SICE Journal of Control, Measurement, and System

Integration

ISSN: 1882-4889 (Print) 1884-9970 (Online) Journal homepage: https://www.tandfonline.com/loi/tmsi20

Economic and Distributed Model Predictive

Control: Recent Developments in Optimization-
Based Control

Matthias A. Müller & Frank Allgöwer

To cite this article: Matthias A. Müller & Frank Allgöwer (2017) Economic and
Distributed Model Predictive Control: Recent Developments in Optimization-Based Control, SICE
Journal of Control, Measurement, and System Integration, 10:2, 39-52, DOI: 10.9746/jcmsi.10.39

To link to this article: https://doi.org/10.9746/jcmsi.10.39

Published online: 18 Jan 2021.

Submit your article to this journal

Article views: 77

View related articles

View Crossmark data

Citing articles: 20 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tmsi20
SICE Journal of Control, Measurement, and System Integration, Vol. 10, No. 2, pp. 039–052, March 2017

Economic and Distributed Model Predictive Control: Recent Developments

in Optimization-Based Control

Matthias A. MÜLLER ∗ and Frank ALLGÖWER ∗

Abstract : The goal of this paper is to give an overview of some recent developments in the field of model predictive
control. After a brief introduction to the basic concepts and available stability results, we in particular set our focus
on the areas of distributed and economic model predictive control, where more general control objectives than setpoint
stabilization are typically of interest.

Key Words : model predictive control, distributed MPC, economic MPC, nonlinear systems.

1. Introduction mization of the achievable profit in general. While the former

Model predictive control (MPC), also called receding hori-
zon control, is a very successful modern control technology. Its
basic idea is as follows: at each sampling instant, the future be-
havior of the system is predicted over some finite time horizon
using some prediction model, and an open-loop optimal control
problem is solved to determine the optimal input trajectory over
this horizon. Then, the first part of this optimal input is applied
to the system until the next sampling instant, at which the hori-
zon is shifted and the whole procedure is repeated again. The
main advantages of MPC and the reasons for its widespread
success include (i) guarantees for closed-loop satisfaction of
hard input and state constraints, (ii) the possibility to directly
include the optimization of some performance criterion in the
controller design, and (iii) its applicability to nonlinear systems
with possibly multiple inputs. In the last decades, a very rich
body of literature has emerged discussing both various theoret-
ical issues such as stability and robustness as well as successful
applications in many industrial fields, see, e.g., the books and
survey articles [1]–[9] and the references therein.
Besides a brief introduction to the basic concepts and avail-
able stability results in MPC (see Section 2), the main focus
of this review article are the areas of distributed and economic
model predictive control, which have both received a signif-
icant amount of attention in recent years. One unifying fea-
ture of the underlying control tasks in both these areas is that
typically more general objectives than the classical goal of set-
point stabilization are considered. For example, in a typical dis-
tributed cooperative control task, multiple dynamical systems
have to achieve some common cooperative goal such as multi-
vehicle platooning and formation stabilization, or distributed
power generation. In other situations, some general objective
is considered which is possibly related to the economics of the
system, such as maximization of a certain product in the pro-
cess industry, minimization of energy consumption, or maxi-
∗ control in general, and in particular to the areas of distributed
Institute for Systems Theory and Automatic Control, University
of Stuttgart, 70550 Stuttgart, Germany
E-mail: matthias.mueller@ist.uni-stuttgart.de,
frank.allgower@ist.uni-stuttgart.de
(Received November 28, 2016)
(Revised December 22, 2016)
setting can be handled within a suitable distributed MPC frame-
work, the latter is the distinctive feature of so-called economic
MPC schemes.
When considering large-scale systems, the development of a
single (centralized) model predictive controller for the overall
system is in general not possible or not desirable, since, e.g.,
such a central control entity having enough computational re-
sources is not available, such an approach is not scalable, or
the information/communication flow between the systems is re-
stricted to a certain (given) communication topology. Hence, in
distributed MPC, the main challenge is how the repeated online
optimization can be performed in a distributed fashion such that
for the overall closed-loop system, desired guarantees such as
constraint satisfaction, stability, or achievement of the cooper-
ative control objective can still be established. To this end, var-
ious solution approaches suitable for different system classes
and interconnection structures have been proposed; Section 3
provides an overview of some of these approaches.
As motivated above, in economic MPC some general cost
function is used for controller design, which is not necessarily
positive definite with respect to some setpoint to be stabilized.
This results in the fact that the closed-loop system might not
converge to some steady-state, since the optimal behavior of
the system can be more complex (e.g., periodic), depending on
the system dynamics, the cost function and the constraints. Im-
portant questions in such a setting are in particular the classifi-
cation of the optimal operating behavior as well as establishing
desired closed-loop performance and convergence guarantees.
These issues will be discussed in Section 4.
We close this section by noting that the goal of this article is
not to give an exhaustive overview of all the available results
in the above areas, which by the immense number of existing
publications is almost impossible. Rather, we attempt to sum-
marize the main ideas in a coherent fashion such as to provide
the reader a concise introduction to the field of model predictive
and economic MPC. The exposition will be done for discrete-
time systems, as is the case in most of the available results in
the literature. Nevertheless, the same (or similar) results are
typically also available for continuous-time systems.

JCMSI 0002/17/1002–0039 
c 2016 SICE
 40 SICE JCMSI, Vol. 10, No. 2, March 2017

1.1 Notation
Let I[a,b] denote the set of integers in the interval [a, b] ⊆ R,
and I≥a the set of integers greater than or equal to a. Boldface
symbols denote finite sequences v : I[0,N] → Rn for N ∈ I≥0 ,
or infinite sequences v : I≥0 → Rn , i.e., v = {v(0), . . . , v(N)} or
v = {v(0), v(1), . . . }, respectively.
2. MPC - Basic Principle and Stability
In this section, we give a brief introduction to the basic prin-
ciple of MPC and available stability results. In the following,
the system dynamics are given by a nonlinear difference equa-
tion of the form
x(t + 1) = f (x(t), u(t)), x(0) = x0 , (1)
where f : Rn × Rm → Rn , x(t) ∈ Rn and u(t) ∈ Rm are the
system state and the control input, respectively, at time t ∈ I≥0 ,
and x0 ∈ Rn is the initial condition. The system has to satisfy
pointwise-in-time state and input constraints
x(t) ∈ X, u(t) ∈ U (2)
for all t ∈ I≥0 , where X ⊆ Rn and U ⊆ Rm are the state and
input constraint sets, respectively1 . In order to define the model
predictive control law, at each time t ∈ I≥0 with measured state
x(t), for a given prediction horizon N ∈ I≥1 the following opti-
mization problem is solved:
Problem 1
minimize JN (x(t), u(t))
u(t)
subject to
x(k + 1|t) = f (x(k|t), u(k|t)), k ∈ I[0,N−1] , (3a)
x(0|t) = x(t), (3b)
x(k|t) ∈ X, u(k|t) ∈ U, k ∈ I[0,N−1] , (3c)
x(N|t) ∈ X ,f
(3d)
where

N−1
JN (x(t), u(t)) := (x(k|t), u(k|t)) + V f (x(N|t)). (4)
k=0
In this problem, u(t) := {u(0|t), . . . , u(N − 1|t)} and x(t) :=
{x(0|t), . . . , x(N|t)} are input and corresponding state sequences
predicted at time t over the prediction horizon N. The con-
straints (3) ensure that predicted states and inputs have to obey
the system model (1) and satisfy the given state and input con-
straints, and that the final predicted state is contained in the
terminal region X f ⊆ X. Furthermore, the stage cost function
 : X × U → R specifies the considered performance crite-
rion, and V f : X f → R is the terminal cost function. The
terminal constraint (3d) and the terminal cost V f are used in
various schemes in the literature in order to guarantee closed-
loop stability (cf. Subsection 2.1); other approaches omit these
additional ingredients (cf. Subsection 2.2), as is also typically
1
Instead of decoupled state and input constraints, also more gen-
eral (possibly coupled) constraints of the form (x(t), u(t)) ∈ Z ⊆
Rn × Rm for all t ∈ I≥0 can be considered.
done in industrial practice. Under some mild technical condi-
tions (such as continuity of the involved functions and com-
pactness/closedness of the involved sets), Problem 1 is either
infeasible or a (possibly non-unique) minimizer exists, which is
denoted by u0 (t) := {u0 (0|t), . . . , u0 (N − 1|t)} and in which case
x0 (t) := {x0 (0|t), . . . , x0 (N|t)} and JN0 (x(t)) := JN (x(t), u0 (t))
denote the corresponding optimal state sequence and optimal
value function, respectively. A standard model predictive con-
troller is now defined by the following algorithm.
Algorithm 1 Consider system (1). At each time t ∈ I≥0 , mea-
sure the state x(t), solve Problem 1 and apply the control input
u MPC (x(t)) := u0 (0|t).
Algorithm 1 results in the closed-loop system
x(t + 1) = f (x(t), u MPC (x(t))), x(0) = x0 . (5)
In the following, we discuss how desired guarantees for this
closed-loop system can be derived. The classical control objec-
tive, which is considered in most of the available MPC litera-
ture, is to asymptotically stabilize a given equilibrium setpoint
of system (1), which without loss of generality is assumed to
be the origin. In this case, a positive definite stage cost func-
tion  needs to be chosen, i.e., (0, 0) = 0 and (x, u) > 0 for all
(x, u)  0. However, in general the closed loop (5) is not neces-
sarily asymptotically stable if Problem 1 is not designed prop-
erly. In fact, it is rather easy to construct examples where this is
the case; for an experimental study of this fact, see, e.g., [10].
In the literature, two main approaches can be found how stabil-
ity in MPC can be established. One uses a suitable choice of
the terminal region and cost, while the other establishes stabil-
ity without these ingredients by a suitable (long enough) choice
of the prediction horizon N.
2.1 Stability in MPC with Additional (Terminal) Con-
straints
When using a framework with a terminal constraint (3d) and
a terminal cost function, the main assumptions to ensure stabil-
ity are as follows.
Assumption 1 There exists a local auxiliary control law u =
κ f (x) such that the following is satisfied inside the terminal re-
gion, i.e., for all x ∈ X f :
i) κ f (x) ∈ U,
ii) f (x, κ f (x)) ∈ X f ,
iii) V f ( f (x, κ f (x))) − V f (x) ≤ −(x, κ f (x)).
Assumption 1 means that the local auxiliary controller satis-
fies the input constraints (when applied inside the terminal re-
gion), that the terminal region is invariant under application of
the local auxiliary controller, and that the terminal cost serves
as a control Lyapunov function inside the terminal region. Note
that the auxiliary local controller is never applied to the system,
but is only needed in order to properly design the terminal re-
gion X f and the terminal cost V f . If Assumption 1 (and some
additional, nonrestrictive technical assumptions) are satisfied,
one can show that (i) Problem 1 is feasible for all t ∈ I≥0 if it is
initially feasible (i.e., for t = 0) and (ii) the origin is an asymp-
totically stable equilibrium for the closed-loop system (5). A
 SICE JCMSI, Vol. 10, No. 2, March 2017
proof of this result was first obtained in [11] for continuous-
time systems (compare also [3],[12]); for the discrete-time set-
ting, see, e.g., [2],[4],[5]. The survey paper [2] also gives a
good overview on the historical development of stability results
in MPC. The main idea in order to prove the above result on
recursive feasibility and asymptotic stability is to construct a
feasible candidate solution to Problem 1 at time t + 1 from
the optimal solution at time t. In particular, such a feasible
candidate solution ũ(t + 1) is given by appending the tail of
the previously optimal solution by the local control law, i.e.,
ũ(t + 1) = {u0 (1|t), . . . , u0 (N − 1|t), κ f (x0 (N|t))}. Feasibility of
this candidate solution follows from feasibility of the optimal
solution u0 (t) at time t and properties i) and ii) of Assumption 1.
Given this feasible candidate solution and the assumption of
initial feasibility, recursive feasibility can easily be established
by induction. Furthermore, one can use the optimal value func-
tion JN0 as a Lyapunov function for the closed-loop system (5).
In particular, the above defined feasible (but in general subop-
timal) candidate solution ũ results in a cost decrease given by
JN0 (x(t + 1)) ≤ JN (x(t + 1), ũ(t + 1)) ≤ JN0 (x(t)) − (x(t), u MPC (t)),
from which asymptotic stability can be concluded.
A constructive method how to compute X f , V f and κ f such
that Assumption 1 is satisfied was first proposed in [11] (for
continuous-time systems). This procedure is based on the as-
sumption that the linearization of system (1) at the origin is
stabilizable and considers the case where the stage cost func-
tion  is quadratic. The local auxiliary controller is chosen as a
linear control law κ f (x) = K x which stabilizes the linearization.
Then, a quadratic terminal cost V f is computed such that con-
dition iii) of Assumption 1 is satisfied for the linearized system
by solving a Lyapunov equation, and the terminal region X f
is chosen as a sublevel set of V f . Since the linearization error
consists of quadratic (and higher order) terms, one can show
that condition iii) of Assumption 1 is also satisfied for the orig-
inal nonlinear system if a small enough sublevel set is chosen
as terminal region. Furthermore, by assuming that 0 is con-
tained in the input constraint set U, it follows that condition i)
is also satisfied for X f small enough. Finally, condition ii) fol-
lows from satisfaction of iii) and the fact that X f is a sublevel
set of V f . A discrete-time version of this procedure can, e.g.,
be found in [4]; also, various different approaches and gener-
alizations to obtain possibly larger terminal sets are available,
see, e.g., [12]–[14].
2.2 Stability in MPC without Terminal Constraints
The second main approach to ensure stability omits the use
of a terminal constraint (3d) and a terminal cost in (4). In-
stead, closed-loop stability is typically shown by choosing the
prediction horizon N large enough, given that some controlla-
bility condition is satisfied. While early results showed that un-
der certain conditions, such a large enough prediction horizon
exists (see, e.g., [15],[16]), more recent publications also give
constructive (although possibly conservative) estimates for the
required prediction horizon length using a suitable controlla-
bility condition [5],[17]–[21]. This controllability condition is
typically formulated as the existence of a suitable upper bound
on the optimal value function:
Assumption 2 There exist L > 0 and numbers 0 ≤ LN ≤ L,
such that for all x ∈ X and all N ∈ I≥0 , the following inequality
41
holds:
JN0 (x) ≤ LN · inf (x, u). (6)
u∈U
Assumption 2 can, e.g., be verified using certain asymptotic
or exponential controllability conditions, although this can be
a difficult task in general. Using Assumption 2, one can
show that there exists a number α(N) such that the relaxed
dynamic programming inequality JN0 (x(t + 1)) ≤ JN0 (x(t)) −
α(N)(x(t), u MPC (t)) holds. In case that α(N) > 0, again asymp-
totic stability follows.
Furthermore, α(N) provides a suboptimality estimate for
the resulting MPC closed-loop cost compared to the in-
MPC
finite horizon optimal cost. Namely, let J∞ (x0 ) :=
∞
t=0 (x(t), u
MPC
(x(t))) denote the closed-loop cost resulting
from application of Algorithm 1 (with prediction horizon N),
and J∞ 0
(x0 ) be the infinite horizon optimal cost. Then from
the relaxed dynamic programming inequality it follows that
α(N)J∞ MPC
(x0 ) ≤ J∞0
(x0 ) for all x0 ∈ X. In [5],[20], a detailed
analysis can be found how α(N) can be obtained using Assump-
tion 2 (see also the continuous-time version [21]). In particular,
one can show that α(N) → 1 for N → ∞. This implies that (i)
closed-loop stability is guaranteed for a large enough prediction
horizon N (since for such N, α(N) > 0) and (ii) for N → ∞, the
MPC closed loop recovers infinite horizon optimality.
2.3 Discussion and Further Results
We now briefly discuss some of the advantages and disad-
vantages of the above described methods to establish stability
in MPC. The main benefits of the framework of Subsection 2.1
with terminal constraints and a terminal cost are that a shorter
prediction horizon than in a setting without terminal constraints
might be sufficient to ensure closed-loop stability, and that a
systematic procedure how to satisfy the crucial Assumption 1
is available for a large class of systems (in particular, in case
that the linearization is stabilizable as described above). Fur-
thermore, this assumption allows to establish recursive feasi-
bility of the repeatedly solved optimization problem as well as
closed-loop stability in a straightforward manner. On the other
hand, adding a terminal constraint might be restrictive and the
feasible set (i.e., the set of states for which a solution to Prob-
lem 1 exists) might be unnecessarily small, depending on the
size of the terminal region. Also, the computational complex-
ity is increased through the additional terminal constraint. The
approach of Subsection 2.2 does not exhibit these drawbacks
through omitting the use of a terminal constraint, resulting in a
simpler optimization problem and a possibly larger feasible re-
gion. On the other hand, recursive feasibility is not as straight-
forward, but requires additional assumptions and/or arguments,
see, e.g., [5],[15],[22]. Furthermore, the required controllabil-
ity condition (Assumption 2) can be difficult to verify, and for
general nonlinear systems, no systematic procedure exists to
this end. An additional advantage of MPC without terminal
constraints is that closed-loop performance estimates can be
given via the suboptimality index α(N). When using the set-
ting of Subsection 2.1, the establishment of such closed-loop
performance estimates is not as straightforward and in general
depends on properties of the terminal cost V f . Certain subopti-
mality estimates can be established if the terminal cost is a good
approximation of the infinite horizon optimal cost, see, e.g., [5].
 42 SICE JCMSI, Vol. 10, No. 2, March 2017
In conclusion, each of the two presented schemes has its advan-
tages and disadvantages, and the choice of a suitable stabilizing
MPC scheme depends on the considered application.
An alternative to the two stabilizing MPC approaches de-
scribed above are Lyapunov-based MPC methods (see, e.g.,
[23]). Here, a (global) control Lyapunov function is assumed
to be known, and instead of terminal constraints, a constraint
requiring a certain decrease of the control Lyapunov function
is included into Problem 1. Furthermore, the basic stabilizing
MPC approaches discussed in this section have been extended
in many different ways. For example, output feedback MPC
schemes, the stabilization of more general sets and trajectory
tracking problems have been studied, as well as MPC schemes
for different system classes such as hybrid, periodic, or time-
delay systems. Also, important issues concerning robustness in
the presence of uncertainties and disturbances have been inves-
tigated both in a deterministic and a stochastic framework. All
these areas are currently still very active fields of research, and
it is well beyond the scope of this article to discuss all these
approaches in detail; a recent survey summarizing some of the
existing results (with a particular focus on MPC for uncertain
systems) can be found in [9].
3. Distributed MPC
When considering large-scale systems, the development of
a single (centralized) model predictive controller, i.e., repeat-
edly solving Problem 1 for the overall system, is in general
not possible or not desirable. This can be due to several rea-
sons. For example, such a central entity having enough compu-
tational power might not be available. In other situations, the
communication infrastructure is such that information can only
be shared locally between subsystems, but not with some cen-
tral coordinating unit. Furthermore, in a competitive setting,
systems might not want to share all available information with
everyone due to privacy issues. In such cases, distributed MPC
schemes are of interest. Here, the main challenge is how to de-
sign suitable optimization problems (similar to Problem 1) and
how to develop suitable algorithms for a distributed solution,
such that guarantees for the overall closed-loop system can be
given.
In such a distributed setting, coupling/interactions between
subsystems can be of different nature. First, physical couplings
can exist, i.e., the dynamics of the individual subsystems are
coupled. This means that the dynamics of the individual sub-
systems depend on other subsystems’ states and/or inputs, i.e.,
the dynamics of each subsystem i is given by
xi (t + 1) = fi (x(t), u(t)), (7)
where x denotes the state of the overall system x = [x1T . . . xTP ]T
and P is the number of subsystems, and similarly u denotes the
input of the overall system. Such couplings appear in many
applications such as, e.g., large chemical plants consisting of
multiple interconnected reactors, in electrical grids, or in cou-
pled mechanical systems. In other settings, the overall system
is composed of a team of dynamically independent subsystems
whose dynamics only depend on the own system state and in-
put, i.e.,
xi (t + 1) = fi (xi (t), ui (t)). (8)
This is typically the case in multi-agent settings such as multi-
robot coordination or formation flight. Second, couplings be-
tween the subsystems can stem from coupling constraints. Typ-
ical examples include collision avoidance or connectivity main-
tenance constraints. Third, a common objective is another
source of coupling between subsystems, resulting in couplings
in the cost functions Ji,N of the individual subsystems. In the
above mentioned examples, this could be a system-wide perfor-
mance optimization of a chemical plant or a certain cooperative
task such as formation flight. Finally, the available communi-
cation topology, i.e., which systems can exchange information
with each other, is crucial for the selection of a suitable dis-
tributed MPC scheme. In general, the graph representing this
communication topology need not be the same as the one result-
ing from the three sources of couplings between the subsystems
discussed above. However, in many of the available results, it
is assumed that each system can (at least) communicate with
every other system to which it is coupled. A distinctive fea-
ture of distributed MPC compared to other distributed control
methods is that the information, which is exchanged between
subsystems, typically consists of predicted state and/or input
trajectories (compared to, e.g., only the current state), which
allows satisfaction of constraints despite couplings and allows
for improved performance.
The existing types of coupling between the subsystems as
well as the available communication infrastructure are crucial
for designing suitable optimization problems (similar to Prob-
lem 1) for each of the subsystems and for developing suitable
algorithms for a distributed solution. In recent years, a large
body of literature with hundreds of papers has been developed
for various different settings including different types of sys-
tem dynamics, couplings, and control objectives and using dif-
ferent distributed solution methods, see, e.g., the survey arti-
cles [24],[25] and the edited volume [26], which presents 35
distributed MPC approaches using a unified notation. In the
following, we do not attempt to provide an exhaustive survey
of all available distributed MPC results, but rather give a brief
overview of how the different approaches can be classified in
terms of the employed solution strategies (see Subsection 3.1).
Furthermore, in contrast to the above mentioned existing sur-
veys, we put a particular focus on distributed MPC schemes
which have been developed for more general cooperative con-
trol tasks than setpoint stabilization (see Subsection 3.2).
3.1 Solution Approaches in Distributed MPC
One possibility to classify distributed MPC schemes is ac-
cording to the employed solution strategy, i.e., how the dis-
tributed MPC problem is formulated and solved, and what
the main distinctive feature is regarding the establishment of
closed-loop guarantees. In the following, we discuss four dif-
ferent classes, to which most of the existing distributed MPC
schemes can be allocated. First, we consider iterative schemes,
where some sort of distributed optimization algorithm is used.
After that, in non-iterative schemes, we distinguish between se-
quential approaches, schemes based on robustness considera-
tions, and methods using additional consistency constraints. Fi-
nally, we discuss various advantages and disadvantages of the
different schemes.
Iterative methods. Many distributed MPC schemes use iter-
ative methods. Here, in each time step the optimization prob-
 SICE JCMSI, Vol. 10, No. 2, March 2017
lem for the overall system, i.e., Problem 1, is solved by some
sort of distributed optimization algorithm. This means that at
each time t, all subsystems iteratively solve an optimization
problem (in parallel) and exchange information multiple times
such that the solutions converge to the solution of the central-
ized, overall optimization problem. Distributed MPC schemes
belonging to this category can, e.g., be found in [27]–[29] using
a Jacobi method or in [30],[31] based on dual decomposition.
The alternating direction method of multipliers (ADMM) was
employed in [32], whereas in [33] and [34] the authors use a
sensitivity-based coordination scheme and a block-coordinate
descend method, respectively. In [35], the authors present
a framework for the distributed synthesis of suitable termi-
nal regions and terminal cost functions, which can then be
used for distributed MPC using suitable distributed optimiza-
tion algorithms. Finally, Lyapunov-based techniques are used
in [36],[37].
Most of the existing iterative distributed MPC schemes are
designed for linear systems and the obtained results build on
convexity of the problem (which is in particular needed for
convergence of the distributed optimization algorithms); there
are few exceptions which also treat nonlinear systems such
as [28],[36]. Some of the available schemes need the require-
ment that the distributed optimization algorithm converges at
each time t, which might be impractical since a large number
of iterations might be necessary. Other schemes provide fea-
sibility and stability already for a finite number of iterations,
some even for the case of only one iteration. Ensuring satis-
faction of coupling constraints is typically difficult, at least in
case of a finite number of iterations; a notable exception can be
found in [31].
Sequential methods. A different approach to distributed
MPC are sequential methods. Here, the distinctive feature is
that the subsystems do not solve their optimization problems
in parallel but sequentially, i.e., only one subsystem optimizes
while all subsystems keep their predicted trajectories fixed. Us-
ing such a strategy allows to establish feasibility and fulfillment
of coupling constraints in a rather straightforward fashion. A
first sequential distributed MPC scheme of this type appeared
in [38], where only couplings due to constraints were consid-
ered. These ideas were subsequently extended to more general
settings, including other cooperative control tasks than stabi-
lization and uncertain system dynamics, see, e.g., [39]–[43]. A
benefit of such sequential schemes is that they easily carry over
to nonlinear systems and also coupling constraints can be taken
into account in a straightforward fashion. On the other hand,
such approaches are typically not well scalable if all subsystems
solve their optimization problem at each time t in a sequential
order, or the performance might be compromised if only one (or
few) subsystems are allowed to optimize at each time t, while
the other subsystems follow their previously computed trajec-
tories. Depending on the interconnection structure, this draw-
back can be overcome since subsystems which are not coupled
to each other can optimize in parallel without compromising
the theoretical feasibility and stability guarantees. The above
schemes were developed for settings where no couplings in the
dynamics are present. If this is the case, further negotiation be-
tween the systems [44] or a special system structure [36],[45]
are necessary in order to establish closed-loop guarantees.
43
Approaches based on robustness considerations. Another
class of distributed MPC schemes is based on robustness con-
siderations. Here, the influences of neighboring subsystems are
(partially) treated as unknown disturbances, and certain con-
cepts for the analysis and controller design of uncertain sys-
tems are employed either within the analysis of the resulting
closed-loop system, or already in the design of the optimiza-
tion problems in the distributed MPC scheme. For example,
in [46] the authors show for nonlinear subsystems coupled by
a common objective that each subsystem is input-to-state sta-
ble (ISS) with respect to the “disturbance” being composed
of the influences of neighboring subsystems, and by a small
gain argument that in turn also the overall closed-loop system
is asymptotically stable. In [47], a tube-based robust MPC ap-
proach is utilized. Each subsystem assumes that trajectories of
neighboring subsystems are contained within a tube around a
communicated nominal trajectory, and ensures that its own tra-
jectory is contained in such a tube. This design can be used
in order to show asymptotic stability of the resulting overall
closed-loop system. In the distributed MPC scheme presented
in [48], subsystems exchange a sequence of sets (called “con-
tracts”), in which predicted trajectories are guaranteed to be
contained. For the resulting closed-loop system, ISS with re-
spect to the “contract size” can be established. While in the
above references, predicted trajectories (and/or tubes contain-
ing them) are exchanged between the subsystems, the idea of
treating influences of neighboring subsystems as disturbances
can also be employed in an even simpler, yet more conserva-
tive, setting, where no communication between the subsystems
takes place. Such schemes are typically referred to as decen-
tralized MPC schemes, see, e.g., [49],[50], where min-max or
again tube-based MPC techniques, respectively, are employed.
Methods employing consistency constraints. Finally, some
distributed MPC approaches impose additional constraints on
predicted state and/or input sequences in order to establish de-
sired closed-loop properties. One typical form of such con-
straints is to require predicted trajectories to be close enough to
previously communicated ones, see, e.g., [51]–[53]. This en-
sures that at each time t, the new optimal trajectory computed
by each subsystem i is close enough to what the other subsys-
tems assumed for this subsystem, which is why such constraints
are sometimes called “consistency constraints” or “compatibil-
ity constraints” in the literature. A similar approach was taken
in [54], where overall closed-loop stability was shown if a cer-
tain compatibility condition is satisfied. A different approach
is taken in [55] and [56] for systems only coupled via a com-
mon objective, where a constraint tightening on predicted in-
puts and states, respectively, is employed in order to obtain de-
sired closed-loop guarantees.
Discussion. We now briefly compare the above distin-
guished classes of distributed MPC schemes. In general, the
best system-wide performance will be obtained by iterative
schemes. As indicated above, for enough iterations the per-
formance will converge to that of a centralized MPC approach;
this is typically not the case in the other distributed MPC frame-
works. Hence if both enough communication and computa-
tional capacity are available for possibly multiple iterations per
time instant, iterative schemes can be well suited. On the other
hand, they might not be applicable when considering nonlin-
ear systems or if coupling constraints are present. Sequen-
 44 SICE JCMSI, Vol. 10, No. 2, March 2017
tial schemes have the advantage that possibly less communi-
cation (typically only once per time instant) is needed, and also
coupling constraints as well as nonlinear system dynamics can
be taken into account in a fairly straightforward fashion. On
the other hand, couplings in the dynamics might be difficult
to consider, and there is also a potential scalability problem
as discussed above. Approaches based on robustness consid-
eration also need less communication than iterative schemes,
and compared to sequential schemes, they might be better scal-
able (depending on the system structure) and also couplings
in the system dynamics can be considered. However, such
schemes can be quite conservative and hence might only work
if the coupling strength is rather weak. Similar considerations
also hold for methods based on additional (consistency) con-
straints. Namely, while they potentially share similar advan-
tages as robustness-based approaches compared to iterative and
sequential schemes, the degrees of freedom within the opti-
mization might be quite small, depending on the exact formu-
lation of the consistency constraints, resulting in a quite con-
servative solution with diminished performance. In summary,
each of the different classes of distributed MPC schemes has its
own advantages and disadvantages, and the choice of a suitable
distributed MPC algorithm heavily depends on the properties
of the considered application, i.e., on the considered system
class, the types of couplings, and the available computational
resources and communication infrastructure.
3.2 Distributed MPC for General Cooperative Control
Tasks
The vast majority of existing distributed MPC schemes, in-
cluding most of those referenced above, consider the classical
control objective of stabilizing the overall system at some (a
priori given) setpoint. However, when considering typical co-
operative control problems, often different, more general, con-
trol objectives are of interest. This is in particular the case in
settings where a team of dynamically independent subsystems
has to pursue a common, cooperative task such as vehicle pla-
tooning or formation flight, corresponding to couplings in con-
straints and cost functions. Cooperative control tasks of par-
ticular importance, which have received significant attention
in recent years, are consensus and synchronization problems,
where the systems have to agree on an a priori unknown state
(or output) value or trajectory. These type of problems appear in
many applications such as multi-vehicle platooning and forma-
tion stabilization, distributed sensor fusion, the synchronization
of coupled oscillators and others (see, e.g., [57]). When con-
sidering such general cooperative control objectives, typically
different methods have to be used for the MPC design and/or
closed-loop analysis compared to the case where a setpoint (or
trajectory) to be stabilized is known a priori. For example, the
design of suitable terminal cost functions and terminal regions
has to be suitably adapted to such a setting. Furthermore, while
some approaches still use the optimal value function as a Lya-
punov function for convergence analysis, others rather exploit
certain geometric properties of the resulting closed-loop dy-
namics. In the following, we briefly review some distributed
MPC schemes which were specifically developed for such more
general control objectives than setpoint stabilization.
In [58], consensus for single and double integrators subject
to input constraints is considered, exploiting certain geometric
properties of the system trajectories resulting from the rather
simple dynamics. Reaching consensus via distributed MPC is
also considered in [59] for single integrators, for double inte-
grators [60], for scalar linear systems [61], and for general lin-
ear systems [62]. Here, the closed-loop analysis is based on the
explicit solution to the respective optimization problems, which
is characterized for the case that no constraints are present. A
similar approach is taken in [63], where consensus for double-
integrators is shown in the presence of input constraints which
are active only a finite number of times, and in [64], where
flocking of double-integrators is considered. In [65], for gen-
eral linear systems, the authors calculate an optimal consensus
point at each time step by iteratively solving a centralized op-
timization problem, where this point is used as setpoint in the
MPC formulation. A theoretical analysis of this scheme was
done in [66], where convergence to a common consensus point
corresponding to an equilibrium of the system is established,
both for the case where the negotiations about the optimal con-
sensus point have converged in each time step, as well as when
this negotiation is terminated prior to convergence. In [67], a
consensus problem is solved for linear systems using standard
methods, and then the resulting solution is used as a reference
signal in a tracking MPC framework. Furthermore, in [68] the
authors propose a robust distributed MPC scheme for linear
time-varying uncertain systems to achieve practical consensus.
The sequential distributed MPC scheme in [40] is designed
for general cooperative control tasks which can be formulated
as the stabilization of some (possibly non-compact) set, which
includes consensus/synchronization as a special case. Conver-
gence to the desired set, i.e., fulfillment of the cooperative con-
trol task, is shown for nonlinear systems including coupling
constraints under suitable assumptions on the terminal regions
and terminal cost functions. Furthermore, it is shown how these
conditions can be satisfied for the special cooperative control
task of consensus/synchronization. Distributed MPC for vehi-
cle platooning is considered in [69]. Stability and string sta-
bility for the resulting closed-loop system is achieved by pe-
nalizing deviations from previously communicated trajectories
in the cost function. In [70], the authors consider distributed
MPC for two single-integrator systems with a possibly com-
petitive objective, such as different desired distances to each
other. Finally, also the framework of economic MPC (compare
Section 4), where some general performance criterion is con-
sidered, can be exploited for certain cooperative control prob-
lems. To this end, recently some first distributed economic
MPC schemes have been proposed in the literature, which will
be discussed at the end of Section 4.
4. Economic MPC
Economic MPC is a variant of MPC where compared to the
classical stabilizing MPC setting as in Section 2, some more
general performance criterion is considered. The motivation
(and also the naming) for this more general MPC formulation
came from the process industry. There, the classical approach
is to apply a two-layer control structure, where at the upper
layer the so-called real-time optimization (RTO) determines
economically optimal setpoints or trajectories for the system,
which are then tracked by a controller on the lower level, see,
e.g., [71]–[74] and the references therein. However, it was
noticed that the performance can possibly be significantly im-
 SICE JCMSI, Vol. 10, No. 2, March 2017
proved if the economic cost function is directly incorporated
into the lower level MPC controller [73],[75],[76]. On a tech-
nical level, the only difference of economic MPC compared to
standard stabilizing MPC is that the stage cost function  in
Problem 1 is not chosen to be positive definite with respect
to an a priori given setpoint as in Section 2, but can be some
general function (possibly related to the economics of the con-
sidered system). This results in the fact that the closed-loop
system might not converge to some equilibrium setpoint, since
other trajectories (e.g., periodic) exist resulting in a better per-
formance. Important questions in this context are (i) how to
classify the optimal operating behavior, (ii) whether the closed-
loop system “finds” the optimal operating behavior (i.e., con-
verges there), and (iii) what statements about closed-loop per-
formance can be made. These issues will be discussed in the
following subsections.
4.1 Classifying Optimal Operating Behaviors
In general, the optimal operating behavior depends on the
given system dynamics f , the stage cost function , and the in-
put and state constraint sets U and X. The most simple case
is when steady-state operation is optimal, i.e., the optimal op-
erating behavior is to stay at the best steady-state. This is the
best studied case in the literature, which we also first discuss
in the following. We then mention possible extensions to more
general cases (such as optimal periodic operation).
Let Z0 denote the largest forward invariant set contained
in X × U, i.e., the set which contains all elements in X × U
which are part of a feasible state/input sequence pair:
 trollability conditions, necessity of dissipativity for optimal
Z0 := (x, u) ∈ X × U : ∃v : I≥0 → U s.t.
(z(0), v(0)) = (x, u), z(t + 1) = f (z(t), v(t)),
 with respect to the supply rate s(x, u) = (x, u) − (x∗ , u∗ ) on the
(z(t), v(t)) ∈ X × U ∀t ∈ I≥0 , ⊆ X × U. (9)
Furthermore, let S be defined as the set of all feasible
state/input equilibrium pairs of system (1), i.e.,
S := {(x, u) : x ∈ X, u ∈ U, x = f (x, u)} ⊆ Z0 , (10)
which is assumed to be non-empty. Under some mild technical
conditions (such as, e.g., continuity of f and  and compact-
ness of X and U), existence of a (possibly non-unique) optimal
state/input equilibrium pair (x∗ , u∗ ) is guaranteed, i.e., (x∗ , u∗ )
satisfies
(x∗ , u∗ ) = inf (x, u). (11)
(x,u)∈S
Optimal steady-state operation can now been defined as fol-
lows [75],[77].
Definition 3 The system (1) is optimally operated at steady-
state, if for each solution satisfying (x(t), u(t)) ∈ X × U for all
t ∈ I≥0 the following holds:
T
(x(t), u(t))
lim inf t=0
≥ (x∗ , u∗ ). (12)
T →∞ T +1
The definition of optimal operation at steady-state is such that
no feasible solution to system (1) leads to an average perfor-
mance (measured in terms of the stage cost ) which is better
than the optimal steady-state cost (x∗ , u∗ ), which is the average
45
cost when operating the system at the optimal steady-state x∗ .
In order to study optimal steady-state operation of a system, a
certain dissipativity condition has turned out to play a crucial
role. The concept of dissipativity dates back to Willems [78]
(see also [79] for a discrete time version) and is as follows.
Definition 4 The system (1) is dissipative with respect to the
supply rate s : Rn ×Rm → R if there exists a storage function λ :
Rn → R≥0 such that the following inequality is satisfied for all
(x, u) ∈ Rn × Rm :
λ( f (x, u)) − λ(x) ≤ s(x, u). (13)
If there exists ρ ∈ K∞ such that for all (x, u) ∈ Rn × Rm
λ( f (x, u)) − λ(x) ≤ −ρ(|x − x∗ |) + s(x, u), (14)
then system (1) is strictly dissipative.
If the dissipation inequality (13) is only satisfied on some
set W ⊆ Rn × Rm , we say that system (1) is dissipative on W.
Given the above notation and definitions, it was shown in [75]
(see also [80] for a continuous-time version) that if system (1)
is dissipative on Z0 with respect to the supply rate s(x, u) =
(x, u) − (x∗ , u∗ ), then it is optimally operated at steady-state2 .
Furthermore, if system (1) is strictly dissipative with respect
to this supply rate, a slightly stronger property than optimal
steady-state operation is satisfied, which was termed (uniform)
suboptimal operation off steady-state [75],[81]. In the most
general case, the converse result is not true, which was shown
by two counterexamples in [82]. However, under suitable con-
steady-state operation can indeed be established. Namely, if
system (1) is optimally operated at steady-state, it is dissipative
intersection of the M-step controllable and M-step reachable
set, for arbitrary M ∈ I≥1 (cf. [83] for a more precise statement
of this result). Furthermore, if system (1) is uniformly subop-
timally operated off steady-state and locally controllable at x∗ ,
then it is strictly dissipative on Z0 [81]. In summary, dissipa-
tivity with respect to the supply rate s(x, u) = (x, u) − (x∗ , u∗ )
serves as an (almost) equivalent characterization of the property
of optimal steady-state operation.
The more general case where the system is optimally op-
erated at a periodic orbit can be treated in a similar fashion.
Namely, one can define optimal operation of system (1) at a P-
periodic orbit Π analogously to Definition 3, but with (x∗ , u∗ )
on the right hand side of (12) replaced by the average cost
of the P-periodic orbit Π. Then, a suitable cyclic dissipativ-
ity condition or a dissipativity condition for a suitably defined
P-step system can be used as a sufficient and necessary con-
dition (the latter again under a suitable controllability assump-
tion) for optimal P-periodic operation [81],[84]–[86]. Further-
more, a (partial) generalization of the above results to the case
of time-varying systems and cost functions is available [87],
where again a suitable dissipativity condition is used as a suf-
ficient condition to establish optimal operation of the system at
some optimal time-varying trajectory.
2
We remark that in the stabilizing MPC setting of Section 2, i.e.,
with positive definite stage cost , strict dissipativity with re-
spect to the supply rate s(x, u) = (x, u) − (x∗ , u∗ ) trivially fol-
lows with λ ≡ 0.
 46 SICE JCMSI, Vol. 10, No. 2, March 2017
4.2 Closed-Loop Convergence
Having established a systems-theoretic characterization of
optimal operating behaviors, we now discuss whether the
closed-loop system (5) resulting from application of an eco-
nomic MPC scheme (Algorithm 1) converges to the optimal op-
erating behavior. Again, the exposition follows the basic case
where steady-state operation is optimal, and we subsequently
discuss available extensions to more general cases.
In case that steady-state operation is optimal, it is desirable
that the closed-loop system converges to the optimal steady-
state x∗ . Again, different economic MPC schemes are available
in the literature, in particular again with and without an ad-
ditional terminal constraint and/or terminal cost in Problem 1.
The main difficulty in both approaches is the following: since
the stage cost  is in general not positive definite with re-
spect to x∗ , the optimal value function JN0 might increase along
closed-loop trajectories and can therefore not be used as a Lya-
punov function, as was done in Section 2. Convergence to the
optimal steady-state was first established in [88] for linear sys-
tems and strictly convex cost functions, by using convexity ar-
guments. A major step forward in the general nonlinear case
was done in [89] under a strong duality condition, which was
later generalized in [75] to a strict dissipativity condition, both
using a terminal equality constraint (i.e., X f = {x∗ }). Namely,
in these references the rotated cost function
L(x, u) := (x, u) − (x∗ , u∗ ) + λ(x) − λ( f (x, u)) (15)
is considered. Under the assumption that system (1) is strictly
dissipative with respect to the supply rate s(x, u) = (x, u) −
(x∗ , u∗ ), from (14) it directly follows that the rotated stage
cost L is positive definite with respect to x∗ . Furthermore, it
can be shown that when replacing the cost function  in (4)
by the rotated cost L, for this modified optimization problem
one still obtains the same solution as for the original optimiza-
tion problem. This is due to the fact that the constraints are
the same, and one can show that thanks to the terminal equal-
ity constraint, the cost functions only differ by a constant term.
This implies that for the convergence analysis, the modified op-
timization problem using the rotated stage cost function L can
be used. But since L is positive definite as discussed above, re-
sults from standard stabilizing MPC can be used in order to
establish closed-loop asymptotic convergence to the optimal
steady-state x∗ . In case that a (nontrivial) terminal region in-
stead of a terminal equality constraint shall be used, again As-
sumption 1 (with −(x, κ f (x)) on the right hand side of iii) re-
placed by −(x, κ f (x)) + (x∗ , u∗ )) has to be satisfied. Again,
since  is in general not positive definite with respect to x∗ , a
different procedure than the one described in Subsection 2.1
has to be used to calculate a suitable terminal region and termi-
nal cost satisfying Assumption 1 as shown in [90] (see also [87]
for a continuous-time version).
In economic MPC schemes without terminal constraints, the
main insight employed in order to establish closed-loop con-
vergence to x∗ is that the optimal solutions to Problem 1 exhibit
a certain turnpike behavior. Here, turnpike behavior [91]–[93]
means that the open-loop optimal trajectories x0 resulting from
application of optimal solutions u0 to Problem 1 spend most
of the time in a neighborhood E of x∗ , where (an upper bound
on) the number of time instants outside this neighborhood de-
pends on the size of E, but is independent of the prediction
horizon N. In [94],[95], it was shown that strict dissipativity
with respect to the supply rate s(x, u) = (x, u) − (x∗ , u∗ ) to-
gether with suitable controllability assumptions implies such a
turnpike behavior (see also [80] for a continuous-time version);
converse statements were recently studied in [96]. Using this
turnpike property, (practical) asymptotic stability of the optimal
steady-state x∗ for the closed-loop system (5) is shown in [97],
using again the optimal value function of a suitably modified
optimization problem as a (practical) Lyapunov function (see
also [98] for a continuous-time version). Here, the size of the
neighborhood into which the closed loop converges depends on
the prediction horizon N, and goes to zero as N → ∞. Inter-
estingly, a candidate solution ũ(t + 1) for the next time instant
t +1 is constructed by appending an additional control value not
necessarily at the end as described in Subsection 2.1 (and as is
also the case in economic MPC with terminal constraints), but
at some point where the optimal open-loop trajectory predicted
at time t, x0 (t), is close to x∗ . Existence of such a time point is
guaranteed by the turnpike property.
Similar to Subsection 4.1, generalizations of the above re-
sults are available to cases where different operating behaviors
than steady-state operation are optimal. In case of optimal peri-
odic operation, closed-loop convergence to the optimal periodic
orbit was established in [85],[99]–[102] using suitable (peri-
odic) terminal constraints and in [86] without such additional
constraints exploiting a certain periodic turnpike property. Fur-
thermore, in the case of time-varying systems and cost func-
tions, closed-loop convergence to some optimal time-varying
optimal trajectory is established in [87] using again a suitable
time-varying dissipativity condition and suitable time-varying
terminal constraints.
4.3 Closed-Loop Performance
As discussed above, the primary objective in economic MPC
is not the stabilization of an a priori given setpoint (or trajec-
tory), but rather the optimization of some general performance
criterion. Hence closed-loop performance guarantees are of
particular interest. The first such result was obtained for a set-
ting with terminal (equality) constraint [75]. Namely, there it
was shown that when using the optimal steady-state x∗ as a ter-
minal equality constraint, the asymptotic average performance
of the resulting closed-loop system (5) is better than or equal to
the optimal steady-state cost, i.e.,
T
(x(t), u MPC (x(t)))
lim sup t=0 ≤ (x∗ , u∗ ). (16)
T →∞ T +1
This result was extended in [90] to a setting using a terminal
region instead of a terminal equality constraint. Furthermore,
in [75] it was also shown that when using an arbitrary periodic
orbit as a periodic terminal constraint, again (16) is satisfied
with (x∗ , u∗ ) on the right hand side replaced by the average
cost along this periodic orbit. The proof of (16) uses the same
candidate input sequence ũ(t+1) as defined in Subsection 2.1 to
show that JN0 (x(t + 1)) ≤ JN0 (x(t)) − (x(t), u MPC (t)) + (x∗ , u∗ ),
from which under some mild technical assumptions (16) fol-
lows. In case that steady-state operation is optimal (compare
Subsection 4.1), (12) together with (16) implies that the closed-
loop exhibits optimal asymptotic average performance. Simi-
lar results are also available for economic MPC schemes with-
out terminal constraints. Namely, for both the cases of opti-
 SICE JCMSI, Vol. 10, No. 2, March 2017
mal steady-state [94] and optimal periodic operation [86], the
same conditions which allow to conclude practical convergence
to the optimal operating behavior (see Subsection 4.2) also al-
low to conclude that the closed-loop asymptotic average perfor-
mance is approximately optimal (with vanishing error term as
N → ∞).
While asymptotic average performance results allow for
statements about the long-term behavior of the closed-loop sys-
tem, they do not allow for statements about transient (finite-
time) performance. Transient performance estimates were ob-
tained for the case where steady-state operation is optimal for
both settings with [103] and without [97] terminal constraints.
Namely, there it was shown that among all trajectories which
steer the system into a certain neighborhood of the optimal
steady-state in a certain number of steps K, the closed-loop (5)
yields the best performance over this time interval up to an error
term depending on K and N (see [97],[103] for a more precise
statement). This means that the closed-loop also exhibits an
approximately optimal transient performance. The generaliza-
tion of this result to the case where steady-state operation is not
optimal is still open. A different finite-time performance esti-
mate was obtained in [104]. There, the authors establish that
the finite-time performance of a certain shrinking-horizon eco-
nomic MPC scheme is at least as good as the one of a certain
stabilizing MPC controller.
4.4 Discussion
The results of the preceding subsections show that dissipa-
tivity plays a crucial role in the context of economic MPC.
Namely, certain (strict) dissipativity conditions can both be
used to classify different optimal operating behaviors as well
as to show that the closed loop converges there as desired.
In [89],[95], it is shown that a linear system with a strictly con-
vex stage cost  subject to convex constraints (with nonempty
interior) is strictly dissipative with respect to the supply rate
s(x, u) = (x, u) − (x∗ , u∗ ) (using a linear storage function λ)
and hence optimally operated at steady-state. For general non-
linear systems, computing a suitable storage function λ in or-
der to verify the above dissipativity conditions can be a diffi-
cult task, and systematic approaches are only available for cer-
tain special system classes such as, e.g., sum of squares tech-
niques for polynomial systems [105]. On the other hand, ex-
plicit knowledge of a storage function λ is not necessary for the
implementation of economic MPC, and the results of the pre-
vious subsections also allow (at least partially) to conclude a
desired closed-loop behavior without having to verify the strict
dissipation inequality (14). Namely, if the optimal operating
behavior for a system is steady-state or periodic operation (in
its strict form), then the corresponding strict dissipativity con-
dition is satisfied (cf. Subsection 4.1), which in turn can be used
to conclude that the closed-loop system converges to the opti-
mal steady-state / periodic orbit (cf. Subsection 4.2). Loosely
speaking, this means that the closed-loop system ”does the right
thing“, i.e., it ”finds“ the optimal operating behavior. How-
ever, in a setting with terminal constraints, these conclusions
are limited to the case where the optimal steady-state / periodic
orbit is known a priori, since it has to be used as a terminal
constraint3 . This is not the case when using an economic MPC
scheme without terminal constraints; here, however, the closed-
loop convergence results of Subsection 4.2 crucially depend on
47
the assumption that an optimal solution to Problem 1 can be
found, which might not be the case in the general nonlinear
(nonconvex) case.
4.5 Further Results
We now briefly discuss some further available results in
the field of economic MPC. Lyapunov-based economic MPC
schemes are, e.g., developed in [104],[106],[107]. Here, con-
straints based on a known control Lyapunov function are added
to the repeatedly solved optimization problem to ensure that
the system stays within a certain stability region or converges
there. Economic MPC using a generalized terminal constraint
was, e.g., considered in [101],[108]–[111]. Here, the predicted
terminal state is not required to be equal to the optimal steady-
state or periodic orbit, but only to some (arbitrary) steady-
state [108]–[110] or periodic orbit [101],[111]. Advantages
compared to ”standard“ terminal constraints are that the opti-
mal steady-state/periodic orbit need not be known a priori, and
the feasible set can be larger. Closed-loop performance and
convergence guarantees can again be established under suit-
able conditions. Generalizations of some of the above results to
time-varying cost functions, systems, and/or constraints were,
e.g., developed in [87],[107],[109],[112],[113]. Besides stan-
dard pointwise-in-time constraints (2), also (less restrictive)
average constraints can be of interest in economic MPC. In
[75],[114],[115] it was shown how Problem 1 can be modified
such that the resulting closed-loop system satisfies these con-
straints. The articles [116],[117] consider combined economic
and tracking cost functions and establish closed-loop conver-
gence or ultimate boundedness depending on how the economic
and tracking part are weighted. For the case where steady-state
operation is optimal, a tracking MPC problem is formulated
in [118],[119] which is shown to be (locally) equivalent to eco-
nomic MPC.
The above results have been developed for nominal systems,
i.e., without uncertainties or disturbances. As was discussed
in [120], just transferring robust MPC approaches from a sta-
bilizing to an economic context might not result in an opti-
mal performance, and hence novel approaches have to be de-
veloped. Some robust and stochastic economic MPC schemes
can, e.g., be found in [111],[120]–[126]. Some of these ref-
erences focus on applications, while others establish theoret-
ical properties such as closed-loop performance and conver-
gence guarantees in the presence of disturbances. Concern-
ing applications, economic MPC schemes have successfully
been applied in many different fields, and it is well beyond the
scope of this article to mention all of these. Examples include,
among others, the process industry [106],[127], building cli-
mate control [128], the control of wind turbines [129], supply
chain management [130], refrigeration systems [131], power
systems [132], and the development of sustainable climate poli-
cies [133].
Finally, concerning distributed economic MPC, there are
only few theoretical results available in the literature. In
[134],[135], the approach of [27] was extended to a setting with
3
If a ”wrong“ terminal constraint is used (e.g., the optimal
steady-state x∗ is used as terminal constraint even though
steady-state operation is not optimal), closed-loop convergence
to the optimal operating behavior cannot be guaranteed. Never-
theless, the asymptotic average performance result (16) holds.
 48 SICE JCMSI, Vol. 10, No. 2, March 2017
economic objective function. A distributed economic MPC
scheme for output stabilization in a network of competing sys-
tems was proposed in [136]. Furthermore, the distributed im-
plementation of a Lyapunov-based economic MPC scheme is
considered in [137], and a hierarchical structure is proposed
in [138], where a centralized economic MPC problem is solved
on the upper level on a slow time scale, while a distributed
controller on the lower level is implemented on a faster time
scale to react on disturbances. Finally, in [139],[140] a dis-
tributed economic MPC scheme is considered for a setting of
self-interested systems which have to satisfy some cooperative
task, where the overall optimal steady-state is not known a pri-
ori, but negotiated online while the systems already take control
actions.
5. Conclusions
This paper is intended to give an overview of recent devel-
opments in the field of model predictive control. A particular
focus was set on the areas of distributed and economic MPC,
where typically more general control objectives than setpoint
stabilization are of interest. In conclusion, both areas have seen
significant progress within the last years and various important
results have been obtained regarding the design and the anal-
ysis of suitable MPC schemes for such settings. Furthermore,
an ever-growing number of successful implementations of both
distributed and economic MPC schemes in a multitude of dif-
ferent application areas has been reported in the literature.
Both fields of distributed and economic MPC are currently
still very active fields of research, and many interesting issues
deserve further attention in the future. For example, further re-
search is needed in both distributed and economic MPC in the
context of uncertain systems or in a time-varying and/or adap-
tive setting. Also, the presented results at the intersection of
distributed and economic MPC only constitute a first step on
the way towards a general cooperative control and coordination
framework for multi-agent systems, especially in a dynamic
setting where the optimal operating behavior of the overall sys-
tem is unknown a priori and may also change online, e.g., due
to alterations in the number or interconnection structure of the
single subsystems. Finally, a further development of efficient
algorithms for real-time implementation of both distributed and
economic MPC schemes in challenging applications is needed.
References
[1] J.M. Maciejowski: Predictive control with constraints, Pear-
son Education, 2002.
[2] D.Q. Mayne, J.B. Rawlings, C.V. Rao, and P.O.M. Scokaert:
Constrained model predictive control: Stability and optimal-
ity, Automatica, Vol. 36, No. 6, pp. 789–814, 2000.
[3] R. Findeisen, L. Imsland, F. Allgöwer, and B.A. Foss: State
and output feedback nonlinear model predictive control: An
overview, European Journal of Control, Vol. 9, No. 2-3,
pp. 190–206, 2003.
[4] J.B. Rawlings and D.Q. Mayne: Model Predictive Control:
Theory and Design, Madison, WI: Nob Hill Publishing, 2009.
[5] L. Grüne and J. Pannek: Nonlinear Model Predictive Control.
Theory and Algorithms, London: Springer, 2011.
[6] E.F. Camacho and C. Bordons: Model Predictive Control,
London: Springer, 2004.
[7] S.J. Qin and T.A. Badgwell: A survey of industrial model
predictive control technology, Control Engineering Practice,
Vol. 11, No. 7, pp. 733–764, 2003.
[8] M.L. Darby and M. Nikolaou: MPC: Current practice and
challenges, Control Engineering Practice, Vol. 20, No. 4,
pp. 328–342, 2012.
[9] D.Q. Mayne: Model predictive control: Recent developments
and future promise, Automatica, Vol. 50, No. 12, pp. 2967–
2986, 2014.
[10] T. Raff, S. Huber, Z.K. Nagy, and F. Allgöwer: Nonlinear
model predictive control of a four tank system: An experi-
mental stability study, Proceedings of the IEEE Conference
on Control Applications, pp. 237–242, 2006.
[11] H. Chen and F. Allgöwer: A quasi-infinite horizon nonlin-
ear model predictive control scheme with guaranteed stability,
Automatica, Vol. 34, No. 10, pp. 1205–1217, 1998.
[12] F.A.C.C. Fontes: A general framework to design stabiliz-
ing nonlinear model predictive controllers, Systems & Control
Letters, Vol. 42, No. 2, pp. 127–143, 2001.
[13] W.H. Chen, J. O’Reilly, and D.J. Ballance: On the termi-
nal region of model predictive control for non-linear systems
with input/state constraints, International Journal of Adaptive
Control and Signal Processing, Vol. 17, No. 3, pp. 195–207,
2003.
[14] S. Yu, H. Chen, C. Böhm, and F. Allgöwer: Enlarging the
terminal region of NMPC with parameter-dependent terminal
control law, Nonlinear Model Predictive Control - Towards
New Challenging Applications, ser. Lecture Notes in Control
and Information Sciences, L. Magni, D.M. Raimondo, and F.
Allgöwer, Eds., pp. 69–78, Springer, 2009.
[15] J.A. Primbs and V. Nevistić: Feasibility and stability of con-
strained finite receding horizon control, Automatica, Vol. 36,
No. 7, pp. 965–971, 2000.
[16] A. Jadbabaie and J. Hauser: On the stability of receding hori-
zon control with a general terminal cost, IEEE Transactions
on Automatic Control, Vol. 50, No. 5, pp. 674–678, 2005.
[17] G. Grimm, M.J. Messina, S.E. Tuna, and A.R. Teel: Model
predictive control: For want of a local control lyapunov func-
tion, all is not lost, IEEE Transactions on Automatic Control,
Vol. 50, No. 5, pp. 546–558, 2005.
[18] S.E. Tuna, M.J. Messina, and A.R. Teel: Shorter horizons for
model predictive control, Proceedings of the American Con-
trol Conference, pp. 863–868, June 2006.
[19] L. Grüne and A. Rantzer: On the infinite horizon performance
of receding horizon controllers, IEEE Transactions on Auto-
matic Control, Vol. 53, No. 9, pp. 2100–2111, 2008.
[20] L. Grüne: Analysis and design of unconstrained nonlinear
MPC schemes for finite and infinite dimensional systems,
SIAM Journal on Control and Optimization, Vol. 48, No. 2,
pp. 1206–1228, 2009.
[21] M. Reble and F. Allgöwer: Unconstrained model predictive
control and suboptimality estimates for nonlinear continuous-
time systems, Automatica, Vol. 48, No. 8, pp. 1812–1817,
2012.
[22] A. Boccia, L. Grüne, and K. Worthmann: Stability and fea-
sibility of state constrained MPC without stabilizing terminal
constraints, Systems & Control Letters, Vol. 72, pp. 14–21,
2014.
[23] P. Mhaskar, N.H. El-Farra, and P.D. Christofides: Stabiliza-
tion of nonlinear systems with state and control constraints
using Lyapunov-based predictive control, Systems & Control
Letters, Vol. 55, No. 8, pp. 650–659, 2006.
[24] R. Scattolini: Architectures for distributed and hierarchical
model predictive control - a review, Journal of Process Con-
trol, Vol. 19, No. 5, pp. 723–731, 2009.
[25] P.D. Christofides, R. Scattolini, D. Muñoz de la Peña, and J.
Liu: Distributed model predictive control: A tutorial review
and future research directions, Computers & Chemical Engi-
neering, Vol. 51, pp. 21–41, 2013.
[26] J.M. Maestre and R.R. Negenborn, Eds.: Distributed Model
 SICE JCMSI, Vol. 10, No. 2, March 2017
Predictive Control Made Easy, Springer, 2014.
[27] B.T. Stewart, A.N. Venkat, J.B. Rawlings, S.J. Wright, and
G. Pannocchia: Cooperative distributed model predictive con-
trol, Systems & Control Letters, Vol. 59, No. 8, pp. 460–469,
2010.
[28] B.T. Stewart, S.J. Wright, and J.B. Rawlings: Cooperative dis-
tributed model predictive control for nonlinear systems, Jour-
nal of Process Control, Vol. 21, No. 5, pp. 698–704, 2011.
[29] D. Groß and O. Stursberg: On the convergence rate of a
Jacobi algorithm for cooperative distributed MPC, Proceed-
ings of the 52nd IEEE Conference on Decision and Control,
pp. 1508–1513, 2013.
[30] P. Giselsson and A. Rantzer: On feasibility, stability and per-
formance in distributed model predictive control, IEEE Trans-
actions on Automatic Control, Vol. 59, No. 4, pp. 1031–1036,
2014.
[31] M.D. Doan, T. Keviczky, and B. De Schutter: A dual
decomposition-based optimization method with guaranteed
primal feasibility for hierarchical MPC problems, Proceed-
ings of the 18th IFAC World Congress, pp. 392–397, 2011.
[32] M. Kögel and R. Findeisen: Cooperative distributed MPC us-
ing the alternating direction multiplier method, Proceedings
of the 8th IFAC Symposium on Advanced Control of Chemical
Processes, pp. 445–450, 2012.
[33] H. Scheu and W. Marquardt: Sensitivity-based coordination
in distributed model predictive control, Journal of Process
Control, Vol. 21, No. 5, pp. 715–728, 2011.
[34] I. Necoara: Suboptimal distributed MPC based on a block-
coordinate descent method with feasibility and stability guar-
antees, Proceedings of the 51st IEEE Conference on Decision
and Control, pp. 4480–4485, 2012.
[35] C. Conte, C.N. Jones, M. Morari, and M.N. Zeilinger: Dis-
tributed synthesis and stability of cooperative distributed
model predictive control for linear systems, Automatica,
Vol. 69, pp. 117–125, 2016.
[36] J. Liu, X. Chen, D. Muñoz de la Peña, and P.D. Christofides:
Sequential and iterative architectures for distributed model
predictive control of nonlinear process systems, AIChE Jour-
nal, Vol. 56, No. 8, pp. 2137–2149, 2010.
[37] J. Liu, X. Chen, D. Muñoz de la Peña, and P.D. Christofides:
Iterative distributed model predictive control of nonlinear sys-
tems: Handling asynchronous, delayed measurements, IEEE
Transactions on Automatic Control, Vol. 57, No. 2, pp. 528–
534, 2012.
[38] A. Richards and J.P. How: Robust distributed model predic-
tive control, International Journal of Control, Vol. 80, No. 9,
pp. 1517–1531, 2007.
[39] L. Grüne and K. Worthmann: A distributed NMPC scheme
without stabilizing terminal constraints, Distributed Decision
Making and Control, ser. Lecture Notes in Control and Infor-
mation Sciences, R. Johansson and A. Rantzer, Eds., Vol. 417,
pp. 261–287, Springer, 2012.
[40] M.A. Müller, M. Reble, and F. Allgöwer: Cooperative con-
trol of dynamically decoupled systems via distributed model
predictive control, International Journal of Robust and Non-
linear Control, Vol. 22, No. 12, pp. 1376–1397, 2012.
[41] P. Trodden and A. Richards: Distributed model predictive
control of linear systems with persistent disturbances, Inter-
national Journal of Control, Vol. 83, No. 8, pp. 1653–1663,
2010.
[42] M.A. Müller, B. Schürmann, and F. Allgöwer: Robust co-
operative control of dynamically decoupled systems via dis-
tributed MPC, Proceedings of the 4th IFAC Nonlinear Model
Predictive Control Conference, pp. 412–417, 2012.
[43] M. Bürger, G. Notarstefano, and F. Allgöwer: From non-
cooperative to cooperative distributed MPC: A simplicial ap-
proximation perspective, Proceedings of the European Con-
49
trol Conference, pp. 2795–2800, 2013.
[44] J.M. Maestre, D. Muñoz de la Peña, E.F. Camacho, and
T. Alamo: Distributed model predictive control based on
agent negotiation, Journal of Process Control, Vol. 21, No. 5,
pp. 685–697, 2011.
[45] P. Varutti, B. Kern, and R. Findeisen: Dissipativity-based dis-
tributed nonlinear predictive control for cascaded systems,
Proceedings of the 8th IFAC Symposium on Advanced Con-
trol of Chemical Processes, pp. 439–444, 2012.
[46] E. Franco, L. Magni, T. Parisini, M.M. Polycarpou, and
D.M. Raimondo: Cooperative constrained control of dis-
tributed agents with nonlinear dynamics and delayed infor-
mation exchange: A stabilizing receding-horizon approach,
IEEE Transactions on Automatic Control, Vol. 53, No. 1,
pp. 324–338, 2008.
[47] M. Farina and R. Scattolini: Distributed predictive control:
A non-cooperative algorithm with neighbor-to-neighbor com-
munication for linear systems, Automatica, Vol. 48, No. 6,
pp. 1088–1096, 2012.
[48] S. Lucia, M. Kögel, and R. Findeisen: Contract-based predic-
tive control of distributed systems with plug and play capabil-
ities, Proceedings of the 5th IFAC Conference on Nonlinear
Model Predictive Control, pp. 205–211, 2015.
[49] D.M. Raimondo, L. Magni, and R. Scattolini: Decentralized
MPC of nonlinear systems: An input-to-state stability ap-
proach, International Journal of Robust and Nonlinear Con-
trol, Vol. 17, No. 17, pp. 1651–1667, 2007.
[50] S. Riverso, M. Farina, and G. Ferrari-Trecate: Plug-and-
play decentralized model predictive control for linear systems,
IEEE Transactions on Automatic Control, Vol. 58, No. 10,
pp. 2608–2614, 2013.
[51] W.B. Dunbar and R.M. Murray: Distributed receding horizon
control for multi-vehicle formation stabilization, Automatica,
Vol. 42, No. 4, pp. 549–558, 2006.
[52] W.B. Dunbar: Distributed receding horizon control of dynam-
ically coupled nonlinear systems, IEEE Transactions on Au-
tomatic Control, Vol. 52, No. 7, pp. 1249–1263, 2007.
[53] M. Zhao and B. Ding: Distributed model predictive control
for constrained nonlinear systems with decoupled local dy-
namics, ISA Transactions, Vol. 55, pp. 1–12, 2015.
[54] T. Keviczky, F. Borrelli, and G.J. Balas: Decentralized reced-
ing horizon control for large scale dynamically decoupled sys-
tems, Automatica, Vol. 42, No. 12, pp. 2105–2115, 2006.
[55] C. Wang and C.-J. Ong: Distributed model predictive control
of dynamically decoupled systems with coupled cost, Auto-
matica, Vol. 46, No. 12, pp. 2053–2058, 2010.
[56] H. Li and Y. Shi: Robust distributed model predictive control
of constrained continuous-time nonlinear systems: A robust-
ness constraint approach, IEEE Transactions on Automatic
Control, Vol. 59, No. 6, pp. 1673–1678, 2014.
[57] R. Olfati-Saber, J.A. Fax, and R.M. Murray: Consensus and
cooperation in networked multi-agent systems, Proceedings
of the IEEE, Vol. 95, No. 1, pp. 215–233, 2007.
[58] G. Ferrari-Trecate, L. Galbusera, M.P.E. Marciandi, and R.
Scattolini: Model predictive control schemes for consensus
in multi-agent systems with single- and double-integrator dy-
namics, IEEE Transactions on Automatic Control, Vol. 54,
No. 11, pp. 2560–2572, 2009.
[59] J. Zhan and X. Li: Consensus of sampled-data multi-agent
networking systems via model predictive control, Automatica,
Vol. 49, No. 8, pp. 2502–2507, 2013.
[60] L. Zhou and S. Li: Distributed model predictive control for
consensus of sampled-data multi-agent systems with double-
integrator dynamics, IET Control Theory Applications, Vol. 9,
No. 12, pp. 1774–1780, 2015.
[61] H. Li, Y. Shi, and W. Yan: On neighbor information utilization
in distributed receding horizon control for consensus-seeking,
 50 SICE JCMSI, Vol. 10, No. 2, March 2017
IEEE Transactions on Cybernetics, Vol. 46, No. 9, pp. 2019–
2027, 2016.
[62] H. Li and W. Yan: Receding horizon control based consen-
sus scheme in general linear multi-agent systems, Automatica,
Vol. 56, pp. 12–18, 2015.
[63] Z. Cheng, H.T. Zhang, M.C. Fan, and G. Chen: Distributed
consensus of multi-agent systems with input constraints: A
model predictive control approach, IEEE Transactions on Cir-
cuits and Systems I: Regular Papers, Vol. 62, No. 3, pp. 825–
834, 2015.
[64] L. Zhou and S. Li: Distributed model predictive control
for multi-agent flocking via neighbor screening optimiza-
tion, International Journal of Robust and Nonlinear Control,
RNC3606, 2016.
[65] B.J. Johansson, A. Speranzon, M. Johansson, and K.H.
Johansson: Distributed model predictive consensus, Proceed-
ings of the 17th International Symposium on Mathematical
Theory of Networks and Systems, pp. 2438–2444, 2006.
[66] T. Keviczky and K.H. Johansson: A study on distributed
model predictive consensus, Proceedings of the 17th IFAC
World Congress, pp. 1516–1521, 2008.
[67] J. Lee, J.-S. Kim, H. Song, and H. Shim: A constrained con-
sensus problem using MPC, International Journal of Control,
Automation and Systems, Vol. 9, No. 5, pp. 952–957, 2011.
[68] A. Gautam, Y.-C. Chu, and Y.C. Soh: Robust H∞ receding
horizon control for a class of coordinated control problems
involving dynamically decoupled subsystems, IEEE Transac-
tions on Automatic Control, Vol. 59, No. 1, pp. 134–149, Jan.
2014.
[69] W.B. Dunbar and D.S. Caveney: Distributed receding horizon
control of vehicle platoons: Stability and string stability, IEEE
Transactions on Automatic Control, Vol. 57, No. 3, pp. 620–
633, 2012.
[70] A. Carron and E. Franco: Receding horizon control of a two-
agent system with competitive objectives, Proceedings of the
American Control Conference, pp. 2533–2538, 2013.
[71] T. Backx, O. Bosgra, and W. Marquardt: Integration of model
predictive control and optimization of processes, Proceedings
of the IFAC Symposium on Advanced Control of Chemical
Processes, pp. 249–260, 2000.
[72] S. Engell: Feedback control for optimal process operation,
Journal of Process Control, Vol. 17, No. 3, pp. 203–219,
2007.
[73] J.B. Rawlings and R. Amrit: Optimizing process economic
performance using model predictive control, Nonlinear Model
Predictive Control - Towards New Challenging Applications,
ser. Lecture Notes in Control and Information Sciences, L.
Magni, D.M. Raimondo, and F. Allgöwer, Eds., pp. 119–138,
Springer, 2009.
[74] J.V. Kadam and W. Marquardt: Integration of economical op-
timization and control for intentionally transient process oper-
ation, Assessment and Future Directions of Nonlinear Model
Predictive Control, ser. Lecture Notes in Control and Infor-
mation Sciences, R. Findeisen, F. Allgöwer, and L.T. Biegler,
Eds., pp. 419–434, Springer, 2007.
[75] D. Angeli, R. Amrit, and J.B. Rawlings: On average per-
formance and stability of economic model predictive con-
trol, IEEE Transactions on Automatic Control, Vol. 57, No. 7,
pp. 1615–1626, 2012.
[76] M. Ellis, H. Durand, and P.D. Christofides: A tutorial review
of economic model predictive control methods, Journal of
Process Control, Vol. 24, No. 8, pp. 1156–1178, 2014.
[77] D. Angeli and J.B. Rawlings: Receding horizon cost op-
timization and control for nonlinear plants, Proceedings of
the 8th IFAC Symposium on Nonlinear Control Systems,
pp. 1217–1223, 2010.
[78] J.C. Willems: Dissipative dynamical systems - part i: General
theory, Archive for Rational Mechanics and Analysis, Vol. 45,
No. 5, pp. 321–351, 1972.
[79] C.I. Byrnes and W. Lin: Losslessness, feedback equiva-
lence, and the global stabilization of discrete-time nonlinear
systems, IEEE Transactions on Automatic Control, Vol. 39,
No. 1, pp. 83–98, 1994.
[80] T. Faulwasser, M. Korda, C.N. Jones, and D. Bonvin: Turn-
pike and dissipativity properties in dynamic real-time opti-
mization and economic MPC, Proceedings of the 53rd IEEE
Conference on Decision and Control, pp. 2734–2739, 2014.
[81] M.A. Müller, L. Grüne, and F. Allgöwer: On the role of dis-
sipativity in economic model predictive control, Proceedings
of the 5th IFAC Conference on Nonlinear Model Predictive
Control, pp. 110–116, 2015.
[82] M.A. Müller, D. Angeli, and F. Allgöwer: On convergence of
averagely constrained economic MPC and necessity of dissi-
pativity for optimal steady-state operation, Proceedings of the
American Control Conference, pp. 3147–3152, 2013.
[83] M.A. Müller, D. Angeli, and F. Allgöwer: On necessity and
robustness of dissipativity in economic model predictive con-
trol, IEEE Transactions on Automatic Control, Vol. 60, No. 6,
pp. 1671–1676, 2015.
[84] L. Grüne and M. Zanon: Periodic optimal control, dissipa-
tivity and MPC, Proceedings of the 21st International Sym-
posium on Mathematical Theory of Networks and Systems,
pp. 1804–1807, 2014.
[85] M. Zanon, L. Grüne, and M. Diehl: Periodic optimal control,
dissipativity and MPC, IEEE Transactions on Automatic Con-
trol, 2017, to appear.
[86] M.A. Müller and L. Grüne: Economic model predictive con-
trol without terminal constraints for optimal periodic behav-
ior, Automatica, Vol. 70, pp. 128–139, 2016.
[87] A. Alessandretti, A.P. Aguiar, and C.N. Jones: On conver-
gence and performance certification of a continuous-time eco-
nomic model predictive control scheme with time-varying
performance index, Automatica, Vol. 68, pp. 305–313, 2016.
[88] J.B. Rawlings, D. Bonné, J.B. Jørgensen, A.N. Venkat, and
S.B. Jørgensen: Unreachable setpoints in model predictive
control, IEEE Transactions on Automatic Control, Vol. 53,
No. 9, pp. 2209–2215, 2008.
[89] M. Diehl, R. Amrit, and J.B. Rawlings: A Lyapunov func-
tion for economic optimizing model predictive control, IEEE
Transactions on Automatic Control, Vol. 56, No. 3, pp. 703–
707, 2011.
[90] R. Amrit, J.B. Rawlings, and D. Angeli: Economic optimiza-
tion using model predictive control with a terminal cost, An-
nual Reviews in Control, Vol. 35, No. 2, pp. 178–186, 2011.
[91] R. Dorfman, P.A. Samuelson, and R.M. Solow: Linear Pro-
gramming and Economic Analysis, New York: Dover Publi-
cations, 1987 reprint of the 1958 original.
[92] A.J. Zaslavski: Turnpike Properties in the Calculus of Varia-
tions and Optimal Control, New York: Springer, 2006.
[93] E. Trélat and E. Zuazua: The turnpike property in finite-
dimensional nonlinear optimal control, Journal of Differential
Equations, Vol. 258, No. 1, pp. 81–114, 2015.
[94] L. Grüne: Economic receding horizon control without ter-
minal constraints, Automatica, Vol. 49, No. 3, pp. 725–734,
2013.
[95] T. Damm, L. Grüne, M. Stieler, and K. Worthmann: An expo-
nential turnpike theorem for dissipative discrete time optimal
control problems, SIAM Journal on Control and Optimization,
Vol. 52, No. 3, pp. 1935–1957, 2014.
[96] L. Grüne and M.A. Müller: On the relation between strict dis-
sipativity and turnpike properties, System & Control Letters,
Vol. 90, pp. 45–53, 2016.
[97] L. Grüne and M. Stieler: Asymptotic stability and transient
optimality of economic MPC without terminal constraints,
 SICE JCMSI, Vol. 10, No. 2, March 2017
Journal of Process Control, Vol. 24, No. 8, pp. 1187–1196,
2014.
[98] T. Faulwasser and D. Bonvin: On the design of economic
NMPC based on approximate turnpike properties, Proceed-
ings of the 54th IEEE Conference on Decision and Control,
pp. 4964–4970, 2015.
[99] R. Huang, E. Harinath, and L.T. Biegler: Lyapunov stability
of economically oriented NMPC for cyclic processes, Journal
of Process Control, Vol. 21, No. 4, pp. 501–509, 2011.
[100] M. Zanon, S. Gros, and M. Diehl: A Lyapunov function for
periodic economic optimizing model predictive control, Pro-
ceedings of the 52nd IEEE Conference on Decision and Con-
trol, pp. 5107–5112, 2013.
[101] D. Limon, M. Pereira, D. Muñoz de la Peña, T. Alamo, and
J.M. Grosso: Single-layer economic model predictive control
for periodic operation, Journal of Process Control, Vol. 24,
No. 8, pp. 1207–1224, 2014.
[102] B. Houska: Enforcing asymptotic orbital stability of eco-
nomic model predictive control, Automatica, Vol. 57, pp. 45–
50, 2015.
[103] L. Grüne and A. Panin: On non-averaged performance of eco-
nomic MPC with terminal conditions, Proceedings of the 54th
IEEE Conference on Decision and Control, pp. 4332–4337,
Dec. 2015.
[104] M. Ellis and P.D. Christofides: On finite-time and infinite-
time cost improvement of economic model predictive control
for nonlinear systems, Automatica, Vol. 50, No. 10, pp. 2561–
2569, 2014.
[105] C. Ebenbauer and F. Allgöwer: Analysis and design of poly-
nomial control systems using dissipation inequalities and sum
of squares, Computers & Chemical Engineering, Vol. 30,
No. 10-12, pp. 1590–1602, 2006.
[106] M. Heidarinejad, J. Liu, and P.D. Christofides: Economic
model predictive control of nonlinear process systems us-
ing Lyapunov techniques, AIChE Journal, Vol. 58, No. 3,
pp. 855–870, 2012.
[107] M. Ellis and P.D. Christofides: Economic model predictive
control with time-varying objective function for nonlinear
process systems, AIChE Journal, Vol. 60, No. 2, pp. 507–519,
2014.
[108] L. Fagiano and A.R. Teel: Generalized terminal state con-
straint for model predictive control, Automatica, Vol. 49,
No. 9, pp. 2622–2631, 2013.
[109] A. Ferramosca, D. Limon, and E.F. Camacho: Economic
MPC for a changing economic criterion for linear systems,
IEEE Transactions on Automatic Control, Vol. 59, No. 10,
pp. 2657–2667, 2014.
[110] M.A. Müller, D. Angeli, and F. Allgöwer: Economic model
predictive control with self-tuning terminal cost, European
Journal of Control, Vol. 19, No. 5, pp. 408–416, 2013.
[111] T.J. Broomhead, C. Manzie, R.C. Shekhar, and P. Hield: Ro-
bust periodic economic MPC for linear systems, Automatica,
Vol. 60, pp. 30–37, 2015.
[112] D. Angeli, A. Casavola, and F. Tedesco: On average per-
formance of economic model predictive control with time-
varying cost and terminal constraints, Proceedings of the
American Control Conference, pp. 2974–2979, July 2015.
[113] D. Angeli, A. Casavola, and F. Tedesco: Economic model
predictive control with parameter-varying cost and guaranteed
average performance, Proceedings of the 54th IEEE Confer-
ence on Decision and Control, pp. 4314–4319, 2015.
[114] M.A. Müller, D. Angeli, F. Allgöwer, R. Amrit, and J.B.
Rawlings: Convergence in economic model predictive con-
trol with average constraints, Automatica, Vol. 50, No. 12,
pp. 3100–3111, 2014.
[115] M.A. Müller, D. Angeli, and F. Allgöwer: Transient average
constraints in economic model predictive control, Automatica,
51
Vol. 50, No. 11, pp. 2943–2950, 2014.
[116] A. Alessandretti, A.P. Aguiar, and C.N. Jones: A model pre-
dictive control scheme with ultimate bound for economic op-
timization, Proceedings of the American Control Conference,
pp. 1653–1658, July 2015.
[117] J.P. Maree and L. Imsland: Combined economic and regu-
latory predictive control, Automatica, Vol. 69, pp. 342–347,
2016.
[118] M. Zanon, S. Gros, and M. Diehl: Indefinite linear MPC and
approximated economic MPC for nonlinear systems, Journal
of Process Control, Vol. 24, No. 8, pp. 1273–1281, 2014.
[119] M. Zanon, S. Gros, and M. Diehl: A tracking MPC formu-
lation that is locally equivalent to economic MPC, Journal of
Process Control, Vol. 45, pp. 30–42, 2016.
[120] F.A. Bayer, M.A. Müller, and F. Allgöwer: Tube-based ro-
bust economic model predictive control, Journal of Process
Control, Vol. 24, No. 8, pp. 1237–1246, 2014.
[121] F.A. Bayer, M. Lorenzen, M.A. Müller, and F. Allgöwer: Ro-
bust economic model predictive control using stochastic in-
formation, Automatica, Vol. 74, pp. 151–161, 2016.
[122] T.G. Hovgaard, L.F.S. Larsen, and J.B. Jørgensen: Robust
economic MPC for a power management scenario with un-
certainties, Proceedings of the 50th IEEE Conference on
Decision and Control and European Control Conference,
pp. 1515–1520, 2011.
[123] R. Huang, L.T. Biegler, and E. Harinath: Robust stability
of economically oriented infinite horizon NMPC that include
cyclic processes, Journal of Process Control, Vol. 22, No. 1,
pp. 51–59, 2012.
[124] L.E. Sokoler, B. Dammann, H. Madsen, and J.B. Jørgensen:
A mean-variance criterion for economic model predictive
control of stochastic linear systems, Proceedings of the 53rd
Conference on Decision and Control, pp. 5907–5914, 2014.
[125] S. Lucia, J.A.E. Andersson, H. Brandt, M. Diehl, and S.
Engell: Handling uncertainty in economic nonlinear model
predictive control: A comparative case study, Journal of Pro-
cess Control, Vol. 24, No. 8, pp. 1247–1259, 2014.
[126] A. Marquez, J. Patiño, and J. Espinosa: Min-max economic
model predictive control, Proceedings of the 53rd IEEE Con-
ference on Decision and Control, pp. 4410–4415, Dec. 2014.
[127] R. Amrit, J.B. Rawlings, and L.T. Biegler: Optimizing pro-
cess economics online using model predictive control, Com-
puters & Chemical Engineering, Vol. 58, pp. 334–343, 2013.
[128] D.I. Mendoza-Serrano and D.J. Chmielewski: Smart grid co-
ordination in building HVAC systems: EMPC and the impact
of forecasting, Journal of Process Control, Vol. 24, No. 8,
pp. 1301–1310, 2014.
[129] S. Gros: An economic NMPC formulation for wind turbine
control, Proceedings of the 52nd IEEE Conference on Deci-
sion and Control, pp. 1001–1006, 2013.
[130] K. Subramanian, J.B. Rawlings, and C.T. Maravelias: Eco-
nomic model predictive control for inventory management in
supply chains, Computers & Chemical Engineering, Vol. 64,
pp. 71–80, 2014.
[131] T.G. Hovgaard, L.F.S. Larsen, K. Edlund, and J.B. Jørgensen:
Model predictive control technologies for efficient and flex-
ible power consumption in refrigeration systems, Energy,
Vol. 44, No. 1, pp. 105–116, 2012.
[132] L.E. Sokoler, K. Edlund, and J.B. Jørgensen: Application of
economic MPC to frequency control in a single-area power
system, Proceedings of the 54th IEEE Conference on Deci-
sion and Control, pp. 2635–2642. 2015.
[133] B. Chu, S. Duncan, A. Papachristodoulou, and C. Hepburn:
Using economic model predictive control to design sustain-
able policies for mitigating climate change, Proceedings of
the 51st IEEE Conference on Decision and Control, pp. 406–
411, 2012.
 52 SICE JCMSI, Vol. 10, No. 2, March 2017

[134] J. Lee and D. Angeli: Cooperative distributed model pre-

dictive control for linear plants subject to convex economic
objectives, Proceedings of the 50th IEEE Conference on
Decision and Control and European Control Conference,
pp. 3434–3439, 2011.
[135] J. Lee and D. Angeli: Distributed cooperative nonlinear eco-
nomic MPC, Proceedings of the 20th International Sympo-
sium on Mathematical Theory of Networks and Systems, 2012.
[136] P.A.A. Driessen, R.M. Hermans, and P.P.J. van den Bosch:
Distributed economic model predictive control of networks in
competitive environments, Proceedings of the 51st IEEE Con-
ference on Decision and Control, pp. 266–271, 2012.
[137] X. Chen, M. Heidarinejad, J. Liu, and P.D. Christofides: Dis-
tributed economic MPC: Application to a nonlinear chemical
process network, Journal of Process Control, Vol. 22, No. 4,
pp. 689–699, 2012.
[138] I.J. Wolf, H. Scheu, and W. Marquardt: A hierarchical dis-
tributed economic NMPC architecture based on neighboring-
extremal updates, Proceedings of the American Control Con-
ference, pp. 4155–4160, 2012.
[139] M.A. Müller and F. Allgöwer: Distributed economic MPC: A
framework for cooperative control problems, Proceedings of
the 19th IFAC World Congress, pp. 1029–1034, 2014.
[140] P.N. Köhler, M.A. Müller, and F. Allgöwer: A distributed
economic MPC scheme for coordination of self-interested
systems, Proceedings of the American Control Conference,
pp. 889–894, 2016.

Matthias A. MÜLLER
He received a Diploma degree in Engineering Cyber-
netics from the University of Stuttgart, Germany, and an
M.S. in Electrical and Computer Engineering from the
University of Illinois at Urbana-Champaign, US, both in
2009. In 2014, he obtained a Ph.D. in Mechanical Engi-
neering, also from the University of Stuttgart, Germany,
for which he received the 2015 European Ph.D. award
on control for complex and heterogeneous systems. He is currently work-
ing as a lecturer and postdoctoral researcher (Akademischer Rat) at the
Institute for Systems Theory and Automatic Control at the University of
Stuttgart. In 2012, he was a visiting researcher at the Imperial College,
London, UK. He is a member of the Eliteprogram for PostDocs of the
Baden-Württemberg Foundation and was a semi-plenary speaker at the 5th
IFAC Conference on Nonlinear Model Predictive Control 2015. His re-
search interests include nonlinear control and estimation, model predictive
control, distributed control and switched systems.

Frank ALLGÖWER
He studied Engineering Cybernetics and Applied
Mathematics in Stuttgart and at the University of Cali-
fornia, Los Angeles (UCLA), respectively, and received
his Ph.D. degree from the University of Stuttgart in Ger-
many. He is the Director of the Institute for Systems The-
ory and Automatic Control and Executive Director of the
Stuttgart Research Centre Systems Biology at the Univer-
sity of Stuttgart. His research interests include cooperative control, predic-
tive control, and nonlinear control with application to a wide range of fields
including systems biology. At present he is President Elect of the Interna-
tional Federation of Automatic Control IFAC and serves as Vice President
of the German Research Foundation DFG.