External authors:

• Eduardo Rodriguez ( Universidad de Chile )
• Enrique Lopez Droguett ( Universidad de Chile , University of California Los Angeles )

Abstract:

Accurate solar-plant design requires detailed measurement campaigns to determine the site's radiative conditions. In the absence of empirical data, researchers employ separation models to estimate solar radiation components by calibrating polynomial coefficients with local meteorological data. Previous studies have adjusted these coefficients for various climate zones using the K & ouml;ppen & Geiger classification, originally devised to demarcate regions based on plant distributions. Consequently, applying this classification to solar radiation may merge areas with different radiative characteristics, resulting in flawed assessments. This study describes a clustering technique that treats solar radiation as temporal data through the Discrete Fourier Transform and the Time Series Feature Extraction Library. By selecting input variables based on atmospheric attenuation and sky conditions, the K-means algorithm identified six clusters as the optimal solution, validated with a widely used separation model adjusted for the new clusters. The results of the estimation for the Cluster-adjusted model were then compared to the same separation model adjusted to the K & ouml;ppen & Geiger classification. The Cluster-adjusted model showed superior performance in 34 of 50 stations, which shows that grouping meteorological stations according to their radiative characteristics achieves better results than dividing them on the basis of climate.