Superstructure-based approaches for process synthesis are powerful because they can account for complex interactions between design decisions. However, they usually lead to mixed-integer nonlinear programs that are hard to solve, especially when realistic unit operation models are used. To address this challenge, we develop a strategy where complex unit models are replaced with surrogate models built using data generated via commercial process simulators or experiments. We study aspects such as the systematic design of unit surrogate models, the generation of simulation data, the selection of the surrogate type (e.g., Artificial-Neural-Network (ANN)), and the required model ﬁtting. Also, we present how these surrogate models can be reformulated and incorporated into mathematical programming superstructure formulations.
Henao CA, Maravelias CT. Surrogate-Based Superstructure Optimization Framework. AIChE J., 57(5), 1216-1232, 2011.