A new Adaptive methodology of Global-to-Direct irradiance based on clustering and kernel machines techniques

Martín Gastón (1) , Íñigo Pagola (1) , Carlos M. Fernández-Peruchena (1) , Lourdes Ramírez (1) , Fermin Mallor (2)

1 Department of Solar Thermal Energy [Sevilla]
2 UPNA - Universidad Pública de Navarra [Espagne] = Public University of Navarra

Nowadays, there is a growing need of data related to beam solar radiation. The market situation of the solar technologies as well as the premium prices in countries like Spain drives to the necessity to characterize the expected DNI in a specific location with short measurement campaign. Due to it is impossible to wait ten or more years for measure the DNI in a specific location, methodologies for deriving DNI representative series are needed. To estimate beam radiation from global radiation, there can be used information derived from satellite images, or data from well known data bases of global solar radiation. In both cases global-to-beam irradiance models are needed. This work presents a novelty methodology to transform hourly global irradiance registers in a concrete emplacement, into beam irradiance data. In front of classical global to direct models where general curves are fitting to simulate the direct irradiance behavior, the present methodology allows us to fit the "better curve" to each available observation. To implement this method, a set of ground measurements of Beam and Global irradiance were available and different mathematical techniques have been used. Concretely, topics of measure theory, cluster methods and kernel machines learning are included in the present methodology. So, this work is presented using public access data which allow any contrast of the methodology results.

Mots clés: Global Irradiance; Beam Irradiance; k nearest neighbor; Machine learning; kernel methods; SVM

Domaines: Ingénierie de l'environnement

Martín Gastón, Íñigo Pagola, Carlos M. Fernández-Peruchena, Lourdes Ramírez, Fermin Mallor. A new Adaptive methodology of Global-to-Direct irradiance based on clustering and kernel machines techniques. 15th SolarPACES Conference, Sep 2010, Berlin, Germany. pp.11693. ⟨hal-00919064⟩