Skip to Main content Skip to Navigation
Journal articles

Anomaly Detection for ICS Based on Deep Learning: A Use Case for Aeronautical Radar Data

Abstract : Industrial Control Systems (ICS) are no longer restricted to industrial production. They are also at the heart of safety-critical systems and carry out key information that require strong need in terms of availability and integrity. Furthermore, they are gradually connected with the Internet. In the context of Air Traffic Management, safety critical data are generally time series which contain periodic events. Anomalies can hardly be detected as we only have a little knowledge of the traffic characteristic and the kind of anomalies we might encounter. Consequently, detecting them is challenging as it requires high detection accuracy currently unfeasible with traditional methods based on anomaly signatures or predictions. To cope with this issue, we introduce an anomaly detection method for ICS based on Long Short Term Memory (LSTM) that outperforms the accuracy of traditional ones. We experiment and develop our method with one major dataset containing French civil radar aviation data. We then evaluate our scheme with different datasets containing ICS monitoring data (publicly available predictable time series data) and show that our autoencoder can detect anomalies from predictable times series and present a higher detection rate on average than traditional detection methods.
Document type :
Journal articles
Complete list of metadata
Contributor : Emmanuel Lochin Connect in order to contact the contributor
Submitted on : Friday, January 28, 2022 - 5:02:41 PM
Last modification on : Monday, January 31, 2022 - 11:39:39 AM
Long-term archiving on: : Friday, April 29, 2022 - 9:41:00 PM


Files produced by the author(s)




Theobald de Riberolles, Yunkai Zou, Guthemberg Silvestre, Emmanuel Lochin, Jiefu Song. Anomaly Detection for ICS Based on Deep Learning: A Use Case for Aeronautical Radar Data. Annals of Telecommunications - annales des télécommunications, Springer, 2022, ⟨10.1007/s12243-021-00902-7⟩. ⟨hal-03533871⟩



Record views


Files downloads