Soft-sensing Research For Effluent BOD Based On A Self-organizing Recurrent RBF Neural Network

In recent years,with the rapid development of our country,the industrial sewage and domestic sewage in our country are increasing day by day.The problem of water pollution in our country has become very serious.The problem of sewage treatment has appeared more and more in people's vision.Whether the sewage can be effectively treated is closely related to people's future life.Biochemical oxygen demand(BOD)is the amount of oxygen consumed to convert organic oxygen into inorganic matter.It reflects the pollution degree of sewage and has been proved to be an important variable in water quality management and planning.The accurate measurement of BOD becomes a necessary condition for ensuring the normal operation of the sewage treatment control system and the effluent quality.Therefore,in order to solve the problem that BOD is difficult to be accurately measured in the actual sewage treatment process,this paper proposes a soft-sensing model of effluent BOD based on self-organizing recurrent RBF neural network.Firstly,the model selects the input variables of the neural network through the gravitational search algorithm(GSA),and then uses the directed information and mutual information to dynamically adjust the structure of the recurrent RBF neural network(RRBFNN).Finally,the network is used to predict the effluent BOD in the sewage treatment process,and the development of the effluent BOD soft-sensing intelligent system is completed,realizing the accurate prediction of effluent BOD.The research work of this thesis is mainly divided into the following points:(1)Research on feature selection method based on GSA.In this paper,a feature selection method of effluent BOD based on GSA is proposed.Firstly,the method takes the root mean square error(RMSE)of the neural network as the fitness function of the particle.After that,the particle position information is converted by sigmoid function,and the feature variable selection is converted into a 0-1 combination optimization problem to select the optimal feature subset.So as to avoid the problems of complex calculation and the like caused by too many input variables in the network.Finally,the optimal feature subset selected by this method is used as the input of recurrent RBF neural network to model the effluent BOD.(2)The research on self-organizing recurrent RBF neural network designing based on directed information.This paper proposes a recurrent RBF neural network structure adjustment algorithm(DI-RRBF)based on directed information(DI).The structural adjustment and parameter learning of the network are carried out at the same time.Firstly,in the structural adjustment,the directed information among hidden layer neurons,output layer neurons and feedback values is used to obtain the network directional information intensity,and the neurons with large directed information values are split to meet the requirements of the network for prediction accuracy and complexity.At the same time,two neurons with large mutual information values are combined to simplify the structure of the network so as to determine the structure of the recurrent RBF neural network.At the same time,gradient descent algorithm is adopted in the parameter learning part for training to ensure the accuracy in the structural adjustment process of the recurrent RBF neural network.(3)The research on effluent BOD soft-sensing model based on DI-RRBF neural network.In order to accurately predict the BOD of effluent from sewage treatment,firstly,the input variables are selected by gravitational search algorithm,and the optimal combination of variables selected after feature selection is used as the input of DI-RRBF neural network and effluent BOD is used as the output.After that,a soft-sensing model for BOD of effluent based on DI-RRBF neural network is established.Finally,the proposed soft-sensing model of effluent BOD is applied to actual wastewater treatment data and compared with other methods.The experimental results show that the proposed model can effectively predict the accurate BOD of effluent.(4)The development of a soft-sensing intelligent system for effluent BOD.This paper designs and develops the effluent BOD soft-sensing intelligent system,which mainly includes a user management module,a BOD soft-sensing system home page module,and a display module for important parameter data and prediction curves in sewage treatment.The system connects Visual Studio 2010 software with SQL Server 2008 database to realize user registration and login and display of relevant data in sewage.In BOD prediction module,the mixed programming techniques of C# language and MATLAB is used to realize the training and prediction error curve display by calling MATLAB program of DI-RRBF neural network and realize the visualization of soft measurement system.