Hybrid data-driven optimisation approach for pressure swing adsorption

Pressure swing adsorption (PSA) is a vital gas separation technology widely utilised in various industrial sectors. Optimal design of PSA can be effectively achieved through mathematical modelling and optimisation methods. However, optimising PSA using detailed (rigorous) models directly presents computational challenges due to their inherent complexity arising from high nonlinearity and stiffness. Accordingly, we propose two hybrid data-driven optimisation approaches that utilise surrogate models to determine the optimal PSA design. While the first one called enhanced hybrid data-driven optimisation framework is an enhancement to our previously developed hybrid optimisation framework, the other named hybrid Bayesian optimisation approach is developed based on Bayesian optimisation. The computational results demonstrate that the second approach yields superior designs to the first one as the cost is reduced by 5.5 % to 18.3 %, but the first approach is more computationally efficient than the second one by 34.8 % to 36.5 %. It is also found that the best design generated by the first approach shows noticeable deviation of 13.9 % from that obtained by the rigorous approach that solves the rigorous model with a genetic algorithm and a sequential quadratic programming algorithm, whilst the best design obtained from the second one is within 4 % of that from the rigorous approach. Both the proposed methods significantly reduce computational effort, with hybrid Bayesian optimisation and the enhanced hybrid optimisation framework requiring only 17.9 % and 11.2 % of the time taken by the rigorous approach, respectively.