Skip to Main content Skip to Navigation
Theses

Des agents-modèles pour la modélisation et la simulation de systèmes complexes : application à l'écosystémique des pêches

Abstract : We study the modelling process for the simulation of complex systems. The modelling and simulation processes are complex, therefore the modelling has to be rationalized and control gained over it. Our proposition is to conceive the complex model as a society of models in interaction. We apply the pattern oriented approach to rationalize the modelling by focusing it on the data.
We use the agent approach to distribute the observation and control in the global model. The society of models thus enables the emergence of the patterns through the agents' activities. The \artefacts, applied to the modelling, support the data, the patterns and the agents' interactions. The coupling of the agents is therefore done through the \artefacts, which is different from an approach based on events. We identify three types of agents:
1° the model-agents that execute the model of a phenomenon;
2° the observator-agents that build up observables,
3° the controller-agents that locally control the global model.
The agents are also able to adapt to their environment using operations to modify their perceptions and actions. The complex model can be therefore incrementally built, the agents' behaviors can easily be modified, and an agent can be added or removed for the global model without impact on the structure of the other agents. At last, we identify four types of behaviors for model-agents: 1° based on differential equations (intensional); 2° individual based (emergent); 3° based on raw data or pre-calculated (extensional); 4° played by an expert (participatory). To go even further in the rationalization of the modelling, we exhibit computational biases induced by the translation of conceptuel models of agent populations into computational models. We show that our models are very sensitive to the studied biases. It is necessary for thematicians to know the sources of those biases and the simulation tool must be specialized in order to: 1° explicit the model, 2° garantee the identified properties and 3° make systematic analysis. This is what we have done in our simulator. Finally, this work takes place in the ANR project Chaloupe. The challenge here is to envision the co-viability of ecological populations and of their exploitations by social groups (e.g. fisheries). We show our simulation results, confirming by this mean that the society of models enables a flexible and rigorous modelling.
Document type :
Theses
Complete list of metadatas

Cited literature [99 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00376168
Contributor : Stéphane Bonneaud <>
Submitted on : Thursday, April 16, 2009 - 6:22:41 PM
Last modification on : Monday, October 19, 2020 - 10:55:40 AM
Long-term archiving on: : Thursday, June 10, 2010 - 8:47:15 PM

Identifiers

  • HAL Id : tel-00376168, version 1

Collections

Citation

Stéphane Bonneaud. Des agents-modèles pour la modélisation et la simulation de systèmes complexes : application à l'écosystémique des pêches. Autre [cs.OH]. Université de Bretagne occidentale - Brest, 2008. Français. ⟨tel-00376168⟩

Share

Metrics

Record views

350

Files downloads

1499