On-line parameter identification and design of multivariable discrete- time regulator

Abstract

This paper describes an efficient algorithm for parameter estimation and control of a multivariable discrete-time systems using input-output data. Parallel recursive least square algorithms based on special observable canonical innovation model are considered to estimate model parameters and subsequently parallel state estimation schemes are developed by partitioning the conventional Kalman-filter equation. Proposed observable canonical structure turns out to be more simpler than that of Desai and Mahalanabis [4] which in turn reduces the overall computational burden in the parameter estimation stage. Parameter estimates are then utilized to implement a simple controller that minimizes a single-stage performance index [4]. This scheme has some limitations and more specifically, the parameter estimates do not converge to true value for a weakly controllable system [1]. To avoid this difficulty, a numerically reliable and stable controller based on singular value decomposition (SVD) is considered. Simulation studies for parameter estimation and control of discrete-time systems are carried out by considering power system and flight control problems in order to test the effectiveness of the proposed method.