Anomaly Detection Method for Multivariate Time Series Data of Oil and Gas Stations Based on Digital Twin and MTAD-GAN

Due to the complexity of oil and gas station system, operational data, with various temporal dependencies inter-metric dependencies, has characteristics diverse patterns, variable working conditions imbalance, which brings great challenges multivariate time series anomaly detection. Moreover, time-series reconstruction information data from digital twin space can be used identify interpret anomalies. Therefore, this paper proposes a twin-driven MTAD-GAN (Multivariate Time Series Data Anomaly Detection GAN) detection method. Firstly, framework consisting model, virtual-real synchronization algorithm, strategy realistic is constructed, an efficient mapping achieved by embedding stochastic Petri net (SPN) describe station-operating logic behavior. Secondly, based on potential correlation complementarity among variables, we present method reconstruct error combining mechanism knowledge graph attention Hawkes judge abnormal samples given threshold. The experimental results show that achieve accurate identification anomalous complex performance improved about 2.6% compared other methods machine learning deep learning, proves effectiveness