Surrogate‐based superstructure optimization framework

Abstract

in Wiley Online Library (wileyonlinelibrary.com).In principle, optimization-based ''superstructure'' methods for process synthesis can be more powerful than sequential-conceptual methods as they account for all complex interactions between design decisions. However, these methods have not been widely adopted because they 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 superstructure-based strategy where complex unit models are replaced with surrogate models built from data generated via commercial process simulators. In developing this strategy, we study aspects such as the systematic design of process unit surrogate models, the generation of simulation data, the selection of the surrogate's structure, and the required model fitting. We also present how these models can be reformulated and incorporated into mathematical programming superstructure formulations. Finally, we discuss the application of the proposed strategy to a number of applications.

Global optimization advances in Mixed-Integer Nonlinear Programming, MINLP, and Constrained Derivative-Free Optimization, CDFO

“…In addition, there exist several methodologies using surrogate models to approximate a part of an original MINLP model which is treated as a black-box, and develop an iterative approach collecting additional samples in order to improve convergence towards a global solution (Davis and Ierapetritou, 2007Henao and Maravelias, 2011). Finally, Garcia-Palomares et al (2006) propose combining global search with a final local search; a recent development combines global search and local direct-search optimization in an box-constrained DFO algorithm (GLODS) which aims to identify multiple local solutions without using random multi-start methods (Custódio and Madeira, 2015).…”

Large scale optimization of a sour water stripping plant using surrogate models

“…To deal with uncertainty in black-box systems, D. Huang et al (2006) used Kriging models on complete processes. Henao and Maravelias (2011) employed disaggregated models for each unit in a flowsheet using artificial neural networks. Quirante et al (2015) used Kriging interpolation for the rigorous design of distillation columns and distillation sequences, explicitly including integer variables in the surrogate model.…”

Review of Mixed‐Integer Nonlinear and Generalized Disjunctive Programming Methods

“…There are two general frameworks that are normally used to tackle these problems: decomposition techniques [97][98][99][100], and surrogate models [101][102][103][104]. There are many processes in which flowsheet optimization has been applied over the last 30 years.…”