Computational Intelligence Based Snow Cover Prediction for Photovoltaic Systems

In northern snow-prone areas, photovoltaic (PV) systems are getting more popular. Accumulations of snow on panels after snowfall events, as a major challenge for PV systems’ efficient use in these regions, can attenuate or obstruct solar radiation reaching the surface of the PV cells and cause a significant reduction in the PV system’s power generation. This is an important issue in PV power forecasting (PVPF) for PV-penetrated power systems’ scheduling. To address this issue, data-driven short-term snow cover prediction models for PV systems are proposed in this paper. According to the best of our knowledge, utilizing computational intelligence techniques to predict the presence of a snow cover on PV panels with an hourly resolution solely based on the main meteorological parameters is performed for the first time in the literature. The output of these models can be used as an input for the PVPF stage and help to reduce PVPF errors in snow conditions by enabling the implementation of PVPF approaches compatible with the characteristics of snow-covered PV systems. The study is performed on the historical dataset of electrical and meteorological parameters of a PV system in Canada over 3 years. By applying 5-fold cross-validation and hyperparameter tuning, the best hourly snow cover prediction accuracy, 96%, has been obtained by the developed gradient boosting tree model. Testing this model on the unseen data of 2 other PV systems has resulted in 80% and 78% accuracy for snow cover prediction.