AE-BiLSTM: Multivariate Time-Series EMI Anomaly Detection in 5G-R High-Speed Rail Wireless Communications

With the global expansion of high-speed rail (HSR) and the integration of the latest wireless communication networks into the railway system, establishing a secure bidirectional communication link between moving trains and base stations (BSs) is vital to ensure real-time control. The increasing complexity of contemporary railway systems and heightened exposure to electromagnetic interference (EMI) have led to operational disruptions and security risks. This paper introduces a real-time anomaly detection approach that utilizes a deep learning algorithm based on autoencoder (AE) and long short-term memory (LSTM). By analyzing multivariate time series characteristics, the method simultaneously examines the time and frequency domains at a finer resolution, achieving a desirable trade-off between false alarms and missed anomalies. Specifically, our approach enhances accuracy by 5%, reaching 93.24% in comparison with some state-of-the-art methods. The online detection takes 4.51 ms, meeting the security latency requirements. This highlights the potential for timely detection of unforeseen EMI incidents in diverse scenarios and at varying speeds.