A comparative analysis of particle swarm optimization and differential evolution algorithms for parameter estimation in nonlinear dynamic systems

Research output: Contribution to journalArticlepeer-review

19 Scopus citations

Abstract

The use of evolutionary algorithms has been popular in recent years for solving the inverse problem of identifying system parameters given the chaotic response of a dynamical system. The inverse problem is reformulated as a minimization problem and population-based optimizers such as evolutionary algorithms have been shown to be efficient solvers of the minimization problem. However, to the best of our knowledge, there has been no published work that evaluates the efficacy of using the two most popular evolutionary techniques - particle swarm optimization and differential evolution algorithm, on a wide range of parameter estimation problems. In this paper, the two methods along with their variants (for a total of seven algorithms) are applied to fifteen different parameter estimation problems of varying degrees of complexity. Estimation results are analyzed using nonparametric statistical methods to identify if an algorithm is statistically superior to others over the class of problems analyzed. Results based on parameter estimation quality suggest that there are significant differences between the algorithms with the newer, more sophisticated algorithms performing better than their canonical versions. More importantly, significant differences were also found among variants of the particle swarm optimizer and the best performing differential evolution algorithm.

Original languageEnglish (US)
Pages (from-to)65-83
Number of pages19
JournalChaos, Solitons and Fractals
Volume58
DOIs
StatePublished - Jan 2014

All Science Journal Classification (ASJC) codes

  • Statistical and Nonlinear Physics
  • General Mathematics
  • General Physics and Astronomy
  • Applied Mathematics

Fingerprint

Dive into the research topics of 'A comparative analysis of particle swarm optimization and differential evolution algorithms for parameter estimation in nonlinear dynamic systems'. Together they form a unique fingerprint.

Cite this