Evaluation of different PV power forecasting systems on grid-connected PV plants in the Mediterranean Area.

The increasing spread of renewable generation requires a deep transformation in electrical generation, transmission and distribution. The negative impact of renewables on electric grid, due to their non-programmability, can be mitigated with reliable power forecasting systems. This study compares different probabilistic forecasting systems of photovoltaic production for the next two days with high temporal resolution (15 minutes or 30 minutes). They are tested on three Mediterranean plants. This paper points out how different performances can be achieved according to model configuration, predictors used as input, climatic characteristics and size of plant. A new approach, based on an optimal combination of various models, is also presented. Finally, the study highlights the importance of an accurate characterization of meteorological conditions in order to achieve an improvement in renewable power forecasting.