MULTIVARIABLE CLOSED-LOOP SYSTEM IDENTIFICATION USING ITERATIVE LEAKY LEAST MEAN SQUARES METHOD

Channel isolation usmg partial correlation analysis and estimation of model
parameters in the conventional multivariable closed-loop system identification
approaches use the method of Least Square (LS) with the limitations, viz. (I) large
estimation bias due to the highly correlated input signals, low Signal to Noise Ratio
(SNR) and correlated noise signals; and (2) inconsistent parameter estimates due to
noise and unmeasured disturbances. In this thesis. iterative Leaky Least Mean Squares
(LLMS) based methods are proposed to address the limitations ofLS method in MultiInput
Multi-Output (MIMO) closed-loop system identification. In this research. novel
algorithms have been developed to: (I) isolate the less interacting channe Is using a
modified partial correlation algorithm. (2) achieve unbiased and consistent parameter
estimates using an iterative LLMS algorithm and (3) develop parsimonious models for
closed-loop MIMO systems. All the proposed algorithms are demonstrated with the help
of extensive simulation case studies and a real pilot-scale distillation column with
appropriate validation procedures.