GSA-SOM: a metaheuristic optimisation algorithm guided by machine learning and application to aerodynamic design

Alejandro Gonzalez-Perez, Christian B Allen, Daniel J Poole

Research output: Chapter in Book/Report/Conference proceedingConference Contribution (Conference Proceeding)

1 Citation (Scopus)
233 Downloads (Pure)

Abstract

Typical algorithms used in optimisation problems can be classified into gradient-based or agent-based optimisers. Gradient-based optimisers boast very fast convergence due to their ability to extract topological information from the local gradient of the objective function, but can often get trapped in a local minimum. Agent-based optimisers use a series of agents to traverse the search space stochastically and are thus much more prone to find the global minimum, but their cost may be unacceptable in typical engineering optimisation problems. The question remains on whether convergence of agent-based algorithms can be accelerated using design space information. A novel optimisation algorithm is introduced that uses machine learning in the form of a Self-Organising Map (SOM) to extract topological information from the optimisation design space and guide the agents of a Gravitational Search Algorithm (GSA). The proposed optimiser is benchmarked against a set of unconstrained analytical optimisation functions, and it is shown to outperform classical agent-based optimisers. The algorithm is further applied to an engineering design problem in the form of constrained transonic aerodynamic shape optimisation, confirming its robustness and higher performance.
Original languageEnglish
Title of host publication2021 AIAA Aviation Forum and Exposition
Subtitle of host publicationSession: Aerodynamic Design I
PublisherAmerican Institute of Aeronautics and Astronautics Inc. (AIAA)
ISBN (Electronic) 9781624106101
DOIs
Publication statusPublished - 28 Jul 2021

Bibliographical note

1st Place Winner of 2021 AIAA Applied Aerodynamics Student Paper Competition

Fingerprint

Dive into the research topics of 'GSA-SOM: a metaheuristic optimisation algorithm guided by machine learning and application to aerodynamic design'. Together they form a unique fingerprint.

Cite this