Research on Aircraft System Anomaly Detection Method Based on MHAVGAE Model

With the increasing complexity of aircraft systems, there are different degrees of correlation in monitoring aircraft systems, which brings great challenges to anomaly detection in complex systems. In this paper, the method of anomaly detection for complex aircraft systems is studied, and an anomaly detection model based on variational graph autoencoder considering multi-variable coupling is established. Firstly, the detrended correlation quantification system is used to monitor the coupling relationship between variables, and the coupling relationship network is constructed. Secondly, the multi-head attention mechanism is introduced into the variogram autoencoder anomaly detection model to fully extract the hidden features of the network nodes, and quantitatively analyze the changes of the coupling relationship network during the operation of the aircraft system to accurately characterize the system health state. The reconstruction probability is taken as the evaluation index of system running state, and the threshold value of fault warning is determined by the criterion of Gauss distribution theory to realize the timely detection of system anomalies.