Research on Intelligent Classification Methods for Nuclear Power Plant Operating Condition Change Alarms

Under complex environmental operating conditions, nuclear power devices are prone to faults. At the same time, due to the operational requirements for changing conditions, these devices frequently shift their operating modes, which leads to fluctuations in system parameters and triggers alarms that affect operators’ ability to diagnose faults. This paper proposes an intelligent anomaly classification framework based on a hybrid algorithm of Denoising Autoencoder (DAE) and Isolation Forest (iForest), constructing a time-series feature fusion model. The DAE is used to extract multidimensional parameter features in high-noise environments, and, combined with the unsupervised anomaly scoring mechanism of iForest, it achieves precise differentiation between anomalous events and non-steady parameter fluctuations. This approach helps operators quickly identify operational risks, enhancing their ability to detect hidden faults while maintaining the flexibility of changing operating conditions in nuclear power devices, and ultimately improving the capability to cope with risks under complex conditions.