Email this article Intelligent segmentation of anomalies in oceanographic data: comparative analysis of BiLSTM and TCN on time and time-frequency representations
- Authors: Korotchenko R.A.1, Kosheleva A.V.1
-
Affiliations:
- V. I. Il’ichev Pacific Oceanological Institute, Far Eastern Branch, Russian Academy of Sciences
- Issue: Vol 32, No 6 (2026)
- Pages: 300-308
- Section: Neural network technologies
- Published: 09.06.2026
- URL: https://journals.eco-vector.com/1684-6400/article/view/708389
- DOI: https://doi.org/10.17587/it.32.300-308
- ID: 708389
Cite item
Open Access
Access granted
Abstract
This paper concerns the use of Deep Learning methods and Neural Networks for the task of segmentation and classification of hydrophysical time series in order to identify anomalies of a certain type caused by hardware failures. Two models based on Bidirectional Long-Short Term Memory networks and one based on a Temporal Convolutional Network were used. We made the performance comparisons and assessed the detection success. The results obtained demonstrate the effectiveness of the proposed approaches, and it was found that time-frequency analysis significantly improves the models’ ability to detect anomalies of different types.
Full Text
About the authors
R. A. Korotchenko
V. I. Il’ichev Pacific Oceanological Institute, Far Eastern Branch, Russian Academy of Sciences
Email: kosheleva@poi.dvo.ru
Ph.D., Senior Researcher
Russian Federation, Vladivostok, 690041A. V. Kosheleva
V. I. Il’ichev Pacific Oceanological Institute, Far Eastern Branch, Russian Academy of Sciences
Author for correspondence.
Email: kosheleva@poi.dvo.ru
Researcher
Russian Federation, Vladivostok, 690041References
- Kosheleva A. V., Lazaryuk А. Yu., Yaroshchuk I. O., Korotchenko R. A., Gulin O. E., Samchenko A. N., Pivovarov A. A., Shvyrev А. N. Acoustic-oceanological simulation of seawater hydrophysical parameters on temperature measurements in the Sea of Japan shelf zone, Vestnik Dalnevostochnogo otdeleniya Rossijskoj akademii nauk, 2018, vol. 197, no. 1. pp. 25—30 (in Russian).
- Korotchenko R. A., Kosheleva А. V. Integrated Method for Moderation of Hydrological Data Using Singular Spectrum Analysis and Neural Networks, Informacionnye tehnologii, 2023, vol. 29, no. 3. pp. 149—156, doi: 10.17587/it.29.149—156 (in Russian).
- Bruce P., Bruce A., Gedeck P. Practical Statistics for Data Scientists: 50+ Essential Concepts Using R and Python, 2-nd ed., Boston, O’Reilly Media, 2020, 360 p.
- Bendat J. S., Piersol А. G. Random Data: Analysis and Measurement Procedures, New York, John Wiley and Sons, 1986.
- Chandola V., Banerjee A., Kumar V. Anomaly detection: А survey, ACM computing surveys, 2009, vol. 41, no. 3, pp. 1—58, doi: 10.1145/1541880.1541882.
- Salvador S., Chan Ph. Learning States and Rules for Detecting Anomalies in Time Series, Applied Intelligence, 2005, vol. 23, no. 3, pp. 241—255, doi: 10.1007/s10489-005-4610-3.
- Goodfellow I., Bengio Y., Courville А. Deep Learning, Massachusetts, MIT Press, 2016, 801 p.
- Gift N. Pragmatic AI: Machine Learning and Cloud Technologies, St. Petersburg, Piter, 2019, 304 p. (in Russian).
- Malcev V. A., Murzagulov D. A., Zamyatin А. V. Detecting anomalies in process signals using an autoencoder-type neural network architecture, Avtomatizaciya v promyshlennosti, 2020, no. 10, pp. 54—58, doi: 10.25728/avtprom.2020.10.10 (in Russian).
- Beale M., Hagan M., Demuth Н. Deep Learning Toolbox. User’s Guide, MathWorks, 2024, 5456 p.
- Lea C., Flynn M. D., Vidal R., Reiter A., Hager G. D. Temporal Convolutional Networks for Action Segmentation and Detection, 2016, doi: 10.48550/arXiv.1611.05267.
- Mallat S. G. А Wavelet Tour of Signal Processing, San Diego, Academic Press, 1998, 577 p.
- Auger F., Flandrin P, Lin Y., McLaughlin S., Meignen S., Oberlin T., Wu Н. Time-frequency reassignment and synchrosqueezing: An overview, IEEE Signal Processing Magazine, 2013, vol. 30, no. 6, pp. 32—41, doi: 10.1109/MSP.2013.2265316.
- Oberlin T., Meignen S., Perrier V. The fourier-based synchrosqueezing transform, IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2014, pp. 315—319, doi: 10.1109/icassp.2014.6853609.
- Thakur G., Brevdo E., Fuckar N. S. et al. The synchrosqueezing algorithm for time-varying spectral analysis: Robustness properties and new paleoclimate applications, Signal Processing, 2013, vol. 93, no. 5, pp. 1079—1094, doi: 10.1016/j.sigpro.2012.11.029.
- Yaroshchuk I. O., Kosheleva A. V., Lazaryuk А. Yu., Dolgikh G. I., Pivovarov A. A., Samchenko A. N., Shvyrev A. N., Gulin O. E., Korotchenko R. A. Estimation of Seawater Hydrophysical Characteristics from Thermistor Strings and CTD Data in the Sea of Japan Shelf Zone, Journal of Marine Science and Engineering, 2023, vol. 11, no. 6, art. no. 1204, doi: 10.3390/jmse11061204.
Supplementary files
Supplementary Files
Action
1.
JATS XML
2.
Fig. 1. A fragment of a synthetic time series saturated with “flash” anomalies
Download (96KB)
3.
Fig. 2. FSST spectrogram of a fragment of a time series containing anomalies
Download (126KB)
4.
Fig. 3. BiLSTM network diagram
Download (68KB)
5.
Rice. 4. TCN-FSST network diagram
Download (506KB)
6.
Fig. 5. Classification results. Error matrix: a — BiLSTM-TS; b — BiLSTM-FSST; c — TCN-FSST; d — TCN-FSST with weight balancing
Download (133KB)
7.
Fig. 6. Processing of test time series by TCN-FSST network with weight balancing
Download (156KB)
