Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: In this paper, the past and current issues involved in the design of decentralized networked control sys-
Received 21 April 2011 tems are reviewed. The basic models of interconnected systems described as continuous-time linear
Accepted 21 August 2011 time-invariant systems in the time domain serve as a framework for the inclusion of communication
channels in the decentralized feedback loop. The I/O-oriented models and the interaction oriented mod-
els with disjoint subsystems and interactions are distinguished. The overview is focused on packet drop-
Keywords: outs, transmission delays, and quantization effects which are included in the time-driven design of
Decentralization
feedback loop components. Single- and multiple-packet transmissions are considered in this contents.
Communication
Large-scale complex systems
The design of decentralized state feedback gain matrices with delayed feedback uses the methodology
Networked control systems of sampled-data feedback design for continuous-time systems, while the decentralized H1 quantizer
Delays design is based on the static output controller. The Liapunov stability approach results in computationally
Quantizers efficient decentralized control design strategies described by using linear matrix inequalities.
Ó 2012 Elsevier Ltd. All rights reserved.
1. Introduction each of which receives the measurement data and influences the
control input only of the attached subsystem. The information flow
A fundamental feature of present-day technological, societal or from the plant through the controller in the feedback loop is divided
environmental processes is a high degree of their complexity. Com- into separated parallel flows through the control stations. For feed-
plexity is the main problem in recent system theory and their appli- back control, the use of decentralized controllers is a remarkable
cations. The notion of complexity is used in widely different feature of large scale systems.
contexts. A notable characteristic of most large scale complex sys- Classical as well as modern system theories have dealt typically
tems is that centrality fails due to the lack of centralized information with problems where the communication links between the plant
or centralized computing capability. Many real world problems are and the controller can be considered as transparent ones, that is
considered large scale complex by nature and not by choice. Such with the case of wired connections. Various large scale system
systems cannot be dealt with as an unstructured black box by applications of computer networks as well as a significant progress
one-shot methods. Instead, the analysis and synthesis of large scale in technology, where the system to be controlled, actuators, sen-
complex systems take the advantage of structural properties of the sors, and controllers are spatially distributed, motivate the usage
plant in order to achieve numerical or conceptual simplifications of of a communication network. The operation of closed-loop systems
the overall problem. The notion of large scale system is a very sub- is coordinated by a communication network to achieve desired
jective one. There is no accepted definition for what constitutes a overall objectives. The move towards networks at all levels proba-
large scale complex system. Therefore, a more pragmatic view bly belongs to the most notable trend in recent system theory.
was adopted: a system is considered as large scale complex when- Control networks can replace point-to-point wired systems while
ever it is necessary to partition the given analysis or synthesis prob- providing a number of advantages. The simplest but most impor-
lem in order to come up with manageable subproblems. However, tant advantage is the reduced volume of wiring which results in
such process requires new approaches for dealing with high dimen- increasing reliability and lower implementation costs. An another
sionality of the system equations, different types of uncertainties of significant advantage is that networks enable complex distributed
the model, information structure constraints, and delays as funda- control. Other advantages of communication networks include en-
mental characteristics of complexity. In contrast to this, small scale hanced interchangeability of devices, improved reconfiguration of
complex systems refer to a low dimensionality of the systems which control systems, and increased capability for maintenance. The ba-
exhibit complex dynamic behavior. A decentralized controller is a sic problem within the framework of networked control systems is
feedback controller which consists of independent control stations, the timing problem, i.e to find the communication frequency of the
networked feedback loops under which the stability as well as the
⇑ Corresponding author. desired level of performance remain ensured. Typical phenomena
E-mail addresses: bakule@utia.cas.cz (L. Bakule), papik@utia.cas.cz (M. Papík). concerning network channels include random delays, packet
1367-5788/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.arcontrol.2012.03.001
2 L. Bakule, M. Papík / Annual Reviews in Control 36 (2012) 1–10
Note only that due to space limitations, a large number of re- yi ðtÞ ¼ C si xðtÞ þ Dsij uj ðtÞ i ¼ 1; . . . ; N
j¼1
cent promising results have been omitted.
L. Bakule, M. Papík / Annual Reviews in Control 36 (2012) 1–10 3
The relation of the system matrices in (1) and (2) are given as decentralized decision making, the decision units are completely
independent or at least almost independent. Such a complete divi-
Bs ¼ ðBs1 ; Bs2 ; . . . ; BsN Þ
0 1 0 1 sion is possible only for specific problems. It means that the infor-
C s1 Ds11 Ds12 ... Ds1N mation flow among the decision units can be divided into
BC C BD Ds22 ... Ds2N C
B s2 C B s12 C ð3Þ completely independent parts. The concepts and methods for
Cs ¼ B C
B .. C Ds ¼ B ..
B .. .. C
@ . A @ . . . ... C
A
reformulating a control problem as a set of independent subprob-
lems including the solution of these subproblems are usually re-
C sN DsN1 DsN2 . . . DsNN ferred as decentralized or distributed control. Decentralized
Note that this model does not show how the overall system dynam- structures usually take into account the above decomposition algo-
ics depends on the subsystems because the system is considered as rithms. They result in three basic particularly desirable forms of
a whole. the gain matrix: block-diagonal, BBD, and overlapping structures.
Robustness and model simplification are important tools for treat-
ing with high dimensionality and uncertainties of large scale sys-
2.3. Interaction-oriented model
tems. Robustness analysis evaluates dependencies of the solution
on parametric or structural uncertainties, while approximation
Consider the structure of large scale system decomposed in the
deals with the substitution of the given model by similar but sim-
disjoint structure of subsystems and their interconnection. Many
pler model possessing the same main dynamic properties.
large scale systems are considered as interconnection of subsys-
Delays play a key role in the control design of large scale sys-
tems, where every subsystem represents a dynamic system of its
tems. Their increasing importance is underlined mainly by current
own dynamics. The overall plant is a structured dynamic system
technological trends in digital networks and wireless communica-
described as
tion. Decentralized control is superior over centralized control also
P
N from this point of view because of the reduction of shared local
x_ i ðtÞ ¼ Ai xi ðtÞ þ Bi ui ðtÞ þ Aij xj ðtÞ xi ðt o Þ ¼ xio
j¼1 ð4Þ information as can be found for instance in Bakule and Rodellar
(1996), Bakule, Rodellar, and Rossell (2002), Bakule, Rodellar, and
yi ðtÞ ¼ C ri xi ðtÞ þ Dri ui ðtÞ i ¼ 1; . . . ; N _
Rossell (2006), Iftar (2008), Ghosh, Das, and Ray (2009) and Wu
where xi ðtÞ 2 Rni ; ui ðtÞ 2 Rmi , and yi ðtÞ 2 Rqi are vectors of the sub- (2009).
system states, subsystem control inputs, and subsystem outputs, These general methodologies are widely used in theory and
respectively. practice. There is available a number of survey papers as well as
To overcome the difficulties caused by the complexity issues several books which provide further information about this topic
arising in large scale complex systems, several basic methodologies as presented by Šiljak (1991), Lunze (1992), Jamshidi (1997), Bak-
have been and still being developed. The accumulated experience ule (2008), Zečević and Šiljak (2010) and Mahmoud (2011).
in large scale complex systems control results in three main well
known methodologies as follows
3. Communication in control
– Decomposition
– Decentralization Communication and control are related namely in networked
– Robustness and model simplification control systems (NCS). NCS are control systems composed of the
plant, actuators, sensors, and controllers which are connected to
Decomposition methods utilize the internal system structure, i.e. the plant via some form of communication medium. The compo-
the division of plant into subsystems and interconnections. It nents of such systems are spatially distributed, they may operate
means that the plant is not considered as a single object, but as a in an asynchronous manner, but their operation is coordinated in
mixture of different interacting subsystems. Subsystems can be de- some way to achieve satisfactorily desired overall objective.
fined by the main building blocks of the plant (physical decompo- Decentralized NCS (DNCS) are the control systems with multi-
sitions) or are imposed for computational reasons (numerical ple control stations while transmitting control signals through a
decompositions). The main advantage of decomposition ap- network, i.e. date signals are transmitted to multiple controllers
proaches is the reduction of complexity in the analysis and synthe- in the feedback loop. DNCS combine the advantages of the central-
sis of large scale control problems. The decomposition methods ized NCS and the decentralized control systems. Such a combina-
concern the interaction-oriented models. They can be classified tion enables to cut unnecessary wiring, reduce the complexity
according to the properties of subsystem-interconnection struc- and the overall system cost when designing and implementing
tures as control systems. Recently, DNCS design methods are subject to
intensive research. For instance, the projects ICT Research
– Disjoint subsystems (2011b), ICT Research (2011a) and DFG–Priority Program (2011)
– Overlapping subsystems illustrate this attempt. 2-channel DNCS structure is shown in Fig. 1.
– Symmetric composite systems There is no unique theory in the area of networked control sys-
– Multi-time scale systems tems (NCS) up to now. Recent surveys on this topic can be found
– Hierarchically structured systems in Hristu-Varsakelis and Levine (2005), Bemporad, Heemels, and
Johansson (2010), Baillieul and Antsaklis (2007), Hespanha, Nag-
One effective approach for partitioning the systems exploits the hshtabrizi, and Xu (2007), Wei, Xue, and Da-zhi (2007), Matveev
sparsity of large-scale state space models. The decomposition and Savkin (2009), Goodwin, Silva, and Quevedo (2008) and Nešić
algorithms are based on graph-theoretic representations such as and Liberzon (2009).
epsilon decompositions, Bordered Block Diagonal (BBD) ordering, There are available several results dealing with the timing prob-
and input/output constrained decompositions. lem for the time-driven decentralized networked systems. Bakule
Decentralization concerns the information structure of the deci- (2008) and Xu and Hespanha (2004) introduce DNCS problems.
sion process. It generally means that the overall system task is bro- Decentralized stabilization of NCS using periodically time-varying
ken down into several subsystem tasks so that the solution of local controller is presented in Jiang, Zhou, and Zhang (2008), while
subsystem tasks satisfactorily solves the overall system task. In Zhong, Stefanovski, Dimirovski, and Zhao (2009) and Menon and
4 L. Bakule, M. Papík / Annual Reviews in Control 36 (2012) 1–10
q(x) = 0 holds for x in some neighborhood of the origin. For in- consist of the constant communication delays denoted as dci and
stance, the quantizer with rectangular quantization regions satis- the delays caused by data packet dropout dkiD. The input of ith local
fies the above requirements with its values in each quantization controller is xi(tk) = xi(tk dkiD dci). The control input has the form
region belonging to that region.
ui ðtÞ ¼ K i xi ðt k Þ t 2 ½t k ; tkþ1 Þ k ¼ 1; 2; . . . ð19Þ
Consider the quantized measurements in the form
T T
where xðt k Þ ¼ xT1 ðt k Þ xT2 ðt k Þ and uðtÞ ¼ uT1 ðtÞ uT2 ðtÞ consist of ni-
z
ql ðzÞ,lq ð17Þ and mi-dimensional vectors, respectively. x1(tk) and x2(tk) are trans-
l mitted over different channels. Each channel produces its own data
where l > 0 is a scalar parameter. Supposing l = 0 means that the packet dropouts and delays. Suppose also that the acknowledgment
output of the quantizer equals zero. The range of the quantizer is (ACK) about data losses is always available to the sender of the
Ml and the quantization error is Dl. The parameter l can be under- plant. Thus, the network with a multiple-packet transmission is
stood as a ‘‘zoom’’ variable. It means that when l increases, then the modeled as a switch. It is illustrated in Fig. 5 for 2-channels with
zooming goes out and a new quantizer with larger range and larger a buffer B.
quantization error is obtained. When l decreases, then the zooming Denote the corresponding time-varying delays di(t) = t tk
goes in and a quantizer with smaller range but also with smaller dkiD dci, where 1 6 dki 6 (tk1 dci)/D. Suppose that the number
quantization error is constructed. l can be updated in the depen- of packet dropouts is bounded so that it satisfies the constraints
dence on the system local state (or the local measurement output).
0 < di ðtÞ 6 d ð20Þ
In this sense, another state of the closed-loop system can be
reached. The most common static quantizers are illustrated in Fig. 4. is a given positive constant.
for i = 1, 2, where d
A common definition of a quantizer has been introduced by Lib- Therefore, the input to the controller includes two different de-
erzon (2003). Zhu and Yang (2008a) and Wen and Yang (2009) con- lays has the form
sider quantized H1 control for the NCSs, while the elimination of
quantized input effect is considered in Yun, Choi, and Park
x1 ðt k Þ x1 ðt d1 ðkÞÞ
xðtk Þ ¼ ¼ t 2 ½tk ; t kþ1 Þ k ¼ 1; 2; . . .
(2009). Nešić and Liberzon (2009) presents a unified approach to x2 ðt k Þ x2 ðt d2 ðkÞÞ
the analysis and synthesis of networked and quantized control ð21Þ
systems.
Consider now the closed-loop overall system (18)–(21) as follows
where Ps > 0, Rsi P 0, and Ssi > 0 are constant matrices of appropriate
dimensions. Denote the matrices G1i ¼ ð0 . . . Imi . . . 0ÞT and
G2i ¼ ð0 . . . Ini . . . 0Þ, where the identity matrices are located at
the ith position, i 2 {1, 2}.
The system (18) with a given matrix U and a scalar d > 0 is
asymptotically stabilized by the controller (19) for all dðtÞ 6 d if
Fig. 4. Uniform (upper) and logarithmic (lower) quantizers. Fig. 5. NCS with multiple packet transmission.
L. Bakule, M. Papík / Annual Reviews in Control 36 (2012) 1–10 7
there exist symmetric matrices Ps > 0, Rsi P 0, Ssi > 0 and block stabilize the closed-loop system (31) and the standard relation
diagonal matrices Ys, Xs satisfying the relation kD + C1(sI Ac)1B1k1 < c holds. Then, there exist positive definite
P<0 ð25Þ matrices P, R satisfying the matrix inequality
0 1
where ATc P þ PAc þ 1c C T1 C 1 þ R PB1 þ 1c C T1 D
@ A<0 ð32Þ
P1 P2 BT1 P þ 1c DT C 1 cI þ 1c DT D
P¼ VW ðWVÞT
P3
0 1 as follows from the bounded real lemma.
P
2
B ðRsi Ssi Þ Ps C Suppose that each local quantizer is located between sensors
P1 ¼ B C
i¼1
B C and a local controller in the local feedback loop, where each quan-
@ 2 P
2 A
2d Ssi tizer is described in the same way as a centralized quantizer by
i¼1 (13)–(16) with the quantized measurements (17). Suppose the
Ssi 0 availability of only quantized local output information. The static
P2 ¼ ð P21 P22 ÞP2i ¼ output feedback (30) can be modified using the quantized informa-
0 0 ð26Þ
tion on y as
2Ssi Ssi
P3 ¼ diagðP31 P32 ÞP3i ¼
Rsi Ssi y1 y2
u1 ðtÞ ¼ K 1 l1 q1 u2 ðtÞ ¼ K 2 l2 q2 ð33Þ
V ¼ UT I 0 . . . 0 l1 l2
W ¼ ð H 1 H2 N 1 0 N 2 0 Þ The closed-loop system (31), (33) with any fixed positive scalars l1
P2 P
2 and l2 has the form
H1 ¼ A GT2i X si G2i H2 ¼ GT2i X si G2i
i¼1 i¼1 _
xðtÞ ¼ Ac xðtÞ þ B1 wðtÞ þ Fðl; yÞ
Ni ¼ Bs Di G1i Y si G2i i ¼ 1; 2 ð34Þ
zðtÞ ¼ C 1 xðtÞ þ DwðtÞ
The gain matrix Ks is given as where
K s ¼ Y s X 1
s ð27Þ Fðl; yðtÞÞ ¼ F 1 ðl1 ; y1 ðtÞÞ þ F 2 ðl2 ; y2 ðtÞÞ
where F 1 ðl1 ; y1 ðtÞÞ ¼ l1 B21 K 1 q1 yl1 ðtÞ yl1 ðtÞ
1 1 ð35Þ
Y s ¼ diagðY s1 Y s2 ÞX s ¼ diagðX s1 X s2 Þ ð28Þ
F 2 ðl2 ; y2 ðtÞÞ ¼ l2 B22 K 2 q2 yl2 ðtÞ yl2 ðtÞ
The relations P s ¼ X Ts PX s ; Rsi ¼ X Ts Ri X s ; Ssi ¼ X Ts Si X s between the 2 2
matrices in (24) and (25) are introduced by Zhu and Yang The goal is to find the parameters l1 depending on the local output
(2008b). Communication requirements for decentralized control y1(t) as well as l2 depending on the local output y2(t) guaranteeing
systems with noiseless digital channels and bounded system noise the closed-loop system stability with the H1 disturbance attenua-
are presented by Yüksel and Basßar (2006, 2007). Decentralized sta- tion level c.
bilization of multi-channel systems with communication con- Given two quantizers Mi sufficiently larger than Di as
straints is studied by Matveev and Savkin (2009).
kPB2i K i kkC 2i k
M i > 4 Di i ¼ 1; 2 ð36Þ
4.2. Quantization km ðRÞ
Then, there exist positive scalars li such that
Consider the decentralized quantizer design for a 2-channel
system in the form
jyi j km ðRÞð1 eÞ
6 li 6 jy j i ¼ 1; 2 ð37Þ
_ Mi 4Di kPB2i K i kkC 2i k i
xðtÞ ¼ AxðtÞ þ B1 wðtÞ þ B21 u1 ðtÞ þ B22 u2 ðtÞ
zðtÞ ¼ C 1 xðtÞ þ DwðtÞ for any e 2 (0, 1) asymptotically stabilizing the closed loop system
ð29Þ (34) with the H1 disturbance attenuation level c as proved by Zhai,
y1 ðtÞ ¼ C 21 xðtÞ
Chen, and Gui (2009).
y2 ðtÞ ¼ C 22 xðtÞ
where xðtÞ 2 Rn , u1 ðtÞ 2 Rm1 and u2 ðtÞ 2 Rm2 are the system states, 5. Interaction oriented systems
the control inputs of channel 1 and 2, respectively. wðtÞ 2 Rh ,
zðtÞ 2 Rp , y1 ðtÞ 2 Rq1 and y2 ðtÞ 2 Rq2 are the disturbance input, the 5.1. Delays in the loop
disturbance input, the local measurement outputs of channel 1
and 2, respectively. A, B1, B21, B22, C1, C21, C22 and D are constant Consider the system (4) composed of 2 two interconnected sub-
matrices of appropriate dimension. Suppose that the triple systems with Cri = 0, Dri = 0, i = 1, 2 as
h iT
A; ½B21 B22 ; C T21 C T22 is stabilizable and detectable. _
xðtÞ ¼ AxðtÞ þ Br uðtÞ ð38Þ
where xðt k Þ ¼ ðxT1 ðt k ÞxT2 ðt k ÞÞT , uðtÞ ¼ ðuT1 ðtÞ uT2 ðtÞÞT and cept of quadratic invariance. In Langbort, Gupta, and Murray
(2006), the authors developed the LMI type condition for the exis-
x1 ðtk Þ x1 ðt d1 ðtÞÞ
xðt k Þ ¼ ¼ t 2 ½t k ; tkþ1 Þ k ¼ 1; 2; . . . tence of a decentralized controller for heterogenous interconnected
x2 ðtk Þ x2 ðt d2 ðtÞÞ subsystems with failing communication channels. Stubbs, Vla-
ð41Þ dimerou, Fulford, Strick, and Dullerud (2006) describes the hover-
craft flexible testbed for decentralized control of multiple
The closed-loop overall system (38)–(41) has the form
autonomous vehicles that are networked in a wireless manner
_
xðtÞ ¼ AxðtÞ þ Br K r xðt k Þ ¼ AxðtÞ þ Br K r C 1 xðt d1 ðtÞÞ and can be commanded
0
þ Br K r C 2 xðt d2 ðtÞÞxðt o Þ ¼ U ðt o Þ t o 2 ½d; ð42Þ
k
5.2. Quantization
where Di, Ci are defined by (23). The essential difference between
the systems (18) and (20) is in the matrices Bs and Br, where Consider the decentralized quantizer design for two intercon-
Bsi ¼ ð0 . . . Imi . . . 0ÞT Bri with the identity matrices are located at nected subsystem as follows
the ith position. It leads to an analogous way of reasoning with
the procedure described by the relation (24)–(28), when using the x_ i ðtÞ ¼ Aii xi ðtÞ þ Aij xj ðtÞ þ B1i wi ðtÞ þ B2i ui ðtÞ
same notation for G1i, G2i. To simplify suppose the structure of the zi ðtÞ ¼ C i xi ðtÞ þ Di wi ðtÞ ð47Þ
Liapunov–Krasovskii functional candidate Vr(t) for the system (42)
yi ðtÞ ¼ Ei xi ðtÞ i; j ¼ 1; 2
identical with (24). It means only to substitute Vs(t), Ps, Rri, Ssi by
Vr(t), Pr = diag(Pr1 Pr2) > 0, Rri P 0, Ssi > 0, respectively.
> 0 is where xi ðtÞ 2 Rni , ui ðtÞ 2 Rmi , wi ðtÞ 2 Rhi , and zi ðtÞ 2 Rpi are the sub-
The system (38) with a given matrix Ur and a scalar d
if system states, the control inputs, the disturbance inputs, and the
asymptotically stabilized by the controller (40) for all dðtÞ 6 d
controlled outputs, respectively. Aij, Bij, Ci, Di and Ei are constant
there exist symmetric matrices Pri > 0, Rri P 0, Sri > 0, and block
matrices.
diagonal matrices Yr, Xr satisfying the relation
Suppose the availability of a decentralized static output feed-
C<0 ð43Þ back controller for the system (47) as
where
u1 ðtÞ ¼ K 1 y1 ðtÞ u2 ðtÞ ¼ K 2 y2 ðtÞ ð48Þ
!
C1 C2
C¼ V r W r ðW r V r ÞT asymptotically stabilizing the closed-loop system (47), (48) with
C3
0 2 1 the H1 disturbance attenuation level c. Denote as
P
B ðRri Sri Þ Pr C _
xðtÞ ¼ Ac xðtÞ þ B1 wðtÞ
ð49Þ
C1 ¼ B C
i¼1
B C zðtÞ ¼ CxðtÞ þ DwðtÞ
@ P
2 A
2
2d Sri
i¼1 the closed-loop system, where x(t) = (x1(t)T x2(t)T)T, w(t) = (w1(t)T
!
Sri 0 w2(t)T)T, z(t) = (z1(t)T z2(t)T)T and
C2 ¼ ð C21 C22 ÞC2i ¼
0 0 ð44Þ !
! A11 þ B21 K 1 E1 A12
2Sri Sri A¼
C3 ¼ diagðC31 C32 ÞC3i ¼ A21 A22 þ B22 K 2 E2 ð50Þ
Rri Sri
B1 ¼ diagðB11 B12 Þ C ¼ diagðC 1 C 2 Þ D ¼ diagðD1 D2 Þ
V r ¼ U Tr I 0 ... 0
W r ¼ ð Hr1 Hr2 Nr1 0 Nr2 0Þ Consider that each local feedback loop includes two quantizers. One
P
2 P
2 quantizer is located between subsystem sensors and a local control-
Hr1 ¼ A GT2i X ri G2i Hr2 ¼ GT2i X ri G2i ler, while an another quantizer is located between a local controller
i¼1 i¼1
and subsystem inputs. Suppose that each quantizer is described by
Nri ¼ Br Di G1i Y ri G2i i ¼ 1; 2 (13)–(16) with the quantized measurements (17) is adopted to the
subsystem setting for each local feedback loop. More precisely, de-
The gain matrix Kr is given as
note the quantization ranges Mi1, Mi2 and the quantization errors
K r ¼ Y r X 1
r ð45Þ Di1, Di2, where Mi1, Di1 describe the quantizer located between
the sensors and the controller in the ith feedback loop. Suppose
where two quantizer variables li1, li2 accordingly. The input to the ith
Y r ¼ diagðY r1 Y r2 Þ X r ¼ diagðX r1 X r2 Þ ð46Þ controller has the form
The DNCS design has been considered in several publications up to yi
yci ðtÞ ¼ li1 qi1 ð51Þ
now. The dynamics of local estimator-controller scheme with com- li1
munication among vehicles in cooperative formations are studied
by Smith and Hadaegh (2007). Narendra, Oleng, and Mukhopadhy- while the input to the ith subsystem is
ay (2006) demonstrates through simulation that the significant
K i yci K i li1 qi1 ðyi =li1 Þ
improvement in transient responses of the disjoint subsystems ui ðtÞ ¼ li2 qi2 ¼ li2 qi2 ð52Þ
can be achieved with communication at relatively few instants of li2 li2
time. Decentralized control problem over periodic fast switching
network is presented by Roberts and Stilwell (2005). It is applied The closed-loop system (47)–(52) with any fixed positive scalars l1
to autonomous vehicle platoon with the network characterized and l2 has the form
by using the graph theoretic issues. Rotkowitz and Lall (2006) con- _
xðtÞ ¼ Ac xðtÞ þ B1 wðtÞ þ Tðl; yÞ
siders the problem of decentralized control minimizing a norm of ð53Þ
zðtÞ ¼ C 1 xðtÞ þ DwðtÞ
the closed-loop subject to a subspace constraint by using the con-
L. Bakule, M. Papík / Annual Reviews in Control 36 (2012) 1–10 9
where networked control systems and its limits are the key issues deter-
mined primarily by delays, dropped packets, and quantizations.
T 1 ðl11 ; l12 ; y1 ðtÞÞ
Tðl; yðtÞÞ ¼ Reliability, sensor failures, actuator constraints as well as the per-
T 2 ðl21 ; l22 ; y2 ðtÞÞ
0 1 formance degradation under communication failures, reconfigura-
l12 B21 q12 K 1 l11 ql1112ðy1 =l11 Þ K 1ly121 ðtÞ ble control, and the emphasis on increased autonomy belong to
¼@ A ð54Þ other important current research challenges.
l22 B22 q22 K 2 l21 ql21 ðy2 =l21 Þ K 2ly2 ðtÞ
22 22
The goal is to find the parameters li1 and li2 depending on the local Acknowledgement
output yi(t) for i = 1, 2 guaranteeing the closed-loop system stability
with the H1 disturbance attenuation level c. L. Bakule is grateful to the Ministry of Education, Youth and
Applying the bounded real lemma on the interconnected Sports of the Czech Republic (MEYS) by its support through the
closed-loop system (50) leads to the existence of positive definite Grant LG12014.
block diagonal matrices P = diag(P1 P2), R = diag(R1 R2) satisfying
the matrix inequality References
0 1
ATc P þ PAc þ 1 T
c C1 C1 þR PB1 þ 1 T
c C1 D Abdollahi, F., & Khorasani, K. (2008). A decentralized H1 routing control strategy for
@ A<0 ð55Þ mobile networked multi-agents. In Proceedings of the 2009 American control
BT1 P 1
þ c D C1 T 1
cI þ c D D T
conference (pp. 1555–1560). MO, USA: St. Louis.
Alessio, A., & Bemporad, A. (2008). Stability conditions for decentralized model
Given two quantizers Mi1 sufficiently larger than Di in the ith local predictive control under packet drop communication. In Proceedings of the 2008
feedback loop satisfying the relation American control conference (pp. 3577–3582). Seattle, WA, USA.
Baillieul, J., & Antsaklis, P. (2007). Control and communication challenges in
kP i B2i kkEi k networked real-time systems. Proceedings of the IEEE, 95(1), 9–28.
M i1 > 2Di i ¼ 1; 2 ð56Þ Bakule, L. (2008). Decentralized control: An overview. Annual Reviews in Control, 32,
xi km ðRi Þ 87–98.
Bakule, L., & de la Sen, M. (2009). Decentralized stabilization of networked complex
where xi is a positive tuning parameter relating the parameters li1 composite systems with nonlinear perturbations. In Proceedings of the 7th IEEE
and li2 as follows international conference on control and automation (pp. 2272–2277).
Christchurch, New Zealand.
M i2 ¼ ðDi1 þ2M
xi
i1 ÞkK i k
i ¼ 1; 2 Bakule, L., & de la Sen, M. (2010). Decentralized resilient H1 observer-based control
ð57Þ for a class of uncertain interconnected networked systems. In Proceedings of the
Di1 kK i k
Di ¼ Di2 þ xi li2 ¼ xi li1 2010 American control conference (pp. 1338–1343). Baltimore, MD, USA.
Bakule, L., & de la Sen, M. (2011). Decentralized H1 quantizers design for networked
complex composite systems. In Proceedings of the 18th world congress of the IFAC
Then, there exist positive scalars l1i,l2i which depend on the local
(pp. 2674–2679). Milano, Italy.
output yi(t) such that the closed-loop system (53) is asymptotically Bakule, L., & Rodellar, J. (1996). Decentralized control design of uncertain nominally
stable with the H1 disturbance attenuation level c. The range of l1i linear symmetric composite systems. IEE Proceedings – Control Theory and
which depends on the local output yi(t) can be selected as Applications, 143, 539–630.
Bakule, L., Rodellar, J., & Rossell, J. (2002). Overlapping quadratic optimal control of
jyi j xi km ðRi Þ linear time-varying commutative systems. SIAM Journal on Control and
6 l1i 6 jy j i ¼ 1; 2 ð58Þ Optimization, 40(5), 1611–1627.
M i1 2Di kPi B2i kkEi k i Bakule, L., Rodellar, J., & Rossell, J. (2006). Robust overlapping guaranteed cost
control of uncertain state–delay discrete–time systems. IEEE Transactions on
A simple choice of l1i has the form Automatic Control, 51(12), 1943–1950.
Barcelli, D., & Bemporad, A. (2009). Decentralized model predictive control of
2jyi j dynamically-coupled linear systems: Tracking under packet loss. In Proceedings
l1i ¼ ð59Þ
M i1 þ 2Di ðkPi B2i kkEi k=xi km ðRi ÞÞ of the 1st IFAC workshop on estimation and control of networked systems (pp. 204–
209). Venice, Italy.
as presented by Chen, Zhai, Gui, Yang, and Liu (2010). Bakule and de Bemporad, A., Heemels, M., & Johansson, M. (Eds.). (2010). Networked control
systems. Springer-Verlag.
la Sen (2011) further extended the quantizer design for networked Chen, H., Wang, W., & Li, Z. (2009). Synthetic modeling and control of networked
symmetric composite systems. control systems with multi-packet transmission. In Proceedings of the
international conference on advanced computer control 2009 (pp. 661–665).
Singapore.
6. Conclusion Chen, N., Zhai, G., Gui, W., Yang, C., & Liu, W. (2010). Decentralised H1 quantisers
design for uncertain interconnected networked systems. IET Control Theory and
In this paper, the basic past and current issues involved in the Applications, 4(2), 177–185.
DFG–Priority Program 1305 (2011). Control theory of digitally networked
design of decentralized networked control systems have been re- dynamical systems. Bonn.
viewed. The most simple structures of large scale systems have Donkers, M., & Heemels, W. (2012). Output-based event-triggered control with
been considered, that is the I/O-oriented models and the interac- guaranteed L1-gain and improved and decentralised event-triggering. IEEE
Transactions on Automatic Control, http://dx.doi.org/10.1109/TAC.2011.2174696.
tion oriented models based on the disjoint decomposition. It was Ghosh, S., Das, S. K., & Ray, G. (2009). Decentralized stabilization of uncertain
presented how to treat with the effects of packet dropouts, systems with interconnections and feedback delays: An LMI approach. IEEE
transmission delays, and quantization within the decentralized Transactions on Automatic Control, 54(4), 905–912.
Goodwin, G., Silva, E., & Quevedo, D. (2008). A brief introduction to the analysis and
NCS design. Single- and multiple-packet transmissions have been design of networked control systems. In Proceedings of the 2008 Chinese control
included in the DNCS design for two prototype cases. The first case and decision conference (pp. 1–13). Yantai, PR China.
deals with the design of decentralized state feedback gain matrices Gusrialdi, A., & Hirche, S. (2010). Performance-oriented communication topology
design for large-scale interconnected systems. In Proceedings of the 49th IEEE
with delayed feedback uses the methodology of sampled-data conference on decision and control (pp. 5707–5713). Atlanta, GA, USA.
feedback design for continuous-time systems, while the second Hespanha, J., Naghshtabrizi, P., & Xu, Y. (2007). A survey of recent results in
case presents the decentralized H1 quantizer design based on networked control systems. Proceedings of the IEEE, 95(1), 138–162.
Hristu-Varsakelis, D., & Levine, W. (Eds.). (2005). Handbook of networked and
given static output gain matrices stabilizing the closed-loop sys-
embedded control systems. Boston: Birkhäuser.
tem. The main objective was to show that LMI techniques provide Hu, L., Bai, T., Shi, P., & Wu, Z. (2007). Sampled-data control of networked linear
a general design framework to solve the DNCS design problems. control systems. Automatica, 43, 903–911.
Future research effort should consider the analysis and synthe- ICT Research in FP7 (2011a). HYCON2: Highly-complex and networked control
systems. European Community, Bruxelles.
sis of complex strongly interconnected subsystems controlled over ICT Research in FP7 (2011b). WIDE: Decentralized and wireless control of large-
realistic decentralized communication channels. Performance of scale systems. European Community, Bruxelles.
10 L. Bakule, M. Papík / Annual Reviews in Control 36 (2012) 1–10
_
Iftar, A. (2008). Decentralized robust control of large-scale time-delay systems. In Witrant, E., D’Innocenzo, A., Sandou, G., Santucci, F., Benedetto, M., Isakson, A., et al.
Proceedings of the 17th IFAC world congress (pp. 9332–9337). Seoul, Korea. (2010). Wireless ventilation control for large-scale systems: The mining
Jamshidi, M. (1997). Large-scale systems: Modeling. Control and fuzzy logic. Upper industrial case. International Journal of Robust and Nonlinear Control, 20,
Saddle River, New Jersey: Prentice Hall. 226–251.
Jiang, C., Zhou, D., & Zhang, Q. (2008). Stabilization of networked control systems via Wu, H. (2009). Decentralized robust tracking control of uncertain large scale
time-varying local controller. In Proceedings of the 7th world congress on systems with multiple delays in the interconnections. Kybernetika, 45(1),
intelligent control and automation (pp. 7965–7069). Chongquing, PR China. 121–136.
Kurose, J., & Rose, K. (2005). Computer networking: A top-down approach featuring the Wu, J., & Chen, T. (2007). Design of networked control systems with packet dropout.
internet. Boston: Pearson/Addison Wesley. IEEE Transactions on Automatic Control, 52(7), 1314–1319.
Langbort, C., Gupta, V., & Murray, R. (2006). Distributed control over failing Xie, N., & Xia, B. (2009). Non-fragile control for networked control systems with
channels. In P. Ansaklis & P. Tabuada (Eds.), Proceedings of networked embedded fast-varying delay and packet-dropout. In Proceedings of the 2009 international
sensing and control. LNCIS (Vol. 331, pp. 325–342). Berlin: Springer. conference on information engineering and computer science (pp. 1–4). Wuhan, PR
Liberzon, D. (2003). Switching in systems and control. Boston: Birkhäuser. China.
Lunze, J. (1992). Feedback control of large-scale systems. London: Prentice Hall. Xiong, J., & Lam, J. (2007). Stabilization of linear systems over networks with
Mahmoud, M. (2011). Decentralized control and filtering in interconnected dynamical bounded packet loss. Automatica, 43, 80–87.
systems. Boca Raton: CRC Press. Xiong, J., & Lam, J. (2009). Stabilization of networked control systems with a logic
Matveev, A., & Savkin, A. (2009). Estimation and control over communication ZOH. IEEE Transactions on Automatic Control, 54(2), 358–363.
networks. Boston: Birkhäuser. Xu, Y., & Hespanha, J. (2004). Communication logics for networked control systems.
Menon, P., & Edwards, C. (2009). Decentralised static output feedback stabilisation In Proceedings of the 2004 American control conference (pp. 572–577). Boston,
and synchronisation of networks. Automatica, 45(12), 2910–2916. MA, USA.
Nair, G., Evans, R., & Caines, P. (2004). Stabilising decentralised linear systems under Yu, M., Wang, L., Chu, T., & Hao, F. (2005). Stabilization of networked control
data rate constraints. In Proceedings of the IEEE conference on decision and control systems with data packet dropout and transmission delays: Continuous-time
(pp. 3992–3997). Paradise Island, Bahamas. case. European Journal of Control, 11, 40–49.
Narendra, K., Oleng, N., & Mukhopadhyay, S. (2006). Decentralized adaptive control Yu, X., & Tomsovic, K. (2004). Application of linear matrix inequalities for load
with partial communication. IEE Proceedings – Control Theory and Applications, frequency control with communication delays. IEEE Transactions on Power
153(5), 546–555. Systems, 19(3), 1508–1515.
Nešić, D., & Liberzon, D. (2009). A unified framework for design and analysis of Yue, D., Han, Q.-L., & Peng, C. (2004). State feedback controller design of networked
networked and quantized control systems. IEEE Transactions on Automatic control systems. IEEE Transactions on Circuits and Systems – II: Express Briefs,
Control, 54(4), 732–747. 51(11), 640–644.
Ohori, A., Kogiso, K., & Sugimoto, K. (2007). Stabilization of a network control Yüksel, S., & Basßar, T. (2006). On the absence of rate loss in decentralized sensor and
system under bounded disturbances with unreliable communication links via controller structure for asymptotic stability. In Proceedings of the American
common Lyapunov function approach. In SICE Annual Conference 2007 (pp. control conference (pp. 5562–5567). Minneapolis, MN, USA.
1898–1903). Takamatsu, Kagawa, Japan. Yüksel, S., & Basßar, T. (2007). Communication constraints for decentralized
Roberts, D., & Stilwell, D. (2005). Control of an autonomous vehicle platoon with a stabilizability with time-invariant policies. IEEE Transactions on Automatic
switched communication network. In Proceedings of the 2005 American control Control, 52(6), 1060–1066.
conference (pp. 4333–4338). Portland, OR. Yun, S., Choi, Y., & Park, P. (2009). H2 control of continous-time uncertain linear
Rotkowitz, M., & Lall, S. (2006). A characterization of convex problems in systems with input quantization and matched disturbance. Automatica, 45,
decentralized control. IEEE Transactions on Automatic Control, 51(2), 274–286. 2435–2439.
Seth, S., Lynch, J., & Tilbury, D. (2005). Wirelessly networked distributed controllers Zečević, A., & Šiljak, D. (2010). Control of complex systems. Structural constraints and
for real-time control of civil structures. In Proceedings of the American control uncertainty. Berlin–Heidelberg: Springer-Verlag.
conference (pp. 2946–2952). Portland, OR. Zhai, G., Chen, N., & Gui, W. (2009). A study on decentralized H1 feedback control
Smith, R., & Hadaegh, F. (2007). Closed-loop dynamics of cooperative vehicle systems with local quantizers. Kybernetika, 45(1), 137–150.
formations with parallel estimators and communication. IEEE Transactions on Zhong, W., Stefanovski, J., Dimirovski, G., & Zhao, J. (2009). Decentralized control
Automatic Control, 52(8), 1404–1414. and synchronization of time-varying complex dynamic systems. Kybernetika,
Stubbs, A., Vladimerou, V., Fulford, A., Strick, J., & Dullerud, G. E. (2006). A hovercraft 45(1), 151–167.
testbed for networked and decentralized control. IEEE Control Systems Magazine, Zhu, X.-L., & Yang, G.-H. (2008a). Quantized H1 controller design for networked
26(3), 56–69. control systems. In Proceedings of the 2008 Chinese control and decision
Sun, Y., & El-Farra, N. (2009). Robust quasi-decentralized networked control of conference (pp. 357–362). Yantai, PR China.
process systems. In Proceedings of the 48th IEEE conference on decision and Zhu, X.-L., & Yang, G.-H. (2008b). Stability analysis and state feedback control of
control and the 28th Chinese control conference (pp. 5851–5856). Shanghai, PR networked control systems with multi-packet transmission. In Proceedings of
China. the 2008 American control conference (pp. 3133–3138). Seattle, WA, USA.
Tatikonda, S. (2003). Some scaling properties of large distributed control systems. In
Proceedings of the 42nd IEEE conference on decision and control (pp. 3142–3147). Lubomír Bakule received his Ph.D. degree in Control Engineering from the
Maui, HI, USA. Czechoslovak Academy of Sciences, Prague, Czechoslovakia in 1974. Since 1974 he
Šiljak, D. (1991). Decentralized control of complex systems. New York: Academic has been with the Institute of Information Theory and Automation, Academy of
Press. Sciences of the Czech Republic, where he is Chief Researcher. He has held numerous
Wang, X., & Lemmon, M. (2011). Event-triggering in distributed networked control visiting appointments abroad. He has published over 150 research papers and has
systems. IEEE Transactions on Automatic Control, 56(3), 586–601. led many research projects. His research interests are in the theory and applications
Wang, Y., & Yang, G. (2009). State feedback control synthesis for networked control of robust decentralized control of large-scale systems, networked control systems,
systems with packet dropout. Asian Journal of Control, 11(1), 49–58. delayed and switching control systems, and structural control. He is Vice-Chair of
Wei, J. (2008). Stability analysis of decentralized networked control systems. In: the IFAC TC on Large Scale Complex Systems.
Proceedings of the 7th world congress on intelligent control and automation (pp.
5477–5482). Chongquing, PR China. Martin Papík received his Ph.D. degree in System Engineering and Informatics from
Wei, L., Xue, D.-Y., Da-zhi, E. (2007). Some basic issues in networked control the Czech University of Life Sciences, Prague in 2010. Since 2005 he has been with
systems. In Proceedings of the 2nd IEEE conference on industrial electronics and the Institute of Information Theory and Automation, Academy of Sciences of the
applications (pp. 2098–2102). Harbin, PR China. Czech Republic, where he is Research Assistant. His research interests are in the
Wen, D.-L., & Yang, G.-H. (2009). Quantized H1 control for networked control theory and applications of decentralized control, supervision, Petri nets, computer
systems with random delays. In Proceedings of the 21th Chinese control and networks, and complex systems.
decision conference (pp. 588–592). Guilin, PR China.