What Streaming Anomaly Detection Finds (and Misses) in Industrial Time Series

2026-09-30

AI Summary

This study addresses the absence of a universally optimal approach for industrial time-series streaming anomaly detection by conducting an empirical investigation using real-world data from nuclear power plants. Methodologically, it integrates unsupervised learning with ensemble techniques to systematically compare the automated detection performance of streaming methods against online time-series anomaly detection (TSAD) models. The results demonstrate that online TSAD models achieve superior detection consistency, while the proposed ensemble strategy exhibits strong robustness. Overall, this work provides empirical evidence to guide the selection between streaming and online anomaly detection approaches in industrial scenarios, further validating the effectiveness of ensemble learning in enhancing detection stability.

Abstract

EDF relies on continuous monitoring of its power plants to detect anomalies as soon as they occur. Given the absence of a universally optimal streaming method in unsupervised settings, we compare streaming methods with state-of-the-art TSAD models deployed online on a real nuclear power plant dataset. This work also evaluates Automated Anomaly Detection in a streaming context. Results show higher consistency for online TSAD and strong robustness from ensembling strategies.

Problem

• streaming anomaly detection
• industrial time series
• unsupervised learning
• time series anomaly detection
• automated anomaly detection

Innovation

• streaming anomaly detection
• time series anomaly detection
• automated anomaly detection
• ensemble strategies
• industrial time series