Fig. 4: Hyper-parameter study for the heliostat array. | Nature Communications

Fig. 4: Hyper-parameter study for the heliostat array.

From: Automatic heliostat learning for in situ concentrating solar power plant metrology with differentiable ray tracing

Fig. 4

a Comparison of the irradiance profile on three days of the year, i.e., the shortest, longest and at equinox. The ground truth is a simulation based on a deflectometry measured surface, contrasted with a perfectly planar heliostat and the proposed differentiable ray tracer trained on 16 images. Our method predicts complex structures accurately on a fine-grained level. b Simulated and inferred irradiance profiles at different distances between heliostat and calibration target. c Quantitative assessment of prediction quality for varying distances and number of training images. Comparisons in all subfigures are made with test images distinct from the training set. To facilitate an enhanced visual comparison of the focal spot details, the images were cropped. The depicted loss corresponds to the image of the complete calibration target. The numerical values can be taken from the Hyperparameter section in the Supplementary.

Back to article page