The present paper evaluates the soiling losses of a 3.25‐MW photovoltaic (PV) system installed in central Chile, 200 km north of Santiago, and analyzes the nonuniform soiling deposition between the various strings for a period of 3 years. A robust methodology is developed to extract, in the most systematic way, 142 reliable soiling profiles from the 256 PV power time series recorded on site. It is found that, if unmitigated, soiling would reduce the annual DC energy generation by 8%, with a factor of 2× between the losses of the most and least affected strings. Most of the losses are registered on the edges of the plant, closer to traffic and unpaved roads. The most soiling intense months are in summer, result of the infrequent rainfalls and of the high concentrations of suspended particles that characterize this season. The revenues and the costs of different manual cleaning frequencies are evaluated and compared to identify the optimal soiling mitigation strategy for this site. Three cleanings per year are found to return the highest profits for the economic conditions considered in this study. However, a sensitivity analysis shows how different cleaning costs and electricity prices would affect the soiling mitigation strategy. In addition, in light of the nonuniform soiling deposition distribution, the possibility of cleaning only selected strings rather than the full PV plant is discussed.

Several models have been presented in the recent years to estimate the magnitude of soiling from environmental parameters. However, these models are often based on data from a single site, or at most a few sites, and only limited data are, as of yet, available on their uncertainty. The present work aims to present a first comparative analysis of soiling estimation models, using measured soiling data from various locations in the USA. The study also investigates the impact that the source of the input data can have on the estimation. The results show that the model selection is only one of the factors that can affect the evaluation. Indeed, the use of satellite-derived or ground-mounted particulate matter data can lead to the generation of different soiling maps, with factors greater than 2× between the modeled losses. The current challenges and the unanswered questions that can bias soiling estimation are discussed. Additionally, potential research directions to improve the quality of soiling modeling are identified.

In this study, the impact of pollen as a PV soiling agent is investigated. The performance data of five utility-scale PV plants in North Carolina, USA, was collected and analyzed using two soiling extraction methods. Satellite and environmental data, including pollen counts, cropland, and vegetation, was also collected and analyzed to identify impacts to soiling losses. During the spring peak pollen season, performance losses of >15% were observed at all five sites. Partial performance recoveries following the pollen season were slow, with lack of correlation with rainfall. This means that the statistical soiling estimation methods that assume abrupt performance recovery from rain are not appropriate for pollen-impacted solar sites. When manual cleanings were performed on site the performance recovery ranged from 5% to 11% indicating persistent soiling impacts are present in this region. The results of this work provide new insights into the phenomenon of pollen deposition on PV systems, demonstrating that 1) soiling can also affect systems located in rainy locations and 2) that its effects cannot be determined using the current estimation methodologies.

An advanced control operation center to enable corrective, preventive and predictive maintenance, while also ensuring optimal photovoltaic (PV) plant performance was developed in this work. The developed software solution hosts innovative algorithms able to ensure data quality, while also allowing early failure and performance loss diagnosis without disrupting the normal operation of the PV plant. It is primarily based on real-time analysis of measurement data, machine learning and statistical analysis. The solution was validated experimentally against field measurements from an operating PV power plant of 1.8 $\mathbf{MW}_{\mathbf{p}}$ installed in Greece. The results showed technical availability and energy yield improvements of the test PV plant by handling intelligently the detected faults through the smart ticketing system. Optimal maintenance planning (e.g., optimum hardware replacement/maintenance, cleaning schedules, etc.) can thus lead to a reduction of operation and maintenance (O&M) costs and hence directly impacting positively the levelised cost of electricity (LCOE).