Deep Learning for the Analysis of Solar Radiation Prediction with Different Time Horizons and Data Acquisition Frequencies

This study aims to develop and compare different AI systems for predicting solar radiation and evaluate their performance across different prediction horizons. Predicting solar radiation is of crucial importance for harnessing renewable energy sources. The models were designed to predict radiation over 6-h and 15-min horizons with the lowest possible error. The impact of prediction horizon and data acquisition frequency on prediction accuracy is discussed, emphasizing the need to consider the number of parameters and training time when comparing models. To improve the accuracy of short-term solar radiation predictions, five deep learning models, including classical, convolutional, and recurrent neural networks, were analyzed. The accuracy of the predictions was compared using two error metrics: root mean square error and mean absolute error.