Modèles autorégressifs à changements de régimes markoviens. Applications aux séries tempo-relles de vent

In this thesis, several original Markov switching autoregressive model are proposed for wind time series. The first chapter is devoted to a theoretical study of these models. We focus mainly on the problems of the numerical calculation of the maximum likelihood estimators, of the asymp-totic behavior of these estimators and finally of model selection and validation. In the second chapter, we propose various Markov switching autoregressive model to describe the evolution of the wind in a fixed point, and then in the third chapter its space-time evolution. For each suggested model, we check the physical interpretability of the various parameters, and their capacity to simulate realistic artificial sequences. The obtained results are compared to those corresponding to the models which are usually used in the literature.