Towards Intra-Hour Solar Forecasting using Two Sky Imagers at a Large Solar Power Plant

Abstract

A pair of sky imagers has been deployed at a 48MW photovoltaic (PV) facility in Henderson, Nevada. The sky images will be analyzed using image processing techniques developed for short term forecasting of plant power output. Cloud maps generated from the two instruments will be combined to generate a single large cloud map over the site to provide coverage of the 1.3 km 2 plant. Future cloud positions are determined by looking at the motion of the clouds over consecutive image pairs and a forecasted position is generated. The ground projected cloud position, a binary map showing cloud shadows with a granular resolution of 6.25 m 2 , is forecast for 0 to 15 minutes. The forecasts are evaluated using AC inverter power production measurements, 15 plane-of-array reference modules measuring global irradiance (GI) and five weather stations measuring GI and global horizontal irradiance. The binary map of the shadows, showing either clear or cloudy conditions, is assigned an optical depth for each condition as determined from the recent time history of observations. A probability density function of observed optical depth for each condition is constructed from the last 10 minutes of measurement and the most frequent occurrence is selected as that condition's optical depth. The optical depths are then used to compute power output for the each power measurement, as well the aggregated plant output. Power output is evaluated for a 0 to 15 minute forecast horizon.