Enhancing heliostat calibration on low data by fusing robotic rigid body kinematics with neural networks

Abstract

Solar tower power plants rely on very precise alignments of their heliostats for efficient operation. To achieve this, the heliostats have to be calibrated regularly. Open-loop calibration procedures are the most common type due to their cost-effectiveness. Two main approaches to these algorithms exist: geometry-based robotic kinematics and neural network-based models. While the former is reliable and requires little data, it only yields moderate accuracy. The latter, however, promises higher accuracies but is data-hungry and unreliable. We here present a two-layer hybrid model that combines a well-established geometric model for pre-alignment with a neural network disturbance model whose impact is gradually adapted through a regularization sweep. This approach ensures that the prediction accuracy is, in the worst-case, equivalent to that of the rigid-body model. Moreover, it helps to identify deficiencies that may have been overlooked by the physical approach. Especially, it is capable to compute deviation from the geometry models averaged optimum. For testing, real measurement data from daily heliostat calibration at the solar tower in Jülich is used. To report accuracies which hold true throughout the year a special training/validation data sampling is employed, which allows for a conservative performance assumption. The results demonstrate that the Hybrid-model outperforms rigid-body models starting from the first measurement, achieving a top performance below 0.7 milliradians. In conclusion, the proposed hybrid model provides a cost effective in-situ solution for heliostat calibration with highest accuracies on low data in solar tower power plants for all open-loop calibration methods.