The authors present a machine learning approach for the extraction of radiowave propagation models in tunnels. To that end, they discuss three challenges related to the application of machine learning to general wireless propagation problems: how to efficiently specify the input to the model, which learning method to use and what output functions to seek. The input that any propagation modelling tool (be it a ray-tracer, a full-wave method or a parabolic equation solver) uses, can be considered as visual, in the form of an image or a point cloud of the environment under consideration. Therefore, they propose an artificial neural network structure that generalises well to various geometries. The desired output can be values of the electromagnetic field components across the channel or just a path loss model. They apply these ideas to the case of arched tunnels for the first time. They consider cases where the geometric parameters of the tunnel, the position of the receiver and the frequency of operation are parts of a model trained by a vector parabolic equation solver. The model is evaluated using solver-generated as well as measured data. The numerical results demonstrate that this approach combines computational efficiency with high accuracy.

中文翻译：

---

隧道中无线电波传播的人工神经网络模型

作者提出了一种机器学习方法，用于提取隧道中的无线电波传播模型。为此，他们讨论了与将机器学习应用于一般无线传播问题相关的三个挑战：如何有效地指定模型的输入，使用哪种学习方法以及寻找哪种输出功能。任何传播建模工具（无论是光线示踪剂，全波方法还是抛物线方程求解器）使用的输入都可以视为可视图像，可以是所考虑环境的图像或点云形式。因此，他们提出了一种人工神经网络结构，可以很好地推广到各种几何形状。期望的输出可以是跨通道的电磁场分量的值，也可以是路径损耗模型。他们首次将这些想法应用于拱形隧道。他们考虑了以下情况：隧道的几何参数，接收器的位置和运行频率是矢量抛物线方程求解器训练的模型的一部分。使用求解器生成的以及测量数据评估模型。数值结果表明，该方法将计算效率与高精度相结合。