Reduced-Space Multi-Fidelity Bayesian Optimization of Process Simulation Models
The paper proposes RS-MFBO to cut costly high-fidelity process-simulator runs without giving up much optimization performance.
The method combines global sensitivity analysis for dimensionality reduction with a fidelity-aware Gaussian process model. It uses cost-aware sampling, plus cooldown and promotion mechanisms, to decide when to query cheap approximations versus expensive simulations. The authors test it on a plasmid DNA bioprocess in SuperPro Designer and a green fuel synthesis plant in Aspen HYSYS. The arXiv version notes acceptance at LION 20 and a corrected typo in the multi-fidelity covariance kernel description. ArXiv · AI/CL/LG's note
The method combines global sensitivity analysis for dimensionality reduction with a fidelity-aware Gaussian process model. It uses cost-aware sampling, plus cooldown and promotion mechanisms, to decide when to query cheap approximations versus expensive simulations. The authors test it on a plasmid DNA bioprocess in SuperPro Designer and a green fuel synthesis plant in Aspen HYSYS. The arXiv version notes acceptance at LION 20 and a corrected typo in the multi-fidelity covariance kernel description. ArXiv · AI/CL/LG's note
score 4