Transformer Neural Network Architecture for Forecasting of Colombian Solar Irradiance

Renewable resources for electrical energy generation are each time more demanded. Solar irradiation is widely used on these days to compute the possible energy generation. However, the current climate change makes the measuring of the availability of this supply a challenge. For this, forecasting models can be employed to determine what so convenient could be the projection of the generation. This paper shows an approach based on comparison of two neural networks architecture for forecasting of solar irradiance, which can be a resource for photovoltaic generation. Long short-term memory and transformer models were analyzed for determine what network holds better performance in this specific case. Information from three days and a transformer neural network with eight heads presented the best result for the forecasting.