Scalable Anomaly Detection for Smart City Infrastructure Networks

Abstract

Dynamically detecting anomalies can be difficult in very large-scale infrastructure networks. The authors' approach addresses spatiotemporal anomaly detection in a smarter city context with large numbers of sensors deployed. They propose a scalable, hybrid Internet infrastructure for dynamically detecting potential anomalies in real time using stream processing. The infrastructure enables analytically inspecting and comparing anomalies globally using large-scale array processing. Deployed on a real pipe network topology of 1,891 nodes, this approach can effectively detect and characterize anomalies while minimizing the amount of data shared across the network.