Method for detecting anomalies in geomagnetic field variations based on artificial neural network
The paper proposes a method for anomaly detection in geomagnetic data based on the classical autoencoder architecture. The training data consisted of daily variations in the geomagnetic field on quiet days for 2020, 2021, and 2022, collected from the Ak-Suu base station of the geomagnetic monitori...
Saved in:
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Дальневосточного отделения Российской академии наук, Южно-Сахалинск, Федеральное государственное бюджетное учреждение науки Институт морской геологии и геофизики
2024-12-01
|
| Series: | Геосистемы переходных зон |
| Subjects: | |
| Online Access: | http://journal.imgg.ru/web/full/f2024-4-6.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832591956961656832 |
|---|---|
| author | Imashev, Sanjar A. |
| author_facet | Imashev, Sanjar A. |
| author_sort | Imashev, Sanjar A. |
| collection | DOAJ |
| description | The paper proposes a method for anomaly detection in geomagnetic data based on the classical
autoencoder architecture. The training data consisted of daily variations in the geomagnetic field on quiet days
for 2020, 2021, and 2022, collected from the Ak-Suu base station of the geomagnetic monitoring network of the
Research Station of the Russian Academy of Sciences in Bishkek. The neural network has five hidden layers with
a total of ~3.5‧106
trainable parameters. The trained model accurately reproduces typical features of normal data,
whereas in the presence of anomalies it shows a decline in reconstruction quality. This property of the
autoencoder was used to classify the data into two categories: normal and anomalous. The reconstruction error,
measured as the Mean Absolute Error (MAE), was used as the anomaly metric. In particular, the MAE value of
0.109 was used as the threshold for class separation. Testing the model on the data from the Ak-Suu station for
2017, 2018, and 2019 demonstrated good results. Binary classification metrics such as recall and F1-score were
notably high: 0.965 and 0.918 for the 2017 data, 0.982 and 0.933 for the 2018 data, and 0.970 and 0.935 for the
2019 data, respectively. |
| format | Article |
| id | doaj-art-68c1a964db6242ce882723db9feb0d17 |
| institution | Kabale University |
| issn | 2541-8912 2713-2161 |
| language | English |
| publishDate | 2024-12-01 |
| publisher | Дальневосточного отделения Российской академии наук, Южно-Сахалинск, Федеральное государственное бюджетное учреждение науки Институт морской геологии и геофизики |
| record_format | Article |
| series | Геосистемы переходных зон |
| spelling | doaj-art-68c1a964db6242ce882723db9feb0d172025-01-22T02:41:08ZengДальневосточного отделения Российской академии наук, Южно-Сахалинск, Федеральное государственное бюджетное учреждение науки Институт морской геологии и геофизикиГеосистемы переходных зон2541-89122713-21612024-12-018343356https://doi.org/10.30730/gtrz.2024.8.4.343-356Method for detecting anomalies in geomagnetic field variations based on artificial neural networkImashev, Sanjar A.0https://orcid.org/0000-0003-3293-3764Research Station of the Russian Academy of Sciences in Bishkek, Bishkek city, KyrgyzstanThe paper proposes a method for anomaly detection in geomagnetic data based on the classical autoencoder architecture. The training data consisted of daily variations in the geomagnetic field on quiet days for 2020, 2021, and 2022, collected from the Ak-Suu base station of the geomagnetic monitoring network of the Research Station of the Russian Academy of Sciences in Bishkek. The neural network has five hidden layers with a total of ~3.5‧106 trainable parameters. The trained model accurately reproduces typical features of normal data, whereas in the presence of anomalies it shows a decline in reconstruction quality. This property of the autoencoder was used to classify the data into two categories: normal and anomalous. The reconstruction error, measured as the Mean Absolute Error (MAE), was used as the anomaly metric. In particular, the MAE value of 0.109 was used as the threshold for class separation. Testing the model on the data from the Ak-Suu station for 2017, 2018, and 2019 demonstrated good results. Binary classification metrics such as recall and F1-score were notably high: 0.965 and 0.918 for the 2017 data, 0.982 and 0.933 for the 2018 data, and 0.970 and 0.935 for the 2019 data, respectively.http://journal.imgg.ru/web/full/f2024-4-6.pdfanomalygeomagnetic fieldvariational seriesneural networkautoencoderconfusion matrix |
| spellingShingle | Imashev, Sanjar A. Method for detecting anomalies in geomagnetic field variations based on artificial neural network Геосистемы переходных зон anomaly geomagnetic field variational series neural network autoencoder confusion matrix |
| title | Method for detecting anomalies in geomagnetic field variations
based on artificial neural network |
| title_full | Method for detecting anomalies in geomagnetic field variations
based on artificial neural network |
| title_fullStr | Method for detecting anomalies in geomagnetic field variations
based on artificial neural network |
| title_full_unstemmed | Method for detecting anomalies in geomagnetic field variations
based on artificial neural network |
| title_short | Method for detecting anomalies in geomagnetic field variations
based on artificial neural network |
| title_sort | method for detecting anomalies in geomagnetic field variations based on artificial neural network |
| topic | anomaly geomagnetic field variational series neural network autoencoder confusion matrix |
| url | http://journal.imgg.ru/web/full/f2024-4-6.pdf |
| work_keys_str_mv | AT imashevsanjara methodfordetectinganomaliesingeomagneticfieldvariationsbasedonartificialneuralnetwork |