Achieving efficient flight data anomaly detection is a key technology for monitoring aircraft flight state and identifying unsafe flight behavior and is a basic requirement for intelligent flight control of future aircraft. Flight data belongs to multivariate time series, which has inherent attributes such as long series, many variables, lack of labels, unbalanced distribution of abnormal and normal samples, etc. At the same time, the high accuracy requirement of flight data abnormality detection and the limitation of computing resources bring great challenges for practical applications.
To address these issues, this paper proposes an unsupervised flight data anomaly detection method based on Transformer and adversarial training. This method is based on the Transformer architecture, uses the attention mechanism to model long-term dependencies in the time dimension, and introduces a convolutional neural network block to better capture local contextual features within the sequence window. Adopting two-stage adversarial training with a loss constraint mechanism enhances the stability and generalization of the model.
This paper validates the model using real-world datasets, including MSL, SMAP, and SMD. The results show that the detection effect is better than LSTM-VAE, MAD-GAN, and OmniAnomaly. In addition, the model is trained and tested using QAR data. The experimental results show that the anomaly detection method proposed in this paper can accurately locate the time of anomaly occurrence in-flight data with a prediction accuracy of 0.9017.
Transformer-Based Method for Unsupervised Anomaly Detection of Flight Data
Lect. Notes Electrical Eng.
Asia-Pacific International Symposium on Aerospace Technology ; 2023 ; Lingshui, China October 16, 2023 - October 18, 2023
2023 Asia-Pacific International Symposium on Aerospace Technology (APISAT 2023) Proceedings ; Kapitel : 142 ; 1816-1826
02.07.2024
11 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Proximity landing stage anomaly detection method based on flight data
Europäisches Patentamt | 2025
|UNSUPERVISED ANOMALY DETECTION FOR AUTONOMOUS VEHICLES
Europäisches Patentamt | 2021
|UNSUPERVISED ANOMALY DETECTION FOR AUTONOMOUS VEHICLES
Europäisches Patentamt | 2021
|Unsupervised anomaly detection for autonomous vehicles
Europäisches Patentamt | 2023
|