ADS-B anomaly data detection model based on VAE-SVDD

Abstract

As a key technology of the new generation air traffic surveillance system, ADS-B (Automatic Dependent Surveillance-Broadcast) is vulnerable to cyber security challenges because it lacks data integrity and authentication mechanism. For detecting ADS-B data attacks accurately, an anomaly detection model is proposed which fully considers temporal correlations and distribution characteristics of ADS-B data. First, VAE (Variational AutoEncoder) is used to reconstruct ADS-B data so that the reconstructed values can be obtained. Then, for the sake of solving the adaptive problem of anomaly detection threshold, the difference values between the reconstructed values and the actual values are put into SVDD (Support Vector Data Description) for training, and a hypersphere classifier that can detect ADS-B anomaly data is obtained. In addition, in order to prevent overfitting and underfitting, appropriate reconstructed values are selected which can reduce FPR (False Positive Rate) and FNR (False Negative Rate) of anomaly detection. Experiments show that the VAE-SVDD model can detect ADS-B anomaly data which is generated by attacks such as random position deviation and constant position deviation. Moreover, compared with other machine learning methods, this model is not only more adaptable, but also has a lower FPR and FNR.