Using sequential approximate optimization and a genetic algorithm to calibrate agent-based models

Roberto Borquez, Enrique Canessa, Carlos Barra, Sergio Chaigneau

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

We present a Genetic Algorithm (GA) tool that uses Sequential Approximate Optimization (SAO) to calibrate Agent-Based Models (ABMs). The SAO/GA searches through a user-defined set of input parameters to an ABM, delivering values for those parameters so that the output time series of an ABM match the real system's time series to certain precision. SAO/GA calculates a meta-model of the real and ABM's time series and optimizes that model. This allows SAO/GA to stabilize the ABMï¿1/2s time series and assure a higher probability of convergence, even under highly variable ABM's outputs. The results show that SAO/GA exhibits a higher convergence probability, but requires a rather long computational time to reach the stopping condition, although that long time is not so excessive to preclude SAO/GA practical use.

Original languageEnglish
Title of host publicationProceedings - 2015 34th International Conference of the Chilean Computer Science Society, SCCC 2015
PublisherIEEE Computer Society
ISBN (Electronic)9781467398176
DOIs
StatePublished - 23 Feb 2016
Externally publishedYes
Event34th International Conference of the Chilean Computer Science Society, SCCC 2015 - Santiago, Chile
Duration: 9 Nov 201513 Nov 2015

Publication series

NameProceedings - International Conference of the Chilean Computer Science Society, SCCC
Volume2016-February
ISSN (Print)1522-4902

Conference

Conference34th International Conference of the Chilean Computer Science Society, SCCC 2015
Country/TerritoryChile
CitySantiago
Period9/11/1513/11/15

Keywords

  • Agent-based modelling
  • calibration
  • genetic algorithms
  • sequential approximate optimization

Fingerprint

Dive into the research topics of 'Using sequential approximate optimization and a genetic algorithm to calibrate agent-based models'. Together they form a unique fingerprint.

Cite this