Evaluation of the level of prediction errors of wind and PV in a future with a large amount of renewables

Abstract : In a future with a large amount of renewable energies integrated into the distribution system there are many kind of new constraint that cannot be considered apriori as un-active in the power distribution system operation and planning. This includes short term (few hours in advance) forecast errors of the production. However their importance is subject to the so called aggregation effect and cannot be considered as a linear function of the installed capacity. Furthermore this aggregation effect highly depends on the considered geographical perimeter. In this paper we presents a methodology to estimate the level of short term forecast error at any spatial scale and we apply the methodology in the case of France, and regions of France. Before simulating forecast error, we describe the methodology used to simulate hourly wind power and solar power production at any location. Wind power is simulated through the use of a statistically calibrated power curve a spatio-temporal field of win simulation obtained from the use of a meteorological model in reanalysis mode. Solar power simulation are based on the use of SODA estimation of surface solar irradiation at a hourly resolution and at the spatial scale of 3km. SODA data rely on the use of satellite images and physical modeling of radiative transfer in the atmosphere. Both models to simulate production are validated on real data. Our methodology for simulating forecast error mimic real forecasting performed on the simulated production data. The real forecasting model is simplified in order to allow a large quantity of computation repetitions. In the case of PV forecasting as well as in the case of Wind Power forecasting, it uses as input two explanatory variables: meteorological forecast of ECMWF (of wind speed for wind power and of surface solar irradiation fot PC) and closed past production measurement. The simulation is performed along several years and thousands of different implantation configurations. The distribution of errors is analyzed as a function of the spatial scale and the chosen horizon.