Mixed-Integer Evolution Strategy for Chemical Plant Optimization with Simulators

Abstract

The optimization of chemical engineering plants is still a challenging task. Economical evaluations of a process flowsheet using rigorous simulation models are very time consuming. Furthermore, many different types of parameters can be involved into the optimization procedure, resulting in highly restricted mixed-integer nonlinear objective functions.

Evolution Strategies (ES) are a promising robust and flexible optimization technique for such problems. Motivated by a typical chemical process optimization problem, in this paper a non standard ES is presented, which deals with nominal discrete, metric integer and metric continuous parameters taken from limited domains. Genetic operators from literature are combined and adapted.

Experimental results on test functions and an application example — the parameter optimization of a HDA process — show the robust convergence behaviour of the algorithm even for small population sizes.