Renewable energy sources, such as wind and solar, are characterized by non-programmability (FRNP) and considerable randomness. Electricity demand, on the other hand, is subject to fluctuations that depend on domestic and industrial consumption. To cope with these variabilities, it is necessary to have a flexible and resilient net grid. In this context, the interconnection of systems can enable rapid information exchange, which is a key component for the development and implementation of predictive systems related to generation, demand, and real-time quantification of dispatchable flexible resources.

The activities illustrated here were devoted to the analysis, development, and operational implementation of various short (up to 3 days ahead issued twice a day) and very short (up to 6 hours ahead updated every 15’) term predictive techniques. Spatial scales have varied from local, such as distinct wind farms and interconnections / feeder of a Medium Voltage (MV) line, to market areas. Short-term forecasts can be hourly or quarter-hourly, depending on the granularity of the acquired measurements. The very short-term, on the other hand, is characterized always by a temporal resolution of 15 minutes.

The examined techniques were based on machine learning, such as Random Forest and Gradient Boosting Machines, statistical techniques, such as functional Principal Component Analysis, Analog Ensemble, and the One-Edge technique developed by the University of Pavia, autoregressive methods, such as ARIMAX, and hybrid methods combining several methods.

When a set of elementary forecasts were available, it was possible to activate an optimization mechanism, e.g., by a Quantile Random Forest technique, to obtain not only a “deterministic” forecast, but also the corresponding bands of reliability or the expected variability for the days following the target day.

The continuous activation of a very short-term forecast – which must necessarily require limited computational resources to meet delivery time constraints – driven by the short-term one, makes it possible to reduce the latter’s systematic errors and to cope, in real time, with structural changes or weather events that were not adequately predicted.

The document is available on the site in Italian