Learning identification of a class of time-varying ARMAX systems

This paper presents a learning identification method for a class of ARMAX systems with time-varying unknowns. The least squares learning algorithm is derived on the basis of repetitive operations over a pre-specified finite time interval. Sufficient conditions are presented for repetitive consistency of the learning algorithm. It is shown that the complete estimation can be achieved, and the estimates converge to the time-varying values of the parameters over the entire interval. In addition, the learning identification method is shown to be applicable to periodically time-varying systems.