Prediction and analysis of separation performance of on-board hollow fiber membrane

Due to the long experimental period and complex operating conditions, it is necessary to study the precise prediction of membrane performance using limited data in airborne hollow fiber membrane separation performance experiments. An artificial neural network-based prediction model for hollow fiber membrane separation performance has been developed. Error back propagation (BP) neural networks, Elman neural networks, and genetic algorithm optimized BP (GA-BP) neural networks have been constructed to investigate the variation of airborne hollow fiber membrane separation performance with bleed pressure, flight altitude, and bleed flow rate. Compare and predict the performance of hollow fiber membranes under different algorithms. The research results indicate that the three network algorithms constructed can accurately predict the separation characteristics of hollow fiber membranes, with an mean absolute percentage error (MAPE) of less than 1%. With an average prediction accuracy 1.43% greater than the other two, GA-BP neural networks typically produce the best prediction results when selecting 5%, 10%, and 20% of the complete dataset. Based on the research results, suitable algorithms can be selected in practical work to process membrane experimental data and efficiently analyze performance.