PhD graduated
Team : SMA
Departure date : 05/01/2018

Supervision : Zahia GUESSOUM

Co-supervision : AISSANI MOKHTARI Aïcha

Vers une approche d’ingénierie multi-agents à base de lignes de produits logiciels

Multi-Agent Systems (MAS) represent a technology and an ideal solution that has already proved positive for the modeling of complex systems.
The engineering of MAS or AOSE (Agent Oriented Software Engineering) offers different methodologies, meta-models, templates and reuse patterns that facilitate their development and accelerate their acceptance within the software industry.
However, the existing approaches to MAS engineering do not allow the management and development of similar applications known as MAS families. These applications have some commonalities, as well as differences called variability.
The management of variability can be done at different levels such as design and development, except that it is not taken into account in existing approaches.
In order to compensate for the lack of variability management within multi-agent families at the level of agent-oriented approaches, SPL (Software Product Lines) engineering turns out to be the appropriate solution for which the management of variability remains. a key element. In this context, the exploitation of SPL engineering techniques within the framework of multi- agent approaches is known as MAS-PL (Multi-agent systems Product Lines) engineering. This thesis subject is part of this thematic of MAS-PL approaches meant to enhance the management of variability within families of MAS; what, consequently, improves the aspects of reuse revolving around variability.
This is how our approach, which is based on the general SPL process,in favor of an improvement, pushes the limits of current MAS-PL approaches.
Thence,the originality of our approach is the fact that it allows reuse independent of the application domain. To do this, we have considered ,for instance, not a single but several meta-models resulting from agent-oriented engineering, and this unlike MAS-PL approaches such as PASSI-PL or GAIA-PL which only take into account a single meta-model without these concepts are being specified in terms of variability.
This variability can be specified, for example, by the feature models or FM (feature models) which are the most used in this field, and which we consequently used. The particularity of our approach is that we have divided the analysis (resp. Implementation) of the domain into analysis (resp. Implementation), generic (reusable independently of the domain) and specific. This breakdown appears within the FM as well as at the level of implementation artifacts. This is how the present thesis subject offers a dual educational and scientific interest, both for academia and that of scientific research and in the field of the software industry aiming to make the best use of SPLE techniques for the community of designers and developers of MAS. Based on this approach, our research was illustrated by three case studies (management of the timetable, the booktrading, and the multi-agent Contest).
It is useful to underline that the evaluation of our approach was articulated around the derivation of a multi-agent line for the Contest. This is how this multi-agent line was tested by deriving a subset variants based on the set of valid configurations obtained.
Likewise, the variants were deployed after their derivation, thus showing that each of them is simulated within the Contest environment.
The results of our work have shown that our approach improves the reuse rates of characteristics.

Defence : 12/14/2020

Jury members :

Stéphane GALLAND [Rapporteur]
Adbelhak-Djamel SERIAI [Rapporteur]

2015-2020 Publications