Machine Learning-Based Power Loss Analysis in Photovoltaic Systems

This research estimates the power loss in photovoltaic (PV) systems through a Random Forest model by using 15-min interval data for two years (2021–2023) of operation from a university campus located in Delhi, India. Irradiance, temperature and the dust concentration were used in the analysis to calculate and estimate actual and theoretical potentials and power losses based on factual, modelled and theoretical determinations under Standard test conditions (STC). Seasonal changes are of particular interest and therefore, the power loss is to be analyzed based on three power quality reports obtained over three different seasons in 2023. The results of the model show the Mean absolute error (MAE) to be 29.145 W along with Root mean square error (RMSE) to be 57.72 W. Interactions with power loss observed seasonally proved to have variations in dust and power loss, temperature, and so on; all of which confirmed the effectiveness of Random Forest model in providing valuable predictions of season-sensitive data. These results would be significant to enhance the efficiency of the PV system and how energy technologies for dusty areas should be designed.