Key points are not available for this paper at this time.
An adaptive framework for day-ahead forecasting of available solar energy is proposed based on a combination of data-analytic approaches consisting of artificial intelligence and statistical techniques. Models are developed and validated utilizing a large dataset from the National Renewable Energy Laboratory (NREL) archive, the Automated Surface Observing System, and the solar position and intensity calculator (i.e., NREL-SOLPOS) sampled at 1-min intervals during eight years (2005-2012) for a site in Golden, CO, USA. The methodology is now ready for testing and validation in San Antonio, TX, USA, with data collected in the largest solar photovoltaic plant in TX, Alamo 1, which is the first solar plant in TX connected to the transmission grid allowing solar energy bidding into the market. A uniqueness of the methodology developed is that an integrated serial time-domain analysis coupled with multivariate analysis was used for preprocessing. The resulting enhanced dataset is used for adaptive training of the neural-network-based forecast engine. Standard performance measures are obtained. The forecast results are compared to those of the state of the art on day-ahead solar energy forecasting methodologies used in Austria and other European Union members in order to provide a clear understanding of capabilities of the proposed solar energy forecasting framework.
Manjili et al. (Mon,) studied this question.