Extended Surrogate Assisted Continuous Process Optimisation

Abstract

Chemical plant design and optimisation have proven to be challenging due to the uncertainty in real-world systems. Research to date has mostly assumed that the system under evaluation behaves deterministically when performing optimisation. This assumption limits the applicability of the results to the real world. Stochastic representations yield more characteristic results for these types of systems. This paper expands on recent research into the optimisation of chemical plant design by further testing the algorithm in using more complex variants of the original problem. The results illustrate the benefits of using Genetic Algorithms as an optimisation framework for complex stochastic chemical plant systems. We further find that combining a Genetic Algorithm framework with Machine Learning Surrogate models as a substitute for a long-running stochastic simulation yields significant computational efficiency improvements (1.74-1.95 times the speedup). This improvement is despite the increased complexity introduced in the problems that were optimised. It is worth noting the algorithm confirms its robustness towards more complex stochastic chemical plant systems and its flexibility in being able to be applied to different chemical plant systems.