Multivariable process identification for mpc: the asymptotic method and its applications

Abstract

Abstract:In this work we will introduce the asymptotic method (ASYM) of identification and provide two case studies. The ASYM was developed for multivariable process identification for model based control. The method calculates time domain parametric models using frequency domain criterion. Fundamental problems, such as test signal design for control, model order/structure selection, parameter estimation and model error quantification, are solved in a systematic manner. The method can supply not only input/output model and unmeasured disturbance model which are asymptotic maximum likelihood estimates, but also the upper bound matrix for the model errors that can be used for model validation and robustness analysis. To demonstrate the use of the method for model predictive control (MPC), the identification of a Shell benchmark process (a simulated distillation column) and an industrial application to a crude unit atmospheric tower will be presented.

Identification of multi-input systems: variance analysis and input design issues

“…, one can examine whether or not the input u 2 reduces the variance of some or all of the estimated coefficients of the polynomials A, B 1 and C. We mention that the gradients appearing in (10) have been computed in [4] for all commonly used polynomial model structures (FIR, ARX, OE, BJ, ARMAX).…”

“…In the literature, it is argued that in industry the guideline for identification practice is still single-variable based thinking [10]. However, several authors propose to excite all inputs simultaneously [11,2].…”

From experiment design to closed-loop control

“…The method ASYM, see Zhu (1998Zhu ( , 2001, is a fully integrated method for identification for control based on this approach and, hence, very close in spirit to the use of the separation principle discussed in Section 8.1. The input design is based on high-order optimal input design; see Section 11.1.…”

“…Since little effort has been spent so far on packaging algorithms in a way that meets the requirements of industrial practitioners, cf. Rivera and Jun (2000) and Zhu (1998), it is not surprising that, as reported in Zhu (1998), system identification still has had relatively little impact on industrial control engineers in general. However, as hopefully evidenced by this survey, the progress in this research area is very promising and there certainly are interesting years ahead!…”

Implementation of MPC on a Deethanizer at Kårstø Gas Plant

“…The dynamic models are identified by Tai-Ji ID (Zhu, 1998). The Tai-Ji ID identification is based on the asymptotic method (ASYM), which calculates time domain parametric models using frequency domain criterion.…”