Artificial neural network based optimization of a six-step two-bed pressure swing adsorption system for hydrogen purification

Abstract The pressure swing adsorption (PSA) system is widely applied to separate and purify hydrogen from gaseous mixtures. The extended Langmuir equation fitted from the extended Langmuir-Freundlich isotherm has been used to predict the adsorption isothermal of hydrogen and methane on the zeolite 5A adsorbent bed. A six-step two-bed PSA model for hydrogen purification is developed and validated by comparing its simulation results with other works. The effects of the adsorption pressure, the P/F ratio, the adsorption step time and the pressure equalization time on the performance of the hydrogen purification system are studied. A four-step two-bed PSA model is taken into consideration, and the six-step PSA system shows higher about 13% hydrogen recovery than the four-step PSA system. The performance of the vacuum pressure swing adsorption (VPSA) system is compared with that of the PSA system, the VPSA system shows higher hydrogen purity than the PSA system. Based on the validated PSA model, a dataset has been produced to train the artificial neural network (ANN) model. The effects of the number of neurons in the hidden layer and the number of samples used for training ANN model on the predicted performance of ANN model are investigated. Then, the well-trained ANN model with 6 neurons in the hidden layer is applied to predict the performance of the PSA system for hydrogen purification. Multi-objective optimization of hydrogen purification system is performed based on the trained ANN model. The artificial neural network can be considered as a very effective method for predicting and optimizing the performance of the PSA system for hydrogen purification.