Anomaly detection refers to the problem of finding patterns in data that do not conform to expected behavior. Nowadays anomaly detection method deployed to production is based on reconstruction error generated by LSTM sequence modeling. Recently, the remarkable improvement achieved by BERT model in language translation demonstrated that the self-attention-based transformer is superior to LSTM models, due to its ignoring distance. In this paper, we continue the research on transformer and propose a transformer-based architecture, masked time series modeling, applying transformer in data stream, which has two novel components (1) the attention mechanism used for updating timestep in parallel and (2) the mask strategy used to detect the anomaly in advanced time. We compared the performances of our method with state-of-the-art AD methods on challenging public NASA telemetry dataset. The experiment results demonstrated our method saves about 80% time cost because of parallel computing compared with LSTM methods and achieves 0.78 F1 point-based score, moreover achieving a better score on range-based indicators.
Spacecraft Anomaly Detection via Transformer Reconstruction Error
Lect. Notes Electrical Eng.
International Conference on Aerospace System Science and Engineering ; 2019 ; Toronto, ON, Canada July 30, 2019 - August 01, 2019
Proceedings of the International Conference on Aerospace System Science and Engineering 2019 ; Kapitel : 28 ; 351-362
2020-03-01
12 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Spacecraft , Anomaly detection , Transformer , Mask , Deep learning , Attention mechanism Physics , Space Sciences (including Extraterrestrial Physics, Space Exploration and Astronautics) , Aerospace Technology and Astronautics , Control and Systems Theory , Vibration, Dynamical Systems, Control , Engineering Design , Physics and Astronomy
Spacecraft Anomaly Detection via Transformer Reconstruction Error
British Library Conference Proceedings | 2020
|NTRS | 2021
|NOAA's spacecraft anomaly data base
AIAA | 1990
|