Intelligent Identification Of Sludge Bulking Based On A Knowledge-transfer Fuzzy Neural Network

Sludge bulking,with the characteristic of high incidence and wide range,is a very difficult problem occured in the process of activated sludge treatment system.Generally,10~30 days are needed for the wastewater treatment system to recover from sludge bulking.Due to the accurate mathematical model of sludge bulking is difficult to be obtained.In addition,the recognition method of sludge bulking,with low precision and poor stability,is still an unsolved problem.Facing the identification problem of sludge bulking,this paper established a knowledge-transfer learning self-organizing fuzzy neural network(KL-SOFNN)as an intelligent identification method for predicting sludge bulking risk online.The main research work of this paper includes the following points:1.Analysis the characteristic variables of sludge bulking.Through the study of influencial factors of sludge bulking,principal component analysis method is adopted to define the characteristics variables of sludge bulking.Firstly,a data collection platform is set up for the real-time acquisition of process variables data information.Secondly,based on principal component analysis method,the related degree between the variables and sludge bulking is analyzed for extracting the characteristic variables of the sludge bulking.Finally,a multivariate local quadratic polynomial regression method is proposed to describe the characteristic model of sludge bulking.The proposed characteristic model is applied for a real wastewater treatment process.The results demonstrate that the characteristic variables of sludge bulking can be used efficiently to build the model of sludge bulking.2.Design of a self organizing fuzzy neural network based on knowledge transfer learning.Traditional fuzzy neural network has poor performance on handling insufficient data.To address this problem,a knowledge transfer learning based self-organizing fuzzy neural network(KL-SOFNN)is studied and developed.At first,a fuzzy neural network is established with knowledge transfer properties and the ptimization objective function with the knowledge is designed.Moreover,an adaptive second-order algorithm is implicated to adjust the parameters of the KL-SOFNN.The importance index of the hidden layer neurons is applied for the growth and reduce of the fuzzy neural network structure.The structure of KL-SOFNN is adjusted with networks parameter learning.Finally,this method is verified by applying it to a standard testing function.The experimental results shows the effectiveness of this method in learning efficiency and prediction accuracy.3.Intelligent identification of sludge bulking based on self-organizing fuzzy neural network.In order to effectively identify sludge bulking,an intelligent model,based on KL-SOFNN,is developed to predict sludge bulking risk.First of all,the current model and the reference model of sludge bulking risk are established using the selected characteristic variables,respectively.Moreover,KL-SOFNN is adopted to establish the knowledge transfer model of sludge bulking.Finally,this proposed sludge bulking risk model is applied to a real wastewater treatment plant.The experimental results indicate that our proposed method has better generalization performance.4.The design of the sludge bulking intelligent measuring system.In order to apply the intelligent identification of sludge bulking for the practical wastewater treatment plant.An intelligent sludge bulking risk prediction system based on KL-SOFNN is developed.First of all,a hardware system was built to obtain the data of related variables in real-time.Then the data collection module,data transmission part,soft-sensor part and visualization module are encapsulated into the system.Finally,using C# language to develop the user interface part to compute and display the online detecting results of sludge bulking risk,combined with real-time process variable data related to sludge bulking.Therefore,the security of the wastewater treatment is realized.