Characterizing soiling losses for photovoltaic systems in dry climates: A case study in Cyprus

• Published in: Solar energy 2023-05, Vol.255, p.243-256
• Main Authors: Lopez-Lorente, Javier, Polo, Jesús, Martín-Chivelet, Nuria, Norton, Matthew, Livera, Andreas, Makrides, George, Georghiou, George E.
• Format: Article
• Language: English
• Subjects: Dry climate; Machine learning; Photovoltaic; Soiling
• ISSN: 0038-092X; 1471-1257
• DOI: 10.1016/j.solener.2023.03.034

•Evaluation of an outdoor soiling test-bench over a 2-year period in Cyprus.•Soiling rates from 0.039 to 0.535%/day depending on the season and dust episodes.•Comparison of 6 soiling models based on physical and machine learning approaches.•Physical models performed slightly better than machine learning models.•Comparison of soiling loss models with field and satellite-derived weather data. Ensuring optimal performance of solar photovoltaic (PV) systems requires the extensive assessment and understanding of losses of different origin that affect these installations. Soiling is a key loss factor influencing the performance of PV systems, particularly in arid and dry climatic regions, and its thorough knowledge and modelling aspects including the seasonal evolution is challenging for the early stages of energy prospecting for PV power plants. The purpose of this study is to address this fundamental challenge by evaluating the loss of soiling and the performance of six soiling models based on both physical and machine learning (ML) approaches. Specifically, the case study is a soiling test-bench experimental apparatus installed at the outdoor test facility of the University of Cyprus in Nicosia, Cyprus. The climatic conditions of the site represent a dry climate with high PV potential due to high irradiation levels. The obtained results reported soiling rates ranging from 0.039%/day to 0.535%/day depending on the season and the presence of dust episodes. The average yield daily and monthly soiling losses were 1.9% and 2.4% over a 2-year period, respectively. Furthermore, the comparative analysis of the different soiling models illustrated that the physical models achieved slightly better performance than the ML models with root mean square error (RMSE) of 1.16% and 0.83% for daily and monthly losses, respectively. Finally, the findings provide evidence and useful information on the performance and limitations of the different soiling models for fielded PV systems located in arid and dry climatic zones.