Deriving multi-agent system behavior

The multi-agent systems (MAS) have become a very powerful paradigm in the development of complex computer systems. The modeling of these systems can reduce this complexity during the development phases. The Model-Driven Architecture (MDA) approach can be used to resolve this problem. It allows designers to model their systems at different abstraction levels, providing them with automatic model transformations to incrementally refine abstract models into concrete ones. This paper presents a MDA approach to develop multi-agent system with the derivation of the behavior of a given system agent from its global requirements. The suggested approach is based on the definition of an appropriate requirements meta-model (Computational Independent Meta-Model CIMM) and the definition of a target design meta-model (Platform Independent Meta-model PIMM). The CIM models are specified using UML activity diagram extended with collaborations to describe the system global behavior. The agent's behavior model (PIM) is in the form of distributed UML state machines. Automatic model transformations between these two models have been designed in order to govern the derivation process. A real application of telediagnosis in neuroscience has been developed using this approach.