Practical Consideration for Identification of Decentralised Control Systems

Abstract

In practice, there is a great number of complex systems (not only by physical size, but also by the nature of models that describe systems) which need to be controlled. The identification of the original multivariable system is a difficult problem. Therefore, the idea is to decompose the original problem into a series of simpler subproblems. The ideas, in system identification, are borrowed from the methods for multivariable controller design: hierarchical and decentralized controllers. In this paper, the concept of decentralization has used. The original multivariable system is decomposed into a series of simpler subsystems with single input and single output. Then, the identification algorithms are proposed for the case when stochastic disturbance have non-Gaussian distribution. The algorithms of stochastic approximation and extended least squared are considered. Based on simulation, convergence speed and accuracy of estimation are compared. The concept of algorithms with extended memory is discussed.