A Surrogate-Assisted Approach for the Optimal Synthesis of Refinery Hydrogen Networks

This paper presents a surrogate-assisted methodology for the optimal synthesis of refinery hydrogen networks, which fully considers the impact of vapor–liquid equilibrium in the flash separation unit, to study the interactions between the varying K-values and the optimal configuration of the overall hydrogen network. In addition, the formations of light hydrocarbons and hydrogen sulfide in the hydroprocessing reactors are calculated by the proposed method. To deliver a more accurate and efficient modeling and optimization, high dimensional model representation (HDMR) is applied to approximate the first-principle-based models of the hydroprocessing reactor and flash separation unit. The established surrogate models are integrated with other realistic system constraints, thus formulating a nonlinear programming model for the optimization of refinery hydrogen networks. The approach is illustrated by its application to a case study, in which the results demonstrate that it yields significant differences in the component concentrations of the reactor inlet streams compared to the previous study. The differences further lead to different configurations of the optimal hydrogen network. Therefore, one can conclude that the proposed method can deliver more accurate and realistic results.