European Journal of Operational Research 174 (2006) 668674

www.elsevier.com/locate/ejor

Book reviews

G.P. Liu, J.B. Yang, J.F. Whidborne, Multiobjec- 2. Content of the book
tive optimisation and control, Series: Engineering
Systems Modelling and Control, Research Studies The book comprises 12 chapters. In the
Press, 2003, ISBN 0 86380 264 8 Introduction (Chapter 1), a brief review is made
of multiobjective optimisation, encompassing tra-
1. Introduction ditional mathematical programming based ap-
proaches and genetic algorithms (GAs). A review
This book covers the central concepts of multi- of multiobjective control is also made, including
objective optimisation and control techniques. references to trade-os arising in control systems,
Multiobjective optimisation models and methodo- robust control, critical control, eigenstructure
logical approaches are increasingly applied to the assignment, PID control, nonlinear system identi-
design of control systems. The authors justify the cation, and fault diagnosis. These are the main top-
need to consider explicitly multiple objectives in ics developed in the subsequent chapters.
control system design based on the common real- Chapters 2 and 3 introduce the basic concepts
istic argument: multiple and conicting design and methods for unconstrained and constrained
objectives are at stake and inevitable trade-os nonlinear optimisation, respectively. These chap-
arise (for example, between output performance ters, as it is understandable in the scope of the
and stability robustness). Besides theoretical book, do not develop these topics in-depth, but
foundations, the book presents a large range of de- they oer the reader a comprehensive review of
sign algorithms and describes applications of mul- concepts and methodological approaches which
tiobjective optimisation and control to several are necessary for the following chapters. In Chap-
areas. ter 2 one-search dimensional methods such as
The book can be used for classroom work at Fibonaccis and golden section search are pre-
the undergraduate or postgraduate levels, pro- sented as well as unconstrained methods such as
vided the students possess a technical back- steepest descent, Newtons and quasi-Newtons
ground in linear control systems, linear algebra methods. In Chapter 3, KuhnTucker necessary
and calculus. The book can also be useful to conditions and sucient conditions (both based
researchers, control system designers and profes- on Hessian matricessecond order derivatives
sionals of several engineering disciplines from and saddle points) are introduced. The presenta-
which practical optimisation and control applica- tion divides the approaches in primal methods
tions emerge. (sequential linear programming and sequential
The book has 320 pages and it is priced at 65 at quadratic programming) and dual methods
the publisher web site www.research-studies-press.- (sequential unconstrained minimisation tech-
co.uk (an electronic version is available at 60). niques: exterior penalty, interior penalty; and
The authors are G.P. Liu (University of Lagrangean). The authors justify the interest of
Nottingham, UK), J.B. Yang (UMIST, UK), these approaches on the basis of their utilisation
and J.F. Whidborne (Kings College London, in optimisation software packages and GA-based
UK). procedures.
 Book reviews / European Journal of Operational Research 174 (2006) 668674 669

The basic concepts and methods in multiobjec- Chapter 8 discusses multiobjective robust con-
tive optimisation are introduced in Chapter 4. Par- trol design for multivariable systems using eigen-
ticular attention is paid to methods based on Lp structure assignment (a design technique which
norms and interactive methods. In the rst cate- may be used to assign the entire eigenstructure of
gory the following approaches are mentioned: ref- a closed-loop linear system via a constant gain full
erence point methods (for which the minmax, state or output feedback control law). The individ-
minimizing the maximum distance to the ideal ual eigenvalue sensitivities and the sensitivity func-
solution can be considered a particular case), goal tions of the closed-loop system in the frequency
attainment and goal programming. In the second domain, and the controller gain constraint have
category four interactive methods are presented: been considered as the multiobjective performance
Georions, STEM, ISTM and an interactive criteria (expressed as a set of inequalities).
procedure based on the (utility) gradient projection Chapter 9 presents a PI controller design using
method. In the framework of the interactive meth- multiobjective optimisation (addressing the ALS-
ods described, it is discussed how the design engi- TOM benchmark challenge on gasier control).
neers preferences can be elicited and used in the The parameters of the multi-input multi-output
operational framework of the procedures. PI controller must satisfy a set of performance cri-
Chapter 5 makes an introduction to GAs in the teria derived from the system specications on
realm of optimisation, spanning the basic structure outputs, inputs and input rates. The nonlinear gas-
of GAs, population representation, computation ier is linearised in three load operating points
of tness functions to assess the performance of (100%, 50% and 0%) at which the degree of satis-
individuals, and the role and working of the oper- faction of specications is measured. Results are
ators selection, crossover and mutation. Some ap- provided comparing the multiobjective optimal-
proaches to multiobjective optimisation using GAs tuning PI controller with xed parameter control-
are then described. lers regarding computation eort, stability and
Chapter 6 presents a hybrid approach to multi- performance.
objective robust control system design combining In Chapter 10 a multiobjective GA is pre-
the exibility of numerical (parameter) optimisa- sented to compute optimal Finite Word-Length
tion techniques (method of inequalities) with ana- structures for PI digital controllers. The method
lytical optimisation (H1, LQG). This combined is illustrated through its application to the imple-
approach is illustrated using two design case stud- mentation of PID controllers for a steel rolling mill
ies (reported elsewhere in the literature) concern- and a benchmark problem. Parameterisation is-
ing a control system for a high purity distillation sues linked to Finite Word-Length implementa-
column and the suspension controller for a mag- tions are relevant because of the degradation
netic levitated vehicle. Pointers to MATLAB rou- (even instability) experienced by closed-loop con-
tines are given. The combined approach enables to trol systems when innite precision controllers
include desirable features such as making explicit are implemented by using xed-point digital pro-
the design trade-os and the closed-loop perfor- cessors (due to the nite precision of the parame-
mance, including time domain and frequency do- ters representation). The method exploits the use
main performance indices. of binary numbers in both the implementation of
Chapter 7 introduces critical control systems and Finite Word-Length controllers and representa-
discusses robust control design of critical systems tion in GAs. The method requires the solution of
with external uncertainties (environmental condi- a linear system similarity completion problem,
tions, disturbances, etc.) and internal uncertainties which is illustrated for the case of a two state
(in the mathematical models). The robust control single-input single-output system.
design problem for critical systems is formulated Chapter 11 addresses the problems of model
considering explicitly a set of performance criteria, selection and identication of nonlinear systems
including output performance in the time domain (a nonlinear functional approximation prob-
and robust performance in the frequency domain. lem) using neural networks and GAs. Multiple
670 Book reviews / European Journal of Operational Research 174 (2006) 668674

performance criteria based on the L2 and L1- grated and the book can be read sequentially or
norm distances between the actual system and only the relevant chapters, which are reasonably
the nonlinear model are considered, as well as a self-contained for someone with technical back-
model complexity indicator (number of nonlinear ground on control system theory. Nevertheless,
units in the model). The method is based on GA in some parts of the book this generates sections
optimisation, involving computing solutions to that are too condensed (which is understandable
the linear system completion problem, to obtain in the scope of journal or conference papers),
the nonlinear function units associated with the but that the authors could have developed further,
simplest model required for approximation. A namely having in mind the use of this book
neural network is used for model representation, for classroom work specially at undergraduate
which is the Gaussian radial basis function. An levels.
illustrative example is reported regarding the iden- Although this strongly depends on the orienta-
tication of a large pilot scale liquid level nonlin- tion given by lecturers to the courses, I have seri-
ear system. ous doubts the book could full the purpose of
Chapter 12 deals with the design of optimal supporting undergraduate students unless it is a
residuals for diagnosing incipient faults. Perfor- strongly control-oriented undergraduate degree.
mance indices are incorporated into the fault ob- However, I found the book a valuable tool for
server design to reduce false and missed alarm graduate students in control related topics in sev-
rates. Five performance indices are expressed in eral engineering disciplines.
the frequency domain, thus accounting for the fre- In my opinion, the review of interactive meth-
quency distribution of faults, noise and modelling ods in Chapter 4 is somewhat poor. Despite the
inaccuracies. Even though the problem is recogni- authors claim that attention is paid to interactive
sed as multiobjective in nature, the authors trans- methods, there are several interactive procedures
form the performance criteria into a set of which are not even mentioned or listed in the ref-
constraints. The problem becomes then nding a erence list. The same happens in Chapter 5 regard-
solution (parameter set) with all performance indi- ing the introduction to GAs. This section is
ces within the acceptable region dened by the surprisingly short and the reference list does not
constraint set. A GA is used to compute that solu- include some important review papers and books
tion. An example is given concerning the fault on multiobjective optimisation using GAs, which
diagnosis of a ight control system. is an exciting and active area both regarding re-
search and applications, including in control as it
is clearly shown in this book.
3. Discussion Section 7.5 on multiobjective critical control is
also too condensed. A generous expansion of this
The main merit of this book is to present in a section in the context of this chapter would be wel-
consistent manner multiobjective optimisation in come in future editions of this book. In Chapter 8,
the realm of control problems. Most books on I also found the description of the controller de-
multiobjective optimisation are too mathematical sign using GAs too scarce to be really useful for
or management-oriented and this book contrib- students and design engineers. Again, the authors
utes to ll the gap regarding engineering applica- should have complemented and expanded the texts
tions of multiobjective optimisation models and published elsewhere to make them more readable
methods, namely in the scope of control systems. and self-contained.
Moreover, it reects both theoretical and practical In summary, I think this book is a very interest-
insights of multiobjective optimisation and ing and useful reference oering a broad view of
control. the range of problems and approaches in multiob-
The book content is mainly based of work jective optimisation and control. Moreover, up-
carried out by the authors and already published to-date references are provided (except for the
elsewhere. However, I found the chapters well inte- criticisms I made above) which enable the reader
 Book reviews / European Journal of Operational Research 174 (2006) 668674 671

to develop further his/her own specic interests in Polo IIPinhal de Marrocos

this eld. 3030 Coimbra
Portugal
Carlos Henggeler Antunes E-mail address: [email protected]
Department of Electrical Engineering
and Computers Available online 15 June 2005
University of Coimbra
doi:10.1016/j.ejor.2005.04.003

Saul I. Gass, Arjang A. Assad, An Annotated 2. Content of the book

Timeline of Operations Research, An Informal
History, Kluwer, 2005 The title of the book accurately reects its con-
tent. The annotated timeline is divided into eight
1. Introduction periods:

This volume is a new contribution to the grow- (1) Operations research precursors from 1564 to
ing literature on the history of OR. The authors 1873 (18 pages).
have tried to organize the development of our dis- (2) Operations research precursors from 1881 to
cipline on a timeline, emphasizing a number of 1935 (25 pages).
key events. For each of these events, the authors (3) Birth of operations research from 1936 to
provide a brief explanation of why it is important 1946 (15 pages).
(from 10 lines to one page) and a number of refer- (4) Expansion of operations research from 1947
ences. For most events, a picture of the authors of to 1950 (19 pages).
the discovery is also available together with a short (5) Mathematical, algorithmic and professional
biographical sketch and/or representative quotes. development of operations research from
This book is authored by Saul I. Gass and 1951 to 1956 (31 pages).
Arjang A. Assad. The name of Saul Gass will (6) International activities, algorithms, applica-
probably be familiar to all readers of this Journal tions, and operations texts and monographs
through his many papers and books and, in partic- from 1957 to 1963 (29 pages).
ular, his Encyclopedia of Operations Research and (7) Methods, applications and publications from
Management Science (co-edited with C.M. Harris, 1964 to 1978 (29 pages).
and published by Kluwer). He is currently Professor (8) Methods, applications, technology and pub-
Emeritus at the University of Maryland at College lications from 1979 to 2004 (33 pages).
Park. Saul Gass is a past president of ORSA and a
past vice-president of both INFORMS and The book ends with a very useful index of acro-
IFORS. Arjang Assad is professor of Management nyms, names, and subjects.
Science at the University of Maryland at College The timeline for the prehistory of OR (up to
Park. He has published two books and numerous 1936) reveals a rich picture, including many
research papers. His interests include Total Qual- great names: in Probability (Cardano, Pascal, Huy-
ity, vehicle routing and operations management. gens, Jakob, Nicolaus and Daniel Bernoulli, de
This volume has 210 pages and is priced at Moivre, Buon, Bayes, Laplace, Poisson, Galton,
79.95 (ISBN 1-4020-8112-X). A cheaper paper- Bachelier, Markov, Erlang, Levy, de Finetti, Chap-
back edition is also available at 29.95 (ISBN man, Kolmogorov, Pollaczeck), Statistics (Bosco-
1-4020-8116-2). vich, Graunt, Legendre, Gauss, Quetelet, Neyman,