Streaming methods improve anomaly detection in nuclear plants

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

Machine LearningArtificial Intelligence

Summary

Detecting unusual events quickly in power plant data is important for safety and efficiency. The authors compared several streaming anomaly detection methods on data from a real nuclear power plant. They found that online methods designed for time series give more consistent results. Combining multiple methods also made the detection more reliable. This study helps identify which streaming techniques work best for monitoring complex industrial systems.

What this means in practice

  • For industrial plant operators: Improve real-time detection of faults and anomalies in industrial sensor data streams using more consistent online time series anomaly detection methods.
  • For energy grid monitoring teams: Deploy ensemble-based streaming anomaly detection tools to enhance robustness of anomaly alerts across power generation assets.

Tested on one dataset.

Authors

Magali Parrino, Antoine Ajenjo, Emmanuel Remy, Pierre Stephan, Paul Boniol

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.