Assessment of Machine Learning Algorithms for Predicting Potential Solar and Wind Energy Locations

Reaching sustainability and tackling climate change need switching to renewable energy. Still, the erratic character of these energy sources poses problems for energy management. Precise forecasting can facilitate the more efficient integration of these sources, increasing system efficiency and energy production. Predictions of solar irradiance, wind speed, and load have all shown promise for accuracy improvement using machine learning (ML) methods. This study investigates the use of ML techniques, specifically Random Forest (RF), Support Vector Machine (SVM), and Multi-Layer Perceptron (MLP), for predicting potential solar and wind energy sites in Morocco. The performance of these ML algorithms was assessed using metrics including sensitivity (Se), specificity (SP), Positive Predictive Value (Po), and kappa coefficient (k ∗). Results showed that the RF algorithm demonstrated superior accuracy in classifying the geospatial dataset related to the wind farm, achieving Se and Sp of 0.97 and 0.90, respectively. Following this, the MLP algorithm achieves Se and Sp values of 0.77 and 0.80, respectively. Lastly, the SVM algorithm presented Se and Sp values of 0.68 and 0.74, respectively. The highest overall precision in predicting suitable locations of power plants was achieved using the RF classifier with 90% and 87% for wind and solar, respectively, followed by the MLP classifier with 78% and 71% for wind and solar, respectively. In contrast, the SVM classifier provided the lowest Po, with 72% and 70%, for wind and solar, respectively. The k* indices for wind and solar modeling using the RF algorithm were 0.78 and 0.77 for wind and solar, respectively, highlighting its potential for predicting solar and wind farm sites. The main findings of the study are that the RF algorithm outperformed the other ML algorithms in predicting potential solar and wind energy sites.