A Deep Learning Framework for Safety Monitoring of a Railway Section

This paper presents an approach for anomaly detection and forecasting that aims to protect the integrity and operational continuity of IoT critical infrastructure, specifically for a railway use case. This system is designed to monitor in real-time the sensor data to detect any deviations that may indicate a potential data tampering attack or a sensor malfunction. This early identification and notification of such anomalies is crucial for preventing unauthorized access and mitigating the risks associated with data tampering attacks. The proposed system is comprises of two primary components: a forecasting component and an anomaly detection component. The Forecasting Component uses a Long Short-Term Memory (LSTM) model to predict future sensor values, while the Anomaly Detection Component employs the Tukey's fence method to identify sensor measurements that significantly deviate from normal behaviour. A railway use case is included to demonstrate the practical application of the deep learning framework. These components were evaluated and both demonstrated excellent performance. The Forecasting Component provided highly accurate predictions of future sensor values, while the Anomaly Detection Component effectively identified deviations from normal patterns. The evaluation results confirmed the system's ability to detect significant anomalies and its capability to maintain operational integrity and security in IoT critical infrastructures.