Evaluation of classification methods according to solar radiation features from the viewpoint of the production of parabolic trough CSP plants

In this work, the representativeness of the day-types classified according to the solar radiation features by two classification methods is evaluated from the perspective of the production of two parabolic trough plants. A new methodology to characterize the representativeness of the day-types using a novel index is proposed, based on the characterization of the daily production. As a previous step to the use of a classification method, the evaluation methodology helps to select the most adequate model and to improve it from the perspective of a concentrated solar power project. This methodology is applied to 16 years of measurements from Seville (Spain) classified by two methods: a method based on daily clearness index values (kt), and a method that uses clustering techniques to define the day-types. From the application of the methodology to the clustering classification some improvements are identified and applied. As a result, from the 10 day-types identified by the clustering classification method a new classification based on 8 day-types with different features for each type of plant is proposed. The use of this classification to estimate the daily yield outperforms the results obtained with the kt classification, with a mean yearly RMSE value more than 20% lower.