Architecture logicielle pour l'informatique diffuse: Modelisation du contexte et adaptation dynamique des services

In a pervasive computing system, a set of smart devices communicate and collaborate together by sensing the global context and reacting proactively (without an explicit intervention of the user) in order to provide adapted services to the user and applications. Devices in pervasive system must be context-aware. Thus context is a key concept in such systems and requires a good understanding and use of it. This term is still general and vague in spite of the large number of proposed definitions. In this thesis, we present a service based definition of context that is abstract enough to be used in pervasive computing and limit the set of information required to make service adaptation. The design and implementation of pervasive computing systems requires new software tools, and architectures specifically to support their development. These tools must support characteristics of such systems, such as mobility and context-awareness. In this thesis we propose a multi-agent service oriented architecture for pervasive computing systems. The architecture takes into account the dynamic aspect of pervasive systems. It is more generic (applicable to a large variety of applications) and modular which enhance its reusability. The architecture was modeled using colored Petri net formalism and simulated with the CPN-Tools based on real Scenarios of pervasive computing. The first task of designing pervasive computing applications consists of understanding context and establishes its components. Context must be well understood and modeled in an appropriate form which enhance its sharing between devices and provide a high level of abstraction. The most interesting methods for modeling context are those based on ontology. However the majority of the proposed methods fail in proposing a generic ontology for context which limit their usability and keep them specific to a particular domain. Among contributions of this thesis is the proposition of a new service ontology for context modeling in a pervasive system. This ontology is enough generic, simple and encloses the main concepts of a pervasive computing system. In addition, it enhances the context sharing between devices. Devices of a pervasive computing system must acquire some intelligence which permits them to sense the current context and trigger the appropriate service or provide a service in a better suitable form. The adaptation task must be context-aware and takes into account the limited resources of devices in a pervasive system which is generally hand-held. In this thesis we propose two approaches of dynamic services adaptation in a pervasive system. The first one is based on machine learning and was validated by a typical application scenario. The second one is a new architecture for service adaptation which aware of both context and limited resources of devices. The architecture was modeled using colored Petri net formalism and simulated with the CPN-Tools based on real scenarios of pervasive computing.;Key words. pervasive computing, context, architecture, ontology, adaptation.