Résumé
This thesis addresses questions related to the modeling and simulation of complex systems whichare based on the multi-agents system (MAS) paradigm. Multi-Agent Based Simulations (MABS) relyon the idea that it is possible to directly represent the behavior and the interactions of a set ofautonomous entities situated in a common environment. One critical issue is the difficulty to accuratelyachieve replication of published models. This raises problems for verification and validation of MABS.In this thesis, we advocate that MABS verification and validation problems are mainly related to thegap that exists between models specifications and the computational structures which are used toexecute them. The motivation of our work is to take steps toward an effective mapping between modelsspecifications and concrete software structures considering two complementary approaches. The firstapproach proposes generic simulator engineering tools. The idea is to make explicit the computationalstructures which are used for the implementation. The second approach relies on the study of todayMAS modeling means : we propose a reflexion about the MAS paradigm and we identify a set ofmodeling constraints which are related to it. We then propose modeling principles based on (1) theInfluence/Reaction approach and (2) a formal model of the environment for MASs, namely MIC ∗ .We show how our approach handles the various identified modeling constraints and we illustrate itsfeasibility through a concrete example.