| Publikationstyp: | Konferenz: Paper |
| Art der Begutachtung: | Peer review (Publikation) |
| Titel: | Deep Gaussian mixture model : a novelty detection method for time series |
| Autor/-in: | Brunner, Stefan Frischknecht-Gruber, Carmen Reif, Monika Ulrike Senn, Christoph |
| et. al: | No |
| DOI: | 10.3850/978-981-18-5183-4_R22-12-325-cd |
| Tagungsband: | Proceedings of the 32nd European Safety and Reliability Conference |
| Herausgeber/-in des übergeordneten Werkes: | Leva, Maria Chiara Patelli, Edoardo Podofillini, Luca Wilson, Simon |
| Seite(n): | 1291 |
| Seiten bis: | 1298 |
| Angaben zur Konferenz: | 32nd European Safety and Reliability Conference (ESREL 2022), Dublin, Ireland, 28 August - 1 September 2022 |
| Erscheinungsdatum: | 2022 |
| Verlag / Hrsg. Institution: | Research Publishing |
| Verlag / Hrsg. Institution: | Singapore |
| ISBN: | 978-981-18-5183-4 |
| Sprache: | Englisch |
| Schlagwörter: | Novelty detection; Time series; Anomaly detection; Deep learning |
| Fachgebiet (DDC): | 510: Mathematik |
| Zusammenfassung: | Safety-critical systems require continuous monitoring of operations in order to detect potential abnormal behavior at an early stage. This is done for personnel safety reasons, to prevent environmental damage and to avoid system downtimes or minimize them respectively. During system monitoring, incoming sensor signals are processed and analyzed. In the event of deviating behaviour, a decision-making system is employed to assess an appropriate response. If a system fails to react or responds too late when required, this can result in human injuries, environmental and in significant financial damage. Whereby it should be noted that erroneous alarms can also lead to damage, such as unnecessary downtimes. Usually there is a decent knowledge about the normal state of a system, while often there is little known about previously unseen and therefore unknown states. Novelty detection as the name implies is concerned with the detection of previously unknown data, identifying whether the data lies inside or outside of the desired range and therefore represents a novelty or an anomaly. This paper discusses the performance of different novelty detection methods for stationary and non-stationary time series. Depending on the observed time series, the advantages and disadvantages are analyzed and outlined. We present the Deep Gaussian Mixture Model (DGMM) for novelty detection, which allows to predict unknown as well as known states in time series. The DGMM consists of a combination of an under-complete autoencoder with multiple hidden layers and a Gaussian mixture model (GMM), using the capabilities offered by each method for the respective components of the time series. The obtained results on both synthetic and original data for stationary and non-stationary time series demonstrate the benefits of the DGMM in novelty detection. |
| URI: | https://digitalcollection.zhaw.ch/handle/11475/27676 |
| Volltext Version: | Publizierte Version |
| Lizenz (gemäss Verlagsvertrag): | Lizenz gemäss Verlagsvertrag |
| Departement: | School of Engineering |
| Organisationseinheit: | Institut für Angewandte Mathematik und Physik (IAMP) |
| Enthalten in den Sammlungen: | Publikationen School of Engineering |
Dateien zu dieser Ressource:
Es gibt keine Dateien zu dieser Ressource.
Zur Langanzeige
Brunner, S., Frischknecht-Gruber, C., Reif, M. U., & Senn, C. (2022). Deep Gaussian mixture model : a novelty detection method for time series [Conference paper]. In M. C. Leva, E. Patelli, L. Podofillini, & S. Wilson (Eds.), Proceedings of the 32nd European Safety and Reliability Conference (pp. 1291–1298). Research Publishing. https://doi.org/10.3850/978-981-18-5183-4_R22-12-325-cd
Brunner, S. et al. (2022) ‘Deep Gaussian mixture model : a novelty detection method for time series’, in M.C. Leva et al. (eds) Proceedings of the 32nd European Safety and Reliability Conference. Singapore: Research Publishing, pp. 1291–1298. Available at: https://doi.org/10.3850/978-981-18-5183-4_R22-12-325-cd.
S. Brunner, C. Frischknecht-Gruber, M. U. Reif, and C. Senn, “Deep Gaussian mixture model : a novelty detection method for time series,” in Proceedings of the 32nd European Safety and Reliability Conference, 2022, pp. 1291–1298. doi: 10.3850/978-981-18-5183-4_R22-12-325-cd.
BRUNNER, Stefan, Carmen FRISCHKNECHT-GRUBER, Monika Ulrike REIF und Christoph SENN, 2022. Deep Gaussian mixture model : a novelty detection method for time series. In: Maria Chiara LEVA, Edoardo PATELLI, Luca PODOFILLINI und Simon WILSON (Hrsg.), Proceedings of the 32nd European Safety and Reliability Conference. Conference paper. Singapore: Research Publishing. 2022. S. 1291–1298. ISBN 978-981-18-5183-4
Brunner, Stefan, Carmen Frischknecht-Gruber, Monika Ulrike Reif, and Christoph Senn. 2022. “Deep Gaussian Mixture Model : A Novelty Detection Method for Time Series.” Conference paper. In Proceedings of the 32nd European Safety and Reliability Conference, edited by Maria Chiara Leva, Edoardo Patelli, Luca Podofillini, and Simon Wilson, 1291–98. Singapore: Research Publishing. https://doi.org/10.3850/978-981-18-5183-4_R22-12-325-cd.
Brunner, Stefan, et al. “Deep Gaussian Mixture Model : A Novelty Detection Method for Time Series.” Proceedings of the 32nd European Safety and Reliability Conference, edited by Maria Chiara Leva et al., Research Publishing, 2022, pp. 1291–98, https://doi.org/10.3850/978-981-18-5183-4_R22-12-325-cd.
Alle Ressourcen in diesem Repository sind urheberrechtlich geschützt, soweit nicht anderweitig angezeigt.