988 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO.

2, MAY 2010

Model Predictive Control-Based Real-Time

Power System Protection Schemes
Licheng Jin, Member, IEEE, Ratnesh Kumar, Fellow, IEEE, and Nicola Elia, Member, IEEE

Abstract—The objective of power system controls is to keep the The traditional SPS is determined offline and is rule based
electrical flow as well as voltage magnitudes within acceptable [1]–[3]. A rule-based system protection scheme relies on
limits in spite of the load and network topology changes. The con- voltage, or their rate of change levels, or line flow limits. For
trol of voltage level is accomplished by controlling the production,
absorption as well as flow of reactive power at various locations example, if the measured voltage is lower than a specific value,
in the system. This paper presents an approach to determine a or the line flow exceeds the line rating limit, a predefined SPS is
real-time system protection scheme to prevent voltage instability triggered (such as adjustment of generator outputs or load shed-
and maintain a desired amount of post-transient voltage stability ding). The limitation of the rule-based SPSs lies in the use of
margin (an index of system security) following the occurrence of a limited local information. In contrast, a real-time SPS computes
contingency by means of reactive power control. This approach is
based on the model predictive control (MPC) theory. According and carries out control actions based on global state information
to an economic criterion and control effectiveness, a control in response to an impending contingency detected by an online
switching strategy consisting of a sequence of amounts of the dynamic security assessment program. Recent advances in
shunt capacitors to switch is identified for voltage restoration. The monitoring, communication, and computing technologies have
effect of the capacitive control on voltage recovery is measured via greatly facilitated the implementation of real-time SPSs [4].
trajectory sensitivity. The sensitivity of voltage stability margin
with respect to the capacitive control is used to construct a security A real-time system protection scheme for voltage stabiliza-
constraint for post-fault operation in the MPC formulation. The tion is studied in this work. The control of voltage level is ac-
efficacy of the proposed approach is illustrated through applica- complished by controlling the production, absorption, and flow
tions to the WECC system for enhancing the voltage performance of reactive power at various locations in the system. With regard
and to the 39-bus New England system for preventing voltage to a power system, sources and/or sinks of reactive power, such
collapse.
as shunt capacitors, shunt reactors, synchronous condensers,
Index Terms—Model predictive control, power system, and static var compensators (SVCs) are used to control voltage
switching control, trajectory sensitivity, voltage stability margin,
level. In literature, many algorithms [5]–[7] have been devel-
voltage stabilization.
oped to determine the amounts and locations of shunt reactive
power compensation devices needed for maintaining a satisfac-
I. INTRODUCTION tory voltage profile, while minimizing their cost.
Most these work however are based on static analysis, which
means that the voltage performance criteria could be met only
S a result of deregulation as well as increasing demands,
A power systems operate close to their capacity. Although
power systems are designed with proper planning and with
if the system reaches a post-contingency stable operating point.
However, if the disturbances are sever, the power system may
lose stability. Under this situation, the control strategy to restore
proper stability margin, the instability can still occur under the stable equilibrium point requires a dynamic analysis.
certain severe disturbances. It is imperative that schemes for Model predictive control has been applied in power system
power system protection be in place to mitigate their cata- voltage control based on dynamic analysis. Reference [8]
strophic effects such as large scale shutdowns and collapses. presents a method of coordination of load shedding, capacitor
The objective of SPSs is to detect a potential instability or a switching and tap changers using model preventive control.
safety/security degradation of a power system and carry out the The prediction of states is based on the numerical simulation
necessary control actions to mitigate their effects (such as a of nonlinear differential algebraic equations (DAEs) together
partial shutdown or a total collapse). with Euler state prediction. A tree search method is adopted to
solve the optimization. Reference [9] proposes a coordination
Manuscript received November 01, 2007; revised August 17, 2008. First of generator voltage setting points, load shedding and ULTCs
published December 01, 2009; current version published April 21, 2010.
using a heuristic search and the predictive control. The predic-
This work was supported in part by the National Science Foundation under
Grants NSF-ECCS-0424048, NSF-ECCS-0601570, NSF-ECCS-0801763, tion of states is based on the linearization of nonlinear DAEs.
NSF-CCF-0811541, and NSF-ECCS-0926029. Paper no. TPWRS-00811-2007. Reference [10] presents an optimal coordinated voltage control
L. Jin is with the California ISO, Folsom, CA 95630 USA (e-mail: ljin@caiso.
using model predictive control. The controls used include: shunt
com).
R. Kumar and N. Elia are with the Department of Electrical and Com- capacitors, load shedding, tap changers and generator voltage
puter Engineering, Iowa State University, Ames, IA 50011 USA (e-mail: setting points. The prediction of voltage trajectory is based on
[email protected]; [email protected]). the Euler state prediction. The optimization problem is solved
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. by a pseudo gradient evolutionary programming (PGEP) tech-
Digital Object Identifier 10.1109/TPWRS.2009.2034748 nique. In [11] and [12], authors present a method to compute a
0885-8950/$26.00 © 2009 IEEE

Authorized licensed use limited to: Cardiff University. Downloaded on April 27,2023 at 22:20:59 UTC from IEEE Xplore. Restrictions apply.
JIN et al.: MODEL PREDICTIVE CONTROL-BASED REAL-TIME POWER SYSTEM PROTECTION SCHEMES 989

voltage emergency control strategy based on model predictive

control. The prediction of the output trajectories is based on
trajectory sensitivity. However, in these two papers, the authors
employ a simplified model predictive control, which computes
the control actions only at the initial time and implements it
over the entire control horizon. A voltage stabilization control
strategy is also proposed in [13] based on load shedding, where
the objective function is to minimize the amount of load shed-
ding required to restore the voltages. It shows load shedding
is an effective voltage control under emergency condition.
Reference [14] presents a MPC-based voltage control design
using trajectory sensitivities. The controls are reference voltage
of automatic voltage regulators and load shedding.
In this paper, we design a control scheme to restore voltage
following a contingency and to maintain a pre-specified amount
of post-transient voltage stability margin. Voltage stability
margin is an indication of how far the post-transient operating
point is from the voltage collapse point. It is an index of system
security. The derived control strategy not only considers the Fig. 1. Principle of MPC.
dynamic performance of voltages after a contingency, but also
takes into account the degree of post-transient power system
security. The computation of the control strategy is based on previous MPC-based control study, the robustness anal-
model predictive control (MPC). Shunt capacitors are adopted ysis of MPC-based control with respect to load variation
as reactive power compensation devices because they have is meaningful.
been widely used to enhance voltage stability. The control 5) Reference [14] presented a voltage control strategy based
design problem is to determine a capacitor switching sequence on reference voltage of automatic voltage regulators and
and amounts given their locations and capacities to satisfy load shedding. Reference [13] applied MPC for load-shed-
the requirements of voltage performance and voltage stability ding computation. This paper presents an MPC-based
margin. Trajectory sensitivities are used to estimate the effect shunt capacitor control for voltage stabilization. Shunt
of controls on the voltage behavior in a linear manner. The capacitors are commonly used in North America power
features of our work, compared to the prior works dealing with grid for voltage control.
dynamic analysis, is summarized as follows.
1) The control strategy not only prevents voltage instability,
II. BACKGROUND
but also maintains a desired amount of post-transient
voltage stability margin. Voltage stability margin sensitiv-
A. Model Predictive Control
ities are used to characterize the effect of control variables
on stability margin enhancement. Prior works involving MPC refers to a class of algorithms that compute a sequence
dynamic analysis for voltage stabilization did not include of manipulated variable adjustments in order to optimize the
voltage stability margin as a control objective. future behavior of a plant. An introduction to the basic concepts
2) A decreasing horizon MPC is used, in which the control of MPC and a formulation can be found in [15]. The principle
horizon decreases from one iteration to the next. This of MPC is graphically depicted in Fig. 1. Here represents the
modification not only reduces the computation time, but state variable that needs to be controlled to a specific range. The
also helps the convergence of the optimization process. available control is represented by variable .
To the best of our knowledge, this feature of MPC has not At a current time , the MPC solves an optimization problem
been considered in prior works on stabilization of power over a finite prediction horizon with respect to a pre-
systems. determined objective function such that the predicted state vari-
3) This paper provides a comparative study between tradi- able can optimally stay close to a reference trajectory.
tional local state-feedback-based shunt capacitor control The control is computed over a control horizon ,
and MPC-based global shunt capacitor control. To the best which is smaller than the prediction horizon . If there
of our knowledge, none of the prior works demonstrate were no disturbances, no model-plant mismatch and the pre-
why a global model-predictive-based shunt capacitor con- diction horizon is infinite, one could apply the control strategy
trol design, that is arguably more complex, can have ad- found at current time for all times . However, due
vantages over a local state feedback control design. to the disturbances, model-plant mismatch and finite prediction
4) This paper also studies the effect of load variations on the horizon, the true system behavior is different from the predicted
performance of a MPC-based voltage control. Both base- behavior. In order to incorporate the feedback information about
line increase in load, and random fluctuations are consid- the true system state, the computed optimal control is imple-
ered. Since load model plays a critical role in voltage sta- mented only until the next measurement instant , at
bility and load variation has not been considered in any which point the entire computation is repeated.

Authorized licensed use limited to: Cardiff University. Downloaded on April 27,2023 at 22:20:59 UTC from IEEE Xplore. Restrictions apply.
990 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 2, MAY 2010

In a MPC, the optimization problem to be solved at time

can be formulated as follows:

(1)

subject to

(2)
(3)
(4)
(5)

Here, and are the control and prediction horizon with Fig. 2. Application of trajectory sensitivity in system behavior prediction.
. denotes the estimated state and represents “es-
timated” control. (The true state may be different and the true
control matches the estimated control only during the first sam- Detailed information about trajectory sensitivity theory can
pling period.) be found in [16]. The trajectory sensitivity can be solved nu-
Equation (1) represents the cost function of the MPC opti- merically. Reference [17] provides a methodology for the com-
mization. Equation (2) represents the dynamic system model putation of trajectory sensitivity. When time domain simulation
with initial state . Equations (3) and (4) represent the con- of a power system is based on trapezoidal numerical integra-
straints on the control input during the prediction horizon. Equa- tion, the calculation of trajectory sensitivity requires solving a
tion (5) indicates the state operation requirement during the pre- set of linear equations, thus costing a little time. In our work, we
diction horizon. extended the Power System Analysis Tool [18] (a MATLAB-
based tool) to do trajectory sensitivity calculation and the MPC
B. Trajectory Sensitivity optimization.
Consider a differential algebraic equation (DAE) of a system Fig. 2 illustrates the application of trajectory sensitivity in
evaluating the effect of controls on system behavior. The tra-
(6) jectory of the nominal system represents the behavior under
(7) the control . When the control is increased by at time
, the change in predicted system behavior based on sensitivity
where is a vector of state variables, is a vector of algebraic analysis at time , can be approximated as .
variables, and is a vector of control variables. Trajectory sensi- Here is the trajectory sensitivity of the state variable at time
tivity considers the influence of small variations in the control
with respect to the control at time . Similarly if we increase
(and any other variable of interest) on the solution of the state
the control by at time , the change in the state
(6) and (7). Let be a nominal value of , and assume that the
variable at time is represented by . Here,
nominal system in (8) and (9) has a unique solution
over : is the trajectory sensitivity of the state variable at time
with respect to the control at time .
(8)
(9) C. Voltage Stability Margin Sensitivity
Consider a system with DAE model
Then the system in (6) and (7) has a unique solution
over that is related to as

(10)
(11)
where represents a vector of state variables, represents a set
Here is called the trajectory sensi- of algebraic variables, is a vector of control variables, and
tivities of state variables with respect to variable and is a parameter.
is the trajectory sensitivities of algebraic vari- Let be a vector of variables which are parame-
ables with respect to variable . terized by and a change in which (due to a change in ) affects
The evolution of trajectory sensitivities can be obtained by the system stability (For the power system application, this will
differentiating (6) and (7) with respect to the control variables consist of load and generation power.) The th component of
and is expressed as is denoted as which increases linearly with as

(12)
(13)

Authorized licensed use limited to: Cardiff University. Downloaded on April 27,2023 at 22:20:59 UTC from IEEE Xplore. Restrictions apply.
JIN et al.: MODEL PREDICTIVE CONTROL-BASED REAL-TIME POWER SYSTEM PROTECTION SCHEMES 991

Here, is a constant and represents the base case value nominal power system evolves according to (6) and
of the th component of . (7). Here, is the
If increases slowly and continuously, a bifurcation point is
control variable (i.e., amounts of shunt capacitors
reached beyond which the system loses stability. Let be the
value of at this point, then this implies that currently in use). is the amounts of shunt ca-
pacitors that exist at time 0. is the
amounts of shunt capacitors that were added over
time . Time domain simulation is used
to obtain the trajectory of the nominal system (6)
has no solution when . The voltage stability margin and (7), starting from the state at time to
refers to the distance from the voltage collapse point to the the end of prediction horizon . At the same
present total system real power load. It is expressed as time, the trajectory sensitivity of bus voltages with
respect to the shunt capacitors to be added at in-
stants , is obtained
and denoted as (see below for the expla-
nation of notation). In addition, the sensitivity of
The rate change of stability margin with respect to the control voltage stability margin with respect to shunt capac-
variable is known as the margin sensitivity with respect to itor at location is calculated based on a continua-
tion power flow program. It is expressed as
in the optimization.
Step 2) At time , solve the optimization problem over
the prediction horizon and the control
horizon as stated in (14)–(19). The
Reference [19] presented a detailed derivation of the sensitivity objective function is composed of two parts. The
calculation. first term is the trajectory deviation, the second
III. PROBLEM FORMULATION AND SOLUTION term is the cost of controls. The combination of the
deviation of voltages from nominal values and the
The purpose of this work is to find an effective and economic control cost needs to be minimized. The number of
control strategy for controlling the shunt capacitors so as to sat- candidate control locations and their upper limits
isfy the requirements of voltage performance and voltage sta- are determined through a prior planning step (see,
bility margin. For analyzing voltage performance following dis- for example, [22] and [23]). The total number of
turbances, we model generator and automatic voltage regulator control variables in the optimization is the number
(AVR) as well as aggregated exponential dynamic load models of candidate control locations times the number of
[20], [21]. The overall power system is represented by a set of control steps. The optimization is solved in Matlab,
DAEs as in (6) and (7). Here is a vector of states including and it does converge to a global minimum.
state variables in generator dynamic models, AVR models and Minimize (with respect to )
dynamic load models such as, rotor angles and angular speeds of
generators, outputs of AVRs, and active power recovery and re-
active power recovery of dynamic load models. is a vector of
algebraic variables such as bus voltage magnitudes and phase (14)
angles. The vector indicates the output of shunt capacitors.
The computation is iterative over a finite control horizon, where
in each step a quadratic programming problem is solved to com- subject to
pute the amounts of shunt capacitors to be added in that step. The
quadratic programming formulation is valid when the capacitor
control is continuous as in SVC. Even in the case where capac- (15)
itor control is discrete, we can still proceed by assuming con-
tinuous control so as to compute an optimal control by solving
a quadratic programming relaxation. Then for implementation, (16)
the nearest discrete control value can be applied. Any error will
get propagated to a following control step, and where it will get
corrected. The control is piecewise constant, changing only at
the sampling times. Let be the prediction horizon, be (17)
the control horizon, be the control sampling interval, and
be the total number of control steps. The proce-
dure to determine the control strategy at time based on MPC
is as follows.
Step 1) At time (i.e., the th sampling instant), an (18)
estimate of the current state is obtained. The (19)

Authorized licensed use limited to: Cardiff University. Downloaded on April 27,2023 at 22:20:59 UTC from IEEE Xplore. Restrictions apply.
992 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 2, MAY 2010

• is the weight matrix. is the predicted

voltage vector at the control sampling time
that contains all the bus voltages in the system at
time t. is a desired reference voltage. is
the control matrix calculated at time .
• is the weighted cost of control to be
added at time .
• is the total number of control variables, i.e.,
the number of shunt capacitor locations.
• is the total number of control steps.
• is the entry , which is the amount of
control to be added at time .
• is the minimum amount of control
to be added at any step.
• is the maximum amount of control Fig. 3. Structure of a real-time SPS.
to be added at any step.
• is the amount of control implemented
at the control sampling point , . margin of the power system. If a contingency is identified
• is the minimum amount of control where the system will become unstable, the MPC-based SPS
that must be used, typically 0. computation will get triggered at the time an identified critical
• is the maximum available amount of contingency occurs. The steps of the MPC computation in the
control . th iteration include the following.
• is the voltage of bus at time • Estimate static variables such as voltage magnitudes and
, of the nominal system of time . angles at time as well as the dynamic variables
• is the minimum voltage at bus desired such as generator angles, velocities and real and reactive
at time . load recovery.
• is the maximum voltage at bus desired • Run time-domain simulation to compute the system tra-
at time . jectory given the current state. This step also requires the
• is the trajectory sensitivity of voltage at knowledge of a complete system model (including both dy-
bus at time with respect to namic and static components).
control added at time . • Obtain trajectory sensitivities of voltage with respect to the
• is the voltage stability margin at time control variables as a by-product of the time-domain sim-
. ulation performed in the previous step. This is required for
• is the stability margin sensitivity with re- the prediction of system response given a certain control
spect to capacitor at time . strategy.
• is the desirable stability margin for the • Solve the quadratic programming optimization problem
system. and implement the first step of the control.
Step 3) At time , a solution of the optimization problem • Repeat the above steps at each sampling point until the end
(14)–(19) computes a sequence of controls . of control horizon.
Add only the first control at time and ob- Remark: While we suggested an online computation of MPC-
serve or estimate the system state at time based SPS above, it is also possible to do this computation of-
. fline based on the predicted (rather estimated) values of the
Step 4) Increase by and repeat steps (1)–(3) until the states and trajectory sensitivities.
.
IV. APPLICATION TO WECC AND TO NEW ENGLAND SYSTEMS
A. Implementation
The proposed method has been applied to the WECC nine-bus
The functional structure of a real-time SPS is shown in system as well as to the New England 39-bus system. The de-
Fig. 3. Line flow, bus voltage information, switch status as sired voltage stability margin is chosen to be 35%. The exponen-
well as phase measurement unit (PMU) measurements are tial recovery load model is used in both cases. The parameters
sent to a control center through communication channels of a of the load model are as follows:
SCADA system. These measurements plus a network model
are used by the state estimator (SE) for filtering out the noise
and making best use of the measured data. The results from the
state estimator are used for power flow analysis. A power flow The parameters in MPC optimization are determined based
solution is then used by an online dynamic security assessment on the following considerations. Any voltage instability fol-
program to initialize the state variables of the dynamic models. lowing a contingency must be stabilized in a certain time
Further, it uses system models and disturbance information duration (typically the time in which voltage will decrease by
to perform the contingency analysis to evaluate the security 15%). This is the prediction horizon . The control should be

Authorized licensed use limited to: Cardiff University. Downloaded on April 27,2023 at 22:20:59 UTC from IEEE Xplore. Restrictions apply.
JIN et al.: MODEL PREDICTIVE CONTROL-BASED REAL-TIME POWER SYSTEM PROTECTION SCHEMES 993

Fig. 4. WECC three-generator nine-bus test system.

Fig. 5. Voltage behavior of WECC system without MPC control.

exercised on a time horizon , which is shorter than the predic-

tion horizon, typically the time in which voltage will decrease
by 10% (if no control is applied). A discrete-time control must trol actions are required to satisfy the criterion that the voltages
be applied within this duration at a sample-rate high enough outlined above remain above 0.95 p.u.
to adequately react to the changing voltage trajectory, as well as 3) Simulation Result: In this example, we have chosen pre-
to allow accurate enough predictions of the voltage trajectory diction horizon to be 40 s (the time in which voltage drops
based on the linearization of the trajectory-sensitivity. This by nearly 15% at bus 5). has been chosen to be 35 s. We
dictates the sampling duration . The number of sampling found that a sampling duration of works well for this
point is then determined as the ratio of and the sampling example, and so we have the number of control steps:
duration . . Model predictive control approach de-
termines the amounts of shunt capacitors to be added at each
sampling instant so as to recover the local voltages. Although
A. WECC Three-Generator Nine-Bus Test System the capacitors have a positive effect on low voltage problems,
the maximum capacitor to be added at any step was set
1) System Description: Fig. 4 is a representation of the to be 0.1 p.u. This is because if large amounts of capacitors are
WECC three-generator nine-bus system. A fourth-order model added at one time, an over-voltage may occur, which has a bad
is used for modeling each of the three generators. The state effect on the electrical devices of the power system. During the
variables include the rotor angle , the rotor speed , the optimization, we set the lower bound of all bus voltages to be
-axis transient voltage , and the -axis transient voltage . 0.95 p.u. and upper bound of load bus voltages to be 1.05 p.u.
AVR defines the primary voltage regulation for generator 1. For other buses, such as a generator bus, we set the maximum
The continuously acting regulator and exciter model [24] is voltage magnitude to be 1.08 p.u., a bit higher than a load bus.
employed in this study. It is represented by a four-dimensional These settings are practical. Fig. 6 shows the bus voltages after
state equation. The loads at buses 5, 6, and 8 are taken to be MPC-based control was implemented starting at time .
exponential recovery dynamic load and each load is described From the figure, we can see that all the bus voltages were re-
by a two-dimensional state equation. Therefore, the total di- stored to above 0.95 p.u. and the oscillations of the voltages dis-
mension number of the state space is 22. At buses 5, 7, and 8, appeared within 35 s.
there exist shunt capacitors for voltage regulation. These are The control strategy is shown in Table I. Suppose the con-
the control variables. Under normal conditions, all of the shunt trol action starts right after the fault is cleared. The first control
capacitors are disconnected. action happens at . 0.1 p.u. capacitors at buses 5 and
2) Fault Scenario: We consider a three-phase fault at bus 5 7, 0.0648 p.u. capacitors at bus 8 were added. The sample du-
at , which is cleared at by the tripping of the ration is 7 s as explained in the last paragraph. Therefore, the
line between bus 4 and bus 5. Based on the time domain simu- second control action happens at . The third, fourth,
lation, the voltages at buses 5, 7, and 8 are shown in Fig. 5 and and fifth control steps happen at 15.2 s, 22.2 s, and 29.2 s, re-
are not satisfactory. At , the voltages begin to drop dra- spectively. The post-fault system with this control strategy has a
matically due to the three phase to ground fault. At , 52.6% voltage stability margin. Therefore, this optimal control
the voltages start to recover since the fault gets cleared. How- not only improves voltage performance but also has a satisfac-
ever, the voltages begin to oscillate. Voltages begin to decline tory voltage stability margin.
gradually 15 s later. The dynamic load models result in slightly
recovery load consumption, which deteriorate the voltage con-
B. New England Ten-Generator 39-Bus Test System
dition. These three voltages fall out of the lower limit 0.95 p.u.
1 min later. According to the system’s operational criteria, the 1) System Description: Fig. 7 represents the New England
load bus voltages must be above 0.95 p.u. Therefore, some con- ten-generator 39-bus system. All the generator models are taken

Authorized licensed use limited to: Cardiff University. Downloaded on April 27,2023 at 22:20:59 UTC from IEEE Xplore. Restrictions apply.
994 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 2, MAY 2010

Fig. 6. Voltage behavior of WECC system with MPC control.

TABLE I
RESULTING CONTROL STRATEGY FOR WECC SYSTEM

Fig. 7. New England ten-generator 39-bus test system.

to have a fourth-order state-space consisting of the rotor angle

, the rotor speed , and the -axis transient voltage and the
-axis transient voltage. The exception is the generator at bus
39 for which only a third-order model is used that does not in-
clude the d-axis transient voltage as part of the state-space. In
addition, all the generators except generators at buses 34 and
37 have automatic voltage regulators (AVRs), which are repre-
sented by fourth-order models. The load models used in the time
domain simulation are exponential recovery dynamic loads. The
total dimension of the state space is 131. The control variables
include the shunt capacitors that are located at buses 16, 20, 22,
23, and 34. Under normal conditions, none of the shunt capaci-
tors is in use.
2) Fault Scenario: The contingency considered here is a
three-phase to ground fault at bus 21 at , which is Fig. 8. Voltage behavior of New England system without MPC control.
cleared at and by the tripping of the transmission
line between bus 21 and bus 22. The voltage drops dramatically
when the fault occurs as seen in Fig. 8. After the fault is cleared the post-fault power system has a voltage stability margin
at 1.02 s, the voltages recover around 0.95 p.u., although of 32.4%. However as can be seen from simulation (which
some oscillations proceed. About 30 s later, the oscillations considers the dynamic evolution), the system is unable to reach
disappear, but all the voltages start to decline very slowly. Then the associated post-fault equilibrium point. This illustrates the
around 2 min later, the voltages collapse. One reason for the limitation of the control design based on a purely static anal-
voltage recovery is the presence of generator automatic voltage ysis. Through our MPC-based approach (which incorporates
regulators. When the system voltage drops following the fault, the dynamic analysis) we are able to ensure that the post-fault
AVRs start to increase the generator excitation voltages so as system has a desired voltage stability margin of 35%, and the
to support the system voltage. However, AVRs have their upper system is able to reach the associated post-fault equilibrium
limits. At the same time, the exponential recovery of the loads point.
during the voltage disturbance worsen the operation of the 3) Simulation Result: In this example, we have chosen pre-
system. The system can not fully recover from the contingency diction horizon to be 130 s (the time in which voltage drops
considering these two factors, which lead to the voltage col- by nearly 15% at bus 20). has been chosen to be 120 s.
lapse. According to a continuation power-flow-based analysis, We found that a sample duration of works well

Authorized licensed use limited to: Cardiff University. Downloaded on April 27,2023 at 22:20:59 UTC from IEEE Xplore. Restrictions apply.
JIN et al.: MODEL PREDICTIVE CONTROL-BASED REAL-TIME POWER SYSTEM PROTECTION SCHEMES 995

Fig. 9. Voltage behavior of New England system with MPC control. Fig. 10. Voltage behavior of WECC system under local feedback control.

TABLE II
RESULTING CONTROL STRATEGY FOR NEW ENGLAND SYSTEM

for this example, and so we have the number of control steps:

. The control strategy is deter-
mined by our model predictive control approach. The system re-
sponse with MPC in place is shown in Fig. 9. The corresponding
control strategy is shown in Table II. The first control step hap-
pens at 1.2 s. Since the sampling interval is 20 s, the second con-
trol happens at 21.2 s. The third, fourth, and fifth control steps
happen at 41.2 s, 61.2 s, 81.2 s, and 101.2 s, respectively. The
post-fault power system has a voltage stability margin of 35.0%,
which is the required value. Fig. 11. Output of SVC based on local feedback control.

V. COMPARISON WITH TRADITIONAL

to illustrate this point. Suppose there are two shunt capacitors
LOCAL FEEDBACK CONTROL
in the WECC system which are located at buses 5 and 6. Both
Shunt capacitors such as SVCs can also be used as in the SVCs have a capacity of 0.5 p.u. The system data and fault sce-
setting of traditional local feedback control. The mathematical nario are the same as in Section IV. Under no fault, the voltage
formulation of the local feedback control can be expressed as magnitudes of buses 5 and 6 are 0.9819 p.u. and 0.9981 p.u.,
follows: respectively. For local feedback control we set these values as
the reference voltages at buses 5 and 6, respectively. Therefore,
(20) when there is no fault, the outputs of the shunt capacitors are
zero. Suppose the regulation gain is 100 and the regulation
where is regulation gain, is reference voltage, and time constant is 0.5 s. After the fault happens, Fig. 10 depicts
is regulation time constant. is the voltage magnitude of the the dynamic behavior of voltage magnitudes at buses 5 and 6.
regulated bus. is the control amount. For the traditional local Although the voltage at bus 6 is acceptable, the voltage at bus
feedback control, shunt capacitor adjusts its output based on the 5 is unsatisfactory . Fig. 11 shows the outputs of
voltage of the controlled bus. the SVCs at buses 5 and 6. Under no fault, the output of the
Compared with traditional local feedback control, the pro- SVCs are zero. When fault happens, the voltage at bus 6 drops
posed control scheme is more effective since it involves global dramatically, and the output of SVC at bus 6 increases immedi-
state feedback and global control. The state of the entire power ately based on the local feedback control to boost the voltage at
system is taken into consideration in deciding the global con- bus 6. However, after around 5 s, the output of the SVC returns
trol. The WECC system discussed in Section IV can be used to zero since the voltage at bus 6 is greater than the reference

Authorized licensed use limited to: Cardiff University. Downloaded on April 27,2023 at 22:20:59 UTC from IEEE Xplore. Restrictions apply.
996 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 2, MAY 2010

Fig. 12. Voltage behavior of WECC system with MPC. Fig. 13. Voltage behavior with the designed control under 1% load increase.

value. The output of the SVC at bus 5 reaches its maximum

value. Yet the voltage magnitude at bus 5 remains below the de-
sired value. From this simulation, we can see that local feedback
control-based SVC only maintains the voltage of the regulated
bus. It does not offer control for any unsatisfactory voltage be-
havior at other buses.
For the case discussed above, we also use the proposed MPC-
based method to design control. The parameters for the MPC
control are the same as in Section IV. The resulting control
strategy is described as follows. At time 1.2 s, 0.5 p.u. of SVC
control is added at bus 5 and 0.123 p.u. of SVC control is added
at bus 6. At times 8.2 s and 15.2 s, no control is added. At time
22.2 s, 0.0053 p.u. control is added at bus 6. At time 29.2 s, an-
other 0.005 p.u. of SVC control is added at bus 6. The voltage
behaviors at buses 5 and 6 under the MPC-based control design
are shown in Fig. 12. From this figure, we can see that voltage
behavior at bus 5 under the proposed method is better than that
Fig. 14. Voltage behavior with the designed control under 3% load increase.
under local feedback control. From this simulation, we can see
that the main difference between the proposed method and the
local feedback control is that the proposed method makes use of
of the designed control. The second part studies the robustness
all the available controls in the system to improve voltage per-
of the designed control when random disturbances happen on
formance of all the buses. In contrast, the local feedback con-
the dynamic state variables of the load model.
trol-based method makes use of only the local controls. (In the
above example, control at bus 6 is not being used to compensate A. Base Case Load Increase
for performance at bus 5.)
This part studies the robustness of the designed control when
the total base case load increases. Fig. 13 shows the voltage
VI. ROBUSTNESS STUDY behavior of 1% load increase. From this figure, we can see that
The impact of data uncertainty on the performance of the control scheme is still valid under the small load variation.
model-based control methodologies is an important issue. In Fig. 14 indicates the voltage behavior of the same system with
this section, the designed control is tested for robustness over 3% load increase. Although the voltages are stable, the voltage
different operating conditions using time domain simulation. magnitude on bus 5 is lower than 0.95 p.u. This study shows,
Our study is based on the nine-bus three-generator WECC under small load variation, the designed control is still valid.
example of Section IV for which the control scheme is as
shown in Table I. Since load plays an important role in voltage B. Random Disturbance on an Individual Load
stability problem, our robustness study mainly focuses on the Besides the base case load change, we also study the effect
effect of load change, and consists of two parts. The first part is of random disturbance on the dynamic state variable of an indi-
to study the effect of base case load variation on the robustness vidual load. Assume the random disturbance is represented by a

Authorized licensed use limited to: Cardiff University. Downloaded on April 27,2023 at 22:20:59 UTC from IEEE Xplore. Restrictions apply.
JIN et al.: MODEL PREDICTIVE CONTROL-BASED REAL-TIME POWER SYSTEM PROTECTION SCHEMES 997

stability margin. The stabilizing control is achieved through

the economic use of shunt capacitors. Trajectory sensitivity is
used to establish the relationship between the control systems
and the voltage recovery, which provides for a more accurate
measurement of the influence of control. The sensitivity of
Voltage stability margin with respect to control systems is
used to add a security constraint. Further the feature of model
predictive control, that at each control instant implements only
the first step of the computed control sequence, corrects the
errors brought by the approximation of system models. The
WECC and New England systems are employed to illustrate
the effectiveness of the control strategies.
In the numerical simulation, maximum value for control vari-
ables for each step is set as 0.1. This value is used only for illus-
tration. In fact our formulation allows arbitrary values for the
bounds, denoted , in (15) and (16). By solving
the quadratic programming problem of the formulation given
Fig. 15. Voltage behavior with the designed control with 0.7258 increase for
by (14)–(19), an optimal solution is always obtained.
dynamic state variable 17.
REFERENCES
[1] S. Bernard, G. Trudel, and G. Scott, “A 735 kv shunt reactors au-
tomatic switching system for Hydro-Quebec network,” IEEE Trans.
Power Syst., vol. 11, no. 4, pp. 2024–2030, Nov. 1996.
[2] G. Trudel, S. Bernard, and G. Scott, “Hydro-Quebec’s defence plan
against extreme contingencies,” IEEE Trans. Power Syst., vol. 14, no.
3, pp. 958–965, Aug. 1999.
[3] B. Otomega and T. Van Cutsem, “Undervoltage load shedding using
distributed controllers,” IEEE Trans. Power Syst., vol. 22, no. 4, pp.
1898–1907, Nov. 2007.
[4] K. Tomsovic, D. E. Bakken, V. Venkatasubramanian, and A. Bose,
“Designing the next generation of real-time control, communication,
and computations for large power systems,” Proc. IEEE, vol. 93, no. 5,
pp. 965–979, May 2005.
[5] N. Yorino, E. E. El-Araby, H. Sasaki, and S. Harada, “A new formula-
tion for facts allocation for security enhancement against voltage col-
lapse,” IEEE Trans. Power Syst., vol. 18, no. 1, pp. 3–10, Feb. 2003.
[6] Y. L. Chen, “Weak bus-oriented optimal multi-objective VaR plan-
ning,” IEEE Trans. Power Syst., vol. 11, no. 4, pp. 1885–1890, Nov.
1996.
[7] S. Granville and M. C. Abib Lima, “Application of decomposition tech-
niques to VaR planning: Methodological and computational aspects,”
IEEE Trans. Power Syst., vol. 9, no. 4, pp. 1780–1787, Nov. 1994.
Fig. 16. Voltage behavior with the designed control with 0.5883 decrease for [8] M. Larsson, D. J. Hill, and G. Olsson, “Emergency voltage control
dynamic state variable 17. using search and predictive control,” Int. J. Power Energy Syst., vol.
24, no. 2, pp. 121–130, Feb. 2002.
[9] M. Larsson and D. Karlsson, “Coordinated system protection scheme
against voltage collapse using heuristic search and predictive control,”
statistical variable with normal distribution whose mean is zero IEEE Trans. Power Syst., vol. 18, no. 3, pp. 1001–1006, Aug. 2003.
and variance is 1. In our study, a Matlab function Normrnd is [10] J. Y. Wen, Q. H. Wu, D. R. Turner, S. J. Cheng, and J. Fitch, “Optimal
used to generate the disturbance. The disturbance is imposed coordinated voltage control for power system voltage stability,” IEEE
to the active power recovery of the load at bus 6 at control Trans. Power Syst., vol. 19, no. 2, pp. 1115–1122, May 2004.
[11] M. Zima, P. Korba, and G. Andersson, “Power systems voltage emer-
sampling point 3. The first disturbance generated by the Matlab gency control approach using trajectory sensitivities,” in Proc. 2003
program is 0.7258 increase of the dynamic state variable of the IEEE Conf. Control Applications, Istanbul, Turkey, 2003, pp. 189–194.
exponential recovery load model at bus 6. The voltage behavior [12] M. Zima and G. Andersson, “Stability assessment and emergency con-
trol method using trajectory sensitivities,” in Proc. 2003 IEEE Bologna
under such disturbance is shown in Fig. 15. The second distur- Power Tech Conf., Bologna, Italy, Jun. 2003, pp. 7–7.
bance is 0.5883 decrease of the same dynamic state variable. [13] I. A. Hiskens and B. Gong, “MPC-based load shedding for voltage sta-
Fig. 16 indicates the dynamic voltage behavior under the distur- bility enhancement,” in Proc. 44th IEEE Conf. Decision and Control,
bance. These two figures show that the designed control has a Seville, Spain, Dec. 2005, pp. 4463–4468.
[14] M. Zima and G. Andersson, “Model predictive control employing
certain robustness against random load disturbances. trajectory sensitivities for power systems applications,” in Proc. 44th
IEEE Conf. Decision and Control, Seville, Spain, Dec. 2005, pp.
VII. CONCLUSION AND DISCUSSION 4452–4456.
[15] J. Rawlings, “Tutorial overview of model predictive control,” IEEE
This paper proposes a model predictive control design
Control Syst. Mag., vol. 20, no. 3, pp. 38–52, Jun. 2000.
scheme to restore voltage following a contingency and to [16] H. K. Khalil, Nonlinear System, 3rd ed. Englewood Cliffs, NJ: Pren-
maintain a pre-specified amount of post-transient voltage tice-Hall, 2002.

Authorized licensed use limited to: Cardiff University. Downloaded on April 27,2023 at 22:20:59 UTC from IEEE Xplore. Restrictions apply.
998 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 2, MAY 2010

[17] I. A. Hiskens and M. A. Pai, “Trajectory sensitivity analysis of hybrid Ratnesh Kumar (S’87–M’90–SM’00–F’07) re-
systems,” IEEE Trans. Circuits Syst., vol. 47, no. 2, pp. 204–220, Feb. ceived the B.Tech. degree in electrical engineering
2000. from the Indian Institute of Technology, Kanpur,
[18] F. Milano, “An open source power system analysis toolbox,” IEEE India, in 1987 and the M.S. and the Ph.D. degrees
Trans. Power Syst., vol. 20, no. 3, pp. 1199–1206, Aug. 2005. in electrical and computer engineering from the
[19] T. Smed, “Feasible eigenvalue sensitivity for large power systems,” University of Texas at Austin in 1989 and 1991,
IEEE Trans. Power Syst., vol. 8, no. 2, pp. 555–563, May 1993. respectively.
[20] D. Karlsson and D. Hill, “Modeling and identification of nonlinear dy- From 1991–2002, he was on the faculty of the Uni-
namic loads in power systems,” IEEE Trans. Power Syst., vol. 9, no. 1, versity of Kentucky, and since 2002, he is on the fac-
pp. 157–166, Feb. 1994. ulty of the Iowa State University, Ames. He has held
visiting position at the Institute of Systems Research
[21] D. Hill, “Nonlinear dynamic load models with recovery for voltage
at the University of Maryland at College Park, the Applied Research Labora-
stability studies,” IEEE Trans. Power Syst., vol. 8, no. 1, pp. 166–176,
tory at the Pennsylvania State University, the NASA Ames Research Center,
Feb. 1993.
the Argonne National Laboratory—West, and the United Technology Research
[22] H. Liu, L. Jin, J. D. McCalley, R. Kumar, and V. Ajjarapu, “Plan- Center. His primary research interest is in reactive, real-time, and hybrid sys-
ning minimum reactive compensation to mitigate voltage instability,” tems and their applications to embedded software, web services, power systems,
in Proc. 2006 IEEE Power Eng. Soc. General Meeting, Montreal, QC, and autonomous systems. He is coauthor of the book Modeling and Control of
Canada, 2006, pp. 4452–4456. Logical Discrete Event Systems (Norwell, MA: Kluwer, 1995).
[23] H. Liu, L. Jin, J. McCalley, R. Kumar, V. Ajjarapu, and N. Elia, “Plan- Dr. Kumar was a recipient of the Microelectronics and Computer Devel-
ning reconfigurable reactive control for voltage stability limited power opment (MCD) Fellowship from the University of Texas at Austin, and was
systems,” IEEE Trans. Power Syst., vol. 24, no. 2, pp. 1029–1037, May awarded the Lalit Narain Das Memorial Gold Medal for the Best EE Student and
2009. the Ratan Swarup Memorial Gold Medal for the Best All-Rounder Student from
[24] IEEE Committee Report, “Computer representation of excitation the Indian Institute of Technology, Kanpur, India. He is a recipient of the NSF
systems,” IEEE Trans. Power App. Syst., vol. PAS-87, no. 6, pp. Research Initiation Award and NASA-ASEE summer faculty fellowship award.
1460–1464, Jun. 1968. He serves on the program committee for the IEEE Control Systems Society,
the International Workshop on Discrete Event Systems, and the IEEE Work-
shop on Software Cybernetics. He is or has been an associate editor of SIAM
Journal on Control and Optimization, the IEEE TRANSACTIONS ON ROBOTICS
AND AUTOMATION, Journal of Discrete Event Dynamical Systems, and the IEEE
Control Systems Society.

Nicola Elia (M’00) received the Laurea degree in electrical engineering from
Politecnico of Turin, Turin, Italy, in 1987 and the Ph.D. degree in electrical en-
gineering and computer science from the Massachusetts Institute of Technology
(MIT), Cambridge, in 1996.
He was a Control Engineer at the Fiat Research Center from 1987 to 1990.
He was a Postdoctoral Associate at the Laboratory for Information and Deci-
Licheng Jin (S’04–M’07) received the B.S. degree and the M.S. degree from sion Systems, MIT, from 1996 to 1999. Presently, he is Associate Professor with
Zhejiang University, Hangzhou, China, in 2000 and 2003, respectively, and the the Department of Electrical and Computer Engineering, Iowa State University,
M.S. degree in economics and Ph.D. degree in electrical engineering from Iowa Ames. His research interests include computational methods for controller de-
State University, Ames, in 2007 and 2009, respectively. sign, communication systems with access to feedback, and control with com-
She has been working as a Network Application Engineer in California ISO, munication constraints, and networked systems.
Folsom, CA, since 2007. Dr. Elia received the NSF CAREER Award in 2001.

Authorized licensed use limited to: Cardiff University. Downloaded on April 27,2023 at 22:20:59 UTC from IEEE Xplore. Restrictions apply.