Survey of time series anomaly detection for industrial sensor networks

The deep integration of industrial control systems and information networks drives the trend towards networking and intelligence in future industrial development. Industrial sensor networks, crucial for industrial system networking, raise concerns in industrial security, particularly regarding data security. Anomalies in industrial sensor network data impact the physical, information, and network security of industrial control systems. Industrial sensor network anomaly detection, addressing network attacks and physical faults, involves analyzing complex, multi-layered, and multi-scale sensor time series data to discover hidden anomalous logic and fault causes. The causes of anomalies in industrial sensor networks were summarized, research progress in industrial sensor network anomaly detection was reviewed systematically, and key technologies and typical methods were explained categorically from three perspectives: time series features, spatiotemporal multiscale, and non-structured graph representation. The developmental trajectories and major breakthroughs of various existing methods were analyzed and consolidated. Datasets and evaluation metrics currently used for industrial sensor networks were introduced, the detection performance of existing methods was summarized, and through comparative analysis of experimental results, the characteristics and technical focuses of each method were highlighted. The application prospects of existing work were pointed out and the challenges faced by current anomaly detection methods in practical applications were outlined. Future development trends and research directions for industrial sensor network anomaly detection were suggested.