Identifying snow in photovoltaic monitoring data for improved snow loss modeling and snow detection. (15th July 2021)

• Record Type: Journal Article
• Authors: Øgaard, Mari B.; Aarseth, Bjørn L.; Skomedal, Åsmund F.; Riise, Heine N.; Sartori, Sabrina; Selj, Josefine H.
• Is Part Of: Solar energy. Volume 223(2021)
• Page Start: 238
• Page End: 247
• Publication Date: 2021-07-15
• Subjects: Photovoltaic systems -- Snow -- Snow loss modeling -- Snow detection -- PV performance -- Soiling; Solar energy -- Periodicals; Solar engines -- Periodicals; 621.47
• DOI: 10.1016/j.solener.2021.05.023
• Languages: English
• ISSNs: 0038-092X

Abstract

Highlights: Snow introduces recognizable signatures in PV data, varying with type of snow cover. PV snow losses are influenced by the transmittance and nonuniformity of snow cover. The Marion snow loss model yields more accurate results than purely empirical models. The natural snow clearing rate of PV modules depend on snow cover depth. Accounting for snow depth in snow clearing modeling improves the Marion model. Abstract: As cost reductions have made photovoltaics (PV) a favorable choice also in colder climates, the number of PV plants in regions with snowfalls is increasing rapidly. Snow coverage on the PV modules will lead to significant power losses, which must be estimated and accounted for in order to achieve accurate energy yield assessment and production forecasts. Additionally, detection and separation of snow loss from other system losses is necessary to establish robust operation and maintenance (O&M) routines and performance evaluations. Snow loss models have been suggested in the literature, but developing general models is challenging, and validation of the models are lacking. Characterization and detection of snow events in PV data has not been widely discussed. In this paper, we identify the signatures in PV data caused by different types of snow cover, evaluate and improve snow loss modeling, and develop snow detection. The analysis is based on five years of data from a commercial PV system in Norway. In an evaluation of four snow loss models, the Marion model yields the best results. We find that system design and snow depth influence the natural snow clearing, and by expanding the Marion model to take this into account, the error in the modeled absolute loss for the tested system is reduced from 23% to 3%. Based on the improved modeling and the identified data signatures we detect 97% of the snow losses in the dataset. Endogenous snow detection constitutes a cost-effective improvement to current monitoring systems.