Simulation a evenements discrets pour la commande temps reel de systemes dynamiques complexes

Several systems involved in different application fields such as production, transportation or services are now considered as discrete event systems with complex dynamic behavior. Our thesis is directed towards this growing research area which seems to attract a lot of interest since the emergence of technological advances. One of the major contributions of our work is the proposed architecture for real-time control of these systems. This architecture is mainly inspired form the model-based predictive control scheme developed by the automation scientist. However, to ensure a reliable prediction model of the complex and dynamic systems, our control scheme is based on a discrete event simulation rather than an analytical model. Even though two decades have now passed since this approach was proposed, very few industrial implantations have been reported. The scarcity of this promising approach is mainly due to the lack of a formal specification of such control architecture. The purpose of our work is to open up new generic architecture avenues for the implementation of an efficient simulation-based real-time control system. The structural design and the behavior of the proposed architecture are built upon the object-oriented modelling methodology.;A prototype application of mine transportation system is chosen to experiment the capability of the control architecture. This choice is mainly motivated by the critics formulated around the unprofitable huge investments made on information technologies acquisition in the mining industry. For example, recent studies demonstrated that accurate online data delivered from positioning systems and sensors is provided for tracking instantaneous state of trucks, but this information is not effectively used in the real-time dispatching systems. Thus, a problem of high unpredictability level caused by truck bunched together in platoons remains elusive in the mining industry.;In order to embed the real time control architecture in the mining environment, we first developed a discrete event simulation model. The development of a high fidelity simulation model dedicated to the control purpose constitutes an important issue in our work. For this purpose, we firstly demonstrate that the classical modelling approach of the considered transport dispatching problems could lead to biased simulation results. Then, based on the object oriented methodology, we integrate a microscopic traffic model in the classical transportation model. We have noticed that the fleet dispatching problem was mostly addressed by the operational research community while this problem of traffic constitutes a civil engineering field of research. Thus, our work comes to bring those two communities by considering the traffic in the network as an inherent part of the internal transportation systems. The same critics concerning the weakness of the dispatching models in considering the actual traffic state in the internal transportation networks are also founded for the transport of containers at transhipment terminals. Thus, in order to ensure the reusability of our approach, we provide the conceptual model based on the object-oriented approach. After the implementation and validation of the high granularity simulation model, we investigated its capability as an observer of the traffic state in the control scheme. Results proved that our simulator could indeed reproduce important traffic behaviors of platoon formation and propagation due to the longitudinal trucks interactions in the internal transportation network. With the addition of a routing decision to the original control law of truck dispatching, it results in an interesting control system for real-time dispatching and routing. Finally, we demonstrate the efficiency of such real-time truck routing in alleviating traffic congestion problems arising in vehicle-based internal transport systems. (Abstract shortened by UMI.);In the architecture, control laws are computed concurrently with the objective of minimizing the deviation between the controlled system achieved performances and the off-line forecasted performances. To solve this minimization problem, we incorporated an intelligent optimization module based on the simulated annealing metaheuristic. Another important contribution of our work is the effective use of a metaheuristic (the simulated annealing) in the optimization module to achieve intelligent control. Previous researches discard the use of metaheuristics at this control stage due to the following limitations: integration complexity and the long simulation based optimization run-time. In our work, we demonstrate how this long run-time could substantially be decreased by developing a specific simulation model designed for the control purpose. Furthermore, we propose to resolve the integration problem by embedding a multiprocessing technique.