A Deep Learning Approach for Climate Parameter Estimations and Renewable Energy Sources

This paper introduces a novel deep learning approach for predicting global solar radiation and temperature. We propose an architecture based on Gated Recurrent Unit (GRU) neural networks able to refine weather predictions returned by the MM5 Regional Climate Model. Measured values from a weather station and outputs from the MM5 system are used to train and validate the model. The forecasting capability is assessed for three-day estimations. The results demonstrate that the model, by correcting MM5’s periodic tendencies to underestimate or overestimate the outputs, leads to a more accurate forecasting of the weather variables.These more precise predictions are then adopted for calculating the electrical energy production of a photovoltaic cell. Also in this case, the proposed model allows better results than the traditional MM5 system, enabling adaptive adjustments in intelligent energy systems.