Standard

Application of evolutionary algorithms to optimise one- and two-dimensional gradient chromatographic separations. / Huygens, Bram; Efthymiadis, Kyriakos; Nowé, Ann; Desmet, Gert.

In: Journal of Chromatography. A, Vol. 1628, 461435, 27.09.2020.

Research output: Contribution to journalArticle

Harvard

APA

Vancouver

Author

BibTeX

@article{981f091ebd6d49dba77ffbdc82cb0a83,
title = "Application of evolutionary algorithms to optimise one- and two-dimensional gradient chromatographic separations",
abstract = "We report on the performance of three classes of evolutionary algorithms (genetic algorithms (GA), evolution strategies (ES) and covariance matrix adaptation evolution strategy (CMA-ES)) as a means to enhance searches in the method development spaces of 1D- and 2D-chromatography. After optimisation of the design parameters of the different algorithms, they were benchmarked against the performance of a plain grid search. It was found that all three classes significantly outperform the plain grid search, especially in terms of the number of search runs needed to achieve a given separation quality. As soon as more than 100 search runs are needed, the ES algorithm clearly outperforms the GA and CMA-ES algorithms, with the latter performing very well for short searches (<50 search runs) but being susceptible to convergence to local optima for longer searches. It was also found that the performance of the ES and GA algorithms, as well as the grid search, follow a hyperbolic law in the large search run number limit, such that the convergence rate parameter of this hyperbolic function can be used to quantify the difference in required number of search runs for these algorithms. In agreement with one's physical expectations, it was also found that the general advantage of the GA and ES algorithms over the grid search, as well as their mutual performance differences, grow with increasing difficulty of the separation problem.",
author = "Bram Huygens and Kyriakos Efthymiadis and Ann Now{\'e} and Gert Desmet",
note = "Copyright {\circledC} 2020. Published by Elsevier B.V.",
year = "2020",
month = "9",
day = "27",
doi = "10.1016/j.chroma.2020.461435",
language = "English",
volume = "1628",
journal = "Journal of Chromatography. A",
issn = "0021-9673",
publisher = "Elsevier",

}

RIS

TY - JOUR

T1 - Application of evolutionary algorithms to optimise one- and two-dimensional gradient chromatographic separations

AU - Huygens, Bram

AU - Efthymiadis, Kyriakos

AU - Nowé, Ann

AU - Desmet, Gert

N1 - Copyright © 2020. Published by Elsevier B.V.

PY - 2020/9/27

Y1 - 2020/9/27

N2 - We report on the performance of three classes of evolutionary algorithms (genetic algorithms (GA), evolution strategies (ES) and covariance matrix adaptation evolution strategy (CMA-ES)) as a means to enhance searches in the method development spaces of 1D- and 2D-chromatography. After optimisation of the design parameters of the different algorithms, they were benchmarked against the performance of a plain grid search. It was found that all three classes significantly outperform the plain grid search, especially in terms of the number of search runs needed to achieve a given separation quality. As soon as more than 100 search runs are needed, the ES algorithm clearly outperforms the GA and CMA-ES algorithms, with the latter performing very well for short searches (<50 search runs) but being susceptible to convergence to local optima for longer searches. It was also found that the performance of the ES and GA algorithms, as well as the grid search, follow a hyperbolic law in the large search run number limit, such that the convergence rate parameter of this hyperbolic function can be used to quantify the difference in required number of search runs for these algorithms. In agreement with one's physical expectations, it was also found that the general advantage of the GA and ES algorithms over the grid search, as well as their mutual performance differences, grow with increasing difficulty of the separation problem.

AB - We report on the performance of three classes of evolutionary algorithms (genetic algorithms (GA), evolution strategies (ES) and covariance matrix adaptation evolution strategy (CMA-ES)) as a means to enhance searches in the method development spaces of 1D- and 2D-chromatography. After optimisation of the design parameters of the different algorithms, they were benchmarked against the performance of a plain grid search. It was found that all three classes significantly outperform the plain grid search, especially in terms of the number of search runs needed to achieve a given separation quality. As soon as more than 100 search runs are needed, the ES algorithm clearly outperforms the GA and CMA-ES algorithms, with the latter performing very well for short searches (<50 search runs) but being susceptible to convergence to local optima for longer searches. It was also found that the performance of the ES and GA algorithms, as well as the grid search, follow a hyperbolic law in the large search run number limit, such that the convergence rate parameter of this hyperbolic function can be used to quantify the difference in required number of search runs for these algorithms. In agreement with one's physical expectations, it was also found that the general advantage of the GA and ES algorithms over the grid search, as well as their mutual performance differences, grow with increasing difficulty of the separation problem.

U2 - 10.1016/j.chroma.2020.461435

DO - 10.1016/j.chroma.2020.461435

M3 - Article

C2 - 32822975

VL - 1628

JO - Journal of Chromatography. A

JF - Journal of Chromatography. A

SN - 0021-9673

M1 - 461435

ER -

ID: 53743883