Improving Hour Ahead Solar Irradiance Forecast Using Ensemble Method in French Guiana
| Autor: | Salloum, M., Diallo, M., Primerose, A., Linguet, L. |
|---|---|
| Jazyk: | angličtina |
| Rok vydání: | 2019 |
| Předmět: | |
| ISSN: | 1466-1470 |
| DOI: | 10.4229/eupvsec20192019-5cv.3.6 |
| Popis: | 36th European Photovoltaic Solar Energy Conference and Exhibition; 1466-1470 The objective of this study is to improve the hour ahead solar irradiance forecasts in the inter tropical zone using ensembles of forecasts from several models in French Guiana. The first model of the ensemble is a physical model: Weather Research and Forecasting model (WRF), the second and third model are statistical: the persistence and Auto Regression. The fourth model is and hybrid physical statistical model. It is obtained by post processing WRF output with a Kalman filter bias correction method. We use a Ridge Regression method to linearly combine all the forecasts with weights varying by models. We validated our method over three months of GHI measurements with hourly data of global horizontal irradiance (GHI) from six stations of the French national weather service. We used the MAE, MBE and RMSE and clear sky index to compute the statistics. For all sky conditions and stations merged, we found that statistical and hybrid statistical models perform better than WRF under low variability sky conditions. For highly variable sky conditions WRF outperform other models. For all sky conditions merged, WRF Kalman has the lowest RMSE and MAE, respectively 29% and 22%. We also found that aggregating the ensemble members using a Ridge regression method decreases RMSE and MAE of the most accurate single model by approximately 5%. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|