Multi-Objective Calibration For Agent-Based Models
Multi-Objective Calibration For Agent-Based Models
Agent-based modelling is already proving to be an immensely useful tool for scientific and industrial modelling applications. Whilst the building of such models will always be something between an art and a science, once a detailed model has been built, the process of parameter calibration should be performed as precisely as possible. This task is often made difficult by the proliferation of model parameters with non-linear interactions. In addition to this, these models generate a large number of outputs, and their ‘accuracy’ can be measured by many different, often conflicting, criteria. In this paper we demonstrate the use of multi-objective optimisation tools to calibrate just such an agent-based model. We use an agent-based model of a financial market as an exemplar and calibrate the model using a multi-objective genetic algorithm. The technique is automated and requires no explicit weighting of criteria prior to calibration. The final choice of parameter set can be made after calibration with the additional input of the domain expert.
Agent based model, calibration, pareto-optimality, genetic algorithm
Rogers, Alex
f9130bc6-da32-474e-9fab-6c6cb8077fdc
von Tessin, Peter
e0a34d94-3fa9-453a-879a-5c5c3b8d2e5e
2004
Rogers, Alex
f9130bc6-da32-474e-9fab-6c6cb8077fdc
von Tessin, Peter
e0a34d94-3fa9-453a-879a-5c5c3b8d2e5e
Rogers, Alex and von Tessin, Peter
(2004)
Multi-Objective Calibration For Agent-Based Models.
Agent-Based Simulation 5, Lisbon, Portugal.
Record type:
Conference or Workshop Item
(Paper)
Abstract
Agent-based modelling is already proving to be an immensely useful tool for scientific and industrial modelling applications. Whilst the building of such models will always be something between an art and a science, once a detailed model has been built, the process of parameter calibration should be performed as precisely as possible. This task is often made difficult by the proliferation of model parameters with non-linear interactions. In addition to this, these models generate a large number of outputs, and their ‘accuracy’ can be measured by many different, often conflicting, criteria. In this paper we demonstrate the use of multi-objective optimisation tools to calibrate just such an agent-based model. We use an agent-based model of a financial market as an exemplar and calibrate the model using a multi-objective genetic algorithm. The technique is automated and requires no explicit weighting of criteria prior to calibration. The final choice of parameter set can be made after calibration with the additional input of the domain expert.
More information
Published date: 2004
Additional Information:
Event Dates: May 2004
Venue - Dates:
Agent-Based Simulation 5, Lisbon, Portugal, 2004-05-01
Keywords:
Agent based model, calibration, pareto-optimality, genetic algorithm
Organisations:
Agents, Interactions & Complexity
Identifiers
Local EPrints ID: 261547
URI: http://eprints.soton.ac.uk/id/eprint/261547
PURE UUID: 95e2c10f-0a94-4096-b98b-8cea5299f3d6
Catalogue record
Date deposited: 07 Feb 2006
Last modified: 14 Mar 2024 06:55
Export record
Contributors
Author:
Alex Rogers
Author:
Peter von Tessin
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