Sunahara Memorial Lecture

 

Grey-box identification based on horizon estimation

and nonlinear optimization

 

Dr. Alf Isaksson

(Corporate Research Fellow, ABB Corporate Research

and Director Process Industry Center, Linköping University)

 

 

Abstract

 

Obtaining models for real industrial plants there are two main alternatives available:

  • Deriving a model from first principles using laws of physics, chemistry, etc, so-called white-box modelling
  • Estimating an empirical model from experimental data -- black-box modelling

In general, a white-box model becomes a set of nonlinear Differential and Algebraic Equations (DAE). A black-box model on the other hand is typically linear, although non-linear black-box structures do exist (e.g. neural nets), and are usually estimated in a stochastic framework. This talk will focus on the combination of these two techniques, i.e. to start with a physical model, but adding black-box elements to this model. This is often referred to as grey-box modelling or grey-box identification.
As described in the book by Bohlin (2006), grey-box modelling requires a number of steps which in turn need software support:

    1. Physical modelling

    2. Numerical estimation of model parameters

    3. Model calibration

    4. Model validation

The lecture will go through these steps and discuss how they may and should be done. The main focus will be on steps 1 and 2. For the physical modelling the use of object oriented modelling in general and the modelling language Modelica in particular will be described.
The traditional solution for parameter estimation is to use a maximum likelihood criterion formed by running an extended Kalman filter. It will be argued that an interesting alternative is to work directly with a discretized version of the nonlinear model and fit both states and parameters directly using nonlinear optimization.

 

Reference

Bohlin T. (2006). Practical Grey-box Process Identification -- Theory and Applications. Springer.

 

 

 

Dr. Alf

 

Biography of Dr. Alf Isaksson

 

Dr. Alf Isaksson was born in Sweden on 6 June, 1959. He got Ph.D in Automatic Control, Linköping University in 1988. He was a full professor in Royal Institute of Technology in Stockholm in 1999-2001, and he entered ABB Corporate Research, Vasteras, Sweden in 2001. Now he is Corporate Research Fellow, ABB Corporate Research and Director Process Industry Center, Linköping University (PIC-LI). Dr. Isaksson has written 25 journal articles, ca 60 conference articles and 6 patent applications. His research field focuses on automatic control and system identification. He is an associate editor IEEE Transactions of Control Systems Technology, and a member of the international programme committee for several conferences including the bi-annual conference series Control Systems devoted to Pulp & Paper industry, since 1996 (incl chairman of committee 2002). He has been a senior member of IEEE, industry vice-chair for the IFAC Symposium SAFEPROCESS 2009 in Barcelona.