Recursive Subspace System Identification (RSSID) algorithms

Abstract

The goal of system identification is to find mathematical equation that gives approximation to the actual behavior of real systems. In this thesis, a recursive subspace model identification algorithm is presented that recursively identifies both linear and nonlinear systems. Each recursion step consisted of two-stages: first, the innovation form of the stochastic system was estimated, then the model Matrices was estimated. Much attention is paid to the computational cost and the performance of the models derived from the developed identification algorithm and a comparison to existing traditional methods as well as the neural network algorithm was made using various monte-carlo simulations on different laboratory data. It is observed that the proposed algorithm performed better than some traditional methods in some conditions and was reasonable good on other conditions or process types and is therefore very reliable.