Abstract

The global capacity of solar photovoltaic (PV) systems has increased exponentially in recent years and is expected to grow more than tenfold by 2050. With the increase in PV penetration levels into the existing electricity grid, weather-induced variability in power output has become extremely significant. Variability in the amount of power generated can impact the supply -demand ratio and grid stability by influencing voltage and frequency. To address sudden changes in the voltage caused due to ramps (sudden fluctuations in the power output), plant operators usually use ramp control devices. Battery energy storage (BES) systems are effective for storing excess generation during clear-sky days and then supplying this excess during energy deficit periods. Variability in solar resources is studied on multiple time-scales depending on the requirement. Understanding short-term solar resource variability (seconds to minutes) is important for ramp detection and forecasting, while long-term variability studies (daily to decadal scale) is important for resource estimation, site assessment, system design and estimating storage requirements. In the past, there have been several studies dedicated towards developing and improving the clear sky models. The latest clear-sky models take into account the complex atmospheric composition, are dependent on the location and can precisely forecast clear-sky irradiance. However, despite having some of the best-performing clear-sky models, identifying daily sky conditions remains a challenge.
To date, all the studies dedicated towards identifying sky-conditions are based on high resolution temporal data ranging from seconds to minutes. This study aims to propose a new sky-classification scheme that can be homogenously applicable to datasets with any temporal resolution globally. This study uses a clustering technique to classify sky conditions into five days: clear days, overcast days, low intermittent days, high intermittent days and highly variable days. On performing a Chi-square test on the training and test sets, we obtain chi-square statistic values are greater than 3 with p-value > 0.05. This indicates that the distribution of the training and test clusters are similar indicating the robustness of the proposed sky-classification scheme. This proposed classification scheme can be homogenously applied to any dataset globally despite their temporal resolution. We demonstrate the use of this scheme for resource allocation, site selection, understanding future intermittency due to climate change and cloud enhancement effects.