See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/321098170

A Comparative Study of LQR, LQG, and Integral LQG Controller for Frequency
Control of Interconnected Smart Grid

Conference Paper · December 2017

DOI: 10.1109/EICT.2017.8275216

CITATIONS READS

11 538

4 authors, including:

Mizanur Rahman Subrata Kumar Sarker

Rajshahi University of Engineering & Technology Varendra University
6 PUBLICATIONS   17 CITATIONS    42 PUBLICATIONS   151 CITATIONS   

SEE PROFILE SEE PROFILE

Sajal Kumar Das

Rajshahi University of Engineering & Technology
64 PUBLICATIONS   377 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Brain Signal Analysis and Detection View project

Robust Control of Electro-Mechanical Systems View project

All content following this page was uploaded by Mizanur Rahman on 14 August 2018.

The user has requested enhancement of the downloaded file.

2017 3rd International Conference on Electrical Information and Communication Technology (EICT), 7-9 December 2017,
Khulna, Bangladesh

A Comparative Study of LQR, LQG, and Integral LQG

Controller for Frequency Control of Interconnected Smart
Grid

Mizanur Rahman∗ , Subroto K. Sarkar∗ Sajal K. Das∗ and Yuan Miao†

∗ Departmentof Mechatronics Engineering, Rajshahi University of Engineering & Technology, Bangladesh.
† Departmentof Information Technology, College of Engineering and Science, Victoria University, Australia.
Email: [email protected] ; [email protected]; [email protected]; [email protected]

Abstract—Renewable energy plays the vital role to increase technology as power engineering, information communica-
the power generation of smart grid. A smart grid consists tion control and instrumentation.
of smart meters, smart appliances, energy efficient resources The components use in a smart grid are electric power
and renewable energy resources. The integration of renewable generators, electric power substation, transmission and dis-
energy in a smart grid is challenging because of its frequency tribution lines, controllers, smart meter and collector nodes
fluctuation in the power grid. The Plug-in hybrid electric [?]. Electronic controller uses in power generators and elec-
vehicles (PHEV) are expected to install in customer side. The tric power substation to control the generation and flow
PHEV can be applied to stabilize the power and frequency of electric power. The collector nodes communicate with
fluctuation. This paper shows a comparative study of Linear each of other using Zigbee or two way mesh wireless
Quadratic Regulator (LQR), Linear Gaussian Control (LQG) communication network. Advanced Metering Infrastructure
and Integral Linear Quadratic Gaussian Controller (ILQG) (AMI) uses to communicate with collector nodes and utility
to control the frequency of interconnected power systems. grid. Smart grid uses a control system that allows intelligent
In order to maintain the frequency stabilization, the three monitoring mechanism and it makes electric power flow
controllers are developed in this paper. The performance of flexible and detail.
these controllers is presented using Matlab simulation. The PHEV is connected to smart grid due to excessive usage
simulation results show that the integral linear quadratic of smart grid power [?]. It makes additional load on a
gaussian controller provides best as compared to LQR and
smart grid and increases time distribution and reduces the
life of the power equipments. The function of PHEV is
LQG controller.
to reduce the transportation cost and integrates the energy
storage unit in power grid. The renewable energy penetrates
Keywords: Load-frequency control, LQR, LQG, Integral
in a smart grid and it causes frequency fluctuation. The
Linear Quadratic Regulator, Smart grid, Plug-in hybrid
frequency fluctuation makes system unstability. The energy
electric vehicles.
storage units increase the reliability and efficiency of grid
and store energy from grid and renewable sources. During
1. Introduction stored energy period, the performance of PHEV can be
deviated due to variation of renewable energy sources such
as wind power and solar strength. The variation of these
Over the past few decades, there have been a significant factors ensures the frequency fluctuation that makes the grid
change in the way to produce and transmit electricity [1]. instability [3].
A traditional grid uses electromechanical and solid state A number of researches have been done to control the
devices to produce power which distributes through the frequency stabilization of plug-in hybrid electric vehicles
communication and protection equipment. The generation during charging period. The bidirectional charging and dis-
of power from grid faces several problems [2]. In order charging power control of PHEV has been proposed to con-
to remove the problems, a hub is investigated to connect trol the power fluctuation. The design of charging controller
the generation of power, transmission and consumption is done based on proportional control theory. It ensures the
equipment which caries the high amount cost. Smart power system stability only when the generation of power is greater
grid is a great solution to achieve efficient, reliable, and than the demand of power, whereas it can not control the
communication-rich system. A smart grid is one kind of system stability. This limitation restricts the application of
electrical grid which can deliver electricity from multiple charging controller. The charging or power controller also
suppliers to end user using two way communication, mul- presents the unbalance frequency in the presence of various
tiple distributed intelligent entities and large scale real time uncertainty such as the variation of wind power parameters
data collection. Smart grid technology is a multidisciplinary [?]. However, the lack of robustness is another drawback of

978-1-5386-2307-7/17/$31.00
c 2017 IEEE 1
charging controller. MW respectively. The power capacities of WF, THP and
Robust H∞ [4] controller has been proposed to over- load in area 2 are 2530 MW, 5560 MW and 7090 MW
come the limitation of charging controller. The advantage of respectively [?]. The number of PHEV in area 1 is 640000
H∞ controller over other controller is that it can applicable and the number of PHEV in area 2 is 200000 respectively.
to problem involving a multivariate systems. The design The rated power capacity for one PHEV is 5 KW. Hence
of H∞ controller is done using H∞ norms minimization total power capacities of PHEV in both areas are 3200 MW
which deals with the require of high level of mathematical and 100 MW respectively.
knowledge. However, in many practical control constraints, Each area has local control center which sends the
the design of H∞ controller is challenging. Fuzzy logic control signal via smart meter to the PHEV. In both areas,
controller has been used for frequency stabilization for the frequency fluctuates because of intermittent wind power
two area interconnected system. The fuzzy logic can deal and random load changes. The capacity of turbine (TB)
with change in system parameter. Since the parameters of and governor (GOV) is inadequate to keep the frequency
controller can be generally changed very quickly, which fluctuation in a reasonable range. In both areas, the PHEV
make control algorithm complicated and unstable transient is installed to corresponds with this problem. The dynamic
response. Model predictive controller (MPC) can be ap- response of PHEV is faster than TB and GOV of THP. The
plied for desirable performance in multi area power system. PHEV is used for suppressing the peak value of frequency
The MPC controller can give fast response and robustness deviation against the abrupt load change. On the other hand,
against system parameter uncertainties but MPC suffers the GOV and TB of THP are corresponds to eliminate the
calculation burden problem. stead-state error of frequency deviation. As a result, the
The linear quadratic regulator (LQR) is an optimal con- TB and GOV design are discarded in the control design
troller that provides practical feedback gain. The controller of PHEV.
has been proposed for frequency stabilization and it can The PHEV is in charge of seducing the peak value
minimize the cost function. The performance of LQR may of frequency deviation against the abrupt load changes.
be deviated due to the presence of system noise. The LQG Linearized model of two area interconnected power system
controller is quite similar with LQR controller. It is able is shown in Figure 1. In both areas of Figure 1 the generator
to provide stable performance under system noise and un- and the frequency sensitive load are represented by the first
certainty. However, integral LQG controller can give better order transfer function with inertia constant M and damping
performance than the others. The motivation for designing a coefficient D. Using first order transfer function with gain
integral LQG controller is to provide large bandwidth, large KP HEV , i =1,2 and time constant Tp is in series with
gain and phase margin. controller with gain and time constant. The model of wind
In this paper the comparative study of LQR, LQG and power and load changes are considered as ∆PW and ∆PL
integral LQG controller is presented for frequency stabi- respectively. In both areas the presented controller is 1st
lization of a two area interconnected system. The aim of order transfer function with time constant TACEi and the
this paper is to develop high performance controller for area control error is (ACEi ). Here,
frequency stabilization within interconnected power system. ACE1 = fo Ksystem1 ∆f1 − ∆P21
The goal of designing controller is to achieve fast transient ACE2 = fo Ksystem2 ∆f2 − a21 ∆P21
response, zero steady state error and remove the frequency where, Ksystem1 and Ksystem2 are the system constant of
fluctuation problem. The contribution of this paper is to area 1 and 2 respectively, fo is the normal system frequency,
find the robust control strategy for frequency stabilization a21 is the area capacity ratio, ∆P21 is the deviation of tie
of smart grid within PHEV. The controller is designed to line power from area 2 to area 1, ∆f1 and ∆f2 is the devia-
achieve the robust performance of frequency stabilization. tion of frequency in area 1 and area 2 and T is synchronizing
The results presented in this paper provide the optimum power coefficient. The area control error (ACEi ) sends the
frequency stabilization for the interconnected system. The signal from controller to PHEV using smart meters. Signal
comparison analysis is important for smart grid as it paves is sent to PHEV via the smart meter. The load frequency
the way for frequency stabilization for high performance. control (LFC) fLF C1 , fLF C2 is considered as a input signal
The rest of the paper is organized as follows. Section
for the area 1 and area 2 respectively. The system parameters
2 presents the modeling of smart grid. Controller design
are given in Table 1.
for the system is carried out in Section 3. Performance
evaluation of the controllers for smart grid is presented in Consider the following state space model for two area
Section 4. The paper is concluded in Section 5. interconnected linear time invariant system

2. System Modelling ∆Ẋ = A∆X + B∆u (1)

Two area interconnected power system shown in Figure

1 is used for the study of smart grid. Each area consists of
∆Y = C∆X (2)
wind farm (WF), thermal power plant (THP), load frequency
control (LFC), PHEV and load. In area 1, Power capacities
of WF, THP and load are 12038 MW, 33090 MW and 7090 Where A is the system matrix, B is the input matrix and C

2
 Wind power change
PW 1 +

Load
change  PL1 _

PPHEV 1
Area 1
LFC 1 fo +
1
 f1
1 / K PHEV1 _
f o K system1 + TD s  1

 f1
1 1
Controller 1 / K PHEV 1
 P 21
+
TACE1s  1 M 1s  D1
_ f LFC 1
PHEV 1
Power
+ system 1
_ 1 1 0.7 +
5 0.25s  1 9s  1
25 s
_ +
0 .2 s  1 _
0 .3
Thermal power plant 1
+

T
_

 P21 Wind power change s

 PW 2 + Synchronizing
Area capacity power
a21 Load  PL 2 _ coefficient
ratio
change
_

25 _
0 .2 s  1 _ 1 1 0.7 +
5 0.25s  1 9s  1
s +
+
_

f2
Thermal power plant 2 0 .3 1
M 2 s  D2
a 21  P21 _
f PHEV 2 Power
f2
LFC 2
1
TACE 2 s  1
Controller 1 / K PHEV 2
+ system 2
f o K system 2 +
1 / K PHEV 2 1 _
TD s  1
fo
LFC 2
+
 PPHEV 2 Area 2

Figure 1: Linearized model of two-area interconnected system

TABLE 1: System parameter values

Parameters Area 1 Area 2
 
0 0
Reference frequency f0 (Hz) 50 50 0 0
Inertia Constant M (s) 9 9 K3

0
  
0 1 0 0 0 0 0
Frequency characteristic of load except B= 0 0 ;C =
 
2 2 0 0 0 0 0 1 0
PHEVD(pu) 0 0 

ACE calculation time constant TACE (s) 10 10 0 0 
Frequency bias factor K( system) 10 10 0 K4
Frequency Characteristic per PHEV 2.5
2.5 Here, K1 = Ksystem1 , K2 = Ksystem2 , K3 =
(KW/Hz)
Number of PHEV 640000 640000 KP HEV 1 , K4 = KP HEV 2 and T2 = TACE2 the state
Total frequency characteristic KP HEV 1600 1600 vector ∆X = [∆f1 , ACE1 , ∆PP HEV 2 ]T . The input vec-
1
tor and the output vector for the two area system is
Time constant of PHEV Tp 1
∆u = [fLF C1 , fLF C2 ]T and ∆Y = [ACE1 , ACE2 ]T ,
Tie-line stiffness coefficient T (s) 5 5
respectively.
Area capacity ratio a(21) 0.17 0.17

3. Controller Design
is the output matrix as follows:
 −D1
0 −1 −1
0 0 0
 3.1. LQR Controller Design
M1 M1 M1
 −fo K1 −1 0 −1
0 0 0 
 T1 T1 T1  The LQR is defined as where the system dynamics
 K 3 fo
 0 −1 0 0 0 0 
 are described as set of differential equations and the cost
A= −T 0 0 0 T 0 0 ;
function is described as quadratic function [5]. Using mathe-
−a21 −D2 −1 
 
 0 0 0 0 matical algorithm the cost function has been minimized with
 M2 M2 M2 
 0 −a21 −fo K1 −1
0 0 T2 T2 T2 0 
0 0 0 0 K4 fo 0 −1
3
 TABLE 2: Design parameter values for integral LQG
Design R
γ1 γ2 γ3 Q
parameter
Value 1 × 1012 1 × 10−8 1 × 106 1 × 107 5 × 10−3

weighting factors. The performance criterion of equation (1) vector at iteration n-1 including measurement n-1.
and (2) determined by the cost function is as follows: Step 2: parameter covariance matrix propagation from iter-
Z ∞ ation n-1 to n:
J∞ = [xT (t)Qx(t) + uT (t)Ru(t)]dt (3)
0 P̂ − (tn ) = φ.(τ )P + t(n−1) φ.(τ ) + Qtn (10)
where Q is non negative definite and R is R = RT > 0 T
where φ(τ )φ(τ ) is the state propagation matrix and the
positive definite. Now the optimal control position of J is transpose calculated for time spacing t. P − (tn ) is the
found using the state feedback law. u(t) = −Kx(t) where parameter covariance matrix at iteration n not including
k is the controller gain that is determined by following way measurement n. Q(tn ) is the process covariance matrix.
K = R−1 B T P (4) Step 3: Kalman gain computation:

and here P is the unique positive definite solution to K(tn ) = P − (tn )H T [HP − (tn )H T + R(tn )]−1 (11)
Algebric Riccati Equcation (ARE). The ARE is as follows where P − (tn ) is the parameter covariance matrix at iteration
T
A P + P A − P BR −1 T
B P +Q=0 (5) n not including measurement n. The design matrix transpose
is HH T . The Kalman Gain and measurement covariance
3.2. LQG Controller Design matrix are K(tn ) and R(tn ) respectively.
Step 4: state vector update:
The LQG controller is the combination of Kalman fil- x̂+ (tn ) = x̂− (tn ) + K(tn )[y(tn ) − H x̂− (tn )] (12)
ter with Linear Quadratic Regulator (LQR). LQG control
can be applied to linear time invariant systems as well as where the estimate of the state vectors at iteration n and not
linear time variant systems. It deals with uncertain linear including measurement n are x̂+ (tn ) and x̂− (tn ) respec-
systems disturbed by additive white gaussian noise, having tively. The measurement vector is represented as y(tn ). The
incomplete state information [?]. Practically it is used for kalman gain is K(tn ) and The design matrix is shown as
predicting future courses of dynamic systems that are likely H.
as: Step 5: Parameter covariance matrix update:
ẋ = F x + Gu + w (6) P (t+ −
n ) = [I − K(tn )H]P (tn ) (13)
the measurement may be linearly related. where P (t+ −
n ) and P (tn ) is the parameter covariance matrix
z = Hx + v (7) at iteration n, not including measurement n. If kalman gain
is deliberately set sub-optimal then use
The process noise matrix Q with noise vector w and mea-
−
surement noise matrix R is related to the measurement noise P (t+
n ) = [I − K(tn )H]P (tn )[I − K(tn )H]
T
(14)
vector v according to
Q = E[wwT ] Time Update(Predictor): update expected value of x is
T
(8) shown as follows
R = E[vv ]
Here, E represents expected value or average value of the x̂+ = Ax̂(k−1) + Buk (15)
parameter of the system. The update error covariance matrix is p represented as
A kalman filter is optimal recursive data processing
algorithm. It computes all avilable data, regardless of their Pk− (+) = AP(k−1) AT + Q (16)
precession, to estimate the current value of the variable of The optimal kalman gain Kk is as
interest. Operation of kalman filter can be divided with few
steps: (1) state vector propagation, (2) parameter covariance Kk = Pk− H T (HPk H T + R)−1 (17)
matrix propagation, (3) Compute Kalman Gain, (4) State
vector update, (5) Parameter Covariance Matrix update 3.3. Integral Linear Quadratic Gaussian Controller
Step 1: State Vector Propagation: Design
x̂− (tn ) = φ.(τ )x̂+ + tn−1
(9) First consider the state space model for the system is
R = E[vv T ].
ẋ(t) = Ax(t) + Bu(t) + D1 w(t) (18)
where x̂− (tn ) estimate of state vector at iteration n and φ(τ )
is state propagation matrix and x̂+ (tn ) estimate of the state y(t) = Cx(t) + D2 w(t) (19)

4
where x represents state vector, u is the input to the system, TABLE 3: Settling Time Comparison
y is the measured output from the plant and w is a gaussian settling settling settling
settling
white noise disturbance acting on the system. A represents time for
time for time for time for
the system matrix, B is the input matrix, C is the output input 1 input 1 input 1
Controller input 1
and and and
matrix, and D1 and D2 is the system noise matrices. An to output
output 2 output 3 output 4
optimal LQG controller design will begin form that state 1 (in sec)
(in sec) (in sec) (in sec)
space model. Desiging of optimal LQG controller with the LQR
4.16 347 374 0.675
feedback controller designed in such a way that it minimizes Controller
LQG 8.26
cost function. Controller
4.77 6.83 11.5
1 T ILQG
Z
0.0761 1.03 1.3 1.38
(x(t)T Qx(t)+u(t)T Ru(t))dt . (20)

J = lim E Controller
T →+∞ T 0
Here, Q ≥ 0 and R > 0 are symmetric weighting matrices.
E[.] is the expected value. In the cost function (20), the where PK ≥ is the solution of the following Riccati equa-
term xT Qx corresponds to a requirement to minimize the tion:
states of the system. the term uT Ru corresponds to the −1
P˙K = Af PK + PK ATf − PK CfT RK Cf PK + QK (27)
requirement to minimize the size of control inputs. The
selection of matrices Q and R in the cost function (20) Here, QK ≥ 0 and RK > 0 are process and measurement
depends on the desired performance objective of the system. noise matrices defined as
Minimizing the tracking error between the command signal  2
γ2 0

2
and measured output is the main control objective. Integarl QK = γ1 RK = (28)
0 γ32
controller can be used in smart grid, because it has excellent
low frequency tracking performance. The updated plant with where γ1 is the standard deviation associated with the
an integral controller of the plant can written as process noise w1 . γ2 and γ3 are the standard deviation as-
sociated with the output sensor noise w2 and the augmented
x˙f (t) = Af xf (t) + Bf u(t) + Bf w1 (21) integral output sensor noise is w3 .
zf (t) = Cf xf (t) + v (22) The feedback gain matrix L in (25) is given by
Here, v = [w2 w3 ]TR is the augmented state and output
−1
vectors are xf = [x ydt]T and zf = [z1 z2 ]T . The L = RC B f PC (29)
matrices
  for the augmented
  are defined as Af =
system
where PC ≥ 0 is the solution of the following matrix riccati
A 0 B C 0
, Bf = , and Cf = where A, B , C equation:
C 0 0 C I
−1
are defined in (18)-(19). P˙C = Af PC + PC ATf − PC CfT RC Cf PC + QC (30)
The quantity w1 is the mechanical noise entering into
the system which is assumed as the gaussian white noise Here, RC = r is the controller weighting matrix and
with variance γ12 . The quantity w2 is the sensor noise from 
1 0

the output y of the system which is also assumed to be QC = CfT C (31)
0 q f
gaussian white noise with variance γ22 . For the augmented
plant outlined in (21)-(22), the integral LQG controller is The controller design parameters γ1 , γ2 , γ3 , r and q are
designed by defining a quadratic cost function of the form adjusted for good controller performance. This includes a
requirement that the control system has a suitable gain and
Z T phase margins and a reasonable controller bandwidth.
1
J = lim E (xT Qx + f (y)T QI f (y) + uT Ru)dt,
T →+∞ T 0
(23) 4. Performance Evaluation
RT
where f (y) = 0 y(τ )dτ , Q ≥ 0, QI ≥ 0, and R > 0
are weighting matrices associated with the state, the integral The performance of the controllers is investigated in
state, and the control input respectively. this section. The simulation results of the proposed con-
The integral LQG controller is represented as trollers are presented using Matlab. The results are shown
for individual controller. The parameter values of Table 1 is
b˙f (t) = Af x
x bf (t) + Bf u(t) + K(zf − Cf x
bf ) (24) used to measure the performance of proposed system. The
u(t) = −Lx bf . (25) performance of the controller can be varied due to change
in system parameter, renewable energy source that ensures
Here, x
bf is the augmented state vector estimated by linear the instable performance of smart grid.
Kalman filter and K is the associated Kalman gain defined The open-loop step response for two area interconnected
as system is shown in Figure 2(a) that provides undesirable
−1 nature of responses. This causes poor frequency stabilization
K = PK CfT RK (26) for the system. The closed-loop step response for two area

5
 In(1) In(2)
2
1
6
Open-loop Step Response
Closed-loop Step Response
0

Amplitude
5

4 −1
Pitch Angle (rad)

3
2
2 1
1
0
0
−1
-1
0 5 10 15 20 25
0 5 0 5
Time (seconds) Time (seconds)
(a) (b)

Figure 2: (a) The dashed line (- -) represent open-loop step response for the plant, (b) The blue (- -) solid line , green solid
line (-) and red solid (-) line represent closed-loop step response of integral LQG , LQG and LQR controller respectively

proposed controllers for frequency stabilization of two area

In(1) In(2) interconnected power system.
500
5. Conclusion
0
Magnitude (dB)

The design of robust frequency controller is presented

in this paper for frequency stabilization control for a two
−500
area interconnected system . The comparative study of con-
trollers is designed for stabilization of frequency fluctuation
0 in the smart grid with renewable energy. The results of
this paper is of interest to engineering community because
−200
the comparative study presented in the paper provides the
way of optimum frequency stabilization for a two area
−400 −5 −5 0 5
10 10
0
10 10
5
10 10 interconnected power system. Simulation results in two area
Frequency interconnected power system and comparison study shows
Figure 3: Comparison of bode diagram. The green (- - the superior robustness has integral linear quadratic gaussian
) dashed line represents LQR controller. The black (- -) controller. The controller has superior performance against
dashed line shows LQG controller and red (- -) dashed line system parameter variation in comparison with conventional
represents integral LQG controller. The blue (-) solid line controllers.
line represents open loop system response.
References

interconnected system given in Figure 2(b). The comparison [1] A. S. Ali, Smart grids: opportunities, developments, and trends.
Springer, 2013.
results are reported in Table 3. The settling time for integral
LQG is very small for every output with respect to every [2] A. Yokoyama, “Smarter grid ii,” IEEJ J, vol. 30, no. 3, pp. 163–167,
2010.
input and makes it superior controller in compared with
LQR and LQG controller. The integral LQG controller has [3] M. Takagi, K. Yamaji, and H. Yamamoto, “Power system stabilization
by charging power management of plug-in hybrid electric vehicles
zero steady state error and less rising time for the system. with lfc signal,” in Vehicle Power and Propulsion Conference, 2009.
The transient time period is also very small of integral VPPC’09. IEEE. IEEE, 2009, pp. 822–826.
LQG controller in comparison with others.The closed-loop [4] J. Pahasa and I. Ngamroo, “Phevs bidirectional charging/discharging
frequency response of two area interconnected system is and soc control for microgrid frequency stabilization using multiple
shown in Figure 3. The proposed controllers have stable mpc,” IEEE Transactions on Smart Grid, vol. 6, no. 2, pp. 526–533,
poles for the nominal system. The integral LQG controller 2015.
is able to provide high gain and phase margin for each output [5] H. Ko, K. Lee, and H. Kim, “An intelligent based lqr controller de-
with its corresponding input. For all cases, the integral LQG sign to power system stabilization,” Electric Power Systems Research,
vol. 71, no. 1, pp. 1–9, 2004.
controller shows the high performance compared to others

View publication stats