Use of Econometrics Models to Forecast Short Term Solar Irradiance
Résumé
Between 2006 and 2011, the French government set up an incentive policy in order to develop the electricity production from photovoltaics. In the overseas territories, like Reunion Island, the feedin tariffs proposed for the next 20 years were specifically high. It resulted an exponential increase of the installed PV systems. For these small non-interconnected grids, an important penetration rate of such an intermittent energy can destabilize the supply-demand balance and may lead to an electricity outage risk. As a consequence, a regulatory limit of 30% of the instantaneous power produced from intermittent renewable (solar, wind and waves) was defined in order to avoid this risk. For instance, this legal constraint was reached in 2012 in Reunion. In this context, the forecasting of solar irradiance is essential in order to increase the penetration rate of PV output power into the grid. More precisely, accurate solar forecasts will help the grid operator to better manage the means of production. Forecasts are also needed in order to optimize the operation of grid connected storage energy systems. In order to cope with specific plant operations, forecasts must be provided with different granularities and horizons. In this work, we will focus on forecasts from 10 minutes to to 4 hours ahead with a 10 minutes granularity. The chosen forecasting time horizon will permit to monitor the production and to adjust the scheduling. In this paper, we introduce a novel approach based on econometrics models to forecast the global solar irradiance. The use of econometrics methods is justified by the similar behavior exhibited by the clear sky index and log return time series Our approach combine 2 models: a ARMA to forecast the mean of solar irradiance and and heteroskedastic model to forecast the volatility.
Origine : Fichiers produits par l'(les) auteur(s)