DATA-DRIVEN POST-PROCESSING OF ENSEMBLE FORECASTS FOR IMPROVED ACCURACY IN SOLAR FORECASTING

Abstract

High penetration of PV in the power grid can create significant stability-related challenges due to the intermittent nature of PV power. Solar forecasting (solar irradiance or PV power forecasting) is one of the most promising mitigation strategies to address these challenges, but generating high-accuracy forecasts is challenging, and thus post-processing of forecasts becomes crucial. Hence, this work proposes novel data-driven post-processing techniques to improve the quality of solar irradiance and PV power forecasts. The proposed post-processing techniques improved the quality of raw solar forecasts and help to mitigate the impacts of the following two major problems in solar forecasting: (1) under-dispersion problem of data-driven forecasting models in ensemble generation, and (2) aggregate-consistency problem observed in regional solar forecasting applications. The findings and the results presented in this work may help forecast practitioners and system operators in improving the quality of data-driven solar forecasts and ultimately maintaining reliable grid operations.