Optimized machine learning approach for intelligent dust contamination prediction in solar photovoltaic systems

Solar photovoltaic systems suffer significant losses in their performance due to contamination by dust, bird droppings and other pollutants, resulting in reduced power output and greater regular maintenance requirements. Thus, the key requirement for the proper functioning of a PV system is correct prediction of contamination which allows performing timely maintenance for better working of the PV system. In this paper, an approach based on the combination of machine learning and particle swarm optimization presented with the aim to predict contamination conditions in a solar PV system. A thorough validation of the proposed model done using SVM, Random Forest, XGBoost, CatBoost, PSO-SVM, PSO-RF, PSO-XGBoost and CNN-PSO models. The methodology consists of feature preparation and implementation followed by machine learning models that compared in terms of performance features. Among all the assessed models, the CatBoost model showed the best performance with a prediction accuracy of 99.43%.The findings demonstrate the potential of ensemble-based ML, particularly CatBoost, for intelligent PV contamination monitoring and predictive maintenance, enabling earlier identification of contamination conditions and supporting improved energy-generation performance and reduced operational losses.