Data obtained through complex simulations are routinely used to build models and inform process decisions. For many applications, use of the full fidelity system simulation is simply too expensive. A number of techniques can be used to generate reduced-order models of the original system. However, techniques commonly used, such as artificial neural networks and support vector machines, result in surrogate models that are too complex and overfit the data. A simple algebraic surrogate model makes the relationship between inputs and outputs available in closed form, and is much more amenable to the tasks of optimization, control, and prediction.
We describe an approach to complex energy system optimization based on the ALAMO software for building simple yet accurate surrogate models of simulations. ALAMO is a computational methodology that has been developed to learn algebraic surrogate models from exogenous data . In order to model nonlinear behavior present in process data, ALAMO performs a number of nonlinear transformations of input variables to populate a regression basis set. An optimization-based subset selection method then identifies the optimal subset of regression variables according to a model fitness metric such as Mallows’ Cp, Akaike’s information criterion, or Bayesian information criterion. The resulting models consist of a linear combination of nonlinear transformations of input variables, and their simple algebraic form can help provide insight on the system at hand. ALAMO has been used to successfully generate surrogate models for a bubbling fluidized bed carbon capture reactor. These algebraic surrogates were then used to find the optimal configuration of reactors and regenerators in a chemical process superstructure for a CO2 capture system.
 Cozad, A., N. V. Sahinidis, and D. C. Miller, Automatic learning of algebraic models for optimization, AIChE Journal, 60, 2211-2227, 2014.
See more of this Group/Topical: Topical Conference: Advances in Fossil Energy R&D