A basis expansion approach for detecting transient plant disturbances and jumps

Rapid changes in the behavior of linear systems are usually modeled by sudden jumps in the system's parameters. In this paper, basis expansion techniques are considered to address the more general case, where the transient disturbance cannot be adequately described by simple jumps (piecewise constant parameters). Each parameter's time history is expanded onto a set of basis sequences that can well describe the transient disturbance. Short Time Fourier transform as well as wavelet bases are investigated to model the local cyclostationarity or local erratic behavior of the system's parameters. The resulting time varying system is tested against the time invariant alternative (constant parameters) using variations of the AIC criterion or /spl Aring/strom's F-test.<>