Discovering Multidimensional Time Series Anomalies of a Circulating Water System

To maximize plant availability, complex systems such as nuclear power plants continuously monitor and collect time series data of many components, assets, and systems. This data can provide system engineers with insights into anomalous system behaviors. In recent years, there has been significant progress in time series anomaly detection due to advancements in machine learning and artificial intelligence. However, several challenges have proved to be roadblocks to identifying multidimensional time series anomalies, including high dimensionality in nature and and temporal dependencies across different dimensions. Recent developments in matrix profile correlations shows great promise in discovering multidimensional time series anomalies. In this paper, we propose using matrix profile and model-based system engineering (MBSE) models to discover temporal correlations and dependencies across different dimensions of time series from a nuclear power plant circulating water system. First, we will employ a K-dimensional-profile algorithm based on matrix profile to identify the best K of N anomaly subset with K<N. This assumes that the anomaly will generally only manifest itself on K of the N-dimensional time series. Second, we will study the temporal correlations among the identified anomalies. Finally, we will perform a preliminary study on utilizing MBSE model of circulating water systems for temporal dependencies and root causes analyses. Integrating these models and multidimensional anomaly detections enables system engineers to identify both correlations and cause-and-effect relationships among the monitoring data.