large-scale c 36. large-scale complex...

619

Large-Scale C36. Large-Scale Complex Systems

Florin-Gheorghe Filip, Kauko Leiviskä

Large-scale complex systems (LSS) have tradi-tionally been characterized by large numbers ofvariables, structure of interconnected subsys-tems, and other features that complicate thecontrol models such as nonlinearities, time de-lays, and uncertainties. The decomposition of LSSinto smaller, more manageable subsystems al-lowed for implementing effective decentralizationand coordination mechanisms. The last decade re-vealed new characteristic features of LSS such asthe networked structure, enhanced geographicaldistribution and increased cooperation of subsys-tems, evolutionary development, and higher risksensitivity. This chapter aims to present a balancedreview of several traditional well-establishedmethods and new approaches together with typ-ical applications. First the hierarchical systemsapproach is described and the transition fromcoordinated control to collaborative schemes ishighlighted. Three subclasses of methods thatare widely utilized in LSS – decentralized control,simulation-based, and artificial-intelligence-based schemes – are then reviewed. Several basicaspects of decision support systems (DSS) that are

36.1 Background and Scope .......................... 62036.1.1 Approaches ................................. 62136.1.2 History ........................................ 622

36.2 Methods and Applications ..................... 62236.2.1 Hierarchical Systems Approach ....... 62236.2.2 Other Methods and Applications .... 626

36.3 Case Studies ......................................... 63236.3.1 Case Study 1:

Pulp Mill Production Scheduling..... 63236.3.2 Case Study 2: Decision Support

in Complex Disassembly Lines ........ 63336.3.3 Case Study 3: Time Delay Estimation

in Large-Scale Complex Systems ..... 634

36.4 Emerging Trends .................................. 634

References .................................................. 635

meant to enable effective cooperation betweenman and machine and among the humansin charge with LSS management and controlare briefly exposed. The chapter concludes bypresenting several technology trends in LSS.

There is not yet a universally accepted definition ofthe large-scale complex systems (LSS) though the LSSmovement started more than 40 years ago. However, byconvention, one may say that a particular system isa large and complex one if it possesses one or severalcharacteristic features. For example, according to To-movic [36.1], the set of LSS characteristics includes thestructure of interconnected subsystems and the presenceof multiple objectives, which, sometimes, are vague andeven conflicting. A similar viewpoint is proposed byMahmoud, who describes a LSS as [36.2]:

A system which is composed of a number of smallerconstituents, which serve particular functions, sharecommon resources, are governed by interrelatedgoals and constraints and, consequently, requiremore than one controllers.

Siljak [36.3] states that a LSS is characterized byits high dimensions (large number of variables), con-straints in the information infrastructure, and thepresence of uncertainties. At present there are soft-ware products on the market which can be utilized

PartD

36

620 Part D Automation Design: Theory and Methods for Integration

to solve optimization problems with thousands ofvariables. A good example is Solver.com [36.4].Complications may still be caused by system non-

linearities, time delays, and different time constants,and, especially over recent years, risk sensitivityaspects.

36.1 Background and Scope

In real life one can encounter lots of natural, man-made, and social entities that can be viewed as LSS.From the early years of the LSS movement, the LSSclass has included several particular subclasses suchas: steelworks, petrochemical plants, power systems,transportation networks, water systems, and societalorganizations [36.5–7]. Interest in designing effectivecontrol schemes for such systems was primarily moti-vated by the fact that even small improvements in theLSS operations could lead to large savings and impor-tant economic effects.

The structure of interconnected subsystems has ap-parently been the characteristic feature of LSS to befound in the vast majority of definitions. Several sub-classes of interconnections can be noticed (Fig. 36.1).

Resources

w1

u1 SSy1

SSy2u2

w2 m2m1

y2

y1

a)

w1

u1

u2

SSy1

SSy2

u = h (z)

w2 m2

z1

z2

m1

y2

y1

b)

w1

u1

u2

SSy1

SSy2

s = g (s,z)u = h (s)

w2 m2

z1

z2

m1

y2

y1

c)

Fig. 36.1a–c Interconnection patterns: (a) resource sharing, (b) di-rect interconnection, (c) flexible interconnection; SSy = subsystem,m = control variable, y = output variable, w = disturbance, u =interconnection input, z = interconnection output, g(·) = stock dy-namics function, h(·) = interconnection function

First there are the resource sharing interconnectionsdescribed by Findeisen [36.8], which can be identi-fied at the system level as remarked by Takatsu [36.9].Also, at the system level, subsystems may be inter-connected through their common objectives [36.8].Subsystems may also be interconnected through bufferunits (tanks), which are meant to attenuate the ef-fects of possible differences in the operation regimesof plants which feed or drain the stock in the buffer.This type of flexible interconnection can frequentlybe met in large industrial and related systems suchas refineries, steelworks, and water systems [36.10].The dynamics of the stock value s in the buffer unitcan be modeled by a differential equation. In somecases buffering units are not allowed because of tech-nological reasons; for example, electric power cannotbe stocked at all and reheated ingots in steelworksmust go immediately to rolling mills to be processed.When there are no buffer units, the subsystems arecoupled through direct interconnections, at the processlevel [36.9].

In the 1990s, integration of systems continued andnew paradigms such as the extended/networked/virtualenterprise were articulated to reflect real-life develop-ments. In this context, Mårtenson [36.11] remarkedthat complex systems became even more complex.She provided several arguments to support her remark:first, the ever larger number of interacting subsystemsthat perform various functions and utilize technolo-gies belonging to different domains such as mechanics,electronics, and information and communication tech-nologies (ICT); second, that experts from differentdomains can encounter hard-to-solve communicationproblems; and also, that people in charge of control andmaintenance tasks, who have to treat both routine andemergence situations, possess uneven levels of skillsand training and might even belong to different cultures.

Nowadays, Nof et al. show that [36.12]:

There is the need to create the next generation man-ufacturing systems with higher levels of flexibility,allowing these systems to respond as a componentof enterprise networks in a timely manner to highlydynamic supply-and-demand networked markets.

PartD

36.1

Large-Scale Complex Systems 36.1 Background and Scope 621

Table 36.1 Summary of methods described in this chapter

Decomposition-coordination-based Mesarovic, Macko, Takahara [36.7]; Findeisen et al. [36.16];

methods Titli [36.17]; Jamshidi [36.6]; Brdys, Tatjewski [36.18]

Optimization-based methods Dourado [36.19]; Filip, et al. [36.20]; Filip et al. [36.21];

Guran et al. [36.22]; Peterson [36.23]; Tamura [36.24]

Decentralized control Aybar et al. [36.25]; Bakule [36.26]; Borrelli et al. [36.27];

Inalhan et al. [36.28]; Krishnamurthy et al. [36.29];

Langbort et al. [36.30]; Siljak et al. [36.31]; Siljak et al. [36.32]

Simulation-based methods Arisha and Yong [36.33]; Chong et al. [36.34]; Filip et al. [36.20];

Gupta et al. [36.35]; Julia and Valette [36.36]; Lee et al. [36.37];

Leiviska et al. [36.38, 39]; Liu et al. [36.40];

Ramakrishnan et al. [36.41]; Ramakrishnan and Thakur [36.42];

Taylor [36.43]

Intelligent methods

• Fuzzy logic Ichtev [36.44]; Leiviska [36.45]; Leiviska and Yliniemi [36.46];

• Neural networks Arisha and Yong [36.33]; Azhar et al. [36.47]; Hussain [36.48];

Liu et al. [36.49]

• Genetic algorithms Dehghani et al. [36.50]; El Mdbouly et al. [36.51]; Liu et al. [36.40]

• Agent-based methods Akkiraju et al. [36.52]; Hadeli et al. [36.53]; Heo and Lee [36.54];

Marık and Lazansky [36.55]; Park and Lim [36.56]; Parunak [36.57]

They also emphasize that e-Manufacturing is highlydependent on the efficiency of collaborative human–human and human–machine e-Work. See Chap. 88 onCollaborative e-Work, e-Business, and e-Service. Ingeneral, there is a growing trend to understand thedesign, management, and control aspects of complexsupersystems or systems of systems (SoS). Systemsof systems can be met in space exploration, mili-tary and civil applications such as computer networks,integrated education systems, and air transportation sys-tems. There are several definitions of SoS, most ofthem being articulated in the context of particular ap-plications; for example, Sage and Cuppan [36.13] statethat a SoS is not a monolithic entity and possessesthe majority of the following characteristics: geographicdistribution, operational and management independenceof its subsystems, emergent behavior, and evolutionarydevelopment. All these developments obviously im-ply ever more complex control and decision problems.A particular case which has received a lot of attentionover recent years is large-scale critical infrastructures(communication networks, the Internet, highways, wa-ter systems, and power systems) that serve not only thebusiness sector but society in general [36.14, 15]. All

of these recent developments are likely to provide freshstrong stimuli for new research in the LSS domain.

36.1.1 Approaches

The progresses made in information and commu-nication technologies have enabled the designer toovercome several difficulties he might have encounteredwhen approaching a LSS, in particular those causedby a large number of variables and the low perfor-mance (with respect to throughput and reliability) ofcommunication links. However, as Cassandras pointsout [36.58]:

The complexity of systems designed nowadays ismainly defined by the fact that computational poweralone does not suffice to overcome all difficultiesencountered in analyzing, planning and decision-making in the presence of uncertainties.

A plethora of methods have been proposed overthe last four decades for managing and controllinglarge-scale complex systems such as: decomposition,hierarchical control and optimization, decentralizedcontrol, model reduction, robust control, perturbation-

PartD

36.1


based techniques, usage of artificial-intelligence-basedtechniques, integrated problems of system optimiza-tion and parameter estimation [36.59], and so on. Twocommon ideas can be found in the vast majority ofapproaches proposed so far:

a) Replacing the original problem with a set of sim-pler ones which can be solved with the availabletools and accepting the satisfactory, near optimalsolutions

b) Exploiting the particular structure of each system tothe extent possible.

Table 36.1 presents a summary of the main methodsto be described in this chapter.

36.1.2 History

Though several ideas and methods for controlling LSSswere proposed in the 1960s and even earlier, it is ac-cepted by many authors that the book of Mesarovicet al. published in 1970 [36.7] triggered the LSS move-

ment. The concepts revealed in that book, even thoughthey were strongly criticized in 1972 by Varaiya [36.60](an authority among the pioneers of the LSS move-ment), have inspired many academics and practitioners.A series of books including those of Wismer [36.61],Titli [36.17], Ho and Mitter [36.62], Sage [36.63],Siljak [36.3,64], Singh [36.65], Findeisen et al. [36.16],Jamshidi [36.6], Lunze [36.66], and Brdys and Tat-jewski [36.18] followed on and contributed to theconsolidation of the LSS domain of research and pavedthe way for practical applications.

In 1976, the first International Federation of Au-tomatic Control (IFAC) conference on Large-ScaleSystems: Theory and Applications was held in Udine,Italy. This was followed by a series of symposia whichwere organized by the specialized Technical Commit-tee of IFAC and took place in various cities in Europeand Asia (Toulouse, Warsaw, Zurich, Berlin, Beijing,London, Patras, Bucharest, Osaka, and Gdansk). Thescientific journal Large Scale Systems published byNorth Holland played an important role in the devel-opment of LSS domain, especially in the 1980s.

36.2 Methods and Applications

36.2.1 Hierarchical Systems Approach

The central idea of the hierarchical multilevel systems(HMS) approach to LSS consists of replacing the orig-inal system (and the associated control problem) witha multilevel structure of smaller subsystems (and asso-ciated less complicated problems). The subproblems atthe bottom of the hierarchy are defined by the interven-tions made by the higher-level subproblems, which inturn utilize the feedback information they receive fromthe solutions of the lower-level subproblems.

There are three main subclasses of hierarchieswhich can be obtained in accordance with the complex-ity of description, control task, and organization [36.7].

Levels of DescriptionThe first step in analyzing an LSS and designing thecorresponding control scheme consists of model build-ing. As Steward [36.67] points out, practical experiencewitnessed there is a paradoxical law of systems. If thedescription of the plant is too complicated, then the de-signer is tempted to consider only a part of the system ora limited sets of aspects which characterize its behavior.In this case it is very likely that the very ignored parts

and aspects have a crucial importance. Consequently itemerges that more aspects should be considered, butthis may lead to a problem which is too complex to besolved in due time. To solve the conflict between thenecessary simplicity (to allow for the usage of exist-ing methods and tools with a reasonable consumptionof time and other computer resources) and the accept-able precision (to avoid obtaining wrong or unreliableresults), the LSS can be represented by a family of mod-els. These models reflect the behavior of the LSS asviewed from various perspectives, called [36.7] levels ofdescription or strata, or levels of influence [36.63, 68].The description levels are governed by independentlaws and principles and use different sets of descrip-tive variables. The lower the level is, the more detailedthe description of a certain entity is. A unit placed onthe n-th level may be viewed as a subsystem at leveln −1. For example, the same manufacturing system canbe described from the top stratum in terms of economicand financial models, and, at the same time, by controlvariables (states, controls, and disturbances) as viewedfrom the middle stratum, or by physical and chem-ical variables as viewed from the bottom descriptionlevel (Fig. 36.2).

PartD

36.2

Large-Scale Complex Systems 36.2 Methods and Applications 623

Levels of ControlIn order to act in due time even in emergency situ-ations, when the available data are uncertain and thedecision consequences are not fully explored and eval-uated, a hierarchy of specialized control functions canbe an effective solution as shown by Eckman andLefkowitz [36.70]. Several examples of sets of levels ofcontrol are:

a) Regulation, optimization, and organization [36.71]b) Direct control, supervisory control, optimization,

and coordination [36.72]c) Stabilization, dynamic coordination, static opti-

mization, and dynamic optimization [36.8]d) Measurement and regulation, production planning

and scheduling, and business planning [36.73].

The levels of control, also called layers byMesarovic et al. [36.7], can be the result of a time-scaledecomposition. They can be defined on the basis oftime horizons taken into consideration, or the frequencyof disturbances which may show up in process vari-ables, operation conditions, parameters, and structureof the plant as stated by Schoeffler [36.68], as shownin Fig. 36.2.

Organization layers(echelons)

Generalmanagement

Productioncontrol

Unit processcontrol

Economic representations

Regulation Disturbance frequency in… process variables

… input variables

… plant structure

Optimization

Organization

Control modelsPhysical variables

Description levels(strata)

Control levels(layers)

Fig. 36.2 A hierarchical system approach applied to an industrial plant (after [36.69])

Levels of OrganizationThe hierarchies based on the complexity of orga-nization were proposed in mid 1960s by Brosilowet al. [36.74] and Lasdon and Schoeffler [36.75] andwere formalized in detail by Mesarovic et al. [36.7].The hierarchy with several levels of organization,also called echelons by Mesarovic et al. [36.7], hasbeen, for many years, a natural solution for man-agement of large-scale military, industrial, and socialsystems, which are made up of several intercon-nected subsystems when a centralized scheme can-not be either technically possible or economicallyacceptable.

The central idea of the multiechelon hierarchy isto place the control/decision units, which might havedifferent objectives and information bases, on severallevels of a management and control pyramid. Whilethe multilayer systems implement the vertical divi-sion of the control effort, the multiechelon systemsinclude also a horizontal division of work. Thus, onthe n-th organization level the i-th control unit, CUn

i ,has limited autonomy. It sends coordination signalsdownwards to a well-defined subset of control unitswhich are placed at the level n −1 and it receives co-ordination signals from the corresponding unit placed

PartD

36.2


Coordinator

Level 2

Level 1

w y

CU2 CU3CU1

SSy2

H

SSy3SSy1

m1 m2u1

z1 z2 z3

y1 y2

α2 α3α1

�2 �3�1

y3

m3u3u2

Fig. 36.3 A simple two-level multilevel control system (CU =control unit, SSy = controlled subsystem, H = interconnectionfunction, m = control variable, u = input interconnection variable,z = output interconnection variable, y = output variable, w = dis-turbance)

on level n + 1. The unit on the top of the pyra-mid is called the supremal coordinator and the unitsto be found at the bottom level are called infimalunits.

Manipulationof Complex Mathematical Problems

To take advantage of possible benefits of hierarchicalmultilevel systems a systematic decomposition of theoriginal large-scale system and associated control prob-lem is necessary. There are many situations when thecontrol problem may be formulated as (or reduced to) anoptimization problem (P1), which is, defined in generalterms as

(P1) : extrv

J(v) ; v ∈ V , (36.1)

where v is the decision variable (a scalar, or a vector),V is the admissible variation domain (which can bedefined by differential or difference equations and/or al-gebraic inequations), and J is the performance measure(which can be a function or a functional).

The decomposition methods are based on var-ious combinations of several elementary manipula-tions [36.76]. There are two main subsets of elementarymanipulations:

a) Transformations, which are meant to substitute theoriginal large-scale complex problem by a more ma-nipulable one

b) Decompositions, which are meant to replace a large-scale problem by a number of smaller subproblems.

In the sequel several elementary manipulationswill be reviewed following the lines exposed byWilson [36.76].

The variable transformation replaces the originalproblem (P1) by an equivalent one (P2) through theutilization of a new variable y = f (x) and a new per-formance measure Q(y) and admissible domain Y , sothat there is the inverse function v = f −1(v). The newproblem is defined as

(P2) : extry

Q(y) ;(y ∈ Y ) , (∀y) = f (v)[Q(y) = J(v)] .

(36.2)

The Lagrange transformation can simplify the admissi-ble domain; for example, let the domain V be definedby complicated equalities and inequalities

V = {v : (v ∈ V1) , (g0(v) , g−(v) ≤ 0)

}, (36.3)

where V1 is a certain set, g0 and g− represent equalityand inequality constraints, respectively.

A Lagrangian can be defined

L(v, π, γ ) = J(v)+〈π, g0(v)〉−〈γ , g−(v)〉 , (36.4)

where π are the Lagrange multipliers, γ are the Kuhn–Tucker multipliers, and 〈·, ·〉 is the scalar product.

If L possesses a saddle point, the solution of (P1)is also the solution of the transformed problem (P2)defined as

(P2) : maxπγ

minv

[L(v, π, γ )] ; v ∈ V1 . (36.5)

The manipulation called evolving the problem is uti-lized when not all parameters are known or the prioritiesand the constraints are subject to alteration in time. Insuch situations, the problem is solved even under uncer-tainties and then is reformulated to take into account theaccumulation of new information. The repetitive controlproposed by Findeisen et al. [36.16] is based on sucha transformation.

Having transformed the original problem into a con-venient form, a subset of smaller subproblems can beobtained through decomposition as shown in the sequel.

The partitioning of the large-scale problem can beapplied if several subsets of independent variables can

PartD

36.2


be identified; for example, let (P1) be defined as

(P1) : extr[J1(v1)+ J2(v2)] ; v1 ∈ V 1 ; v2 ∈ V 2 ,

(36.6)

then, two independent subproblems (P21) and (P22) canbe obtained

(P21) : extr[J1(v1)] ; v1 ∈ V 1 , (36.7)

(P22) : extr[J2(v2)] ; v2 ∈ V 2 . (36.8)

This decomposition is utilized in assigning separatesubproblems to the controllers which are situated atthe same level of a hierarchical pyramid or in de-centralized control schemes where the controllers actindependently.

The parametric decomposition divides the large-scale problem into a pair of subproblems by settingtemporary values to a set of coupling parameters. Whilein one problem of the pair the coupling parameters arefixed and all other variables are free, in the second sub-problem they are free and the remaining variables arefixed as solutions of the first subproblem. The two sub-problems are solved through an iterative scheme whichstarts with a set of guessed values of the coupling pa-rameters; for example, let the large-scale problem bedefined as follows

(P1) : extrv

[J(v)] ; v = (α, β) ; (α ∈ A) , (β ∈ B) ;αRβ ,

where α and β are the components of v, A and B aretwo admissible sets, β is the coupling parameter, and Ris a relation between α and β. The problem (P1) can bedivided into the pair of subproblems (P2) and (P3)

(P2) : extrα

[J

(α,

∗β

)]; (α ∈ A) , (αRβ) ,

(P3) : extrβ

[J

(∗α (β), β

)]; (β ∈ B) ,

{∃ ∗

α[(∗

α∈ A)

,(∗α Rβ

)]},

where∗α (β) is the solution of (P2) for the given value

β= ∗β and

∗β is the solution of (P3) for the given value

α= ∗α.The parametric decomposition is utilized to divide

the effort between a coordinating unit and the subsetof coordinated units situated at lower organization level(echelon).

The structural decomposition divides the large-scale problem into a pair of subproblems through

modifying the performance measure and/or con-straints. While one subproblem consists in setting thebest/satisfactory formulation of the performance mea-sure and/or admissible domain, the second one is to findthe solution of the modified problem. This manipulationis utilized to divide the control effort between two levelsof control (layers).

From Coordination to CooperationThe traditional multilevel systems proposed in the1970s to be used for the management and con-trol of large-scale systems can be viewed as purehierarchies [36.77]. They are characterized by thecirculation of feedback and intervention signals onlyalong the vertical axis, up and down, respec-tively, in accordance with traditional concepts ofthe command and control systems. They consti-tuted a theoretical basis for various industrial dis-tributed control systems which possess at highestlevel a powerful minicomputer. Also the multilayerand multiechelon hierarchies served in the 1980sas a conceptual reference model for the efforts todesign computer-integrated manufacturing (CIM) sys-tems [36.78, 79].

Several new schemes have been proposed over thelast 25 years to overcome the drawbacks and limits ofthe practical management and control systems designedin accordance with the concepts of pure hierarchiessuch as: inflexibility, difficult maintenance, and lim-ited robustness to major disturbances. The more recentsolutions exhibit ever more increased communicationand cooperation capabilities of the management andcontrol units. This trend has been supported by theadvances in communication technology and artificial in-telligence; for example, even in 1977, Binder [36.80]introduced the concept of decentralized coordinatedcontrol with cooperation, which allowed limited com-munication among the control unit placed at the samelevel. Several years later, Hatvany [36.81] proposed theheterarchical organization, which allows for exchangeof information among the units placed at various levelsof the hierarchy.

The term holon was first proposed by Koestler in1967 [36.82] with a view to describing a general orga-nization scheme able to explain the evolution and life ofbiological and social systems. A holon cooperates withother holons to build up a larger structure (or to solvea complex problem) and, at the same time, it works to-ward attaining its own objectives and treats the varioussituations it faces without waiting for any instructionsfrom the entities placed at higher levels. A holarchy is

PartD

36.2


a hierarchy made up of holons. It is characterized byseveral features as follows [36.83]:

• It has a tendency to continuously grow up by attract-ing new holons.• The structure of the holarchy may permanentlychange.• There are various patterns of interactions amongholons such as: communication messages, negotia-tions, and even aggressions.• A holon may belong to more than one holarchy if itobserves their operation rules.• Some holarchies may work as pure hierarchies andothers may behave as heterarchical organizations.

Figure 36.4 shows an object-oriented representation ofa holarchy. The rectangles represent various classes ofobjects such as pure hierarchies, heterarchical systems,channels, and holons. This shows that the class of ho-larchies may have particular subclasses such as purehierarchies and heterarchical systems. Also a holarchyis composed of several constituents (subclasses) suchas: holons (at least one coordinator unit and two infi-mal/coordinated units in the care of pure hierarchies)and channels for coordination (in the case of pure hier-archies) or channels for cooperation (in the case of pureheterarchies). Coordination channels link the supremalunit to, at least, two infimal units. While there are, atleast, two such coordination links in the case of purehierarchies, a heterarchical system may have no suchlink. While, at least, one cooperation channel is presentin a heterarchical system, no such a link is allowed ina pure hierarchy.

0+ 2+ 2+ 1+3+

n+

higher class has as particular forms…

higher class is made up of…

there may be n or more objects related to the class…

there may be none, one or more objects related to the class…

Holarchy

Heterarchical systemPure hierarchy

Horizontal channelfor cooperation

Vertical channelfor coordination

Holon

Fig. 36.4 Holarchies: an object-oriented description

Management and control structures based onholarchy concepts were proposed by Van Brusselet al. [36.84], Valckenaers et al. [36.85] for implemen-tation in complex discrete-part manufacturing systems.

To increase the autonomy of the decision andcontrol units and their cooperation the multiagent tech-nology is recommended by Parunak [36.57] and Hadeliet al. [36.53]. An intelligent software agent encapsu-lates its code and data, is able to act in a proactiveway, and cooperates with other agents to achieve a com-mon goal [36.86]. The control structures which utilizethe agent technology have the advantage of simplify-ing industrial transfer by incorporating existing legacysystems, which can be encapsulated in specific agents.Marık and Lazansky [36.55] make a survey of industrialapplications of agent technologies which also considerspros and cons of agent-based systems. They also presenttwo applications:

a) A shipboard automation system which providesflexible and distributed control of a ship’s equip-ment

b) A production planning and scheduling systemwhich is designed for a factory with the possi-bility of influencing the developed schedules bycustomers and suppliers.

36.2.2 Other Methods and Applications

Decentralized ControlFeedback control of large-scale systems poses the stan-dard control problem: to find a controller for a givensystem with control input and control output ensur-ing closed-loop systems stability and reach a suitableinput–output behavior. The fundamental difference be-tween small and large systems is usually described bya pragmatic view: a system is large if it is conceptuallyor computationally attractive to decompose it into inter-connected subsystems. Such subsystems are typically ofsmall size and can be solved easier than the original sys-tem. The subsystem solutions can be combined in somemanner to obtain a satisfactory solution for the overallsystem [36.87].

Decentralized control has consistently been a con-trol of choice for large-scale systems. The prominentreason for adopting this approach is its capability tosolve effectively the particular problems of dimen-sionality, uncertainty, information structure constraints,and time delays. It also attenuates the problems thatcommunication lines may cause. While in the hierar-chical control schemes, as shown above, the control

PartD

36.2


units are coordinated through intervention signals andmay be allowed to exchange cooperation messages,in decentralized control, the units are completely in-dependent or at least almost independent. This meansthat the information flow network among the controlunits can be divided into completely independent par-titions. The units that belong to different subnetworksare completely separate from each other. Only restrictedcommunication at certain time moments or intervals orlimited to small part of information among the unitsis allowed. Decentralized structures are often used buttheir performance is worse compared with the central-ized case. The basic decentralized control schemes areas follows:

• Multichannel system. The global system is con-sidered as one whole. The control inputs and thecontrol outputs operate only locally. This means thateach channel has available only local informationabout the system and influences only a local part ofthe system.• Interconnected systems. The overall system is de-composed according to a selected criterion. Thenlocal controllers are designed for each subsystems.Finally, the local closed-loop subsystems and inter-connections are tested to satisfy the desired overallsystem requirements.

At present a serious problem is the lack of rele-vant theoretic and methodological tools to supportthe scalable solution of new networked complexlarge-scale problems including asynchronous issues.The recent accomplishments are aimed at broaden-ing the scope of decentralized control design methodsusing linear matrix inequalities (LMIs) [36.31], dy-namic interaction coordinator design to ensure thedesired level of interconnections [36.32], advanced de-centralized control strategies for complex switchingsystems [36.26], hybrid large-scale systems [36.27],Petri nets [36.25], large-scale supply chain decentral-ized coordination [36.28, 29], and distributed controlsystems with network communication [36.30].

Simulation-Based Schedulingand Control in LSS

In continuous, large-scale industrial plants such as inchemical, power, and paper industries and waste-watertreatment plants, simulation-based scheduling startsfrom creating scenarios for production and comparingthese scenarios for optimality and availability. Prob-lems can vary from order allocation between multiple

production lines to optimal storage usage and detec-tion and compensating for bottlenecks. Heuristic rulesare usually connected to simulation, making it possi-ble to adjust the production to varying customer needs,minimize the use of raw materials and energy, decreasethe environmental load, stabilize or improve the quality,etc. Early applications in the paper industry are givenby Leiviskä et al. [36.38, 39]. The main problem is tobalance the production and several intermediate stor-ages in (multiple) production lines, and give room formaintenance shutdowns and coordinate production ratechanges. The model is based on the state model withstorage capacities as the state variables and productionrates as the control variables. Heuristics and bottleneckconsiderations are connected to these systems. A newer,agent-based solution has also been proposed [36.52].There are also several classical optimization-based so-lutions for this problem [36.19, 21–24].

Modern chemical batch processes are large scale,complex, serial/parallel, multipurpose processes. Theyare especially common in the food and fine chemi-cals industries. They resemble flexible manufacturingsystems common in electronics production. From thescheduling and control point of view complexity bringsalong also difficult interactions and uncertainty that aredifficult to tackle with conventional tools. Simulation-based scheduling can include as much complexity asneeded, and it is a largely used tool in the eval-uation of the performance of different optimizingsystems. Connecting heuristics or rule-based systemsto simulation makes it also a flexible tool for batchprocess scheduling. Modeling approaches differ, e.g.,real-time simulation using Petri nets [36.36] and thecombination of discrete event simulation with geneticalgorithms for the steel annealing shop have been pro-posed [36.40].

Flexible manufacturing systems, e.g., for compo-nents assembly, offer several difficulties for productionscheduling and control. Dynamic, random nature is onemain concern in operation control. Also quickly chang-ing products and production environments, especiallyin electronics production, lead to a great variabilityin requirements for production control. In real cases,it is also typical that several scenarios must be cre-ated and evaluated. The handling of uncertain andvague information itself causes also problems in real-world applications. Uncertain data has to be extractedfrom data sources avoiding noise, or at least avoidingincreasing it.

Discrete event simulation models the system as itpropagates over time, describing the changes as separate

PartD

36.2


discrete events. This approach also found a lot of appli-cations in manufacturing industries, queuing systems,and so on. An early application to jobshop schedul-ing is presented by Filip et al. [36.20], who utilizevarious combinations of several dispatching rules to cre-ate the list of future events. Taylor [36.17] reportedon an application of discrete event simulation, com-bined with heuristics, to the scheduling of the printedcircuit board (PCB) assembly line. The situation is com-plicated by the fact that the production control mustoperate on three levels: at the system level concerningproduction mix problems, at the cell level for routingproblems, and at the machine level to solve sequenc-ing problems. Discrete event simulation is also the keyelement in the shop floor scheduling system proposedby Gupta et al. [36.35]. The procedure starts by creat-ing feasible schedules for the telephone terminals plant,helps in taking other requirements into account and intackling uncertainties, and makes rescheduling possible.A system integrating simulation and neural networkshas been used in photolithography toolset schedul-ing in wafer production [36.33]. The system uses theweighted-score approach, and the role of the neural net-work is to update the weights set to different selectioncriteria. Fuzzy logic provides the arsenal of methods fordealing with uncertainties. Several examples for PCBproduction are given by Leiviskä [36.45].

Two-stage approaches have been used in bottleneck-based approaches [36.34]. The first-pass simulationrecognizes the bottlenecks, and their operation are opti-mized during the second-pass simulation. Better controlof work in bottlenecks improves the performance of thewhole system. The main dispatching rule is to group to-gether the lots that need the same setups. The systemalso reveals the non-bottleneck machines and makes itpossible to apply different dispatching rules accordingto the process state. The example is from semiconductorproduction.

In practice, scheduling is a part of the decisionhierarchy starting from the enterprise-level strategic de-cisions and going down to machine-level order or toolsscheduling. Simulation is used at different levels of thishierarchy to provide interactive means for guarantee-ing the overall optimality or at least the feasibility ofthe decisions made at different levels. Such integratedand interactive approaches exist also in supply-chainmanagement systems. In large-scale manufacturing sys-tems, supply-chain control must take four interactingfactors into account: suppliers, manufacturing, distri-bution, network, and customers. To control all theseinteractions successfully, various operating factors and

constraints – processing times, production capacities,availability of raw materials, inventory levels, and trans-portation times – must be considered.

Discrete event simulation is also one possibilityto create an object-oriented, scalable, simulation-basedcontrol architecture for supply-chain control [36.41].Requirements for modularity and maintainability alsolead to distributed simulation models, especially whena simulation-based control architecture is controllingsupply chain interactions. This means a modeling tech-nique including a federation of simulation models thatare solved in a coordinated manner. The system archi-tecture is presented in [36.42]. Each supply-chain entityhas two simulation models associated with it – onerunning in real time and the other as a lookahead sim-ulation. The lookahead model is capable of predictingthe impact of a disturbance observed by the real-timemodel. A federation object coordinator (FOC) coor-dinates the real-time simulation models. In this case,a master event calendar allocates interprocess eventsto all simulation models and resynchronizes all simu-lations at the end of every activity [36.37].

In simulation-based control the controller makes de-cisions based both on the current state of the system andfuture scenarios, usually produced by simulation. Here,the techniques for calculation of these scenarios playthe main role. Ramakrishnan and Thakur [36.42] pro-posed the extension sequential dynamic systems (SDS)that they call input–output SDS to model and analyzedistributed control systems and to compensate for theweaknesses of automata-based models. They use thediscrete-part production plant as an example.

Artificial Intelligence-Based Control in LSSArtificial intelligence (AI)-based control in large-scalesystems uses, in practice, all the usual methods of intel-ligent control: fuzzy logic, neural networks, and geneticalgorithms together with different kinds of hybrid solu-tions [36.88]. The complex nature of applications makesthe use of intelligent systems advantageous. Dealingwith this complexity is also the biggest challenge for themethodological development: the large-scale processstructures, complicated interconnections, nonlinearity,and multiple time scales make the systems difficultto model and control. Fuzzy logic control (FLC) hasfound most of its applications in cases which are dif-ficult to model, suffer from uncertainty or imprecision,and where a skilful operator is superior to conven-tional automation systems. Artificial neural networks(ANN) contribute to modeling and forecasting tasksand combined with fuzzy logic in neuro-fuzzy systems

PartD

36.2


combine the benefits of both approaches. Genetic algo-rithms (GA), which are basically optimization systems,are used in tuning models and controllers. See Chap. 14on Artificial Intelligence and Automation for additionalcontent.

As shown above, the control of large-scale industrialplants have usually been based on distributed hardwareand hierarchical design of control functions [36.89,90].The supervisory and local control levels lay under enter-prise and mill-wide control levels. Supervisory controlprovides the local controls with the set points that ful-fil the quality and schedule requirements coming fromthe mill-wide level and help in optimizing the operationof the whole plant. This optimization leaves room forversatile application of intelligent methods. Local unitson the other hand, control the actual process variablesaccording to the set points given by the supervisorycontrol level. Even though the proportional–integral–differential (PID) controller is by far the most importanttool, intelligent control plays an increasing role alsoat the local control level. Intelligent methods havebeen useful in tuning local PID controllers. In practice,fuzzy controllers must have adaptive capabilities. Gainscheduling is a typical approach for large-scale systems,but applications of model reference adaptive control andself-tuning adaptive control exist. Self-tuning has beenused in controlling a pilot-scale rotary drum where thedisturbances are due to long and varying time delaysand changes in the raw materials [36.46].

Model-based control techniques, e.g., model predic-tive control (MPC), have been applied for the control ofprocesses with a long delay or dead time. In MPC, thecontroller based on a plant model determines a manipu-lated variable profile that optimizes some performanceobjectives over the time in question. ANN are usedin replacing the mathematical models in optimizationas shown in a survey made by Hussain [36.48]. AlsoTakagi–Sugeno fuzzy models are used in connectionwith model-based predictive control [36.44]. Hybridsystems include both continuous- and/or discrete-timedynamics together with discrete events. So their stateconsists of real-valued, discrete-valued, and/or logicalvariables. Support vector machines have been used asa part of MPC strategy for hybrid systems [36.91].

Power systems have been an important applica-tion field for intelligent control since 1990s [36.92].Design of centralized controllers is difficult for manyobvious reasons: power systems are large scale and de-centralized by nature. They are also nonlinear and havemultiple dynamics and considerable time delays. De-centralized local control can apply linear models and

purely local measurements. Available transfer capabil-ity (ATC) is a real-time index used in monitoring andcontrolling the power transactions and avoiding over-loading of the transmission lines [36.47]. There aredifficulties in calculating it accurately online for large-scale systems. Decreasing the number of input variablesto only three and using fuzzy modeling helps in this.Simulations show that neural-networks-based local ex-citation controls can take care of interactions betweengenerators and dampen oscillations effectively. Neuralnetworks are used in approximating unknown dynamicsand interconnections [36.40]. The designing of the con-troller for two-area hydrothermal power systems basedon genetic algorithm improves the rise time and settlingtime, and simulations show that the proposed techniqueis superior to the traditional methods [36.51]. A localKalman filter and genetic algorithms estimate all localstates and interactions between subsystems in a large-scale power system. The controller uses these estimates,optimizes a given performance index, and then regulatesthe system states [36.50].

Agent-based technologies have been used in com-plex, distributed systems. Good examples come fromintelligent control of highly distributed systems in thechemical industry and in the area of utility distribu-tion (power, gas, and waste-water treatment). As shownabove, holonic agents take care of machine or cell-level(local) controls, sometimes even integrated with ma-chines. Intelligent agents can be associated with eachmanufacturing unit and they communicate, coordinatetheir activities, and cooperate with each other.

Fault detection and diagnosis (FDD) may be tack-led by decomposing the large-scale problem intosmaller subtasks and performing control and FDD lo-cally [36.93]. Large-scale complex power systems needsystematic tools for protection and control. The su-pervisory control technique and a design procedureof a supervisor that coordinates the behavior of relayagents to isolate fault areas are presented in [36.56].Multiagent systems have also been used in identifica-tion and control of a 600 MW boiler–turbine–generatorunit [36.54]. In this case, online identifiers are usedfor control and offline identifiers for fault diagnosis.Event-based approaches are used for building large-scale distributed systems and applications, especiallyin a networked environment. A hybrid approach ofevent-based communications for real-time manufac-turing supervisory control is applied for large-scalewarehouse management [36.94]. See Chap. 30 on Au-tomating Error and Conflict Prognostics and Preventionfor additional content.

PartD

36.2


Computer-Supported Decision Makingin Large-Scale Complex Systems

As shown above, a possible solution to many LSScontrol problems is the use of artificial-intelligencemethods. However, in the field, due to strange combi-nations of external influences and circumstances, rareor new situations may show up that were not taken intoconsideration at design time. Already in 1990, Martinet al. remarked that [36.95]:

although AI and expert systems were successful insolving problems that resisted to classical numeri-cal methods, their role remains confined to supportfunctions, whereas the belief that evaluation by manof the computerized solutions may become superflu-ous is a very dangerous error.

Based on this observation, Martin et al. [36.95] rec-ommended appropriate automation, which integratestechnical, human, organizational, economical, and cul-tural factors.

The decision support system concept (DSS) ap-peared in the early 1970s. As with any new term, thesignificance of DSS was in the beginning rather vagueand controversial. While some people viewed it as a newredundant term used to describe a subset of manage-ment information systems (MIS), some other argued itwas a new label abusively used by some vendors to takeadvantage of a new fashion. Since then many researchand development activities and applications have wit-nessed that the DSS concept definitely meets a real needand there is a market for it even in the context of real-time applications in the industrial milieu [36.96, 97].

The Nobel Prize winner H. Simon [36.98] identi-fied three steps of the decision-making (DM) process,namely:

a) Intelligence, consisting of activities such as data col-lection and analysis in order to recognize a decisionproblem

b) Design, including activities such as model statementand identification/production and evaluation of var-ious potential solutions to the problem

c) Choice, or selection of a feasible alternative for im-plementation.

Later, he added a fourth step – implementation andresult evaluation – which may correspond to supervi-sory control in industrial milieu. If a decision problemcannot be entirely clarified and all possible decision al-ternatives cannot be fully explored and evaluated before

a choice is made, then the problem is said to be unstruc-tured or semistructured. If the problem were completelystructured, an automatic device could have solved theproblem without any human intervention. On the otherhand, if the problem has no structure at all, nothingbut hazard can help. If the problem is semistructureda computer-aided decision can be envisaged.

Most of the developments in the DSS domain haveinitially addressed business applications not involvingany real-time control. However, even in the early 1980sDSS were reported to be used in manufacturing con-trol [36.20, 99]. In 1987, Bosman [36.100] stated thatcontrol problems could be looked upon as a naturalextension and as a distinct element of planning decision-making processes (DMP). Almost 20 years later, Nofet al. state [36.12]:

. . . the development and application of intelligentdecision support systems can help enterprises copewith problems of uncertainty and complexity, to in-crease efficiency, join competitively in productionnetworks, and improve the scope and quality of theircustomer relations management (CRM).

Real-time decision-making processes (RT DMPs)for control applications are characterized by several par-ticular aspects such as:

a) They involve continuous monitoring of a dynamicenvironment.

b) They are short time horizon oriented and are carriedout on a repetitive basis.

c) They normally occur under time pressure.d) Long-term effects are difficult to predict [36.101].

It is quite unlikely that an econological (eco-nomically logic) approach, involving optimization, betechnically possible for genuine RT DMPs. Satisficingapproaches, which reduce the search space at the ex-pense of the decision quality, or fully automated DMsystems, if taken separately, cannot be accepted either,but for some exceptions. At the same time, one cannotice that genuine RT DMP can show up in crisis sit-uations only; for example, if a process unit must beshut down due to an unexpected event, the productionschedule of the entire plant might become obsolete. Theright decision will be to take the most appropriate com-pensation measures to manage the crisis over the timeperiod needed to recomputed a new schedule or up-date the current one. In this case, a satisficing decisionmay be appropriate. If the crisis situation has been metpreviously and successfully surpassed, an almost auto-mated solution based on past decisions stored in the

PartD

36.2


Table 36.2 Possible task assignment in DSS

Decision steps and activities EU NU NM ES ANN CBR IA

Intelligence

• Setting objectives I M I/M P/M

• Perception of DM situation I M P M/I• Problem recognition I M P M/I

Design

• Model selection E M I I

• Model building M P I/M P

• Model validation I M• Setting alternatives P M P

Choice

• Model experimenting M/I– Model solving E E I

– Result interpreting I P P

– Parameter changing E M/I• Solution adotiing E• Sensitivity analysis M I

Release for implementation E E P

EU – expert user, NU – novice user, NM – numerical model, ES – rule-based expert system, ANN – artificialneural network, CBR – case-based reasoning, GA – genetic algorithm, IA – intelligent agent, P – possible, M –moderate, I – intensive, E – essential

information system can be accepted and validated bythe human operator. On the other hand, the minimiza-tion of the probability of occurrences of crisis situationsshould be considered as one of the inputs (expressed asa set of constraints or/and objectives) in the schedulingproblem [36.96, 102].

In many problems, decisions are made by a group ofpersons instead of an individual. Because the group de-cision is either a combination of individual decisions ora result of the selection of one individual decision, thismay not be rational in Simon’s acceptance. The groupdecision is not necessarily the best choice or a combi-nation of individual decisions, even though those mightbe optimal, because various individuals might havevarious perspectives, goals, information bases, and cri-teria of choice. Therefore, group decisions show a highsocial nature, including possible conflicts of interest,different visions, influences, and relations [36.103].Consequently, a group (or multiparticipant) DSS needsan important communication facility.

The generic framework of a DSS, proposed byBonczek et al. in 1980 [36.104] and refined later by

Holsapple and Whinston [36.105] is quite general andcan accommodate the most recent technologies andarchitectural solutions. It is based on three essentialcomponents. The first one is the language (and com-munications) subsystem (LS). This is used for:

a) Directing data retrieval, allowing the user to invokeone out of a number of report generators

b) Directing numerical or symbolic computation, en-abling the user either to invoke the models by namesor construct model and perform some computationat his/her free will

c) Maintaining knowledge and information in the sys-tem

d) Allowing communication among people in case ofa group DM

e) Personalizing the user interface.

The knowledge subsystem (KS) normally contains:

a) Empirical knowledge about the state of the applica-tion environment in which the DSS operates

PartD

36.2


b) Modeling knowledge, including basic modelingblocks and computerized simulation and optimiza-tion algorithms to use for deriving new knowledgefrom the existing knowledge

c) Derived knowledge containing the constructed mod-els and the results of various computations

d) Meta-knowledge (knowledge about knowledge)supporting model building and experimentation andresult evaluation

e) Linguistic knowledge allowing the adaptation ofsystem vocabulary to a specific application

f) Presentation knowledge to allow for the most appro-priate information presentation to the user.

The third essential component of a DSS is theproblem processing subsystem (PPS), which enablescombinations of abilities and functions such as in-formation acquisition, model formulation, analysis,evaluation, etc.

It has been noticed that some DSS are orientedtowards the left hemisphere of the human brain andsome others are oriented towards the right hemisphere.While in the first case quantitative and computationalaspects are important, in the second pattern recogni-tion and reasoning based on analogy prevail. In thiscontext, there is a significant trend towards combin-

ing numerical models and models that emulate thehuman reasoning to build advanced DSS [36.106].A great number of optimization algorithms have beendeveloped and carefully tested so far. However, their ef-fectiveness in decision making has been limited. Overthe last three decades traditional numerical methodshave, along with databases, been essential ingredients ofDSS. From an information technology perspective, theirmain advantages [36.107] are: compactness, computa-tional efficiency (if the model is correctly formulated),and the market availability of software products. Onthe other hand, they present several disadvantages. Be-cause they are the result of intellectual processes ofabstraction and idealization, they can be applied toproblems which possess a certain structure, which ishardly the case in many real-life problems. In addi-tion, the use of numerical models requires that theuser possesses certain skills to formulate and experi-ment the model. As was shown in the previous section,AI-based methods supporting decision making are al-ready promising alternatives and possible complementsto numerical models. New terms such as tandem sys-tems, or expert DSS (XDSS) have been proposed forsystems that combine numerical models with AI-basedtechniques. An ideal task assignment is given in Ta-ble 36.2 [36.97].

36.3 Case Studies

The following case studies illustrate how combinationsof methods may be utilized to solve large-scale complexproblems.

Digester s2 s1

s4

s3s6

s5

m1 w1

m4 m5

m2

m3

Bleaching

Auxiliaryboiler

Recoveryboiler

Dryingmachine

Caustici-zation

Evaporation

Fig. 36.5 Pulp mill model

36.3.1 Case Study 1:Pulp Mill Production Scheduling

Figure 36.5 shows the pulp mill modeled as a commonstate-space system. The state of the system s(t) is de-scribed by the amount of material in each storage tank.The production rates of the processes are chosen ascontrol variables forming the control vector m(t). Therequired pulp production is usually taken as a determin-istic known disturbance vector w(t).

The operation of the plant presented in Fig. 36.5 isdescribed by the vector–matrix differential equation

ds(t)

dt= Bm(t)+Cw(t) ,

where B and C are coefficient matrices describing therelationships between the model flows (transfer ratios).

PartD

36.3

Large-Scale Complex Systems 36.3 Case Studies 633

Since the most storage tanks have only one input flowand one output flow, most elements in B and C matricesequal zero.

If the steam balance (dashed line in Fig. 36.5) is in-cluded in scheduling, an additional variable describingthe steam development in the auxiliary boiler is re-quired. It is a scalar variable denoted by S. Accordingly,the steam balance is

S(t) = Dm(t)+ Ew(t) .

Note that the right-hand side of the balance includesboth consumption and generation terms. The variablesin the model are constrained by the capacity limits oftanks and processes in the following way

smin ≤ s(t) ≤ smax ,

mmin ≤ m(t) ≤ mmax ,

Smin ≤ S(t) ≤ Smax .

Due to the fact that scheduling is concerned withrelatively long time intervals, no complete and com-plicated process models are necessary. If all the smallstorage tanks are included in the model, the systemdimensions increase and it becomes difficult to dealwith. These tanks also have no meaning from the con-trol point of view. Simpler model follows by combiningsmall storage tanks.

There are several ways to solve the schedulingproblem as shown before. Optimization can benefitfrom decomposition and solving of smaller problemsas described in Sect. 36.2.1. A review of methods ispresented in [36.39]. It seems, however, that no ap-proach alone can deal with this problem successfully.Hybrid systems, consisting of algorithmic, rule-based,and intelligent parts integrated with each other, andalso agent-based systems, could be the best possibleanswer [36.97, 110].

36.3.2 Case Study 2: Decision Supportin Complex Disassembly Lines

In [36.108], the control of a complex industrial disas-sembly process of out-of-use manufactured products isstudied. The disassembly processes are subject to uncer-tainties. The most difficult problem in such systems isthat a disassembly operation can fail at any moment be-cause of the product or component degradation. In thiscase one has to choose between applying an alternativedisassembly destructive operation (dismantling), andaborting the disassembly procedure. This decision mustbe taken in real time because in a used product the com-ponents states are not known from the beginning of the

Product

Model

Planner

Ethernetnetwork

DB + KB

Simulation

DSS

Direct control system

Fig. 36.6 DSS integration in the multilayer control system (af-ter [36.108])

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Delay

Median

Standard

100

90

80

70

60

50

40

30

20

10

0

Fig. 36.7 The results of time delay estimation for one group (af-ter [36.109])

process. The solution is to integrate a decision supportsystem (DSS) in the architecture of a multilayer system.As shown in Fig. 36.6, the control and decision tasks aredistributed among three levels: planning, decision sup-port, and direct control. The disassembly planner givesthe sequence of the components that must be separatedto achieve the target component. The planner fuses theinformation from the artificial vision system with thatcontained in the database for each component or sub-assembly. A model of the product is generated. The DSSintegrates the model and performs the simulation to rec-

PartD

36.3


ommend a good disassembly sequence with respect tothe economical criteria.

36.3.3 Case Study 3: Time Delay Estimationin Large-Scale Complex Systems

In data-based modeling of large-scale complex systems,the exact determination of time delays is extremely dif-ficult. The methods for delay estimation are widelystudied in control engineering, but these studies aremainly limited to the two-variable cases, i. e., estimatingthe delay between the manipulated and the controlledvariable in the feedback control loop. The situation istotally different when dealing with a large number ofvariables grouped in several groups for modeling ormonitoring purposes.

Mäyrä et al. [36.109] discuss a delay estimationscheme combining genetic algorithms and principalcomponent analysis (PCA). Delays are optimized withgenetic algorithms with objective functions based onPCA. Typically, a genetic algorithm maximizes thevariance explained by the first or two first principalcomponents. The paper gives an example using simu-lation data of the paper machine, which includes over50 variables. The variables were first grouped basedon the cross-correlation and graphical analysis into fivegroups, and delays were estimated both for the variablesinside the groups and between the groups. The resultsfor one group of 15 variables are given in Fig. 36.7.The estimation was repeated 60 times and the fig-ure shows the median and standard variance of thesesimulations.

36.4 Emerging Trends

Large-scale complex systems have become a researchand development domain of automation with a seriesof rather established method and technologies and in-dustrial application. Table 36.3 contains a summary ofreferences to basic concepts.

Table 36.3 Key to references on basic concepts

Basic books Mesarovic, Macko, Takahara [36.7]; Wismer [36.61]; Titli [36.17];

Ho and Mitter [36.62]; Sage [36.63]; Siljak [36.3, 64]; Singh [36.65];

Findeisen et al. [36.16]; Jamshidi [36.6]; Lunze [36.66];

Brdys and Tatjewski [36.18]

Hierarchies Mesarovic, Macko, Takahara [36.7]

Strata Sage [36.63]; Schoeffler [36.68]

Layers Findeisen [36.8]; Havlena and Lu [36.73]; Isermann [36.72];

Lefkowitz [36.111]; Schoeffler [36.68]; Brdys and Ulanicki [36.112]

Echelons Brosilow, Lasdon and Pearson [36.74]; Lasdon and Schoeffler [36.75]

Heterarchy Hatvany [36.81]

Holarchy Hop and Schaeffer [36.83]; Koestler [36.82]; Van Brussel et al. [36.84];

Valckenaers et al. [36.85]

Decision support systems Bonczek, Holsapple and Whinston [36.104]; Bosman [36.100];

Chaturverdi et al. [36.101]; De Michelis [36.103]; Dutta [36.107]; Filip [36.96];

Filip et al. [36.21]; Filip, Donciulescu and Filip [36.97];

Holsapple and Whinston [36.105]; Kusiak [36.106]; Martin et al. [36.95];

Nof [36.99]; Nof et al. [36.12]; Simon [36.98]

At present academics and industrial practitioners areworking to adapt the methods and practical solutions inthe LSS field to modern information and communica-tion technologies and new enterprise paradigms. Severalsignificant trends which can be noticed or forecast are:

PartD

36.4

Large-Scale Complex Systems References 635

• A promising modern form to coordinate the ac-tions of the intelligent agents is stigmergy. This isinspired by the behavior of social insects whichuse a form of indirect communication mediatedby an active environment to coordinate their ac-tions [36.53].• Advanced decentralized control strategies for large-scale complex systems have recently been ex-tended into new applied areas, such as flex-ible structures [36.113, 114], Internet conges-tion control [36.115], aerial vehicles [36.116],or traffic control [36.117], to mention a few ofthem.• Recent theoretic achievements in decentralized con-trol can be progressively extended into the areas

of integrated/embedded control, distributed control(over communication networks), hybrid/discrete-event systems and networks, and autonomous sys-tems to serve as a very efficient tool to solve variouslarge-scale control problems.• Incorporation and combination of newly developednumeric optimization and simulation models andsymbolic/and connectionist or agent-based will con-tinue in an effort to reach the unification of humans,numerical models, and AI-based tools.• Mobile communications and web technology willbe ever more considered in LSS management andcontrol applications. In multiparticipant DSS, peo-ple will make co-decisions in virtual teams, nomatter where they are temporarily located.

References

36.1 R. Tomovic: Control of large systems. In: Simulationof Control Systems, ed. by I. Troch (North Holland,Amsterdam 1972) pp. 3–6

36.2 M.S. Mahmoud: Multilevel systems control and ap-plications, IEEE Trans. Syst. Man Cybern. SMC–7,125–143 (1977)

36.3 D.D. Siljak: Large Scale Dynamical Systems: Sta-bility and Structure (North Holland, Amsterdam1978)

36.4 Solver.com: Premium Solver Platform for Excelwww.solver.com (2007)

36.5 M. Athans: Advances and open problems in thecontrol of large-scale systems, Plenary paper, Proc.7th IFAC Congr. (Pergamon, Oxford 1978), 871–2382

36.6 M. Jamshidi: Large Scale Systems: Modeling andControl (North Holland, New York 1983) 2nd edn.(Prentice Hall, Upper Saddle River 1997)

36.7 M.D. Mesarovic, D. Macko, Y. Takahara: Theoryof Hierarchical Multilevel Systems (Academic, NewYork 1970)

36.8 W. Findeisen: Decentralized and hierarchical con-trol under consistence or disagreements of inter-ests, Automatica 18(6), 647–664 (1982)

36.9 S. Takatsu: Coordination principles for two-levelsatisfactory decision-making systems, Syst. Sci.7(3/4), 266–284 (1982)

36.10 F.G. Filip, D.A. Donciulescu: On an online dynamiccoordination method in process industry, IFAC J.Autom. 19(3), 317–320 (1983)

36.11 L. Mårtenson: Are operators in control of com-plex systems?, Proc. 13th IFAC World Congr., Vol. B(Pergamon, Oxford 1990) pp. 259–270

36.12 S.Y. Nof, G. Morel, L. Monostori, A. Molina,F.G. Filip: From plant and logistics control to multi-enterprise collaboration, Annu. Rev. Control 30(1),55–68 (2006)

36.13 A.P. Sage, C.D. Cuppan: On the system engineeringof systems of systems and federations of systems,Inf. Knowl. Syst. Manage. 2(4), 325–349 (2001)

36.14 A.V. Gheorghe: Risks, vulnerability, maintainabil-ity and governance: a new landscape for criticalinfrastructures, Int. J. Crit. Infrastruct. 1(1), 118–124(2004)

36.15 A.V. Gheorghe: Integrated Risk and VulnerabilityManagement Assisted by Decision Support Systems.Relevance and Impact on Governance (Springer,Dordrecht 2005)

36.16 W. Findeisen, M. Brdys, K. Malinowski, P. Tat-jewski, A. Wozniak: Control and Coordination inHierachical Systems (Wiley, Chichester 1980)

36.17 A. Titli: Commande hierarchisée des systemes com-plexes (Dunod Automatique, Paris 1975)

36.18 M. Brdys, P. Tatjewski: Iterative Algorithms forMultilayer Optimizing Control (Imperial College,London 2001)

36.19 A. Dourado Correia: Optimal scheduling and energymanagement in industrial complexes: some newresults and proposals, Preprints CIM Process Man-ufact. Ind. IFAC Workshop (Pergamon Press, Espoo1992) pp. 139–145

36.20 F.G. Filip, G. Neagu, D. Donciulescu: Jobshopscheduling optimization in real-time productioncontrol, Comput. Ind. 4(3), 395–403 (1983)

36.21 F.G. Filip, D. Donciulescu, R. Gaspar, M. Mu-ratcea, L. Orasanu: Multilevel optimisation algo-rithms in computer aided production control inthe process industries, Comput. Ind. 6(1), 47–57(1985)

36.22 M. Guran, F.G. Filip, D.A. Donciulescu, L. Orasanu:Hierarchical optimisation in computer dispatchersystems in process industry, Large Scale Syst. 8,157–167 (1985)

PartD

36

www.solver.com


36.23 J. Pettersson, U. Persson, T. Lindberg, L. Ledung,X. Zhang: Online pulp mill production optimiza-tion, Proc. 16th IFAC World Congr. (Prague 2005),on CD ROM

36.24 H. Tamura: Decentralised optimization for distrib-uted-lag models of discrete systems, Automatica11, 593–602 (1975)

36.25 A. Aybar, A. Iftar, H. Apaydin-Ozkan: Centralizedand decentralized supervisory controller design toenforce boundedness, liveness, and reversibility inPetri nets, Int. J. Control 78, 537–553 (2005)

36.26 L. Bakule: Stabilization of uncertain switched sym-metric composite systems, Nonlinear Anal.: HybridSyst. 1, 188–197 (2007)

36.27 F. Borrelli, T. Keviczky, G.J. Balas, G. Stew-ard, K. Fregene, D. Godbole: Hybrid DecentralizedControl of Large Scale Systems, Hybrid Systems:Computation and Control (Springer, Heidelberg2005) pp. 168–183

36.28 G. Inalham, J. How: Decentralized inventory con-trol for large-scale supply chains, Proc. Am. ControlConf. (Minneapolis 2006) pp. 568–575

36.29 P. Krishnamurthy, F. Khorrami, D. Schoenwald:Computationally tractable inventory control forlarge-scale reverse supply chains, Proc. Am. Con-trol Conf. (Minneapolis 2006) pp. 550–555

36.30 C. Langbort, V. Gupta, R.M. Murray: Distributedcontrol over falling channels. In: Networked Em-bedded Sensing and Control, ed. by P. Antsaklis,P. Tabuada (Springer, Berlin 2006) pp. 325–342

36.31 D.D. Siljak, A.I. Zecevic: Control of large-scale sys-tems: beyond decentralized feedback, Annu. Rev.Control 29, 169–179 (2005)

36.32 D.D. Siljak: Dynamic Graphs. Plenary paper. TheInternational Conference on Hybrid Systems andApplications (University of Louisiana, Lafayette2006)

36.33 A. Arisha, P. Young: Intelligent simulation-basedlot scheduling of photolithography toolsets ina wafer fabrication facility, Proc. 2004 WinterSimul. Conf. (Washington 2004) pp. 1935–1942

36.34 C.S. Chong, A.I. Sivakumar, R. Gay: Simulationbased scheduling using a two-pass approach,Proc. 2003 Winter Simul. Conf. (New Orleans 2003)pp. 1433–1439

36.35 A.K. Gupta, A.I. Sivakumar, S. Sarawgi: Shopfloorscheduling with simulation based proactive deci-sion support, Proc. Winter Simul. Conf. (San Diego2002) pp. 1897–1902

36.36 S. Julia, R. Valette: Real-time scheduling of batchsystems, Simul. Pract. Theory 8, 307–319 (2000)

36.37 S. Lee, S. Ramakrishnan, R.A. Wysk: A federationobject coordinator for simulation based controland analysis, Proc. Winter Simul. Conf. (San Diego2002) pp. 1986–1994

36.38 K. Leiviskä, P. Uronen, H. Komokallio, H. Auras-maa: Heuristic algorithm for production control of

an integrated pulp and paper mill, Large Scale Syst.3, 13–25 (1982)

36.39 K. Leiviskä: Benefits of intelligent produc-tion scheduling methods in pulp mills, Proc.CESA’96 IMACS Multiconf. Comput. Eng. Syst. Appl.Symp. Control Optim. Supervis., Vol. 2 (Lille 1996)pp. 1246–1251

36.40 Q.L. Liu, W. Wang, H.R. Zhan, D.G. Wang, R.G. Liu:Optimal scheduling method for a bell-type batchannealing shop and its application, Control Eng.Pract. 13, 1315–1325 (2005)

36.41 S. Ramakrishnan, S. Lee, R.A. Wysk: Implementa-tion of a simulation-based control architecture forsupply chain interactions, Proc. Winter Simul. Conf.(San Diego 2002) pp. 1667–1674

36.42 S. Ramakrishnan, M. Thakur: An SDS modeling ap-proach for simulation-based control, Proc. WinterSimul. Conf. (Orlando 2005) pp. 1473–1482

36.43 G.D. Taylor Jr: A flexible simulation framework forevaluating multilevel, heuristic-based productioncontrol strategies, Proc. Winter Simul. Conf. (NewOrleans 1990) pp. 567–569

36.44 A. Ichtev, J. Hellendoom, R. Babuska, S. Mollov:Fault-tolerant model-based predictive control us-ing multiple Takagi–Sugeno fuzzy models, Proc.IEEE Int. Conf. Fuzzy Syst. FUZZ-IEEE’02, Vol. 1 (Ho-nolulu 2002) pp. 346–351

36.45 K. Leiviskä: Applications of intelligent systems inelectronics manufacturing, Proc. 2nd Conf. Manag.Control Prod. Logist. MCPL’2000 (Grenoble 2000), onCD-ROM

36.46 K. Leiviskä, L. Yliniemi: Design of adaptive fuzzycontrollers. In: Do Smart Adaptive Systems Ex-ist?, ed. by B. Gabrys, K. Leiviskä, J. Strackeljan(Springer, Berlin 2005) pp. 251–266

36.47 B. Azhar, A.B. Khairuddin, S.S. Ahmed, M.W. Mustafa,A. Zin, H. Ahmad: A novel method for ATC compu-tations in a large-scale power system, IEEE Trans.Power Syst. 19(2), 1150–1158 (2004)

36.48 M.A. Hussain: Review of the applications of neuralnetworks in chemical process control-simulationand online implementation, Artif. Intell. Eng. 13,55–68 (1999)

36.49 W. Liu, J. Sarangapani, G.K. Venayagamoorthy,D.C. Wunsch, D.A. Cartes: Neural network baseddecentralized excitation control of large scalepower systems, Proc. Int. Jt. Conf. Neural Netw.(Vancouver 2006)

36.50 M. Dehghani, A. Afshar, S.K. Nikravesh: Decen-tralized stochastic control of power systems usinggenetic algorithms for interaction estimation, Proc.16th IFAC World Congr. (Prague 2005), on CD ROM

36.51 E.E. El Mdbouly, A.A. Ibrahim, G.Z. El-Far, M. ElNassef: Multilevel optimization control for large-scale systems using genetic algorithms, Proc. 2004Int. Conf. Electr., Electron. Comput. Eng. ICEEC ’04(Cairo 2004) pp. 193–197

PartD

36

Large-Scale Complex Systems References 637

36.52 R. Akkiraju, P. Keskinocak, S. Murthy, F. Wu:An agent-based approach for scheduling multiplemachines, Appl. Intell. 14(2), 135–144 (2001)

36.53 H. Hadeli, P. Valckenaers, C.B. Zamfirescu, H. VanBrussel, B.S. Germain: Self-organising in multi-agent coordination and control using stigmergy.In: Self-Organising Applications: Issues, chal-lenges and trends. Lecture Notes in ArtificialIntelligence, Vol. 2977, ed. by H. Hadeli (Springer,Heidelberg 2004) pp. 325–340

36.54 J.S. Heo, K.Y. Lee: A multi-agent system-basedintelligent identification system for power plantcontrol and fault-diagnosis, Proc. IEEE Power Eng.Soc. Gen. Meet. (Montreal 2006) pp. 1–6

36.55 V. Marík, J. Lazanský: Industrial applications ofagent technologies, Control Eng. Pract. 15(11), 1364–1380 (2007)

36.56 S.J. Park, J.T. Lim: Modelling and control of agent-based power protection systems using supervisors,IEEE Proc. Control Theory Appl. 153, 92–99 (2006)

36.57 H.V.D. Parunak: Practical and Industrial Ap-plications of Agent-Based Systems (IndustrialTechnology Institute, Ann Arbor 1998)

36.58 C.G. Cassandras: Complexity made simple – atsmall price, Proc. 9th IFAC Symp. Large ScaleSystems: Theory and Applications 2001, ed. byF.G. Filip, I. Dumitrache, S. Iliescu (Elsevier, Oxford2001) pp. 1–5

36.59 P.D. Roberts: An algorithm for steady-state sys-tem optimization and parameter estimation, Int.J. Syst. Sci. 10(7), 719–734 (1979)

36.60 P.P. Varaiya: Review of the book Theory of Hi-erarchical Multilevel Systems, IEEE Trans. Autom.Control 17, 280–281 (1972)

36.61 D. Wismer: Optimization Methods for Large ScaleSystems (Mc. Graw Gill, New York 1971)

36.62 Y.C. Ho, S.K. Mitter: Directions in Large-Scale Sys-tems (Plenum, New York 1976)

36.63 A.P. Sage: Methodology for Large Scale Systems(McGraw Hill, New York 1977)

36.64 D.D. Siljak: Decentralized Control of Complex Sys-tems (Academic, Cambridge 1990)

36.65 M.G. Singh: Dynamic Hierarchical Control (NorthHolland, Amsterdam 1978)

36.66 J. Lunze: Feedback Control of Large-Scale Systems(Prentice Hall, New York 1992)

36.67 D. Steward: Systems Analysis and Management:Structure, Strategy and Design (Petrocelli Book,New York 1981)

36.68 J. Schoeffler: Online multilevel systems. In: Opti-mization Methods for Large Scale Systems, ed. byD. Wismer (McGraw Hill, New York 1971) pp. 291–330

36.69 J. Minsker, S. Piggot, G. Freidson: Hierarchical au-tomation control systems for large-scale systemsand applications, Proc. 5th IFAC World Congr. (Paris1972)

36.70 D.P. Eckman, I. Lefkowitz: Principles of modeltechnique in optimizing control, Proc. 1st IFACWorld Congr. (Moscow 1960) pp. 970–974

36.71 I. Lefkowitz: Multilevel approach to control systemdesign, Proc. JACC (1965) pp. 100–109

36.72 R. Isermann: Advanced methods of process com-puter control for industrial, Int. J. Comput. Ind.2(1), 59–72 (1981)

36.73 V. Havlena, J. Lu: A distributed automationframework for plant-wide control, optimisa-tion, scheduling and planning, selected plenaries,semiplenaries, milestones and surveys, Proc. 16thIFAC World Congr., ed. by P. Horacet, M. Simandl,P. Zitek (2005) pp. 80–94

36.74 C.B. Brosilow, L. Ladson, J.D. Pearson: Feasibleoptimization methods for interconnected systems,Proc. Joint Autom. Control Conf. – JACC (Rensse-laer Polytechnic Institute, Troy, New York 1965)pp. 79–84

36.75 L.S. Lasdon, J.D. Schoeffer: A multilevel techniquefor optimization, Proc. JACC (1965) pp. 85–92

36.76 I.D. Wilson: Foundations of hierarchical control,Int. J. Control 29(6), 899 (1979)

36.77 F.G. Vernadat: Enterprise Modelling and Integra-tion Principles and Applications (Chapman Hall,London 1996)

36.78 T.J. Williams: Analysis and Design of Hierarchi-cal Control Systems with Special Reference to SteelPlant Operations (Elsevier, Amsterdam 1985)

36.79 T.J. Williams: A Reference Model for ComputerIntegrated Manufacturing (Instrument Society ofAmerica, Research Triangle Park 1989)

36.80 Z. Binder: Sur l’organisation et la conduite des sys-temes complexes, These de Docteur (LAG, Grenoble1977) in French

36.81 J. Hatvany: Intelligence and cooperation in het-erarchic manufacturing systems, Robot. Comput.Integr. Manuf. 2(2), 101–104 (1985)

36.82 A. Koestler: The Ghost in the Machine (Hutchinson,London 1967)

36.83 M. Hopf, C.F. Schoeffer: Holonic Manufactur-ing Systems, Information Infrastructure Systemsfor Manufacturing (Chapmann Hall, London 1997)pp. 431–438

36.84 H. Van Brussel, P. Valckenaers, J. Wyns: HMS– holonic manufacturing system test case (IMSProject). In: Enterprise Engineering and Inte-gration: Building International Consensus, ed.by K. Kosanke, J.G. Nell (Springer, Berlin 1997)pp. 284–292

36.85 P. Valckenaers, H. Van Brussel, K. Hadeli,O. Bochmann, B.S. Germain, C. Zamfirescu: On thedesign of emergent systems an investigation of in-tegration and interoperability issues, Eng. Appl.Artif. Intell. 16, 377–393 (2003)

36.86 G. Tecuci: Building Intelligent Agents: An Appren-ticeship Multistrategy Learning Theory, Methodol-

PartD

36


ogy, Tool and Case Studies (Academic, New York1998)

36.87 L. Bakule: Complexity-reduced guaranteed costcontrol design for delayed uncertain symmetricallyconnected systems, Proc. 2005 Am. Control Conf.(Portland 2005) pp. 2500–2505

36.88 P.P. Groumpos: Complex systems and intelligentcontrol: issues and challenges, Proc. 9th IFACSymp. Large Scale Syst.: Theory Appl. 2001, ed. byF.G. Filip, I. Dumitrache, S. Iliescu (Elsevier, 2001)pp. 29–36

36.89 A. Kamiya, S.J. Ovaska, R. Roy, S. Kobayashi: Fusionof soft computing and hard computing for large-scale plants: a general model, Appl. Soft Comput.J. 5, 265–279 (2005)

36.90 K. Leiviskä: Control systems. In: Process Control.Papermaking Science and Technology, Book 14, ed.by K. Leiviskä (Fapet Oy, Jyväskylä 1999) pp. 13–17

36.91 B.M. Åkesson, M.J. Nikus, H.T. Toivonen: Ex-plicit model predictive control of a hybrid systemusing support vector machines, Proc. 1st IFACWorkshop Appl. Large Scale Ind. Syst. ALSIS ’06(Helsinki/Stockholm 2006), on CD ROM

36.92 K. Kawai: Knowledge engineering in power-plantcontrol and operation, Control Eng. Pract. 4, 1199–1208 (1996)

36.93 G. Stephanopoulos, J. Romagnoli, E.S. Yoon: OnlineFault Detection and Supervision in the ChemicalProcess Industries 2001 (Jejudo Island, Korea 2004)

36.94 D.H. Zhang, J.B. Zhang, Y.Z. Zhao, M.M. Wong:Event-based communications for equipment su-pervisory control, Proc. 10th IEEE Conf. Emerg.Technol. Fact. Autom. (Catania 2005) pp. 341–347

36.95 T. Martin, J. Kivinen, J.E. Rinjdorp, M.G. Rodd,W.B. Rouse: Appropriate automation integratinghuman, organisation and culture factors, PreprintsIFAC 11th World Congr., Vol. 1 (1990) pp. 47–65

36.96 F.G. Filip: Towards more humanized real-time de-cision support systems. In: Balanced AutomationSystems; Architectures and Design Methods, ed. byL.M. Camarinha-Matos, H. Afsarmanesh (ChapmanHall, London 1995) pp. 230–240

36.97 F.G. Filip, D. Donciulescu, C.I. Filip: Towards intelli-gent real-time decision support systems, Stud. Inf.Control SIC 11(4), 303–312 (2002)

36.98 H. Simon: The New Science of Management Deci-sions (Harper & Row, New York 1960)

36.99 S.Y. Nof: Theory and practice in decision supportfor manufacturing control. In: Data Base Man-agement, ed. by C.W. Holsapple, A.B. Whinston(Reidel, Dordrecht 1981) pp. 325–348

36.100 A. Bosman: Relations between specific DSS, Decis.Support Syst. 3, 213–224 (1987)

36.101 A.R. Charturverdi, G.K. Hutchinson, D.L. Nazareth:Supporting real-time decision-making through

machine learning, Decis. Support Syst. 10, 213–233(1997)

36.102 M. Cioca, L.I. Cioca, S.C. Buraga: Spatial [elements]decision support system used in disaster manage-ment, Digit. EcoSyst. Technol. Conf. 2007. DEST ’07.Inaugural IEEE-IES (Cairn 2006) pp. 607–612

36.103 G. De Michelis: Coordination with cooperativeprocesses. In: Implementing Systems for Sup-port Management Decisions, ed. by P. Humphrey,L. Bannon, A. McCosh, P. Migliarese, J.C. Pomerol(1996) pp. 108–123

36.104 R.H. Bonczek, C.W. Holsapple, A.B. Whinston:Foundations of Decision Support Systems (Aca-demic, New York 1980)

36.105 C.W. Holsapple, A.B. Whinston: Decision Sup-port System: a Knowledge-Based Approach (West,Mineapolis 1996)

36.106 A. Kusiak: Intelligent Management Systems (Pren-tice Hall, Englewood Cliffs 1990)

36.107 A. Dutta: Integrated AI and optimization for de-cision support: a survey, Decis. Support Syst. 18,213–226 (1996)

36.108 L. Duta, J.M. Henrioud, F.G. Filip: Applying equalpiles approach to disassembly line balancing prob-lem, Proc. 16th IFAC World Congr. (Session IndustrialAssembly and Disassembly) (Prague 2005), on CDROM

36.109 O. Mäyrä, T. Ahola, K. Leiviskä: Time delay estima-tion in large data bases, IFAC LSSTA Symp. (Gdansk2007), on CD ROM

36.110 F.G. Filip: System analysis and expert systems tech-niques for operative decision making, J. Syst. Anal.Model. Simul. 8(2), 296–404 (1990)

36.111 I. Lefkowitz: Hierarchical control in large-scaleindustrial systems. In: Large Scale Systems, ed.by D.D. Haimes (North Holland, Amsterdam 1982)pp. 65–98

36.112 M. Brdys, B. Ulanicki: Operatioal Control of WaterSystems (Prentice Hall, New York 1994)

36.113 L. Bakule, F. Paulet-Crainiceanu, J. Rodellar,J.M. Rossell: Overlapping reliable control fora cable-stayed bridge benchmark, IEEE Trans. Con-trol Syst. Technol. 4, 663–669 (2006)

36.114 L. Bakule, J. Rodellar, J.M. Rossell: Robust over-lapping guaranteed cost control of uncertainsteady-state discrete-time systems, IEEE Trans.Autom. Control 12, 1943–1950 (2006)

36.115 R. Srikant: The Mathematics of Internet CongestionControl (Birkhäuser, Boston 2004)

36.116 D.M. Stipanovic, G. Inalham, R. Teo, C.J. Tomlin:Decentralized overlapping control of a formation ofunmanned aerial vehicles, Automatica 40(1), 1285–1296 (2004)

36.117 S.S. Stankovic, M.J. Stanojevic, D.D. Siljak: De-centralized overlapping control of a platoon ofvehicles, IEEE Trans. Control Syst. Technol. 8, 816–832 (2000)

PartD

36

large-scale c 36. large-scale complex...

Documents