Decentralized Control for Urban Drainage Systems Via Population

Dynamics: Bogota Case Study

J. Barreiro-Gomez, G. Obando, G. Riano-Briceno, N. Quijano, and C. Ocampo-Martinez
Abstract Control of Urban Drainage Systems (UDS) is studied for cases in which the distribution of runoff through the
channels of a system is inefficient, i.e. when the capacity of some
structures is not used optimally. A decentralized population
dynamics-based control for UDS is presented, particularly using
the replicator and projection dynamics. For the design, a
methodology to make a partitioning of the system is introduced,
and the design of a population dynamicsbased control per
each partition is proposed. Moreover, a stability analysis of the
closedloop system is made by using passivity theory. Finally,
simulation results show the proposed approach performance in
a segment of the Bogota stormwater UDS case study.

I. I NTRODUCTION
Overflooding events in UDS occur frequently, even when
the intensity of the precipitations that produce them are lower
than the precipitation data that has been used to design the
system [11]. Some of the main reasons of this phenomenon
in urban areas are: the rapid urbanization, infrastructure
complexity, and/or climate changes [11], [3]. For instance,
because of the uncertainty related to the climate changes
at the design stage, the expected probability of failure for
hydraulic structures might increase by a factor of around
30 times [2]. For this reason, the concern of looking for
new design techniques and how to address this problem with
new strategies has become an important research topic in
the field of hydraulic engineering. Most of the proposed
solutions to solve the overflowing problem are related to
procedures at the design stage [1]. However, this approach
might be inefficient for already existing systems, due to the
fact that this approach suggests the need of large expansions
over the system. On the other hand, the control systems
approach suggests to solve the problem of overflowing by
applying an optimal management of the flows throughout
the preestablished system. In a centralized approach, it
is costly and challenging to guarantee the availability of
information in a central manner, which implies also a large
communication network. Furthermore, it is challenging to
J. Barreiro-Gomez and C. Ocampo-Martnez are with the Automatic Control Department, Universitat Polit`ecnica de Catalunya, Institut de Rob`otica i
Inform`atica Industrial (CSIC-UPC), Llorens i Artigas, 4-6, 08028 Barcelona,
Spain. {jbarreiro, cocampo}@iri.upc.edu
J. Barreiro-Gomez, G. Obando, G. Riano-Briceno, and N. Quijano
are with Departamento de Ingeniera Electrica y Electronica. G. RianoBriceno is also with Departamento de Ingeniera Civil, Universidad de los
Andes, Carrera 1A No 18A-10, Colombia {j.barreiro135, ge-oband,
ga.riano949, nquijano}@uniandes.edu.co.
COLCIENCIAS supports J. Barreiro-Gomez and G. Obando. Ag`encia
de Gestio dAjust Universitaris i de Recerca AGAUR supports J. BarreiroGomez. This work has been partially supported by the projects Drenaje
urbano y cambio climatico: Hacia los sistemas de alcantarillados del futuro.
Fase II. COLCIENCIAS 633/2013, ECOCIS (Ref. DPI2013-48243-C2-1R), and EFFINET (Ref. FP7-ICT-2011-8-31855).

design decentralized controllers based on partitioning due

to the dynamical coupling in states that exists throughout
the system. For instance, a general methodology to make a
partitioning of a system in an optimal way is presented in
[6]. The control approach has been widely studied by using
different strategies. In [5], model predictive control is used
to avoid overflows in the UDS. This control strategy is one
of the most used in the literature to solve the UDS control,
and requires a model of the system that suitably represents
its behavior. In contrast, other hydraulic problems have been
solved by using other control techniques. In particular, in
[7] the water distribution problem with a unique source
and different receptor reservoirs has been studied with a
population dynamics approach. As an advantage, the latter
approach does not require of a model of the system.
The main contribution of this paper is the design of a
decentralized population dynamicsbased control for UDS,
that enables a better use of the existing system infrastructure.
In the design process, it is proposed to make a partitioning
of the UDS by using a criterion that only depends on the
topology of the system, different from the one proposed in
[6]. The UDS is divided into different subsystems composed
by a group of reservoirs whose outflows converge to a
same receptor structure. Then, a population dynamics-based
control is designed per each partition. The proposed population dynamics-based controllers are designed for a topology
different from the one treated in [7], where partitions are
not considered, and where flows do not converge to a
reservoir. Then, each local controller is in charge of making
an allocation of the available volume capacities, and the
different local controllers do not communicate with each
other leading to a decentralized controller. Moreover, the
stability of the whole closedloop system is analyzed and
proved via passivity theory, associated to the UDS and the
population dynamics.
The paper is organized as follows. Section II presents the
UDS and a control oriented model. Section III introduces
the proposed partitioning methodology, and the population
dynamics-based controller. Then, stability analysis for each
local closedloop control system is presented by using passivity theory in Section IV. Section V shows the description
of the Bogota case study. In Section VI results are shown
and a discussion is made. Finally, concluding remarks and
future work are presented in Section VII.
Notation: Calligraphy letters are used to denote sets, e.g.,
S. Column vectors are denoted with bold font, e.g., x. The
column vector with n unitary entries is denoted by 1n . The
operator diag() is the diagonal matrix of a vector. Finally,

R0 denotes the set of all the nonnegative real number and

R>0 denotes the set of all positive real numbers.
II. U RBAN D RAINAGE S YSTEMS
Urban Drainage Systems (UDS) are composed by an
arrangement of channels, which are connected by inspection
and collection chambers. The channel arrangement is regulated by using measuring equipment and control structures to
maintain specific operation conditions (e.g., velocity, Froude
number, filling ratio [5]). Normally, the most suitable UDS
topology is given by a tree, i.e., all channels in the UDS
converge to a common drain point. In order to evaluate the
performance of the system under unsteady flow conditions,
two different kind of techniques can be used regarding the
computation of flows along the channels: i) the hydraulic
modeling, and ii) the hydrological modeling.
The hydraulic modeling approach is mainly based on
physics. This model considers the effects of the inertial and
the gravitational forces over the flow. The one dimentional
Saint-Venant equations are used to model this approach
subject to constraints of mass balance and momentum conservation [10]. On the other hand, the hydrological approach
is based on the mass conservation principle, and it can be
considered as a simplification of the hydraulic model. Thus,
for the UDS modeling, the hydrological approach is useful
as a qualitative solution. As an advantage, the hydrological
model has a lower computational burden in comparison with
the computational burden to compute a hydraulic model. For
this reason, the hydrological techniques are mainly used for
controloriented models.
The purpose of the decentralized controller proposed in
this paper is to allocate efficiently the fluid along the UDS
channels. Then, a hydrological technique is appropriate to
characterize the behavior of the system under rain (disturbance) scenarios. In order to characterize the UDS dynamics,
we use the Muskingum linear model. Furthermore, passivity
of this hydrological model is studied in Section IV.
Muskingum Model
The Muskingum linear model for the ith reservoir is given
by the following differential equation and relation between
its inflows and outflows [10], i.e., vi = i (Ii li + Oi (1 li )),
and v i = Ii Oi , where vi is the volume of the
reservoir, i is a parameter for the model calibration, li
is the reservoir length, and Ii and Oi are the inflows
and outflows, respectively. Then, by expressing the outflows as function of vi , Ii , li , and i , it is obtained that
v i = (1 li )1 (Ii vi /i ), then
v i =qin,i Ki xi vi ,

(1)

where qin,i = Ii /(1li ) is the inflow, qout,i = Ki xi vi is the

outflow, and Ki xi = 1/(i i li ). Furthermore, xi [0, 1]
determines the control action over the output gate, which is
related to its percentage of opening, i.e,, zero is completely
closed, and one is completely opened. Ki > 0 scales the
outflow, and it can be seen as the discharge coefficient of
the reservoir.

III. C ONTROL V IA P OPULATION DYNAMICS

The proposed control consists in dividing the UDS into
m different subsystems with the same local topology.
Each local topology has different source reservoirs and one
receptor. Moreover, a population dynamicsbased controller
is designed for each subsystem. In fact, the population
dynamics approach has been already used in hydraulic systems [7]. However, the authors in [7] deal with a different
topology (i.e., one source reservoir and different receptor
reservoirs), and a different control goal without partitioning
of the system. The approach proposed in this paper considers
a decentralized control system for the UDS that is composed
by m local controllers that do not communicate with each
other and operate independently in parallel. This section
introduces the partitioning criterion of the UDS into sub
systems. Furthermore, the population dynamics used in the
control design are presented.
A. System Partitioning
Consider a simple UDS sub-system composed by n source
reservoirs, and only one receptor reservoir as shown in the
partition 1 of Figure 1. This topology is known as the
case of flow convergence1 . Let S = {1, ..., n} be the set
of source reservoirs in the case of flow convergence. The
maximum volume of the ith reservoir is denoted by vi . The
control objective consists in avoiding overflows throughout
the system, i.e., to avoid that the current volume of the
reservoir vi never overcomes its maximum volume capability
vi , for all i S. To achieve this objective, it is proposed to do
an allocation of the available capability of the n reservoirs,
i.e., to distribute the current available volume given by vi vi
in an optimal way by controlling the outflows qout,i , for all
i S.
Each subsystem must correspond to a case of flow
convergence. In order to clarify the partitioning process in
a typical tree topology of a UDS, an arbitrary tree UDS
is presented in Figure 1. At this general example, there
are two stages and there are three partitions. Notice that
the subsystem corresponding to each partition fits the case
of flow convergence. In this process, it is possible to find
a partitioning in which a reservoir is a source and also a
receptor for different partitions in the UDS. For instance, in
the partitioning presented in Figure 1, the gray reservoirs are
receptors for the partitions 1 and 2, and source reservoirs for
the partition 3.
B. Population Dynamics Approach
Consider a population composed by a large and finite
number of agents. Assume that the total amount of agents
represents the total outflow from all the source reservoirs
to the receptor reservoir, which is given by a complete
opening of the gates in the subsystem outflows. Suppose
that agents can select an outflow corresponding to a source
reservoir from the set of n reservoirs S. In the population
1 Due to the fact that there are several outflows converging to a unique
receptor reservoir.

Partition
Partition 11
v1

qout,1

vi

qout,i

Partition
Partition 2

Partition

vn

Population
Dynamics

qout,n
Fig. 2.

vr

Closed-loop control for a partition of the wastewater system.

dynamics given by,

x = diag(x) F 1n x> F ,
Partition
Partition33
Fig. 1.
Partitions over a tree topology. Each partition corresponds to
a flowconvergence topology, with n source reservoirs and one receptor
reservoir. Some reservoirs are source and receptor in different partitions
(gray reservoirs).

context, S represents the set of strategies. The scalar value

xi R0 represents the proportion of agents selecting
the reservoir i S, i.e., the opening proportion of the
ith outflow gate is xi . The population state x Rn0
represents all the proportions assigned to the reservoirs,
i.e., x = [x1 , . . . , xn ]> . Since x represents proportions,
then the set of possible population states is given by the
simplex = {x Rn0 : x> 1n = 1}. Agents in the
population have incentives to select the reservoir outflows
(e.g., in a general control system, the error is an incentive
for the controller to apply more energy to the system and
then correct the states to achieve the desired values). The
incentives associated to a rewarding that the proportion of
agents xi receives for selecting the reservoir i S, are given
by a fitness function fi (x), whose mapping is fi : 7 R.
Moreover, the vector of all the fitness functions is denoted
by F = [f1 fn ]> with mapping F : 7 Rn . The
following fitness function is proposed to control the case of
flow convergence topology:
!
1
xi ,
(2)
fi =
A vvii + 
where A determines the slope rate of the fitness function,
and  is a small factor that avoids an indetermination of fi
when vi = 0. Moreover, the proposed fitness function for
the strategy i S, only depends on the volume vi and the
proportion of agents xi , making it suitable for applying in
cases where only local information is available. The solution
of the population game is a Nash equilibrium denoted by
x in which no agent has incentives to switch from one
strategy to another [8]. Indeed, the solution is characterized
by the condition fi = fj , for all i, j S.
Each partition of the UDS is controlled by a population
dynamics system as presented in Figure 2. All the gates,
defining the outflow of the source reservoir in a partition,
are established by the vector x Rn . These opening grades
of the output gates affect the behavior of the reservoir
volumes, i.e., v = [v1 , . . . , vn ]> . Hence, the variation of the
reservoir volumes modifies the fitness function (2), affecting
the control actions over the output gates x Rn .
For the population dynamics design, it is proposed to
work with two kinds of population dynamics: the replicator

(3)

and the projection dynamics given by,

1
1 n F> 1 n .
(4)
n
These dynamics are of interest in this work since they
share gradient properties [9], and because of their passivity
properties, which are studied in Section IV.
x = F

IV. S TABILITY A NALYSIS

This section studies the stability properties of the closed
loop system controlled via replicator/projection dynamics.
A. Equilibrium Point
First, we characterize the equilibrium point.
Proposition 1: Let f R be a constant.
If x int,

n
>
where int = x R>0 : 1n x = 1 , then the equilibrium point (v , x ) of the closedloop system controlled via
population dynamics (i.e., the closedloop system given by
(1) and (3)/(4)) satisfies the following property:
fi (vi , xi ) = f , for all i = 1, . . . , n,

(5)

where vi and xi denotes the ith element of the vectors v

and x , respectively.
Proof: Consider the replicator dynamics (3). Since
x int by assumption, x = 0 if and only if
F = 1n x > F , where F denotes the vector of fitness
functions in steady state. Taking x > F = f , we conclude
that fi (vi , xi ) = f , for all i = 1, . . . , n.
Furthermore, consider the projection dynamics (4). x = 0
if and only if nF = 1n > F 1n . Taking 1n > F = nf , we
conclude that fi (vi , xi ) = f , for all i = 1, . . . , n.
It is worth noting that the fitness functions associated
with the reservoirs are equalized at equilibrium, i.e, x is
a Nash equilibrium of the underlying population game. The
fact that the fitness functions achieve a same value, means
that there is an equitable assignment of the source reservoir
outflows with respect to their current volumes, i.e., there is
an equitable distribution of the available volume capacities
in the reservoirs. This fact is the core of the proposed control
strategy designed to avoid overflows throughout the UDS.
B. Stability
Now, we use passivity theory to study the stability properties of the equilibrium point characterized before.
Notice that, the UDS dynamics can be written in error
coordinates as follows
e vi = ki evi (xi xi ) ki xi evi ki vi exi , i S,

(6)

where evi = vi vi , and (v1 , . . . , vn , x1 , . . . , xn ) is the

equilibrium point given in Proposition 1. Moreover, notice
that e vi = v i since v i = 0. An important property of this
system is provided in the following lemma.
Lemma 1: Assume that xi (t) 0, for all i = 1, . . . , n;
and vi (t) > 0, for all i = 1, . . . , n, and for all t 0. The
system given in Equation (6) is passive if we define its input
as xx = [x1 x1 , . . . , xn xn ]> and its output as gev =
[g1 (ev1 ), . . . , gn (evn )]> , where gi () is a zero in zero and
strictly increasing continuous function for all i = 1, . . . , n.
Furthermore, the corresponding storage function is positive
definite.
Proof: The proof uses the following storage function
(adapted from [4])
Z ev
n
X
i
1
gi ()d.
(7)
V1 (ev ) =

k
v
0
i=1 i i
V1 (ev ) is positive definite since: i) it is strictly convex
because the derivative of each term in (7) is strictly increasing by assumption; and ii) its minimum is at ev = 0, i.e.,
V |ev =0 = 0.
The derivative of V1 (ev ) along the trajectories of the system stated P
in Equation (6)
evi
n
is given by V 1 (ev )
=
then
i=1 ki vi e vi ,
Pn
Pn xi

V1 (ev ) = i=1 (xi xi )gi (evi ) i=1 v evi gi (evi ),

i
and
V 1 (ev )

gev > (x x )

n
X
xi
ev gi (evi ),
v i
i=1 i

Pn
where i=1 vxi evi gi (evi ) 0 since vi > 0 and xi 0 by
i
assumption. Thus, V 1 (ev ) gev > (x x ). Therefore, the
system is passive.
Additionally, the controllers based on population dynamics
described in (3) and (4) exhibit similar properties. These
properties are summarized in the following lemmas.
Lemma 2: The replicator dynamics system (3) is lossless
if we define its input as F 1n f = [f1 f , . . . , fn
f ]> and its output as x x = [x1 x1 , . . . , xn xn ]> .
Furthermore, the storage function is positive definite.
Proof: First, we notice that adding the same constant to
all fitness function does not affect the behavior of replicator
equation in the simplex . Therefore, the system in (3) can
be rewritten as follows


x i = xi fi f x> (F 1n f ) , i S.
(8)
Now, let us prove that this system is lossless from the input
F f to the output x x . In order to do this, we take the
following positive definite storage function
 
n
X
xi

V2 (x) =
xi ln
.
(9)
xi
i=1
Notice that V2 (x) = 0 for x = x , and V2 (x) > 0
for x 6= x . The derivative of V2 (x) along the trajectoPn x
ries of (8) is given by V 2 (x) = i=1 xii x i , then



Pn
V 2 (x) = i=1 xi fi f x> (F 1n f ) . This
Pn

leads to obtain V2 (x) =

i=1 (xi xi )(fi f ). Finally
V2 (x)

(F 1n f )> (x x ).

Thus, the system is lossless.

Lemma 3: The projection dynamics system (4) is lossless
if we define its input as F 1n f = [f1 f , . . . , fn
f ]> and its output as x x = [x1 x1 , . . . , xn xn ]> .
Furthermore, the storage function is positive definite.
Proof: First, we notice that adding the same constant to
all fitness function does not affect the behavior of projection
dynamics. Therefore, the system in (4) can be rewritten as
follows
1
x i = fi f 1n (F 1n f )> 1n ,
(10)
n
for all i = 1, . . . , n. Now, let us prove that this system is
lossless from the input F f to the output x x . In order
to do this, we take the following positive definite storage
function [8]
n
1X
(xi xi )2 .
(11)
V3 (x) =
2 i=1
Notice that V3 (x) = 0 for x = x , and V3 (x) > 0 for
x 6= x . The derivative of
3 (x) along the trajectories of
PV
n
(10) is given by V 3 (x) = i=1 (xi xi )x i , it follows

n
X
1
(xi xi ) + (x),
V 3 (x) = (F 1n f )> 1n
n
i=1
Pn

where (x) = i=1 (xi xi )(fi f ), then

!
n
n
X
X
1

>

V 3 (x) = (F 1n f ) 1n
xi
xi + (x).
n
i=1
i=1
Pn
Pn

Due to the
fact that
=
i=1 xi
i=1 xi , then
P
n

V3 (x) =
i=1 (xi xi )(fi f ). Finally

V 3 (x) =(F 1n f )> (x x ).

Thus, the system is lossless.

The following results regarding the stability properties of
the closedloop system are based on passivity of the UDS
and controllers.
Theorem 1: The equilibrium point (v , x ), characterized
in Proposition 1, of the closedloop system controlled via
replicator dynamics is stable.
Proof: We use V = V1 + V2 as a Lyapunov function
candidate, where V1 and V2 are given in (7), (9), respectively.
According to Lemma 1 and Lemma 2, we know that the
derivative of V along the trajectories of the closedloop
system is given by V gev > (x x ) + (F 1n f )> (x
x ). Taking gev = F 1n f (notice that ith element of the
vector F 1n f satisfies the conditions in Lemma 1, i.e.,
fi f is continuously increasing and zero in vi = vi ), we
obtain V 0. Therefore, the system is stable.
Theorem 2: The equilibrium point (v , x ), characterized
in Proposition 1, of the closedloop system controlled via
projection dynamics is stable.
Proof: We proceed as in the proof of Theorem 1.

d10
d10

d2
vd23
d3
v3
v3
d4d3
d4

v4v3
v4

x2
x2
c1

x
x3 2
x3

c2
c2
x3
x4
x
c24

x10
x5 x10v
x5
10

v5
v5
dd65
d6
x6 x10
x5 x6
v6
v
v56
d7 c3c3
d67
x6
v7v
67
x7
x7
d8
dd8 c3
7
v8v
78

d8

Q (m3/s)

d11
Time [hours]
c4Theoremx1:
d15
a) ), characterized The direct run-off hydrographs that represent theb)system
equilibrium
(v ,x
x1:
cd4Theorem11
d15 point
11The
The
equilibrium
point
(v
,
x
),
characterized
The
direct
run-off
hydrographs
that
represent
the system
12 v11
inFig.
Proposition
1, ofsystem
the closedloop
system
controlled
via system.
disturbances
presented
in Figure
4. TheThe
maximum
run-off
3.12 a) Equivalent
part x
of15
Bogot
a, Chic
o urban drainage
b) Direct are
runoff
hydrograph
disturbances
di = 0
xthe
15
in d
Proposition
1,
of
theofclosedloop
system
controlled
via disturbances
are presented
in- Disturbances.
Figure 4. The
maximum
run-off
d
v
16
x
15
for
i
=
{1,
...,
16}
in
the
interval
of
time
1.5[hours]

5.25[hours].
12
replicator dynamics
is vstable.
peak is approximately 18 L/s.
d
16
x
15
12
replicator dynamics
is stable.
peak is approximately 18 L/s.
v12
We use V = V1 + V2 as a Lyapunov function
v12Proof:
Proof:x11We use Vd15
= V1 + V2 as a Lyapunov
function
c4Theorem
respectively.
d1
candidate,
where
V1 and
(7),
(9),
1: The
equilibrium
(v
), characterized The direct run-off hydrographs that represent the system
2 are given
v,16x
cpoint
dd
6 inin
dV14
9
c
d2 d1
candidate,
where
V1 and
(7),
respectively.
5
0.016
2 are given
v16(9),
d12
6 2,system
xc15
dVclosedloop
9
xthat
1
c5 Lemma
d3 d2
14 Lemma
According
to
1 and
we
know
the
in Proposition
1, of
the
controlled
via
disturbances
are presented in Figure 4. The maximum run-off
16 =
0.016
x
=
1
d4 d3
Accordingx to Lemma
1
and
Lemma
2,
we
know
that
the
16
0.014
d16closedloop
v15
d5 d4
12V along
derivative
of
the
trajectories
of
the
replicator
dynamics
is
stable.
peak
is
approximately
18
L/s.
0.014
vvderivative
v
9
14
d6 d5
of
V
along
the
trajectories
of
the
closedloop
>

>
0.012

v
v
12
xWe
9 Proof:
system
is given
V V14=
geV
(x
1n f )function
(x
d7 d6
+
V2xas)+
a (F
Lyapunov
x
9 by use

>
v1 >
14
0.012
x9 by V g ev (x
is given
x ) +th(F 1n f ) (x
d8 d7
x
system
14
0.01
d
xcandidate,
).13Taking
gev =VF1 and
1nVf2 are
i th
element
of the
d8
given
in (7),
(9),
respectively.
d9 d1
(notice
c6 that
0.01
9Taking
cwhere
d2
x d
).13
114
thatvi16
element of the
d9
d10
5 gev = F d
0.016
n f (notice
0.008
x1,
=the1
Linear
d3
vector
F 1to
the conditions
According
1 and
LemmaEquivalent
2, inweLemma
know
that
d10
d11
16i.e.,
n fLemma
satisfies
0.008
Equivalent
Linear
d4
vector
F
1
f
satisfies
the
conditions
in
Lemma
1,
i.e.,
Reservoir
d11
d12
n

0.014
fvi13
f
is
continuously
increasing
and
zero
in
v
=
v
),
we
d5
0.006
derivative
of
V
along
the
trajectories
of
the
closedloop
Reservoir
d12
d13
i

iv ), we
v
fvvi13
increasing
and
zero
in
v
=
9 f xis13continuously
0.006
14
d6
Raingauge
d13
d14
i
i
>
stable.
>
0.012

obtain
V is

0.
Therefore,
the
system
is
x
system
given
by
V

g
(x
x
)
+
(F
1
f
)
(x
d7
x
0.004
Raingauge
d15d14
ev x14
n
130. 9Therefore, the
obtain
V
system
is
stable.
d8
0.004
d15
d16

th
d
point (v
Limnimeter
0.01
Theorem
2:
The
equilibrium
,
x
),
characterized
x ).13Taking gev = F 1n f (notice that
d9
d16
0.002
i element of the
Limnimeter
Theorem 2: The
equilibriumcCipoint
(v
, xController
), characterized
d10
0.002
Local
i
0.008
Equivalent
Linear
invector
Proposition
the closedloop
system
controlled
via
F 1n1,f ofsatisfies
the conditions
in
Lemma
1,
i.e.,
d11
c
0
Ci
Local
i
in Proposition
1, of the closedloop
systemController
controlled
via
00
0.5
1
1.5
Reservoir
d12

projection
is stable.
fvi13 f isdynamics
continuously
increasing and
zero in Gate
v = v ), we
Retention
0.006 0
0.5
1
d131.5
projection
is stable.
Retention iGatei
x dynamics
d14
Time [hours]
Proof:
We
proceed asthe
in the
proof
of
Theorem 1.
obtain
V 13
0. Therefore,
system
isRaingauge
stable.

d10

x8
x8 x7

Direct Runoff Hydrograph Disturbances

Direct Runoff Hydrograph Disturbances

Direct Runoff Hydrograph Disturbances

Q (m3/s)

v2v1
v2

c1
c1
x1

0.002

Fig. 3. v
a)11
system of part of the Bogota, Chico urban drainage system. b) Direct runoff hydrograph - Disturbances. The disturbances di = 0
vEquivalent
11
of 1.5[hours]
the Bogota
, Chic
urban drainage system. b) Direct runoff hydrograph - 0Disturbances.
The disturbances
di = 0
0
0.5
forFig.
i =3.{1,a)
...,Equivalent
16} in thesystem
intervalofofpart
time
t o5.25[hours].
for i = {1, ..., 16} in the interval of time 1.5[hours] t 5.25[hours].
Time (hr)

d5
d5

Time (hr)

Time (hr)

0.004
d15
Time [hours]
Proof: We proceed as in the proof
d16
of Theorem 1.
v8
Limnimeter
4. 0.002
Direct runoff hydrograph - b)
Disturbances. The disturbances di =
0
d4
Theorem
point (v
, x ), characterized Fig.
a)a) 2: The equilibrium
Fig.
4.
Direct
runoff
hydrograph
Disturbances.
The
disturbances
d
x8
..., 16} in the interval of time
C ASE
S TUDY
b) 1.5[hours] t 5.25[hours].i = 0
cCii Local
in Proposition 1, ofV.the
closedloop
systemController
controlled via forfori =i ={1,{1,
V. C
ASE S TUDY
0 ..., 16} in the interval of time 1.5[hours] t 5.25[hours].
0
0.5
Fig. 3.v4 a) Equivalent system of part of the Bogot
a, Chic
oshown
urbanindrainage
system.
b)proposed
DirectGate
runoff hydrograph
- Disturbances.
Thevolumes
disturbances
=K
0 i con-1.5
Additionally,
the maximum
vi , 1 anddithe
It has
Section
that the
projection
dynamics
stable. IV
Retention
Fig. 3. a) Equivalent
of part of the Bogot
a,been
Chic
oshown
urbanis
system.
b)proposed
Direct closed
runoff
- Disturbances.
Thevolumes
disturbances
=K
0 i conAdditionally,
the maximum
v , anddithe
It has
indrainage
Section IV
that the
closed hydrograph
x4 in thesystem
[hours]
for i = {1, ..., 16}
interval of time 1.5[hours]
been
t
5.25[hours].
the UDS
are ipresented
in the
loop control
system
for each
then,1. we stants for each reservoir inTime
Proof:
We
proceed
as inpartition
the proofis ofstable,
Theorem
for i = {1, ..., 16} in the interval of time 1.5[hours]
t
5.25[hours].
loop control
system
for each partition is stable, then, we stants for each reservoir in the UDS are presented in the
Fig.
4.
Direct
runoff
hydrograph
Disturbances.
The
disturbances
di = 0
Table
I.
The
K
constants
have
been
found
by
using
EPAapply the
controller
for
a
case
study
to
illustrate
the
usefuli
Direct
Runoff
Hydrograph

Disturbances
b)
Table
I.
The
K
constants
have
been
found
by
using
EPAapplya)the controller for
a
case
study
to
illustrate
the
usefuli
Direct
interval
Disturbances of time 1.5[hours] t 5.25[hours].
for i = {1,
...,Runoff
16}Hydrograph
C ASE S TUDY
which
isin the
a computer
program specialized
in hyness of the model-freeV.decentralized
population dynamics SWMM,
d1
of the model-free decentralized population dynamics
SWMM,
which
is
a
computer
program
specialized
in
hyness
Theorem
1:
The
equilibrium
point
(v
,
x
),
characterized
d1
has

draulic
modelling.
The
program
has
been
used
to
retrieve
Fig.
3.
a)
Equivalent
system
of
part
of
the
Bogot
a
,
Chic
o
urban
drainage
system.
b)
Direct
runoff
hydrograph
Disturbances.
The
disturbances
d
=
0
based
controller
for
UDS.
The
proposed
case
study
consists
Additionally,
the
maximum
volumes
v

,
and
the
K
Fig.
3.
a)
Equivalent
system
of
part
of
the
Bogot
a
,
Chic
o
urban
drainage
system.
b)
Direct
runoff
hydrograph
Disturbances.
The
disturbances
d
=
0
It
been
shown
in
Section
IV
that
the
proposed
closed
d2i
i
i i con0.016 study consists
Theorem 1: The equilibrium point based
(v ,controller
x ), characterized
draulic modelling. The program has been
used
for UDS. The proposed case
d2 to retrieve
d3
0.016
andfor
volume
information,
given
the are
geometric
features
i=
{1,{1,
...,...,
16}
in in
thethe
interval
of of
time
1.5[hours]

5.25[hours].
of
a part
of controlled
the
afor
Colombia
stormwater
UDSthen,
shown
stants
each reservoir
in the
UDS
presented
in the
loop
control
system
each
partition
is stable,
we flow
for
i=
16}
interval
time
1.5[hours]
t Bogot
5.25[hours].
infor
system
via
d4d3th features
of
a part of
the BogotaColombia
stormwater
UDS shown flow and volume information, given the geometric
inProposition
Proposition1,1,ofofthe
theclosedloop
closedloop
system
controlled
viastudyby
0.014
the network.
qout,ihave
= Kbeen
iusing
is
inapply
Figure
The
UDSforis acomposed
subcatchments
I. Direct
The
KSince,
foundthe
by
EPAthe3.controller
case
to 16
illustrate
the useful- ofTable
i vi when
i constants
d5d4
thgate
Runoff Hydrograph
Disturbances
Since,
qout,i = K
i
gate
is
in Figure 3. The UDS is composed by 160.014
subcatchments of the network.
replicator
dynamics
is
stable.
i vi when the
d5
d6 linearity.
TABLE
opened,
Ki can
easily calculated
that
to 16
collectiondecentralized
chambers. Groups
of catchments
SWMM,
which
is abeIcomputer
programassuming
specialized
in hynessdrain
of the
model-free
population
dynamics fully
replicator dynamics is stable.
0.012
d6 linearity.
d1
fully
opened,
K
can
be
easily
calculated
assuming
drain to 16 collection chambers. Groups
that
of
catchments
d7
i
V.
C
ASE
S
TUDY
0.012 the system
Theorem
1:
The
equilibrium
point
(v
,
x
),
characterized
Proof:
We
use
V
=
V
+
V
as
a
Lyapunov
are
approximated
tofunction
reservoirs
as in [6],case
and
It
is
assumed
that
there
is
full
information
within
a
partidraulic
modelling.
The
program
has
been
used
to
retrieve
based
controller
for
UDS.
The proposed
study
consists
d7
d2
1
2
d8
0.016
Proof: We use V = V1 + V2 dynamics
as
Lyapunov
areaapproximated
tofunction
reservoirs as in [6],
and the Coeffi.
system
It
is assumed
there is full
informationmax.
within
a parti0.01
d8 vol.
Res.
max.
vol. that
Res.
Coeffi.
d9d3
are
modeled
with
the Muskingum
linear
reservoir
about
volumes,
i.e.,
eachthe
local
controller
has
flow
and reservoir
volume
information,
given
geometric
features
a part(9),
of controlled
the Bogot
aColombia
stormwater
shown tion
0.01 UDS
in
Proposition
1,VV
theVV
closedloop
system
candidate,
where
and
are given inof
(7),
d9 th3
1of
dynamics
arerespectively.
modeled
withvia
the Muskingum
linear reservoir tion about 3reservoir volumes, i.e., each local d10
controller
has
d4
0.014
candidate,
where
in
(7),
(9),
respectively.
It has been
shown
in2Section
IVmodel
that
the
proposed
closed
1 and
2 are given
presented
Section
II. Each by
equivalent
reservoir
full
about
theqout,i
source
tanks
the
of information
the
Since,
=K
Kiivi within
when the
ipartition.
gate
is
K
v

[m
]
v

]
in Figure
3. TheinUDS
is composed
16
subcatchments
i
i network.
i [m
d10
0.008
d5
d11
model
presented
inthat
Section
II. Each equivalent
reservoir full information about the source tanks within d6
the partition.
replicator
dynamics
is
stable.
According
to
Lemma
1
and
Lemma
2,
we
know
the
0.008
d11
represents
atosource
orthat
receptor
reservoir
of0.012
partition
of The
reservoir
in the system
is checked
fullycapacity
opened,of
K each
can 9
be
easily
calculated
assuming
linearity.
that2,drain
16 collection
chambers.
Groups
catchments
d12
According
to system
Lemma for
1 and
theaploop control
eachLemma
partition
iswe
stable,
then,
we
1a of
0.002332
0.006147
274.80
represents
aknow
source
or receptor
reservoir0.006
of
a partition
of The 114.50
capacity ofi each
reservoir
in the system
is checked
d12
d7
d13
Proof:
use
V the
= Vtrajectories
+ V2 the
as
aof
Lyapunov
as closedloop
introduced
in Section
The
is during
run-offthat
routing
for a 5.25within
hoursa time
derivative
ofofWe
VV along
the
aresystem
approximated
tofunction
reservoirs
as inIII-A.
[6], 0.006
and system
the system
It isthe
assumed
there process,
is full information
partid13
d8
d14
theillustrate
system
as closedloop
introduced
in Section III-A.
The system
is during
the run-off routing
process,
for a 5.25
hours time
derivative
along
of
the
20.01
0.003870
137.40
10
0.005446
183.20
ply the controller
athe
case1trajectories
to
the
usefulness
>study
dynamics
rain
>with
d14 volume
modeled
for(9),
a typical
scenario,
in which0.004
flooding
occurs. window.
Thereservoir
capacity volumes,
is definedi.e.,
as each
the normalized
d9
are
modeled
the Muskingum
linear reservoir
tion about
local d15
controller
has
1 for
candidate,
where
V
and
V
are
given
in
(7),
respectively.
system
is
given
by
V

g
(x
x
)
(F
1
f
)
(x
>
+
rain
>
modeled
for
a
typical
scenario,
in
which
flooding
occurs.
window.
The
capacity
is
defined
as
the
normalized
volume
2
e
n

0.004
v
d15
d10
0.003170
114.50 about
11the source
0.020703
183.20
3
system
given by Vdecentralized
gev (x population
xmodel
) +th
(F
1n in
f Section
) (x II. Each equivalent
of theismodel-free
dynamicsbased
presented
reservoir full information
tanks withind16
the
partition.
0.008

d16
d11
0.002
Lemma
2, iwe
that
the
xAccording
==FF 111
f (notice
the
Taking g
thelement
4 a partition
0.008239
12reservoir
0.001693
274.80
evLemma
nand
represents
aknow
source
orofof
receptor
of
of The 125.95
capacity of each
in the system
is checked
0.002
d12
x).
). Takingtofor
g
(noticethat
that
element
the
ev UDS.
n fproposed
controller
The
case istudy
consists
of areservoir 0.006
d13
50The
0.002217
274.80
13
0.007026
274.80
the system
as closedloop
introduced
in Section III-A.
system
is during
the run-off routing
process,
for a 5.25
hours time
derivative
of
V
along
the
trajectories
of
the
vector
F
1
f
the
conditions
in
Lemma
1,
i.e.,
satisfies
n
00
0.5
1
1.5
d14
vector
F the1nBogot
f satisfies
the
conditions
in
Lemma
1,
i.e.,
part of
a
(Colombia)
stormwater
UDS
shown
in
(hr)
0
0.5 The Time
10.000632

for av typical
scenario,
in which0.004
occurs. window.
capacity
as the normalized
0.008975
274.80
14is defined
807.22
6flooding
d151.5 volume
Time (hr)
is continuously
given by V increasing
g > (x and
xmodeled
) + (F
1=fvrain
)>
fsystem
f is
zero
in
), (x
we
Q (m3/s)

Q (m3/s)

Q (m3/s)

[m3 3 /s]
Q [m3 /s]QQ[m
/s]

Time (hr)

ev
if
d16
f is
continuously
and
zero
in ivin= ivi ), we
Time
iFigure
3a).
The UDS isincreasing
composed
by 16
subcatchments
that
7
0.005185
274.80
15
0.006319
274.80

th
Time[hours]
[hours]

0.002
x
).
Taking
g
=
F
1
f
(notice
that
i
element of the
obtain
V
0.
Therefore,
the
system
is
stable.
e
n
v
obtain
Therefore,
the
system
is
stable.
8
0.004764
274.80
16
0.005782
274.80
drain Vto 160.collection
chambers.
The
groups
of
catchments
4. Direct runoff hydrograph - Disturbances. The disturbances di = 0

Lemma 1, i.e., Fig.

0
vector
F 2:1
fThe
satisfies
the point
conditions
in
Fig. 4. Direct
runoff hydrograph
- Disturbances.
The
disturbances
0
Theorem
equilibrium
(v ,x
characterized
Res.=Reservoir,
max.
maximum
volume di =
0..., 16}
0.5 of
1 t 5.25[hours].
1.5
in the interval
timevol.=
1.5[hours]
Theorem
2:nThe
equilibrium
point
x),
),and
characterized
are
approximated
to reservoirs
as (v
in ,[5],
the systemforfori i=={1,

{1, ..., 16} in the interval of time 1.5[hours] t 5.25[hours].

f
is
continuously
increasing
and
zero
in
v
=
v
),
we
infiniProposition
1,
of
the
closedloop
system
controlled
via
i
i
Time [hours]
Proposition
of the with
closedloop
system controlled
via
dynamics
are 1,
modeled
the Muskingum
linear reservoir
obtain
V
0. Therefore,
the system is stable.
projection
dynamics
isisstable.
projection
dynamics
stable.
model presented
in Section
II. Each equivalent
reservoir
Fig.
4.
Direct
runoff
hydrograph
- Disturbances.
The advantage
disturbances dof
theI.UDS.
To do
so, it is have
proposed
tofound
take
all0 the

i =
Table
The
K
constants
been
by
using
EPATheorem
2:WeThe
equilibrium
point
(v of,ofxTheorem
), characterized
Proof:
proceed
as ininthe
proof
1.1.
i in
i = {1,
..., 16}
the interval have
of timebeen
1.5[hours]
tby
5.25[hours].
Table
I.
The
K
constants
found
using
EPAProof: We
proceed
the
proof
Theorem
i
represents
a source
or as
receptor
reservoir
of a partition
of for
available
volume
capacities program
in the whole
network.
each
which
is a acomputer
specialized
ininAt
hyin Proposition 1, of the closedloop system controlled via SWMM,
SWMM,
which
program
specialized
hythe system as it isisstated
Section III-A. The system isdraulic
partition,
thereisis
a computer
controller
managing
the outflows
of the
C
ASEin
S
TUDY
modelling.
The
program
has
been
used
to
retrieve
projection dynamics V.
stable.
C ASE
S TUDY
draulic
modelling.
The
has been
used to III.
retrieve
modeled for a typicalV.rain
scenario,
in which flooding occurs.flow
source
reservoirs
as itprogram
was
described
in Section
Table
I. volume
The
Ki information,
constants
have
beenthe
found
by using
EPAgiven
geometric
features
Proof:
We proceed
as in IV
thethat
proof
of
Theorem
1.
ItThe
has
been
shown
in
Section
the
proposed
closed
flowand
and
volume
information,
given
the
geometric
features
that
represent
systemofSWMM,
It hasdirect
been run-off
shown inhydrographs
Section IV that
the
proposedthe
closed
th
which
is
a
computer
program
specialized
in
the
network.
Since,
q
=
K
v
when
the
i
gate
isis
out,i
i iv when the ith gatehyloop
control
for
isisstable,
then,
of the network. VI.
Since,
qout,i =AND
K
R ESULTS
i DISCUSSION
i
disturbances
are presented
inpartition
Figure
The
maximum
run-fully
loop
controlsystem
system
foreach
partition
stable,
then,we
we
V.
Ceach
ASE
S
TUDY3b).
draulic
modelling.
The
program
has
been
used
to
retrieve
opened,
K
can
be
easily
calculated
assuming
linearity.
i can be easily calculated assuming linearity.
apply
for
fully opened, K
i
off the
peak
is approximately
18study
L/s.totoillustrate
apply
thecontroller
controller
fora acase
case
study
illustratethe
theusefuluseful- flow
The
system
isinformation,
tested isforfull
a given
typicalthe
raingeometric
scenario
Figure
and
volume
features
ItItisis
assumed
that
within
a(see
It
has
been
shown
in
Section
IV
that
the
proposed
closed
ness
of
the
model-free
decentralized
population
dynamics
assumed
thatthere
there
is fullinformation
information
within
apartipartithcomparison
nessAdditionally,
of the model-free
decentralized
population
the maximum
volumes
vi , and dynamics
the Ki con- of 3b)).
The
results
are
analyzed
by
making
a
the
network.
Since,
q
=
K
v
when
the
i
gate
out,i i.e., each
i i local controller hasis
about
volumes,
loop
control
system
for The
each
partition
iscase
stable,
we I.tion
based
controller
for
UDS.
proposed
case
study
consists
tion
aboutreservoir
reservoir
volumes,
i.e., each
local
controller
has
based
controller
for
UDS.
The
proposed
studythen,
consists
stants
for each
reservoir
in
the
UDS are
presented
in
Table
between
the
case
without
control
(no
active
elements),
fully
opened,
K
can
be
easily
calculated
assuming
linearity.and
i
full information about
the
source
tanks
within
the
partition.
theofof
controller
for
case study
to illustrate
the shown
usefulofapply
a apart
the
aColombia
stormwater
about
the
source
tanks
within
the
partition.
ofThe
part
theBogot
Bogot
aaColombia
stormwater
UDS
shown fulltheinformation
Ki constants
have
been found
by usingUDS
EPA-SWMM,
case
applying
the
decentralized
population
Itcapacity
iscapacity
assumed
there
is
fullininformation
within
adynamicspartiofofthat
each
reservoir
isischecked
of the
model-free
decentralized
population
dynamics The
inness
Figure
3. The
byby
1616subcatchments
The
each
reservoir
inthe
thesystem
system
checked
inwhich
Figure
TheUDS
UDSisprogram
iscomposed
composed
subcatchments
is 3.a computer
specialized
in
hydraulic mod- tion
based
control.
The
results
show
the
total
flood
volume
about
reservoir
volumes,
i.e.,
each
local
controller
hasand
during
the
run-off
routing
process,
for
a
5.25
hours
time
based
controller
for
UDS.
The
proposed
case
study
consists
that
drain
to
16
collection
chambers.
Groups
of
catchments
during
the run-off
routing
process,
for
anetwork.
5.25 hours
time
that
drainThe
to 16
collection
Groups
of catchments
elling.
program
haschambers.
been used
to retrieve
flow andwindow.
the
capacity
of
each
reservoir
in
the
The
biggest
information
aboutisthe
sourceasas
tanks
within
the partition.
The
the
volume
of
aapproximated
part of the Bogot
aColombia
stormwater
shown full
are
totoreservoirs
asasin
the
system
window.
Thecapacity
capacity
isdefined
defined
thenormalized
normalized
volume
3
are
approximated
reservoirs
in[6],
[6],and
andUDS
theof
system
volume
information,
given the geometric
features
the net-ofThe
th
andcapacity
smallest
reservoirs
have ain
capacity
of 807.22m
and
of
each
reservoir
the
system
is
checked
the
i
i.e.
v
=
v
/
v
where
v

isisthe
threservoir
normalized
i
i
i
in
Figure
3.
The
UDS
is
composed
by
16
subcatchments
dynamics
are
modeled
with
the
Muskingum
linear
reservoir
th
of 114.5m
the i 3 ,reservoir
i.e.
v
=
v
/
v
where
v

the
normalized
i
i
i
dynamics
are modeled
Muskingum
linear
reservoir
work. Since,
qout,i = with
Ki vi the
when
the i gate
is fully
opened,maximum
respectively
(see
Table
I).
Table
II
shows
the
during
the
run-off
routing
process,
for
a
5.25
hours
time
volume
of
the
reservoir.
Then,
a
flooding
event
that
drain
to 16 collection
chambers.
of catchments
model
presented
in Section
II.II. Each
equivalent
reservoir
volumeforofthe
thecase
reservoir.
Then,
athere
flooding
event
model
presented
Section
EachGroups
equivalent
reservoir it maximum
Ki can
be easilyincalculated
assuming
linearity. Moreover,
total
overflow
study.
When
is
no
control,
window.
The
capacity
is
defined
as
the
normalized
volume
occurs
when
the
volume
in
a
reservoir
exceeds
its
maximum
are
approximated
toororreservoirs
as
in [6],ofofand
the system
represents
a asource
receptor
reservoir
a apartition
ofof occurs when the volume in a reservoir exceeds its maximum
represents
source
receptor
reservoir
partition
is assumed
that there
is full information
within
a partition
th
the
is 1165.28
and zero
proposed
of
the overflow
ii.e.
i.e. vnormalized
vi when
where the
vi is
the
when
>liters
dynamics
are
modeled
with
the
Muskingum
linear
reservoir
the
system
as
introduced
in
Section
III-A.
The
system
is volume,
volume,
i.e.reservoir
whenvnormalized
vnormalized
>1.1. = vi /
the
system
as introduced
Section
III-A.controller
The system
about
reservoir
volumes, ini.e.,
each local
has isfull maximum
decentralized
control
strategy
is
applied.
It
can
be
seen
volume
of
the
reservoir.
Then,
a
flooding
event
model
presented
in rain
Section
II. Each
equivalent
reservoir
modeled
for
scenario,
ininwhich
flooding
occurs.
The
objective
isistotoavoid
the
overflows
throughout
modeled
fora atypical
typical
rainsource
scenario,
which
flooding
occurs.
Thecontrol
control
objective
avoid
the
overflows
throughout
information
about
the
tanks
within
the partition.
The occurs
that
the
control
strategy
is
very
useful
in
the
reduction
of
when
the
volume
in
a
reservoir
exceeds
its
maximum
represents
a sourcehydrographs
or receptor that
reservoir
of a the
partition
of the
The
direct
represent
UDS.
tototake
The
directrun-off
hydrographs
thatsystem
represent
thesystem
system
theoverflows
UDS.To
Todo
doso,
so,ititisfor
isproposed
proposed
takeadvantage
advantageofofallall
capacity
ofrun-off
each reservoir
in the
is checked
during volume,
objective
the
case
study.
i.e.
when
v
>
1.
the
system are
as presented
introduced
inFigure
Section
III-A.
The system
disturbances
inin
4. The
maximum
run-offis the
available volumenormalized
capacities in the whole network. At
disturbances
presented
Figure
The
maximum
the run-off are
routing
process,
for a 4.5.25
hours
time run-off
window. the available volume capacities in the whole network. At
modeled
for
a
typical
rain
scenario,
in
which
flooding
occurs.
peak
is
approximately
18
L/s.
The
control
objective
to TABLE
avoid the
overflows
each
partition,
there
is
a iscontroller
managing
thethroughout
outflows
th
peak
approximately
18 as
L/s.
IImanaging the outflows
The iscapacity
is defined
the normalized volume of the i each partition, there is a controller
The
direct
run-off
hydrographs
that
represent
the
system
the
UDS.
To
do
so,
it
is
proposed
to
take
advantage
of all
Additionally,
the
maximum
volumes
v

,
and
the
K
conof
the
source
reservoirs
and
distributing
the
current
available
i
i
Control
case
Overflows
[L]
Additionally,
maximum
volumes
of the source reservoirs and distributing the current available
i conreservoir
i.e. vthe
= vi /
vi wherevi ,viand
is the
the K
maximum
normalized
disturbances
are
presented
in
Figure
4.
The
maximum
run-off
the
available
volume
capacities
in
the
whole
network.
At
stants
for
each
reservoir
in
the
UDS
are
presented
in
the
capacity
of
the
source
reservoirs.
Without
Control
1165.28
stants
for ofeach
reservoir in
the aUDS
are presented
in when
the capacity of the source reservoirs.
volume
the reservoir.
Then,
flooding
event occurs
Decentralized
Dynamics
peak is approximately 18 L/s.
each partition,
there with
is aPopulation
controller
managing the0outflows
the volume in a reservoir exceeds its maximum volume, i.e.
Additionally, the maximum volumes vi , and the Ki con- of the source reservoirs and distributing the current available
when v
> 1.
Additionally,
the reservoirs.
evolution of the UDS capacities is prestants fornormalized
each reservoir in the UDS are presented in the capacity
of the source
The control objective is to avoid the overflows throughout sented in Figure 4. It is shown that stormwater is better
Time (hr)

Time (hr)

Q (m3/s)

d2d1
d2

x1
x1

a)
a)

1
Time (hr)

Time [hours]
Time [hours]
b)
b)

Q (m3/s)

v1
v1

v10
v10

0.5
0.5

0.004

Q [m3 /s]
Q [m3 /s] Q [m3 /s]

d1
d1

0.01
0.002
0.002
0.008
0
00
0.006 0

Retention Gate

x4

d11
d11

Local Controller

Q [m3 /

cCii
v4

vnormalized (vi /
vi )
vnormalized (vi /
vi )

0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0
0
0

1
2
1
2

3
43
4

0.6
0.6
0.4
0.4
0.2
0.2

1
1

2
2

3
3

time [hours]
a)

4
4

5
5
9
12
9
13
12
13

0.8
0.8
0.6

0.6
0.4

0
0
0
0
0.2
0.2
0.15

0.4
0.2
0.2
0
0
0
0

0.05
0
0
0
0

time [hours]
d)

3
time2 [hours]
b)

4
4

5
5
10
11
10
11

0.15
0.1
0.1
0.05

1
1

1
1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0
0
0

5
65
76
87
8

1
1

2
3
time
[hours]
c)

4
4

5
5
14
15
14
16
15
16

1
0.81

0.8
0.6
0.6
0.4
0.4
0.2
1

time [hours]
e)

0.2
0
0
0
0

time [hours]
f)

Fig. of
6. the
Evolution
of the 16 normalized
gray correspond
lines correspond
scenario without
control,
and and
blackblack
lines correspond
to scenario
Fig. 4. Evolution
16 normalized
reservoir reservoir
volumes.volumes.
gray lines
to ato scenario
without
control,
lines correspond
to with
a scenario
the decentralized
controller.
e), f)
andshow
f) show
reservoirsfor
forthe
the controllers
controllers 1,1,2,2,
3,3,
4, 4,
5, 5,
andand
6, respectively.
LabelsLabels
refer torefer
the to the
with the decentralized
controller.
FiguresFigures
a), b), a),
c),b),
d),c),e),d),and
the the
reservoirs
6, respectively.
different reservoirs in the UDS.
different reservoirs in the UDS.

by adding
a O.
multi-objective
controller
based on
population
distributed throughout
network for
decentralized
VII. the
C ONCLUSIONS
AND the
FURTHER
WORK
[6] L. Nie,
Lindholm, G. Lindholm,
and E. Syversen.
Impacts
of
climate change on urban drainage systemsa case study in fredrikstad,
control case. The upstream reservoirs get more filled and dynamics.
norway. Urban Water Journal, 6(4):323332, 2009.
It has been shown
that to
thealmost
proposedthe
decentralized
populatheir capacity utilization
increase
double. The
[7] G. Obando, A. Pantoja, and N. Quijano. Building Temperature Control
tion dynamicsbased control is efficient in terms of a better
R EFERENCES
based on Population Dynamics.
IEEE Transactions on Control Systems
flow downstream
is relieved
and flooding
The the
distribution
of wastewater
throughoutistheavoided.
UDS, avoiding
Technology., 22(1):404412, 2014.
[1] [8]
K. Arnbjerg-Nielsen,
Willems,
J. Olsson,
S.wastewater
Beecham,systems.
A. Pathirana,
C. Ocampo-Martinez.P.Model
predictive
control of
use of the proposed
and the casoverflows. decentralized
The partitioning controllers
proposed methodology
allows to
I.B.Springer,
Gregersen,
2010. H. Madsen, V. Nguyen, et al. Impacts of climate
cade topologydesign
guarantees
that the controller
capacity by
of using
the reservoirs
the decentralized
different local
[9]
C. Ocampo-Martinez,
S. Bovo,and
andurban
V. Puig.
Partitioning
approach
change
on rainfall extremes
drainage
systems:
a review.
with a lower
computational
burdenwhen
with respect
oriented
to the&decentralised
predictive
control of
large-scale systems.
upstream is controllers
more effectively
used.
Furthermore,
the to
Water
Science
Technology,
68(1):1628,
2013.
Journal ofD.
Process
Control,A.
21(5):775786,
centralized controller.
it has the
beensteady
shown that [2] R. Ashley,
Balmforth,
Saul, and J.2011.
Blanskby. Flooding in the
system is notacontrolled,
the time Furthermore,
it takes to reach
[10] E. Ramirez-Llanos and N. Quijano. A population dynamics approach
future
predicting
climate
change,
risks and responses
urban areas.
the proposed closedloop system is stable since the system
for the water distribution problem. International
Journal ofinControl,
state is close to 2.5 hours, the response is fast. It seems
Water
Science & Technology,
52(5):265273, 2005.
and population dynamicsbased controllers have passivity
83(9):19471964,
2010.
that the basicproperties.
hydraulic
oriented
to evacuate
the
[3] [11]
K. Berggren,
M. Olofsson,
M. Viklander,
Svensson, dynamics.
and A. GustafsW. H. Sandholm.
Population
games andG.evolutionary
As design
further is
work,
it is proposed
to implement
MIT Press,
2010.drainage systems due to changes in
son.Cambridge,
HydraulicMass.
impacts
on urban
run-off as fast
as
possible.
However,
this
design
approach
the decentralized population dynamicsbased control in a [12] W. H. Sandholm, E. Dokumaci, and R. Lahkar. The projection
rainfall caused by climatic change. Journal of Hydrologic Engineering,
can lead to larger
flooding
in terminal
nodes
as Bogot
seena in
scaleevents
UDS problem
(e.g., the
entire
UDS),
dynamic and the replicator dynamic. Games and Economic Behavior,
17(1):9298,
2011.
64(2):666683, 2008.
advantage
of the
passivity
properties
of the the
closed [4] G. Obando,
Figure 4f). Intaking
the case
where
control
is used,
reaching
A.and
Pantoja,
and N.Evolutionary
Quijano. Building
Temperature
[13] P. D. Taylor
L. B. Jonker.
stable strategies
and gameControl
loopreservoirs
system. Additionally,
some
of the
issues related
based
on Population
Dynamics.
IEEE40(1):145156,
Transactions1978.
on Control Systems
dynamics.
Mathematical
biosciences,
steady state for
takes more
time
compared
to theto the
Technology.,
2014. McGraw-Hill Book Company,
evacuation of the wastewater in reservoirs might be studied [14]
V. Te Chow.22(1):404412,
Open channel hydraulics.
uncontrolled by
case,
and this is a consequence of the retaining
Inc; New York, 1959. Model predictive control of wastewater systems.
adding a multi-objective controller based on population [5] C. Ocampo-Martinez.
[15]
P. Willems,
K. Arnbjerg-Nielsen, J. Olsson, and V. Nguyen. Climate
property provided
by the controller. The control objective
Springer,
2010.
dynamics.
change impact assessment on urban rainfall extremes and urban
[6] C. drainage:
Ocampo-Martinez,
S. Bovo, and V. Puig. Partitioning approach
is satisfied, i.e., reducing overflows by distributing better
Methods and shortcomings. Atmospheric research, 103:106
oriented
to the decentralised predictive control of large-scale systems.
118, 2012.
the water resource. However, asR EFERENCES
future work it would be
Journal of Process Control, 21(5):775786, 2011.
appropriate to establish an additional control objective related
[7] E. Ramirez-Llanos and N. Quijano. A population dynamics approach
Arnbjerg-Nielsen, P. Willems, J. Olsson, S. Beecham, A. Pathirana,
for the water distribution problem. International Journal of Control,
to evacuation [1]
rateK.
throughout
network.
I.B.flow
Gregersen,
H. Madsen,the
V. Nguyen,
et al. Impacts of climate
change on rainfall extremes and urban drainage systems: a review.
[8]
Water Science & Technology, 68(1):1628, 2013.
VII. [2]
C ONCLUSIONS
AND A.
FURTHER
R. Ashley, D. Balmforth,
Saul, and J. WORK
Blanskby. Flooding in the
future predicting climate change, risks and responses in urban areas. [9]
Water Science & Technology, 52(5):265273, 2005.
It has been [3]
shown
that the proposed decentralized populaK. Berggren, M. Olofsson, M. Viklander, G. Svensson, and A. Gustafsson. Hydraulic
on urban drainage
systems
to changes in[10]
tion dynamicsbased
controlimpacts
is efficient
in terms
of due
a better
rainfall caused by climatic change. Journal of Hydrologic Engineering,
distribution of wastewater
throughout
the
UDS,
avoiding
the
17(1):9298, 2011.
J. Hofbauer andproposed
K. Sigmund.methodology
Evolutionary games
and population
overflows. The[4]partitioning
allows
to [11]
dynamics. Cambridge University Press, 1998.
design the decentralized
controller
by
using
different
local
[5] A. Nagurney and D. Zhang. Projected dynamical systems in the
stability analysis, and computation
of fixedrespect
demand traffic
controllers with formulation,
a lower computational
burden with
network equilibria. Transportation Science, 31:147158, 1997.

to a centralized controller. Furthermore, it has been shown

that the proposed closedloop system is stable since the UDS
and population dynamicsbased controllers have passivity
properties. As further work, it is proposed to implement
the decentralized population dynamicsbased control in a
larger scale UDS problem (e.g., the entire Bogota UDS) by
taking advantage of the passivity properties of the closed
loop system. Additionally, some of the issues related to the
evacuation of the wastewater in reservoirs might be studied

83(9):19471964, 2010.
W. H. Sandholm. Population games and evolutionary dynamics.
Cambridge, Mass. MIT Press, 2010.
W. H. Sandholm, E. Dokumaci, and R. Lahkar. The projection
dynamic and the replicator dynamic. Games and Economic Behavior,
64(2):666683, 2008.
V. Te Chow. Open channel hydraulics. McGraw-Hill Book Company,
Inc; New York, 1959.
P. Willems, K. Arnbjerg-Nielsen, J. Olsson, and V. Nguyen. Climate
change impact assessment on urban rainfall extremes and urban
drainage: Methods and shortcomings. Atmospheric research, 103:106
118, 2012.