize the identification and judgment of fuzzy things.It is an advanced information processing system that can be applied to more complex problems and a wider range of fields.However,the existing FNN still has many unsolved research problems in the process of internal dynamic information transfer and structural adaptive adjustment,which limits the modeling ability and application field of FNN to a certain extent.Therefore,it is of great significance to study FNN which can adaptively transfer the internal information of the network and adjust the network structure.To solve the above problems,this paper proposes a self-organizing recurrent fuzzy neural network based on multivariate time series analysis(MTSA-SORFNN)model by studying and analyzing the dynamic characteristics of FNN.In this model,the prediction factors are introduced into the recurrent layer by the wavelet transform fuzzy Markov chain(WTFMC)algorithm,which enhances the adaptability of the recursion link of the network.At the same time,the weighted dynamic time warping(WDTW)algorithm and sensitivity analysis(SA)algorithm are used to optimize the structure of the network.Finally,the network model is used for several benchmark problems and the prediction of key water quality parameters of wastewater treatment,and the development of an intelligent system for soft measurement of key effluent parameters of wastewater treatment is completed.Experimental results show that MTSA-SORFNN shows good performance in both convergence speed and modeling accuracy,and achieves accurate prediction of nonlinear problems.The research content and innovative work of the paper are as follows:(1)Design and research of recursive mechanism based on WTFMC algorithm.In order to solve the problem of self-organizing recurrent fuzzy neural network(SORFNN)which is difficult to adapt to its recursion,a recursion mechanism based on WTFMC algorithm is proposed.Firstly,the fuzzy logic rules of neurons in the hidden layer are recorded in time dimension to construct multivariate time series(MTS);secondly,the MTS is decomposed by wavelet transform,and combined with fuzzy Markov chain to predict the change trend of the subsequence obtained after decomposition;finally,the prediction quantities are combined and brought into the calculation of the recursion layer to enhance the adaptability of the recursion quantity of the network.The experimental results show that the recursive mechanism can detect the internal variation of the network and effectively improve the convergence speed of the network.(2)Design and research of self-organizing mechanism based on WDTW algorithm and SA algorithm.To solve the problem that the structure of SORFNN is difficult to determine,a self-organization mechanism based on WDTW algorithm and SA algorithm is proposed.Firstly,the fuzzy logic rules of hidden layer neurons are recorded in time dimension;secondly,WDTW algorithm is used to analyze the correlation between neurons,so as to guide the combination of neurons;at the same time,SA algorithm is used to calculate the cumulative contribution of neurons,so as to guide the segmentation and deletion of neurons.The experimental results show that the self-organization mechanism can obtain a more concise network structure and effectively improve the prediction accuracy of the network.(3)Research and research of the strategies of structural parameters adjustment of SORFNN.In order to ensure the stability of SORFNN when the structure changes,the adjustment strategies of structural parameters are proposed.Firstly,the corresponding parameters adjustment algorithms are designed for the combining stage,splitting stage and pruning stage of neurons;secondly,the corresponding thresholds adjustments strategy are designed to improve the efficiency of network structure optimization;finally,through the convergence analysis,the stability of the network’s output under this structural parameters adjustment strategies are proved,and the network’s oscillation is avoided.(4)Design and research of the intelligent system for soft-sensing of key effluent parameters of wastewater treatment.In order to solve the problems of complicated operation and difficult real-time and accurate on-line detection of key effluent parameters in wastewater treatment,an intelligent system for soft measurement of key effluent parameters of wastewater treatment based on the MTSA-SORFNN model is designed and developed.According to the analysis of system requirements,it is divided into four modules: user management,data collection,ammonia nitrogen prediction and model theory.The system uses Visual Studio 2010 as the development platform,by calling the SQL Server 2008 database to achieve user information management,water plant data calling and preprocessing,by calling the preset MATLAB program of MTSA-SORFNN to realize the training and prediction of the soft sensor model.Compared with traditional detection methods,the soft measurement system has the advantages of fast detection,low cost and good practicability,and has practical significance for ensuring stable and efficient operation of wastewater treatment plants.