Smart Systems for Urban Water Demand Management Pantelis Sopasakis joint work with A.K. Sampathirao, P. Patrinos & A. Bemporad. IMT Institute for Advanced Studies Lucca Monte Verit´ a, Switzerland, 22-25 Aug 2016
Smart Systems for Urban Water DemandManagement
Pantelis Sopasakisjoint work with A.K. Sampathirao, P. Patrinos & A. Bemporad.
IMT Institute for Advanced Studies LuccaMonte Verita, Switzerland, 22-25 Aug 2016
Today’s talk
We will learn how to:
I model water networks
I identify control objectives
I make decisions under uncertainty
I formulate MPC problems
I devise algorithms to solve them
I parallelise them on GPUs
1 / 89
Introd
uctio
n
Con
trol
Alg
orith
ms
Simul
atio
ns
Wha
t to
mod
el
Lite
ratu
re o
verv
iew
Con
trol-o
rient
ed m
odel
s
Fore
cast
ing
The
MPC
con
cept
Wor
st-c
ase
MPC
Stoc
hast
ic M
PCSc
enar
io-b
ased
MPC
Prox
imal
ope
rato
r
Con
vex
conj
ugat
e
Dua
lity
APG
alg
orith
mD
WN
SM
PC p
robl
ems
Para
llelis
atio
n
KPIs
Sim
ulat
ion
resu
ltsCon
clus
ions
Modoller’s todos
Modelling and Control
of DWNs
PHYSICALPHYSICAL
mass balancespressure dropschlorine balances
2 / 89
Modoller’s todos
Modelling and Control
of DWNs
PHYSICALPHYSICAL
TIME SERIESTIME SERIES
mass balancespressure dropschlorine balances
seasonal ARIMABATS modelsneural networksSVM and other ML
3 / 89
Modoller’s todos
Modelling and Control
of DWNs
PHYSICALPHYSICAL
TIME SERIESTIME SERIES
UNCERTAINTYUNCERTAINTY
mass balancespressure dropschlorine balances
seasonal ARIMABATS modelsneural networksSVM and other ML
parametric PDFsscenario trees
4 / 89
Modoller’s todos
Modelling and Control
of DWNs
PHYSICALPHYSICAL
TIME SERIESTIME SERIES
UNCERTAINTYUNCERTAINTY
CONSTRAINTSCONSTRAINTS
mass balancespressure dropschlorine balances
seasonal ARIMABATS modelsneural networksSVM and other ML
parametric PDFsscenario trees
tanks pumps pressure drops
5 / 89
Modoller’s todos
Modelling and Control
of DWNs
PHYSICALPHYSICAL
TIME SERIESTIME SERIES
UNCERTAINTYUNCERTAINTY
CONSTRAINTSCONSTRAINTSOBJECTIVESOBJECTIVES
mass balancespressure dropschlorine balances
seasonal ARIMABATS modelsneural networksSVM and other ML
parametric PDFsscenario trees
tanks pumps pressure drops
energy consumptionoperating costsmoothness of actuationsafety storage
`
6 / 89
Control of water networks
ControllerDrinking Water
Network
Water demand
Electricity prices
Measurements
Actuation
7 / 89
Taxonomy of control methodologies
Open Loop Closed Loop
CertaintyEquivalent Worst case Stochastic Risk-averse
Control of DWNs
● No feedback● Assumes perfect knowledge● No contingency plan● Requires human intervention Model Predictive
Control
Creasy 1998; Yu et al., 1994; Zessler and Shamir, 1989
8 / 89
Taxonomy of control methodologies
Open Loop Closed Loop
CertaintyEquivalent Worst case Stochastic Risk-averse
Control of DWNs
● Feedback compensates for modelling errors
Model Predictive Control
Creasy 1998; Yu et al., 1994; Zessler and Shamir, 1989
9 / 89
Taxonomy of control methodologies
Open Loop Closed Loop
Model Predictive Control
CertaintyEquivalent Worst case Stochastic Risk-averse
Control of DWNs
● Suitable for MIMO● Control under uncertainty● Goal-driven● Requires simple model
Creasy 1998; Yu et al., 1994; Zessler and Shamir, 1989
10 / 89
Taxonomy of control methodologies
Open Loop Closed Loop
CertaintyEquivalent Worst case Stochastic Risk-averse
Control of DWNs
● Model assumed accurate● Constraints may be violated● Suboptimal (we can do better)
Model Predictive Control
Sampathirao et al., 2014; Bakker et al., 2013; Leirens et al., 2010; Ocampo et al., 2009.
11 / 89
Taxonomy of control methodologies
Open Loop Closed Loop
CertaintyEquivalent Worst case Stochastic Risk-averse
Control of DWNs
● Pessimistic and conservative● Leads to small domain of attraction
Model Predictive Control
Ocampo et al., 2009 and 2010.
12 / 89
Taxonomy of control methodologies
Open Loop Closed Loop
CertaintyEquivalent Worst case Stochastic Risk-averse
Control of DWNs
● Makes use of probabilistic info● Less conservative, more realistic
Model Predictive Control
Sampathirao et al., 2014; Goryashko and Nemirovski, 2014; Tran and Brdys, 2009, Watkins and McKinney, 1997
13 / 89
Taxonomy of control methodologies
Open Loop Closed Loop
CertaintyEquivalent Worst case Stochastic Risk-averse
Control of DWNs
● Beyond the state of art● Uncertainty in uncertainty
Model Predictive Control
Sampathirao et al., 2016
14 / 89
Objectives
To MODEL (water demands, hydraulics, uncertainty, etc), pose astochastic predictive CONTROL problem (define objectives, constraints)
and devise algorithms to SOLVE it numerically.
15 / 89
Intro
duct
ion
Con
trol
Alg
orith
ms
Simul
atio
ns
Wha
t to
mod
el
Lite
ratu
re o
verv
iew
Con
trol-o
rient
ed m
odel
s
Fore
cast
ing
The
MPC
con
cept
Wor
st-c
ase
MPC
Stoc
hast
ic M
PCSc
enar
io-b
ased
MPC
Prox
imal
ope
rato
r
Con
vex
conj
ugat
e
Dua
lity
APG
alg
orith
mD
WN
SM
PC p
robl
ems
Para
llelis
atio
n
KPIs
Sim
ulat
ion
resu
ltsCon
clus
ions
MPC
Our case study
DWN of Barcelona: 63 tanks, 114 pumping stations and valves, 88 demand nodes & 17 pipe intersection nodes.
21 / 89
Control-oriented models
Simple mass balance equation (in discrete time)
xk+1 = Axk +Buk +Gddk,
0 = Euk + Eddk,
xk: tank volumes, uk: flows (controlled by pumping), dk: demands —along with the constraints
xmin ≤ xk ≤ xmax,
umin ≤ uk ≤ umax.
Sampathirao, Sopasakis et al., 2016, 2014; Wang et al., 2014; Ocampo et al., 2010; Ocampo et al., 2009.
22 / 89
Control-oriented models
20 40 60 80 100 1202
3
4
5
6
7x 104 d100CFE
time (h)
m3
20 40 60 80 100 120
200
300
400
500
d114SCL
time (h)
m3
20 40 60 80 100 120
1000
2000
3000
4000
d115CAST
time (h)
m3
20 40 60 80 100 120
0.5
1
1.5
x 104 d130BAR
time (h)
m3
20 40 60 80 100 120500
1000
1500
2000
2500
3000
d132CMF
time (h)
m3
20 40 60 80 100 120
400
600
800
1000d135VIL
time (h)
m3
20 40 60 80 100 120200
400
600
800
1000
d176BARsud
time (h)
m3
20 40 60 80 100 120
5001000
1500
2000
25003000
d450BEG
time (h)
m3
20 40 60 80 100 120
1000
2000
3000
d80GAVi80CAS85
time (h)m
3
SimulatorReal dataUpper limitSafety level
Source: EFFINET, deliverable report D2.123 / 89
Demand forecasting
Demand prediction concept:
dk+j(εj) = dk+j|k + εj
where
1. dk+j : actual demand at time k + j
2. dk+j|k: prediction of dk+j using info up to time k
3. εj : j-step-ahead prediction error
and dk+j|k is a function of observable quantities up to time k.
Sampathirao, Sopasakis et al., 2016.
24 / 89
Demand forecasting
Common approaches:
1. Neglect the error: (ε0, ε1, . . . , εN ) u (0, 0, . . . , 0)
2. Error bounds: (ε0, ε1, . . . , εN ) ∈ E , e.g., εj ∈ [εminj , εmax
j ]
3. Independent normal distributions: εj ∼ N (mj , σ2j )
4. (ε0, ε1, . . . , εN ) is random and admits finitely many values
25 / 89
Demand forecasting
Common approaches:
1. Neglect the error: (ε0, ε1, . . . , εN ) u (0, 0, . . . , 0)
2. Error bounds: (ε0, ε1, . . . , εN ) ∈ E , e.g., εj ∈ [εminj , εmax
j ]
3. Independent normal distributions: εj ∼ N (mj , σ2j )
4. (ε0, ε1, . . . , εN ) is random and admits finitely many values
25 / 89
Demand forecasting
Common approaches:
1. Neglect the error: (ε0, ε1, . . . , εN ) u (0, 0, . . . , 0)
2. Error bounds: (ε0, ε1, . . . , εN ) ∈ E , e.g., εj ∈ [εminj , εmax
j ]
3. Independent normal distributions: εj ∼ N (mj , σ2j )
4. (ε0, ε1, . . . , εN ) is random and admits finitely many values
25 / 89
Demand forecasting
Common approaches:
1. Neglect the error: (ε0, ε1, . . . , εN ) u (0, 0, . . . , 0)
2. Error bounds: (ε0, ε1, . . . , εN ) ∈ E , e.g., εj ∈ [εminj , εmax
j ]
3. Independent normal distributions: εj ∼ N (mj , σ2j )
4. (ε0, ε1, . . . , εN ) is random and admits finitely many values
25 / 89
Error bounds
0 20 40 60 80 100 120 140 160 180 2000
0.01
0.02
0.03
0.04
0.05
0.06
0.07
Time [h]
Wa
ter
De
ma
nd
Flo
w [
m3/h
]
Forecasting of Water Demand
Future Past
26 / 89
Predicted scenarios
Time [hr]
4370 4380 4390 4400 4410 4420 4430
Wa
ter
de
ma
nd
[m
3/s
]
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05
28 / 89
Control objectives
Stage costs:
1. Economic cost: `w(uk, k) = Wα(α1 + α2,k)′uk
2. Smooth operation cost: `∆(∆uk) = ∆u′kWu∆uk
3. Safe operation cost: `S(xk) = Wx‖[xs − xk]+‖4. Total cost: ` = `w + `∆ + `S .
We define ∆uk = uk − uk−1
Sampathirao et al., 2014; Cong Cong et al., 2014
29 / 89
Control objectives
Stage costs:
1. Economic cost: `w(uk, k) = Wα(α1 + α2,k)′uk
2. Smooth operation cost: `∆(∆uk) = ∆u′kWu∆uk
3. Safe operation cost: `S(xk) = Wx‖[xs − xk]+‖4. Total cost: ` = `w + `∆ + `S .
We define ∆uk = uk − uk−1
Sampathirao et al., 2014; Cong Cong et al., 2014
29 / 89
Control objectives
Stage costs:
1. Economic cost: `w(uk, k) = Wα(α1 + α2,k)′uk
2. Smooth operation cost: `∆(∆uk) = ∆u′kWu∆uk
3. Safe operation cost: `S(xk) = Wx‖[xs − xk]+‖
4. Total cost: ` = `w + `∆ + `S .
We define ∆uk = uk − uk−1
Sampathirao et al., 2014; Cong Cong et al., 2014
29 / 89
Control objectives
Stage costs:
1. Economic cost: `w(uk, k) = Wα(α1 + α2,k)′uk
2. Smooth operation cost: `∆(∆uk) = ∆u′kWu∆uk
3. Safe operation cost: `S(xk) = Wx‖[xs − xk]+‖4. Total cost: ` = `w + `∆ + `S .
We define ∆uk = uk − uk−1
Sampathirao et al., 2014; Cong Cong et al., 2014
29 / 89
Model Predictive Control
Problem formulation
minimiseπ=({uk+j|k}j ,{xk+j|k}j)
V (π) :=
N−1∑j=0
`(xk+j|k, uk+j|k, uk+j−1|k, k),
subject to
xk+j+1|k = Axk+j|k +Buk+j|k +Gddk+j|k dynamics
Euk+j|k + Eddk+j|k = 0 algebraic
xmin ≤ xk+j|k ≤ xmax vol. constr.
umin ≤ uk+j|k ≤ umax flow constr.
xk|k = xk, uk−1|k = uk−1 initial cond.
35 / 89
Worst-case MPCProblem formulation
minimiseπ=({uk+j|k}j ,{xk+j|k}j)
maxdk+j|k
V (π),
subject to
dk+j = dk+j|k + εj predictions
εj ∈ Ej err. bounds
xk+j+1|k = Axk+j|k +Buk+j|k +Gddk+j dynamics
Euk+j|k + Eddk+j = 0 algebraic
xmin ≤ xk+j|k ≤ xmax vol. constr.
umin ≤ uk+j|k ≤ umax flow constr.
xk|k = xk, uk−1|k = uk−1 initial cond.
Here uk+j|k is a function of εj – not a fixed value!
37 / 89
Worst-case MPC
Attention! We are looking for control laws (functions) uk+j|k. We mayparametrise (why?) these functions as
uk+j|k = Kjej + bj ,
and solve for Kj and bj .
There exist other parametrisations as well.
38 / 89
Worst-case MPC (tube-based)
Problem formulation
minimiseπ=({uk+j|k}j ,{xk+j|k}j)
V (π),
subject to
xk+j+1|k = Axk+j|k +Buk+j|k +Gddk+j|k dynamics
Euk+j|k + Eddk+j|k = 0 algebraic
xk+j|k ∈ X S volume constr.
umin ≤ uk+j|k ≤ umax flow constr.
xk|k = xk, uk−1|k = uk−1 initial cond.
* the constraint Euk+j|k + Eddk+j|k = 0 (certainty-equivalent) will not be satisfied for all εj
40 / 89
Stochastic MPC
Problem formulation
minimiseπ=({uk+j|k}j ,{xk+j|k}j)
EV (π),
subject to
xk+j+1|k = Axk+j|k +Buk+j|k +Gddk+j|k(εj) dynamics
Euk+j|k + Eddk+j(εj) = 0 alg. cond.
xmin ≤ xk+j|k ≤ xmax vol. constr.
umin ≤ uk+j|k ≤ umax flow constr.
xk|k = xk, uk−1|k = uk−1 initial cond.
again, we’re looking for control laws uk+j|k(εj).
49 / 89
Scenario-based stochastic MPC
Problem formulation:
minimiseπ=({uk+ji|k}i,j ,{xk+ji|k}i,j)
EV (π)︷ ︸︸ ︷N−1∑j=0
µ(j)∑i=1
pij`ij ,
subject to
xik+j+1|k=f(xanc(j+1,i)k+j|k , uik+j|k, d
ik+j|k) dynamics
Euik+j|k + Eddik+j|k = 0 algebraic
xmin ≤ xik+j|k ≤ xmax volume constr.
umin ≤ uik+j|k ≤ umax flow constr.
x1k|k = xk, u
1k−1|k = uk−1 initial cond.
54 / 89
Intro
duct
ion
Con
trol
Alg
orith
ms
Simul
atio
ns
Wha
t to
mod
el
Lite
ratu
re o
verv
iew
Con
trol-o
rient
ed m
odel
s
Fore
cast
ing
The
MPC
con
cept
Wor
st-c
ase
MPC
Stoc
hast
ic M
PCSc
enar
io-b
ased
MPC
Prox
imal
ope
rato
r
Con
vex
conj
ugat
e
Dua
lity
APG
alg
orith
mD
WN
SM
PC p
robl
ems
Para
llelis
atio
n
KPIs
Sim
ulat
ion
resu
ltsCon
clus
ions
theory
The proximal operator
Let g : IRn → IR ∪ {+∞} be a proper, closed function and γ > 0. Define
proxγg(v) = arg minz
{g(z) + 1
2γ ‖z − v‖2}
If this is easy to compute we call g prox-friendly.
Parikh and Boyd, 2014; Combettes and Pesquet, 2010.
56 / 89
The proximal operator
Example 1. Take
g(x) = δ(x | C) =
{0, if x ∈ C+∞, otherwise
Then, proxγg(v) = proj(v | C).
Parikh and Boyd, 2014; Combettes and Pesquet, 2010.
57 / 89
The proximal operator
Example 2. Take
g(x) = d(x | C) = infy∈C‖y − x‖
Then,
proxλg(v) =
{v + proj(v|C)−v
d(v|C) , if d(v | C) > λ
proj(v | C), otherwise
Parikh and Boyd, 2014; Combettes and Pesquet, 2010.
58 / 89
The proximal operator
Key Property. Suppose g is given as
g(x) =
κ∑i=1
gi(xi),
then(proxλg(v))i = proxλgi(vi).
Parikh and Boyd, 2014; Combettes and Pesquet, 2010.
59 / 89
The proximal operator
Key Property. Suppose g is given as
g(x) =
κ∑i=1
gi(xi),
then(proxλg(v))i = proxλgi(vi).
Parikh and Boyd, 2014; Combettes and Pesquet, 2010.
59 / 89
Convex Conjugate
Let f : IR→ IR ∪ {+∞} be convex, proper and closed. We define itsconvex conjugate as
f∗(y) = supx
{x′y − f(x)
}.
When f∗ is differentiable, then
∇f∗(y) = arg minz{x′y + f(z)}.
If f is prox-friendly, then
proxλf (v) + λproxλ−1f∗(λ−1v) = v.
If f is strongly convex, then f∗ is continuously diff/ble (Rockaffelar and Wets, 2009; Prop. 12.60).
60 / 89
Convex Conjugate
Let f : IR→ IR ∪ {+∞} be convex, proper and closed. We define itsconvex conjugate as
f∗(y) = supx
{x′y − f(x)
}.
When f∗ is differentiable, then
∇f∗(y) = arg minz{x′y + f(z)}.
If f is prox-friendly, then
proxλf (v) + λproxλ−1f∗(λ−1v) = v.
If f is strongly convex, then f∗ is continuously diff/ble (Rockaffelar and Wets, 2009; Prop. 12.60).
60 / 89
Convex Conjugate
Let f : IR→ IR ∪ {+∞} be convex, proper and closed. We define itsconvex conjugate as
f∗(y) = supx
{x′y − f(x)
}.
When f∗ is differentiable, then
∇f∗(y) = arg minz{x′y + f(z)}.
If f is prox-friendly, then
proxλf (v) + λproxλ−1f∗(λ−1v) = v.
If f is strongly convex, then f∗ is continuously diff/ble (Rockaffelar and Wets, 2009; Prop. 12.60).
60 / 89
Forward-Backward Splitting
The forward-backward splitting is the representation of an optimisationproblem as follows
minimisez
f(z) + g(z),
where
I f, g: closed, convex
I f : diff/ble with Lipschitz gradient
I g: prox-friendly
FB Splittings are not unique.
61 / 89
Forward-Backward Splitting
Example 1. `1-regularized least squares:
minimisez
12‖Az − b‖
2 + ‖z‖1.
62 / 89
Forward-Backward Splitting
Example 2. Box-constrained QP
minimisez
12z′Qz + q′z + δ(z | C),
where C = {z : zmin ≤ z ≤ zmax}.
63 / 89
Forward-Backward Splitting
Example 3. Constrained QP
minimisez
12z′Qz + q′z + δ(Hz | C).
but, g(z) := δ(Hz | C) is not prox-friendly!
64 / 89
Dual optimisation problem
If g(z) is prox-friendly, but g(Hz) is not we may formulate the dualoptimisation problem
Under certain conditions these two problems have the same minimum and
z? = ∇f∗(−H ′y?).
Fenchel duality generalises Lagrangian duality (Rockafellar, 1972).
65 / 89
Dual optimisation problem
If g(z) is prox-friendly, but g(Hz) is not we may formulate the dualoptimisation problem
Under certain conditions these two problems have the same minimum and
z? = ∇f∗(−H ′y?).
Fenchel duality generalises Lagrangian duality (Rockafellar, 1972).
65 / 89
Proximal gradient algorithm
The proximal gradient method for solving
minimisez
f(z) + g(z)
where f is differentiable with L-Lipschitz gradient runs
zν+1 = proxγg(zν − γ∇f(zν)),
with γ ∈ (0, L−1).
66 / 89
Dual proximal gradient algorithm
The proximal gradient method applied to the dual
minimisez
f∗(−H ′y) + g∗(y)
where f∗ is differentiable with L-Lipschitz gradient is
yν+1 = proxγg∗(yν + γH∇f∗(−H ′yν))
If f is L−1-strongly convex, then f∗ has L-Lipschitz gradient.
67 / 89
Dual proximal gradient algorithm
The dual proximal gradient method
yν+1 = proxγg∗(yν + γH∇f∗(−H ′yν))
can be written as
zν = ∇f∗(−H ′yν)
tν = proxλ−1g(λ−1yν +Hzν)
yν+1 = yν + λ(Hzν − tν).
68 / 89
Dual proximal gradient algorithm
Nesterov’s accelerated proximal gradient method converges as O(1/k2)instead of O(1/k):
wν = yν + θν(θ−1ν−1 − 1)(yν − yν−1)
zν = ∇f∗(−H ′wν)
tν = proxλ−1g(λ−1wν +Hzν)
yν+1 = wν + λ(Hzν − tν)
θν+1 = 12(√θ4ν + 4θ2
ν − θ2ν)
with θ0 = θ−1 = 1 and y0 = y−1 = 0 (Nesterov, 1983).
69 / 89
The DWN control problem
P : minimiseπ=({uk+ji|k}i,j ,{xk+ji|k}i,j)
EV (π)︷ ︸︸ ︷N−1∑j=0
µ(j)∑i=1
pij`ij ,
subject to
xik+j+1|k=f(xanc(j+1,i)k+j|k , uik+j|k, d
ik+j|k) dynamics
Euik+j|k + Eddik+j|k = 0 algebraic
xmin ≤ xik+j|k ≤ xmax volume constr.
umin ≤ uik+j|k ≤ umax flow constr.
x1k|k = xk, u
1k−1|k = uk−1 initial cond.
We will write this problem as: minimisezf(z) + g(Hz).
70 / 89
The DWN control problem
Let z = {xij , uij} and define
f(z) =
N−1∑j=0
µ(j)∑i=1
pij(`w(uij) + `∆(∆uij)) + δ(uij |Φ1(dij))
+ δ(xij+1, uij , x
anc(j+1,i)j |Φ2(dij)),
whereΦ1(d) = {u : Eu+ Edd = 0}
andΦ2(d) = {(x+, x, u) : x+ = Ax+Bu+Gdd}
Sampathirao et al., 2016.
71 / 89
The DWN control problem
In other words:
f(z) = smooth cost + dynamics + alg equations
and
g(z) = all the rest
= nonsmooth cost + constraints
=
N−1∑j=0
µ(j)∑i=1
`S(xij) + δ(xij | X) + δ(uij | U),
but this g is not prox-friendly (it is not separable!).
Sampathirao et al., 2016.
72 / 89
The DWN control problem
We create a copy of {xij}j,i which we denote by χij and introduce
t = ({xij}j,i, {χij}j,i, {uij}j,i)
with xij = χij . Then
g(t) =
N−1∑j=0
µ(j)∑i=1
`S(xij) + δ(χij | X) + δ(uij | U)
is prox-friendly.
It is t = Hz.
73 / 89
Dual gradient computation
To compute the dual gradient we use
∇f∗(y) = arg minz{x′y + f(z)}
= arg minz:dynamics
Euij+Eddij=0
{x′y +∑j,i
quadratic(zij)}
This is an equality-constrained quadratic problem which can be solvedvery efficiently using dynamic programming.
74 / 89
Intro
duct
ion
Con
trol
Algo
rithm
s
Simul
atio
ns
Wha
t to
mod
el
Lite
ratu
re o
verv
iew
Con
trol-o
rient
ed m
odel
s
Fore
cast
ing
The
MPC
con
cept
Wor
st-c
ase
MPC
Stoc
hast
ic M
PCSc
enar
io-b
ased
MPC
Prox
imal
ope
rato
r
Con
vex
conj
ugat
e
Dua
lity
APG
alg
orith
mD
WN
SM
PC p
robl
ems
Para
llelis
atio
n
KPIs
Sim
ulat
ion
resu
ltsCon
clus
ions
it is fast
# scenarios
100 200 300 400 500
Ru
ntim
e (
s)
0
200
400
600
800
1000
1200
Gurobi
APG (500 iterations)
0 100 200 300 400 500
Ru
ntim
e s
lop
e
0
2
4
84 / 89
it is efficient
Number of scenarios
50 100 150 200 250 300
KP
I E/1
00
0
1.55
1.6
1.65
1.7
1.75
1.8
KP
I S/1
000
1
2
3
4
5
6KPIE
KPIS
85 / 89
Closed-loop simulations
20 40 60 80 100 120 140 160
Contr
ol action [%
]
0
0.2
0.4
0.6
0.8
Time [hr]
20 40 60 80 100 120 140 160
Wate
r C
ost [e
.u.]
60
70
80
90
86 / 89
Intro
duct
ion
Con
trol
Algo
rithm
s
Sim
ulat
ions
Wha
t to
mod
el
Lite
ratu
re o
verv
iew
Con
trol-o
rient
ed m
odel
s
Fore
cast
ing
The
MPC
con
cept
Wor
st-c
ase
MPC
Stoc
hast
ic M
PCSc
enar
io-b
ased
MPC
Prox
imal
ope
rato
r
Con
vex
conj
ugat
e
Dua
lity
APG
alg
orith
mD
WN
SM
PC p
robl
ems
Para
llelis
atio
n
KPIs
Sim
ulat
ion
resu
ltsCon
clus
ions
Open problems
I Nonlinear
I Risk-averse
I Distributed
I Robust Economic MPC
I Faster Algorithms
Problem: introduce nonlinear pressure drop equations
87 / 89
Open problems
I Nonlinear
I Risk-averse
I Distributed
I Robust Economic MPC
I Faster Algorithms
Challenge: the distribution of future errors is not exactly known
87 / 89
Open problems
I Nonlinear
I Risk-averse
I Distributed
I Robust Economic MPC
I Faster Algorithms
Problems: spatial decomposition, communication constraints
87 / 89
Open problems
I Nonlinear
I Risk-averse
I Distributed
I Robust Economic MPC
I Faster Algorithms
Questions: performance guarantees, recursive feasibility
87 / 89
Open problems
I Nonlinear
I Risk-averse
I Distributed
I Robust Economic MPC
I Faster Algorithms
Ongoing work: quasi-Newtonian LBFGS-type algorithms
87 / 89
References (i)
1. A. Sampathirao, P. Sopasakis, A. Bemporad, and P. Patrinos, “Fast parallelizable scenario-based stochasticoptimization,” EUCCO 2016, Leuven, Belgium, 2016.
2. A. Sampathirao, P. Sopasakis, A. Bemporad, and P. Patrinos, “GPU-accelerated stochastic predictive control of drinkingwater networks,” IEEE CST (submitted, provisionally accepted), 2016 (on arXiv).
3. A. Sampathirao, J. Grosso, P. Sopasakis, C. Ocampo-Martinez, A. Bemporad, and V. Puig, “Water demand forecastingfor the optimal operation of large-scale drinking water networks: The Barcelona case study,” 19th IFAC World Congress,pp. 10457–10462, 2014.
4. A. Sampathirao, P. Sopasakis, A. Bemporad, and P. Patrinos, “Distributed solution of stochastic optimal controlproblems on GPUs, 54th IEEE Conf. Decision and Control, (Osaka, Japan), Dec 2015.
5. A. Sampathirao, P. Sopasakis, A. Bemporad, and P. Patrinos, “Proximal Quasi-Newton Methods for Scenario-basedStochastic Optimal Control,” IFAC 2017 (submitted).
6. M. Bakker, J. H. G. Vreeburg, L. J. Palmen, V. Sperber, G. Bakker, and L. C. Rietveld, “Better water quality and higherenergy efficiency by using model predictive flow control at water supply systems,” J Wat Supply: Research &Technology – Aqua, 62(1), pp. 1–13, 2013.
7. S. Leirens, C. Zamora, R. Negenborn, and B. De Schutter, “Coordination in urban water supply networks usingdistributed model predictive control,” ACC 2010, (Baltimore, USA), pp. 3957–3962, 2010.
8. C. Ocampo-Martinez, V. Puig, G. Cembrano, R. Creus, and M. Minoves, “Improving water management efficiency byusing optimization-based control strategies: the barcelona case study,” Water Science and Technology: Water Supply,9(5), pp. 565–575, 2009.
9. C. Ocampo-Martinez, V. Fambrini, D. Barcelli, and V. Puig, “Model predictive control of drinking water networks: Ahierarchical and decentralized approach,” ACC 2010, (Baltimore, USA), pp. 3951–3956, 2010.
10. A. Goryashko and A. Nemirovski, “Robust energy cost optimization of water distribution system with uncertaindemand,” Automation and Remote Control 75(10), pp. 1754–1769, 2014.
11. U. Zessler and U. Shamir, “Optimal operation of water distribution systems,” J Wat Resour Plan & Mngmt, 115(6), pp.735–752, 1989.
88 / 89
References (ii)
12. G. Yu, R. Powell, and M. Sterling, “Optimized pump scheduling in water distribution systems,” J Optim Theory & Appl,83(3), pp. 463–488, 1994.
13. V. Tran and M. Brdys, “Optimizing control by robustly feasible model predictive control and application to drinkingwater distribution systems,” Artificial Neural Networks – ICANN 2009, vol. 5769 of Lecture Notes in Computer Science,pp. 823–834, Springer, 2009.
14. J. Watkins, D. and D. McKinney, “Finding robust solutions to water resources problems,” J Wat Res Plan & Mngmt123(1), pp. 49–58, 1997.
15. S. Cong Cong, S. Puig, and G. Cembrano, “Combining CSP and MPC for the operational control of water networks:Application to the Richmond case study,” 19th IFAC World Congress, (Cape Town), pp. 6246–6251, 2014.
16. J. Grosso, C. Ocampo-Mart nez, V. Puig, and B. Joseph, “Chance-constrained model predictive control for drinkingwater networks,” Journal of Process Control 24(5), pp. 504–516, 2014
17. P.L. Combettes and J.-C. Pesquet. “Proximal splitting methods in signal processing,” Technical report, 2010. URLhttp://arxiv.org/abs/0912.3522v4.
18. N. Parikh and S. Boyd, “Proximal algorithms,” Found. Trends Optim 1, pp. 127–239, 2014.
19. R. Rockafellar and J. Wets, “Variational analysis,” Berlin: Springer-Verlag, 3rd ed., 2009.
20. R. Rockafellar, “Convex analysis,” Princeton university press, 1972.
21. Yu. Nesterov, “A method of solving a convex programming problem with convergence rate O(1/k2),” SovietMathematics Doklady 72(2), pp. 372–376, 1983.
89 / 89