Stratospheric airships with a solar array have demonstrated overwhelming superiority in many aspects, such as earth observation, meteorological survey, and communication relay. The solar array supplies sufficient power for the airship to be in flight for months, but excessive heat is also transferred to the airship, causing high overpressure of inner gas. However, the optimal arrangement of the insulation layer on the airship has not been investigated. The theoretical method, including the geometry, thermal, and energy models, is developed and validated. The distribution of the temperatures and power of the solar cells, with different installation angles on the curved surface, is investigated. The thickness of insulation layer has a significant effect on the solar output power and internal pressure of the airship. An optimized configuration of the insulation structure is proposed, in order to improve the total output energy of solar array. The optimized configuration of insulations helps to reduce the structural mass by 24.9% and increase the payload mass by 9%. Moreover, the optimized arrangement improves the output energy of solar array in a year, and the maximum improvement is 8.2% on the winter solstice. The work displays the optimization of the thermal insulation layer for the stratospheric airship with a solar array, in order to improve the everyday energy acquirement during flight in a year.

The stratospheric airship is a type of aerostat that uses solar energy as its power source and can fly continuously for months or even years in near space. The rapid and accurate prediction of the output power of its solar array is the key to maintaining energy balance and extending flight time. This paper establishes an online learning model for predicting the output power of the solar array of stratospheric airships. The readings of radiometers arranged on the surface of the airship are used as features for training the model. The parameters of the model can be updated in real-time during the flight process without retraining the entire model. The effect of radiometer placement on the model accuracy was also analyzed. The results show that for the continuous flight of 40 days, the online learning model can achieve an accuracy of 88% after training with 10 days of flight data and the accuracy basically reaches its highest level after 20 days. In addition, placing the radiometers at the four corners of the array can achieve a higher prediction accuracy of 95%. The online model can also accurately identify and reflect the effect of module efficiency attenuation or damage and maintain high accuracy.

Convolutional neural networks (CNNs) and transformers have achieved great success in hyperspectral image (HSI) classification. However, CNNs are inefficient in establishing long-range dependencies, and transformers may overlook some local information. To overcome these limitations, we propose a U-shaped convolution-aided transformer (UCaT) that incorporates convolutions into a novel transformer architecture to aid classification. The group convolution is employed as parallel local descriptors to extract detailed features, and then the multi-head self-attention recalibrates these features in consistent groups, emphasizing informative features while maintaining the inherent spectral–spatial data structure. Specifically, three components are constructed using particular strategies. First, the spectral groupwise self-attention (spectral-GSA) component is developed for spectral attention, which selectively emphasizes diagnostic spectral features among neighboring bands and reduces the spectral dimension. Then, the spatial dual-scale convolution-aided self-attention (spatial-DCSA) encoder and spatial convolution-aided cross-attention (spatial-CCA) decoder form a U-shaped architecture for per-pixel classifications over HSI patches, where the encoder utilizes a dual-scale strategy to explore information in different scales and the decoder adopts the cross-attention for information fusion. Experimental results on three datasets demonstrate that the proposed UCaT outperforms the competitors. Additionally, a visual explanation of the UCaT is given, showing its ability to build global interactions and capture pixel-level dependencies.