Non-programmable renewable energy sources are continuously increasing at both the global and national levels, driven primarily by the need to reduce greenhouse gas emissions from traditional sources. In Italy, the National Integrated Energy and Climate Plan stipulates that renewable sources should meet approximately 30% of Gross Final Consumption by 2030. This involves increasing installed capacity by about 750 MW per year for wind energy and approximately 3.1 GW per year for solar energy. However, the rise in non-programmable renewable energy sources introduces a series of technical challenges that become increasingly difficult as their penetration into the power system grows, such as the reversal of flows at primary substations, a decrease in rotating equipment, leading to frequency stability issues, and so forth. Managing the grid is thus becoming more complex, requiring consideration of demand and generation at all levels, energy market dynamics, the presence of aggregators of various sizes, and the available flexibility in the territory. Accurate generation and load forecasts are increasingly required, spanning from a few hours (very short term) to several days ahead of real-time power system operation.

The goal for this three-year period is to improve forecasting methods for generation and demand across different spatial and temporal scales. In the current year, the method for determining short-term photovoltaic (PV) forecasts has been updated by optimizing a set of forecasts and also providing information on their reliability. These forecasts are obtained using a direct method that relies on suitable meteorological fields provided by various regional models and global drivers. The method requires a sufficiently extensive set of production measurements. When real-time production measurements are available, very short-term forecasts can also be made. Systems based on autoregressive methods and neural networks have been trained using the best short-term forecast, historical measurements, appropriate intelligent persistence, and multi-model forecasts of certain meteorological variables as regressors, yielding encouraging results. Similar methods have been applied to electrical demand, and a preliminary study on the dependence of national gas demand on degree days has also been conducted.