English   español  
Por favor, use este identificador para citar o enlazar a este item: http://hdl.handle.net/10261/2796
logo share SHARE logo core CORE   Add this article to your Mendeley library MendeleyBASE

Visualizar otros formatos: MARC | Dublin Core | RDF | ORE | MODS | METS | DIDL
Exportar a otros formatos:

Wind Data Mining by Kohonen Neural Networks

AutorFayos, José ; Fayos, Carolina
Fecha de publicación14-feb-2007
EditorPublic Library of Science
CitaciónPLoS ONE. 2007; 2(2): e210.
ResumenTime series of Circulation Weather Type (CWT), including daily averaged wind direction and vorticity, are self-classified by similarity using Kohonen Neural Networks (KNN). It is shown that KNN is able to map by similarity all 7300 five-day CWT sequences during the period of 1975–94, in London, United Kingdom. It gives, as a first result, the most probable wind sequences preceding each one of the 27 CWT Lamb classes in that period. Inversely, as a second result, the observed diffuse correlation between both five-day CWT sequences and the CWT of the 6th day, in the long 20-year period, can be generalized to predict the last from the previous CWT sequence in a different test period, like 1995, as both time series are similar. Although the average prediction error is comparable to that obtained by forecasting standard methods, the KNN approach gives complementary results, as they depend only on an objective classification of observed CWT data, without any model assumption. The 27 CWT of the Lamb Catalogue were coded with binary three-dimensional vectors, pointing to faces, edges and vertex of a "wind-cube", so that similar CWT vectors were close.
DescripciónConceived and designed the experiments: JF CF. Performed the experiments: JF CF. Analyzed the data: JF CF. Contributed reagents/materials/analysis tools: JF CF. Wrote the paper: JF CF.
Aparece en las colecciones: (IQFR) Artículos
Ficheros en este ítem:
Fichero Descripción Tamaño Formato  
fayos.pdf446,37 kBAdobe PDFVista previa
Mostrar el registro completo

Artículos relacionados:

NOTA: Los ítems de Digital.CSIC están protegidos por copyright, con todos los derechos reservados, a menos que se indique lo contrario.