Abstract:Automatic Dependent Surveillance-Broadcast is a critical component of next-generation air traffic management systems. However, its vulnerability to spoofing interference arises from the plaintext broadcasting of ADS-B messages without data encryption and authentication. Aimed at this problem, a spatio-temporal gated variational autoencoder-based anomaly detection algorithm for ADS-B data is proposed. The encoder employs bidirectional long short-term memory to model local temporal features, combines a three-layer eight-head Transformer to extract global spatio-temporal features, and utilizes a gating network to dynamically fuse these features. Variational inference is introduced to generate the latent space distribution, constraining the model's probabilistic modeling of normal flight patterns. The decoder incorporates a cascaded structure of a single-layer LSTM and a two-layer Transformer to simultaneously reconstruct multidimensional flight parameters through fully connected layers. Experimental results demonstrate that the proposed model effectively detects various anomalies in ADS-B data across different attack scenarios, with superior performance over relevant baseline algorithms. This approach provides a feasible solution for enhancing the security of air traffic management systems.
