Abstract
Solar generation will have a significant impact in the electricity supply mix in the upcoming years. Thereby the short-term solar radiation forecast will play a strategic rule in supporting an even more connected and intermittent electrical grid. This paper extends to tropical regions the applicability of the clustering, classification and regression technique, recently proposed in the literature and that presented good results in Europe. Tropical regions usually present lower accuracy in the current numerical models predictions, due to a more sparse instrumentation. We evaluate the predictability of the phases of this approach in two scenarios, by utilizing data from the previous day of the prediction one. Furthermore, we analyzed how the replacement of these variables for the prediction of those equivalent in the prediction day impacts in the forecast skill of the model. We verified how the accuracy increases as the predictions of these new entries improves.
Original language | Portuguese |
---|---|
Publication date | 2017 |
Number of pages | 12 |
Publication status | Published - 2017 |
Event | XLIX Simpósio Brasileiro de Pesquisa Operacional, Sociedade Brasileira de Pesquisa Operacional - Blumenau-SC, Brazil Duration: 27 Aug 2017 → 30 Aug 2017 Conference number: 49 http://www.sbpo2017.iltc.br/index.html |
Conference
Conference | XLIX Simpósio Brasileiro de Pesquisa Operacional, Sociedade Brasileira de Pesquisa Operacional |
---|---|
Number | 49 |
Country/Territory | Brazil |
City | Blumenau-SC |
Period | 27/08/2017 → 30/08/2017 |
Internet address |
Keywords
- Solar radiation prediction
- Support vector machine
- Cluster analysis