The Study Of Filamentous Sludge Bulking Control Strategies Based On Self-organizing Fuzzy Neural Network

As an abnormal phenomenon of wastewater treatment plant(WWTP), filamentous sludge bulking possess the several characteristics obviously, like complex process mechanism, numerous causing factors, frequently occurred and being out of control easily when it has appeared. Almost every WWTPs which have adopted the method of activated sludge process will encounter this problem, and it obstacles the operation of WWTP. Had been given a lots of investments on to this problem, there are also many challenges which have not been covered, such as incomplete analysis of mechanism, without any accurate mathematical model, deficiencies in traditional methods about symptom identification of the filamentous sludge bulking. Besides, whether a measure can be appropriate for preventing the risk of filamentous sludge bulking and improving it or not, are other questions relate to it. In regard of these hard scientific edges, this thesis will further analyze mechanism of sludge bulking, design the model of symptom identification to predict the symptom index based on the self-organizing fuzzy neural network(SOFNN), build a knowledge inference model to solve the problem strategies decision for improving filamentous sludge bulking, and obtain effective method from it in practical. This thesis mainly carries out the following several parts:1. By insight into the mechanism of filamentous sludge bulking, eight types factors are summarized which lead to the phenomenon of filamentous sludge bulking. The relationships between them and symptom index, such as sludge volume index(SVI), which are obtained by analysis of the mechanisms about biochemical reaction, filamentous bacteria growth and sludge settlement. Afterwards, a simplified mechanisms model is proposed, whose parameters can be corrected through the practical data from WWTP. Lastly, the dynamic trends of SVI will be expressed following by all the factors.2. In order to improve accuracy of identification of SVI, and enhance the ability of adaptive to the dynamic about input-output space, this thesis design a type of SOFNN which will realize the structure to be optimal by the growing and pruning neurons in radial basis function(RBF) layer under a right time, and that can be decided by a series of dynamic threshold. In addition, an algorithm called local gradient will be used for updating the parameters of SOFNN to improve the accuracy and reliable of it. Specifically, the optimal weight of SOFNN can be obtained by algorithm of least mean square(LMS), meanwhile, Levenberg-Marquardt(LM) is used for training the parameters of centers and widths, which will improve the learning accuracy globally.3. During analysis of mechanism of filamentous sludge bulking, content involved in knowing that the physical characteristics of it are nonlinear and complexity is also important, and by considering the ability of knowledge acquisition and adaptive learning of SOFNN, this thesis will design the model of SVI identification based on SOFNN, and the model will give the data selection and processing method. Based on the principal component analysis the relationships between factors and SVI, auxiliary variables are obtained from factors, which is set as the model input, and SVI is regarded as the model output. By using the practical data from WWTP, the parameters of model will be calibrated. Then, this model is used for forecasting SVI.4. The methods of improving filamentous sludge bulking are diversity included artificial change process variables, combined with other artificial experience for inhibiting filamentous bacteria overgrowth and improving sludge settlement performance. In order to achieve more effectively and reasonable methods, this thesis directed by different factors and the severity of the phenomenon, will summarize the measures involved in the data information and the artificial experience, and use fuzzy rules to describe it correspondingly. Meanwhile, knowledge inference model will be designed for strategies decision. Then, verified by practical cases, the model can provide control strategies effectively.