Experimental dataset for flux density prediction in Concentrating Solar Power (CSP) plants

This dataset comprises the experimental measurements utilized for the "simulation-to-reality" (sim2Real) step of the Attention-Gated U-Net (AttU-Net) firstly trained on the simulated data from "Concentrating solar power (CSP) plants AI-training dataset for flux density measurements." (10.5281/zenodo.15173386). After proving the success of that training on the peer-review journal publication "Concentrating solar power (CSP) plant data-driven digital twin: A novel method for flux density prediction" (https://doi.org/10.1016/j.rineng.2025.107096), the authors conduct this second training with the aim of reaching experimental accuracies in flux density predictions for solar power towers. The normalized grayscale images were obtained throughout a 3-month experimental campaign which took place in the Solar Tower Jülich (STJ) facility between May, 2025 and July, 2025. The location of this plant is Jülich, Germany (50.915° N, 6.388° E) at 110 metres over sea level and it is equipped with 2016 heliostats of 8 squared-metres. The utilized Lambertian white surface is the 4x4 squared-metres radiation shield of HEHTRES receiver, located at 40.8 metres centred at the western tower of that facility (check layout images).

For repeatibility, this dataset is structured as follows:

• Training images. This folder contains the 572 "input" and "groundtruth" pairs - divided in two subfolders - utilized for training the AttU-Net. Therefore, their resolutions are downsampled to 256x256 pixels. 
  • Input: it contains 572 simulated flux maps under the same input conditions registered at the corresponding experiment. Each image is labelled by using the experiment time (nomenclature: "yymmdd_hhmmss_down.png").
  • Groundtruth: it contains the 572 real images of the radiation shield under irradiation. Each image is labelled by using the experiment time (nomenclature: "yymmdd_hhmmss_down.png").
• Input conditions. This folder contain a subfolder for each of the timesteps recorded in experiments. Inside of each subfolder (nomenclature: "yymmdd-hhmmss") a JSON file named the same way (+"yymmdd-hhmmss.json") is found. It contains the following self-explanatory entries:
  • "datetime". Date and time of the experimental measurement [yymmdd-hhmmss].
  • "radiometer_values". Recorded signals of the 4 radiometers equipped (1 to 4 in clockwise direction beginning from the lower left corner), [kW/m²].
  • "dni_value". Average DNI recorded by two pyrheliometers (located at each end of the heliostat field), [W/m²].
  • "reflectivity_value". Average reflectivity from the heliostat field.
  • "solar_elevation". Sun position at experiment time, [deg].
  • "solar_azimuth". Sun position at experiment time, [deg].
  • "active_heliostats". Heliostats focused toward the receiver.
• Layout figure.
• Validation dataset. GIF from validation (inference) conditions.
• Full dataset movement. GIF from training dataset conditions.