Abstract:Accurate detection and diagnosis of abnormal behaviors such as network attacks from multivariate time series (MTS) are crucial for ensuring the stable and effective operation of industrial cyber-physical systems (CPS). However, existing researches pay little attention to the logical dependencies among system working states, and have difficulties in explaining the evolution mechanisms of abnormal signals. To reveal the spatio-temporal association relationships and evolution mechanisms of the working states of industrial CPS, this paper proposes a fine-grained adaptive anomaly diagnosis method (i.e. MAD-Transformer) to identify and diagnose anomalies in MTS. MAD-Transformer first constructs a temporal state matrix to characterize and estimate the change patterns of the system states in the temporal dimension. Then, to better locate the anomalies, a spatial state matrix is also constructed to capture the inter-sensor state correlation relationships within the system. Subsequently, based on these two types of state matrices, a three-branch structure of series-temporal-spatial attention module is designed to simultaneously capture the series, temporal, and space dependencies among MTS. Afterwards, three associated alignment loss functions and a reconstruction loss are constructed to jointly optimize the model. Finally, anomalies are determined and diagnosed by comparing the residual matrices with the original matrices. We conducted comparative experiments on five publicly datasets spanning three application domains (service monitoring, spatial and earth exploration, and water treatment), along with a petroleum refining simulation dataset collected by ourselves. The results demonstrate that MAD-Transformer can adaptively detect fine-grained anomalies with short duration, and outperforms the state-of-the-art baselines in terms of noise robustness and localization performance.

Hankel Matrix Factorization for Tagged Time Series to Recover Missing Values During Blackouts

HKMF-T: Recover from Blackouts in Tagged Time Series with Hankel Matrix Factorization

HTF: An Effective Algorithm for Time Series to Recover Missing Blocks

Temporal Difference Learning Based Critical Component Identifying Method with Cascading Failure Data in Power Systems

Bayesian High-Rank Hankel Matrix Completion for Nonlinear Synchrophasor Data Recovery

Missing Data Imputation in GNSS Monitoring Time Series Using Temporal and Spatial Hankel Matrix Factorization

Power Grid Missing Data Filling Method Based on Historical Data Mining Assisted Multi-dimensional Scenario Analysis

DBT-DMAE: An Effective Multivariate Time Series Pre-Train Model under Missing Data

Precise Recovery of Corrupted Synchrophasors Based on Autoregressive Bayesian Low-Rank Factorization and Adaptive K-Medoids Clustering

Graph Spatiotemporal Process for Multivariate Time Series Anomaly Detection with Missing Values

Time Series Data-Driven Online Prognosis of Wind Turbine Faults in Presence of SCADA Data Loss

Recovering Multiple Nonnegative Time Series From a Few Temporal Aggregates

Robust Multi-Dimensional Time Series Forecasting

An End-to-End Model for Time Series Classification In the Presence of Missing Values

Unsupervised Spatio-Temporal State Estimation for Fine-grained Adaptive Anomaly Diagnosis of Industrial Cyber-physical Systems

Probabilistic Temporal Fusion Transformers for Large-Scale KPI Anomaly Detection

Cross-modal missing time-series imputation using dense spatio-temporal transformer nets

Bayesian Temporal Factorization for Multidimensional Time Series Prediction

Multivariate Time Series Anomaly Detection with Fourier Time Series Transformer

BayOTIDE: Bayesian Online Multivariate Time series Imputation with functional decomposition