| Work Detail |
Cypriot scientists evaluated six different models used to predict energy losses caused by the accumulation of dust, dirt and other substances on the surface of photovoltaic panels in the islands arid climate.
Around the world, many of the locations that offer the highest solar irradiance are also handicapped by dry and dusty ground conditions, which can cause various problems for PV system performance.
Coping with losses caused by dust buildup on the surface of a module is big business for the PV industry, as these losses can quickly turn into significant revenue losses. Cleaning modules too frequently or investing in the wrong cleaning equipment can also hurt the profitability of a project. For this reason, PV project developers and system operators highly value the ability to accurately predict short- and long-term soiling losses.
Various approaches exist, using different combinations of in situ sensors, historical climate data, local weather data, satellite imagery, etc. A group of scientists led by the University of Cyprus have attempted to compare the accuracy of some of these methods, by comparing fouling loss prediction models with data from a test facility on the University of Cyprus Nicosia campus.
Machine Learning
Dirt losses in the test facility were calculated by comparing a set of clean and uncleaned modules. Six different models were tested for accuracy, three based on a physical model and three based on machine learning.
The three “physical” models are well-established methods for modeling dirt, while the machine learning methods are open source programs being applied to dirt measurement for the first time. In the article “ Characterizing soiling losses for photovoltaic systems in dry climates : A case study in Cyprus ”, published in Solar Energy .
The evaluation showed that the physical models, fed with data observed in the field, achieved the highest precision, with error rates (mean square error) of 1.16% in daily losses due to fouling and 0.83% in daily losses due to soiling. per month for fouling, for the highest performing machine learning model, called CatBoost.
Machine learning approaches, however, were not far behind, with 1.55% error for daily dirt losses and 1.18% for monthly ones. The researchers note that given the deficiencies in the availability of ground-observed data covering an entire site over a sufficient period, machine learning models, based on satellite-collected environmental data, could also be a useful approach.
“Soil modeling with this type of satellite-derived environmental data could help plan O&M strategies and operations throughout the year to minimize soil loss, especially in arid and dusty regions where sudden changes in weather can occur. aerosol load and precipitation are much less frequent,” the researchers explain. |
