Identification and Control of Linear Time-Varying Systems With Varying Uncertainty Degrees in Parameters: A Measurement-Based Approach
Abstract
This paper develops a measurement-based method for parameter identification and control correction of linear time-varying systems with heterogeneous levels of a priori parametric uncertainty. The formulation distinguishes adjustable parameters with known or unknown initial and target values from unknown fixed physical parameters. Using noisy sampled output measurements, a finite-dimensional residual model is constructed to estimate the unknown initial parameter values and physical parameters. An explicit generalized least-squares estimator is then derived through constrained covariance minimization. Under the stated residual-covariance and finite-horizon identifiability conditions, the estimator is finite-sample unbiased and minimum-variance within the class of linear unbiased estimators. Under independent repeated measurement/correction runs, strong and mean-square consistency are further established. The subsequent control-correction problem is formulated as a finite-dimensional quadratic tracking problem. A positive-definiteness condition on the control-information matrix guarantees a unique closed-form global minimizer within the selected control parameterization. A gas turbine engine idle-mode model is used as a numerical case study with synthetic process and measurement disturbances and ten independent stochastic realizations. The results illustrate the identification and correction procedure for constant and linear control parameterizations, with the latter yielding a lower final tracking error functional in the considered numerical case.