Hourly forecasting of the photovoltaic electricity at any latitude using a network of artificial neural networks

Nowadays, special attention is paid to the importance of using photovoltaic (PV) systems to tackle the problem of climate change and the energy crisis. Artificial intelligence is currently used in different science fields for its great potential and accuracy in forecasting problems. In this work, a network of artificial neural networks (ANNs) was trained and validated to forecast the hourly worldwide electrical power produced by various PV modules, with different electrical characteristics. Each ANN describes the worldwide performance of each PV module on the optimal inclination angle. The training data consists of the hourly air temperature, horizontal total solar radiation as input data and electrical power produced as output. The power is obtained from the hourly simulation of PV modules with an electrical circuit model in 24 localities at very different latitudes. The validation and generalization of the network were obtained by considering the six PV modules in further 24 localities and by considering two further PV modules in all 48 localities considered. The excellent results in terms of accuracy metrics confirmed that the network of ANNs is a reliable, simple and accurate tool that can be used to predict the hourly performance of any PV module in any location worldwide.