PV Power Prediction Using Artificial Intelligence Models: A Case Study of Meknes, Morocco
TL;DR
This study demonstrates the effectiveness of gradient boosting models and neural networks for accurately forecasting solar production in Meknes, thus offering promising prospects for the deployment of solar solutions in Morocco.
Abstract
The precise forecasting of photovoltaic (PV) energy production has emerged as a crucial challenge for the optimal management of electrical grids and the stability of energy systems. This study examines the use of AI techniques for short-term forecasting of PV power, employing meteorological data from the NASA POWER satellite database (solar irradiance, air temperature, atmospheric pressure, relative humidity, and wind speed), along with real PV production data gathered in Meknes, Morocco, over a year (February 2015 – March 2016) as part of the national Propre.Ma project. Three predictive models were developed and compared: A Deep Learning model based on an Artificial Neural Network (ANN), a Support Vector Machine (SVM), and a gradient boosting model (LightGBM), evaluated using MAE, MSE, and R 2 . The results indicate that LightGBM performs the best overall (R 2 = 93.1%, MAE = 0.073, MSE = 0.0248), followed by the ANN model (R 2 = 92.5%), while the SVM has the lowest accuracy (R 2 = 81.0%). This study demonstrates the effectiveness of gradient boosting models and neural networks for accurately forecasting solar production in Meknes, thus offering promising prospects for the deployment of solar solutions in Morocco.