Industrial time series determinative anomaly detection based on constraint hypergraph

The explosive growth of time series captured by sensors in industrial pipelines gives rise to the flourish of intelligent industry. Exploiting the value of these time series is conductive to workload balancing and production optimization. Unfortunately, knowledge obtained from the mining process turns out to be insufficient for use due to widespread anomalies, indicating machine breakdown, sensor failure or working status shifts. To tackle this problem, we propose a constraint hypergraph-based method, combining multiple constraints for anomaly detection. We develop strategies for adaptive determinative anomaly detection and anomaly pattern mining. We also investigate the problem of Anomaly Pattern Matching, prove its NP-completeness, and propose algorithms to obtain its global and local optimum. Finally, we demonstrate our approach with three real world datasets from a real powerplant, a chemical production pipeline and a hydraulic system. The experimental results show that our approach can effectively and efficiently work under different circumstances. •The first constraint hypergraph-based anomaly detection method for time series.•A size-adaptive vertex mincover strategy for determinative anomaly detection.•A Dynamic Programming algorithm to match anomaly patterns.•The algorithms outperform traditional algorithm with their own suitable scenarios.