A High-Accuracy Dynamically-Stacked Residual LSTM-Based Radio Wave Propagation Model for Long-Distance Tunnels

This letter proposes a high-accuracy dynamically-stacked residual long short-term memory (DSR-LSTM) based modeling approach for radio wave propagation in long-distance tunnels. The proposed architecture utilizes simulated electric field distribution from the initial segment of the tunnel to learn the spatial evolution patterns of signal strength along the propagation path, thereby enabling accurate prediction of channel characteristics in subsequent tunnel sections. By introducing dynamic skip connections and residual learning, the model effectively captures long-term dependencies and enhances prediction stability. The proposed DSR-LSTM model demonstrates high prediction accuracy across diverse tunnel geometries. Validation against both numerical simulation results and field measurements in actual tunnel environments confirms its accuracy and practical applicability.