Home > Events > 2020 > Improved forecasts of direct normal irradiance

Improved forecasts of direct normal irradiance

To contribute for improved operational strategies of concentrating solar power plants with accurate forecasts of direct normal irradiance, IMDEA Energy has published a work* describing the use of several post-processing methods on numerical weather prediction. Focus is given to a multivariate regression model that uses measured irradiance values from previous hours to improve next-hour predictions, which can be used to refine daily strategies based on day-ahead predictions. Short-term forecasts provided by the Integrated Forecasting System, the global model from the European Centre for Medium-Range Weather Forecasts ECMWF, are used together with ground measurements. As a nowcasting tool, the proposed regression model significantly improves hourly predictions with a skill score of ≈0.84 (i.e. an increase of ≈27.29% towards the original hourly forecasts). Using previous-day measured availability to improve next-day forecasts, the model shows a skill score of ≈0.78 (i.e. an increase of ≈6% towards the original forecasts), being further improved if larger sets of data are used. Through a power plant simulator (i.e. the System Advisor Model), a preliminary economic analysis shows that using improved hourly predictions of electrical energy allows to enhance a power plant’s profit ≈0.44 M€/year, as compared with the original forecasts. Operational strategies are proposed accordingly.

(*) Improved ECMWF forecasts of direct normal irradiance: A tool for better operational strategies in concentrating solar power plants. Francis M. Lopes; Ricardo Conceição; Hugo G. Silva; Rui Salgado; Manuel Collares-Pereira. Renewable Energy, Vol. 163, January 2021, Pages 755-771 https://doi.org/10.1016/j.renene.2020.08.140 

More information: Ricardo Conceição, Postdoctoral researcher, High Temperature Processes Unit, ricardo.conceicao@imdea.org

Hourly clearness indices for Direct Normal Irradiance DNI (kb) between measurements (OBS) and improved DNI predictions with (a) SRM, (b) MRM1 and (c) MRM2 models, where the identity line, y=x, is represented as a red-dashed line. Analysis is performed over one year of data in Évora (southern Portugal), from July 1st 2018 to June 30th 2019. The respective hourly probability density functions (PDF’s) are also depicted (d, e and f).

Event Date: 
Friday, October 2, 2020