Constrained parameter estimation with applications to blending operations

The classical least squares approach to parameter estimation for dynamic models ignores a priori information about the feasible values of the estimated parameters. But in many practical problems, such information is available in the form of upper and lower limits. In this paper, two alternative techniques are evaluated for this important class of constrained parameter estimation problems for dynamic systems. Simulation results for two blending problems illustrate that more accurate parameter estimates and better predictions can be obtained by using a quadratic programming approach.