| Popis: |
Heliostat calibration is a vital task in solar tower plants to ensure high plant efficiencies. Currently calibration is performed at close time intervals for every heliostat, to ensure constantly precise heliostat tracking, because it is assumed that a heliostat's precision tends to degrade over time. Consequently, new heliostat calibration procedures are frequently presented that incorporate time-dependent data set splitting for training and testing. In this study, we present a new data-set splitting method that measures the nearest neighbor distance between each calibration point using sun positions in Euler angles (Azimuth, Elevation). We conducted a comparative analysis with the common time-based split method, and our results demonstrate that neither time nor data set size significantly impacted the tracking accuracies. Instead, the distribution of sun position within the data set emerged as the most important factor. Moreover, our findings suggest that using time as a metric can be misleading when reporting validation accuracies. The proposed method has significant implications for the calibration of heliostats, previous and upcoming publications as well as the daily power plant operation. Utilizing this method, the acquired data sets are expected to achieve higher levels of accuracy while requiring less data. Furthermore, it has the potential to enhance the comparability between publications and enable risk-averse assessments of new methods to ensure stated accuracies and improve model evaluation reliability. |
