A Widely Applicable ADS-B Anomaly Data Detection Method

‍ ‍Automatic Dependent Surveillance-Broadcast （ADS-B） system is widely used in Air Traffic Surveillance and Management because of its high accuracy and efficiency. However， ADS-B relies on Global Navigation Satellite System （GNSS）， and broadcasts messages in plaintext without authentication and encryption. All of these defects make ADS-B system extremely vulnerable to malicious attacks. Several different ADS-B anomaly data detection models based on deep learning were proposed in recent years， but these models are only suitable for the specific flight operations in a fixed route or only effective for the flights in cruise phase， but not for the flights during takeoff and approach phases. In view of this problem， an ADS-B anomaly data detection model which combines K-Means clustering and LSTM neural network is proposed. By dividing the flight trajectory into different flight phases， the model can learn the characteristics of trajectories in different ones. Comparative experiments show that， the hybrid model has a high accuracy and recall rate for ADS-B abnormal data detection in all flight phases of takeoff， cruise and approach landing. The overall detection effect is better， the applicability is wider， and it is suitable for detecting real abnormal events.