Abstract : This thesis is a contribution to the problem of a complex system prognosis. More precisely, it concerns the model-based prognosis approach and the thesis is divided into three main contributions. First of all, a definition of prognosis concept is proposed as a first contribution and is positionned in reference to the diagnosis and predictive diagnosis concepts. For that, a notion of temporal constraint is introduced to give all pertinence to the prediction achieved. It is also shown how prognosis is linked to the finite time reachability notion. The second contribution is dedicated to the use of finite time convergence observer for the prognosis problem. A prognosis methodology is presented for nonlinear multiple time scale systems. Then, a last contribution is introduced through the use of interval observer for the prognosis problem. A pronognosis methodology is proposed for nonlinear uncertain multiple time scale systems. To illustrate the theorical results, simulations are achieved based on a model of an electromechanical oscillator system.