Anomaly detection in multidimensional time series for water injection pump operations based on LSTMA-AE and mechanism constraints
Abstract Addressing the issues of inadequate information exchange among subsequences in the operational time series of water injection pumps, leading to low accuracy and high false alarm rates in anomaly detection, this paper proposes a multidimensional time series anomaly detection method for water...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-01-01
|
| Series: | Scientific Reports |
| Subjects: | |
| Online Access: | https://doi.org/10.1038/s41598-025-85436-x |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Addressing the issues of inadequate information exchange among subsequences in the operational time series of water injection pumps, leading to low accuracy and high false alarm rates in anomaly detection, this paper proposes a multidimensional time series anomaly detection method for water injection pump operations, leveraging Long Short-Term Memory Autoencoder augmented with Attention Mechanism (LSTMA-AE) and mechanistic constraints. The LSTMA-AE framework encompasses three primary modules: a Time Feature Extraction Module (Encoder), an Attention Layer, and a Data Reconstruction Module (Decoder). The Encoder captures temporal dependencies and features within the input sequences, mapping the input data into a higher-dimensional space. The Attention Layer, embedded within the hidden state computation, dynamically adjusts the contribution of each timestep’s input information to the hidden state, thereby enhancing the extraction of vital information while ignoring irrelevant data. The Decoder is responsible for reconstructing the latent representations generated by the Encoder back into the original time series data. By utilizing LSTMA-AE, we aim to improve the accuracy of anomaly detection, while simultaneously employing mechanistic constraints to mitigate false alarm rates. Experimental results demonstrate that this approach significantly outperforms methods such as polynomial interpolation, random forest, and LSTM-AE in terms of anomaly detection accuracy on field datasets from oilfields, accompanied by a notably lower false alarm rate. |
|---|---|
| ISSN: | 2045-2322 |