Classification of days according to DNI profiles using clustering techniques
A methodology to classify days as a function of the state of the sky for Concentrated Solar Power (CSP) plant operation is proposed. For this purpose, three indexes are used to characterize the energy, variability and time distribution of the DNI and to define the type of days by means of clustering...
| Autores: | , , , |
|---|---|
| Tipo de recurso: | artículo |
| Estado: | Versión aceptada para publicación |
| Fecha de publicación: | 2017 |
| País: | España |
| Institución: | Universidad de Sevilla (US) |
| Repositorio: | idUS. Depósito de Investigación de la Universidad de Sevilla |
| OAI Identifier: | oai:idus.us.es:11441/163040 |
| Acceso en línea: | https://hdl.handle.net/11441/163040 https://doi.org/10.1016/j.solener.2017.02.031 |
| Access Level: | acceso abierto |
| Palabra clave: | State of sky DNI Transmittance index Clustering techniques |
| Sumario: | A methodology to classify days as a function of the state of the sky for Concentrated Solar Power (CSP) plant operation is proposed. For this purpose, three indexes are used to characterize the energy, variability and time distribution of the DNI and to define the type of days by means of clustering techniques. Two sets of indexes are tested and compared. The energy of days is represented by the transmittance index, kb. Two indexes are used to characterize the variability of the DNI: persistence index of the instantaneous kb values (POPD) and Variability Index (VI). Equivalent indexes have been previously used to classify the types of days using Global Horizontal Irradiation (GHI). A novel index to define the time distribution of the DNI daily energy is introduced. Clustering analysis is applied to thirteen years (2000–2012) of 10-min DNI measurements recorded in Seville (37.40°N, 6.01°W) by the Group of Thermodynamics and Renewable Energy (GTER) at the University of Seville. The k-medoids algorithm is used for cluster analysis. Through the use of well-known internal validity indexes and with the help of the L-method, the optimum number of clusters (types of days) is found to be 10. The results are compared with the assessment carried out by five experts on a reference set composed of DNI daily curves from two years (2010 and 2011). This comparison reveals a better coincidence when the clustering is performed using VI. |
|---|