ABSTRACT
A time-series analysis of the complete data set would be misleading as data averaging would occur when there are detectable temporal changes in the time-series parameters. So, a statistically sound data analysis technique that is sensitive to nonstationarity in time series data is required. The technique that is discussed here involves fitting an autoregression model to a short initial sample of the data set and then updating the autoregression parameters to track possible temporal changes in these parameters. A sensitive test for a departure from stationarity involves accumulating the residuals and providing a warning signal whenever this sum exceeds a predetermined threshold value. The prediction errors, generated by the adaptive Kalman filter, are used by a sequential decision-making scheme to detect statistically significant changes in these parameters. Residual analysis indicated that the data should be smoothed using a second order moving average filter prior to analysis by the Kalman filter.
