Data-driven strategies for extractive distillation unit optimization

Kaiwen Ma, Nikolaos V. Sahinidis, Rahul Bindlish, Scott J. Bury, Reza Haghpanah, Sreekanth Rajagopalan

Abstract: Highlights•Black-box solvers have advantages over surrogate modeling for low-dimensional optimization.•Modeling with generalized linear models requires fewer data measurements than neural network.•Smooth penalty functions outperform non-smooth ones for black-box optimization.