In railway environments, communication signals may become very weak due to geographical conditions, building structures, or other factors, leading to a decrease in communication quality. The Bidirectional Long Short-term Memory (Bi-LSTM) model was adopted to accurately predict the signal strength of future time steps. By establishing a railway communication network (RCN) signal enhancement system, the performance of the RCN was improved. A large amount of historical data on RCNs was collected and preprocessed. Features related to RCN signals were extracted, and the entire time series data was divided into datasets. By using bidirectional LSTM layers, patterns and features in the sequence were learned, and future signal strength was predicted and analyzed for targeted signal enhancement. The experimental results showed that the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) of Bi-LSTM for signal strength prediction were 0.04 and 0.08, respectively. The average delay improvement rate of Bi-LSTM was 58.6%, and the interference suppression rates of Bi-LSTM model for electromagnetic interference, radio frequency interference, natural environment interference, multipath propagation interference, and mechanical vibration interference were 78.6 dB, 56.2 dB, 67.8 dB, 79.2 dB, and 71.2 dB, respectively. The application of Bi-LSTM model can effectively predict signal strength and provide a new method for signal enhancement in RCNs.
