Finite memory generalized predictive controls for discrete-time state space models

In this paper, a generalized predictive control (GPC) is represented over state space and then is shown to be separated into the receding horizon control (RHC) and the steady-state Kalman filter. By utilizing only the information on the recent finite inputs and outputs, we propose a new finite memory GPC (FMGPC) that consists of the RHC and a finite impulse response (FIR) filter. The proposed FMGPC will be compared with a conventional GPC for an input-output (I/O) model and the existing receding horizon finite memory control (RHFMC) for a state space model.