Data set sampling and its implications on the heliostat calibration

Abstract

The ongoing global energy transition towards sustainable and climate-neutral power generation has led to the increasing adoption of concentrated solar tower power plants, relying on heliostats for precise solar tracking. Heliostat calibration, vital for maintaining accurate alignment, traditionally assumes a decrease in accuracy over time due to various factors. However, the impact of data set sampling on reported tracking accuracy has been overlooked. This paper utilizes a kNN (k-Nearest Neighbors) data set sampling approach to investigate data set distribution's impact on model accuracy. Results indicate that conventional time-dependent sampling can lead to an overestimation of reported accuracies. In contrast, the kNN sampling approach demonstrates a strong correlation between model performance and the proximity of test data to training data. Simulations reveal that reported accuracy scores are influenced by the similarity between training and test data sets. The study highlights the critical importance of considering data set distribution when interpreting accuracy scores. The proposed method improves tracking accuracy and offers a dependable metric for evaluating calibration results. It provides valuable insights to enhance heliostat calibration models, advancing precise solar tracking in concentrated solar tower power plants and supporting the global transition towards sustainable energy solutions.