Available online at www.sciencedirect.

com

ScienceDirect
Procedia Computer Science 109C (2017) 754–762

The 7th International Conference on Sustainable Energy Information Technology

(SEIT 2017)

Implementation of Proportional-Integral-Observer Techniques for

Load Frequency Control of Power System
Arkan Ahmed Husseina*, Sabah Sadeem Saliha, Yaroub Ghazi Ghasma
a
Electrical Engineering Department, Engineering College, Tikrit University,Tikrit,IRAQ

Abstract

This paper presents the implementation of Proportional-Integral-Observer(PI-Observer) based state feedback control for Load
Frequency of single area isolated power system model. The general application of PI-Observer has been realized in areas of
robotic and mechanical system. The PI-Observer is used to estimate both the original states and the unknown inputs generally
disturbances, modeling errors or the other non-linearities present in the system. The estimation error of the PI-Observer is
gradually reduced by the proportional and integral feedback loop that appreciably improves the observer dynamics. A single ar ea
linear power system model is taken as a test case to implement the proposed methodology which is then compared with full state
Luenberger observer based state feedback control for load frequency control of power system. MATLAB/SIMULINK is
extensively used for modeling & simulation of power system and observer. The dynamic responses are compared and discussed
on the basis of error dynamics, system states, peak overshoot and settling time. Lastly it was found that response of PI -observer
based state feedback control scheme is quite appreciable as compared to Luenberger observer based state feedback schemes.

1877-0509 © 2017 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the Conference Program Chairs.
Keywords: Load Frequency Control; Luenberger observer; LQR ;PI-Observer systems; state feedback control.

1. Introduction

In the complex structure of geographically distributed and interconnected power system, the generation and load
within each area is to be matched and the scheduled power interchange and system frequency should be cl ose to their
nominal values for the stable operation of power system. The general idea of controlling the frequency is to maintain
the balance between the generated power and the consumed power. Since the existence of alternating current power
systems, different philosophies have been applied to maintain the supply frequency. The most common control modes
are the Isochronous control, Droop control and Load Frequency Control. In the isochronous control mode, a big

* Corresponding author. Tel.: +9647717170490.

E-mail address: aalganabe@tu.edu.iq

1877-0509 © 2017 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the Conference Program Chairs.
10.1016/j.procs.2017.05.307
 Arkan Ahmed Hussein et al. / Procedia Computer Science 109C (2017) 754–762 755

generator will be assigned the task of maintaining the frequency and the rest of generators will be running at constant
power output. Whereas in the droop control mode, all generators will respond to the frequency deviation. However
Load Frequency control (LFC) is achieved by adding a supervisory control loop to the droop control loop in order to
achieve better performance. The main aim of LFC is to maintain zero steady state frequency deviation and to track the
load demands. The Load Frequency controller design with better performance has received considerable attention
during the past years and many control strategies have been developed for LFC problem 1. The expression of PI-
Observer was founded by Wojciechowski for control single-input single-output (SISO) systems and later it is used by
other to control multivariable systems. The PI-observer and Luenberger observer has the same structure as the
extended state observer for disturbance estimations, however the PI-observer has an extra integral feedback loop of
estimation error. This loop with proportional feedback loop generate the name of PI -observer and offers additional
degrees of freedom for the estimation task in two aspects, first in enhanced robustness of estimations and sec ond in
estimations of step disturbance and variation in system parameters. The PI techniques can be estimate the disturbance
and unknown effects of system in the form of unknown input of the states. In this paper the development of PI -
Observer design and its applications for Load Frequency control for power system is explored and discussed.

2. General design of observer

In practice all the states of the system cannot be measured due to number of reasons including cost or that state
may not be physically measurable therefore a method for reconstructing system states by using measured inputs and
outputs of the system is the design basis of observer technique which is first proposed and developed by Luenberger
in the early sixties of the last century 2. Later PI observer has been introduced to improve the robustness of
estimation against the both known and unknown input effecting to the system, the so-called Unknown Input
Observer (UIO). In this design, the disturbances can be considered as unknown inputs. Se veral paper concentrate on
state estimation with unknown inputs . The observer technique has been continuously developing upgrading and
different directions of observer design are improved, for example optimal observer design and nonlinear observer
design on the basis of different structure of observers for different engineering applications can be classified as
shown in fig 1.

Fig. 1. Classification of Observer

The basic use of observer is to estimate the unmeasured states and can also be used as distur bance estimation
by augmenting an additional state in the basic state space equation of the system3.

2.1. Structure of Luenberger Observer

State reconstruction and estimation are used in different types of applications like signal processing,
telecommunications, and fault detection. The gains of the observer is the main parameter to estimate unmeasured
states from the known inputs and the outputs of the system. An observer estimates all of the system state variables
and has mathematical model to produce an estimate of the actual state vector x however the observer dynamics will
never exactly equal the system dynamics and the performance of the observer depends on the ou tput error which
756 Arkan Ahmed Hussein et al. / Procedia Computer Science 109C (2017) 754–762

should be minimized by tuning the observer gains. The basic arrangement of the Luenberger observer is shown in
fig 2. The Power system has internal state x, input u and output y . the Luenberger observer that has two parts (i) an
exact model of the power system dynamics (A,B,C) (ii) plus an error correcting part K o(y- yest) . The Ko is called the
observer gain and the observer has same internal states x est as that of power system. If L is optimally chosen then the

observer estimates closely matches with power system states.

Fig. 2. Luenberger Observer.

For the system which is expressed as below

!" = 𝐴𝐴 𝐴𝐴 𝐴𝐴 + 𝐴𝐴 𝐴𝐴(𝐴𝐴) (1)
!"

𝐴𝐴 = 𝐴𝐴 𝐴𝐴(𝐴𝐴) (2)
For the estimate of x̂ the state vector is in given below
!"!"#
= 𝐴𝐴 𝐴𝐴!"# + 𝐴𝐴𝐴𝐴 + 𝐴𝐴! (𝐴𝐴 − 𝐴𝐴 𝐴𝐴!"# ) (3)
!"

Where Ko is the observer gain matrix.

From equation (1) and (3) error dynamic is given below.
𝐴𝐴 = 𝐴𝐴 − 𝐴𝐴!"# (4)
!" = 𝐴𝐴 − 𝐴𝐴 ! 𝐴𝐴 (5)
!"

And the equation for Luenberger observer is given below

!"!"#
= (𝐴𝐴 − 𝐴𝐴! ) 𝐴𝐴!"# + 𝐴𝐴𝐴𝐴 + 𝐴𝐴! 𝐴𝐴 (6)
!"

The dynamic behavior of the error depends on the Eigen values which is expressed as in equation (6).
(𝐴𝐴 − 𝐴𝐴! 𝐴𝐴) (7)

2.2. Structure of PI-Observer

Basically it is an extension of Luenberger’s observer and have been first proposed by Wojciechowsky for single
input single output linear time invariant systems. However it has been extended to multivariable systems by
Kaczorek and shafai and carroll for robustness against parameter variation and step disturbances. Some successful
applications are found in connection with full state feedback robustness recovery loop transfer recovery and fault
detection problems4. There are many non-linear systems that cannot be modelled by equation (1) however due to
 Arkan Ahmed Hussein et al. / Procedia Computer Science 109C (2017) 754–762 757

inherent non-linear effects the dynamics of system is represented as below

!" = 𝐴𝐴 𝐴𝐴 𝐴𝐴 + 𝐴𝐴𝐴𝐴(𝐴𝐴) + 𝐴𝐴𝐴𝐴(𝐴𝐴, 𝐴𝐴, 𝐴𝐴) (8)
!"
The vector function f(x, u, t) describes the nonlinearities, unknown inputs and un-modelled dynamics of the
system. It is the nonlinear function of states, control inputs and time. The matrix N is the corresponding distribution
matrix with full column rank. It was discovered by Ackermann and soffker that by assuming the disturbance as
piecewise constant the PI-observer finely reconstruct the states and in combination with controls gives a very good
compensation effects. However the performance of the PI-observer defined by Luenberger in equation (4) can only
be achieved by the optimum gains of PI loop used in the observer. Since PI-observer is an extension of the
Luenberger observer and not only uses the information proportional to estimation but also an integral of the
estimation error that gives a better estimation performance. Fig 3 shows the basic structure of PI-Observer for the
system expressed by equation (1) and (2).

Fig. 3. PI-observer

The dynamics of PI-observer can be expressed as

!"!"#
= 𝐴𝐴 𝐴𝐴!"# + 𝐴𝐴! 𝐴𝐴!"# + 𝐴𝐴𝐴𝐴(𝐴𝐴) + 𝐴𝐴! (𝐴𝐴 − 𝐴𝐴!"# ) (9)
!"

𝐴𝐴!"# = 𝐴𝐴! (𝐴𝐴 − 𝐴𝐴!"# ) (10)

Where 𝐴𝐴!"# = 𝐴𝐴𝐴𝐴!"# and writing equation (8) in matrix form

!"!"#
!" 𝐴𝐴 𝐴𝐴! 𝐴𝐴!"# 𝐴𝐴 𝐴𝐴!
= + 𝐴𝐴 + (𝐴𝐴 − 𝐴𝐴!"#) (11)
!"!"#
0 0 𝐴𝐴!"# 0 𝐴𝐴!
!"

The optimum selection of gains L 1, L2 and L3 will give the approximate estimate of the original states by the
observer.

3. Power system model for LFC

For the stable operation of power system matching of total generation with total load demand and associated
system losses is highly desirable but since the operating point of a power system changes, and hence, these systems
may experience deviations in nominal system frequency and scheduled power exchanges to other areas, which may
yield hazardous effects if not controlled at an appropriate time.
758 Arkan Ahmed Hussein et al. / Procedia Computer Science 109C (2017) 754–762

The supplementary control included to the governor with the help of a signal directly proportional to the
frequency deviation plus its integral constitutes the classical approach to the LFC of power systems . Fig 4 shows the
classical overview of the Load Frequency control loop that is used as secondary control loop adjusts power set
points of the generators to compensate for frequency error that is not corrected by the primary control where as
primary control refers to control actions that are done locally at the power plant level and are based on the set points
for frequency and power.

Fig. 4. Control Structure for frequency control system

LFC problem has been dealt with extensively for more than three decades and most of the work reported has
been performed by considering linearized model of single and multi-area power system model. But researchers also
uses non-linearities such as generation rate constraints(GRC), governor dead-band and boiler dynamics etc for
different LFC schemes5.

3.1 State space model of power system model

The power system model considered being a linear continuous-time dynamic system can be represented by the
standard state space model as-

!" = 𝐴𝐴 𝐴𝐴 𝐴𝐴 + 𝐴𝐴𝐴𝐴(𝐴𝐴) + Γ𝐴𝐴 ! (12)

!"

𝐴𝐴 = 𝐴𝐴𝐴𝐴 (13)
Where X, U, Pd and Y are the state, control, disturbance and output vectors respectively and A, B, Γ and C are the
system, input, disturbance and output matrices respectively. These matrices are constant and of compatible
dimensions associated with them which depend on the system parameters and the operating point 6,7. The state space
model for isolated single area system is single input single output system as shown in figure 1 and the rate of change
of frequency can be written as below-
 K KP
1 P
Δf (t)   Δf(t)  ΔP (t)  ΔP (t) (14)
T T g T d
P P P
The rate of change of mechanical power can be written as below
 1 1
ΔP (t)   ΔP  ΔX (15)
g T g T g
s s
The change of governor position is given as-
 Arkan Ahmed Hussein et al. / Procedia Computer Science 109C (2017) 754–762 759


1 1 1 1 (16)
Δf(t)  ΔE(t)  u(t)
ΔX (t)   T ΔX g  T
g RTg g Tg g
Here integral control of Δf (t) can be written as-
t
ΔE(t)  K  Δf (t )dt (17)
i
0
For regulation the controller should make the below equation zero.

ΔE (t)  K Δf(t) (18)
i
Where Ki is the integral controller gain.
The sate vector for the power system model is given as-
T
x (t)  [ Δf (t) ΔP (t) ΔX (t) ΔE (t)] (19)
g g
And the system matrix is represented as below-

  1/T p 
 
K /T 0 0
pp
  T

,
 0  1/T  1/T 0 B
A  
s s   0 0 1/T 0
 1/T  1/T  g
  1/RTG 0
g g 
 

0  

 
K 0 0
 i 
C 1000  D 0   T
, , 
L
p
  K /T 0 0 0
p
The values for the time constants are given in Appendix A.
The system equation developed above are for single area isolated power system and is used for designing observer
and controller.

4. Observer based state feedback control for power system

Generally this type of scheme is used for disturbance rejection and in this paper the PI-observer based scheme is
implemented for LFC for single area power system model. The LQR technique based on the Algebraic Riccati
Equations (ARE) is used to calculate the state feedback gains for a chosen set of weighting matrices. These
weighting matrices regulate the penalties on the deviation in the trajectories of the es timated state variables and
control signal (u) and the performance index is given as-

1  T T
J (x Qx  u Ru) dt (20)
20
Where Q is nxn positive semi definite symmetric state cost weighting matrix R is mxm positive semi definite
symmetric control cost weighting matrix .The matrices and are chosen appropriately. The optimal controller that
minimizes the cost of the system in state variable form is a function of the present states of the system weighted by
the components of a constant gain matrix K of dimension mxn and can be defined as
𝐴𝐴 = −𝐴𝐴𝐴𝐴 (21)
K is obtained from the solution of the reduced matrix Riccati equation given below.
760 Arkan Ahmed Hussein et al. / Procedia Computer Science 109C (2017) 754–762

𝐴𝐴! 𝐴𝐴! + 𝐴𝐴! 𝐴𝐴 − 𝐴𝐴! 𝐴𝐴𝐴𝐴 !! 𝐴𝐴 ! 𝐴𝐴! + 𝐴𝐴 = 0 (22)

1 T
KR B P1 (23)

The acceptable solution of K is that for which the system remains stable.

5. System and results

The referenced model is developed and simulated in MATLAB/SIMULINK to observe the dynamics of PI -
observer and the associated control scheme with it. All the data referred for power system modelling is given in
Appendix A. At first Luenberger observer and PI-observer is developed for the single area power system model and
the corresponding block diagram of model is shown in fig 5, blocks represent the state space model of power system
with other block represent the Luenberger observer state space model and PI-observer state space model. The gains of
the observer are tuned by LQR technique.

Fig.5. Block diagram of model of power system with observers

For good observer design the gain should be calculated in such a way that the error dynamics should regulate to
zero. So that difference between actual states and the estimated states will converge to zero. The performance of
both the observer are analyzed by the error dynamics shown in fig 6.

Fig.6. Error dynamics of Luenberger and PI-observer.

It is quite clear that the negative peak overshoot of PI-observer is -0.025 which is quite high as compared to
Luenberger observer that has maximum negative overshoot of -0.01, as far as the settling time is concerned the error
 Arkan Ahmed Hussein et al. / Procedia Computer Science 109C (2017) 754–762 761

in PI-Observer settles around 13 sec as compared to Luenberger observer which takes more than 15 sec to converge
to zero. However the response PI-observer is slightly oscillatory due to the presence of proportional loop and the
gains are tuned by the LQR technique. Practically all the states cannot be measured thereby state variable feedback
control cannot be used for control of power system. Instead, available measurements are used to estimate the
unmeasured states and control law can be implemented using estimated states and the state estimation is usually
done by observers or state observers. In this scheme the estimated state by the Luenberger observer is feedback to
system through state control gain K which is calculated by LQR technique through optimum selection of Q and R
matrices. All the system parameters used for simulation are given in Appendix I. The block diagram of model of PI -
observer based state feedback control for power system is shown in fig 7.

Fig. 7. Power system with PI-Observer based state feedback

The scheme is same as above with a minor modification that error in states is fed to PI block of the observer for
convergence of error to zero and the estimated states is feedback to power system model via state gain K calculated
by LQR technique. After a perturbation of 1% in the steady state of power system the response of both the observer
based state feedback control is shown in fig 8, and it is clearly visible that peak overshoot of Luenberger observer
based state feedback control -0.05 which is slightly low as compared to PI-observer based state feedback control
which has value -0.06. Whereas second and third overshoot of PI-observer based state feedback control is slightly
lower than Luenberger observer based control. As far as settling time PI-observer controller settles around 8 sec
with less oscillation as compared to Luenberger-observer which settles around 9.5 sec. Lastly an interesting thing to
note is that the response of PI-Observer is quite faster than Luenberger observer that is due to presence of PI-loop in
the observer. Therefore PI-Observer based state feedback control is appreciable as compared to Luenberger based
state feedback control.

Fig. 8. Power system with Luenberger Observer and PI- Observer based state feedback control.
 762 Arkan Ahmed Hussein et al. / Procedia Computer Science 109C (2017) 754–762

6. Conclusion

The paper discusses, implements and compares the Luenberger observer and PI-Observer based state feedback
control law for linear isolated single area power system model. Modern control theory is used in the development o f
power system model, observer model and finding the optimal gains via LQR technique for observer and state
feedbacks. The Q and R for the observer gains and the controller gains are also selected optimally for the desired
stable closed loop response of the system. From simulation of the developed model it is inferred that response of PI -
observer is appreciable fast as compared to Luenberger observer , however the system response oscillations are more
as compared to latter observer. Since the convergence of estimation error of the PI-Observer takes more time than
Luenberger observer therefore it is expected and found by simulations that PI-Observer based state feedback control
law for power system after 1% perturbation also lags in response as compared to Luenberger observer. In view of
above discussion we can conclude that response of PI-Observer based state feedback control for linear power system
model perform well as compared to Luenberger observer based state feedback law.

Appendix A: Parameters for system simulation

Power rating of control area (P) = 2000MW
Rated frequency (f) = 60Hz
Regualtion constant ( R ) = 2.4 Hz/puMW
Governor time constant (Tg) = 0.08s
Steam turbine time constant (T t) = 0.3s
Time constant of power system (T P) = 20s
Gain for power system (Kp) = 120


1
0
0 0 0
 
Q    ,
1 0 0

0 0 0.01 0 

 
R  0.01
 
0.01 
0 0 0 

References:
1. Ibraheem, P. Kumar, and D. P. Kothari.Recent philosophies of automatic generation control strategies in power systems. IEEE Transactions on
Power Systems.2005; 20(1):346-357.
2. Hosseini S. H. and Etemadi A. H.Adaptive neuro-fuzzy inference system based automatic generation control. Electric Power Systems
Research.2008;78(7): 1230-1239.
3. E .Schrijver, J. Dijk van, and Nijmeijer. H. Equivalence of Disturbance Observer Structures for Linear Systems. In Proceedings of the 39th
IEEE Conference on Decision and Control. Sydney, Australia, 2000; 5: 4518–4519.
4. Doolla. S. and T.S. Bhatti. Automatic generation control of an isolated small-hydro power plant.
Electric Power Systems Research, 2006;76(9): 889-896.
5. Liu, Y., and Zhang. J. Minimal-Order Observer and Output-Feedback Stabilization Control Design of Stochastic Nonlinear Systems. Science in
China Series F: Information Sciences 2004; 47(4): 527–544.
6. Krener, A.J., and Xiao. M. Nonlinear Observer Design in the Siegel Domain. SIAM Journal on Control and Optimization.2002;41(3): 932–953.
7. Duan, G.R., Liu, G.P., and Thompson,S. Eigen structure assignment design for proportional-integral observers: continous-time case. IEEE
proc. Control Theory Appl.2001; 148:.263-267.