Control-relevant identification for constrained and nonlinear systems.

摘要

The need to develop methods that are able to identify process models adequate for advanced control algorithms, like Model Predictive Control, has become clear in recent years. Today, control-relevant identification, which studies the impact of model identification on the controller design and vice versa, provides us with the tools to address this issue.; In the first part of this dissertation, we present work on control-relevant identification of linear constrained systems. Our purpose is to extend control-relevant identification ideas for unconstrained systems to constrained ones. As a first step, control-relevant prefilters based on the unconstrained system are used to pre-treat the data. Then a constrained control-relevant identification criterion is formulated based on constrained control theory and is imposed as a constraint on the classical identification problem. Examples are given to illustrate the proposed scheme.; Volterra series can be used as input-output models of nonlinear systems. However their use has been limited due to the huge number of coefficients that need to be estimated. In the second part of this dissertation, we address this problem by extending the use of generalized orthonormal basis functions to non-linear system identification and discuss the merits of such use. Examples are presented to demonstrate the feasibility and the advantages of the proposed idea.; Some issues on the use of reduced Volterra models for the control of nonlinear processes are also discussed. Controllers based on the inverse of the Volterra model have been proposed but their use is limited due to the need of an invertible linear part. Here we develop some guidelines to avoid such problems by appropriately selecting certain model parameters. A control-relevant approach is also proposed to address this problem and effectively eliminate it. The resulting controller displays a larger stability range and better response performance than those based on classical model identification. Examples verify these statements and demonstrate the validity of this technique.; Volterra models can also facilitate other types of advanced control, like non-linear model predictive control and run-to-run control in semiconductor manufacturing. Due to their structure, analytic estimation of the partial derivatives is straightforward, reducing the computational time and improving the accuracy of the estimate. An example demonstrates the advantages of their use.