EMuLSion, a generic simulation framework in animal epidemiology based on multi-level multi-agent modelling - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2017

EMuLSion, a generic simulation framework in animal epidemiology based on multi-level multi-agent modelling

Résumé

To recommend better control measures in public or animal health, epidemiologists incorporate ever-finer details in their models, from individual diversity to public policies, which often involve several observation scales. Due to the variety of modelling paradigms, it becomes more and more difficult to compare hypotheses and outcomes, or explore new assumptions or scenarios, especially since the increased complexity of simulation programs is not yet counterbalanced by design principles nor by software engineering methods. Thus, any change in the model often induces a substantial coding effort. The EMuLSion framework, developed within the MIHMES project, rely upon the multi-level agent-based paradigm to integrate existing methods within a common interface, provide a separation between concerns and reduce the part of code devoted to model designers. Multi-agent systems are composed of interacting autonomous entities (agents) endowed with behaviors, and are quite flexible since an agent can be used to model any kind of entity. Especially, in multi-level agent-based systems [1], agents can represent several organization or observation levels, from individuals to metapopulations. Thus, multi-level agent-based systems are a convenient way to encompass classical modelling paradigms used in epidemiology (e.g. compartments, IBM, metapopulations). We also advocate for a strong separation of concerns, i.e. a clear separation between a generic simulation engine (code) on the one hand, and an explicit, human-readable and highly modular description of the model structure, parameters and assumptions on the other hand. Thanks to this approach, the EMuLSion framework is able to process epidemiological models with very little specific code development, allowing for quick model refactoring, alternative hypotheses assessment, and objective model comparison. Models can involve several modelling paradigms at the same time, and the framework also provides efficient individual grouping methods to benefit from the detail level of IBM without excessive performance degradation. We illustrate our approach with an application to the Q fever disease in cattle, first to reproduce a within-herd model developed previously [2], study possible simplifications of assumptions, and extend it to the between-herd level.
Fichier non déposé

Dates et versions

hal-01602690 , version 1 (02-10-2017)

Identifiants

  • HAL Id : hal-01602690 , version 1
  • PRODINRA : 404810

Citer

Sébastien Picault, Yu-Lin Huang, Vianney Sicard, François Beaudeau, Pauline Ezanno. EMuLSion, a generic simulation framework in animal epidemiology based on multi-level multi-agent modelling. Modelling in Animal Health conference (ModAH), Jun 2017, Nantes, France. 69 p. ⟨hal-01602690⟩
174 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More