No abstract

Solar tower power plants play a key role to facilitate the ongoing energy transition as they deliver climate neutral electricity and direct heat for chemical processes. These plants generate temperatures over 1000 °C by reflecting sunlight with thousands of mirrors (heliostats) to a receiver. The temper- ature achievable in practice is limited due to the system’s susceptibility to small surface defects and misalignments of individual mirrors, hindering the plant’s full efficiency. We present an inverse render- ing technique that predicts the incident power distribution of each heliostat, including the inaccuracies, based solely on focal spot images that are already acquired in most solar power plants. The method allows reconstructing flawed mirror shapes within sub-mm precision. Applied at the solar tower plant in Juelich, our approach outperforms all alternatives in accuracy and reliability. Our data-driven method is a key ingredient to building digital twins of solar power plants. It can be integrated into the existing infrastructure and plant control at low cost, leading to increased efficiency of existing and decreased expenses for future power plants, the key factors of success in the competitive market. For other fields, our approach can be a blueprint, as we present the option for the very first large-scale indus- trial deployment of differentiable ray tracing. Merging data-intensive Machine Learning with physical modeling creates flexible, data-efficient and trustworthy solutions applicable in science and industry.

A precise and reliable alignment of the two-axis heliostat tracking is of great importance for an efficient operation of solar power towers. In order to minimize the tracking error of heliostats, especially in large plants, it is essential to recalibrate the heliostat control unit regularly. Conventional calibration methods with regression can meet the requirements of frequent and regular use, but they cannot adequately account for the many factors that influence alignment. Deep learning algorithms have made remarkable progress in recent years and have the potential to reduce the number of calibrations over time while reducing tracking errors. However, neural networks are still rarely used for such purposes, because such algorithms usually require an extremely large amount of data to map the individual heliostat errors. We present a comparison of different pre-train studies for neural networks to reduce the amount of data per heliostat which are needed to improve the accuracy compared to a state-of-theart method by applying supervised as well as unsupervised pretraining.