Prediction Of Seawater Desalination Reverse Osmosis Membrane Contamination Based On BP Neural Netword Model

Seawater desalination is an important strategy measure to solve the shortage of fresh water resources and realize the sustainable development of economy and society.Compared with other desalination technologies,reverse osmosis technology has been widely used for its small footprint,high degree of automation,simple operation and management,and high quality of water production.However,the pollution of reverse osmosis membrane will lead to the reduction of membrane flux,the shortening of membrane service life,and the increase of production energy consumption cost,which will affect the efficient and stable operation of seawater desalination system.One of the effective methods to control the pollution of reverse osmosis membrane is to establish a prediction model for the membrane pollution of reverse osmosis system.In this paper,the reverse osmosis system of a seawater desalination plant in Qingdao was taken as the research object,and the components,formation process and influencing factors of reverse osmosis membrane pollution were explored.On this basis,the membrane flux was used as the characteristic index of membrane pollution,the dimension of the correlation variables of membrane flux was reduced,and the membrane flux prediction model of Error Back Propagation（BP）neural network was established.The prediction performance of its network topology structure,network parameter setting and optimization algorithm is evaluated,in order to provide scientific guidance for the membrane pollution prediction of seawater desalination reverse osmosis system.The main conclusions are as follows:（1）The surface of reverse osmosis membrane and pollutants were characterized and analyzed by physical appearance detection,burning loss,SEM-EDS,CA,XRD and FT-IR.The results show that the reverse osmosis membrane is mainly polluted by inorganic pollution composed of inorganic salts such as calcium carbonate and calcium sulfate,and organic pollution composed of a small amount of organic matter such as protein,polysaccharide and humic acid.Combined with the seawater quality analysis results,it can be concluded that the RO membrane pollution is closely related to the inorganic and organic indexes in the influent water quality.（2）Based on the characteristics of reverse osmosis membrane pollution and seawater quality analysis,membrane flux was selected as the characteristic index of membrane pollution,and conductivity,temperature,p H,turbidity,NH₄⁺-N,COD,TDS,TSS,alkalinity,hardness,UV₂₅₄,transmembrane pressure difference and running time were selected as the correlation variables of membrane flux,and the principal component algorithm was used for dimensionality reduction.The results showed that the independent effective information of 13 membrane flux correlation variables was extracted by principal component analysis and projected into 5 comprehensive factors with a cumulative contribution value of 88.268%.The function expressions of 5 principal component factors were constructed,which laid the foundation for the establishment of membrane flux prediction model.（3）The five principal component factors after dimension reduction were used as input variables,and the membrane flux of reverse osmosis system was used as output variable.The reverse osmosis membrane flux prediction model based on BP neural network was established,and the Particle Swarm Optimization（PSO）and Genetic Algorithm（GA）were used to optimize the BP neural network.The prediction performance of the model was evaluated by mean Absolute error（MAE）,mean square error（MSE）and goodness of fit（R²）.The results show that the prediction accuracy and generalization ability of BP neural network are improved after optimized by PSO and GA algorithm.Among them,GA-BP model has the best prediction performance,and its MAE is 0.1084,MSE is 0.0189,R² is 0.9070.The predicted value of membrane flux fits the actual value best,and has high prediction accuracy.（4）The constructed BP,PSO-BP and GA-BP neural network models were applied to the membrane flux prediction of a seawater desalination plant in Qingdao in August,and the prediction results of the three neural network models were compared and analyzed.The results show that GA-BP neural network model can better fit the change trend of membrane flux in August,and its MAE is 0.0613,MSE is 0.0063,R² is 0.8725,and the generalization ability and robustness of the prediction model are strong.The feasibility of the GA-BP neural network model in the membrane flux prediction of the actual seawater desalination reverse osmosis system is further verified.