Evaluation of four global optimisation techniques (ASSA, DE, NA, Tabu Search) as applied to anechoic coating design and inverse problem uncertainty estimation

Authors:

  • Gothäll Hanna
  • Westin Rune

Publish date: 2005-01-01

Report number: FOI-R--1593--SE

Pages: 85

Written in: English

Abstract

During the last ten to fifteen years, simulated annealing and genetic algorithms have become routine tools in the field of underwater acoustics for solving difficult optimisation and inverse problems. However, other global optimisation methods have recently been introduced, some of which have been reported to outperform the previous ones. In the present report, four such more modern global optimisation techniques are tested and compared: Adaptive Simplex Simulated Annealing (ASSA), Differential Evolution (DE), Neighbourhood Algorithm (NA), and Enhanced Continuous Tabu Search (ECTS). The techniques have been tested on synthetic optimisation problems and applied to the design of Alberich anechoic coatings. ASSA and DE performed best of the four algorithms in the synthetic test problems, as well as in the coating design problem. For the other two algorithms, NA and ECTS, further research is desired in order to improve their exploring capabilities. In the context of inverse problems, a solution appraisal stage is important, and an evaluation of a recently developed method for that purpose is reported. A Bayesian inversion problem was formulated concerning the Alberich anechoic coatings, and the optimisation algorithms were applied to obtain least-squares solutions. An extension of NA was then used to get an indication of the variable estimate uncertainties, by resampling the obtained search ensembles. This extension, called NA-Bayes, was found to be useful tool for the solution appraisal stage, provided that the search domain is well sampled by the optimisation technique used.