Multi-objective collaborative optimization of sequential simulated moving bed separation process: a hybrid approach integrating numerical computation and machine learning.

Sequential simulated moving bed (SSMB), although it possesses the advantages of high separation efficiency and relatively low energy-solvent consumption, still encounters optimization challenges due to its numerous independent variables, multiple and conflicting separation objectives, and mass transfer limitation. To address this issue, this study proposes a hybrid approach combining numerical computation and machine learning based on adsorption and kinetic behavior to establish a multi-objective collaborative optimization framework for the SSMB separation process. Initially, SSMB experiments and the corresponding process simulation were conducted and the accuracy of parameter measurement and modeling was verified. Furthermore, numerical calculation approach was constructed through adsorption isotherm and transport-dispersive model, within which a non-dominated sorting genetic algorithm was applied to screen and optimize operating conditions under different separation goals. Then, a machine learning method was developed based on experimental data by using the neural networks and Bayesian algorithms. Finally, the SSMB optimization speed was increased by 125%, and the four-objective optimization was successfully achieved. The purity and recovery of final product exceeded 99% and 97%, respectively.