Evaluation of multivariate autoregressive time series models and their application to reactor noise analysis

The multivariate autoregressive (MAR) time series model is investigated to determine its usefulness for representing continuous random signals. . This is achieved by determining the best criterion of fit, an estimate of the total number of sample points of signals and the choice of sampling interval. Furthermore, the MAR model procedures are employed to investigate the effect of nonstationarity of the data. The autoregressive (AR) time series model is statistically evaluated to compare the error in the AR spectrum with the error estimates of Fourier transform (FFT) method. Moreover, the error in the AR spectrum is calculated using an empirical procedure. Finally, the MAR model procedures are used as a diagnostic tool to derive cause-effect relationship among process variables from the St. Lucie nuclear power plant.

It is determined that the MAR model can adequately represent noisy continuous signals including periodic processes. Proper analysis must be carried out for signals with wide dynamic ranges.