| ADS-B has the advantages of high precision,wide coverage,supporting data sharing and air surveillance,and is the key component of next generation air transport system(NextGen).The ADS-B system on the airborne aircraft can regularly play the aircraft position and flight identification information based on the satellite,but because the ADS-B message protocol lacks the identity verification of the message sender or the ground system,the data is broadcast unencrypted by default,which makes the system vulnerable to various network attacks,receiving false information and forged or repeated data.At present,the main response strategy is to use additional participating nodes or sensors to verify the aircraft position by analyzing physical signals or to modify the current protocol structure to prevent attacks,which is difficult to implement.Based on the characteristics of fast message update and strong time correlation,this paper proposes an ADS-B anomaly trajectory clustering detection method based on multi-dimensional distance from the perspective of ADS-B data.According to the data format of 1090 mhz S-mode message,the ADS-B data is regarded as track by selecting the relevant characteristics of flight data,and an ADS-B abnormal track detection method based on multi-dimensional distance clustering is proposed.Firstly,the Hausdorff distance formula is improved to calculate the multi-dimensional feature similarity of ADS-B trajectory data,and the similarity matrix between the tracks is obtained.Secondly,the multi-dimensional Hausdorff distance is applied to the hierarchical clustering algorithm to replace the original Euclidean distance formula,and the anomaly of ADS-B data in the trajectory is identified by clustering method.Experimental results show that the accuracy of abnormal behavior detection of ADS-B trajectory data is improved by increasing the motion and direction characteristics of trajectory points.For various types of abnormal attacks on ADS-B system,in order to automatically learn the feature expression of ADS-B data from the data,an ADS-B data anomaly detection algorithm based on LSTM-VAE is proposed.In this algorithm,the message sequence data is segmented in the form of sliding window,and the state weakening LSTM mechanism isintroduced to learn the relationship between ADS-B data sequences.Using VAE to learn the potential distribution of eigenvectors and generate new reconstruction data from potential spatial sampling;for abnormal data with extreme behavior(such as position drift,sharp change of speed or sharp turn),the poor model fitting ability makes the difference between reconstruction data and input large.According to the difference,the anomaly discrimination of ADS-B data can be realized.Through the simulation of four attacks,it is verified that this method can detect four attacks with high accuracy and low false alarm rate. |
