Deep Learning Approach for Solar Irradiance Forecasting: A Moroccan Case Study

Due to its influence on applications such as renewable energy generation, solar irradiance data, and meteorological parameters have risen in prominence. However, developing an accurate model for predicting solar irradiance based on multiple weather parameters remains a challenging issue. As a novelty, a multi-horizon forecasting scheme ranging from 1 to 3 days ahead is studied in the present work. A SeqtoSeq model architecture to forecast global horizontal irradiance based on Masen’s dataset. Univariate and multivariate SeqtoSeq models are implemented to forecast global horizontal irradiance (GHI) based on Masen’s dataset. Moreover, the performance of the proposed models was compared against other models such as Multi-Layer Perceptron (MLP), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM). The obtained results reveal that the proposed multivariate model outperforms all the other models and improves prediction in terms of Mean Absolute Error (MAE) by 42.49, 48.29, and 47.78% for 1 day ahead, 2 days ahead, and 3 days ahead, respectively.