Subspace-Driven Learning for Anomaly Detection in Process Transients

Nuclear power plant (NPP) monitoring and diagnostic centers are actively investigating and implementing automated anomaly detection algorithms to help plants catch anomalies sooner, thereby preventing or reducing the duration of unexpected shutdowns. Current machine learning (ML)-based anomaly detection methods are expected to be highly effective during stable, full-power operations because NPPs typically operate as baseload power generators, meaning that extensive operating data on plant equipment are available. However, anomaly detection methods are expected to face significant challenges when it comes to transient conditions (i.e., when the power output falls below full power). This is because plants only occasionally operate at these lower power levels, thus generating sparse transient operational data. This can result in false alarms or missed detections during transient conditions. To address this issue, transfer learning is employed. For the present case, this entails leveraging knowledge (in the form of learned features) derived from stable, full-power operations to improve detection accuracy under transient conditions, despite limited data. For this effort, a novel subspace approach was developed to transfer a subset of the data features from full-power operations over to cases involving transients. This approach was validated through experiments by using synthetic data, and was found to outperform two baseline transfer learning approaches when comparing anomaly detection performance across different percentages of transient data applied to the training process.