ADS-B Spoofing Attack Detection Method Based On Machine Learning

Automatic Dependent Surveillance-Broadcast(ADS-B)is an important part of the next-generation air traffic system.It obtains aircraft parameters through Global Navigation Satellite System(GNSS)and related airborne equipment,then broadcasts the aircraft's position,speed,heading,identification number and other information to other aircraft or ground base stations,so that the controller can monitor the status of the aircraft.However,the ADS-B protocol does not provide relevant information authentication and data encryption,so it is extremely vulnerable to spoofing attack.Based on the analysis of the ADS-B protocol and the technical solution of system security certification and verification,this paper starts with a data perspective and builds a deep learning-based sequence-to-sequence(seq2seq)model according to the fast update and strong time correlation of ADS-B messages to reconstruct the ADS-B data.The cosine similarity is used to measure the reconstruction error,and the training set is used to determine the abnormal threshold.Considering the time dependence of the data,this paper further uses the sliding window as unit to calculate the statistical characteristics of the relevant features in the window to enrich the training features.The model is a codec structure composed of a long short-term memory network(LSTM)and the training data is input in the form of a sliding window,so that the model can better capture the time characteristics of the data.In the test phase,if a sequence containing anomalies is input to the model,its reconstruction error will exceed the anomaly threshold,thereby achieving the effect of anomaly detection.In addition,this paper also uses one class support vector machine(OCSVM),isolation forest(IF),local outlier factor algorithm(LOF),and long-term and short-term memory(LSTM)neural network(Sequential model)to compare with the proposed model.Accuracy,recall,and F1 scores were used as measures for each method.The experimental results show that the seq2 seq model is superior to traditional machine learning methods,and after the data features are expanded,the detection effect is improved.Compared with the existing spoofing attack detection methods,the proposed methoddoes not need to change the protocol of the ADS-B system,nor does it require additional nodes or sensors to participate.It only needs the message sequence from the ADS-B system and output the classified results of abnormal detection through the visualization system.