Single and blended models for day-ahead photovoltaic power forecasting

  1. Antonanzas, J. 1
  2. Urraca, R. 1
  3. Pernía-Espinoza, A. 1
  4. Aldama, A. 1
  5. Fernández-Jiménez, L.A. 1
  6. Martínez-De-pisón, F.J. 1
  1. 1 Universidad de La Rioja
    info

    Universidad de La Rioja

    Logroño, España

    ROR https://ror.org/0553yr311

Journal:
Lecture Notes in Computer Science

ISSN: 0302-9743

Year of publication: 2017

Volume: 10334 LNCS

Pages: 427-434

Type: Article

beta Ver similares en nube de resultados
DOI: 10.1007/978-3-319-59650-1_36 SCOPUS: 2-s2.0-85021762284 GOOGLE SCHOLAR

More publications in: Lecture Notes in Computer Science

Institutional repository: lockOpen access Editor

Abstract

Solar power forecasts are gaining continuous importance as the penetration of solar energy into the grid rises. The natural variability of the solar resource, joined to the difficulties of cloud movement modeling, endow solar power forecasts with a certain level of uncertainty. Important efforts have been carried out in the field to reduce as much as possible the errors. Various approaches have been followed, being the predominant nowadays the use of statistical techniques to model production. In this study, we have performed a comparison study between two extensively used statistical techniques, support vector regression (SVR) machines and random forests, and two other techniques that have been scarcely applied to solar forecasting, deep neural networks and extreme gradient boosting machines. Best results were obtained with the SVR technique, showing a nRMSE of 22.49%. To complete the assessment, a weighted blended model consisting on an average weighted combination of individual predictions was created. This blended model outperformed all the models studied, with a nRMSE of 22.24%. © Springer International Publishing AG 2017.