Most relevant input parameters selection for 10-min global solar irradiation estimation on arbitrary inclined plane using neural networks

Abstract

ABSTRACT This paper propounds the use of ‘Weights’ method to assess the relative importance of input features for estimating 10-min global solar irradiation on an arbitrary tilted plane, using artificial Neural Networks (ANNs). Two different data sources are exerted: satellite data, including different inclinations, are used for learning, while ground data, measured at Bouzaréah radiometric station (Algeria), are used for testing. ‘Weights’ method allowed to identify the relative importance of input variables and rank them according to their order of contribution, thus significantly improving the performance of the model by eliminating the parameters that are poorly correlated with the network output. The results show that the most relevant input data are in the following order: the global horizontal irradiation 35.9%, extraterrestrial irradiation 25.08%, declination 15.81%, elevation of the sun 12.60% and inclination 10.61%. The optimal configuration includes 16 neurons in the hidden layer with nRMSE = 6.37% and R 2 = 0.992.