Electrical Power and Energy Systems 66 (2015) 67–77

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

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

Comparative performance analysis of teaching learning based

optimization for automatic load frequency control of multi-source
power systems
A.K. Barisal ⇑
Department of Electrical Engineering, V. S. S. U. T, Burla, Odisha, India

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 new population based parameter free optimization algorithm as teaching learning
Received 4 October 2013 based optimization (TLBO) and its application to automatic load frequency control (ALFC) of multi-source
Received in revised form 12 September 2014 power system having thermal, hydro and gas power plants. The proposed method is based on the effect of
Accepted 10 October 2014
the influence of teacher on the output of learners and the learners can enhance their knowledge by
interactions among themselves in a class. In this extensive study, the algorithm is applied in multi area
and multi-source realistic power system without and with DC link between two areas in order to tune the
Keywords:
PID controller which is used for automatic generation control (AGC). The potential and effectiveness of
Automatic load frequency control
Automatic generation control
the proposed algorithm is compared with that of differential evolution algorithm (DE) and optimal output
Dynamic performance feedback controller tuning performance for the same power systems. The dynamic performance of
Multi source power system proposed controller is investigated by different cost functions like integral of absolute error (IAE), integral
HVDC link of squared error (ISE), integral of time weighted squared error (ITSE) and integral of time multiplied
Teaching learning based optimization absolute error (ITAE) and the robustness of the optimized controller is verified by its response toward
changing in load and system parameters. It is found that the dynamic performance of the proposed
controller is better than that of recently published DE optimized controller and optimal output feedback
controller and also the proposed system is more robust and stable to wide changes in system loading,
parameters, size and locations of step load perturbation and different cost functions.
Ó 2014 Elsevier Ltd. All rights reserved.

Introduction sources of power generation and their corresponding participation

factor are more important for the study of LFC.
The purpose of an interconnected power system is to generate, Many researchers in the recent past have proposed several
exchange and control of electric energy with nominal system fre- strategies for LFC of power systems in order to maintain the system
quency, voltage profile and tie-line power interchange within their frequency and tie line power flow at their scheduled values during
prescribed limits. The automatic load frequency control which is normal operation and also during small perturbations. To meet the
basically a part of automatic generation control plays an important today’s stringent quality requirements, accurate-tools based realis-
role in power pool by maintaining scheduled system frequency and tic models with faster solution speed; high degree of reliability is
scheduled tie-line power in normal operation and during slow and required. While considerable progress has been made in the devel-
small perturbations. The modern power system is composed of opment of intelligent controllers and their applications to large
multiple sources of generation such as thermal, hydro gas and scale power system still remain a challenging area and a common
renewable energy power plants having many control areas or problem. It is found in literature survey that the early work on AGC
regions representing coherent group of generators. The area con- was started by Cohn [3] but the design of modern optimal control-
trol error as the controlled output of AGC is driven to zero in order ler concept for interconnected power system was initiated by
to make the frequency and tie line power deviations of control area Elgerd and Fosha [4]. The recent past control strategies for auto-
to zeros [1,2]. The environmental drive to promote renewable matic generation control of power system are reported by Kumar
energy invites new players to realistic power system with multiple and Kothari [5] which includes various control aspects for AGC
system incorporating with other additional devices. The gain
scheduling control method for AGC of interconnected power
⇑ Tel.: +91 6632430754; fax: +91 6632430204.
system was proposed by Lee et al. [6]. This control is different from
E-mail address: [email protected]

http://dx.doi.org/10.1016/j.ijepes.2014.10.019
0142-0615/Ó 2014 Elsevier Ltd. All rights reserved.
68 A.K. Barisal / Electrical Power and Energy Systems 66 (2015) 67–77

other control techniques in terms of robustness to wide range of iv. To compare the dynamic performance of TLBO based PID
operating conditions and also easy implementation. The growth controller to that of optimal controller and DE tuned control-
in size and complexity of electric power system due to non linear ler for multi-source multi area power system without and
load characteristics and variable operating points has necessitated with HVDC link.
the use of intelligence based methods to address satisfactorily the v. To carry out the sensitivity analysis for the proposed
performances under small perturbations. The intelligent control- TLBO-PID controller with its optimum parameters and its
lers such as fuzzy logic based gain scheduling [7], two fuzzy rules robustness to wide changes in loading pattern and several
for integral and proportional gains PI controller [8], adaptive fuzzy system parameters from their nominal values and also
gain scheduling method [9], genetic algorithm (GA) based PI and changes in size and locations of load perturbations and
PID controller [10], PSO based controller [11], self tuning fuzzy different cost functions.
type PID controller [12], a reinforcement learning approach [13],
adaptive neuro-fuzzy interference system (ANFIS) [14], evolution- System investigated
ary fuzzy PI controller [15], PSO based controller with fuzzy
application [16], differential evolution (DE) algorithm based PI Single area realistic power system
controller [17], optimal output feedback controller for multi-
source system [18], generalized neural network approach [19], The system under study is a realistic power system which
several classical controllers [20] for AGC studies in interconnected includes reheat thermal, hydro and gas generating units. The
power system, were preferred by researchers from the beginning of linearized model of single area system is given in Refs. [18,22]
these dates. and multi area system shown in Fig. 1 for simulation and LFC
It is found in literature survey that most of the researchers study. The nominal parameters of the system are provided in
adopt thermal–thermal or thermal hydro systems in AGC studies. Ref. [18] for single area and for multi area in appendix. Each unit
Surprisingly, very few papers in literature that considers a single has its regulation parameter and participation factor [22] which
area [18,22] or multi-area without or with HVDC link connecting decide the contribution to the nominal loading.
two areas [21,22] of realistic power system having generation from
thermal, hydro and gas units. Mohanty et al. have considered a two
Multi source multi area realistic power system with HVDC link
area AC–DC system with parallel tie lines for frequency stabiliza-
tion by using DE tuned PID controller [22] that better performs
The two area power system interconnected by parallel AC–DC
than an optimal output feedback controller [21] for similar power
tie lines [21,22] which comprises more practical combination of
system. Gozde et al. proposed the usage of craziness based PSO
generating units such as reheat thermal, hydro and gas units in
algorithm for AGC system for an interconnected thermal power
each area is simulated using MATLAB Simulink. Furthermore, the
plants in the year 2011 [23]. One year after, Gozde et al. [24] again
generators in each area may or may not participate in the LFC task
proposed the artificial bee colony based PI and PID controller
and the participation rates are not same for all participating
parameter tuning and its superior performance compared to PSO
generators. The summation of participation factors of all
with transient response analysis method in 2012.
participating generators is equal to unity in a control area. Transfer
There has been considerable progress in intelligent algorithm
function model of multi-source multi area with HVDC link with
based controller research work attempting to better control for
integral controllers is depicted in Fig. 1.
AGC systems [25]. Teaching learning based optimization (TLBO)
[26] is a new optimization technique developed by Rao et al. and
it has hardly been applied to tune the controller in AGC studies. Control strategy with objective function
The author has proposed a parameter free optimization algorithm
for tuning of controller for load frequency control of the present When designing a controller, the main purpose of controller
scenario above realistic power systems having multiple sources must be foremost in all considerations. This can include system
of power generation and its superior dynamic performances are dynamics, robustness to model uncertainty, ability to follow the
compared to that of recently published optimal controller [21] set point, sensitivity to measurement noise and attenuation of load
and DE tuned controller [22] for the same AGC system. disturbances. The Proportional Integral Derivative controller (PID)
Optimal parameter combinations for the DE-tuned controller is the widespread and popular feedback controller used in many
[22] are experimentally determined by conducting a series of modern industries. The popularity of PID controllers stems in part
experiments through simulation with six different DE parameter to their wide applicability to a variety of single input single output
settings [17] before conducting actual simulation to collect the applications. PID controller is often used when stability and fast
results. However, the proposed TLBO a parameter free algorithm response are required. It is reported in Ref. [22] that the DE tuned
is very simple in concept and easy implementation to tune the PID controller outperforms than the DE tuned PI and I controllers
controller of AGC for the realistic power system. In the present for multi-source multi area realistic power system. Keeping in view
work, a brief overview of TLBO technique with a flow chart for above, The PID controller is employed in the present paper for
the sake of completeness and better reading of the paper and also comparative performance analysis of similar power system.
request the readers to refer to [26,27] for more details of TLBO and There are four kinds of performance criteria, generally consid-
its application. ered in the control design. These are the integral of absolute error
The main investigations of the present work: (IAE), integral of squared error (ISE), integral of time weighted
squared error (ITSE) and integral of time multiplied absolute error
i. To propose a new nature inspired algorithm as teaching (ITAE). However, ITAE and ISE criterions are mostly used in AGC
learning based optimization (TLBO) for the load frequency studies for their better performance as compared to ISTE and IAE
control of the realistic power system. criterion. Systems with ITAE objective functions settle more
ii. To optimize the PID controller gains and study of its dynamic quickly than ISE method. Therefore, ITAE is a better objective func-
performances for above power system. tion among all and considered in present paper for comparative
iii. To compare the dynamic performance of TLBO based PID point of view.
controller to that of optimal controller and DE tuned control- The objective function for multi source power system may be
ler for single area system. defined as
 A.K. Barisal / Electrical Power and Energy Systems 66 (2015) 67–77 69

1/R1 Thermal power plant with Reheat turbine

Area-1
DEL PD1

1/R2 1 Kr*Tr.s+1 1 HVDC1

KT
1/R3 Tsg.s+1 Tr.s+1 Tt.s+1 Kdc1
B1 UT
Tdc1.s+1
Hydro power plant with Governor
ACE-1 Kps1
1 Trs.s+1 -Tws+1
Ki UH KH Tps1.s+1
Tgh.s+1 Trh.s+1 0.5*Tws+1
e s u Powe System1
Controller 1 Xc.s+1 Tcr.s+1 1 DEL F1 (s)
KG
UG bg.s+cg Yc.s+1 Tfc.s+1 Tcd.s+1
Gas Turbine power plant
DEL PT 2*pi*T12

Thermal power plant with Reheat turbine s

UT
a12 HVDC2
a12 1 Kr*Tr.s+1 1
KT Kdc2
Tsg.s+1 Tr.s+1 Tt.s+1
ACE-2 Tdc2.s+1

Hydro power plant with Governor Kps2

Ki UH
1 Trs.s+1 -Tws+1 Tps2.s+1
e s u KH
Power System2
Controller1 Tgh.s+1 Trh.s+1 0.5*Tws+1
DEL PD2 DEL F2 (s)
UG
1 Xc.s+1 Tcr.s+1 1
1/R1 KG
bg.s+cg Yc.s+1 Tfc.s+1 Tcd.s+1
B2 1/R2
1/R3 Gas Turbine power plant Area-2

Fig. 1. Transfer function model of multi-source multi area with HVDC link with integral controllers.

Z t sim
algorithm based optimization are applied increasingly due to their
J ¼ ITAE ¼ ðjDF 1 j þ jDF 2 j þ jDPTie jÞ  t  dt ð1Þ
0
simpler implementing, better performance of converging and less
execution time at present [25]. The controller parameters can be
where DF1 and DF2 are the frequency deviations of area-1 and adjusted very quickly in response to the changes in the plant
area-2, respectively. DPTie is the incremental change in tie line dynamics. The details of proposed TLBO algorithm are depicted in
power and tsim is the time range of simulation. Minimize J subject following section.
to PID

controller gains such as K ij min 6 K i 6 K ij max;

Intelligent algorithms
where i ¼ P; I; D and j ¼ 1; 2; 3: ð2Þ
The differential equation of PID controller of each unit of power Over the last decades, there has been a growing interest by
system that are thermal, hydro and gas with respective control many researchers in analyzing the algorithms inspired by the
inputs as UT, UH and UG may be written as behaviors of natural phenomena. The intelligent algorithms are
Z well suited to solve complex computational problems. These algo-
dACE1 rithms are population based stochastic parallel search algorithms
U T ¼ K P1 ACE1 þ K I1 ACE1 þ K D1 ð3Þ
dt differ from traditional optimization techniques by many aspects
Z
dACE1
U H ¼ K P2 ACE1 þ K I2 ACE1 þ K D2 ð4Þ
dt
UT
Controller-1
Z
dACE1
U G ¼ K P3 ACE1 þ K I3 ACE1 þ K D3 ð5Þ -
dt e UH U
Ref. Controller-2 Power System
The ACE signal is the area control error which includes the data +
about the frequency error and the tie line power error for the UG
related control area. They may be represented in (6) and (7) for Controller-3
area-1 and area-2, respectively. Control parameters

ACE1 ¼ B1 DF 1 þ DPTie ð6Þ Scheduling

TLBO Algorithm for gain
scheduling with ITAE Result
Variables
objective function (.m file)
ACE2 ¼ B2 DF 2  DPTie ð7Þ

The controllers’ gains are tuned by the proposed TLBO algorithm Δf , Δ PTie Controlled output of LFC
according to the ITAE objective function. The typical block diagram
of the gain scheduling control is depicted in Fig. 2. The intelligent Fig. 2. Control strategy for gain scheduling of control parameters.
70 A.K. Barisal / Electrical Power and Energy Systems 66 (2015) 67–77

that are population of solutions, not a single point solution, no

Start
limitation of size of the problem, independent of initial guess of
the variables, simple concept and easy implementation, very close
to hundred percent success rate, significantly fast and robust due Initialize all variables and generate initial
to competition and selection and also incredibly well in solving population array (student x subject)
the realistic AGC problems.

Evaluate the mean grade of each subject (variable)

Teaching learning based optimization

Teaching learning based optimization (TLBO) is a recent Identify the best solution (teacher)
addition to modern heuristic optimization techniques, developed
by Rao et al. [26]. It works on the basis of teaching and learning
mechanism in a class between the teacher and the students. This Modify new solution based on best solution
method is based on the effect of the influence of a teacher in the X new = X old + r ( X teacher − (TF )Mean )
results of students in a class. The teacher is generally considered
as most respected and highly learned person in society who
imparts quality education to their students in the class. The result
of a student is improved not only the quality of teaching delivered Is new solution No
by the teacher but also the collective knowledge of his/her own is better than Reject
and the sharing knowledge of his/her classmates. The result of existing?
the students is finally evaluated on the basis of their outcomes/
grades in the class. Yes
TLBO is a nature inspired, parameter free algorithm which uses
a population of solutions to proceed to the optimal solution. For Replace best solution with new one
TLBO, the population is considered as students in a class and the
control variables are the subjects offered to them. This algorithm
working process is divided into two parts, namely teaching phase Select any two solutions randomly X i and X j
and learning phase are described below. The flow chart of the and modify the new solution as below
proposed TLBO method is depicted in Fig. 3. ( )
X new = X old + r X i − X j if X i better than X j

(A) Teaching Phase: This is the first part of the algorithm in

(
Otherwise, X new = X old + r X j − X i )
which the teacher tries the level best according to his/her
potential to improve the performance of students in a class.
The average result of the class room is improved by the
influence of teacher to some extent ie lkj . If the new average Is new solution No
is better than Reject
grade of jth subject at kth iteration is lknewj , the difference
existing?
between the existing mean and new mean of the jth subject
at the kth iteration may be given as [27]
Yes
 
k
l diffj ¼ rand lknewj  ðT F Þlkj ð8Þ Replace best solution with new one

where TF is the teaching factor which is evaluated randomly

by the following equation.
Is stopping No
T F ¼ roundð1 þ randð0; 1ÞÞ ð9Þ criteria is
satisfied?

The grade of the jth subject of the ith student at k + 1th iteration
is updated by Yes

xkþ1
i;j ¼ xki;j þ lkdiffj ð10Þ Return best solution

(B) Learning phase: It is the last part of the algorithm where Fig. 3. Flow chart of TLBO algorithm.
students upgrade their results by mutual interactions among
themselves. Any two students such as xi, xj are randomly
selected from the class and their grades are updated based
of jth subject of lth student (randomly selected) at kth
on the better student. The learning process may be
iteration. f(xi) is the overall grade point of ith student:
expressed mathematically as follows:
  X i ¼ ½ xi1 xi2 . . . ; xij . . . ; xiND 
xkþ1
ij ¼ xkij þ rand  xkij  xklj if f ðxi Þ < f ðxl Þ ð11Þ
Implementation of TLBO algorithm for LFC
 
Otherwise; xijk¼1 ¼ xkij þ rand  xklj  xkij if f ðxi Þ > f ðxl Þ The computational procedure for proposed TLBO algorithm may
be described in the following steps.
ð12Þ
where xk+1
ij , xkij
are the grade point of the jth subject of the ith Step-1: Input parameters of the system and specify the upper
student at the kth and k + 1th iteration; xklj is the grade point and lower boundaries of each variable.
 A.K. Barisal / Electrical Power and Energy Systems 66 (2015) 67–77 71

Table 1
Tuned control parameters, system modes, minimum damping ratio, settling time, Maximum overshoot and ITAE objective function of multi source single area system.

Methods TLBO-controller DE-controller [22] Optimal control [18]

Controller (I) KI1 KI2 KI3 KI1 KI2 KI3 KI1 KI2 KI3
Parameter 0.0511 0.0041 0.1847 0.0516 0.0071 0.1701 0.1514 0.0131 0.0708
System modes 19.9861 19.9860 19.9859
12.6456 12.6456 12.6429
5.8224 5.8243 5.8415
3.9972 3.9946 3.9735
2.9890 2.9878 3.0062
0.5274 + 0.8597i 0.5377 + 0.8609i 0.5103 + 0.8810i
0.5274–0.8597i 0.5377–0.8609i 0.5103–0.8810i
1.2728 1.2787 1.3331
0.4932 0.4671 0.2824 + 0.0777i
0.1090 0.1101 0.2824–0.0777i
0.0329 0.0334 0.0347
5.0000 5.0000 5.0000
MDR 0.5229 0.0301 0.5012
ITAE 0.5135 0.5165 0.9934
MOS 0.0013 0.0006 0.0093
Settling time (0.002% band) 5.23 5.41 15.75

MDR: Minimum damping Ratio, MOS: Maximum Overshoot.

0.02 Step-3: Evaluate the average grade of each subject. The mean
grade of jth subject is given by
 
0 lj ¼ mean xij ; x2j ; . . . ; . . . . . . xij ; . . . . . . xNPj ð14Þ

-0.02
Step-4: Evaluate the objective function. Find out the best
Δ F (Hz)

solution so far as the teacher and is given by

-0.04
X teacher1 ¼ X jf ðxÞ ¼ min ð15Þ
No control
DE Integral [22] Step-5: Update the grade point of each subject of each individ-
-0.06 TLBO Integral ual student as per Eq. (10).
Optimal Control [21] Step-6: Every student in class improves the grade point of each
-0.08 subject through mutual interaction with other students. The
0 10 20 30 40 50
modification of the grade point with the help of learning phase
T ime (sec) concept may be determined by [26,27].
Fig. 4. Frequency deviations of the single area by using I controller for 1% change in
Step-7: If the stopping criterion (maximum iterations set) is
step load perturbation with AC tie line only. satisfied then print the result. Otherwise, go to step-2.

Results and simulations

Table 2 The present work has been implemented in Matlab-7.10.0.499

Tuned controller parameters (PID) of multi-source multi area system without and (R2010a) environment on a 3.06 GHz, Pentium-IV; with 1 GB
with HVDC link.
RAM PC for the controller parameter tuning in automatic load fre-
Methods TLBO-controller quency control. The model of the system under study has been
Controller parameter Thermal Hydro Gas developed in MATLAB/SIMULINK environment and TLBO program
has been written (in .m file). The developed model is simulated
System with AC lines only KP = 4.1468 KP = 1.0431 KP = 4.7678
KI = 4.0771 KI = 0.6030 KI = 3.7644 in a separate .m file using initial gain scheduling parameters con-
KD = 2.0157 KD = 2.2866 KD = 4.9498 sidering a 1% step load perturbation in area-1 at time, t = 0 s. The
System with AC–DC parallel lines KP = 5.0658 KP = 0.7032 KP = 8.7211 objective function is calculated in .m file and used in optimization
KI = 3.9658 KI = 0.0220 KI = 7.4729 algorithm for tuning the gains of PID controller for power system.
KD = 2.4170 KD = 0.0264 KD = 2.4181

Investigation of tuning performance

Step-2: Initialize randomly the members (students in a class as Single area multi-source power system
row and their subjects as column) of the population array. To compare the dynamic performance of TLBO based I control-
These members (NP) must be feasible candidate solutions (each ler to that of optimal controller and DE tuned controller for single
row) that satisfy the given constraints. The number of subjects area multi-source system is investigated. The proposed optimal
offered to all students is ND. controller parameters, system modes, minimum damping ratio,
settling time, maximum overshoot and ITAE objective function of
Let, Pop ¼ ½ X 1 X 2 . . . X NP  be the trial vector denoting the mem- multi source single area system are provided in Table 1. It is
bers of population to be evolved. evident from the Table 1 that minimum damping ratio (MDR =
X 1 ¼ ½ x11 x12 . . . x1ND  ð13Þ 0.5229) for TLBO optimized I controller is superior to that of DE
optimized controller (MDR = 0.0301) and to optimal controller
The elements of X1 are the grade points of respective subjects (ND) (MDR = 0.502) which is very important for stability point of view.
of first student member. Frequency deviation response of the single area by using I
72 A.K. Barisal / Electrical Power and Energy Systems 66 (2015) 67–77

Table 3
Performance criteria for multi source system with AC lines only.

Objective function TLBO-controller By TLBO DE-controller [22] By TLBO Optimal control [21]
Value % Improvement Value % Improvement Value
ITAE 30.1024  104 32.72 44.7442  104 80.50 15.445  103
ISE 11.1210  102 78.83 52.551  102 94.13 1.7372
ITSE 15.4289  104 45.37 28.23  104 72.32 55.742  104
IAE 11.3410  102 3.28 11.7244  102 53.66 24.475  102
Settling times (0.002% band)
Df1 2.28 39.52 3.77 80.77 11.86
Df2 2.88 37.11 4.58 81.47 14.47
DPTie 0 99 4.29 99 6.71
ACE1 2.3 45.1 4.19 75.37 9.34
ACE2 0 99 2.64 99 19.77

Table 4
Performance criteria for Realistic system with parallel AC–DC lines.

Objective function & parameters TLBO-controller By TLBO DE- controller [22] By TLBO Optimal control [21]
Value % Improvement Value % Improvement Value
ITAE 0.10296 48.18 0.1987 94.59 1.9037
ISE 3.438842  105 82.1 19.22  105 96.37 9.49  104
ITSE 2.197295  105 93.68 34.79  105 99.58 52.74  104
IAE 0.0163 57.55 0.0384 88.19 0.1381
Settling times (0.002% band)
Df1 1.48 23.31 1.93 67.54 4.56
Df2 0 99 4.13 99 5.61
DPTie 0 99 3.81 99 11.36

0.005 0.005

0 0
-0.005
-0.005
Δf 2 (Hz)

-0.01
Δ f (Hz)

-0.01
-0.015
1

-0.015
-0.02 TLBO-PID

TLBO -0.02 DE-PID [22]

-0.025
DE-PID [22] Optimal Control [21]
-0.03 -0.025
Optimal Control [21] 0 5 10 15 20 25 30 35 40
-0.035 Time (sec)
0 5 10 15 20 25 30 35 40
Time (sec) Fig. 6. Frequency deviations of the control area-2 by using PID controller for 1%
change in area-1 with AC tie line only.
Fig. 5. Frequency deviations of the control area-1 by using PID controller for 1%
change in area-1 with AC tie line only.

system with AC lines only. The results are compared with recently
controller for 1% change in step load perturbation with AC tie line published approach (Optimal control) [21] and with DE tuned PID
only is provided in Fig. 4. The settling time of frequency response is controller [22] for the same interconnected power system. In
better than to that of DE controller [22] and Optimal controller, but Table 3, the values of ITAE, ISE, ITSE, IAE, settling time of Df1,
the maximum overshoot is better than optimal controller [21] and Df2, DPTie, ACE1 and ACE2 are improved by 32.72%, 78.83%,
close to DE controller. 45.37%, 3.28%, 39.52%, 37.11%, 99%, 45.1% and 99% respectively
with TLBO tuned PID controller compared to DE tuned PID control-
Multi area multi-source power system without and with HVDC ler [22] and also better than optimal controller [21] for system
This power system comprises two control areas, each area hav- with AC line only. Moreover, the proposed controller outperforms
ing three PID controllers designed for investigation of load fre- than the DE tuned controller and optimal output feedback
quency control. A step load change of 1% in area 1 is considered controller.
at t = 0 s. The optimal controller parameters obtained by TLBO For the system with AC–DC parallel line, Table 4 shows the var-
algorithm for the system with AC lines only and with AC–DC par- ious errors (ITAE, ISE, ITSE and IAE) and settling time of frequency
allel tie lines are reported in Table 2. The standard objective func- and tie line power deviations with the proposed PID controller and
tions and settling times of frequency with 2% tolerance band, tie the results are compared with that of optimal controller and DE
line power deviations and also the settling times of ACEs with tuned controller for the similar system. The values of ITAE, ISE,
the proposed TLBO-PID controller are provided in Table 3 for the ITSE, IAE, settling time of Df1, Df2 and DPTie are improved by
 A.K. Barisal / Electrical Power and Energy Systems 66 (2015) 67–77 73

x 10-3 x 10-3
2 1

0 0

Δ P Tie (pu)
Δ p Tie (pu)

-2 -1

-2 TLBO-PID
-4
TLBO-PID DE-PID[22]
-3
-6 DE-PID [22] Optimal- Control [21]
Optimal Control [21]
-4
-8 0 5 10 15 20 25 30 35 40
0 5 10 15 20 25 30 35 40 Time (sec)
Time (sec)
Fig. 10. Tie line power deviations by using PID controller for 1% change in area-1
Fig. 7. Tie line power deviations by using PID controller for 1% change in area-1 with AC–DC parallel tie lines.
with AC tie line only.

-3
x 10
x 10-3 2
5

0
0
Δ f 1 (pu)

Δ ACE1 (pu) -2
-5

TLBO-PID -4
-10
DE-PID [22] TLBO-PID
Optimal Contro [21] -6 DE-PID [22]
-15 Optimal Control [21]
0 5 10 15 20 25 30 35 40
Time (sec) -8
0 5 10 15 20 25 30 35 40

Fig. 8. Frequency deviations of the control area-1 by using PID controller for 1%
Time (sec)
change in area-1 with AC–DC parallel tie lines.
Fig. 11. Change in area control error of area-1 by using PID controller for 1% change
in area-1 with AC–DC parallel tie lines.

x 10-3
2
-3
x 10
1
3
TLBO-PID
Δ f 2 (pu)

0
DE-PID [22]
2
Optimal Control [21]
-1
Δ ACE (pu)

TLBO-PID
DE-PID [22]
2

-2 1
Optimal Control [21]
-3
0 5 10 15 20 25 30 35 40
0
Time (sec)

Fig. 9. Frequency deviations of the control area-2 by using PID controller for 1%
change in area-1 with AC–DC parallel tie lines.
-1
0 5 10 15 20 25 30 35 40
Time (sec)
48.18%, 82.1%, 93.68%, 57.55%, 23.31%, 99%, and 99% respectively Fig. 12. Change in area control error of area-2 by using PID controller for 1% change
with TLBO tuned PID controller as compared to DE tuned PID in area-1 with AC–DC parallel tie lines.
controller [22] as shown in Table 4.
To verify the superiority of the proposed approach, the dynamic
response of system with AC lines only for step load perturba- responses in terms of settling time, overshoot and under shoot of
tion(SLP) of 1% in area-1 occurring at t = 0 s are shown in Figs. 5–7. frequency responses, tie line power deviations and area control
It is evident from Figs. 5–7 that the superb damping performances errors with system AC–DC parallel lines with 1% SLP in area-1
in term of settling time, overshoot and under shoot of frequency are shown in Figs. 8–12. It is observed that the proposed PID
responses and tie line power deviations are better than that of controller excels in dynamic performance in comparison to DE
DE tuned controller and optimal controller. Similarly, the dynamic controller and optimal controller.
74 A.K. Barisal / Electrical Power and Energy Systems 66 (2015) 67–77

0.005 -3
x 10
1
0
0
-0.005
Δ f 1 (pu)

-1

Δ f (Hz)
-0.01

1
-0.015 -2
1% SLP

-0.02 2% SLP
TLBO-PID
3% SLP -3
DE-PID [22]
-0.025 Optimal Control [21]
0 2 4 6 8 10
-4
T ime (sec) 0 5 10 15 20 25 30 35 40
Time (sec)
Fig. 13. Frequency deviations of the control area-1 by using PID controller for 1–3%
change in area-1 with AC–DC parallel tie lines. Fig. 16. Change in frequency of area-1 for 1% change in area-2.

x 10-3 x 10-3
1 5

0
0
-1
Δ f 2 (pu)

Δ f 2 (Hz)

-2
-5

-3
1% SLP
-10 TLBO-PID
-4 2% SLP
DE-PID [22]
3% SLP
Optimal Control [21]
-5
0 5 10 15 20 25 30 35 40 -15
0 5 10 15 20 25 30
T ime (sec)
Time (sec)
Fig. 14. Frequency deviations of the control area-2 by using PID controller for 1–3%
Fig. 17. Change in frequency of area-2 for 1% change in area-2.
change in area-1 with AC–DC parallel tie lines.

-3 -3
x 10 x 10
0.5 5

0 TLBO-PID
4
Optimal Control [21]
-0.5 DE-PID [22]
Δ P Tie (pu)

3
-1
Δ PTie (pu)

-1.5 2

1% SLP
-2 1
2% SLP
-2.5
3% SLP 0
-3
0 5 10 15 20 25 30 35 40
-1
Time (sec) 0 5 10 15 20 25 30 35 40
Time (sec)
Fig. 15. Tie line power deviations of the control area-1 by using PID controller for
1–3% change in area-1 with AC–DC parallel tie lines. Fig. 18. Change in tie line power for 1% change in area-2.

Investigation of robustness
approach, the location of SLP is changed to area-2 with 1% SLP
For different loading conditions occurring at t = 0 s. Figs. 16–18 show the dynamic responses of
To show the robustness of the proposed controllers, SLP is the power system with AC–DC line with 1% SLP at area-2. It is clear
increased to 1–3% in a step of 1% and its dynamic responses for from Figs. 16–18 that proposed controller outperforms to other
the system with AC–DC parallel line are shown in Figs. 13–15. It two controllers. Similarly the SLP is applied simultaneously in both
is clear that the designed controllers are robust and perform satis- areas and the dynamic responses of frequency and tie line power
factorily for different SLP. To verify the robustness of proposed deviations of the AC–DC line are shown in Figs. 19–21. It is also
 A.K. Barisal / Electrical Power and Energy Systems 66 (2015) 67–77 75

-3
x 10 x 10-3
5 2

0
0
-2
+50 %
Δ f 1 (Hz)

Δ f (Hz)
-4
-5 +25%

1
-6 -25%
TLBO-PID -50%
-10 -8
DE-PID [22]
Optimal Control [21] -10
-15
0 5 10 15 20 -12
0 5 10 15 20 25 30 35 40
Time (sec)
Time (sec)
Fig. 19. Change in frequency of area-1 for 1% change in both areas.
Fig. 22. Change in frequency of area-1 for 0.1 p.u change in area 1 with change in
loading.

-3
x 10
5 -3
x 10
2

0
0
Δ f 2 (Hz)

-2
-5
Δ f 1 (Hz)

-4 +50% of T12

TLBO-PID +25% of T12

-10
-6
DE-PID [22] -25% of T12
Optimal Control [21]
-8 -50% of T12
-15
0 5 10 15 20 25 30 35 40
Time (sec) -10
0 5 10 15 20 25
Fig. 20. Change in frequency of area-2 for 1% change in both areas.
Time (sec)

Fig. 23. Change in frequency of area-1 by using PID controller and by changing the
T12.

-4
x 10
20
-3
x 10
15 2

0
(pu)

10 TLBO-PID
-2
Tie

DE-PID [22]
Δ f 1 (Hz)

+50% of TT
ΔP

5 Optimal -4
Control [21] +25% of TT
-6
0 -25 % of TT
-8
-50% of TT
-5 -10
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
Time (sec) Time (sec)
Fig. 21. Change in tie line power for 1% changes in both areas. Fig. 24. Frequency deviations of the control area-1 by using PID controller and by
changing the turbine time constant of all generators in both areas simultaneously
TT.
clear that better performance is obtained irrespective to the size of
SLP from Figs. 13–15 and the locations of SLP from Figs. 16–21.
To study the effect of variations of loading conditions on the evident from Fig. 22 that there is negligible effect on the variation
system dynamic performances, the operating load condition are of loading conditions on the frequency deviation response with
varied by ±25% from their nominal values taking one at a time controller parameters obtained at nominal values. Consequently,
[17,22,24]. Three PID controllers applied to each area are tuned it is found that the power system behaves robust toward changing
by TLBO algorithm for all loading conditions separately. It is the load.
76 A.K. Barisal / Electrical Power and Energy Systems 66 (2015) 67–77

x 10
-3 behavior of the system with TLBO optimized PID controller for
2 1% step load changes in area-1 at t = 0 s for the above parameters
variation are shown in Figs. 23–25. It can be observed that the
0 proposed control strategy provides a robust and stable control
satisfactorily at nominal parameters and need not be reset for wide
-2 changes in the system parameters.
Δ f 1 (Hz)

-4 +50% For different cost functions

+25% From the very beginning the LFC study, all simulations were
-6 performed using ITAE objective function in TLBO algorithm. In
-25%
-50% order to determine the robustness of power system with AC–DC
-8
parallel tie line toward different cost functions are investigated
using the standard cost functions such as ITAE, ISE, ITSE and IAE.
-10
0 5 10 15 20 25 30 Performance criteria for different cost functions and PID controller
Time (sec) gains for parallel AC–DC lines are provided in Table 5. The results
obtained during simulation for different cost functions to the
Fig. 25. Frequency deviations of the control area-1 by using PID controller and by power system are represented in Fig. 26 and Table 5. Moreover,
changing the frequency bias B in both areas simultaneously. it is seen that the tuning performance of TLBO algorithm is affected
from changing the cost function. The settling time obtained with
ITSE for the frequency deviation and maximum overshoots
Table 5 obtained with ISE for them are better than the others as shown
Performance criteria for different cost functions and PID controller gains for parallel in Fig. 26 and Table 5. It can be evaluated in LFC study, the
AC–DC lines. maximum overshoot is more important than settling time.
Cost functions Thermal Hydro Gas Settling time Therefore, ISE is better objective function in comparison to ITSE
and controller and maximum objective function. The compromise solution with consideration
parameters overshoot to the evaluation of both settling time and maximum overshoot
TLBO-PID controller criteria, ITAE is considered to be as the better cost function.
ITAE KP = 1.6124 KP = 0.2257 KP = 1.7139 However, the power system provides the robustness and stable
KI = 1.9382 KI = 1.3005 KI = 1.8857 ST = 2.17 l toward different cost functions.
KD = 1.0485 KD = 0.3818 KD = 0.2069 MOS = 0.011
ISE KP = 1.7810 KP = 0.1027 KP = 1.5967 ST = 2.59
KI = 1.5979 KI = 0.1458 KI = 1.8860 MOS = 0.01
Conclusion
KD = 1.4687 KD = 0.1771 KD = 1.3674 TLBO algorithm is proposed in this paper to tune the parameters
ITSE KP = 1.5970 KP = 0.4738 KP = 1.9474 ST = 1.94 of I and PID controller. A multi-source single area and multi-source
KI = 1.9750 KI = 1.4045 KI = 1.9446 MOS = 0.0121 multi area without and with HVDC link is considered to demon-
KD = 0.3181 KD = 0.7509 KD = 1.2874
strate the proposed method. An integral time absolute error of
IAE KP = 1.8247 KP = 1.8209 KP = 1.9928 ST = 2.11 the frequency deviation of both areas and tie line power is taken
KI = 1.8353 KI = 1.7877 KI = 1.7805 MOS = 0.0112
KD = 1.2371 KD = 0.6906 KD = 0.7276
as the objective function to improve the system response in terms
of settling time and overshoots. Simulation results emphasis that
the proposed TLBO tuned PID controller is robust in its operation
and gives a superb damping performance for frequency and tie line
For different system parameters power deviation following a step load perturbation (SLP) compared
To study the robustness of power system to wide changes in the to optimal output feedback controller and DE tuned PID controller
system parameters such as turbine time constant of both areas, tie for similar system. Furthermore, it is found that the proposed
line power coefficient and frequency bias parameter of both areas system is more robust and stable to wide changes in system
simultaneously are varied in the range of +50% to 50% from their loading, parameters, size and location of step load perturbation
nominal values in steps of 25% taking one at a time. The dynamic and various cost functions. Besides, the simple concept and
architecture of the proposed controller it can be implemented for
real time application.

-3
x 10 Appendix A
5
The typical values of system under study are given below:

0 f = 60 Hz; B1 = B2 = 0.4312 pu MW/Hz; PR = 2000 MW (rating),

PL = 1840 MW (nominal loading); R1 = R2 = R3 = 2.4 Hz/pu MW;
Δ f (Hz)

Tsg = 0.08 s; Tr = 10 s; Kr = 0.3; Tt = 0.3 s; KT = 0.543478;

-5
KH = 0.326084; KG = 0.130438; Tgh = 0.2 s; Trh = 28.75 s;
1

ITAE
Trs = 5 s; TW = 1 s; bg = 0.5; cg = 1; Xc = 0.6 s; Yc = 1 s;
ISE
-10
Tcr = 0.01 s; Tfc = 0.23s; Tcd = 0.2 s; Tps = 11.49 s;
ITSE Kps = 68.9566 Hz/pu MW; Tdc = 0.2 s; Kdc = 1;
IAE T12 = 0.0433 pu; a12 = 1.
-15
0 10 20 30 40
References
Time (sec)
[1] Elgard OI. Electric energy systems theory. New York: Mc Graw-Hill; 1982. p.
Fig. 26. Frequency deviations of the control area-1 by using PID controller and by 299–362.
different cost functions. [2] Kundur P. Power system stability and control. New York: Mc-Grall Hill; 1994.
 A.K. Barisal / Electrical Power and Energy Systems 66 (2015) 67–77 77

[3] Cohn N. Some aspects of tie-line bias control on interconnected power [16] Ghosal SP. Optimization of PID gains by particle swarm optimization in fuzzy
systems. Am Inst Electr Eng Trans 1957;75:1415–36. based automatic generation control. Electr Power Syst Res 2004;72:203–12.
[4] Elgerd OI, Fosha CE. Optimum megawatt-frequency control of multi-area [17] Rout UK, Sahu RK, Panda S. Design and analysis of differential evolution
electric energy systems. IEEE Trans Power Appl Syst 1970;89(4). algorithm based automatic generation control for interconnected power
[5] Kumar IP, Kothari DP. Recent philosophies of automatic generation control system, Ain Shams Eng J 2012; <http://dx.doi.org/10.1016/j.asej.2012.10.010>.
strategies in power systems. IEEE Trans Power Syst 2005;20(1):346–57. [18] Parmar KPS, Majhi S, Kothari DP. Load frequency control of a realistic power
[6] Lee KA, Yee H, Teo CY. Self-tuning algorithm for automatic generation system with multi-source power generation. Int J Electr Power Energy Syst
control in an interconnected power system. Electr Power Syst Res 1991;20(2): 2012;42:426–33.
157–65. [19] Chaturvedi DK, Satsangi PS, Kalra PK. Load frequency control: a generalized
[7] Rubaai A, Udo V. Self tuning LFC: multilevel adaptive approach. Proc Inst Electr neural network approach. Electr Power Energy Syst 1999;21:405–15.
Eng Gen Trans Distrib 1994;141(4):285–90. [20] Saikia LC, Nanda J, Mishra S. Performance comparison of several classical
[8] Chang CS, Fu W. Area load frequency control using fuzzy gain scheduling of PI controllers in AGC for multi-area interconnected thermal system. Int J Electr
controllers. Electr Power Syst Res 1997;42:145–52. Power Energy Syst 2011;33:394–401.
[9] Cam E, Kocaarslan I. A fuzzy gain scheduling PI controller application for an [21] Parmar KPS, Majhi S, Kothari DP. Improvement of dynamic performance of LFC
interconnected electrical power system. Electr Power Syst Res 2005;73(3): of the two area power system: an analysis using MATLAB. Int J Comp Appl
267–74. 2012;40:28–32.
[10] Pinkang L, Hengjun Z, Yuyun L. Genetic algorithm optimization for AGC of [22] Mohanty B, Panda S, Hota PK. Controller parameters tuning of differential
multi-area power systems. In: Proceedings of IEEE region 10 conference on evolution algorithm and its application to load frequency control of multi-
computers, communications, control and, power engineering-TENCON’02; source power system. Int J Electr Power Energy Syst 2014;54:77–85.
2002. p. 1818–21. [23] Gozde H, Taplamacioglu MC. Automatic generation control application with
[11] Abdel-Magid YL, Abido MA. AGC tuning of interconnected reheat thermal craziness based particle swarm optimization in a thermal power system. Int J
systems with particle swarm optimization. In: Proceedings of the 2003 10th Electr Power Energy Syst 2011;33(1):8–16.
IEEE international conference on electronics, circuits and systems, vol. 1; 2003. [24] Gozde H, Taplamacioglu MC, Kocaarslan I. Comparative performance analysis
p. 376–9. of Artificial Bee Colony algorithm in automatic generation control for
[12] Yes_il E, Guzelkaya M, Eks_in I. Sef-tuning fuzzy PID type load and frequency interconnected reheat thermal power system. Int J Electr Power Energy Syst
controller. Energy Convers Manage 2004;45:377–90. 2012;42:167–78.
[13] Ahamed TPI, Rao PSN, Sastry PS. A reinforcement learning approach to [25] Aström KJ, Hagglund T. PID controllers. USA: Instrument Society of America;
automatic generation control. Electr Power Syst Res 2002;63:9–26. 1995.
[14] Khuntia SR, Panda S. Simulation study for automatic generation control of a [26] Rao RV, Savsani VJ, Vakharia DP. Teaching–learning-based optimization: an
multi-area power system by ANFIS approach. Appl Soft Comput 2012;12: optimization method for continuous non-linear large scale problems. Inform
333–41. Sci 2012;183(1):1–15.
[15] Juang C-F, Lu C-F. Power system load frequency control by evolutionary fuzzy [27] Roy PK. Teaching learning based optimization for short-term hydrothermal
PI controller. In: Proceedings of IEEE international conference on fuzzy scheduling problem considering valve point effect and prohibited discharge
systems. Budapest, Hungary; 2004. p. 715–9. constraint. Int J Electr Power Energy Syst 2013;53(1):10–9.