Research on the Application of Artificial Intelligence Technology in ADS-B Data Anomaly Detection

Automatic Dependent Surveillance Broadcast (ADS-B) system is widely used in the field of aerial surveillance because of its accuracy and high efficiency. However, ADS-B system is easy to be spoofed and interfered because it broadcasts signals in plaintext format and lacks data encryption and message authentication mechanism. This article conducts research on this issue. Firstly, it analyzes the structure of the ADS-B system and discusses its data anomaly problems. It summarizes that researchers at home and abroad generally approach the solution to data anomaly problems from three directions: encrypting the data link, improving the ADS-B hardware and software, and analyzing from the perspective of the ADS-B data fields. Then, it particularly points out that artificial intelligence algorithms also build models from the perspective of the ADS-B data fields to complete ADS-B track data anomaly detection, including algorithms based on clustering, data reconstruction, and data temporal correlation. Finally, this paper compares the detection precision, recall rate, and F1 score of several deep learning algorithm models based on data time-related, and finds that they all have excellent ADS-B data anomaly detection effects, with detection precision all above 92.2%, recall rate all above 84.9%, and F1 score all above 86.8%.