Root-cause analysis for time-series anomalies via spatiotemporal graphical modeling in distributed complex systems

Performance monitoring, anomaly detection, and root-cause analysis in complex cyber–physical systems (CPSs) are often highly intractable due to widely diverse operational modes, disparate data types, fault propagation mechanisms. This paper presents a new data-driven framework for analysis, based on spatiotemporal graphical modeling approach built the concept of symbolic dynamics discovering representing causal interactions among sub-systems CPSs. We formulate problem as minimization via proposed inference metric present two approximate approaches namely sequential state switching (S3, free energy restricted Boltzmann machine, RBM) artificial association (A3, classification using deep neural networks, DNN). Synthetic from cases with failed pattern(s) anomalous node(s) simulated validate approaches. Real dataset Tennessee Eastman process (TEP) is also used comparison other The results show that: (1) S3 A3 can obtain high accuracy under both pattern-based node-based scenarios, addition successfully handling multiple nominal operating (2) tool-chain shown be scalable while maintaining accuracy, (3) robust adaptive different conditions performs better state-of-the-art methods.