Cost contingency analysis of sky-imager-based bias-correction of cloud-cover fraction forecasts

Cloud cover conditions over a photovoltaic (PV) power plant generally reduce its energy production compared to cloud-free skies. Surface weather stations and solar radiation equipment, and more recently All-sky imagers are providing real-time measurements of cloud cover passing over PV areas. As utility-scale PV farms and residential roof top PV panels continue to grow in response to lowering prices and government incentives the use of low-cost sensors to monitor energy resources at the local level is needed. In this study, a sky imagery is used to validate and to correct both the timeseries and the frequency bias of fine resolution short range predictions of cloud cover. The high frequency image data of the fraction of cloud cover is matched with that of the NOAA-High Resolution Rapid Refresh (HRRR) to characterize the forecast errors and uncertainties on a reference surface domain. Bias correction techniques to improve forecast performance are carried out both as a continuous forecast (e.g., RMSE) and as categorical forecasts (e.g., octas of cloud cover octas). The study then quantifies cost reductions associated with increasing forecast accuracy through simulation of prescribed load scenarios and historical electricity prices. Under these scenarios, a formulation of economic value is analyzed for optimal economical decisions.