Soft Measurement Modeling Of Sludge Volume Index Based On Self-organizing Recurrent Fuzzy Neural Network

Sludge bulking has can be classified as a complex system due to the complicatedmechanism and many influencing factors. It is now the widespread phenomenon inWaste-Water Treatment Process (WWTP) basd on activated sludge process. Awastewater treatment plant its nonlinear dynamics, large uncertainty, multiple timescales in the internal process dynamics and multivariable structure. The traditionalmathematical modeling has been challenged, it is difficult to achieve on-lineprediction of sludge volume index. The soft measurement modeling must be set up.The neural network is widely used in the wastewater treatment modeling because ittheoretically has the ability of approximate any nonlinear function. Recursive neuralnetwork has capability to deal with dynamic information. The soft sensor model basedon recursive neural network can well simulate the dynamic changes of WWTP, andimplement the online estimation of water quality parameters.The study mainly around the following several parts carry out:1.Based on in-depth study of the important characteristics and the influencingfactors of Sludge bulking, to difficult problem to achieve on-line prediction of sludgevolume index. Analysis the main influencing factors of the SVI from two aspects ofwater quality and reactor environment. Based on monod equation, combined withASM1and microbial growth mechanism and the mechanism of aeration tank toestablish the mathematical model of activated sludge process. They provide importanttheoretical basis for the intelligent prediction model proposed and improved. In viewof soft measurement technique, a set of relatively systematic soft sensor modelingmethod of sewage water quality parameters is proposed. Soft measurement modelingsteps include: secondary variables selection, data acquisition and preprocessing ofsample data of principal component analysis. Finally complete the SVI softmeasurement model of structure design.2. Based on study of the structure of the fuzzy neural network, has improved thestructure of the recursive fuzzy neural network. By introducing self-feedbackconnection in the rule layer, to obtain a new type of improved recurrent fuzzy neuralnetwork HRFNN. This flexible network structure more in line with the actualcharacteristics of complex systems, making the feedback neural network has moreresilience and ability to perform, can be effectively applied to time-varying dynamic system modeling and forecasting, and a simpler architecture network can be achievedsimultaneously.3. A new structure self-organizing algorithm of recursive fuzzy neural networksis proposed for the traditional recursive fuzzy neural network in the design of thestructure of the existing problems. The firing strength contributes to the matchingdegree of a rule node in layer3. The criterion for deciding if a new fuzzyset should begenerated is proposed. Through examples of nonlinear system identification, Todemonstrate that the new algorithm is superior in terms of compact structure andlearning efficiency.4. A forecast model of SVI is established with the proposed algorithmSOHRFNN. Simulation results show that the model is effective with highperformance compared with HRFNN.