Post Processing of Day-Ahead Solar Irradiance Forecast Using Satellite Derived Data in French Guiana
| Autor: | Macaire, J., Salloum, M., Bechet, J., Zermani, S., Linguet, L. |
|---|---|
| Jazyk: | angličtina |
| Rok vydání: | 2021 |
| Předmět: | |
| DOI: | 10.4229/eupvsec20212021-5bv.4.5 |
| Popis: | 38th European Photovoltaic Solar Energy Conference and Exhibition; 1130-1133 The aim of this work is to improve day-ahead hourly Global horizontal irradiance (GHI) forecast from Weather Research and Forecasting (WRF) model using post-processing techniques trained with satellite estimates of hourly irradiance. Generally, post-processing techniques are trained with ground measures of the irradiance, however, in French Guiana, only 6 ground stations are available. To face this issue, we investigate the use of satellite-derived GHI data, which are available over the whole territory, instead of ground data to fit different post-processing techniques to correct the bias of WRF GHI forecasts. We compared two post-processing techniques, Artificial Neural Network and Support Vector Machine, both trained with ground-based solar irradiance data first, and then with satellite-derived solar irradiance data. We showed that post-processing techniques trained with satellite-derived data improved the accuracy of the model. The relative Root Mean Squared Error (rRMSE) and the relative Mean Bias Error (rMBE) decrease, respectively, from 48.69 % to between 40.85 and 41.81 % and from 19.14 % to between -1.52 and -8.76 %. We also showed that using satellite-derived data degrades only slightly the forecast accuracy compared to the techniques using ground data (around 1 % of difference for the rRMSE). Moreover, Support Vector Machine post- processing model overperformed the Artificial Neural Network model. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|