Electrical Power and Energy Systems 62 (2014) 289–298

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Distributed model predictive load frequency control of multi-area

interconnected power system
Miaomiao Ma a,b,⇑, Hong Chen c, Xiangjie Liu a, Frank Allgöwer b
a
School of Control and Computer Engineering, North China Electric Power University, No. 2 Beinong Road, Huilongguan, Changping District, Beijing 102206, PR China
b
Institute for Systems Theory and Automatic Control, University of Stuttgart, Pfaffenwaldring 9, 70550 Stuttgart, Germany
c
Department of Control Science and Engineering, Jilin University, Campus NanLing, Renmin Str. 5988, 130025 Changchun, PR China

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents a load frequency control (LFC) design using the distributed model predictive control
Received 19 May 2013 (DMPC) technique for the multi-area interconnected power system. The dynamics model of multi-area
Received in revised form 18 April 2014 interconnected power system is introduced, and Generation Rate Constraint (GRC) and load reference set-
Accepted 22 April 2014
point constraint are considered. The overall system is decomposed into several subsystems and each has
its own local area MPC controller. These subsystem-based MPCs exchange their measurements and
predictions by communication and incorporate the information from other controllers into their local
Keywords:
control objective so as to coordinate with each other. Analysis and simulation results for a three-area
Load frequency control
Distributed model predictive control
interconnected power system show possible improvements on closed-loop performance, computational
Generation Rate Constraint burden and robustness, while respecting physical hard constraints.
Interconnected power system Ó 2014 Elsevier Ltd. All rights reserved.

Introduction suggested in order to achieve better control performance, based

on various control techniques such as proportional–integral–deriv-
Power systems are composed of several interconnected subsys- ative (PID) control (e.g. [1–5]), robust control (e.g. [6–12]), fuzzy
tems or control areas, and one area is connected to another by the control (e.g. [13–17]) and sliding-mode control (e.g. [18–20]).
tie-lines. Each area has its own generator or group of generators, However, most control methods are implemented in a centralized
and it is responsible for its own load and scheduled power manner (e.g. [6,13,15,18] and the reference therein). The controller
interchanges with neighboring areas. Because of the differences has the full knowledge about the overall system and computes all
in generation and load in a power system, systems frequency devi- the control inputs for the system. For any system, centralized
ates from its nominal value and active power flow interchanges controller can achieve better performance because the effect of
between areas deviate from their contracted values. Load frequency interconnections among subsystems are taken into account
control (LFC) is an important control problem in the dynamical exactly. Furthermore, any conflicts among controller objectives
operation of interconnected power systems. The purpose of the are resolved optimally. But centralized control is not well suited
LFC is to drive the frequency deviation and the inter-area power for control of large-scale, geographically expansive power systems,
flow through tie-lines to zero by manipulating the load reference due to the required inherent computational complexity, stability
setpoint following a disturbance (e.g. a step-change in the system and robustness, and communication bandwidth limitations [21].
load). Actually, considering the Generation Rate Constraint and the On the other hand, some control methods mentioned above are
load reference setpoint limitation, this task can be theoretically based on the decentralized control framework (e.g. [4,22,10] and
described as a disturbances attenuation problem of large-scale the reference therein). The effects of the interconnected subsys-
systems with state and input constraints. tems are assumed to be negligible and are ignored in the decentral-
Recently, there is a growing interest in the LFC problem of ized control framework. In many situations, however, the previous
power systems and many different control methods have been assumption is not valid and leads to reduced control performance.
To achieve better closed-loop control performance, some level of
communication may be established between the different control-
⇑ Corresponding author at: School of Control and Computer Engineering, North lers, which leads to the distributed control of interconnected
China Electric Power University, No. 2 Beinong Road, Huilongguan, Changping power system. In addition, some classical control (e.g. [1–4]) meth-
District, Beijing 102206, PR China. Tel.: +86 15210122883. ods mentioned above could yield unsatisfactory performance since
E-mail addresses: [email protected], [email protected] (M. Ma),
the effects of nonlinearities such as Generation Rate Constraint and
[email protected] (H. Chen), [email protected] (F. Allgöwer).

http://dx.doi.org/10.1016/j.ijepes.2014.04.050
0142-0615/Ó 2014 Elsevier Ltd. All rights reserved.
290 M. Ma et al. / Electrical Power and Energy Systems 62 (2014) 289–298

load reference setpoint constraint were not considered. In order to The trend of frequency measured in each control area is an indica-
deal with these issues, advanced distributed control strategies tor of not only the mismatch power in the interconnection and but
have to be investigated and implemented. also in the control area. The LFC system in each control area of a
With the on-line solution of the optimization problem, MPC pre- multi-area interconnected power system should control the inter-
sents a possibility of managing on-line the tradeoff between distur- change power with the other control areas as well as its local fre-
bance attenuation and control (and/or state) constraints, which quency. Therefore, the dynamic LFC system model must take into
appears to be an efficient strategy to control many applications in account the tie-line power signal. For this purpose, consider
industry. Recently, some papers have reported the application of Fig. 1, which shows a power system with M-control areas [9].
MPC technique on the LFC issue (e.g. [23–31] and the reference Because LFC operation is limited to relatively small system distur-
therein). In [23], fast response and robustness against parameter bances, for the design of LFC, a simplified and linearized model is
uncertainties and load changes can be obtained using MPC control- usually used [32]. Some basic power systems terminologies are
ler, but, only for single area load frequency control application. In provided in Table 1. The notation D is used to indicate a deviation
[24] the usage of MPC in multi-area power system is discussed, from steady state. For example, Dx represents a deviation in the
but, only by economic viewpoint. It presented a new model predic- angular frequency from its nominal operating value (60 Hz in the
tive LFC including economy logic for LFC cost reduction. In [25], a US).
new state contractive constraint-based predictive control scheme Consider any control area i 2 IIM interconnected to control area
was proposed for LFC of two-area interconnected power system. j; j – i through a tie line. A simplified model for any area-i of M
This model predictive control algorithm consists of a basic finite power system control areas with an aggregated generator unit in
horizon MPC technique and an additional state contractive con- each area is described [9]. The overall generator-load dynamic rela-
straint. The crucial function of the additional state contractive con- tionship between the incremental mismatch power ðDPmechi  DP Li Þ
straint is to guarantee the stability of the control scheme. In [26], and the frequency deviation Dxi can be expressed as
the design of LFC system based on MPC is investigated for improving
1 1 1 1
power system dynamic performance over a wide range of operating Dx
_i¼ DPmechi  a DPLi  a Di Dxi  a DPtie;i : ð1Þ
M ai Mi Mi Mi
conditions. However, the MPC controllers of [25,26] are both imple-
mented in centralized fashions (cent-MPC), which is impractical for The dynamic of the turbine can be expressed as
control of large-scale power systems. For this reason, many decen-
1 1
tralized or distributed MPC structures have been developed and DP_ mechi ¼ DP v i  DPmechi : ð2Þ
applied recently (e.g. [28–30]). A decentralized model predictive
T CHi T CHi
control (decent-MPC) scheme for the LFC of multi-area intercon- The dynamic of the governor can be expressed as
nected power system is presented in [28]. However, each local area
1 1 1
controller is designed independently and does not consider the Gen- DP_ v i ¼ DPref i  f Dx i  DP v i : ð3Þ
eration Rate Constraint that is only imposed on the turbine in the T Gi Ri T Gi T Gi

simulation. This solution may result in poor systemwide control per-

The tie-line power flow between areas i and j can be described as
formance of power system with significantly interacting subsystem.
In [30], Feasible Cooperation-Based MPC method is used in distrib- DP_ ijtie ¼ T ij Dxi  T ij Dxj ; ð4aÞ
uted LFC instead of centralized MPC. In spite of the good effort done
DPijtie ¼ DPjitie : ð4bÞ
in [30], the paper did not deal with the problem of system’s param-
eters mismatch and Generation Rate Constraint. In addition, the The total tie-line power flow between areas-i and the other areas
range of load change used in the cases is very large and inappropriate can be calculated as
for the LFC issue [30].
X
M X
M X
M
In this paper, we propose the LFC method by using DMPC for the DP_ tie;i ¼ DP_ ijtie ¼ T ij Dxi  T ij Dxj : ð5Þ
multi-area interconnected power system, in which the controllers j¼1 j¼1 j¼1
j–i j–i j–i
coordinate with each other by exchanging their information. In our
scheme, the overall system is decomposed into several subsystems,
each of which is dealt with by a local MPC controller. The subsys-
tem-based MPCs exchange their measurements and predictions by
incorporating this information in their local control objectives. Ptie23
Moreover, Generation Rate Constraint and load reference setpoint Control Area 2 Control Area 3
limitation are considered. Comparisons of response to step load
change, computational burden and robustness have been made
between DMPC, cent-MPC and decent-MPC. The results confirm 12
the superiority of the proposed DMPC technique.
Ptie
13
The paper is organized as follows. ‘Multi-area power system’ Ptie
describes the dynamics model of the interconnected power sys-
tems to be studied. In ‘Design of distributed model predictive con-
troller’, we state briefly the DMPC algorithm and the design of Control Area 1
DMPC controller for three-area interconnected power system. Both
simulation and analysis results are given and discussed in ‘Simula-
tion and analysis’.
1M
Ptie
Multi-area power system
Control Area M
Modelling

A large-scale multi-area power system consists of a number of

interconnected control areas which are connected by tie-lines. Fig. 1. Multi-area interconnected power system.
 M. Ma et al. / Electrical Power and Energy Systems 62 (2014) 289–298 291

Table 1 Fig. 2. A practice three-area power system having the following

Basic power systems terminology. nominal parameters listed in Table 2 [30]. The Generation Rate
x Angular frequency of rotating mass Constraints for all the areas are taken into account by adding limiters
d Phase angle of rotating mass to the turbines as shown in Fig. 2. Since a distributed control scheme
Ma Angular momentum is proposed, each control area has its own LFC controller, as it is
D Percent change in load
Percent change in frequency shown in Fig. 2.
P mech Mechanical power
PL Nonfrequency sensitive load
Problem statement
T CH Charging time constant (prime mover)
Pv Steam valve position
P ref Load reference setpoint In each control area, a change in local power demand (load)
Rf Percent change in frequency
Percent change in unit output
alters the nominal operating frequency. The controller in each con-
TG Governor time constant trol area i manipulates the load reference setpoint Pref i to drive the
T ij Tie-line (between areas i and j) stiffness coefficient frequency deviations Dxi and tie-line power flow deviations DP ijtie
P ijtie Tie-line power flow between areas i and j tie to zero. Power flow through the tie-lines gives rise to interac-
P tie;i Total tie-line power flow between areas-i and others tions among the control areas. Hence a load change in area 2, for
instance, causes a transient frequency change in all control areas.
In a multi-area power system, in addition to regulating area fre- In addition, the Generation Rate Constraint and load reference
quency, the interchange power with neighboring area should be setpoint constraint should be taken into consideration.
maintained at scheduled value. Area Control Error (ACE) indicates In summary, we formulate the LFC problem of multi-area inter-
the power mismatch between the area load and generation. The connected power system as a disturbance rejection problem of
ACEi for control area i can be expressed as a summation of fre- large-scale system with state and input constraints as follows:
quency deviation Dxi multiplied by a bias factor Bi and tie-line Design a local distributed controller for each area, in which con-
power change DP tie;i trollers coordinate with each other by exchanging their informa-
tion, such that area control error ACEi , i.e., the frequency
ACEi ¼ Bi Dxi þ DP tie;i : ð6Þ
deviations Dxi and tie-line power flow deviations DPtie;i , is driven
The above frequency response model (1)–(3), (4b), (5), (6) for to zero, while satisfying the Generation Rate Constraint and load
area-i of M power system control areas can be combined in the fol- reference setpoint constraint.
lowing state space model:
2 3 Design of distributed model predictive controller
 MDia 1
M ai
0  M1a
2 36 i i
72 3
Dx _i 6 0 0 7 In this section, we first design the distributed model predictive
7 Dx i
1 1
6  T CH T CH
6 _ 7 6 i i 76 controller for LFC of the three-area interconnected power system
6 DPmechi 7 6 7 DPmechi 7
6 7 ¼ 6  Rf 1T 0  T1G 0 76 7
6 DP_ 7 6 i Gi 76 7 with state and input constraints. The block diagram of DMPC for
74 DP v i 5
i
4 vi 5 6 three-area interconnected power system is illustrated in Fig. 3.
6X M 7
_ 6 7
DPtie;i 4 T ij 0 0 0 5 DPtie;i Then, for comparison, the centralized MPC (cent-MPC) controller
j¼1 and decentralized MPC (decent-MPC) controller for the three-area
j–i
3 2 interconnected power system are designed, respectively.
2 3 2 0 3
0  M1a 6 0 7 With a set of state variables xi ¼ ½Dxi DP mechi DP v i DPtie;i T , con-
6 7 6 i
7 6 7 trol input ui ¼ DP ref;i , disturbance input wi ¼ DPLi and output
607 6 0 7 6 7
6 0 7
þ6 7
6 1 7DPref i þ6
6
7DP Li þ 6
7 7Dwj ð7aÞ yi ¼ ACEi , the partitioned model for subsystem i of the three-area
4 T Gi 5 4 0 5 6 XM 7 power system can be described as the state-space model
6 7
4  T ij 5 X
0 0 j¼1
j–i
x_ i ðtÞ ¼ Aii xi ðtÞ þ Bii ui ðtÞ þ Eii wi ðtÞ þ ðAij xj ðtÞ þ Bij uj ðtÞÞ; ð10aÞ
j–i
3 2 yi ¼ C ii xi ðtÞ
Dxi ð10bÞ
6 DP 7 2 3
6 mechi 7  MDia 1
0  M1a
yi ¼ ACEi ¼ ½ Bi 0 0 1 6 7 ð7bÞ Mai
4 DP v i 5 6 i i
7 2 3 2 1 3
6 0 1
 T CH 1
0 7 0  Ma
DPtie;i 6 T CH 7
6 i i 7 6 7 6 i
7
6 7 6 0 7 6 7
Aii ¼ 6  1 0  T1G 0 7; Bii ¼ 6 1 7; Eii ¼ 6 0 7;
In a power system, power generation change should not exceed the 6 Rfi T Gi i 7 6 7 6 7
6 7 4 T Gi 5 4 0 5
specified maximum rate. A typical value of the Generation Rate Con- 6X M 7
6 7 0 0
straint for thermal units is considered as 10%/min (ffi 0:0017 4 T ij 0 0 0 5
p:u:MW=s) [28]. In order to respect Generation Rate Constraint, the j¼1
j–i
generator output power deviation rate should be limited as [28]
  ð11aÞ
DPv i  DPmechi  2 3
jDP_ mechi j ¼   6 0:0017 p:u:MW=s;
 ð8Þ 0 0 0 0 2 3
T CHi 6 0
6 0 0 077
0
which can be considered as state constraints. In addition, the load 6 7 607
6 7
0 0 0 7; Bij ¼ 6 7
6
Aij ¼ 6 0 7; C ¼ ½ Bi 0 0 1 :
reference setpoint is constrained as [30] 6 XM 7 4 0 5 ii
6 7
jDPref i j 6 0:3; ð9Þ 4  T ij 0 0 05 0
j¼1
j–i
which can be considered as control input constraints. Both (8) and
ð11bÞ
(9) are time-domain hard constraints.
As an example we take the three-area interconnected power The real system evolves in ‘‘real’’ time. The system model is used
system for load frequency control. The detailed block diagram to predict the future ‘‘within’’ the controller and evolves in some fic-
modeling of the three-area power system with GRC is shown in titious time. In order to distinguish clearly between the real system
292 M. Ma et al. / Electrical Power and Energy Systems 62 (2014) 289–298

Fig. 2. Three-area interconnected power system.

subject to
Table 2 X
Model parameters and input constraints for the three area power network model. x_ i ðtÞ ¼ Aii xi ðtÞ þ Bii u
 i ðtÞ þ Eii wi ðtÞ þ ðAij xj ðtÞ þ Bij u
 j ðtÞÞ; ð14aÞ
j–i
D1 ¼ 2 D2 ¼ 2:75 D3 ¼ 2:4
Rf1 ¼ 0:03 Rf2 ¼ 0:07 Rf3 ¼ 0:05 xi ðt; xi ðtÞ; tÞ ¼ xi ðtÞ; ð14bÞ
 
M a1 ¼ 3:5 Ma2 ¼ 4:0 Ma3 ¼ 3:75 xi3 ðsÞ  xi2 ðsÞ
  6 0:0017; s 2 ½t; t þ T p ; ð14cÞ
T CH1 ¼ 50 T CH2 ¼ 10 T CH3 ¼ 30  T 
CHi
T G1 ¼ 40 T G2 ¼ 25 T G3 ¼ 32
R1 ¼ 1 R2 ¼ 1 R3 ¼ 1  i ðsÞ 6 0:3;
u s 2 ½t; t þ T p : ð14dÞ
B1 ¼ 1 B2 ¼ 1 B3 ¼ 1
T 12 ¼ T 13 ¼ 7:54 T 21 ¼ T 23 ¼ 7:54 T 31 ¼ T 32 ¼ 7:54
Communication Network

and the system model, we denote the internal variables in the con-
troller by a bar ( x; u ) to indicate that the predicted values need
not and will not be the same as the actual values. Thus,  xi ð; xi ðtÞ; tÞ
is the predicted trajectory of subsystem (10) starting from the actual DMPC 1 DMPC 2 DMPC 3
state xi ðtÞ at time t and driven by a given open-loop input function
 i ðÞ. We assume that the state variables xi and the disturbance wi
u
can be measured or estimated directly by the controller i.
For each subsystem i ¼ 1; 2; 3, the open-loop optimal control
problem at time t is formulated as Control Area 2

 i ðÞÞ
min J i ðxi ðtÞ; u ð12Þ
 i ðÞ
u
Control Area 1 Control Area 3
with
Z tþTp  
 i ðÞÞ ¼
J i ðxi ðtÞ; u kxi ðs; xðtÞ; tÞk2Q i þ ku
 i ðsÞk2R ds
i
ð13Þ
t Fig. 3. Block diagram of DMPC for three-area interconnected power system.
 M. Ma et al. / Electrical Power and Energy Systems 62 (2014) 289–298 293

In the above problem, T P is a finite prediction horizon, and for sim-

plicity of exposition, the control and predictive horizons are chosen
Start
to have identical values. Q i 2 Rni ni and Ri 2 Rmi mi denote positive
definite and symmetric weighting matrices. They are tuning param-
eters to achieve the desired performance and can be chosen freely.
The weighting matrices Q i and Ri in the objective function (13) are
chosen as Q 1 ¼ Q 2 ¼ Q 3 ¼ diagð1000; 0; 0; 1000Þ and R1 ¼ R2 ¼
R3 ¼ 1. Note the initial condition in (14b): The system model used
to predict the future in the controller is initialized by the actual, Communication
measured system states. In this formulation, the control prediction
horizon and the state prediction horizon are assumed to be the
same. We assume that the state variables xi and the disturbance
wi can be measured or estimated directly by the controller i. The
distributed model predictive controller for the three-area intercon-
nected power system can be designed by the following algorithm. Initialization
The flow chart for Algorithm 1 is also illustrated in Fig. 4.

Algorithm 1. For the ith controller, the DMPC algorithm is

described as follows.

S1 Communication: Send out its previous predictions NO Assignment

xi ðs; xi ðtk1 Þ; tk1 Þ; s 2 ½t k ; t k1 þ T p  to other controllers and
 Optimization
ui (tk ) ui (tk 1)
also receive information  xj ðs; xj ðtk1 Þ; tk1 Þ; s 2 ½tk ; t k1 þ T p 
from other controllers;
S2 Initialization: At sampling time tk , given measured xi ðt k Þ YES
and wi ðtk Þ, set  xi ðtk Þ ¼ xi ðt k Þ and w  i ðt k Þ ¼ wi ðtk Þ;
S3 Optimization: Solve the optimal control problem (12);
S4 Assignment: If the optimal control problem (12) is feasible, Assignment
 i ðtk Þ, otherwise, ui ðt k Þ ¼ u
ui ðt k Þ ¼ u  i ðtk1 Þ; ui (tk ) ui (tk )
S5 Prediction: Predict the future states
xi ðs; xi ðtk Þ; t k Þ; s 2 ½t kþ1 ; tk þ T p ;

S6 Implementation: Application control ui ðt k Þ, set k ¼ k þ 1
and return S1 at the next sample time;

Prediction

Remark 3.1. When we implement this scheme, it is not necessary

for each controller to communicate with all the other controllers.
Controller i only needs to send information to controller j, which
is coupled with subsystem i. If the system is loosely coupled, the
amount of communication will not be large. Implementation

Remark 3.2. The proposed DMPC algorithm without stability con- k k 1

straints does not guarantee closed loop stability theoretically.
Recently, a variety of different strategies have been proposed to Fig. 4. Flow chart of DMPC Algorithm for the ith controller.
achieve the closed loop system stability in the literature (e.g.
[33–37]). For instance, stability was established via consistency
constraints, i.e., at each time step, the newly calculated optimal tra-
jectories of each system must not deviate too much from the ones and
calculated at the previous time step [33] or from the ones that the 2 3 2 3
A11 A12 A13 B11 B12 B13
neighboring systems assumed [34]. Thus, it requires the transmis- 6 7 6 7
A ¼4 A21 A22 A23 5; B ¼ 4 B21 B22 B23 5;
sion of trajectories as opposed to just current state information at
each update, which increases the computation and communication A31 A32 A33 B31 B32 B33
2 3 2 3
requirements. In addition, the introduction of stability constraints E11 0 0 C 11 0 0
may lead to the infeasibility of the DMPC algorithm, which also 6 7 6 7
E ¼4 0 E22
0 5; C ¼ 4 0 C 22 0 5;
affects the applicability of proposed DMPC. Our future work is 0 0 E33 0 0 C 33
focused on pursuing the implementation of DMPC with guarantee-  T   T
T
ing stability and feasibility while reducing the computation and x ¼ x1 xT2 xT3 2 R12 ; u ¼ uT1 uT2 uT3 2 R3 ;
communication requirements.  T  T
w ¼ wT1 wT2 wT3 2 R3 ; y ¼ yT1 yT2 yT3 2 R3 :
As comparison, we design the cent-MPC controller and decent-
MPC controller for the three-area interconnected power system, with state constraints (14c) and input constraints (14d) for
respectively. We describe the three-area interconnected power sys- i ¼ 1; 2; 3. For each subsystem i ¼ 1; 2; 3, the subvectors
tem to be controlled as the following centralized state-space model: xi 2 R4 ; ui 2 R1 ; wi 2 R1 ; yi 2 R1 respectively represent the subsys-
_ tem state, input, disturbance and output vectors. In the centralized
xðtÞ ¼ AxðtÞ þ BuðtÞ þ EwðtÞ ð15aÞ
MPC framework, the MPC for the overall system (15) solves the
yðtÞ ¼ CxðtÞ ð15bÞ following optimization problem:
294 M. Ma et al. / Electrical Power and Energy Systems 62 (2014) 289–298

Fig. 5. Response to step load disturbance in area 2: DMPC (solid line), cent-MPC (dash-dotted line) and decent-MPC (dashed line).

min JðxðtÞ; u
 ðÞÞ ð16Þ subject to
 ðÞ
u
x_ ðtÞ ¼ AxðtÞ þ Bu  ðtÞ þ EwðtÞ; xðt; xðtÞ; tÞ ¼ xðtÞ; ð18aÞ
with  
xi3 ðsÞ  xi2 ðsÞ
  6 0:0017; s 2 ½t; t þ T p ; i ¼ 1; 2; 3: ð18bÞ
Z tþTp    T 
CHi
 ðÞÞ ¼
J ðxðtÞ; u kxðs; xðtÞ; tÞk2Q þ ku
 ðsÞk2R ds ð17Þ
t
 i ðsÞ 6 0:3;
u s 2 ½t; t þ T p ; i ¼ 1; 2; 3: ð18cÞ
 M. Ma et al. / Electrical Power and Energy Systems 62 (2014) 289–298 295

Fig. 6. Response to step load disturbance in area 3: DMPC (solid line), cent-MPC (dash-dotted line) and decent-MPC (dashed line).

The weighting matrices Q and R in the objective function (17) are ignored in this modeling framework. The decentralized model for
chosen as Q ¼ diagð1000; 0; 0; 1000; 1000; 0; 0; 1000; 1000; 0; 0;1000Þ subsystem i ¼ 1; 2; 3 is
and R ¼ diagð1; 1;1Þ.
In the decentralized modeling framework, it is assumed that the x_ i ðtÞ ¼ Aii xi ðtÞ þ Bii ui ðtÞ þ Eii wi ðtÞ; ð19aÞ
interaction between the subsystems is negligible. Subsequently,
the effect of the external subsystem on the local subsystem is yi ¼ C ii xi ðtÞ ð19bÞ
296 M. Ma et al. / Electrical Power and Energy Systems 62 (2014) 289–298

with the system matrices denoted as (11a), state constraints (14c) Table 3
and input constraints (14d). In the decentralized MPC framework, Comparison of elapsed CPU time.

each subsystem-based MPC solves the following optimization Step load change Elapsed CPU time (s)
problem: Area 1 Area 2 Area 3 DMPC Cent-MPC Decent-MPC
 i ðÞÞ
min J i ðxi ðtÞ; u ð20Þ 5% – – 1699 7681 1738
 i ðÞ
u
– 5% – 1493 7325 1339
– – 5% 1582 7441 1676
with
10% – – 1536 7959 1907
Z tþTp   – 10% – 1239 5660 1328
 i ðÞÞ ¼
J i ðxi ðtÞ; u kxi ðs; xðtÞ; tÞk2Q i þ ku
 i ðsÞk2R ds
i
ð21Þ – – 10% 1642 7451 1815
t

subject to the controller calls fmincon1 for solving the minimum problem of a
x_ i ðtÞ ¼ Aii xi ðtÞ þ Bii u
 i ðtÞ þ Eii wi ðtÞ; xi ðt; xi ðtÞ; tÞ ¼ xi ðtÞ; ð22aÞ constrained nonlinear multi-variable function with given numerical
  parameters (optimality tolerance = 106 , relative error toler-
xi3 ðsÞ  xi2 ðsÞ


 6 0:0017; s 2 ½t; t þ T p ;
 ð22bÞ ance = 103 , absolute error tolerances = 106 , etc.). The heavy on-line
T CH
i computation burden arises partially from the M-files management.
 i ðsÞ 6 0:3;
u s 2 ½t; t þ T p : ð22cÞ We compare the proposed DMPC controller with two other control-
lers (cent-MPC and decent-MPC). For a total simulation of 100 s, the
The weighting matrices Q i and Ri in the objective function (21) are
elapsed CPU times for variable step load changes in different areas
also chosen as Q 1 ¼ Q 2 ¼ Q 3 ¼ diagð1000; 0; 0; 1000Þ and
are shown in Table 3, where the cent-MPC and decent-MPC control-
R1 ¼ R2 ¼ R3 ¼ 1.
lers also use the same optimization routine fmincon and the same
integration algorithm with the same numerical parameters. In
Simulation and analysis Table 3, it is clearly seen that the DMPC controller needs nearly as
much CPU time as decent-MPC controller and significantly less
In this section, we firstly investigate the system response to CPU time than cent-MPC controllers. The proposed DMPC algorithm
step load changes. Then, to evaluate the computation performance has significant computational advantages when compared to cent-
of different control methods, a quantitative comparison is per- MPC while achieving the comparative performance with cent-MPC.
formed and results are discussed. Finally, robustness analysis of Moreover, the DMPC controller sometimes even needs less CPU time
the proposed DMPC scheme for uncertain parameters is assessed. than decent-MPC controller. That is because the initial condition
The simulations for DMPC, cent-MPC and decent-MPC are all per- T
(candidate input sequence) ½ui ðk þ 1jkÞ; . . . ; ui ðk þ T P jkÞ; 0 at time
formed with a sampling time of T ¼ 1 s and prediction horizon of k þ 1 for optimization takes part of the optimal solution
T p ¼ 15 s. T
½ui ðkjkÞ; ui ðk þ 1jkÞ; . . . ; ui ðk þ T P jkÞ at time k. The DMPC controllers
consider the interaction between the subsystems exactly at time k.
Response to step load change However, dcent-MPC controllers ignore the effect of the intercon-
nected subsystems, which is impractical for significantly interacting
In this scenario, performance of the proposed DMPC with subsystem. As a result, the initial conditions for time k þ 1 obtained
respect to step load disturbance is investigated. In order to from the optimal solution at time k may deviate too much from the
simulate disturbance on system, 10% load increase is applied optimal solution, which increases the optimization time.
in area 2 and area 3, respectively. The load disturbance rejec-
tion performance of the DMPC formulation is evaluated and Robustness analysis for uncertain parameters
compared against the performance of cent-MPC and decent-
MPC. The relative performance of DMPC, cent-MPC and In the design of the DMPC controller, we consider only the nom-
decent-MPC rejecting the load disturbance in area 2 and area inal case, i.e., no uncertainties are considered. In fact, there do exist
3 is depicted in Figs. 5 and 6 as solid, dash-dotted and dashed uncertainties in multi-area power systems such as uncertainties in
lines, respectively. It is obvious that the closed-loop trajectory the governor and turbine time constants. Hence, it is necessary to
of the DMPC controller obtained by Algorithm 1, is almost analyze the robustness of the designed DMPC controller. To show
indistinguishable from the closed-loop trajectory of cent-MPC. this, we consider the following two cases with parametric uncer-
With the cent-MPC, decent-MPC and DMPC framework, area tainties in governor and turbine time constants of the three areas:
control error disturbance is all rejected in about 60 s for 10%
load increase in area 2 and 70 s for 10% load increase in area Case 1 : T CH1 ¼ 60ðffi þ20% changeÞ;T CH2 ¼ 12ðffi þ20% changeÞ;
3. The left fifth picture of Fig. 5 and the left sixth picture T CH3 ¼ 36ðffi þ20% changeÞ;
of Fig. 6 show that the decent-MPC controllers violate the T G1 ¼ 48ðffi þ20% changeÞ;T G2 ¼ 30ðffi þ20% changeÞ;
Generation Rate Constraint in area 2 and area 3 where the load
T G3 ¼ 38ðffi þ20% changeÞ:
disturbance occurred, respectively. Whereas, under DMPC and
cent-MPC, the Generation Rate Constraints are still satisfied. Case 2 : T CH1 ¼ 40ðffi 20% changeÞ;T CH2 ¼ 8ðffi 20% changeÞ;
Notice from Figs. 5 and 6 that DMPC controller performs nearly T CH3 ¼ 24ðffi 20% changeÞ;
as well as cent-MPC in driving the local frequency changes to T G1 ¼ 32ðffi 20% changeÞ;T G2 ¼ 20ðffi 20% changeÞ;
zero while respecting the Generation Rate Constraint and load T G3 ¼ 26ðffi 20% changeÞ:
reference setpoint constraint. This confirms the performance
advantage of the proposed distributed model predictive control Fig. 7 depicts the response of the DMPC controllers in the pres-
algorithm. ence of above uncertainty, at 10% load change in area 2. The
response with nominal values is plotted in Fig. 5 as dashed line.
Computational burden
1
fmincon attempts to find a constrained minimum of a scalar function of several
All the simulations are executed on the computer with Intel variables starting at an initial estimate. This is generally referred to as constrained
Xeon quad core processor E5640 @2.67 GHz and 48G RAM, where nonlinear optimization or nonlinear programming.
 M. Ma et al. / Electrical Power and Energy Systems 62 (2014) 289–298 297

Fig. 7. Response to step load disturbance in area 2 for Case 1 and Case 2: DMPC (solid line) and decent-MPC (dashed line).

It has been indicated that the DMPC scheme can achieve the desir- bounds, but not to achieve a minimum. The robustness of the
able performance response, in the precondition of keeping genera- proposed DMPC algorithm against wide range of parameter uncer-
tor rate and control inputs within bounds. However, with decent- tainty is validated.
MPC, the GRC and control input constraints cannot be guaranteed
in the presence of parameter uncertainties. In order to guarantee Conclusions
the Generation Rate Constraint, the responses of DP_ mech1 and
DP_ mech2 with DMPC controller become worse than the decent- In this paper we have formulated the LFC problem of multi-area
MPC controller. However, it is required to keep DP_ mechi within interconnected power system as a disturbance rejection problem
298 M. Ma et al. / Electrical Power and Energy Systems 62 (2014) 289–298

of large-scale system with state and input constraints, that repre- [12] Singh VP, Mohanty SR, Kishor N, Ray PK. Robust h-infinity load frequency
control in hybrid distributed generation system. Int J Electr Power Energy Syst
sent requirements on generator rate and control inputs within
2013;46:294–305.
bounds. The DMPC algorithm for this kind of control problem has [13] Kocaarslan I, Cam E. Fuzzy logic controller in interconnected electrical power
been presented, where the interconnected power system is decom- systems for load-frequency control. Int J Electr Power Energy Syst
posed into subsystems, each with its own local area MPC control- 2005;27(8):542–9.
[14] Lee HJ, Park JB, Joo YH. Robust load-frequency control for uncertain nonlinear
ler. These subsystem-based MPCs exchange their measurements power systems: a fuzzy logic approach. Turk J Electr Eng Comput Sci
and coordinate with each other. The design of distributed model 2006;176:3520–37.
predictive controller is discussed on the basis of the three-area [15] Ozkop E, Altas IH, Sharaf AM. Load frequency control in four area power
systems using fuzzy logic PI controller. In: 16th National power systems
interconnected power system. Analysis and simulation results conference; 15–17 December 2010. p. 233–6.
have confirmed the benefits of the designed DMPC controller in [16] Boroujeni SMS, Fayazi H, Hemmati R, Boroujeni HF. Load frequency control in
achieving the comparative performance with the cent-MPC, in multi area electric power system using genetic scaled fuzzy logic. Int J Phys Sci
2011;6(3):377–85.
the precondition of keeping generator rate and load reference set- [17] Venkata Prasanth B, Jayaram Kumar DSV. Robust fuzzy load frequency
point within bounds. Moreover, the proposed DMPC scheme can controller for a two area interconnected power system. J Theor Appl Inform
guarantee the robust performance in the presence of uncertainties Technol 2011:242–52.
[18] Vrdoljak K, Perić N, Petrović I. Sliding mode based load-frequency control in
due to governors and turbines parameters variation and loads power systems. Electric Power Syst Res 2010;80(5):514–27.
disturbances. [19] Kondo H, Suzuki Y, Iwamoto S. A load frequency control design using the
sliding mode control theory with a VSS observer. In: Power & energy society
general meeting. Waseda Univ., Tokyo, Japan: IEEE; 2009. p. 1–5.
Acknowledgements [20] Vrdoljak K, Perić N, Petrović I. Applying optimal sliding mode based load-
frequency control in power systems with controllable hydro power plants.
Automatica 2010;51(1):3–18.
The work is financially supported by Beijing Natural Science [21] Scattolini R. Architectures for distributed and hierarchical model predictive
Foundation (No. 4133090), National Nature Science Foundation control – a review. J Process Control 2009;19:723–91.
of China (No. 61273144), Fundamental Research Funds for the Cen- [22] Sakhavati A, Gharehpetian GB, Hosseini HS. Decentralized robust load-
frequency control of power system based on quantitative feedback theory.
tral Universities (No. 12MS49 and No. 2014MS20) and Overseas Turkish J Electr Eng Comput Sci 2011;19(4):513–30.
Training Project for Young Key Teachers (No. [2011]3022). [23] Mohamed TH, Bevrani H, Hassan AA, Hiyama T. Model predictive based load
frequency control design. In: 16th international conference of electrical
engineering. Busan, Korea; July 2010. p. 1–6.
References [24] Rerkpreedapong D, Atic N, Feliachi A. Economy oriented model predictive load
frequency control. In: 2003 Large engineering systems conference on power
[1] Khodabakhshian A, Golbon N. Unified PID design for load frequency control. In: engineering; 2003. p. 1–6.
Proceedings of the 2004 IEEE international conference on control applications, [25] Kong LF, Xiao L. A new model predictive control scheme-based load-frequency
vol. 2; 2004. p. 1627–32. control. In: IEEE international conference on control and automation; 2007. p.
[2] Salim Ali E, Abd-Elazim SM. Optimal PID tuning for load frequency control 2514–8.
using bacteria foraging optimization algorithm. In: Proceedings of the 14th [26] Yousef AM. Model predictive control approach based load frequency
international middle east power systems conference. Cairo University, Egypt; controller. WSEAS Trans Syst Control 2011;6(7):265–75.
December 19–21 2010. p. 410–5. [27] Liu XJ, Kong XB, Deng XZ. Power system model predictive load frequency
[3] Das K, Das P, Sharma S. Load frequency control using classical controller in an control. In: Proc Am Contr Conf; 27–29 June 2012. p. 6602–7.
isolated single area and two area reheat thermal power system. Int J Emerg [28] Mohamed TH, Bevrani H, Hassan AA, Hiyama T. Decentralized model
Technol Adv Eng 2012;2(3):403–9. predictive based load frequency control in an interconnected power system.
[4] Yazdizadeh A, Ramezani MH, Hamedrahmat E. Decentralized load frequency Energy Convers Manage 2011;52(2):1208–14.
control using a new robust optimal MISO PID controller. Int J Electr Power [29] Shafiee Q, Morattab A, Bevrani H. Decentralized model predictive load-
Energy Syst 2012;35(1):9. frequency control for multi-area interconnected power systems. In: 2011 19th
[5] Debbarma S, Saikia LC, Sinha N. Automatic generation control using two degree Iranian conference on electrical engineering (ICEE); 17–19 May 2011. p. 1–6.
of freedom fractional order pid controller. Int J Electr Power Energy Syst [30] Venkat AN, Hiskens IA, Rawlings JB, Wright SJ. Distributed MPC strategies with
2014;46:120–9. application to power system automatic generation control. IEEE Trans Control
[6] Wang YY, Zhou RJ, Wen CY. Robust load-frequency controller design for power Syst Technol 2008;16(6):1192–206.
systems. Gener Transm Distrib IEE Proc C 1993;140(1):11–6. [31] Shiroei M, Ranjbar AM. Supervisory predictive control of power system load
[7] Rajeeb Dey A, Sandip Ghosh B, Ray CG. A robust H1 load-frequency controller frequency control. Int J Electr Power Energy Syst 2014;61:70–80.
design using LMIs. In: 18th IEEE international conference on control [32] Wood AJ, Wollenberg BF. Power generation operation and control. New
applications part of IEEE multi-conference on systems and control. Saint York: Wiley; 1996.
Petersburg, Russia; July 8–10 2009. p. 1501–4. [33] Dunbar WB, Murray RM. Distributed receding horizon control for multi-
[8] Dong LL, Zhang Y. On design of a robust load frequency controller for vehicle formation stabilization. Automatica 2006;42(4):549–58.
interconnected power systems. In: Proc Am contr conf. Marriott Waterfront, [34] Keviczky T, Borrelli F, Balas GJ. Decentralized receding horizon control for large
Baltimore, MD, USA; June 30–July 02 2010. p. 1731–6. scale dynamically decoupled systems. Automatica 2006;42(12):2105–15.
[9] Bevrani H. Robust power system frequency control. New York Inc: Springer- [35] Richards A, How JP. Robust distributed model predictive control. Int J Control
Verlag; 2009. 2007;80(9):1517–31.
[10] Dong LL, Zhang Y, Gao ZQ. A robust decentralized load frequency controller for [36] Wang C, Ong C-J. Distributed model predictive control of dynamically
interconnected power systems. ISA Trans 2012;51:410–9. decoupled systems with coupled cost. Automatica 2010;46(12):2053–8.
[11] Shiroei M, Toulabi MR, Ranjbar AM. Robust multivariable predictive based load [37] Müller MA, Reble M, Allgöwer F. Cooperative control of dynamically decoupled
frequency control considering generation rate constraint. Int J Electr Power systems via distributed model predictive control. Int J Robust Nonlinear
Energy Syst 2013;46:405–13. Control 2012;22(12):1376–97.