Control-relevant curvefitting for plant-friendly multivariable system identification

A control-relevant parameter estimation algorithm is developed in this paper for curvefitting Empirical Transfer Function Estimates (ETFEs) with orthogonal (i.e., zippered) frequency grids to discrete-time parametric Matrix Fraction Description models. Such ETFEs arise from DFT analysis of identification data generated from constrained, plant-friendly multisine inputs as developed by the authors' previous work. This curvefitter minimizes model estimation error using pre/post frequency-dependent weighting matrices as functions of the closed-loop dynamics. The control-relevant multivariable parameter estimation procedure is illustrated with an example case study based on the Shell Heavy Oil Fractionator problem.