Machine learning–based optimization for hydrogen purification performance of layered bed pressure swing adsorption

Abstract

Summary An adsorption, heat and mass transfer model for the five‐component gas from coal gas (H2/CO2/CH4/CO/N2 = 38/50/1/1/10 vol%) in a layered bed packed with activated carbon and zeolite was established by Aspen Adsorption software. Compared with published experimental results, the hydrogen purification performance by pressure swing adsorption (PSA) in a layered bed was numerically studied. The results show that there is a contradiction between the hydrogen purity and recovery, so the multi‐objective optimization algorithms are needed to optimize the PSA process. Machine learning methods can be used for data analysis and prediction; the polynomial regression (PNR) and artificial neural network (ANN) were used to predict the purification performance of two‐bed six‐step process. Finally, two ANN models combined with sequence quadratic program (SQP) algorithm were used to achieve multi‐objective optimization of hydrogen purification performance. According to the analysis of the optimization results, the ANN models are more suitable for optimizing the purification performance of hydrogen than the PNR model.

Hydrogen Purification Performance Optimization of Vacuum Pressure Swing Adsorption on Different Activated Carbons

“…Studies have shown that ANN models can predict hydrogen purification performance well [36]. The ANN method uses simple mapping to approximate and implement a certain function of complex mapping.…”

Harnessing the power of machine learning for carbon capture, utilisation, and storage (CCUS) – a state-of-the-art review

“…The PSA optimisation using neural networks was ten times faster as compared to using high-fidelity simulations for functional evaluations. Instead of constructing a surrogate model for each performance indicator, Xiao et al 131 used a multi-output feed-forward neural network architecture to predict purity, recovery and productivity in the PSA optimisations. Vo et al 132 formulated an integrated process model based on the combination of different feed-forward neural networks, which represent the input-output mapping structure of cryogenic, membrane and PSA units for hydrogen recovery and CO 2 capture from the tail gas of SMR-based hydrogen plants.…”

Physics-based neural networks for simulation and synthesis of cyclic adsorption processes

“…As a result, the relative error of Pareto solutions in both objectives was less than 1% and accelerated the optimization routine by ten times. Xiao et al 15 instead used a multi-output feed-forward ANN architecture to predict process performances in the PSA optimizations. Pai et al 16 extended the use of feed-forward ANN models to predict the axial profiles of the intensive variables for a four-step VSA process at CSS, and the models were experimentally validated.…”