Optimization of monitoring and early warning technology for mine water disasters using microservices and long short-term memory algorithm

The Journal of Supercomputing, Год журнала: 2025, Номер 81(4)

Опубликована: Фев. 24, 2025

Язык: Английский

TiTAD: Time-Invariant Transformer for Multivariate Time Series Anomaly Detection

Electronics, Год журнала: 2025, Номер 14(7), С. 1401 - 1401

Опубликована: Март 31, 2025

Anomaly detection in multivariate time series data is critical for industrial sectors such as manufacturing and aerospace. While existing methods have achieved notable success specific scenarios, they often narrowly focus on either the temporal or spatial dimensions while overlooking their complex interdependencies. Furthermore, these approaches tend to neglect time-invariant characteristics that are crucial accurately capturing spatio-temporal dynamics of series. To address limitations, this paper introduces Time-invariant Transformer Multivariate Time Series Detection (TiTAD), a novel framework synergizes invariance with modeling. TiTAD leverages Transformer, component excels at extracting both features by incorporating an augmented memory mechanism. This mechanism enhances anomaly identification robustness through synergistic integration heterogeneous feature sets. Additionally, mitigates Transformer’s tendency lose sequence information use Gated Recurrent Unit (GRU), thereby further enhancing model’s capability discern patterns. The inclusion Feature Fusion module within serves refine extracted adjusting weights minimizing redundancy, ensuring most relevant utilized prediction detection. Empirical evaluation three industrial-scale benchmarks (SWaT, WADI, SMD) demonstrates TiTAD’s superior performance compared other methods.

Язык: Английский

MSDG: Multi-Scale Dynamic Graph Neural Network for Industrial Time Series Anomaly Detection

Sensors, Год журнала: 2024, Номер 24(22), С. 7218 - 7218

Опубликована: Ноя. 12, 2024

A large number of sensors are typically installed in industrial plants to collect real-time operational data. These monitor data with time series correlation and spatial over time. In previous studies, GNN has built many successful models deal data, but most these have fixed perspectives struggle capture the dynamic correlations space simultaneously. Therefore, this paper constructs a multi-scale graph neural network (MSDG) for anomaly detection sensor First, sliding window mechanism is proposed input different scale into corresponding network. Then, constructed spatial–temporal dependencies multivariate Finally, model comprehensively considers extracted features sequence reconstruction utilizes errors detection. Experiments been conducted on three real public datasets, results show that method outperforms mainstream methods.

Язык: Английский