A model to make probabilistic forecasts of PV production from a plan-oriented parameterization of solar irradiance
Résumé
Today, solar power (PV) capacity is undergoing a fast growth. The development of network management systems facilitating its penetration in the electricity distribution network may rely on individual forecasts for each solar plant connected to the grid. Recent research work has undertaken the development of dedicated short-term (from a few hours to a few days ahead) PV forecasting models based on Numerical Weather Predictions (NWP) as input, basically solar irradiation forecasts. Those models generally provide single point forecasts with no uncertainty estimation. In this work, we propose a short-term probabilistic PV forecasting model. It relies on clear-sky irradiance estimates derived for a plant's orientation. Then, we propose to model the power's distribution with a multi-components mixture, differencing clear-sky from cloudy conditions, and whose shape is parameterized by the clear-sky irradiance level. This parameterization must allow to capture the changes in the power's distribution shape both with the hour of daytime and atmospheric conditions. Forecasting the power's distribution then relies on forecasting production's regimes associated to different sky-clarity conditions, from a day to another. To that purpose, we consider different data-driven regime-switching approaches. We show preliminary results obtained with production data from a 200 kWc plant, and surface solar irradiation forecasts from the ECMWF (European Center for Medium-Range Weather Forecasts).