Results and interpretation of multivariate autoregressive analysis applied to the loft reactor process noise data

Abstract

Multivariate noise analysis of power reactor signals is useful for characterizing baseline operation and plant diagnostics, for isolating process anomaly from sensor maloperation, and for automated plant monitoring. We have developed a systematic and reliable procedure of accomplishing these tasks by improving upon the previously established techniques of empirical modeling of fluctuation signals in power reactors. The application of the algorithm to data from the Loss-of-Fluid Test (LOFT) reactor showed that earlier results (based on physical modeling) regarding the perturbation sources in a pressurized water reactor (PWR) affecting coolant temperature and neutron power fluctuations can be explained using multivariate autoregressive (MAR) analysis. This methodology has important implications regarding plant diagnostics, and system or sensor anomaly detection.