Application Research Of PSO-GA-Elman And CFD In MBR Simulation

In the sewage,there are a lot of solid suspended solids and dissolved substances.The cake layer is deposited on the surface of the membrane under the action of permeate flow and shear stress.It accumulates continuously during the process resulting in a decrease in membrane permeability.This step is called membrane fouling.Eliminating membrane fouling by scouring,increasing pressure and replacing membranes,these methods can increase the cost of invest.Therefore,exploring the influenced factors of membrane fouling and the function of organic substances about membranes in wasted-water have become an important part of research on membrane bio-reactors.Membrane flux is an important parameter to measure the degree of membrane fouling.In this thesis,the degree of membrane fouling is judged by establishing PSO-GA-Elman and PSO-RNN membrane flux prediction models.For the recurrent neural network,it achieves a ring-shaped structure,which leads to become sensitive to historical data.And the relevance of each data is also improved.Elman as a typical recurrent neural network has the advantages of fast convergence,high precision of prediction results and high sensitivity to historical data.But Elman network uses BP algorithm for data correction,so it still has the disadvantage of falling into a local minimum.Particle swarm optimization can quickly derive global optimal values,and genetic algorithm has global convergence.Combine these algorithms to improve the convergence speed of the genetic algorithm and reduce the probability that the particle swarm algorithm falling into the local optimal value.Changing the initial weights of the Elman neural network with the results obtained from PSO-GA,it can avoid falling into local minimum to a certain extent and improve prediction accuracy to some extent.Comparison of prediction results of membrane flux prediction models,the PSO-GA-Elman membrane flux prediction model is 14%more accurate than the Elman model and 20%higher than the BP network model.Based on the concept of Elman neural network,an RNN network is designed with two hidden layers.The prediction results show that the accuracy of the PSO-RNN prediction model is higher than that of the unmodified RNN model.In addition,to explore the internal conditions of the reactor by simulation software,we can get an ability to study the hydraulic characteristics of the membrane components inside the reactor.This thesis uses simulation software(Ansys Fluent)to model the actual MBR reactor.Establishing a membrane bioreactor model based on hollow fiber,the model is based on 20 hollow fibers arranged in a vertical pattern.Calculate the simulation model according to the calculation steps such as modeling,meshing,and determining boundary conditions to study the relationship between water inlet velocity and model height.The simulation results show that the simulation model can reach the maximum water production when the ratio of the inlet water velocity to the model height is 1:20.