A Robust Approach for Daily Solar Irradiance Clustering

Variability in solar irradiance is a major concern for power utilities and island grids. Clustering of historical irradiance data can help develop a better understanding of the patterns in a given location. This work uses a simple, yet effective scalar metric named Daily Irradiance Index (DII) obtained by comparing the measured irradiance level with the clear sky irradiance level. We explore a two-layered clustering approach for a data set from NREL's Solar Radiation Research Laboratory for Golden, Colorado. The first layer is designed to identify the seasons and in the second layer seasonal days are clustered based on the irradiance levels as measured by DII. For the first layer of clustering, we find the use of attributes from clear sky irradiance is sufficient to clearly identify the seasons. In the second layer of clustering, based on the cloud activities, days are clustered into high, medium and low irradiance levels corresponding to clear, intermittent, and cloudy days. The quality of the clusters is monitored through the well-known Silhouette score, Calinski-Harabasz score and Davies-Bouldin score. The effectiveness of employing clear sky irradiance, specifically sunshine duration, as a means of identifying solar seasons was demonstrated.