An indirect approach to forecast produced power on photovoltaic plants under uneven shading conditions

In this paper a forecasting method is proposed for the prediction of the generated power in photovoltaic systems. The approach exploits the combination of a virtual irradiance sensing methodology and a neural network forecasting system. The strength of this approach resides in its capability to support forecasting in presence of distributed shading patterns along the PV plant, without the necessity of external pyranometers or a complex data acquisition system. The technique was validated experimentally by forecasting the produced power on a PV array mounted on the building roof at the ENEA research center of Casaccia (Rome, Italy), and shows very good results in the forecasting of One Day- Ahead PV generated power. The results obtained validate this approach as a competitive option to pyranometer-based monitoring in PV installations.