Optimization under Constraints of Distributed Complex Problems using Cooperative Self-Organization JURY

We solve problems and make decisions all day long: at home, at work, while playing. Some problems and decisions are very challenging: What is the best sequence of actions to reach a goal or the best itinerary to deliver orders given the weather, the traffic and the hour? How to choose the best time for a meeting knowing the availability of concerned people and meeting rooms with adequate material? How to improve product manufacturing performances or to refine a computational model given delays to satisfy, interdependencies between parameter and multi-disciplinary aspects? etc. Problems that are characterized by a high level of complexity due to the heterogeneity and diversity of the participating actors such as humans or electronic devises, to the increasing volume of manipulated data and their distribution and to the dynamics of the applications environments. Classical solving approaches have shown their limits to cope with this growing complexity. Thus, the scientific community has been interested, for the last several years, in the development of new solutions based on computation distribution and control decentralisation, which are more appropriate for solving such problems. The AMAS (Adaptive Multi-Agent-Systems) theory developed by the SMAC team, proposes to build solutions based on self-adaptive multi-agent systems using cooperative self-organisation. In such systems, cooperative interacting agents pursue local goals. By their interactions, the robustness of the system and its capacities to adapt to dynamic environments are increased. Thus, the global function of the system emerges. This theory have shown its adequacy to solve a large variety of complex and dynamic problems, but it remains at a high abstraction level, requiring AMAS experts for its application. This work proposes a specialisation of this theory for complex optimisation problem solving under constraints characterized by multi-disciplinary and multi-objective criteria. This will make the usage of this theory accessible to different non-AMAS experts engineers confronted to such problems. Thus, the AMAS4Opt agent model with cooperative, local and generic behaviours and interactions has been defined. Such behaviours and interactions can be instantiated and extended for solving different complex optimisation problems. Once identified, they have been instantiated and tested on two well-known optimisation problems: scheduling in manufacturing control characterised by high level of dynamics and complex product design characterised by the volume of interrelated data. Finally, in order to show the robustness and adequacy of the developed solutions, a set of evaluation criteria is proposed to underline the advantages and limits of adaptive systems and to compare them with already existing systems. Cooperative Self-Organisation for Optimization under Constraints iii