Efficient hybrid strategy for simultaneous design of refinery hydrogen networks and pressure swing adsorption unit

• Published in: Journal of cleaner production Vol. 466; p. 142858
• Main Authors: Huang, Lingjun, Hong, Xiaodong, Liao, Zuwei, Yang, Yao, Wang, Jingdai, Yang, Yongrong
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 10.08.2024
• Subjects: Bayesian optimization; Constrained black-box optimization; Hybrid strategy; Hydrogen network; Pressure swing adsorption; Constrained black-box optimization; Hydrogen network; Bayesian optimization; Hybrid strategy; Pressure swing adsorption
• ISSN: 0959-6526; 1879-1786
• DOI: 10.1016/j.jclepro.2024.142858

Refineries face increasing hydrogen demand due to the consumption of heavy and sour crude oil. This study presents a novel hybrid method to optimize hydrogen network configurations with a rigorous pressure swing adsorption (PSA) process model. This method utilizes Bayesian optimization to address the upper-level unknown constraints black-box optimization problem while employing deterministic algorithms to manage the lower-level nonlinear programming problem. Furthermore, infeasible regions are transformed into feasible ones by introducing a dummy stream, and the costs incurred during the transformation process are accurately calculated. The proposed method is validated using two literature cases. In Case 1, Bayesian optimization outperformed the Kriging model, resulting in a 7.84% reduction in total annual cost (TAC). In Case 2, a comparison of four search strategies revealed the developed two-phase strategy as the most effective one. This method offers a powerful and practical tool for optimizing real-world hydrogen networks with PSA processes. [Display omitted] •A hybrid method for simultaneous design of hydrogen network and PSA unit is proposed.•A constrained black-box optimization problem is addressed by Bayesian Optimization.•Infeasible solutions are converted into feasible ones, reflecting their deviation.•A two-phase strategy of Bayesian Optimization is developed and approved efficient.•Solutions with smaller cost is obtained, compared to the ones using surrogate models.