Research On MBR Membrane Fouling Prediction Based On XGBoost Hybrid Model

An important human challenge is water pollution since the 21 st century.In recent years,limited water resources have been continuously polluted,causing deterioration of water quality and pollution of water sources in China.With the rapid development of economy in our country,the amount of sewage discharged has increased day by day,and its composition has been complicated year by year.Membrane bioreactor(MBR)process is widely used for its advantages of high quality effluent.However,the MBR process has caused problems such as an increase in transmembrane pressure difference and a decrease in membrane permeability due to membrane fouling,which has affected the continuous and stable operation of MBR.Recent studies on MBR membrane fouling have shown that parameters related to membrane blockage have become a more critical factor which continued affecting MBR's stable operation.First of all,the MBR process and the factors influencing the permeability of MBR membrane were studied and analyzed in this paper.Combined with the powerful learning ability of machine learning algorithm,the membrane permeability was taken as the parameter to measure the degree of membrane pollution,and based on the MBR process history data,the prediction models of the traditional gradient enhancement algorithm GBDT and the limit gradient enhancement algorithm XGBoost membrane permeability were established respectively.Comparing the prediction accuracy of the two models,the results show that the prediction accuracy and efficiency of XGBoost model are better than GBDT model,which confirmed the feasibility of the XGBoost model for MBR membrane fouling research.Secondly,because there are many types of XGBoost parameters,the range of values is large,and the adjustment is cumbersome,so the parameters have a greater impact on the prediction performance of the algorithm.And the traditional global search algorithms are less efficient in terms of parameter optimization.Therefore,this paper uses genetic algorithms to optimize the parameters in the XGBoost model and builds a GA-XGBoost parameter optimization model.Then the model is tested by training and compared with the original model before optimization.The experimental results show that the optimized hybrid model has a significant improvement in prediction accuracy and meets the expected goals.Finally,according to the genetic algorithm,which involves a large number of individual calculations,the operation time is relatively high when the problem is more complex,the algorithm's ability to explore the problem solution space is limited,and it is easy to converge prematurely.Therefore,this article combines genetic ideas with the bat algorithm.After the bat algorithm performs individual perturbations,it performs individual crossover and mutation operations on random individuals to increase the evolutionary ability of bat individuals,thereby increasing population diversity,and maintaining the optimization ability of bat populations.This paper proposes to apply the genetic bat algorithm to the parameter optimization problem of the XGBoost model and obtain experimental results under the same conditions.Compared with the three models of GBDT,XGBoost and GA-XGBoost,the prediction accuracy has been improved by 57.42%,19.11%,4.13%.It shows that the hybrid genetic bat algorithm has greatly improved the parameter optimization ability of the XGBoost model,realized more accurate prediction of MBR membrane pollution data,and provided a certain theoretical basis for the next research in the field of MBR membrane pollution.