Anomaly Detection for Aircraft Based on Multivariate LSTM and Predictive Residual Test
Qingzhen Zhang, Wei Wu, Gongcheng Zhou, Gang Xiang, Ruishi Lin, Langfu Cui, Dongpeng Li, Yu Peng, Guizhen Yu · 2023
With the increasing complexity and high reliability requirements of aircraft, researching more accurate anomaly detection methods and real-time monitoring of aircraft status is of great significance for ensuring equipment safety. This paper proposes a method based on multivariate LSTM and residual testing for anomaly detection in high-dimensional complex data. Firstly, by utilizing maximum mutual information to analyze the correlation of multidimensional data, a multivariate LSTM model is established based on high correlation parameters to predict the parameters. Then, combined with the idea of statistical hypothesis testing, test the statistical distribution of residuals on the training data of the prediction model using Kolmogorov-Smirnov method, and the dynamic threshold for anomaly detection is obtained through significance testing. Finally, real-time calculation of residuals between online monitoring data and predicted values is calculated, and anomaly detection is achieved by comparing the residual with threshold. The method proposed in this paper exhibits excellent anomaly detection performance on real flight data of aircraft.