Autonomous Anomaly Detection for Continuous Streams: Enhancing Nuclear Reactor Monitoring
Discretizes real-time signals, handles caching of historical data stream, and uses a unsupervised machine learning model, Isolation Forest, to autonomously flag anomalies
The Autonomous Anomaly Detection for Continuous Stream (AADCS) software utilizes an unsupervised machine learning algorithm, Isolation Forest, to autonomously detect anomalies in nuclear reactor operations. By monitoring real-time data from a digital twin of the AGN-201 nuclear reactor, the software identifies operational deviations that may signal safety, security, or safeguard risks. It operates without the need for labeled data, providing a lightweight yet effective complement to traditional physics-based anomaly detection methods, thereby enhancing nuclear facility monitoring and safety.
The software implements the Isolation Forest algorithm within the digital twin, capturing operational data such as control rod positions, reactor power, and temperature. By detecting patterns that deviate from expected behavior, the algorithm assigns anomaly scores based on the isolation of rare events. Designed for nuclear facility operators, regulatory bodies like the IAEA, and research institutions, this tool offers real-time anomaly detection, enhances operational safety, aids in regulatory compliance, and supports researchers in improving reactor safety protocols.
For questions regarding AADCS, please click the "Contact Us" button.
This software is available at no cost. For access, please visit our Github Repository: https://github.com/IdahoLabResearch/AcCCS