Diagnostic Analyses of a Nuclear Power Plant Using Multivariate Autoregressive Processes

Abstract

Multivariate autoregressive (AR) procedures are introduced as diagnostic tools to extract dynamic,characteristics for detection of malfunctions of a boiling water reactor (BWR) power plant. The problem of estimating AR matrices is equivalent to identifying, from measured random signals of a BWR station, the dynamic parameters of a stationary linear discrete time system derived from an unmeasured uncorrelated white-noise process. To explain the characteristics of a derived AR spectra , a general multiple-input, single-output model is discussed. The experiments were carried out in a 460-MW(e) BWR station. The power spectral density of the averaged neutron flux is decomposed into terms corresponding to sources of noise at points of measurement, where ^{the origin of  the noise neutron} fluctuation is studied. It is shown fom the analysis that a disturbance of high intensity in neutron fluctuation of the BWR is not Caused by the process var Such Core flow but is possibly caused by the inherent noise.specifically defined in this paper, of the neutron flux itself.