Research On The Application Of Integrated Algorithm And Generalized Regression Neural Network In MB

In the 21st century when water resources are scarce,how to recycle water resources has become a hot research direction.As a new type of sewage treatment process,MBR membrane bioreactor has the advantages of less floor space,good effluent quality,and convenient management compared with traditional sewage treatment methods.It is currently widely used.However,membrane fouling will occur during the operation of the MBR system,leading to problems such as a decrease in membrane flux and water production,and an increase in cost,which seriously affects the popularization and use of MBR technology.An effective measure to reduce membrane fouling is that on the basis of predicting the degree of membrane fouling,workers can obtain relevant information earlier and take relevant measures such as cleaning the membrane.Aiming at the problem of membrane fouling,this article refers to related documents to explain the formation mechanism of membrane fouling.It is found that the degree of membrane fouling can be characterized by the size of membrane permeability.Therefore,this article combines the actual treatment data of modern sewage treatment plants to extract the related membrane fouling The key variable is to establish a prediction model of membrane water permeability through integrated learning intelligent algorithm.This paper uses random forest,XGBoost and LightGBM algorithms to establish membrane pollution prediction models.First,the collected data is used to eliminate abnormal data using criteria,and then PCA principal component analysis and feature selection are used to screen out key variables that have a greater impact on water permeability.They are sludge concentration,nitrate NO3-N in aerobic zone,aeration volume,production water pressure,production water flow,influent COD,and influent PH.Then input the data to the model for training.Because the integrated model has many adjustable parameters,and different parameters have a greater impact on the effect of the model,this article uses grid search combined with cross-validation to find the optimal parameters to improve the model’s performance.Forecast effect.Finally,this article uses Mean Absolute Error(MAE),Mean Square Error(MSE),coefficient of determination,and training time to evaluate the above three models.Finally,it is found that the LightGBM model has lower errors,which is better than XGBoost and random forest models.Accurate,and the training speed of the LightGBM algorithm is faster than the other two models.This article also refers to traditional water quality prediction methods and modern intelligent algorithm water quality prediction methods,uses matlab to build a sewage treatment water quality prediction model based on generalized regression neural network(GRNN),and uses the firefly swarm intelligent algorithm to find the best smoothing factor and optimize The original GRNN model compares the effects of the two models.Compared with the GRNN model,the final GSO-GRNN model reduces the average absolute error by 1.06 and the average error percentage(mape)by 9.97%.The model has good predictive performance and can better let the staff understand the MBR system water production.Water quality has practical application value.