Int. J. Appl. Math. Comput. Sci., 2016, Vol. 26, No.

3, 641654
DOI: 10.1515/amcs-2016-0044

RELIABILITYBASED ECONOMIC MODEL PREDICTIVE CONTROL FOR

GENERALISED FLOWBASED NETWORKS INCLUDING ACTUATORS
HEALTHAWARE CAPABILITIES
J UAN M. GROSSO a , C ARLOS OCAMPO-MARTINEZ a , V ICEN PUIG a,

Automatic Control Department, Institute of Robotics and Industrial Informatics (CSIC-UPC)

Polytechnic University of Catalonia (UPC), Llorens and Artigas 46, 08028 Barcelona, Spain
e-mail: {jgrosso,cocampo,vpuig}@iri.upc.edu

This paper proposes a reliability-based economic model predictive control (MPC) strategy for the management of generalised flow-based networks, integrating some ideas on network service reliability, dynamic safety stock planning, and
degradation of equipment health. The proposed strategy is based on a single-layer economic optimisation problem with dynamic constraints, which includes two enhancements with respect to existing approaches. The first enhancement considers
chance-constraint programming to compute an optimal inventory replenishment policy based on a desired risk acceptability
level, leading to dynamical allocation of safety stocks in flow-based networks to satisfy non-stationary flow demands. The
second enhancement computes a smart distribution of the control effort and maximises actuators availability by estimating
their degradation and reliability. The proposed approach is illustrated with an application of water transport networks using
the Barcelona network as the case study considered.
Keywords: model predictive control, flow-based networks, dynamic safety stocks, actuator health, service reliability,
chance constraints, economic optimisation.

1. Introduction
The normal functioning of modern society strongly relies
on many instances of networks, e.g., communication
networks, electrical power networks, public transport
networks, road-traffic networks, water networks, oil and
gas networks, or supply chains, among other things.
Consequently, such networks are critical infrastructures
(Negenborn and Hellendoorn, 2010), and maintaining an
efficient, reliable and sustainable operation is a must for
all network managers (Kyriakides and Polycarpou, 2015).
Although critical infrastructures are conceived and
designed to supply different specific services, many of the
problems that trigger their operation (e.g., minimisation of
displacement times, maximisation of plants throughput,
minimisation of energy consumption, maximisation of
demand satisfaction, etc.) share a common feature:
some commodity (or many at the same time), e.g.,
water, oil, energy, products, among any other real or
abstract entities, needs to be transported through the
network infrastructure. Such similarity gave raise to

Corresponding author

the concept of generalised flow-based networks and to

classical network flow problems (cf. Ford and Fulkerson,
1962; Papageorgiou, 1984; Ahuja et al., 1993) that aim to
specify some control inputs influencing the flow process
in the network so as to optimise a given performance
criterion subject to constraints and to continuously
varying conditions of both deterministic and probabilistic
nature.
The management of generalised flow-based networks
is a complex task and has become a research subject
worldwide. Strategical and tactical decisions in physical
network operation can be addressed by different methods
proposed within the supply-chain theory (Papageorgiou,
2009), but the mathematical tools available in control
systems theory have shown to be more suitable to handle
the problem consisting of time variance, uncertainties,
delays, dimensionality and lack of system information
(see, e.g., Ortega and Lin, 2004; Sarimveis et al., 2008;
Schwartz and Rivera, 2010; Subramanian et al., 2013).
Most of the approaches developed in the aforementioned
references for the control of dynamic networks are mainly
focused on performance and robustness, and the control

Unauthenticated
Download Date | 11/25/16 10:31 PM

642
strategy is often implemented in a multi-layer control
architecture. However, due to the stochastic nature
of customer demands and the ageing behaviour of the
infrastructure components, there is an important issue
that needs to be considered in the design of the control
strategy, which is system/service reliability. This latter
aspect is the main focus of this paper.
Generally, reliability can be defined as the
probability that units, components, equipments and
systems will accomplish their intended function for a
specified period of time under some operating conditions
and specific environments (Gertsbakh, 2010). Thus,
from the perspective of supply-chain engineering,
reliability analysis of a generalised flow-based network
may be associated with the -service level (type I)
(Goetschalckx, 2011), which is an event-oriented
performance criterion that measures the probability that
all customer demands will be completely served within a
given time interval from the stock on hand without delay,
under normal and emergency conditions.
Service reliability and economic optimisation in
flow-based networks have been important research topics
in the field of inventory management for planning against
uncertainty in demand and/or supply. The main strategy
reported in the literature to assure a service level in
flow-based networks consists in performing demand
forecasting to guarantee a safety stock in storage units
(if existing) as a countermeasure to secure network
performance against forecast inaccuracy. Obtaining
and using advanced demand information enable network
operators to be more responsive to customer needs and
to improve inventory management (zer, 2003). The
interaction between forecasting and stock control is
well reviewed by Betts (2011), Guide and Srivastava
(2000), Kanet et al. (2010), Osman and Demirli (2012),
Schoenmeyr and Graves (2009), Strijbosch et al. (2011)
and the references therein. Nevertheless, to guarantee
a service level in flow-based networks, the control
strategies should consider not only demand uncertainty
but also network topological reliability, which refers to
the probability that a network is connected given its
components probability to remain operative at any time.
To the best of the authors knowledge, reliability
and degradation models of system and components
have not been addressed simultaneously with dynamic
safety stock planning in the framework of generalised
flow-based networks control. Reliability in flow-based
networks is commonly analysed off-line, i.e., a posteriori of the operation cycle, but without a measure
of capacity degradation that may exist in the actuators
of the network. Relevant attempts to compute the
required safety stocks considering the networks health
were presented by Blanchini et al. (1997; 2000) for the
control of production-distribution systems with uncertain
demands and system failures. In these works, necessary

J.M. Grosso et al.

and sufficient conditions to drive and keep the state
within the least storage level are obtained, but under
the requirement that the controller must be aware of
the demand uncertainty bounds and the actuator failure
configuration, which are not always possible to identify
and isolate.
Most of other approaches that study
component-health management and system reliability lie
within the framework of fault-tolerant control or in the
field of maintenance scheduling (see, e.g., the works
of Guida and Giorgio (1995), Martorell et al. (1999),
Gallestey et al. (2002), Khelassi et al. (2011), Pereira
et al. (2010), Chamseddine et al. (2014) and the references
therein), but they do not consider demand uncertainty.
Several economic-oriented controllers have been
recently proposed within the MPC framework (Ellis et al.,
2014), but without considering reliability issues. Both
safety stock and actuator lifetime share the fact that they
are conflicting with the economic performance of the
system. Therefore, it is desired to have a flexible control
strategy that allows a trade-off between the economic
optimisation and the reliability of the system. To achieve
this aim, only a two-layer hierarchical control strategy has
been proposed by Grosso et al. (2012) for network flow
optimisation considering both economic and reliability
criteria. In such a work, first an upper layer performs
a local steady-state economic optimisation to set up a
uniform back-off of a demand satisfaction constraint due
to an assumption of stationary demand uncertainty. At the
same stage, a deterministic model of actuator degradation
is used to monitor the system health and to set up the
maximum allowable degradation of the actuators at each
time step to distribute the overall control effort. Later, in a
lower layer, an economic MPC algorithm is implemented
to compute optimal control set-points that minimise a
multi-objective cost function.
The main contribution of this paper consists in
an improved reliability-based economic MPC strategy
that is aware of the actuator health and allows
dynamic management of risk for non-stationary demand
uncertainty, extending the results presented by Grosso
et al. (2012; 2014). Specifically, the two-layer control
architecture proposed by Grosso et al. (2012) is here
simplified and reduced to a less conservative single-layer
stochastic approach following the chance-constrained
MPC approach presented by Grosso et al. (2014).
The actuator-health management policy used in this
paper follows the one introduced by Pereira et al.
(2010), but considers stochastic actuator-degradation
models and probabilistic actuator lifetime constraints
rather than deterministic ones. The customer service
level is guaranteed here by means of probabilistic
demand satisfaction constraints. The proposed MPC
controller optimises directly the economic (possibly
multi-objective) performance of the network operation
instead of the commonly used tracking cost function.

Unauthenticated
Download Date | 11/25/16 10:31 PM

Reliability-based economic model predictive control for generalised flow-based networks...

The reliability-based tuning strategy proposed by Khelassi
et al. (2011) is here used as part of the constrained
optimisation problem to contribute in the optimal
allocation of the control effort. The Barcelona water
network is used to illustrate and assess the proposed
approach.
The remainder of this paper is organised as follows.
Section 2 briefly describes a control-oriented model
of generalised flow-based networks, and states the
safety stock allocation policy and the actuator-health
management policy.
Section 3 is devoted to the
formulation of the proposed reliability-based economic
MPC strategy. Section 4 describes the case study where
the effectiveness of the proposed approach is analysed via
simulations. Finally, Section 5 highlights the concluding
remarks that can be drawn from the results presented in
this paper, as well as some ideas for future research.
Notation. Throughout this paper, R, Rn , Rmn and R+
denote the field of real numbers, the set of column real
vectors of length n, the set of m by n real matrices and the
set of non-negative real numbers, respectively, while Z+
denotes the set of non-negative integer numbers including
zero. Define Z[a,b] := {x Z+ | a x b} for
some a, b Z+ and Zc := {x Z+ | x c} for
some c Z+ . For a vector x Rn , x(i) denotes the
i-th element of x. Similarly, X(i) denotes the i-th row of
a matrix X Rnm . Additionally,  Z denotes the
weighted 2-norm of a vector, i.e., xZ = (x Zx)1/2 .
If not otherwise noted, all vectors are column vectors.
Transposition is denoted by the superscript  and the
operators <, , =, >, denote element-wise relations of
vectors. Moreover, 0 denotes a zero column vector and I
the identity matrix, both of appropriate dimensions. For a
given vector x Rn , let diag(x) denote a diagonal matrix
in Rnn whose main diagonal contains the elements of x.
For a symmetric matrix Z Rnn , let Z  0 ( 0)
denote that Z is positive definite (semi-definite).

2. Problem statement
Consider a generalised flow-based network being denoted
as N = (G, p, S), which consists of a directed graph
G = (V, A) formed by a finite set of nodes V Z1 ,
and a finite set of arcs A V V, with an arc a A
being an ordered link between a pair of nodes (i, j) with
i, j V, whose order indicates the direction of the flow
between the two nodes. The network has a special subset
of nodes S V called terminals. A terminal is either
a source or a sink. The set of source nodes is denoted
as S + and the set of sink nodes is denoted as S , and it
follows that S = S + S . The rest of nodes i V \ S,
are called intermediate nodes. These latter nodes can be
further classified according to their flow storage capacity
into dynamic nodes and static nodes. The dynamic nodes

643

have non-zero storage capacity, while in the static ones

the transshipment of the commodity is immediate. The
functioning of the network is driven by a vector function p
containing the functions that define the dynamic attributes
of the graph, i.e., capacities, transit times, gains, supplies,
demands. It is supposed here that only the attributes
conforming p are time varying, while the structure of the
network (defined by G and S) remains unchanged.
In this paper, the following initial assumptions are
considered regarding the network operation.
Assumption 1. The network operates in a push-flow
regime with zero transit time for all a A.
Assumption 2. The flow through each arc a A is
controlled by an actuator for all a = (i, j) with i, j
{V \ S }. The flow does not experience any gain or loss
while traversing an arc.
In order to derive a control-oriented model, define
the state vector x Rn to represent the storage at the
dynamic nodes. Similarly, define the vector u Rm of
controlled inputs as the collection of the flow rate through
the arcs (i, j) Au := {(i, j) A such that i, j
V \ S }, and the vector d Rp of uncontrolled inputs
(demands) as the collection of flow rate through the arcs
(i, j) Ad := {(i, j) A such that i V \ S and j
S }. Following flow/mass balance principles as well as
Assumptions 1 and 2, a discrete-time model based on
linear difference-algebraic equations can be formulated
for the network N as follows:


xk+1 = Axk + Buk + Bd dk ,

0 = Eu uk + Ed dk ,

(1a)
(1b)

where k Z+ is the current time step while A, B, Bd ,

Eu and Ed are matrices of compatible dimensions dictated
by the network topology. Specifically, (1a) represents the
mass balance at dynamic nodes while (1b) represents the
mass balance at static nodes. The system is subject to
state and input constraints considered here in the form of
convex polyhedra defined as
xk X := {x Rn | Gx g},
uk U := {u Rm | Hu h},

(2a)
(2b)

for all k, where G Rrx n , g Rrx , H Rru m ,

h Rru , being rx Z+ and ru Z+ the number of
state and input constraints, respectively.
Assumption 3. The states in x and the demands in d are
measured at any time step k Z+ .
Assumption 4. The realisation of demands at any time
step k Z+ can be decomposed as
dk = dk + ek ,

Unauthenticated
Download Date | 11/25/16 10:31 PM

(3)

J.M. Grosso et al.

644
General Management Policies
P rices
Service L evel
Objectives
P riorities

Actuator Health
Monitoring

Wu,k
zk

Reliability-based Economic MPC

u

k

xk

k
d

Demand
Forecasting

Local Feedback Control

uk

Disturbance
(measured demands)

VN0 (k, x)

dk

xk
Plant

Fig. 1. Reliability-based MPC structure.

where dk Rp is the vector of expected disturbances

and ek Rp is the vector of forecasting errors with
non-stationary uncertainty and a known (or approximated)
quasi-concave probability distribution D(0, (e(j),k )).
The stochastic nature of each j-th row of dk is described
then by d(j),k Di (d(j),k , (e(j),k )), where d(j),k
denotes its mean and (e(j),k ) its variance.
The control goal is to minimise a convex (possibly
multi-objective and time-varying) stage cost function  :
Z+ X U R+ , which might bear any functional
relationship to the economics of the system operation.
To do so, the control strategy addressed in this paper is
based on the control system structure shown in Fig. 1,
where the information given by forecasting demand and
actuator health estimation modules is used within an
economic MPC controller to ensure a given service level
in the network. In such a structure, zk is the state of
the cumulative degradation of actuators and Wu,k is a
reliability-dependant matrix that adjusts the tuning of the
MPC controller (see Section 2.2 for details). Moreover,
uk and VN0 denote respectively the optimal control action
computed in the predictive control block and the optimal
value of the cost function that is embedded and minimised
within the proposed MPC controller (see Section 3).
2.1. Safety stock allocation policy. There is often
the need for guaranteeing a safety stock at each
storage node of a generalised flow-based network in
order to decrease the probability of stock-outs (when
a node has insufficient resources to satisfy either
external demands or the flow requested by other
intermediate nodes) due to possible uncertainties in the
network. As discussed in Section 1, stock allocation
problems have been addressed before in the literature of
supply chain management, where solutions are mainly
based on inventory planning strategies that incorporate,
within deterministic formulations, safety mechanisms
to cope with randomness and risks associated with
network operation (Christopher, 2005). Most techniques
from inventory management suppose a hierarchical

and descendant flow of products, even in multi-stage

multi-echelon schemes, in a way that predicted safety
stock changes are easily communicated backwards in
order to support availability of quantities when they are
needed (Kanet et al., 2010). Nevertheless, this behaviour
is not true in real large-scale generalised flow-based
networks since a meshed topology with multi-directional
flows between nodes prevails instead of spread tree
configurations.
To circumvent the aforementioned limitation
and determine optimal dynamic safety stocks, the
chance-constrained MPC strategy described by Grosso
et al. (2014) is used here. Such a strategy relaxes the
original state constraint (2a) by using probabilistic
statements, leading to the form of the so-called
(probabilistic) chance constraint, i.e.,
xk {x Rn | P[G(j) x g(j) , j Z[1,rx ] ] 1x },
(4)
for all k, where P denotes the probability operator,
x (0, 1) is the risk acceptability level of constraint
violation for the states, and G(j) and g(j) denote the
j-th row of G and g, respectively. This requires that all
rows j have to be jointly fulfilled with the probability
1 x . A lower x implies a harder constraint. As
discussed by Grosso et al. (2014), the constraint (4) is
difficult to be addressed since it lacks analytic expressions
due to the multivariate probability distributions involved.
Nevertheless, there are tractable approximations that
can be derived if each element of the demand vector
follows a log-concave univariate distribution with a
known stochastic description; see the work of Grosso
et al. (2014, Section 3) for details. Specifically, (4) can
be enforced by the following constraints:
(1 x,j ), (5)
G(j) (Axk + Buk ) g(j) FG1
(j) Bd dk
rx


x,j x ,

(6)

j=1

0 x,j 1,

(7)
FG1
()
(j) Bd dk

for all j Z[1,rx ] , where FG(j) Bd dk () and

are the cumulative distribution and the left-quantile
function of G(j) Bd dk , respectively. The constraints (5)
are the deterministic equivalent of the set of rx resultant
individual chance constraints. Moreover, (6) and (7) are
conditions imposed to bound the new single risks in such a
way that the joint risk bound is not violated. Any solution
that satisfies the above constraints is guaranteed to satisfy
(4). As suggested by Nemirovski and Shapiro (2006,
Remark 2.1), assigning a fixed and equal value of risk
to each individual constraint, i.e., x,j = x /rx for all
j Z[1,rx ] , satisfies (6) and (7).
In this way, the safety stocks are optimally allocated
and represented by the constraint back-off effect caused

Unauthenticated
Download Date | 11/25/16 10:31 PM

Reliability-based economic model predictive control for generalised flow-based networks...

by the term FG(j) Bd dk (1x,j ) in (4). Hence, the original
state constraint set X is contracted by the effect of the
rx deterministic equivalents in (5) and replaced with the
stochastic feasibility set given by
Xs,k
:= {xk Rn | uk U, such that
G(j) (Axk + Buk ) g(j) FG1

(j) Bd dk

(1 x,j ),

j Z[1,rx ] and Eu uk + Ed dk = 0},

for all k, where d = E[d] is the first moment of d.
From convexity of G(j) xk+1 g(j) and the log-concavity
assumption of the distribution, it follows that the set
Xs,k is convex when non-empty for all x,j (0, 1) in
most distribution functions (Kall and Mayer, 2005). For
some particular distributions, e.g., Gaussian, convexity is
retained for x,j (0, 0.5].
Remark 1. This strategy deals specifically with storage
node reliability (assuming their faulty behaviour as the
inability to satisfy their own demands), which is affected
by both the capacity and reliability of the elements
supplying flow to them. If the flow capacity is less than
the average demand, no storage unit will probably be large
enough to provide a sustained service.
2.2.
Actuator-health management policy. Unless
some damage mitigating policy is adopted to ensure
the availability of actuators for a given maintenance
horizon, their inherent degradation could compromise the
overall service reliability of the network. Therefore,
system safety can be enhanced by taking into account
the health of the components explicitly in controller
design. Several models have been proposed in the
literature to describe reliability and ageing of actuators
under nominal operation; see the works of Gorjian
et al. (2009), Guida and Giorgio (1995), and Letot and
Dehombreux (2012) for a review. Nevertheless, as pointed
out by Khelassi et al. (2011) and Martorell et al. (1999),
a realistic health measurement should also include the
trend of actuator ageing according to the variation of
the operating conditions. Rates of degradation can be
assumed constant for some equipment, but others present
a highly variable and non-linear rate depending on the
degradation mechanism and the local conditions. For
the sake of simplicity, the linear proportional degradation
model presented by Pereira et al. (2010) and its uniform
rationing heuristic are adopted in this paper, but with
the inclusion of an additive uncertainty. The approach
considers the health condition of each actuator being
described by a wear process with the rate associated with
the exerted control effort as follows:
zk+1 = zk + |uk | + k ,

(8)

645

where zk Rm denotes the state of cumulative

degradation of actuators at time step k and :=
diag(1 , . . . , m ) is a diagonal matrix of constant
degradation coefficients i R, i Z[1,m] , associated
with the m actuators. Moreover, Rm is a random
vector whose components lie in a normal distribution
N (0, (i) ).
Degradation of each actuator will accumulate until
the element reaches a state in which it will not perform
its function at an acceptable level. At such a point, it
can be considered that the actuator operation may be
compromising the network supply service unless demands
result reachable from other redundant flow paths or a
fault-tolerant mechanism is activated. Therefore, instead
of incurring into a failure that requires corrective control
actions, a preventive strategy can be implemented to
improve the overall system reliability by guaranteeing
that each actuator remains available until the instant of a
programmed maintenance intervention.
To circumvent the system availability problem,
an obvious approach is to constrain the accumulated
degradation of actuators at each time instant to remain
below a safe threshold until a predefined maintenance
horizon is reached. Here, the health management is
considered to be ruled by the probabilistic version of the
constraints proposed by Pereira et al. (2010), that is,
P[zk+N |k zmax,k ] (1 z ),
zk
z
,
zmax,k := zk + N thresh
M +N k

(9)
(10)

where N Z+ is a prediction horizon used for prognosis,

z (0, 1) is a risk acceptability level, zmax,k Rm
is the vector of maximum accumulated degradation of
actuators allowed for the time step k, and zthresh Rm
is the vector of thresholds for the terminal degradation at
a maintenance horizon M Z+ . Notice that (9) restricts
the predicted accumulated degradation of actuators health
at N -steps ahead from the current time step k and
its deterministic equivalent can be obtained similarly to
Section 2.1. The right-hand side of (10) is a uniform
rationing of the remaining allowed degradation (zthresh
zk ) that is updated at each time step according to the
control actions applied and ensures that zk zthresh for
k = M.
Remark 2. Despite the inherent relation, a degraded
state is not the same as a faulty state (see Hsu et al.,
1991). In fact, under nominal conditions of operation,
degradation always precedes failure. When a component
is degraded, maintenance actions should be executed to
improve its performance to acceptable levels, but when
the component is faulty, repairing actions are needed to
restore its functionality.
Keeping in mind the difference between degraded
and faulty states, it can be noticed that the strategy for

Unauthenticated
Download Date | 11/25/16 10:31 PM

J.M. Grosso et al.

646
uniform rationing of degradation should be complemented
with an another other safety mechanism to incorporate the
remaining useful life of the actuators on the basis of their
reliability and keep them available as long as possible.
Accordingly, here the improvement of the safety and
reliability of a generalised flow-based network is proposed
using a smarter control allocation policy following the
results of Khelassi et al. (2011) and the proportional
hazard model reported by Weber et al. (2012). The main
idea is to add to the process cost function a penalisation
on control actions, which is weighted with a matrix Wu
Rmm
that depends directly on actuators reliability. This
+
strategy leads to a smart use of actuators minimising the
frequency of unscheduled downtimes and related costs.
Consider that actuators reliability can be estimated
for the variable operating conditions with the following
modified exponential distribution:




(11)
ui 2 kt ,
Ri,k = exp 0i exp i 
with i Z[1,m] , where 0i R+ is the nominal failure
rate of the i-th actuator, i R+ is a shape parameter
of the actuator failure for an expected life tM Z+ , and
exp(i 
ui 2 ) R+ is the load function that modifies the
failure rate according to the root-mean-square (denoted by
u
i ) of the control actions applied from the initial time
until the time step k. From (11), it follows that the
cumulative probability of the failure rate can be written
as Fi,k = 1 Ri,k . Hence, the optimal control actions
can be distributed among actuators so that components
with larger accumulated damage are relieved. This can be
achieved by adding to the original economic cost function
a weighed term for the suppression of control moves, i.e.,
uk 2Wu,k , in which the weighing matrix is given by
Wu,k := diag (w1 , w2 , . . . , wm ) ,

(12)

where wi,k = Fi,k = 1 Ri,k for i Z[1,m] . Notice

that the weighing matrix is re-computed on-line at each
time step k to take into account the variation of the control
actions and actuators reliability. Hence, this weighing
strategy allows us to improve system availability, i.e., to
retain the operability of the network elements for longer
times.

3. Reliability-based economic MPC

problem
After discussing reliability aspects of storage and
supply infrastructure, next the setting of the proposed
reliability-based economic MPC controller is shown,
which incorporates into its optimisation problem both
the dynamic safety stock policy and the actuator-health
management policy, in order to improve the flow
supply service level in a given network, facing demand

uncertainty and equipment wear. The design of the

controller is based on Interpretation 1.
Interpretation 1 . (Sup-Inf type information) At any time
step k, when computing the corresponding controlled flow
uk , both the state xk and the demand (uncontrolled flow)
dk are known. Future demands dk+i are unknown for all
i Z+ , but forecast information of their first two moments (i.e., the expected value and the variance) is available for a given prediction horizon N Z+ . The controller has also knowledge of the current estimated accumulated degradation zk of the network actuators.
k =
Therefore, for a given demand sequence d
{dk+i|k }iZ[0,N 1] , estimated actuator degradation zk ,
acceptable risk levels x and z , and reliability-based
weight Wu,k , the proposed approach relies on solving the
following optimisation problem at each time step k:
min
x

N
1


[(k + i, xk+i|k , Wu,k uk+i|k )

uk ,k ,k

i=0

x
+ uk+i|k 2Wu,k + k+i|k
2Wx
z
+ k+i|k
2Wz ],

(13a)

subject to (10), (12),

xk+i+1|k = Axk+i|k + Buk+i|k + Bd dk+i|k ,

(13b)

zk+i+1|k = zk+i|k + |uk+i|k |,

Eu uk+i|k + Ed dk+i|k = 0,

(13c)
(13d)

G(j) (Axk+i|k + Buk+i|k )

(13e)

g(j)

xk,j (x )

x
k+i|k
,

uk+i|k = uk+i|k uk+i1|k ,

z(l),k+N |k zmax(l),k

zk,l (z )

(13f)
+

z
k+i|k
,

uk+i|k U,
x
k+i|k

0,

(13g)
(13h)

z
k+i|k

0,

(13i)

(xk|k , zk|k , uk1|k , dk|k ) = (xk , zk , uk1 , dk ),

(13j)

for all i Z[0,N 1] , j Z[1,rx ] and l Z[1,m] ,

x
}iZ[0,N 1]
where uk = {uk+i|k }iZ[0,N 1] , kx = {k+i|k
z
z
and k = {k+i|k }iZ[0,N 1] are the decision variables,
with uk being the sequence of controlled flows while
kx and kz sequences of slack variables introduced to
retain feasibility of the optimisation problem. Moreover,
dk+i|k is the forecasted demand for the i-step ahead of k.
Additionally, the terms


1 x
xk,j (x ) = FG1
(j) Bd dk+i
rx N
and


z 
zk,l (z ) = F1
1

(l)
mN

Unauthenticated
Download Date | 11/25/16 10:31 PM

Reliability-based economic model predictive control for generalised flow-based networks...

are the quantile functions involved in the stateand actuator-health deterministic equivalent constraints.
Weighing matrices Wx Rnn
and Wz Rmm
are
+
+
used to manage the penalisation of the slack variables
x
z
k+i|k
and k+i|k
, while Wu,k Rmm
is the
+
reliability-based weighing matrix introduced to relieve
the actuators with larger accumulated degradation. The
constraint (13j) represents the measurements available at
time step k.
Denote by (uk , kx , kz ) the optimal solution of (13)
at time step k. Then, following the MPC philosophy, only
the first optimal control action is applied, i.e., uk = uk|k .
Remark 3. The core of the proposed reliability-based
economic MPC approach relies on the dynamic handling
of constraints that allows a trade-off between reliability
and economic optimisation to obtain an enhanced
robust performance. Note that the worse the demand
forecasting and actuator degradation models, the stricter
the constraints and the more conservative control policy.
The proposed controller gives just an enhancement of
robustness, without guaranteeing robust feasibility and
stability. In particular, the authors have addressed the case
of economic recursive feasibility for periodic operation in
different works by means of periodic terminal equality or
inequality constraints; see preliminary results of Grosso
(2015) and Limon et al. (2014). Such references do not
include explicitly the reliability component, but it can
be incorporated in the recursively feasible schemes by
augmenting the state vector with the degradation state z.

4. Numerical results
In this section, the performance of the proposed
reliability-based economic MPC approach is assessed
with a case study consisting of a large-scale real system
reported by Ocampo-Martinez et al. (2009), specifically,
the Barcelona drinking water network (DWN). The
general role of this system is the spatial and temporal
re-allocation of water resources from both superficial (i.e.,
rivers) and underground water sources (i.e., wells) to
distribution nodes located all over the city. The structure
of this network (i.e., its directed graph G and the set
S of source and sink nodes) can be obtained from the
layout shown in Fig. 2 and its model in the form of (1)
can be derived by setting the state xk R63 as the
volume (in m3 ) of water stored in tanks at time step k,
the control input uk R114 as the flow rate through all
network actuators (expressed in m3 /s) and the measured
disturbance dk R88 as the flow rate of customer
demands (expressed in m3 /s). This network is currently
managed by AGBAR1 and supplies potable water to
the Metropolitan Area of Barcelona (Catalunya, Spain).
1

Aguas de Barcelona S.A. Company, which manages the drinking

water transport and distribution in Barcelona (Spain).

647

The main control task for managers is to economically

optimise the network flows while satisfying customer
demands. These demands are characterised by patterns of
water usage and can be forecasted by different methods,
(see, e.g., Billings and Jones, 2008; Sampathirao et al.,
2014).
In this way, the function  in (13a) is defined as
 := c
u,k Wu,k uk t and represents the economic cost of
network operation at each time step k, which depends on
the reliability-based weight Wu,k defined in (12) and on
a time-of-use pricing
scheme
driven by a time-varying


of the water flow,
price cu,k := c1 + c2,k R114
+
which in this application takes into account a fixed
water production/treatment price c1 R114
+ and a water
pumping price c2,k R114
+ . This latter price is time
dependant because it changes according to the electricity
tariff, which is assumed to be periodically time varying.
All prices are given in economic units per cubic meter
(e.u./m3) due to confidentiality reasons. The state and
input constraint sets for this case study are given by X =
{x R63 | xs,k x xmax } and U = {u R114 | 0
u umax }, respectively, where xs,k R63
+ is a desired
63
time-varying safety threshold, xmax R is the vector
of maximum storage capacity in tanks (expressed in m3 )
and umax R114 is the vector of maximum flow rates
of actuators (expressed in m3 /s). The prediction horizon
and the sampling time used in the simulations are N = 24
hours and t = 1 hour, respectively. The simulation
horizon was ns = 96 hours.
To analyse and highlight the benefits of the proposed
reliability-based economic MPC approach, a numeric
comparison with respect to baseline control strategies that
were previously reported for the same case study is shown
in Table 1. Specifically, the assessed approaches are the
following.
Certainty-equivalent economic MPC (CE-MPC). This
approach was proposed by Ocampo-Martinez et al.
(2009). It does not consider uncertainty explicitly in
the controller design and might require on-line tuning
to ensure an appropriate robust performance. In fact,
the common action to deal with demand uncertainty for
such an approach is to heuristically define a conservative
constant safety threshold xs,k = xmax for all k, with
(0, 1), and incorporate a constraint of the form
xk xs,k (or a softened version of it).
Chance-constrained economic MPC (CC-MPC). This
approach was proposed by Grosso et al. (2014). It
incorporates robustness only for demand uncertainty by
replacing the state deterministic constraints with chance
constraints. In this approach, every constraint that
involves random variables is dynamically managed by the
CC-MPC controller causing a back-off with respect to the
original hard constraints. The level of back-off is variable
and depends on the volatility of the forecasted demand at

Unauthenticated
Download Date | 11/25/16 10:31 PM

J.M. Grosso et al.

648
d1

d14

d15

c361CGY

c82PAL

d2

c450BEG

c369BEG

AportA

aMS

d23

x14

u1

d450BEG

d11

d369BEG
iBegues4

d3

d4

d5

C175LOR

C185VIL

c190SCL

d361CGY

iBegues3

u14

d263CES

c255BEG

d205FON

d18

iPalleja2

c374CGL

c313CGL

iPalleja4

x4

x3

x5

d374CGL

d268CGY
d175LOR

d185VIL

d255BEG

d190SCL

u11
iCescaplina2

d12
u4

u3

iOrioles

c268CGY

u16

d200CGY

d150SBO

d135VIL

d114SCL

d184BEG

d246CGY

d29
c320MGB

C135VIL

c114SCL

iCescaplina1

d7

d8

d9

c246CGY

d19

u12

C184ESP

d10

c252CGL

d20
u18

iCanGuey1d2

u20

iCanGuey1d5

d147SCC

u9

iStaClmCervello

d13

c140LLO

u7

iStBoi

d83

u100
x62

aPousCAST

u92

vGava100a80

aPousE3456

iMasJove
c115CAST

u129

vPousEstrella

u77

c175BVI

d88

iMontemar

d85

vCanRoca

u94

u99

c145MMA

iGava4

d115CAST
iCastelldefels8

d87

aCAST8

u95

iCastelldefels

x61
u101

x60

d80GAVi80
CAS85

d225GUI

u59

iGuinardera2

u53
vTerStaColoma

aPousB

c197GUI

iEtapBesos

vPsgStJoan

c120POM

u67
x55

d71

d197GUI

iBonavista

u64
d120POM

iMorera

x48

d63
u61

x53

u62

vConflent

N17
iGuinardera1

u88

d70BBE

c70FLL

u127

d73

d76

d78

c135SCG

x51
d190TCA

iTorreoCastell

iMntjcStaAmalia

x50
d144TPI

iMntjcTresPins

c55BAR

d69

x54

C171SAM

c190TCA

d67

d66

d117MTG

u126
c117MTG

c144TPI

vSJDTot

Reservoir

Demand
sector

d65

v117Montigala

d68

Source
(Superficial or
Underground)

d64

u63
vMontigala

u69

c70CFE
cRECARREGA

u65

d171SAM

u128

v55BAR

iSJD10

u104

x52

u68

c70BBE
v70CFE

d10COR

vTrinitat70

d77

Model Parameters Summary

LEGEND: Infrastructures

Last Update:
23 Mar, 2011
Juan Manuel Grosso
SAC IRI UPC

u60

d151BON

u66

x58

N12
iSJDSpf

x47

c151BON

x49

vTorrassa

vCerdaTraja

u87

x59
iSJD50

u103

d62

d75

vBesosStaColoma

u86

iCornella70

dPLANTA

AportLL2

d61
c225GUI

c100BES

N13

u102

d60
AportT

135MGA

d70

u125

c100BLLsc

iCornella50

iSJDSub

d202CRU

N15
vCncpcioArenal

u85

c202CRU

vTer

vBesosMontcCerd

u70
N14

v70LLFLL

u89

AportLL1

x46

u58

u50

vTrinitat100

vSifoTer

u52

u97
d84

d59
d58

u57

vTerMontcada

N16

u51

vZonaFranca

iSJD70

c70LLO

c101MIR

iCanRuti
vTrinitat130

d72
c100BLLnord

iCornella100

c169CME

d169CME
iTorreBaro1

x38
u71

u82
u81

x45

u56
N6

u55

vTerCarmel

u49

u72

u84

d54REL

u80

iLaSentiu

c130LSE

d86

d130BAR

d74

N11

iBellamar

u54

vTrinitat200

iRoquetes

iCornella130

vRossichMaq

u83
N10

d130LSE

x63

u79

x57

u46

c176BARcentre

iCollblanc

iRelleu

u98

vCerdanyola90

u47

d45
N4

u74

d82

u96

vCanyars

vBaroStLluis

d80

u73
d79

d90CERD

u48
c130BAR

vCornella100

c80GAVi80
CAS85

x44

d50
c200BARnord

N5

u114

u112

d100CFE

u78

iEstrella3456

u93

iStaMaMontcada

x37

u113

vPortola

vCollblanc

iPousCAST
d145MMA

u45

d200ALT

x32
u111

c100CFE

d184SMM

u44
iVallensana1

vEsplugues

u75
N9

c184SMM

d57

iCerdUAB

u41
vMinaCiutat

d44

iAltures

u106

iEstrella12

d215VALL

C215VALL

u124

u42

d46
c200ALT

u110

N7

iCerdSabadell

c150ALT

u130
vAltures

u36

vBonanova

iFinestrelles200

u107

x56

u91

d32
d300BAR

u109

u76

aPousE12

N3

iFinestrelles300

c100LLO
vSJD

d54

iCerdMontflorit

c250VASAB

iStGenis1

x31

iEsplugues

N8

u43

d55

iTorreBaro2

u40

u119

d176BARsud

d200BLL

u90

x42
u123

iFinestrelles176

iCanRoig

d197BET

u37
iCarmel
vHorta

c176BARsud

vFinestrllEsplg

d81

u122
iBoscVilaro

u121
iVallensana2

x43

u105

u21

d132CMF

x39

d42

iTibidabo

vFinestrelles

d31

vFontSanta

x21

c195Tor

u13

c147SCC

iUAB

d49

c176BARnord

u35

u108

d195TOR

d22

x13

iStCliment1

u8

iViladecans1

d437VVI

u120

c197BET

c200BARsc

c300BAR

d33

iCanGuell1

x33

u10
iBegues1

x29

d52
c260VALL

d260SGE

d43

d37

iMasGuimbau1

u32
iBellsoleig
c200BLL

d21

u19

x36

c275BEV

x41

u33

d200BSO

x20

d48
c238UAB

d53

c132CMF

iFlorMaig

d47
c250TBA

d255CAR

C437VVI

u34

x30
c200CGY
c150SBO

x28

d252CGL

x19

x12

x35

u118

d320MGB

d28
c110PAP

c200BSO
d205CES

iPapiol1

d110PAP

d34

c255CAR

iMasGuimbau2

u30

iPapiol2AGBAR

x18

c541TIB

u31

x25

x27

u17
iCanGuell2d3

iCanGuey2

x10

x9

d27

d41

iVallvidrera

d175PAP

c70PAL

u29

iCanGuell2d5

u15

iBegues2

iStCliment2

x8

x7

d36

u28

x17

d30

d56

u39

iStGenis2

c260SGE

c400MGB

x26

vPapiolATLL

c205CES

u6

u5

iViladecans2

vPalleja70

d313CGL

x16

x15

d16

N2

u38

d40

iMasGuimbau3
d400MGB

N1

x6

d35
c475MGB

c175i135PAP

u27

d200FDM

iPalleja3

u117

d26

u26

d328SGE

d320FON

iPalleja1

iMinaSeix

u25
vAbrera

d17

c200FDM

x40

x34

u116

u24

d125PAL
vPallejaATLL

iCanGuey3

x11

d38
c356FON

x24

u23

d51

c328SGE

c320FON

x23

u22

u115

d39

d25

c205FON

x22

c263CES

u2

d6

d24

c125PAL

x2

x1

Distribution
node

Pumping
station

63
114
130
88
17

Valve

STATE VARIABLES (x)

CONTROL ACTIONS (u)
FLOW ACTUATORS (u)
DISTURBANCES (d)
NODES (N)

Fig. 2. Barcelona DWN layout.

each prediction step. The approach relies on a prediction

model of the stochastic properties of disturbances, which
should be running in parallel with the MPC model. The
CC-MPC controller is not aware of the health of the
network actuators.
Reliability-based economic MPC with stationary uncertainty (RB-MPC). This approach was proposed by Grosso
et al. (2012). It uses the original output bounds, but
incorporates a dynamic state soft constraint to guarantee a
desired service level under demand uncertainty. In this
approach, the stochastic description of demands, used
to define the soft constraint, is computed a posteriori
before each MPC execution, based on the sample mean
and sample deviation of water demands. Uncertainty is
considered stationary within the MPC algorithm and, as
a consequence, the controller keeps a uniform back-off
of demand, whose amount represents the safety stock
along the prediction horizon. Additionally, this approach
incorporates also the actuator-health management policy
of Section 2.2, but using a deterministic actuator
degradation model.
Reliability-based economic MPC using chanceconstraints (RB-SMPC). This is the approach proposed

by this paper, which relies on solving the problem (13). It

considers non-stationary stochastic demand uncertainty
and stochastic actuator degradation. Hence, the base
stock constraint, the hard bounds of the states and the
terminal constraint of actuator degradation are in the form
of chance constraints (see Section 2).
The numeric assessment of the aforementioned
approaches is carried out through different key
performance indicators (KPIs), which are defined as
follows:
s
1 
c
u,k uk t,
ns + 1

KPIE :=

KPIU :=
KPIS :=

k=0
ns
m 


1
ns + 1

ns
n 



2
u(i),k ,

(14a)

(14b)

i=1 k=0


max 0, xs(i),k x(i),k ,

(14c)

i=1 k=0
s
1 
z(i),k ,
ns + 1 i=1

KPIZ :=

(14d)

k=0

Unauthenticated
Download Date | 11/25/16 10:31 PM

Reliability-based economic model predictive control for generalised flow-based networks...


ns

KPIV :=

649

Robust Constraints Handling

vk ,

(14e)

500
x

min,RBSMPC@80%

k=1

KPIO := topt,k ,

max,RBSMPC@80%

(14f)

xmin,RBSMPC@95%

400

x
x
x
x

300

Volume [m3]

where KPIE is the average economic performance of

the DWN operation, KPIU measures the smoothness
of the control actions, KPIS is the amount of water
used from safety stocks, KPIZ accounts for the average
degradation of actuators, KPIV measures the number
of safety constraint violations that have occurred during
the simulation, with vk being the number of tanks that
required the use of their safety stock at time step k, and
KPIO determines the difficulty to solve the optimisation
tasks involved in each strategy accounting topt,k as the
average time that takes to solve the corresponding MPC
optimisation problem. A lower KPI value represents a
better performance result. Simulations have been carried
out using = {80, 95}% for RB-MPC and =
{5, 20}% for both the CC-MPC and RB-SMPC (where
x = and z = ). In addition, Table 3 discloses
details of the production and operational costs related
to each strategy, which are the primary objectives for
DWN managers. Furthermore, Table 2 summarises the
capabilities handled by each controller. This qualitative
information complements the quantitative evaluation of
the assessed strategies in order to highlight the benefits
of the proposed RB-SMPC design.
An important aspect in any MPC controller is
the handling of constraints. In the Barcelona DWN,
manipulated variables can always be kept within bounds
by the controller, but output constraints, which are
subject to measured and/or unmeasured uncertainties,
must be properly handled. Since the baseline CE-MPC
approach relies on proper tuning of heuristic safety
stocks, its robustness and economic performance might
be compromised. Contrarily, the RB-MPC, CC-MPC
and RB-SMPC approaches focus on economic robust
performance of the DWN. They enhance the robustness
of the baseline CE-MPC by performing a dynamic
handling of constraints while keeping tractability of the
optimisation problems even for the large-scale model of
the case study. In particular, the RB-SMPC approach
proposed in this paper integrates the health-aware
capabilities of the RB-MPC approach with the stochastic
technique of the CC-MPC approach. Figure 3 shows the
mechanism that both RB-MPC and RB-SMPC approaches
use to guarantee a service level in the DWN and to
avoid the violation of real output constraints due to
uncertainty. The plot shows the response of both
controllers for a forecasted demand with confidence
levels of 80% and 95%.
Notice that both the
approaches dynamically generate a back-off of the
original constraints. An important observation regarding
the handling of constraints by both RB-MPC and
RB-SMPC controllers is the inherent relation between

max,RBSMPC@95%
s,RBSMPC@80%
s,RBSMPC@95%
s,RBMPC@80%

xs,RBMPC@95%
d

net

x
x

200

min,real
max,real

100

10

20

30

40

50

60

70

80

90

100

Time [h]

Fig. 3. Risk averse mechanism using the RB-MPC and RBSMPC approaches.

the service level in RB-MPC and the joint risk level

in RB-SMPC. Despite being defined under different
philosophies, both parameters represent a measure of
reliability for the DWN function. Nevertheless, Fig. 3
shows that the dynamic safety stock computed by the
RB-MPC controller is uniform and more conservative
than the one computed by RB-SMPC, which increases
according to the forecast error along the prediction
horizon.
This fact highlights the importance of a
suitable forecasting model and the effect of the explicit
propagation of uncertainty within the MPC model. In
general, decreasing the value of the service level, e.g.,
from 95% to 80% (equivalent to increasing the value of
the risk level from 5% to 20%), causes a reduction in the
safety stock and leads the base stock closer to the demand
pattern, which means that the probability of not achieving
the customer requirements increases due to the demand
uncertainty.
After reviewing the results in Tables 1 and 2, it can
be said that the robustness enhancements of the MPC
strategy proposed in this paper outperform the CE-MPC
controller in terms of reliability, i.e., CE-MPC may have
low values in most of KPIs but without any guarantee of
reliability and robust or probabilistic feasibility. Despite
having the lowest KPIS , the baseline CE-MPC approach is
the one that presents the highest number of soft constraint
violations, which means that the safety thresholds might
be overestimated (as observed in several tanks in the
DWN), causing more oscillations in the excursion of
water, or keeping states near the threshold with easiness
of activating the constraints in the controller. Therefore,
the baseline CE-MPC approach, with fixed and empirical
safety stocks, limits the economic optimisation. Instead,
the CC-MPC approach reached the lowest KPIE (in both
80% and 95% risk levels) by incorporating robust and

Unauthenticated
Download Date | 11/25/16 10:31 PM

J.M. Grosso et al.

650
Table 1. Comparison of the performance of controllers.
KPIE
KPIS
KPIU
KPIZ
KPIV
KPIO

Controller
CE-MPC
CC-MPC@5%
CC-MPC@20%
RB-MPC@95%
RB-MPC@80%
RB-SMPC@5%
RB-SMPC@20%

2442.97
2390.57
2362.64
2569.59
2560.72
2761.48
2560.36

0.18011
9421.46
710.22
3029.94
1625.29
3364.82
4946.13

0.8419
1.0223
1.1556
2.1023
2.0665
2.8664
2.2038

0.1374
0.1373
0.1374
0.1098
0.1187
0.1270
0.1076

2245
1822
1960
1699
1761
1710
1715

1.83
2.65
2.41
9.18
9.17
2.50
2.67

Simulation
Time
202.37
624.36
603.61
892.34
891.38
603.91
629.35

Table 2. Comparison of capabilities handled by each controller.

Controller
Dynamic
Dynamic
Actuator Smart
safety stocks output bounds
health
tuning
CE-MPC

CC-MPC

RB-MPC

RB-SMPC

: handled.
Table 3. Comparison of daily average costs of MPC strategies.
MPC approach Water average cost Electric average cost Daily average cost
(e.u./day)
(e.u./day)
(e.u./day)
CE-MPC
29037.21
29594.14
58631.35
CC-MPC
27706.72
29666.85
57373.58
RB-MPC
42072.97
19597.27
61670.25
RB-SMPC
53179.29
13096.23
66275.53
e.u.: economic units.

optimal safety stocks to face demand uncertainty with

minimum storage of water. This stochastic approach has
lower KPIV , i.e., it reduces the number of violations
of the base stocks but increases the amount of safety
stocks used to meet demands (higher KPIS ). This
is expected behaviour due to the policy of minimum
storage behind the computation of the base stocks, which
prefers using the safety stocks instead of keeping more
volume of water than required. The lower cost of water
in the Barcelona DWN (see Table 3), comparing both
the CE-MPC and CC-MPC approaches, reinforces this
observation. The main disadvantage of these cheaper
controllers is that control actions are computed based on
economic criteria, accounting for tanks reliability but not
for actuators reliability. This fact leads to higher values of
the KPIZ , i.e., the controllers overexploit those actuators
that have lower operational costs, accelerating their wear
and compromising the service reliability.
In order to manage the overall system reliability,
the RB-MPC and RB-SMPC controllers incorporate
actuator-health models and restrict their maximal
cumulative degradation at each time step to ensure
their proper functioning until a maintenance horizon
is reached. As seen in Table 3, the ability to compute

control actions for efficient management of actuator

reliability implies an important reduction in the electric
costs. This improvement is achieved at the expenses of
(i) an increment in KPIE due to the higher water cost,
(ii) an increment in KPIU due to the distribution of
control effort that avoids (if possible) constant control
actions that could cause an imbalance degradation of
actuators, and (iii) an increment in KPIS due to the
narrowing of constraints. It is important to point out
that the RB-MPC controller has greater KPIO than the
RB-SMPC one. The reason is that the former has to solve
a bi-level optimisation problem on-line, compared with
the RB-SMPC controller, which just requires solving
a single optimisation problem. Nonetheless, all the
compared controllers are suitable for real-time control
considering the sampling time in the DWN is one hour.
As can be expected, the RB-MPC and RB-SMPC
approaches have higher KPIE than the CC-MPC
approach.
The reason is that the inclusion of
actuator degradation constraints leads to control actions
that sacrifice (if necessary) economic performance in
order to guarantee the availability of actuators for a
given maintenance horizon. The rationing of actuator
degradation also leads to an increase in control action

Unauthenticated
Download Date | 11/25/16 10:31 PM

Reliability-based economic model predictive control for generalised flow-based networks...

Tank # 31

651

Tank # 36

6000

Volume [m3]

Volume [m3]

400
4000

2000

300
200
100

24

48

72

96

24

48

Tank # 49

48
Time [h]

72

96

xmin

xs

dnet

3000
Volume [m3]

Volume [m3]

96

Tank # 60

40
30
20

2000
1000

10
0

72

24

48
Time [h]

xCCMPC

72

96

xRBSMPC

xCEMPC

24

xmax

(a)
Actuator # 81

Actuator # 83
0.05

0.05

0.2
0.1

Flow [m3/h]

Flow [m3/h]

0.3

2
0

0.1
0

24

48

0.05
96

72

24

Actuator # 84

0.05
96

1.5
0

0.05

0.5
Flow [m3/h]

Flow [m3/h]

72

Actuator # 87
0.05

0.5
0

48

0.4
0.3

0.2
0.1

24

CCMPC

48
Time [h]

RBSMPC

0.05
96

72

CEMPC

24

max

48
Time [h]

min

72

0.05
96

Electric tariff pattern

(b)

Fig. 4. Operation of the Barcelona DWN with stochastic strategies: management of water storage with stochastic strategies (a), management of actuators with stochastic strategies (b).

smoothness (KPIU ), specially due to the operation

of pumps associated with Tank 55 to Tank 63 in the
bottom-right part of the DWN diagram (see Fig. 2).
With the CC-MPC approach, the volumes of water
in the aforementioned tanks are managed near the
safety constraints without complete replenishments, while
with the RB-SMPC approach the excursion of water
is periodic within the full range of operation. The
actuator-health management policy forces cycling the

operation of several pumps instead of keeping some of

them always active, and therefore requires exploiting
the full capacity of the related tanks. Furthermore, the
safety performance indicator (KPIS ) is drastically higher
in the CC-MPC approach; the reason is that the water
volume in tanks tends to keep longer time in the limit of
constraints, which leads to an increase in the frequency
of violation of safety thresholds. Figure 4 illustrates the
mentioned behaviour of the system. In general, chance

Unauthenticated
Download Date | 11/25/16 10:31 PM

J.M. Grosso et al.

652
3

constraints cause an optimal back-off from real constraints

as a risk-averse mechanism to face the non-stationary
uncertainty involved in the prediction of states.

x 10

z (81)

CCMPC

z (83)

In general, the proposed RB-SMPC approach

leads to a higher total closed-loop operational cost if
considering only the water and electric costs as indicators
for economic performance. This is the price to be paid
for enlarging the availability of the actuators by using the
proposed health-aware policy. Nevertheless, the economic
advantages of the RB-SMPC approach might be seen
when considering the long term operation, e.g., N 
k < M , where high corrective maintenance costs (due
to the possibly overexploitation and consequent failure
of actuators) could appear if the actuator-degradation
management policy is not considered. Therefore, the
RB-SMPC controller indirectly takes into account the
cost maintenance tasks by ensuring that the actuators will
be available until a pre-scheduled maintenance horizon
M , and consequently it might lead to a better long term
closed-loop economic performance.
Figure 5 shows the accumulated degradation of a
set of redundant actuators. Notice how the RB-SMPC
approach smartly decides to decrease the rate of
degradation of Actuator 87 (pump) by distributing the
control effort among the other three plotted actuators
(which are valves that have smaller coefficients of
degradation) according to their flow capacity. This
behaviour is equivalent to the one obtained with RB-MPC,
the difference being that the chance-constrained approach
narrows the maximum level of degradation allowed at
each time step according to the uncertainty in the health
prediction model of actuators. The wear process with
both the CE-MPC and CC-MPC approaches is neglected,
compromising the reliability of the supply infrastructure
even if safety stocks are optimally computed for a reliable
service.
Looking at the results discussed before, the
RB-SMPC strategy should be preferred given its
tractability for large-scale systems (as shown with the
Barcelona DWN case study) and ability to handle
probabilistic constraints related to demand and service
reliability.

CCMPC

z (84)

5
Accumulated Health Degradation (Z)

Table 3 details the water production and electricity

costs of each strategy. The CC-MPC approach has quite
similar costs to those of the baseline CE-MPC approach,
but with the benefit of better handling of constraints,
automatic computation of safety stocks and management
of risk near to the output bounds. On the other hand, the
RB-SMPC approach achieves a constant improvement in
electric costs, although at the expense of increasing stored
volumes of water (no matter how expensive the source
could be) and consequently water costs.

CCMPC

z (87)

CCMPC

z
4

(81)
RBSMPC

z (83)

RBSMPC

z
3

(84)
RBSMPC

z (87)

RBSMPC

10

20

30

40

50
Time [h]

60

70

80

90

100

Fig. 5. Degradation of a set of redundant actuators under

stochastic approaches.

5. Conclusions
In this paper, a reliability-based economic MPC
approach relying on chance-constrained programming
has been proposed to deal with the management
of generalised flow-based networks, considering both
demand uncertainty and actuator-health degradation.
The approach avoids relying on heuristic fixed safety
volumes such as those used in the CE-MPC or RB-MPC
schemes proposed in previous publications, which is
traduced in better robust economic performance. This
latter is achieved by incorporating dynamic planning of
safety stocks and actuators health monitoring, to assure
reliability in the flow supply and to minimise operational
costs for a given customer service level.
According to the results obtained with the case study
considered, the methodology is applicable to real-size
problems. The level of the resultant back-off is variable
and depends on the volatility of the forecasted demand
and actuator degradation at each prediction step as well
as the suitability of the probabilistic distributions used to
model uncertainties. The fact of unbounded disturbances
in the system precludes the guarantee of robust feasibility
with these schemes. Hence, the approach proposed
in this paper is based on a service-level guarantee and
probabilistic feasibility. Even when RB-SMPC increased
the operational costs by around 2.5%, it allowed improving service reliability by more than 90% when compared
with a baseline CE-MPC setting.
Future research will be focused on incorporation of
parametric uncertainty and unmeasured disturbances in
the model, in addition to deriving conditions for robust
feasibility and stability. From the economic point of
view, considering plant equipment depreciation and actua-

Unauthenticated
Download Date | 11/25/16 10:31 PM

Reliability-based economic model predictive control for generalised flow-based networks...

tor ageing models enriched with the effect of maintenance
quality and costs could be advantageous for network
management. Moreover, it is of interest to extend
the results and develop non-centralised stochastic MPC
controllers for large-scale complex flow networks.

Acknowledgment
The authors wish to thank the projects that funded this
research, i.e., the Spanish projects ECOCIS (CICYT
DPI2013-48243-C2-1-R) and HARCRICS (CICYT
DPI2014-58104-R), both of the Spanish Ministry of
Economy and Competitiveness.

References
Ahuja, R., Magnanti, T. and Orlin, J. (1993).
Network
Flows: Theory, Algorithms, and Applications, Prentice
Hall, Englewood Cliffs, NJ.
Betts, J. (2011). A robust approximation for setting target
inventory levels in a constrained production environment,
Procedia Computer Science 4: 12621271.
Billings, R. and Jones, C. (2008). Forecasting Urban Water
Demand, 2nd Edn., American Water Works Association,
Denver, CO.
Blanchini, F., Miani, S. and Ukovich, W. (2000). Control of
production-distribution systems with unknown inputs and
system failures, IEEE Transactions on Automatic Control
45(6): 10721081.
Blanchini, F., Rinaldi, F. and Ukovich, W. (1997). Least
inventory control of multistorage systems with
non-stochastic unknown inputs, IEEE Transactions
on Robotics and Automation 13(5): 633645.
Chamseddine, A., Theilliol, D., Sadeghzadeh, I., Zhang, Y.
and Weber, P. (2014). Optimal reliability design for
over-actuated systems based on the MIT rule: Application
to an octocopter helicopter testbed, Reliability Engineering
& System Safety 132: 196206.
Christopher, M. (2005). Logistics & Supply Chain Management: Creating Value-Adding Networks, 3rd Edn., Pearson
Education Limited, Dorchester.
Ellis, M., Durand, H. and Christofides, P. (2014). A tutorial
review of economic model predictive control methods,
Journal of Process Control 24(8): 11561178.
Ford, L.R. and Fulkerson, D.R. (1962). Flows in Networks,
Princeton University Press, Princeton, NJ.
Gallestey, E., Stothert, A., Antoine, M. and Morton, S. (2002).
Model predictive control and the optimization of power
plant load while considering lifetime consumption, IEEE
Transactions on Power Systems 17(1): 186191.
Gertsbakh, I. (2010). Reliability Theory: With Applications to
Preventive Maintenance, Springer, Berlin.
Goetschalckx, M. (2011). Advanced supply chain models,
in F. Hillier and C. Price (Eds.), Supply Chain Engineering, International Series in Operations Research &
Management Science, Vol. 161, Springer US, New York,
NY, pp. 615670.

653

Gorjian, N., Ma, L., Mittinty, M., Yarlagadda, P. and Sun, Y.

(2009). A review on degradation models in reliability
analysis, 4th World Congress on Engineering Asset Management, Athens, Greece, pp. 369384.
Grosso, J. (2015).
On Model Predictive Control
for Economic and Robust Operation of Generalized
Flow-based
Networks,
Ph.D.
thesis,
Universitat Politcnica de Catalunya, Barcelona,
http://hdl.handle.net/10803/288218.
Grosso, J., Ocampo-Martinez, C. and Puig, V. (2012). A service
reliability model predictive control with dynamic safety
stocks and actuators health monitoring for drinking water
networks, 51st IEEE Annual Conference on Decision and
Control (CDC), Maui, HI, USA, pp. 45684573.
Grosso, J., Ocampo-Martinez, C., Puig, V. and Joseph, B.
(2014). Chance-constrained model predictive control
for drinking water networks, Journal of Process Control
24(5): 504516.
Guida, M. and Giorgio, M. (1995). Reliability analysis of
accelerated life-test data from a repairable system, IEEE
Transactions on Reliability 44(2): 337346.
Guide, V. and Srivastava, R. (2000). A review of techniques for
buffering against uncertainty with MRP systems, Production Planning & Control 11(3): 223233.
Hsu, F., Vesely, W., Grove, E., Subudhi, M. and Samanta,
P. (1991).
Degradation modeling: Extensions and
applications, Technical report, Brookhaven National
Laboratory, Ridge, NY.
Kall, P. and Mayer, J. (2005). Stochastic Linear Programming, International Series in Operations Research &
Management Science, No. 80, Springer, New York, NY.
Kanet, J., Gorman, M. and Stlein, M. (2010). Dynamic
planned safety stocks in supply networks, International
Journal of Production Research 48(22): 68596880.
Khelassi, A., Theilliol, D., Weber, P. and Sauter, D. (2011).
A novel active fault tolerant control design with respect
to actuators reliability, 50th IEEE Conference on Decision and Control and European Control Conference (CDCECC), Orlando, FL, USA, pp. 22692274.
Kyriakides, E. and Polycarpou, M. (2015). Intelligent Monitoring, Control, and Security of Critical Infrastructure
Systems, Studies in Computational Intelligence, Vol. 565,
Springer Verlag, Berlin/Heidelberg.
Letot, C. and Dehombreux, P. (2012). Dynamic reliability
degradation based models and maintenance optimization,
Proc. 9th National Congress on Theoretical and Applied
Mechanics (NCTAM), Brussels, Belgium, pp. 19.
Limon, D., Pereira, M., de la Pea, D.M., Alamo, T. and
Grosso, J. (2014). Single-layer economic model predictive
control for periodic operation, Journal of Process Control
24(8): 12071224.
Martorell, S., Sanchez, A. and Serradell, V. (1999).
Age-dependent reliability model considering effects of
maintenance and working conditions, Reliability Engineering and System Safety 64(1): 1931.

Unauthenticated
Download Date | 11/25/16 10:31 PM

654

J.M. Grosso et al.

Negenborn, R. and Hellendoorn, H. (2010). Intelligence in

transportation infrastructures via model-based predictive
control, in R.R. Negenborn et al. (Eds.), Intelligent Infrastructures, Springer, Dordrecht, pp. 324.

Subramanian, K., Rawlings, J., Maravelias, C., Flores-Cerrillo,

J. and Megan, L. (2013). Integration of control theory and
scheduling methods for supply chain management, Computers & Chemical Engineering 51(0): 420.

Nemirovski, A. and Shapiro, A. (2006). Convex approximations

of chance constrained programs, SIAM Journal on Optimization 17(4): 969996.

Weber, P., Boussaid, B., Khelassi, A., Theilliol, D. and Aubrun,

C. (2012). Reconfigurable control design with integration
of a reference governor and reliability indicators, International Journal of Applied Mathematics and Computer Science 22(1): 139148, DOI: 10.2478/v10006-012-0010-0.

Ocampo-Martinez, C., Puig, V., Cembrano, G., Creus, R.

and Minoves, M. (2009). Improving water management
efficiency by using optimization-based control strategies:
The Barcelona case study, Water Science & Technology:
Water Supply 9(5): 565575.
Ortega, M. and Lin, L. (2004). Control theory applications to
the production-inventory problem: A review, International
Journal of Production Research 42(11): 23032322.
Osman, H. and Demirli, K. (2012).
Integrated safety
stock optimization for multiple sourced stockpoints facing
variable demand and lead time, International Journal of
Production Economics 135(1): 299307.
zer, . (2003). Replenishment strategies for distribution
systems under advance demand information, Management
Science 49(3): 255272.
Papageorgiou, L. (2009). Supply chain optimisation for the
process industries: Advances and opportunities, Computers & Chemical Engineering 33(12): 19311938.
Papageorgiou, M. (1984). Optimal control of generalized
flow networks, in P. Thoft-Christensen (Ed.), System Modelling and Optimization, Lecture Notes in
Control and Information Sciences, Vol. 59, Springer,
Berlin/Heidelberg, pp. 373382.
Pereira, E., Galvao, R. and Yoneyama, T. (2010). Model
predictive control using prognosis and health monitoring
of actuators, 2010 IEEE International Symposium on Industrial Electronics (ISIE), Bari, Italy, pp. 237243.
Sampathirao, A., Grosso, J., Sopasakis, P., Ocampo-Martinez,
C., Bemporad, A. and Puig, V. (2014). Water demand
forecasting for the optimal operation of large-scale
drinking water networks: The Barcelona case study,
19th IFAC World Congress, Cape Town, South Africa,
pp. 1045710462.

Juan M. Grosso received his mechatronics

and energy engineering degrees from Universidad Autnoma de Bucaramanga (UNAB) in
2008 and 2010, respectively. In 2012, he received his Masters degree in automatic control and robotics from Universitat Politcnica de
Catalunya (UPC). In 2015, he received his Ph.D.
degree in automatic control, vision and robotics,
also from UPC, working at Institut de Robtica i
Informtica Industrial (IRI), CSIC-UPC. His research interests include data analytics, systems identification, model predictive control, cyber-physical systems, non-centralized control, forecasting and fault-tolerant control.
Carlos Ocampo-Martinez received his electronics engineering degree and the M.Sc. degree
in industrial automation from the National University of Colombia (Manizales) in 2001 and
2003, respectively. In 2007, he received his Ph.D.
degree in control engineering from the Technical
University of Catalonia (UPC). In 20072008,
he held a postdoctoral position at the University
of Newcastle (Australia), and in 20082011 at
the Spanish Research Council (CSIC), Institut de
Robtica i Informtica Industrial (IRI), CSIC-UPC. Currently, he is an
associate professor at the UPC Automatic Control Department. Since
2014, he has also been the deputy director of IRI. His research interests
include constrained predictive control, large-scale systems and industrial
applications.

Schwartz, J. and Rivera, D. (2010). A process control approach

to tactical inventory management in production-inventory
systems, International Journal of Production Economics
125(1): 111124.

Vicen Puig received the telecommunications

engineering degree in 1993 and the Ph.D. degree
in automatic control, vision, and robotics in 1999,
both from Universitat Politcnica de Catalunya
(UPC). He is a professor at the Automatic Control Department and a researcher at Institut de
Robtica i Informtica Industrial, both at UPC.
He is the chair of the Automatic Control Department and the head of the research group on advanced control systems at UPC. He has developed important scientific contributions in the areas of fault diagnosis and
fault tolerant control, using interval and linear-parameter-varying models
using set-based approaches. He has participated in more than 20 European and national research projects in the last decade. He has also led
many private contracts with several companies and has published more
than 100 journal articles as well as over 350 contributions in international
conference/workshop proceedings. He has supervised over 15 Ph.D. dissertations and over 40 Masters theses/final projects. He is currently the
vice-chair of the IFAC Safeprocess TC Committee 6.4 (20142017). He
is the chair of the 3rd IEEE Conference on Control and Fault-Tolerant
Systems (SysTol 2016) and the IPC chair of the IFAC Safeprocess 2018.

Strijbosch, L., Syntetos, A., Boylan, J. and Janssen, E. (2011).

On the interaction between forecasting and stock control:
The case of non-stationary demand, International Journal
of Production Economics 133(1): 470480.

Received: 13 April 2015

Revised: 29 December 2015
Accepted: 16 February 2016

Sarimveis, H., Patrinos, P., Tarantilis, C. and Kiranoudis, C.

(2008). Dynamic modeling and control of supply chain
systems: A review, Computers & Operations Research
35(11): 35303561.
Schoenmeyr, T. and Graves, S. (2009). Strategic safety stocks
in supply chains with evolving forecasts, Manufacturing &
Service Operations Management 11(4): 657673.

Unauthenticated
Download Date | 11/25/16 10:31 PM