ANN-Based Model for Daily Solar Radiation Prediction with A Low Number of Hidden Neurons And Optimal Inputs

Abstract

Solar radiation prediction has been the focus of many studies over the past years due to its usefulness for clean energy generation through photovoltaic (PV) systems. The prediction result is important for both standalone and grid-connected PV systems as it is used for the design of these systems, for making power dispatching plans in hybrid systems, as well as for potential future PV system feasibility. In this article, an Artificial Neural Network (ANN)-based model for daily global solar radiation prediction is developed. This model is trained with a back-propagation training algorithm and make prediction using meteorological variables as inputs. While keeping a good accuracy level, the model is built using a low number of neurons in the hidden layer of the ANN. Therefore, the proposed model is simpler as compared to many existing models. First, the minimum and the maximum numbers of hidden neurons are calculated (page 4). Second, simulations based on a trial-and-error method can show us the good number of hidden neurons. A simple model is preferred than a complex one when the performance is same or almost. In addition, a simple model is easy to understand and to implement and also allows quicker modifications when needed.