Improving Power Grids Transient Stability Via Model

Predictive Control
A. Bonfiglio, A. Oliveri, R. Procopio, F. Delfino, G.B. Denegri, M. Invernizzi and M. Storace
DITEN, University of Genoa
Genoa, Italy
{a.bonfiglio, alberto.oliveri}@unige.it

Abstract— This paper proposes the application of Model adjusting the system operating point (mainly active power
Predictive Control (MPC) to ensure the stability of an electric settings) in order to make it able to withstand a set of probable
grid following fault conditions. The technique is firstly described contingencies without losing stability. In the recent years,
as a general tool; subsequently, the chosen model of the power different algorithms have been developed in order to insert
system is detailed together with the specific requirements to be
met by the control algorithm. Finally, the application of MPC
into an Optimal Power Flow problem specific constraints
strategy to a WECC (Western Electricity Coordinating Council) accounting for stability issues (see [8]-[10] for instance). The
test case network shows that the effects of a contingency, that second group of methods is the so-called emergency control,
would bring the system to instability without any external action, as it operates in real-time and designs an action (typically fast
are mitigated by the control action, leading the network to a new generator tripping and load shedding) in order not to lose the
working point, as close as possible to the previous one and, most stability of a power system after the occurrence of a
of all, without exceeding the stability limits. contingency.
The method proposed in this paper belongs to the latter
Keywords - MPC, stability, emergency control, load shedding
category and proposes the application of the Model Predictive
I. INTRODUCTION Control (MPC) technique to ensure the stability of an electric
grid following fault conditions.
Modern power networks are large and complex systems whose MPC is an increasingly popular technique in industry for the
operation should be reliable, secure, and economical [1]. In feedback control of multivariable systems subject to
particular, maintaining transient stability is a fundamental constraints on states and inputs [11]. The control action is
requirement of interconnected power system operation, as it obtained, at each sampling time, by minimizing a proper cost
concerns the ability of a system to withstand severe function, which accounts for the predicted evolution of the
disturbances while ensuring continuity of service. As a system within a given time horizon. This technique has also
consequence, many different power system researchers have been applied to electric power systems, e.g. for HVDC (High
focused their attention on solving the two basic problems of (i) Voltage Direct Current) control to avoid loss of synchronism
defining efficient methods to assess the stability of a power phenomena [12]. The application of MPC strategy can
system subject to a particular contingency and (ii) determining represent a solution to regulate the states of the synchronous
a suitable action to guarantee it. As far as the first problem is generators in a continuous way, in order to avoid the loss of
concerned, the conventional, world-wide accepted approach synchronism and allowing a faster restoration of the system
relies on time-domain numerical integration. The main appeal after the fault removal.
of this approach is its ability to handle any power system The paper is organized as follows: in section II, the main
modeling and to provide detailed time-domain descriptions of features of the MPC strategy are described, while section III
the physical phenomena. The main weakness is the lack of details the model chosen for the test case network adopted to
sensitivity and control information. Also, time-domain validate the developed control system. Finally, section IV
methods remain computationally demanding, despite the presents the obtained results before drawing some concluding
progress of computer performance in recent years. To alleviate remarks (section V).
the above weaknesses, Lyapunov-like direct methods started
being developed in the early sixties [2]-[4]. More recently, II. MODEL PREDICTIVE CONTROL
Lyapunov-like hybrid methods have been adopted; among
them, it is worth mentioning the SIngle Machine Equivalent A. MPC for linear systems
(SIME), which is a hybrid time-domain/direct method [5]-[7]. Let us consider a linear discrete-time system in the form:
In short, it relies on time-domain simulations with early ⎧ xk +1 = Axk + Buk
termination and calculation of stability limits dictated by the ⎨ (1)
equal-area criterion. ⎩ yk = Cxk
About the second problem, there are basically two categories where , and denote the system
of methods for maintaining the stability of a power system. states, inputs and outputs, respectively, at time , being
The first one is the so-called preventive control and consists of the system sampling time. If the system is continuous-time, it

Authorized licensed use limited to: Universite du Quebec a Trois-Rivieres. Downloaded on May 01,2024 at 11:38:34 UTC from IEEE Xplore. Restrictions apply.
must be discretized. Obviously, matrices A and B in the formulation to piecewise-affine (PWA) systems, i.e., systems
discrete-time case are different from the corresponding ones in with different linear dynamics, according to one or more
the continuous-time. decision variables. Physical systems with impacts can be, in
Starting from a measurement of we want to compute a general, described by PWA models: two different dynamics
control action which brings the state to , such that (i) are used before and after the impact.
lim and (ii) the following constraints are In this case, problem (5) can be recast as a Mixed Integer
satisfied: Quadratic or Linear Program (MIQP/MILP) and can be solved
H u uk + i ≤ K u , i = 0,…, ∞ (2) numerically with available software tools; for instance, a very
efficient commercial tool (free for academia) is IBM ILOG
H x xk +i ≤ K x , i = 1,…, ∞ (3)
CPLEX [14]. Obviously the computational effort is much
Notice that this is a general formulation and includes typical larger than for solving a QP or LP.
saturation constraints in the form: For a generic nonlinear system, a systematic solution does not
⎧ xmin ≤ x ≤ xmax exist at present. A possibility is to approximate the nonlinear
⎨ (4) system with a PWA system. A finer approximation allows
⎩umin ≤ u ≤ umax
achieving better control performances at the cost of a higher
If the state is not directly measurable, it must be estimated, complexity of the MIQP/MILP problem.
e.g., through an observer. The MPC technique allows solving Another strategy consists in linearizing the system at each
this constrained regulation problem in a systematic way, by sampling instant around the current state (successive
minimizing the following cost function: linearization [15]). In this way a linear model is used for
min
uk ,…,uk + N −1
{( x k + N |k − xref )
T
( )
QN xk + N |k − xref + prediction (so a QP/LP problem is solved), which is updated at
each step.
(5) Consider a nonlinear system:
}
N −1
⎡ x
∑( ) ( )
T
Q xk + i|k − xref + ( uk + i ) R (uk + i ) ⎤⎥
T
+ k + i |k − xref xk +1 = f ( xk , uk ) (7)
i =0
⎣⎢ ⎦
being : a nonlinear function. The first equation
subject to:
of (6) can be substituted, in this case, with the truncated
⎧ xk +i +1|k = Axk + i|k + Buk + i Taylor expansion of (7):
⎪
⎨ H u uk +i ≤ Ku , i = 0,…, N − 1
⎪H x
(6) ( )
xk + i +1|k = f xk + i|k , uk + i ≅ f ( xk , uk ) +
(8)
⎩ x k + i|k ≤ K x , i = 0,…, N + J x ( xk , uk ) ( xk + i|k − xk ) + J u ( xk , uk )( uk + i − uk )
Here | denotes the predicted state at time
starting from a measurement (or estimation) of the state at
where , , , is the Jacobian
time . , and are
matrix of function computed at point , .
symmetric positive-semidefinite ( must be positive-definite)
Equation (8) is linear and therefore problem (5) reduces to a
matrices defining the cost function, and denotes the QP or LP. This approach works properly if the system does
prediction horizon (equal to the control horizon, in this not exhibit strong nonlinearities.
formulation). Given a measurement of , function (5) only
depends on control moves ,…, and assumes its C. Discussion
minimum value when | ( 0, … , and As shown in the previous sections, MPC technique easily
0 ( 0, … , 1). By solving the optimization allows computing regulators for constrained MIMO linear or
problem, a sequence of control actions ,…, is PWA systems. The main drawback of MPC is the online
obtained. Among these, only the first one ( ) is applied to the computation time required to solve the optimization problem,
system, all the others are discarded. At this point, problem (5) which prevents its applicability to systems with fast dynamics,
is solved again starting from the new state . i.e., with low sampling times (< 1 ms). A solution to overcome
With some algebra, the above optimization problem can be this issue has been proposed in [16], where the optimization
recast as a quadratic program (QP), whose numerical solution problem is solved offline for a given set of states. An explicit
can be found with many efficient algorithms and through solution is therefore obtained for the controller, which results
several free or commercial software tools. A formulation of being a PWA function of the system states. Digital circuit
problem (5) based on 1- or ∞-norm can also be carried out, architectures have been designed for the online computation of
which leads at the end to a linear program (LP). The interested explicit MPC, which exhibit latency times in the order of
reader can refer to [11] to gain insight into MPC. microseconds [17]. Nevertheless, explicit MPC is only
applicable for small-size systems (n < 10), since the
B. MPC for nonlinear systems
complexity of the resulting PWA function grows fast with .
In the previous section we showed how to compute an MPC Another issue can arise from the fact that MPC is a model-
control law for the regulation of a linear system. Obviously, based control algorithm: therefore, if the model of the physical
physical processes cannot always be described with linear process is not accurate, the control implementation can exhibit
models. In [13], it is shown how to generalize the MPC poor performances, even leading to instability. A classical

Authorized licensed use limited to: Universite du Quebec a Trois-Rivieres. Downloaded on May 01,2024 at 11:38:34 UTC from IEEE Xplore. Restrictions apply.
control algorithm such as PID (proportional-integral- ⎧d
derivative) is very simple and fast to implement and is not ⎪ dt δ i ( t ) = ωn ⎡⎣ωi ( t ) − 1⎤⎦
tailored to a physical model, but only relies on measured data. ⎪
Nevertheless, constraints satisfaction is not directly included ⎪ P ( t ) − P ( t ) = 2 H d ω ( t ) i = 1,..., M
⎪ i ei i
dt
i
in the control formulation, even if it can be managed through ⎪⎪
anti-windup [18]. Moreover it is not straightforward to design d
⎨ Pi ( t ) = Ai ( t ) (9)
PID controllers for MIMO systems. ⎪ dt
As will be shown in the following section, the model of the ⎪δ ( 0 ) = δ
distribution grid is MIMO, constrained and slightly non-linear. ⎪ i 0i

Moreover, the sampling time of the system is 30 ms. For these ⎪ωi ( 0 ) = ω0i
⎪
reasons, the implicit MPC appears to be a good choice for ⎪⎩ Pi ( 0 ) = P0i
regulating the system.
where ωn is the rated angular frequency, whereas Hi and Pei
III. BASE POWER SYSTEM MODEL are the inertia constant and the electric power of the i-th
machine, respectively. Terms indicate the controlled inputs.
In this section, the general model of the power system for System (9) can be solved together with the following set of
application of MPC to problems of transient stability and algebraic equations (where the * symbol denotes complex
emergency control is briefly recalled. In the scheme shown in conjugate.) relating the electric power and the state variables:
Fig. 1, M synchronous machines are connected to a passive ⎧ M
⎫
network, described by its extended admittance matrix, in Pei ( t ) = Re ⎨E 'i e jδi (t ) ∑Y *E,ik E 'k e− jδk (t ) ⎬ =
which all passive loads have been converted into constant ⎩ k =1 ⎭
admittances. Moreover, each machine is described by its ⎧ M
− jδ t ⎫
= Re ⎨E 'i e i ( ) ∑( GE,ik − jBE,ik ) E 'k e k ( ) ⎬ =
jδ t
classical second-order model [19], in which the only state ⎩ k =1 ⎭ (10)
variables are the machine rotor angle δi measured with respect M
to a common load flow reference (the network, if present) and = E 'i ∑⎡⎣GE,ik E 'k cos (δi ( t ) − δk ( t ) ) + BE,ik E 'k sin (δi ( t ) − δk ( t ) ) ⎤⎦
the rotating electric speed ωi. In this context, the classical k =1

input for each machine is typically the prime mover i = 1,..., M

mechanical power Pi; however, here, since the physical input The terms YE ,ik = ( GE ,ik + jBE ,ik ) in (10) are the elements of the
is not really the power but rather the valves opening rate,
another state equation has been added for each machine, in admittance matrix reduced at the internal nodes of the
order to relate the controlled input to the prime mover synchronous machines on the basis of their transient
mechanical power. From an electric point of view, each reactances. Moreover, it accounts for the equivalent
machine can be represented by a Thévenin-like equivalent admittance of loads. Substituting (10) into (9), we obtain:
circuit, consisting of an electromotive force (with constant ⎧d
amplitude E’i during the transient) behind the so-called ⎪ dt δ i ( t ) = ωn ⎣⎡ωi ( t ) − ωe ⎦⎤
transient reactance X'di (see [19] for details). The values of E’i ⎪
E , ik E 'k cos ( δ i ( t ) − δ k ( t ) )
⎪ M ⎡G ⎤
are calculated from the initial load flow results by means of ⎪ Pi ( t ) − E 'i ∑ ⎢ ⎥
the X'di. ⎪d k =1 ⎢ + BE ,ik E 'k sin ( δ i ( t ) − δ k ( t ) ) ⎥
⎣ ⎦
⎪ ωi ( t ) =
⎪ dt 2H i
⎨
⎪ d P (t ) = A ( t ) (11)
⎪ dt i i

⎪
⎪δ i ( 0 ) = δ 0i
⎪ω 0 = ω
⎪ i( ) 0i

⎪ Pi ( 0 ) = P0i
⎩
i = 1,..., M
which can be recast in the form
dx
= f c ( x, u ) (12)
dt
where
x = [ δ1 , δ2 ,..., δM , ω1 , ω2 ,..., ωM , P1 , P2 ,...., PM ]
T
Fig. 1. Layout of the electric network used for the control implementation. (13)
u = [ A1 , A2 ,..., AM ]
T
Under these assumptions, the system of ordinary differential (14)
equations representing (in p.u.) the network dynamics is:

Authorized licensed use limited to: Universite du Quebec a Trois-Rivieres. Downloaded on May 01,2024 at 11:38:34 UTC from IEEE Xplore. Restrictions apply.
and then discretized in order to obtain (7). The machine rotor B. Controller design and simulation results
angles (not directly measurable), can be estimated by We designed an implicit MPC controller for the stabilization
integration of speed measured data (see first equation in (11)). of the power grid by successive linearization of the system
dynamics (11). The controller goal is to bring all velocities to
IV. RESULTS
1, trying to keep the mechanical powers as close as possible
In this section, the test case network adopted to validate the to the load flow values , which allows defining the vector
proposed procedure is firstly described (subsection A), then, xref appearing in (5) as follows:
details on the controller design are provided and the results T
highlighting the effectiveness of the developed algorithm are xref = ⎡⎣ 0, 0,..., 0,1,1,...,1, P01 , P02 ,..., P0, M ⎤⎦ (16)
shown (subsection B). Moreover, the problem constraints are:
A. Description ⎧−π + (δ 0i − δCOI ( 0 ) ) ≤ δ i ( t ) − δ COI ( t ) ≤ π − (δ 0i − δCOI ( 0 ) )
⎪
The MPC application for the transient stability analysis has ⎪ωmin ≤ ωi ( t ) ≤ ωmax
been tested on the WECC (Western Electricity Coordinating ⎪
Council) three-machine, nine-bus system shown in Fig. 2. The ⎨ Pi ,min ≤ Pi ( t ) ≤ Pi ,max (17)
⎪
full dynamic data of this system can be found in [20], ⎪ Ai ,min ≤ Ai ( t ) ≤ Ai ,max
generator cost data and limits are defined in [21] and [22], and ⎪i = 1,...,3
the pre-contingency working point is defined in [9]. ⎩
The considered contingency is a three-phase symmetrical fault The first inequality of (17) is to be met to prevent the
occurring at bus seven after 0.1 s from the beginning of the machines from losing stability, whereas the second constraint
simulation and cleared after 0.3 s from the application of the maintains the speed in a suitable range thus not causing the
fault with a simultaneous removal of line 5-7. protection trip. The third and fourth inequalities impose
bounds on both the prime mover mechanical power and its
time derivative . The values of the bounds are defined in
Table I.
TABLE I. VALUES OF CONTROL CONSTRAINTS USED FOR MPC
IMPLEMENTATION

PARAMETER VALUE
0.95 p.u.
1.05 p.u.
Fig. 2. Test case network implemented for the simulations.
, , , 0 p.u.
Figure 3 represents the behaviour of the three machine angles , 1.8 p.u.
with respect to the centre of inertia (COI) defined as: , 1.2 p.u.
3
2.4 p.u.
∑ H δ (t ) i i
,
-10 p.u./s
δ COI ( t ) = i =1
(15) , , ,
3
2.5 p.u./s
∑H
i =1
i
, , ,

under the assumption of keeping constant the mechanical System (11) has been discretized with zero-order hold method
powers of the three machines. The figure shows that the by considering a sampling time of 30 ms, thus leading to a
system is not able to withstand the contingency in the absence nonlinear system in the form of (7).
of a corrective action. For the definition of the controller, the following assumptions
have been made:
1. The system state x is completely measurable
2. The admittance matrix YE is exactly known with a
delay of 3 sampling periods (i.e., 90 ms)
The effects of computation delays were also considered. The
formulations described in section II assume that is
available at the same instant in which the state is acquired.
This is obviously untrue in practice, since the solution of
problem (5) needs some time. It is therefore reasonable to
impose that the controller computed at step is then applied at
step 1. At step k, state is measured, a prediction of
Fig. 3. Evolution of (in thousands of degrees) for the three machines
without control action. Machine 1: gray solid line; machine 2: black solid line; is performed through equation (8) and the controller
machine 3: black dashed line. The vertical solid lines indicate the fault is computed. This control action will be then applied at
occurring instant and the fault-clearing one, respectively. step 1. For our case study, problem (5)-(6) has been
solved with the following parameters, chosen heuristically:

Authorized licensed use limited to: Universite du Quebec a Trois-Rivieres. Downloaded on May 01,2024 at 11:38:34 UTC from IEEE Xplore. Restrictions apply.
 ⎧QN = Q = diag (0, 0, 0,105 ,105 ,105 , 20, 20, 20) Fig. 5 shows the computation times required to solve problem
⎪ −2 −2 −2
(5) during the simulation. The times are always below the
⎨ R = diag (10 ,10 ,10 ) (18) system sampling time (here 30 ms).
⎪ N = 10
⎩
The optimization problem has been solved with the MATLAB
interface to CPLEX solver [14].
Fig. 4 shows the evolution of (top plot), (middle
plot) and (bottom plot), 1,2,3, having applied the MPC
control. It can be noticed that the angles remain bounded and
the velocities approach 1; moreover, the mechanical powers
are adjusted in order to maintain stability and they are
eventually set to a slightly lower value with respect to the load Fig. 5. Computation times needed to solve problem (5) at each sampling
flow value P0i. We also remark that all imposed constraints are instant k.
met. The vertical solid lines indicate the fault occurring instant
and the fault-clearing one, respectively, whereas the vertical Two more simulations have then been performed, in which
dashed lines point out the instants when the controller acquires two classical control techniques (the total and partial fast
the updated admittance matrix. valving [19]) have been adopted. 90 ms after the fault, the
mechanical powers of all machines are brought to 0 for T
seconds and then they are set back to a well-defined value of
mechanical power. T is chosen equal to 1.5 s [19]. The total
fast valving imposes a rigid profile to production units such
that, after the clearing time, the mechanical power of
generating units is brought back to the pre-fault value. In the
partial fast valving the restoration value is imposed at the 60%
of the pre-fault value. The results of the full fast valving are
depicted in Fig. 6.

Fig. 4. Evolution of δ δCOI (in degrees), ω, (in p.u.) for the three
machines with MPC control. Machine 1: gray solid lines; machine 2: black
solid lines; machine 3: black dashed lines.

The adopted simulation time, equal to 5 s, is beyond the

frame of validity of the proposed second-order dynamic model
(which usually is acceptable for a time interval of about 1 s).
Consequently, simulation results for a longer period of time
(when transient stability is already decided) have not to be
considered fully reliable. This fact does not represent a critical
aspect for the achievement of transient stability, which is Fig. 6. Evolution of δ δCOI (in thousands of degrees), ω, (in p.u.) for the
three machines with fast valving control with T 1.5s. Machine 1: gray solid
usually gathered within the first few cycles from the lines; machine 2: black solid lines; machine 3: black dashed lines.
perturbation. A model upgrade for the power system which
accounts for the dynamics of the E’i terms (still maintaining The simulation results highlight that the system stability is not
the same second-order model for the control synthesis) would achieved since the machine angle diverge with respect to the
represent a robustness test of the proposed MPC approach . The solid vertical lines have the same meaning as for
over a longer simulation horizon.

Authorized licensed use limited to: Universite du Quebec a Trois-Rivieres. Downloaded on May 01,2024 at 11:38:34 UTC from IEEE Xplore. Restrictions apply.
Fig. 3 and Fig. 4, whereas the dashed vertical line indicates the physical constraints that, in this specific application, prevent
intervention of the fast valving. the system from losing stability. The performance of the
The angle stability is anyway guaranteed by the application of designed controller has been tested on a three-machine and
the partial fast valving (see Fig. 7). This action does not permit nine-bus test case network, which, after a fault, would not be
the system frequency recovery to 1 p.u., but this is an expected able to withstand the contingency in the absence of a proper
behavior for the proposed configuration that does not account corrective action. The performed simulations highlight the
for the frequency regulation. Frequency regulation acts over a effectiveness of the control system in comparison with the
longer time interval than the one of transient stability and more traditional fast valving. Future work will concern the
surely requires a different representation of the system possibility of verifying the robustness of the algorithm
dynamics. The comparison between the traditional fast valving performing a sensitivity analysis to exploit the possibility of
and the MPC control technique highlights that, with a dynamic the controller to manage both parametric and model
adaptive control of the mechanical power of the generators, it uncertainties. The implementation of a more detailed model of
is possible to reach a working operation asset after the the synchronous machines may also allow the application of
contingency that is much closer to the pre-perturbation one the proposed control technique for different purposes such as
(the three machines reach respectively 92%, 93% and 96% of secondary regulation and the damping of electro-mechanical
the load flow value against the 60% of the partial fast valving) oscillations. Moreover, the application of the technique to a
avoiding the loss of synchronism. Of course the final steady wider network will be crucial to establish its limits in a real-
state condition should be checked against voltage profile and time application, also in relation to the delays introduced by
line loading constraints, not considered in this analysis but measured data transmission, which can affect state estimation.
easily adoptable in a more comprehensive analysis. Moreover,
the action of MPC technique has, in the validity time frame of REFERENCES
the model, a strong damping effect on the system dynamic [1] V. Vittal, “Consequence and impact of electric utility
evolution. industry restructuring on transient stability and small
signal stability analysis,” Proc. IEEE, vol. 88, no. 2, pp.
196–207, Feb. 2000.
[2] M. A. Pai, "Energy Function Analysis for Power System
Stability," Kluwer Academic Publishers, Boston, 1989.
[3] A. A. Fouad and V. Vittal, "Power System Transient
Stability Analysis Using the Transient Energy Function
Methods," Prentice Hall Inc., Englewood Cliffs, NJ,
1992.
[4] M. Pavella and P. G. Murthy, "Transient Stability of
Power Systems: Theory and Practice," J. Wiley & Sons,
Chichester, UK, 1994.
[5] Y. Zhang, P. Rousseaux, I. Wehenkel and M. Pavella,
"SIME: A comprehensive approach to fast transient
stability assessment," Proc. of lEE-Japan, Power and
Energy '96, Osaka, Japan, pp. 1177-182, 1996.
[6] A. Bonfiglio, F. Delfino, M. Invernizzi, A. Perfumo and
R. Procopio, “A Feedback Linearization Scheme for the
Control of Synchronous Generators,” Electric Power
Components and System, vol. 40, no. 16, pp. 1842-1869,
2012.
[7] Y. Zhang, P. Rousseaux, I. Wehenkel and M. Pavella,
"SIME: A hybrid approach to fast transient stability
assessment and contingency selection," Electrical Power
& Energy Systems, Vol. 19, No. 3, pp. 195-208, 1997.
Fig. 7. Evolution of δ δCOI (in thousands of degrees), ω, (in p.u.) for the [8] D. Gan, R. J. Thomas and R. D. Zimmerman, "Stability-
three machines with fast valving control with T 1. Machine 1: gray solid constrained optimal power flow," IEEE Trans. Power
lines; machine 2: black solid lines; machine 3: black dashed lines.
Syst., vol. 15, no. 2, pp. 535–540, May 2000.
V. CONCLUSION [9] R. Zárate-Miñano, T. Van Cutsem, F. Milano and A. J.
This paper details the application of Model Predictive Control Conejo, "Securing transient stability using time-domain
(MPC) to ensure the transient stability of a power system after simulations within an optimal power flow," IEEE Trans.
Power Syst., vol. 25, no. 1, pp. 243–253, Feb. 2010
a contingency. Such technique basically performs the control
action by minimizing a suitable function written in terms of a [10] A. Pizano-Martinez, C. R. Fuerte-Esquivel and D. Ruiz-
quadratic form of the state and the input under a set of Vega, "A New Practical Approach to Transient Stability-

Authorized licensed use limited to: Universite du Quebec a Trois-Rivieres. Downloaded on May 01,2024 at 11:38:34 UTC from IEEE Xplore. Restrictions apply.
 constrained Optimal Power Flow," IEEE Trans. Power [17] A. Oliveri and M. Storace, "Hardware-in-the-loop
Syst., vol. 26, no. 3, pp. 1686–1696, Aug. 2011. simulations of circuit architectures for the computation
[11] E. Camacho and C. Bordons, "Model Predictive of exact and approximate explicit MPC control
Control," ser. Advanced Textbooks in Control and Signal functions," proc. 19th IEEE International Conference on
Processing, 2nd ed., London, U.K.: Springer-Verlag, Electronics, Circuits and Systems (ICECS) 2012, Seville,
2004. Spain, pp.380-383, 9-12 Dec. 2012.
[12] Y. Phulpin, J. Hazra and D. Ernst "Model predictive [18] Y. Peng, V. Damir and H. Raymond "Anti-windup,
control of HVDC power flow to improve transient bumpless, and conditioned transfer techniques for PID
stability" Proc. of the Second IEEE International controllers," IEEE Control Systems, vol. 16, no. 4, pp.
Conference on Smart Grid Communications (IEEE 48-57, 1996.
SmartGridComm), Brussels, Belgium, October 17-20, [19] P. Kundur, “Power System Stability and Control,”
2011, pp. 611-616. McGraw-Hill, New York, 1994.
[13] A. Bemporad and M. Morari, "Control of systems [20] P. W. Sauer and M. A. Pai, ‘‘Power System Dynamics
integrating logic, dynamics, and constraints," Automatica and Stability,’’ Upper Saddle River, NJ: Prentice-Hall,
vol. 35, no. 3, pp. 407-427, 1999. 1998.
[14] IBM ILOG CPLEX. Available at http://www- [21] T. B. Nguyen and M. A. Pai, ‘‘Dynamic security
01.ibm.com/software/commerce/optimization/cplex- constrained rescheduling of power systems using
optimizer trajectory sensitivities,’’ IEEE Trans. Power Syst., vol.
18, no. 2, pp. 848---854, May 2003.
[15] M. A. Henson, "Nonlinear model predictive control:
[22] H. R. Cai, C. Y. Chung and K. P. Wong, ‘‘Application
current status and future directions," Computers and
of differential evolution algorithm for transient stability
Chemical Engineering, vol. 23, no. 2, pp. 187-202, 1998.
constrained optimal power flow,’’ IEEE Trans. Power
[16] A. Alessio and A. Bemporad, “A survey on explicit Syst., vol. 23, no. 2, pp. 719---728, May 2008.
model predictive control,” in Nonlinear Model Predictive
Control: Towards New Challenging Applications, ser.
Lecture Notes in Control and Information Sciences, D.
R. L. Magni and F. Allgower, Eds. Berlin, Heidelberg,
Germany: Springer-Verlag, 2009, vol. 384, pp. 345–369.

Authorized licensed use limited to: Universite du Quebec a Trois-Rivieres. Downloaded on May 01,2024 at 11:38:34 UTC from IEEE Xplore. Restrictions apply.