Trust Regions in Surrogate-Assisted Local Search for Industrial Columns’ Mass Transfer Efficiencies Estimation

Tray efficiencies are important structural parameters for distillations' modeling and optimization. However, determining tray efficiencies in industrial practice remains a challenge. A new method of estimating tray efficiencies by using industrial operating data is proposed in this study. The proposed approach called trust regions in surrogate-assisted local search (TRSALS) is developed on the basis of industrial operating data. In TRSALS, the tray efficiencies are estimated by solving a constrained optimization problem, which is based on the combination of Kriging surrogate based global search and RBF surrogate-based trust-region local search. The estimated key tray temperature profile is compared with that of the actual one of the industrial column to verify the feasibility of the proposed method. Experimental results show that the performance of TRSALS was outstanding compared with that of a Kriging surrogate-only-based algorithm.