Clustering weather situations with respect to prediction of solar irradiance by multiple NWP models

With the photovoltaic (PV) and concentrating solar power (CSP) forming a growing portion of European power sources, there is a strong demand for a reliable prediction of solar power. Such predictions are mostly provided by physical or statistical models, both of which rely on accurate forecast of solar irradiance. For the short-to-medium-term forecast horizon (hours to days), irradiance forecast is provided mostly by numerical weather prediction (NWP) models. However, in spite of a recent effort to improve irradiance prediction within current NWP models, its quality is still not satisfactory and it is responsible for a majority of uncertainty in photovoltaic power forecasting. A promising method of improving NWP solar irradiance prediction is multi-model approach. This paper presents preliminary results from a data mining approach to combining irradiance forecasts from multiple NWP models.