9th International Symposium on Advanced Control of Chemical Processes

9th
Symposium
on
June
7-10, 2015. Whistler,
British
Columbia,Control
Canadaof
9th International
International
Symposium
on Advanced
Advanced
Control
of Chemical
Chemical Processes
Processes
9th
International
Symposium
on
Advanced
Control
of
Chemical
Processes
June
7-10,
2015.
Whistler,
British
Columbia,
Canada
Available
online
June
7-10,
2015.
Whistler,
British
Columbia,
Canada
June 7-10, 2015. Whistler, British Columbia, Canada at www.sciencedirect.com

ScienceDirect
IFAC-PapersOnLine 48-8 (2015) 551555

Distributed
Model Predictive
Control
Distributed
Distributed Model
Model Predictive
Predictive Control
Control
Based
on
Nash
Optimality
for
Large
Scale
Based
on
Nash
Optimality
for
Large
Based on Nash Optimality for Large Scale
Scale
Irrigation
Systems

Irrigation
Irrigation Systems
Systems
Rongchao Zhang Andong Liu Li Yu Wen-an Zhang
Rongchao
Zhang Andong
Liu Li
Yu Wen-an
Zhang
Rongchao
Rongchao Zhang
Zhang Andong
Andong Liu
Liu Li
Li Yu
Yu Wen-an
Wen-an Zhang
Zhang

Department of Automation, Zhejiang University of Technology,

Department of Automation, Zhejiang University of Technology,

Department
of Automation,
Automation,
Zhejiang
University
of Technology,
Technology,
Zhejiang
Provincial
United KeyZhejiang
Laboratory
of Embedded
Systems,
Department
of
University
of
Zhejiang
Provincial
United
Key
Laboratory
of
Embedded
Systems,
Zhejiang
Provincial
United
Key
Laboratory
of
Embedded
Systems,
Hangzhou
310023, P.R.
China,
(e-mail:
zhangrongchao@163.com)
Zhejiang
Provincial
United
Key
Laboratory
of
Embedded
Systems,
Hangzhou
310023,
P.R.
China,
(e-mail:
zhangrongchao@163.com)
Hangzhou
310023,
P.R.
China,
(e-mail:
zhangrongchao@163.com)
Hangzhou 310023, P.R. China, (e-mail: zhangrongchao@163.com)

Abstract: Irrigation system is a large scale system, consisting of many interacting channels, and
Abstract:
Irrigation
system
is
a large
scale
system,
consisting
many
interacting
channels,
and
Abstract:
Irrigation
system
is
scale
consisting
of
many
channels,
and
spanning vast
geographical
areas.
In practice,
the water
level isof
expected
to the desired
reference
Abstract:
Irrigation
systemareas.
is a
a large
large
scale system,
system,
consisting
ofexpected
many interacting
interacting
channels,
and
spanning
vast
geographical
In
practice,
the
water
level
is
to
the
desired
reference
spanning
vast
geographical
areas.
In
practice,
the
water
level
is expected
expected
to the
theisdesired
desired
reference
values forvast
safety
and eciency.
InIn
this
paper, the
thewater
modellevel
of irrigation
system
modeled
by the
spanning
geographical
areas.
practice,
is
to
reference
values
for
safety
and
eciency.
In
this
paper,
irrigation
system
is
modeled by
the
values
for
safety
and
In
paper,
the
model
of
irrigation
system
is
the
principle
on
conservation
of mass.
Because
ofthe
themodel
strongof
coupling
between
neighboring
pools
values
foron
safety
and eciency.
eciency.
In this
this
paper,of
the
model
ofcoupling
irrigationbetween
system neighboring
is modeled
modeled by
by
the
principle
conservation
of
mass.
Because
the
strong
pools
principle
on
conservation
of
mass.
Because
of
the
strong
coupling
between
neighboring
pools
in
irrigation
system
and
large
disturbance
from
the
environment,
a
non-cooperative
distributed
principle
on
conservation
of
mass.
Because
of
the
strong
coupling
between
neighboring
pools
in
irrigation
system
and
large
disturbance
from
the
environment,
aa non-cooperative
distributed
in
irrigation
system
and
disturbance
from
environment,
distributed
model
predictive
control
(NDMPC)
algorithm
based
on Nash optimality
is proposed
for the
in
irrigation
system
and large
large
disturbance
from the
the
environment,
a non-cooperative
non-cooperative
distributed
model
predictive
control
(NDMPC)
algorithm
based
on
Nash
optimality
is
proposed
for
model
predictive
control
(NDMPC)
algorithm
based on
on
Nash
optimality
is proposed
proposed
for the
the
large scale
irrigation
system
to design
the controller
for Nash
setting
points regulation.
Finally,
a
model
predictive
control
(NDMPC)
algorithm
based
optimality
is
for
the
large
scale
irrigation
design
the
controller
for
setting
regulation.
Finally,
aa
large
scale
irrigation
system
to
design
the
controller
for
setting
points
regulation.
Finally,
simulation
example
is system
given toto
demonstrate
the eectiveness
of thepoints
proposed
algorithm.
large
scale
irrigation
system
to
design
the
controller
for
setting
points
regulation.
Finally,
a
simulation
example
is
given to
demonstrate the
eectiveness
of
the
proposed
algorithm.
simulation
example
is
to
the
eectiveness
of
the
proposed
algorithm.
simulation
example
is given
given
to demonstrate
demonstrate
the Control)
eectiveness
of by
theElsevier
proposed
2015, IFAC
(International
Federation
of Automatic
Hosting
Ltd. algorithm.
All rights reserved.
Keywords: Distributed model predictive control, Irrigation systems, Nash optimality, Large
Keywords:
Distributed
model
predictive
control,
Irrigation
systems,
Nash
optimality,
Keywords:
Distributed model
model predictive
predictive control,
control, Irrigation
Irrigation systems,
systems, Nash
Nash optimality,
optimality, Large
Large
scale systems.
Keywords:
Distributed
Large
scale
systems.
scale
systems.
scale systems.
1. INTRODUCTION
There are lots of results for large scale irrigation system. In
1.
There
are
lots
of
results
large
scale
irrigation
In
1. INTRODUCTION
INTRODUCTION
There
are
lots
of
for
large
irrigation
system.
In
Cantoni
al (2007).,
thefor
irrigation
system
modelsystem.
had been
1.
INTRODUCTION
There
areet
lots
of results
results
for
large scale
scale
irrigation
system.
In
Cantoni
et
al
(2007).,
the
irrigation
system
model
had
been
Cantoni
et
al
(2007).,
the
irrigation
system
model
had
been
described
detailedly
and
a
closed-loop
centralized
control
et
al
(2007).,
the
irrigation
system
model
had
been
Water is one of the most vital elements in human life. Cantoni
described
detailedly
and
a closed-loop
centralized
control
Water
one
of
vital
elements
in
life.
described
detailedly
and
centralized
control
and a decentralized
control algorithm
openWater
isother
one things,
of the
the most
most
vital for
elements
in human
human
life. algorithm
described
detailedly
and aa closed-loop
closed-loop
centralizedof
control
Amongis
it is used
drinking,
agriculture,
Water
is
one
of
the
most
vital
elements
in
human
life.
algorithm
and
a
decentralized
control
algorithm
of
openAmong
other
things,
it
is
used
for
drinking,
agriculture,
algorithm
and
a
decentralized
control
algorithm
of
openwater
channels
had
been
proposed
from
the
perspective
of
Among
other
things,
it
is
used
for
drinking,
agriculture,
algorithm
and
a
decentralized
control
algorithm
of
opentransportation,
recreation,
and energy
production.
Irriga- water channels had been proposed from the perspective
Among
other things,
it is used
for drinking,
agriculture,
of
transportation,
recreation,
and
energy
production.
Irrigawater
channels
had
been
proposed
from
the
perspective
of
large
scale
irrigation
network
management.
In
Bolea
et
al
transportation,
recreation,
and
energy
production.
Irrigawater
channels
had
been
proposed
from
the
perspective
of
tion
systems
consist
of
water
bodies,
such
as
lakes
and
transportation,
recreation,
and
energy
production.
Irrigalarge
scale
irrigation
network
management.
In
Bolea
et
al
tion
systems
consist
of
water
bodies,
such
as
lakes
and
large
scale
irrigation
network
management.
In
Bolea
et
al
(2014).,
the
irrigation
systems
were
suitably
represented
tion
systems
consist by
of natural
water bodies,
bodies,
such as
as lakes
lakes
and large
scale
irrigation
network
management.
In
Bolea
et
al
reservoirs,
connected
water channels
and mantion
systems
consist
of
water
such
and
(2014).,
the
irrigation
systems
were
suitably
represented
reservoirs,
connected
by
natural
water
channels
and
man(2014).,
the
irrigation
systems
were
suitably
represented
for
control
purposes
by
using
linear
parameter-varying
reservoirs,
connected
by
natural
water
channels
and
man(2014).,
the
irrigation
systems
were
suitably
represented
made channels.
The by
water
owswater
in the
rivers and
pools
reservoirs,
connected
natural
channels
and
manfor
purposes
by
using
linear
made
The
ows
in
rivers
and
for control
control
purposes
bysystem
using parameters
linear parameter-varying
parameter-varying
(LPV)
models,
and the
can be easily
made
channels.
The water
water
ows and
in the
the
rivers
and pools
pools for
control
purposes
by
using
linear
parameter-varying
can bechannels.
manipulated
by pumps
gates
(Agriculture
made
channels.
The
water
ows
in
the
rivers
and
pools
(LPV)
models,
and
the
system
parameters
can
easily
can
be
manipulated
by
pumps
and
gates
(Agriculture
(LPV)
models,
and
the
system
parameters
can
be
easily
identied
from
input-output
data
by
means
of be
classical
can
be
manipulated
by
pumps
and
gates
(Agriculture
(LPV)
models,
and
the
system
parameters
can
be
easily
and
Water,
2010).
Water
is
drawn
from
the
reservoir
and
can
be
manipulated
by
pumps
and
gates
(Agriculture
identied
from
input-output
data
by
means
of
classical
and
Water,
2010).
Water
is
drawn
from
the
reservoir
and
identied
from
input-output
data
by
means
of
classical
identication
techniques,
moreover,
a
decentralized
conand
Water, through
2010). Water
Water
is drawn
drawn
from
the
reservoir
and identied
from
input-output
data
by
means
of
classical
distributed
the main
channel
andthe
many
secondary
and
Water,
2010).
is
from
reservoir
and
identication
techniques,
moreover,
aa decentralized
condistributed
through
the
main
channel
and
many
secondary
identication
techniques,
moreover,
decentralized
controller
is
designed
for
the
irrigation
system.
Because
of
the
distributed
through
the
main
channel
and
many
secondary
identication
techniques,
moreover,
a
decentralized
channels to through
farms. Along
thechannel
channels,
mechanical
gates troller is designed for the irrigation system. Because ofcondistributed
the main
andmechanical
many secondary
the
channels
to
farms.
Along
the
channels,
gates
troller is
is designed
designed
for the
the irrigation
irrigation
system.
Because of
ofburthe
expensive
communication
cost
and
huge
computation
channels
to
farms.
Along
the
channels,
mechanical
gates
troller
for
system.
Because
the
are
installed
to
regulate
the
ow.
A
stretch
of
water
bechannels
to farms.
Alongthe
theow.
channels,
mechanical
gates
expensive
communication
cost
and
huge
computation
burare
installed
to
regulate
A
stretch
of
water
beexpensive
communication
cost
and
huge
computation
burden,
the
centralized
control
algorithm
is
confronted
with
are
installed
to
regulate
the
ow.
A
stretch
of
water
beexpensive
communication
cost
and
huge
computation
burtween
two
neighboring
gates
is
called
a
pool.
An
irrigation
are
installed
to regulate
the isow.
A stretch
of water
be- den, the centralized control algorithm is confronted with
tween
neighboring
gates
irrigation
den,
the
centralized
control
is
with
diculties
in practical
applications.
In the decentween
two
neighboring
gates
is called
called a
a pool.
pool.
An
irrigation
den,
the
centralized
control algorithm
algorithm
is confronted
confronted
with
systemtwo
is usually
largely
gravity-fed(i.e.
thereAn
is no
pump- great
tween
two
neighboring
gates
is
called
a
pool.
An
irrigation
great
diculties
in
practical
applications.
In
the
decensystem
is
usually
largely
gravity-fed(i.e.
there
is
no
pumpgreat
diculties
in
practical
applications.
In
the
decentralized
control,
there
is
no
information
transfer
between
system
is
usually
largely
gravity-fed(i.e.
there
is
no
pumpgreat
diculties
in
practical
applications.
In
the
decening); to issatisfy
water-demands
from farms
andistonodecrease
system
usually
largely
gravity-fed(i.e.
there
pumptralized
control, there
there
islead
no to
information
transfer
between
ing);
satisfy
from
to
control,
no
information
transfer
between
local controllers,
whichis
a poor system
performance
ing);
to
satisfy water-demands
water-demands
from
farms
and
to decrease
decrease
tralized
control, there
islead
no to
information
transfer
between
waterto
wastage,
the water levels in
thefarms
pools and
should
be regu- tralized
ing);
to
satisfy
water-demands
from
farms
and
to
decrease
local
controllers,
which
aaetpoor
system
performance
water
wastage,
the
water
levels
in
the
pools
should
be
regulocal
controllers,
which
lead
to
poor
system
performance
of
irrigation
system.
In
Farina
al
(2011).,
a
distributed
water
wastage,
the
water
levels
in
the
pools
should
be
regulocal
controllers,
which
lead
to
a
poor
system
performance
lated
to
certain
set
points
(Li
and
Schutter,
2011;
Rostalski
water
wastage,
the
water levels
in the
pools should
be regu- of irrigation system. In Farina et al (2011)., a distributed
lated
certain
set
(Li
2011;
of
irrigation
system.
In
et
(2011).,
aa distributed
horizon
estimation
(MHE)
proposed
lated
to
certain
set points
points
(Li and
and
Schutter,
2011; Rostalski
Rostalski
of
irrigation
system.
In Farina
Farina
et al
al method
(2011)., was
distributed
et al.,to
2008).
Since
most farms
sitSchutter,
at the downstream
ends moving
lated
to
certain
set
points
(Li
and
Schutter,
2011;
Rostalski
moving
horizon
estimation
(MHE)
method
was
proposed
et
al.,
2008).
Since
most
farms
sit
at
the
downstream
ends
moving
horizon
estimation
(MHE)
method
was
proposed
for
discrete-time
nonlinear
systems
decomposed
into couet
al.,
2008).
Since
most
farms
sit
at
the
downstream
ends
moving
horizon
estimation
(MHE)
method
was
of pools,
it is Since
more important
tosit
control
downstream
water
et
al., 2008).
most farmsto
at thedownstream
downstreamwater
ends for discrete-time nonlinear systems decomposed proposed
into coucouof
pools,
it
is
more
important
control
for
discrete-time
nonlinear
systems
decomposed
into
pled
pools
with
non-overlapping
states.
of
pools,
it
is
more
important
to
control
downstream
water
for
discrete-time
nonlinear
systems
decomposed
into coulevels
by
controlling
the
gate
opening.
To
avoid
excessive
of
pools,
itcontrolling
is more important
to
control To
downstream
water pled pools with non-overlapping states.
levels
by
the
gate
opening.
avoid
excessive
pled
pools
with
non-overlapping
states.
levels
by controlling
controlling
the large
gate opening.
opening.
To avoid
avoid
excessive
pools withetnon-overlapping
states.
communication
load for
scale irrigation
systems,
de- pled
levels
by
the
gate
To
excessive
In Negenborn
al (2006)., a novel
DMPC algorithm was
communication
load
scale
deIn
Negenborn
et
al
(2006).,
a
novel
DMPC
algorithm
was
communication
load for
for large
large
scale irrigation
irrigation systems,
systems,
de- In
centralized control(there
is no communication
between any
communication
load
for
large
scale
irrigation
systems,
deNegenborn
et
al
(2006).,
a
novel
was
presented
based
on(2006).,
Lagrange
dual DMPC
theory algorithm
for transport
In
Negenborn
et
al
a
novel
DMPC
algorithm
was
centralized
control(there
is
no
communication
between
any
based
on
Lagrange
dual
theory
for
transport
centralized
control(there
is no
no communication
communication
between any
any presented
local controller)
and distributed
control(communication
centralized
control(there
is
between
presented
based
on
Lagrange
dual
theory
for
transport
systems
to
handle
couplings
between
subsystems.
An
itpresented
based
on
Lagrange
dual
theory
for
transport
local
controller)
and
distributed
control(communication
to
handle
couplings
between
subsystems.
An
itlocal
controller)
and distributed
distributed
control(communication
delivercontroller)
between neighboring
local controllers)
are preferred systems
local
and
control(communication
systems
to
handle
couplings
between
subsystems.
An
iterative
DMPC
algorithm
was
developed
based
on
Nash
systems
to handle
couplings
between
subsystems.
An
itdeliver
between
neighboring
local
controllers)
are
preferred
erative
DMPC
algorithm
was
developed
based
on
Nash
deliver
between
neighboring
local
controllers)
are
preferred
to
centralized
control.
Moreover,
an
irrigation
channel
is
a
deliver
between
neighboring
local
controllers)
are
preferred
erative
DMPC
algorithm
was
developed
based
on
Nash
optimality
for
large
scale
power
system
to
tackle
the
state
erative
DMPC
algorithm
was
developed
based
on
Nash
to
centralized
control.
Moreover,
an
irrigation
channel
is
a
optimality
for
large
scale
power
system
to
tackle
the
state
to
centralized
control.
Moreover,
anbetween
irrigation
channel
is
system
presenting
strong
coupling
pools,
i.e. is
theaa optimality
to
centralized
control.
Moreover,
an
irrigation
channel
for
scale
system
to
the
coupling between
the relevant
for large
largesubsystems,
scale power
powerand
system
to tackle
tacklecomputathe state
state
system
strong
pools,
i.e.
coupling
between
subsystems,
and
the
relevant
computasystem
presenting
strongtocoupling
coupling
between
pools,
i.e. the
the optimality
ow intopresenting
a pool is equal
the ow between
out of the
neighboring
system
presenting
strong
coupling
between
pools,
i.e.
the
coupling
between
subsystems,
and
the
relevant
computation
convergence
and
the
nominal
stability
condition
were
coupling
between
subsystems,
and
the
relevant
computaow
into
a
pool
is
equal
to
the
ow
out
of
the
neighboring
convergence
and
the
nominal
stability
condition
were
ow
into aapool.
pool With
is equal
equal
to gate
the ow
ow
out of
ofcontrolling,
the neighboring
neighboring
upstream
local
opening
when tion
ow
into
pool
is
to
the
out
the
tion
convergence
and
the
nominal
stability
condition
were
presented
for
unconstrained
DMPC
(Li
et
al.,
2005;
Zhang
convergence
and the nominal
stability
condition
were
upstream
pool.
local
gate
opening
controlling,
when
presented
for
unconstrained
DMPC
(Li
et
al.,
2005;
Zhang
upstream
pool.atWith
With
local
gate pool
opening
controlling,
when tion
otakes occur
a down
stream
(Litrico
and Fromion,
upstream
pool.
With
local
gate
opening
controlling,
when
presented
for
unconstrained
DMPC
(Li
et
al.,
2005;
Zhang
and
Li,
2007).
Considering
dierent
forms
of
collaborapresented
for
unconstrained
DMPC
(Li
et
al.,
2005;
Zhang
otakes
occur
at
a
down
stream
pool
(Litrico
and
Fromion,
and
Li,
2007).
Considering
dierent
forms
of
collaboraotakes
occur at
at aa down
down stream
stream pool
pool (Litrico
(Litrico and
and Fromion,
Fromion, and
2003, 2005).
otakes
occur
dierent
of
tion, Li,
the2007).
DMPCConsidering
can be classied
intoforms
two categories:
1)
and
Li,
2007).
Considering
dierent
forms
of collaboracollabora2003,
tion,
the
DMPC
can
be
classied
into
two
categories:
1)
2003, 2005).
2005).
2003,
2005).
tion,
the
DMPC
can
be
classied
into
two
categories:
1)
Distributed
algorithms
where
each
local
controller
minition,
the
DMPC
can
be
classied
into
two
categories:
1)
Distributed
algorithms
where
each
local
controller
miniDistributed
algorithms
where
each
local
controller
minimizes
a
global
cost
function
(cooperating
algorithms).
In
Distributed
algorithms
where
each
local
controller
mini This research work was supported by the National Natural Science
mizes
global
cost
(cooperating
algorithms).
In

mizes aaa et
global
cost function
function
(cooperating
algorithms).
In
Stewart
al (2010).,
a cooperative
linear DMPC
strategy
This
research
work
was
supported
by
the
National
Science

mizes
global
cost
function
(cooperating
algorithms).
In
Foundation
of China
the Grants
andNatural
61273117,
and
This
work
was
by
the
Natural
Science

Stewart
et
al
(2010).,
a
cooperative
linear
DMPC
strategy
This research
research
work under
was supported
supported
by 61403344
the National
National
Natural
Science
Stewart
et
al
(2010).,
a
cooperative
linear
DMPC
strategy
Foundation
of
China
under
the
Grants
61403344
and
61273117,
and
was presented
for anyanite
numberlinear
of subsystems
and a
Stewart
et
al
(2010).,
cooperative
DMPC
strategy
the
Zhejiang
Qianjiang
talent
project(QJD1302012).
Foundation
of
China
under
the
Grants
61403344
and
61273117,
and
Foundation
China under
theproject(QJD1302012).
Grants 61403344 and 61273117, and
was
presented for
any nite
number of
subsystems and
a
the
ZhejiangofQianjiang
talent
was
was presented
presented for
for any
any nite
nite number
number of
of subsystems
subsystems and
and aa
the
the Zhejiang
Zhejiang Qianjiang
Qianjiang talent
talent project(QJD1302012).
project(QJD1302012).
Copyright
2015 IFAC
552 Hosting by Elsevier Ltd. All rights reserved.
2405-8963 2015,
IFAC (International Federation of Automatic Control)
Copyright
2015
IFAC
552
Copyright

2015
IFAC
552
Peer review
of International Federation of Automatic
Copyright
under
2015 responsibility
IFAC
552Control.
10.1016/j.ifacol.2015.09.025

IFAC ADCHEM 2015

552
June 7-10, 2015. Whistler, BC, Canada

Rongchao Zhang et al. / IFAC-PapersOnLine 48-8 (2015) 551555

stabilizability condition was also presented. A cooperative

DMPC algorithm was investigated to provide an approach
based on negotiation, cooperation and learning techniques
(Javalera et al., 2010). In Morosan et al (2010)., a cooperative DMPC strategy had been proposed for building
temperature regulation using electrical convectors. 2) Distributed algorithms where each local controller minimizes
a local performance index (non-cooperating algorithms).
A NMPC algorithm was investigated to deal with optimal
control of large-scale power systems (Camponogara et al.,
2002). In Mercangoz and Doyle (2007), a NDMPC algorithm was proposed for liquid level control system with
four tanks.
There are also some results about DMPC for large scale
irrigation system. In Alvarez et al (2013)., a NDMPC
algorithm had been proposed to the control of an accurate
model of an actual irrigation system in Spain, but the cooperative DMPC algorithm increases much of the computation burden to the local controller. A cooperating DMPC
algorithm based on Lagrange dual theory was proposed
for irrigation system (Negenborn et al., 2009), but the
proposed algorithm was complicated with slow converging
speed. To the best of authors knowledge, although some
of results have been available about DMPC for irrigation
system, there are still some problems needing much further
investigation, such as handing the strong coupling of the
pools and large disturbance from environment in irrigation
system, and reducing computation burden and communication cost in controller design. Besides, the irrigation
system performance under the controlling of the existence
algorithms is not satisfactory and lots of algorithms are
still too complicated. NDMPC based on Nash optimality
is a powerful approach to design controller with low computation burden and communication cost, fast converging
speed and preferable system performance. In view of the
above discussion, we are motivated to study NDMPC
based on Nash optimality for the large scale irrigation
systems.
2. MODELING OF THE IRRIGATION SYSTEM
An irrigation system consists of a main channel and several
secondary channels, see Fig. 1. An irrigation channel is in
fact a string of pools, and Fig. 2 shows the sideview of
pools and the illustration of associated variables. With in
the context of control design for set point regulation and
load disturbance rejection, the following model is given
based in principle on conservation of mass
1
x i (t) = xi (t) + ui (t)
i
1
vi,h (t) + di (t)

h
i
i
s.t.

di (t) is the disturbance. Xi , Ui is the feasible domain of

xi (t), ui (t).

Fig.1. Irrigation system.

Fig.2. Illustration of pool i and associated variables

Discretize system (1) with the sampling time T , the
discrete time state space model can be described as follows
T
xi (k + 1) = xi (k) + ui (k)
i
T

vi,h (k) + T di (k)

hi
i
s.t.

(2)

xi (k) Xi
ui (k) Ui

3. NDMPC BASED ON NASH OPTIMALITY

ALGORITHM DESIGN
3.1 Local performance index of pool

(1)

xi (t) Xi
ui (t) Ui

where xi (t) is the water level of pool i, i (1, ..., n), i

is the surface area of pool i, i is the set of neighboring
downstream pool i, ui (t) is the control decision of pool
i (the ow from the neighboring upstream pool j to the
pool i), vi,h (t) is the ow from the pool i to the neighboring
downstream pool h with vj,i (t) = ui (t) and vi,h (t) = uh (t),
553

Considering the strong coupling between neighboring

pools, communication constrains between pools and large
disturbance in irrigation system, the NDMPC based on
Nash optimality algorithm will be an ecient approach
to handle these problems in controlling irrigation system.
The structure of the NDMPC is shown in Fig 3. Because
of the spacial structure of irrigation system, there is only
control decision coupling between neighboring pools, and
the direction of the information transfer is one-way, from
downstream to upstream. Based on Nash optimality algorithm, the solution of NDMPC is computed in iterative
form. In order to describe the algorithm clearly, the system
model of pool can be rewritten as follows:

IFAC ADCHEM 2015

June 7-10, 2015. Whistler, BC, Canada

Rongchao Zhang et al. / IFAC-PapersOnLine 48-8 (2015) 551555

xi (k + 1) = xi (k) + Bi ui (k)
T

uh (k) + T di (k)
hi
i

T
i ,

uh (k) = vi,h (k).

Fig.3. Illustration of NDMPC structure in irrigation

system
According to (3), each local controller computes the local
control decision based on neighborhood communication,
and the optimize problem for local performance index is
dened as follows
min

ui (k|k)

Ji (k)
(4)

..
.

ui (k+M |k)

where
Ji (k) =

xi (k + s|k) xi,r (k + s) Qi
s=1

M 1
+
ui (k + s|k) Ri

(5)

s=0

iu
i (k)
xi (k + s|k) = xi (k) + B

iu
h (k)
B

h
i

(7)

where ki,1 and ki,2 is the local controller gains of pool i.

3.2 Algorithm design

xi (k) Xi
ui (k) Ui

s.t.
where Bi =

ui (k) = ki,1 xi (k) + ki,2

(3)

553

(6)

and u
i (k) = [ui (k|k) , , ui (k + s 1|k)] ,
T

Bi = [Bi , , Bi ] . when s M , then s = M . xi (k + s|k)

sBi

and ui (k + s|k) are the predictive water level and control

decision, respectively, at time instant k + s based on the
water level at time instant k of pool i. xi,r is the set
point of water level for pool i. Qi and Ri are the weight
coecients matrices of water level and control decision
with appropriate dimension, P is the prediction horizon,
M is the control horizon and M P . The objective of the
each local controller is to regulate the pools water level to
the set point while keeping the performance index minimal
and satisfying the above constraints. So the local control
decisions ui (k) is computed by solving local optimization
min Ji (k) with constraints in the NDMPC based on Nash
optimality. The form of the local controller is given
554

The Nash optimality algorithm is one of the iterative

algorithms, which is developed to seek the local optimal
control decision for each subsystem at each sampling time.
Each subsystem resolves its local optimization problem
with the local optimal control decision of its neighbours
available. Then each subsystem compares the newly computed control decision with that obtained in last iteration
and checks if the terminal condition is satised. If the
algorithm is convergent, all the terminal conditions of all
the subsystems will be satised and the iteration will be
terminated; otherwise, each subsystem then transfer the
newly computed control decision to its neighbours and
resolves its local problem with the updated values for
neighbours. This optimization process will be repeated at
the next sampling time. The overall control system will
converge at Nash equilibrium. However, the Nash optimal
solution to the local optimization problem may not equal
to the global optimal control decision. In large scale irrigation systems, the goal of communication and exchanging
solutions among neighboring local controllers is to achieve
the optimal solutions of the pools in an iterative fashion.
Based on Nash optimality, an online iterative algorithm
for NDMPC is proposed to seek the local optimal control
decision for each pool at each time instant. The algorithm
is summarized as Algorithm 1.
Algorithm 1
S1 : Initialization:
At time instant k, each local controller measure its
initial water level xi (k|k), initialize its predictive conT
trol decision u
pi (k) = [ui (k|k), ..., ui (k + M 1|k)] ,
p = 0, here p is the iteration times.
S2 : Communication:
Each local controller transmits the predictive control
decision u
pi (k) to neighbors. Simultaneously, local
controller receive its neighbors predictive control
decisions.
S3 : Local optimization:
Each local controller solves its local optimization
problem objective (4) subject to (5) and (6) to derive
its local optimal control decision u
p+1
(k).
i
S4 : Checking and updating:

p+1
Given error accuracy i , if
pi (k) i ,
ui (k) u
Set the local optimal control decision for time instant
(k|k), then go to S5. Else, go
k with ui (k|k) = up+1
i
to the next iteration p = p + 1, go to S2.
S5 : Receding horizon:
Let [ui (k + 1| k + 1), ..., ui (k + M |k + 1)] =
[ui (k + 1|k), .., ui (k + M 1|k), ui (k + M 1|k)].
Move horizon to the next sampling time k = k + 1,
go to S2, and repeat the above steps.
3.3 Stability analysis
The model of the overall irrigation system can be described
as

IFAC ADCHEM 2015

554
June 7-10, 2015. Whistler, BC, Canada

Rongchao Zhang et al. / IFAC-PapersOnLine 48-8 (2015) 551555

x (k + 1) = x (k) + Bu (k)
= (I + BK1 ) x (k) + K2
xX
uU
T

b11
..
B= .
bn1

..
.

b1n

i
.. , b =
T
ij
.

i
bnn

i = j,
j
i,

else.

According to the above analysis, the theorem of system

stability can be summarized as follows
Theorem 1. The nominal stability of the irrigation system
can be guaranteed if only if the following condition
|m (I + BK1 )| < 1
m = 1, ..., N

(9)

holds, where the eigenvalues m , m (1, ..., n) of the

above matrix are all in the unit circle.

5.6

4. SIMULATION
In this section, an irrigation system is used to illustrative
the eectiveness of the proposed NDMPC based on Nash
optimally algorithm for the large scale irrigation system.
A schematic of this process is shown in Fig. 4, and the
parameter values of this irrigation system is given in Table
1. It can be seen that the irrigation system is composed
by one reservoir and ve pools, and each pool can supply
water to one or serval farms. The water level can be adjust
by the gates (Each gate is controlled by local controller).

5.5
water level
set point

5.4
5.3

x2

It should be noted that the stability condition in (9) is

global because each subsystems controller gain ki,1 , i
(1, ..., n) computed by distributed algorithm illustrated in
last section should satisfy the global stability condition.

x1

Let M = P = 5, Qi = 400, Ri = 1, i (1, ..., 5). Then, the

controllers are designed by Algorithm 1, which presented
in section 3. The simulation results are shown in Fig. 5
and Fig. 6. Water level response of irrigation system in
case of starting at the point which departure from the set
points is shown in Fig. 5, and the corresponding control
decision is shown in Fig. 6. It can seen from Fig. 5 and
Fig. 6 that the water level would be adjust to the set
points quickly by using the proposed algorithm when the
irrigation system start at the point which departure from
the set points. At t = 150s, there is a disturbance at pool
5, which may be the water releasing to the nearby farms.
The water levels of pool 5, pool 4 and pool 1 are aected
by the disturbance, because the upstream pools need to
convey water down to the downstream pools to makeup
the shortage of water . As shown in Fig. 5, it is more
slower for the water levels of the upstream pools converges
to the set points, but the oset of the water levels do not
amplify but minish in the upstream pools, compared to
the converge speed in downstream pools, and all the water
levels of pools converges to the set points gradually.

50

100

150

200

250

300

5.2
0
4.55

50

100

150

200

250

300

50

100

150

200

250

300

50

100

150

200

250

300

50

100

150

200

250

300

5.3
5.25

4.5

x3

where, K1 = [k1,1 , ..., kn,1 ] , K2 = [k1,2 , ..., kn,2 ] ,

T
T
x = [x1 , ..., xn ] , u = [u1 , ..., un ] , X = [X1 , ...,
T
T
Xn ] , U = [U1 , ..., Un ] . I is the identity matrix with
appropriate dimension.

4.45
4.4
5

x4

s.t.

Table 1: Parameter values of irrigation system

Pool Surface area Set point of water level
1
210 m2
5.5 m
2
160 m2
5.3 m
3
170 m2
4.5 m
4
180 m2
5.0 m
5
180 m2
5.0 m

(8)

4.95
4.9
4.85

x5

5
4.9
4.8
4.7

time t/s

Fig.5. Water level response of irrigation system

5. CONCLUSION
In this paper, a NDMPC based on Nash optimality is
proposed to manage large scale irrigation system. Firstly,
a common irrigation system model is presented. Based on
the irrigation system model, a NDMPC based on Nash optimality algorithm is presented to improve the system performance under the situation of strong coupling between

Fig.4. Schematic description of the irrigation system in

simulation
555

IFAC ADCHEM 2015

June 7-10, 2015. Whistler, BC, Canada

Rongchao Zhang et al. / IFAC-PapersOnLine 48-8 (2015) 551555

pools, communication constrains and large disturbance

from environment. Finally, a simulation example is given
to demonstrate the eectiveness of the proposed algorithm.
In addition, there are still some problems needing much
further investigation, such as the irrigation system model
should be describe more practically. More complex and
practical case studies with time delay will be considered
in the future.

u1

control decision

u2

0
0
1.5

50

100

150

200

250

300

50

100

150

200

250

300

0
0
1.5

50

100

150

200

250

300

50

100

150

200

250

300

50

100

150

200

250

300

1
0.5
0
1.5

u3

u4

0.5

1
0.5
0

u5

0
3
2
1
0

time t/s

Fig.6. Control decision of the irrigation system

REFERENCES
A. Alvarez, M.A. Ridao, D.R. Ramirez and L. Snchez.
Constrained predictive control of an irrigation canal.
Journal of Irrigation and Drainage Engineering, 2013,
139(10), 841854.
B.T. Stewart , A.N. Venkat, J.B. Rawlings, S.J. Wright
and G. Pannocchi. Cooperative distributed model predictive control. Systems Control Letters, 2010, 59(8),
460469.
E. Camponogara, D. Jia, B.H. Krogh and S. Talukdar. Distributed model predictive control. IEEE Transactions
on Control Systems, 2002, 22(1), 4452.
F. Fele, J.M. Maestre, S.M. Hashemy, D.M. de la Pena and
E.F. Camachoa. Coalitional model predictive control of
an irrigation canal. Journal of Process Control, 2014,
24(14), 314325.
M. Cantoni, E. Weyer, Y.P. Li, S.K. Ooi, I. Mareels and
M. Ryan, Control of large-scale irrigation networks.
Proceedings of the IEEE, 2007, 95(1), 7591 .
M. Farina, G. Ferrari-Trecate, C. Romani and R. Scattolini. Moving horizon estimation for distributed nonlinear systems with application to cascade river reaches.
Journal of Process Control, 2011, 21(5), 767774.
M. Mercangoz and F.J. Doyle. Distributed model predictive control of an experimental four-tank system.
Journal of Process Control, 2007, 17(30), 297308.
P.D. Morosan, R. Bourdais, D. Dumur and J. Buisson.
Building temperature regulation using a distributed
556

555

model predictive control. ENERGY AND BUILDINGS,

2010, 42(9), 14451452.
P. Rostalski, G. Papafotiou, C. Setz, A. Heinrich and M.
Morari. Application of model predictive control to a
cascade of river power plants. in Proceedings of the 17th
World Congress, 2008, Seoul, Corea, 1187611983.
R.R. Negenborn, B.D. Schutter and H. Hellendoorn.
Multi-agent model predictive control for transportation
networks. in Proceedings of the 2006 IEEE International
Conference on Networking, Sensing and Control, 2006,
Ft. Lauderdale, FL, 296301.
R.R. Negenborn, P.-J. van Overloop, T. Keviczky and
B.D. Schutter. Distributed model predictive control of
irrigation canals. Networks and Heterogeneous Media,
2009, 4(2), 359380.
S.Y. Li, Y. Zhang and Q.M. Zhu. Nash-optimization
enhanced distributed model predictive control applied
to the Shell benchmark problem. Information Sciences,
2005, 170(2), 329349.
V. Javalera, B. Spain and V. Puig. Negotiation and learning in distributed MPC of large scale systems. in Proceedings of American Control Conference, 2010, Baltimore, MD, 31683173.
Water
Encyclopedia:
Science
and
Issues.
/http:www.wateren
cyclopedia.com/A-Bi/Agriculture-and-water.html,
February.
X. Litrico and V.Fromion. Advanced control politics and
optimal performance for an irrigation canal. In Proceedings of the 2003 ECC, 2003, Cambridge, UK.
X. Litrico and V.Fromion. Design of structured multivariable controllers for irrigation canals. In Proceedings of
44th IEEE CDC-ECC, 2005, Seville, Spain, December,
18811886.
Y. Bolea, V. Puig and J. Bles. Linear parameter varying modeling and identication for real-time control
of open-ow irrigation canals. Environmental modelling
and Software, 2014, 53, 8797.
Y.P. Li and B.D. Schutter. Stability and performance an
analysis of an irrigation channel with distributed control. Control Engineering Practice, 2011, 19(10), 1147
1156.
Y. Zhang and S.Y. Li. Networked model predictive control
based on neighborhood optimization for serially connected large-scale processes. Journal of Process Control,
2007, 17(1), 3751.