[pt] Diferentes procedimentos têm sido propostos para a modelagem e previsão de séries temporais sendo que nos anos recentes muitos dos métodos mais importantes têm sido formulados na representação espaço de estado. A principal vantagem de tal abordagem é que se pode usar o Filtro de Kalman diretamente para, seqüencialmente, atualizar o vetor de estado. Apresentamos de forma sistemática a abordagem para a previsão de Séries Temporais não- Estacionárias formulada na representação de espaço de estado desenvolvida por P.Young. A novidade desta abordagem não está na natureza dos algoritmos recursivos, e sim na maneira como os hiperparâmetros são obtidos. Modelling and forecasting of Time Series have been approached in many different ways. Lately, the most important approaches have been formulated in a state space framework. The state space representation enables the state vector to be sequentially updated in time via the Kalman filter. In this dissertation, we present in a systematic way an approach to modelling and forecasting of non-stationary time series, formulated in state space terms, and due to P. Young. The novelty of this methodology is neither the nature fo the time series models nor the recursive algorithms, but on how the hyperparameters are estimated / [en] Modelling and forecasting of times Series have been approached in many different ways. Lately, the most important approaches have been formulated in a space framework. The state space representation enables the state vector to be sequencially updated in time via the Kalman filter. In this dissertation, we present in a systematic way an approach to modelling and forecasting of non-stationary time series, formulated in state space terms, and due to P. Young. The novelty of this methodology is neither the nature of the time series models nor the recursive algorithms, but on how the hyperparameteres are estimated