Inference of a continuous auto-regressive model for the forecasting of nonstationary stochastic processes deriving from energy demand in electrical networks

This paper discusses the application of Hilbertian auto regressive models to medium term forecasting of electric energy demand, that is, one week-ahead prediction. These models are aimed at predicting whole future trajectories of continuous stochastic processes and can be useful in order to forecast not only the aggregate figures of energy demand (e.g., mean levels and load peaks), but also the time evolution of electrical quantities. Consideration on the optimum number of weeks which should be employed in order to achieve the minimum-variance prediction as well as on the robustness of the model to particular data and outliers, due, for example, to sudden weather changes, are reported