Process identification in on-line optimizing control, an application to a heat pump

Abstract

The objective of this paper is to focus on on-line state and parameter estimation in connection with on-line model-based optimizing control of continuous processes. A nonlinear programming approach is used to estimate unmeasured state variables and parameters in systems modelled by nonlinear differential-algebraic equations. The nonlinear dynamic model is discretized by orthogonal collocation on finite elements, and the moving-horizon approach is used to reduce the dimension of the final optimization problem. A priori parameter information is included in the minimization criterion (Bayes estimation), and this makes the estimation problem more robust with respect to missing process excitations and over-parameterization. The a priori parameter covariance matrix is treated as a tuning matrix, where the diagonal elements can be set acording to the amount of information in the measurements. The updated steady-state part of the process model is used to optimize the economic performance of the process, where new optimum set-points are calculated for the regulatory control system. The sequential quadratic programming method (SQP) is used to solve this nonlinear optimization problem, where the objective function, the model equations, and the operational feasibility constraints are solved simultaneously in an "infeasible path" approach. The identification and on-line optimizing control approach are illustrated with an example from an experimental water-to-water heat pump unit.