Multivariable
System Identification
for Process ControlYucai Zhu (Elsevier Science Ltd, Oxford, UK, 372 pages, ISBN: 0-08-043985-3)
Preface
Systems and control theory has experienced a continuous development in the last few decades. The state space approach and Kalman filter are the products of the 1960s which, for the first time, made it possible to solve general linear multivariable control problems. Since 1970 adaptive control theory and techniques have been developed. In the 1980s robust control and H-infinity control of multivariable systems were developed. Fault detection and diagnosis techniques were also developed in this period. These new techniques are very promising for industrial applications and have attracted much interest from the academic researchers. The impact of these developments on process industries, however, has been very limited. When we visit plants in process industries, we find that a typical modern computer control system is a combination of the state-of-the-art computer technology and classical PID (proportional, integral and differential) control algorithms which are the restrictive single variable control techniques of the 1940s and 1950s.
Many possible reasons for this failure of technology transfer can be identified. One important reason is the lack of accurate dynamic models of industrial processes, since all the above mentioned modern techniques are model-based and need reasonably accurate process models. Another reason is the lack of good communication between the modern control community and process industries.
However, one process industry has made a distinction. In the last decade, model predictive control (MPC) technology has gained its industrial position in the refinery and petrochemical industry, and has started to attract interest from other process industries. There are over 500 control engineers from contracting and operation companies several thousands of applications have been reported. Most often, an MPC controller uses a linear dynamic model of the process that is obtained by way of black-box identification. However, due to various reasons, the cost of current MPC identification is very high and many trials and errors have to be made by the user. The test time is rather long (from several weeks to several months) and the tests are carried out manually around the clock. This, on the one hand, demands very high commitment of engineers and operators and, on the other hand, makes MPC project planning difficult. It is believed nowadays that process modeling and identification is the most difficult and time consuming part of an MPC project. Wide spread applications of MPC technology call for more effective and efficient identification technology.
Process identification is the field of mathematical modeling using test data. This branch of automatic control has been very actively developed in the last three decades, with many books published on the topic. Most of these books have very high academic quality. However, they are too theoretically oriented for industrial users and for undergraduate students. Therefore, the purpose of this book is to fill the gap between theory and application and to provide industrial solutions to process identification that are based on sound scientific theory. We will study various identification methods; both linear and block-oriented nonlinear models will be treated. We will present, in detail, project procedures for multivariable process identification for control. Identification test design and model validation will be emphasized. The book is organized in a way that is reader friendly and easy to use for engineers and students. We will start with the simplest method, and then gradually introduce other methods. In this way, one can bring more physical insight to the reader and some mathematics can be avoided. Each method is treated in a single chapter or section, and experiment design is explained before discussing any identification algorithms. The many simulation examples and actual industrial case studies will show the power and efficiency of process identification and will make the theory more applicable. Matlab® M files are included that will help the reader to learn identification in a computing environment.
M-files of the book (for Matlab® 6.5)
Go back to Tai-Ji Control home page