作者: Jerome R. Bellegarda , David C. Farden
DOI: 10.23919/ACC.1988.4789920
关键词:
摘要: Tools are presented to reliably identify a time-varying autoregressive (AR) model for realization of stochastic process with an arbitrary non-stationarity. Only limited priori knowledge about the nature non-stationarity, namely expected maximum rate change parameters, is necessary estimate these parameters on-line. The criterion considered constrained least squares cost functional which incorporates equal weight all instantaneous errors up time observation. constraint specified from using (non-unique) backward state-space description parameter variation. A doubly recursive algorithm based on smoothing theory derived find quasi-optimal solution estimation problem. Associated trade-offs discussed various non-stationary environments.