The magnetic field plays an essential role in the initiation and evolution of different solar phenomena in the corona. The structure and evolution of the 3D coronal magnetic field are still not very well known. A way to ascertain the 3D structure of the coronal magnetic field is by performing magnetic field extrapolations from the photosphere to the corona. In previous work, it was shown that by prescribing the 3D-reconstructed loops’ geometry, the magnetic field extrapolation produces a solution with a better agreement between the modeled field and the reconstructed loops. This also improves the quality of the field extrapolation. Stereoscopy, which uses at least two view directions, is the traditional method for performing 3D coronal loop reconstruction. When only one vantage point of the coronal loops is available, other 3D reconstruction methods must be applied. Within this work, we present a method for the 3D loop reconstruction based on machine learning. Our purpose for developing this method is to use as many observed coronal loops in space and time for the modeling of the coronal magnetic field. Our results show that we can build machine-learning models that can retrieve 3D loops based only on their projection information. Ultimately, the neural network model will be able to use only 2D information of the coronal loops, identified, traced, and extracted from the extreme-ultraviolet images, for the calculation of their 3D geometry.

磁场在日冕中不同太阳现象的发生和演化中起着至关重要的作用。3D日冕磁场的结构和演化仍然不是很清楚。确定日冕磁场 3D 结构的一种方法是执行从光球层到日冕的磁场外推。在以前的工作中，表明通过规定 3D 重建回路的几何形状，磁场外推产生了一个在建模场和重建回路之间具有更好一致性的解决方案。这也提高了场外推的质量。使用至少两个视图方向的立体镜是执行 3D 冠状环重建的传统方法。当只有一个冠状环的有利位置可用时，必须应用其他 3D 重建方法。在这项工作中，我们提出了一种基于机器学习的 3D 循环重建方法。我们开发这种方法的目的是使用尽可能多的空间和时间观测到的日冕环来模拟日冕磁场。我们的结果表明，我们可以构建机器学习模型，该模型可以仅根据投影信息检索 3D 循环。最终，神经网络模型将能够仅使用从极紫外图像中识别、追踪和提取的日冕环的 2D 信息来计算它们的 3D 几何形状。我们开发这种方法的目的是在空间和时间上使用尽可能多的观测日冕环来模拟日冕磁场。我们的结果表明，我们可以构建机器学习模型，该模型可以仅根据投影信息检索 3D 循环。最终，神经网络模型将能够仅使用从极紫外图像中识别、追踪和提取的日冕环的 2D 信息来计算它们的 3D 几何形状。我们开发这种方法的目的是在空间和时间上使用尽可能多的观测日冕环来模拟日冕磁场。我们的结果表明，我们可以构建机器学习模型，该模型可以仅根据投影信息检索 3D 循环。最终，神经网络模型将能够仅使用从极紫外图像中识别、追踪和提取的日冕环的 2D 信息来计算它们的 3D 几何形状。