Improving the optimisation performance of an ensemble of radial basis functions
Improving the optimisation performance of an ensemble of radial basis functions
In this paper we investigate surrogate-based optimisation performance using two different ensemble approaches, and a novel update strategy based on the local Pearson correlation coefficient. The ?first ensemble, is based on a selective approach, where ns RBFs are constructed and the most accurate RBF is selected for prediction at each iteration, while the others are ignored. The second
ensemble uses a combined approach, which takes advantage of ns different RBFs, in the hope of reducing errors in the prediction through a weighted combination of the RBFs used. The update strategy uses the local Pearson correlation coefficient as a constraint to ignore domain areas where
there is disagreement between the surrogates. In total the performance of six different approaches are investigated, using ?five analytical test functions with 2 to 50 dimensions, and one engineering problem related to the frequency response of a satellite boom with 2 to 40 dimensions.
Stramacchia, Michele
a82506fd-6885-4567-a510-17e3fbb46ef2
Toal, David
dc67543d-69d2-4f27-a469-42195fa31a68
Keane, Andy
26d7fa33-5415-4910-89d8-fb3620413def
Stramacchia, Michele
a82506fd-6885-4567-a510-17e3fbb46ef2
Toal, David
dc67543d-69d2-4f27-a469-42195fa31a68
Keane, Andy
26d7fa33-5415-4910-89d8-fb3620413def
Stramacchia, Michele, Toal, David and Keane, Andy
(2016)
Improving the optimisation performance of an ensemble of radial basis functions.
EngOpt 2016 - 5th International Conference on Engineering Optimization, Iguassu Falls, Brazil.
19 - 23 Jun 2016.
(In Press)
Record type:
Conference or Workshop Item
(Paper)
Abstract
In this paper we investigate surrogate-based optimisation performance using two different ensemble approaches, and a novel update strategy based on the local Pearson correlation coefficient. The ?first ensemble, is based on a selective approach, where ns RBFs are constructed and the most accurate RBF is selected for prediction at each iteration, while the others are ignored. The second
ensemble uses a combined approach, which takes advantage of ns different RBFs, in the hope of reducing errors in the prediction through a weighted combination of the RBFs used. The update strategy uses the local Pearson correlation coefficient as a constraint to ignore domain areas where
there is disagreement between the surrogates. In total the performance of six different approaches are investigated, using ?five analytical test functions with 2 to 50 dimensions, and one engineering problem related to the frequency response of a satellite boom with 2 to 40 dimensions.
Text
Conference_Paper_EngOpt2016_Brazil.pdf
- Accepted Manuscript
More information
Accepted/In Press date: 19 June 2016
Venue - Dates:
EngOpt 2016 - 5th International Conference on Engineering Optimization, Iguassu Falls, Brazil, 2016-06-19 - 2016-06-23
Organisations:
Computational Engineering & Design Group
Identifiers
Local EPrints ID: 398609
URI: http://eprints.soton.ac.uk/id/eprint/398609
PURE UUID: 82a4f562-b075-4b7e-8116-a4bcfdeea887
Catalogue record
Date deposited: 27 Jul 2016 15:56
Last modified: 15 Mar 2024 03:29
Export record
Contributors
Author:
Michele Stramacchia
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics
