The invention belongs to the technical field of
spacecraft design and thermal management, and provides a
spacecraft temperature field modeling method and
system based on device
point cloud data and 3D U-Net, and the method comprises the steps: carrying out the preprocessing of the
point cloud data, and determining a feature
tensor; then, according to the feature
tensor and a preset 3D U-Net network, predicting a three-dimensional temperature field as an initial temperature field; and finally, according to the initial temperature field, adjusting the center coordinate of the device through
reinforcement learning, minimizing the volume of the high-temperature region, and generating a new temperature field. A high-resolution temperature field can be directly generated from
point cloud data of a plurality of devices in a
spacecraft by utilizing the powerful three-dimensional
convolutional neural network nonlinear fitting capability of the 3D U-Net network, and the
parallel computing capability of
deep learning is combined, so that the computing time is remarkably shortened, and the prediction precision under a complex
layout is improved; meanwhile, the coordinates of the device are efficiently adjusted through
reinforcement learning, the highest temperature is minimized, and the risk of heat concentration is reduced.