Quantification of Snow Losses Using Limited PV Production Data

As photovoltaic installations are increasingly deployed in northern latitudes it is essential to accurately identify and quantify the snow related power losses to mitigate financial and operational risks for photovoltaic system owners and financiers. Typically, generic snow loss estimates are used for financing PV projects with varying accuracy. This study proposes an innovative methodology to identify snow-covered production and quantify snow losses using limited data from monitoring sites. Specifically, it utilizes photovoltaic power production, plane of array irradiance, and ambient temperature measurements. The method involves two steps: using Monte Carlo Markov Chain to generate samples from snow covered days, then using the statistical properties of generated samples to identify snow signatures from photovoltaic production data. If the standard deviation and the mean of daily photovoltaic production are within the confidence intervals obtained via Monte Carlo Markov Chain sampling the day is labeled as snow-covered. After labeling snow-covered days, we remove them and use two imputation strategies namely the k-Nearest Neighbor and Multivariate Imputation by Chained Equations to fill in the removed values. Finally, the difference between the actual and expected production is used as a quantification of snow related losses. The first results on labeled snow give reasonable estimates of snow related power losses. The approach will be applied on utility scale photovoltaic plants.