New approach to noncausal identification of nonstationary stochastic systems subject to both smooth and abrupt parameter changes

In this paper we consider the problem of finite-interval parameter smoothing for a class of nonstationary linear stochastic systems subject to both smooth and abrupt parameter changes. The proposed parallel estimation scheme combines the estimates yielded by several exponentially weighted basis function algorithms. The resulting smoother automatically adjusts its smoothing bandwidth to the type and rate of nonstationarity of the identified system. It also allows one to account for the distribution of the measurement noise.