Assessment of Random Forest Algorithm Efficacy for Photovoltaic Production Prediction During Saharan Dust Intrusion Episodes: A Case Study Including Wet Deposition

This study assesses the efficacy of the Random Forest (RF) algorithm in predicting photovoltaic (PV) energy production during Saharan dust intrusion episodes, including days of wet deposition. The research focuses on a PV plant in Tenerife, Canary Islands, during the period July 24 to September 20, 2015, encompassing a significant dust event. The RF model was applied to both daily and hourly energy production forecasts, incorporating various meteorological and atmospheric parameters as input features. For daily predictions, the model achieved high accuracy (97.58%) using only four key features: solar insolation hours, incident shortwave radiation, wind velocity, and maximum gust velocity. Hourly predictions required ten features to attain comparable accuracy (95.86%). The study demonstrates the RF algorithm's robustness in forecasting PV output under complex atmospheric conditions, even with limited non-in-situ data. Key words. African dust, Random Forest, Photovoltaic energy, energy production, prediction