Multimodel Design of Large Scale Systems with Multiple Decision Makers.

摘要

The central theme of this thesis is multimodeling. It is concerned with modeling and control strategy interaction in a multimodel context. Realistic situations are studied, which allow the decision makers to use different simplified models of the system. Three different approaches to multimodeling are examined. Firstly, within the framework of multiparameter singular perturbations, we demonstrate the well-posedness of an a-priori selected multimodeling scheme, for a class of Nash and team problems. This establishes, in some sense, the robustness of this multimodeling to a class of solution concepts and information patterns. Secondly, for a class of weakly-coupled Markov chains, we use a perturbational approach to develop an efficient algorithm for computing near-optimal incentive policies, which allows for multimodeling on the part of the decision makers. Finally, for a class of linear-quadratic problems, we use an input-output approach to restructure the problem, and choose appropriate admissible strategies which induce multimodel solutions. (Author)