A neural network approach for generating solar irradiation artificial series

Abstract

In this paper a relevant problem in the photovoltaic solar energy field is considered: the generation of artificial series of hourly solar irradiation. The proposed methodology artificially generates series following the average tendency of the hourly radiation series k _t in a given place. This is obtained by making use of a set of historical values of this series in such place (for training purposes) as well as the daily clarity index K _T of the year to be generated. This information is employed for the supervised training of a proposed neural network model. The neural model employs a well known paradigm, called Multilayer Perceptron (MLP), in a feedback architecture. The generation method is based on the MLP ability to extract, from a sufficiently general training set, the existing relationships between variables whose interdependence is unknown a priori. This way, the presented design methodology can implicitly include all the available information. Simulation results show the good performance of the irradiation series generator, and the general applicability of this methodology in the estimation of highly complex temporal series.