Forecasting Daily Global Solar Radiation On A Horizontal Surface Using Artificial Intelligence Methods

Résumé: Since the creation of the earth, the sun has been an inexhaustible source of energy. For this reason, this research work is focused on exploiting this star to maximize its use by predicting solar radiation. In this work, the main objective was to improve the accuracy of solar radiation forecasts for the city of Batna, Algeria by exploiting the advanced capabilities of artificial intelligence techniques. These data span a decade (1996-2005) and are sourced from the Helio-Clim1 database. In this work, several artificial intelligence models are presented, divided into two categories: classical model, such as regression, and intelligent models, which are in turn divided into two subcategories, including individual models (Fuzzy Logic, MLP, RNN, CNN, LSTM, DT, M5, GBDT, XGBoost, CatBoost, RF, SVR and KNN) as well as hybrid models (The average, The geometric average, The harmonic average and The weighted average). The results of our analyses indicate that the multilayer neural network (MLP) model, integrating multiple meteorological parameters (temperature, humidity, pressure, wind speed, wind direction and rain) with the following numerical results: MSE=9.014, MAE=2.180, RMSE=3.002 and R2=0.918 are the best performing approach for predicting global solar radiation in Batna.

Publié dans la revue: