Applied Soft Computing 119 (2022) 108574

Contents lists available at ScienceDirect

Applied Soft Computing

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

Renewable sources-based automatic load frequency control of

interconnected systems using chaotic atom search optimization
∗
Andrew Xavier Raj Irudayaraj a , , Noor Izzri Abdul Wahab a , Manoharan Premkumar b ,
Mohd Amran Mohd Radzi a , Nasri Bin Sulaiman a , Veerapandiyan Veerasamy c ,
Rizwan A. Farade d , Mohammad Zohrul Islam a
a
Advanced Lightning, Power and Energy Research (ALPER), Department of Electrical and Electronic Engineering, Faculty of Engineering, Universiti
Putra Malaysia (UPM), 43400 UPM Serdang, Selangor, Malaysia
b
Department of Electrical and Electronic Engineering, Dayananda Sagar College of Engineering, Bengaluru, Karnataka 560078, India
c
School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore
d
Department of Electrical Engineering, School of Engineering and Technology, Anjuman-I-Islam’s Kalsekar Technical Campus, Mumbai 410206, India

article info a b s t r a c t

Article history: This paper proposes an improved form of chaotic based atom search optimization (IASO) algorithm
Received 27 July 2021 by adapting one-dimensional (1D) chaotic map (tent, sine and logistic) to improve the search ability
Received in revised form 26 January 2022 by intensifying the exploration and exploitation phase. The IASO avoids premature convergence and
Accepted 3 February 2022
trapping into local optima. Initially, the proposed IASO is validated using a classical benchmark function
Available online 9 February 2022
and its performance is compared with ASO algorithm. Test results indicate that the proposed algorithm
Keywords: outperforms in terms of mean, standard deviation, and best values. Further, the proposed technique
Automatic load frequency control is used to design the parameters of fractional-order proportional integral derivative controller for
Improved atom search optimization automatic load frequency control (ALFC) of multi-area, multi-source hybrid power system (HPS) by
Fractional order proportional integral minimizing the integral time absolute error. The results obtained show that the proposed control
derivative controller scheme improves the frequency response of the system by 48 %, 70 %, 15 % and 69 % in terms of
Hybrid power system settling time, peak undershoot, steady state error value and control effort, respectively compared to
ASO. Moreover, the sensitivity analysis is carried out by considering ±25 % variation in HPS parameters
and the real-time applicability is tested with Malaysian meteorological data of solar radiation and wind
speed variation. These analysis indicates that the transient oscillations are damped out with minimum
settling time and the system regains to stable operating conditions. Further, the evaluation of transient
and steady-state performance indices shows that the tent map-based IASO is found to be more efficient
for obtaining the optimal solution in solving the ALFC problems. In addition, the stability of the system
is analysed by approximating the fractional-order transfer function based on the oustaloup filter in
frequency domain.
© 2022 Elsevier B.V. All rights reserved.

Code metadata power being exchanged over tie-lines to balance the genera-
tion and load demands. However, the area load changes and
Permanent link to reproducible Capsule: https://doi.org/10. unusual events, such as power outages, cause mismatches in
24433/CO.2246117.v1. frequency and scheduled power interchanges across areas. These
mismatches can be controlled using a supplementary control
1. Introduction called Automatic Load Frequency Control (also called Automatic
Generation control). ALFC is an essential mechanism in electric
An interconnected power network consists of several areas energy system to regulate the generator power and demand
or regions depicting the coherent group of generators with in each area of interconnected system to maintain the system
frequency at nominal value and tie-line power exchange being
The code (and data) in this article has been certified as Reproducible by its scheduled value [1,2]. Owing to the depletion of fossil fuel
Code Ocean: (https://codeocean.com/). More information on the Reproducibility
resources, environmental emissions, and ever increasing in en-
Badge Initiative is available at https://www.elsevier.com/physical-sciences-and-
engineering/computer-science/journals. ergy demands paved the way for integration of RES into the
∗ Corresponding author. conventional grid system. However, the RES like solar and wind
E-mail address: [email protected] (A.X.R. Irudayaraj). power leads to uncertain variation in power generation due to

https://doi.org/10.1016/j.asoc.2022.108574
1568-4946/© 2022 Elsevier B.V. All rights reserved.
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

technique to be employed in each control area of interconnected

Abbreviations system to maintain the generation-load balance under severe
ACE Area Control Error disturbances [3].
AE Aqua Electrolyzer Numerous control strategies have been proposed by several
researchers for satisfactory operation of ALFC. A conventional
AGC Automatic Generation Control
PI/PID control applied for two-area thermal system by Hote
ALFC Automatic Load Frequency Control
et al. [4]. Optimal control like linear quadratic regulator/linear
ALO Ant Lion Optimization quadratic gaussian control was applied to a four-area thermal
ANN Artificial Neural Network system by Arya et al. [5]. Robust control like H2 /H∞ control [6],
ASO Atom Search Optimization SMC [7], model predictive control [8], and hybrid control algo-
BBBC Big Bang Big Crunch Algorithm rithms [9] were applied to study the system dynamics under
BD Boiler Dynamics varying disturbances. Among these controllers, the PI/PID con-
BES Battery Energy Storage troller is widely used due to its simplicity and adaptability for
CE Control Effort practical implementation [4]. On the other hand, tuning the gain
DEG Diesel Energy Generator parameters of PID controller for large power system is a crucial
task. The improper tuning of controller parameters for ALFC will
FC Fuel cell
show the poor dynamic response in frequency regulation that
FESS Flywheel Energy Storage System
may lead to system instability. To resolve this, some studies were
FLC Fuzzy Logic Control
reported in the literature to determine the gain values of the
FOPID Fractional Order Proportional Integral controller by various methods such as ANN [10], and FLC [11].
Derivative Despite their favourable, the computational time necessary to
GA Genetic Algorithm select a rule base fuzzy logic and large training database for
GDB Governor Dead Band ANN limits their use in actual LFC studies. Furthermore, fuzzy
GRC Generator Rate Constraints parameters such as inputs, membership function, scaling factors,
GWO Grey Wolf Optimization and rule base are empirically chosen [12]. However, there is
HHO Harris Hawks Optimization no specific mathematical definition for the selection which may
HPS Hybrid Power System sometimes lead to poor ALFC performance.
I Integral On the other hand, different heuristic-based stochastic opti-
mization techniques were used to optimize the gain parameters
IAE Integral Absolute Error
of PID/modified PID/ FOPID controllers for ALFC application as de-
ICA Imperialist Competitive Algorithm
picted in Table 1. The table describes the methodology, controller
ISE Integral Square Error type, system model considered, and the limitations of control
ITAE Integral Time Absolute Error technique used for various studies. However, the PID controller
ITSE Integral Time Square Error does not provide effective control for coupled, uncertain, and
MFO Moth Flame Optimization nonlinear systems. Further, for a process with time delay, un-
MVO Multi Verse Optimization stable transfer functions, and resonance, the controller’s robust-
MPA Marine Predator Algorithm ness and disturbance rejection capability provide unsatisfactory
PEV Plugin Electric Vehicle performance [13].
PI Proportional Integral Hence, to enhance the performance of the PID controller,
PIs Performance Indices FOPID controller was proposed by Podlubny [25] based on the
concept of fractional calculus where the order of differ-integral
PID Proportional Integral Derivative
terms is non-integer. The predominant feature of FOPID controller
PIDD Proportional Integral Double Derivative
over PID controller is, it has two additional tuning parameters
|P − U | Magnitude of Peak Undershoots which provide more flexibility for change in system dynamics.
PSO Particle Swarm Optimization Furthermore, the controller has good disturbance rejection and
QO Jaya Quai Oppositional Jaya better output frequency response than the PID controller [26]. The
RES Renewable Energy Sources application of FOPID controller is extended to various ALFC sys-
RFB Redox Flow Battery tems in conjunction with computational intelligence algorithms
RNG Random Number Generator as listed in Table 1. However, the convergence properties of
SE Steady state Error these stochastic optimization algorithms are strongly connected
SLP Step Load Perturbation to the random numbers generated by the RNG in each algorithm.
Because of this RNG, the solutions generated by the optimization
SMC Sliding Mode Control
algorithms are not diversified to hit every mode in the multi-
SMES Super Magnetic Energy Storage
modal objective landscape in search space [27]. Sometimes, it
ST Settling Time
affects the exploration phase in the early stages of iterations,
STPG Solar Thermal Power Generation leading to premature convergence and trapping into local op-
WOA Whale Optimization Algorithm timum. The embedding of chaos function into the optimization
WTG Wind Turbine Generator algorithm is one of the most effective method to solve the prob-
WTPG Wind Turbine Power Generation lem of local minima. The chaos has the properties of ergodicity,
2DOF-PID 2 Degree of Freedom PID dynamic and non-repetitive [28]. The dynamic property ensures
that the algorithm provides a range of solutions and searches
distinct landscapes in the search space, while the ergodicity and
non-repetition accelerate the search. To incorporate chaos in
meteorological conditions which introduces frequency oscillation metaheuristics algorithm, chaotic maps that generate chaotic se-
that need to be controlled else resulting in damage of equip- quences instead of RNG has been embedded which improves
ment’s and blackout. [2]. To cater this issue, a robust control the global convergence by avoiding local optima [29]. However,
2
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

Table 1
Comparative review of different optimization algorithms used in ALFC study.
Ref. No. Methodology Controller System model Limitations
[14] PSO and PSO-GSA PID Two area thermal system considering PSO suffers from poor exploration
GRC and GDB were studied. and takes a longer time to reach
global minima for a highly nonlinear
and large dimensional systems like
ALFC. This is overcome by PSO-GSA
technique; however, the study is
limited with conventional power
generating sources.
[15] TLBO PIDD Two equal area with thermal, hydro TLBO has two phases such as
and wind in area-1 and thermal, teaching and learning phase, so it
hydro and diesel plants in area-2 takes more run time for a complex
considering all the nonlinearities of system. Further, the study lacks to
thermal system were studied. consider the uncertainties in the
wind power generation.
[16] WOA PID Two equal area multi source system During later stages of WOA iteration
consisting of WTPG, PV, wave and convergence, whales are get attracted
thermal power plant incorporating by the coefficient vector then the
GRC in area −1 and WTPG, PV and whole whale population is easily falls
Thermal plant with GRC in area-2 into local optimum for a high
were considered. dimensional optimization problem.
The study also fails to consider the
real-time variations in the solar/wind
power variations.
[17] QOJAYA 2DOF-PID Two equal area multi source system A single quasi oppositional based
comprises of thermal and hydro technique is employed in Jaya
power plant in each area considering algorithm to optimize the gain
GRC, GDB and BD were studied. parameters of 2DOF-PID controller.
However, relying on single quasi
oppositional learning technique may
not be sufficiently effective for the
problems contains many peaks.
Moreover, if the peaks are closer
together a single quasi oppositional
based technique may not be able to
prevent entrapment in local optima.
[18] MVO PIDD Four equal area power system with To produce children, the MVO
area-1 and 2 consists of thermal algorithm always takes the current
power plant. While area-3 and 4 global optimal as the parent
operating with hydro-type power generation. The benefit of this
plant without considering GRC, GDB behaviour is that it allows for rapid
and BD. convergence in the early stages of
the process. However, there is no
inter-species information
communication when the algorithm
converges to a certain stage, it is
difficult to escape from the local
optimal.
[19] MPA PID Two equal area multi source system The main disadvantage of MPA is the
consisting of STPP, PV, SMES and number of iterations is divided into
thermal power plant incorporating two phases. As a result, large number
GDB and GRC in area −1 and WTPG, of iterations were necessary,
STPP, BES and Thermal plant in especially for a complex and high
area-2 were considered. dimensional problem like ALFC with
large uncertainties.
[20] PSO FOPID A HPS model with different The downsides of PSO algorithm are
generation systems like WTG, PV and that it is easy to slip into a local
DEG with energy storage systems optimum in a high-dimensional
such as AE, FC, FESS and BES were environment and that the iterative
studied. process has a slow convergence rate.
However, the study lags in
considering real-time variation in RE
sources.
[21] ASO FOPID Two equal area HPS model with In ASO, the movement of atoms is
STPG, WTPG, PEV and thermal power determined by their velocities,
system considering GRC and GDB because of interaction forces and
were discussed. constraint forces, as well as their
masses, throughout all iterations; a
slight change in velocities leads to
poor exploitation in later iterations.

(continued on next page)

3
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

Table 1 (continued).
Ref. No. Methodology Controller System model Limitations
[22] GWO FOPID A HPS model with power generation The GWO algorithm updates the
sources like WTG, PV, PEV and DEG individual wolves locations using the
with Energy storage devices such as average position, neglecting the Grey
AE, FC, FESS and BES were wolf hierarchy, which causes GWO to
considered. fall into local extremes.
[23] ALO FOPID Two equal area multi source system The random walking mechanism in
consisting of thermal, hydro and ALO algorithm employs the
wind power plant were studied. maximum iteration number to
construct the ant’s random walking
route. This strategy is ineffective in
terms of the algorithm’s run time.
[24] MFO FOPID Two equal area HPS model with MFO is ineffective in solving the high
STPG, WTPG, PEV and thermal power dimensional constrained search space
system considering GRC, GDB and BD problem and suffers from lack of
were studied. diversity.

chaotic embedded metaheuristic algorithm studies are very lim- order system to analyse its closed loop stability [35]. Further,
ited in the literature for ALFC application. In [30], Khokhar et al. this analysis requires the reduction of higher order system into
applied iterative map based chaotic crow search algorithm to lower order system to analyse the frequency stability. Therefore,
tune a hybrid fuzzy proportional derivative-tilt integral derivative a Hankel method of model order reduction with oustaloup tech-
controller for a simple standalone microgrid. Pan et al. applied nique have been used to analyse the system stability with FOPID
henon and logistic map based PSO to tune the parameters of controller. The prime contributions of the paper are as follows:
FOPID controller for a microgrid model [20]. Similarly, a two-
dimensional sine logistic map-based sine cosine algorithm is used • To improve the performance of ASO algorithm with different
to obtain the gain parameters of PID controller for a stand-alone chaotic maps (like logistic, tent, and sine map) and applying
microgrid system [31]. However, none of the literature discussed it to tune the parameters of FOPID controller for ALFC of HPS.
above applied chaotic based optimization algorithm for a large, • To demonstrate the efficacy of the proposed IASO over ASO
complex interconnected system. This motivates the authors to using five classical benchmark functions in terms of mean
apply a chaotic map embedded metaheuristic algorithm for a and standard deviation values.
complex nonlinear interconnected system. • To prove the robustness of the proposed tuning method of
ASO is a recently developed physics-based swarm-inspired IASO for HPS by sensitivity analysis for wide changes in
optimization algorithm introduced by Weiguo Zhao [32]. It is system loading, inertia constant and uncertain variations in
influenced by molecular dynamics based on the constraint and wind and solar power for its real-time applicability obtained
interaction forces of the atoms. Recently, the authors in the previ- for optimal tuned controller values.
ous study have applied ASO to optimize the parameters of FOPID • To compare the efficacy of proposed IASO tuned FOPID
controller to minimize the ACE of the HPS model with electric controller with ASO optimized FOPID controller in terms of
vehicle. The results obtained reveal that ASO tuned FOPID con- steady state and transient performance indices. The results
trollers perform better than other intelligence techniques such of frequency and tie-line power deviation shows that the
as PSO, BBBC, ICA, HHO and ALO [21]. Nevertheless, the veloc- performance of IASO tuned FOPID controller is significantly
ity updating equation is only updated by each atom’s velocity improved in terms of ST, |P − U | and SE by an amount of
and acceleration that does not deliver the required population 9% to 48%, 3% to 70% and 4% to 15%, respectively than ASO
diversity for atoms to exchange the information during early tuned FOPID controller.
iteration stages which affecting the exploration phase and causing • To propose an oustaloup method of approximation with
premature convergence. In the later stages of the iteration, the Hankel based model order reduction to analyse the closed
weak repulsive force that cannot help the atoms to escape from loop stability of interconnected multi-area power system for
the local optimum, affecting the exploitation [33]. Therefore, to ALFC of HPS.
balance the exploration and exploitation of ASO, a chaotic map
is applied to control the parameters in the velocity by updating The paper is organized as follows: Section 2 describes the ALFC
the equation to avoid premature convergence and trapping of model of ith area multi-area multi-source power system con-
local optimum. Motivated by the literature study, this article uses sisting of conventional and renewable power generation source.
three 1D chaotic maps such as tent map, sine map, and logistic Section 3 establishes the control scheme and the proposed control
map [34], as these maps improve the computational time, simple techniques of IASO. The simulation results of two-area HPS and
dynamic structure, and ease of implementation than other 1D three-area multisource power system for the proposed IASO and
chaotic maps. This enhances the performance of ASO technique ASO are portrayed in Section 4. Section 5 presents the closed loop
which helps to tune the optimal parameters of FOPID controller stability analysis of HPS with fractional controller. Finally, the
for ALFC application. However, analysing the closed loop stability conclusion and future scope of the work is summarized in last
of the interconnected system with fractional order controller is section.
another significant challenge. The authors in [21], analysed the
closed loop stability of HPS using Matignon’s theorem of stability. 2. System model studied
But this analysis requires transformation of variables from s-
plane to a σ -plane and then the stability of the system is analysed In this study, an ith area HPS is considered with reheat ther-
by encircling of closed loop poles. The technique become complex mal, gas, and hydro unit, and RES like solar thermal and wind
for a system having large uncertainties with ‘n’ number of vari- power generation with energy storage devices as shown in Figs. 1
ables which need to be transformed to a σ -plane. Hence, a simple and 2. In worldwide, most of the base load demand is met by the
recursive filter of oustaloup based approximation technique has above-mentioned conventional plants. The country like Malaysia
been applied to approximate the fractional order into an integer where the major power generations are from thermal, hydro,
4
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

Fig. 1. Block diagram of multi-area multi-source HPS.

and gas power plants to meet the base load demand with the STPG plant can be linearized with a few approximations, and the
installed capacity of thermal, hydro, and gas generation units transfer function model is described as below [40]:
in Malaysia being 13.5 GW, 6.7 GW and 42.3 GW, respectively ∆PSTPG Ks KT
which supplying 62.78% of total load demand [36]. Further, the Gs (s) = = (1)
∆Psolar 1 + sTs 1 + sTT
remaining part of load demand is met by RES like solar, wind,
biogas, and biodiesel, etc. Considering these aspects, the pro- where KS and KT signify the gain constant, TS and TT indicate the
posed study incorporates STPG and WTPG with energy storage time constant of solar collector and steam turbine, respectively.
devices such as AE and FC in the HPS model for ALFC operation.
In case of thermal unit, the non-linearities namely BD, GRC, 2.2. Wind turbine generation
and GDB are considered for realistic analysis of the system. In
conventional steam plants, the change in generation is actuated Wind power has been steadily increased in recent years as a
mature form of renewable energy power generation, and its pen-
by turbine control valves, and boiler controls respond instantly
etration into the electrical grid keeps rising. Although, the wind
to changes in steam flow and pressure deviation. As the boiler
power has benefits in terms of environmental impact and supply,
dynamics generate steam under pressure, the pressure effects of
its intermittent nature in power output causes overburdening
a drum type boiler are taken into consideration in this study.
of interconnected tie-lines, resulting in frequency deviation of
The primary reason to consider GRC in this study is to simulate
the power system. As the wind speed changes, a pitch control
the practical constraints on the sensitivity of the turbine. The
mechanism is activated, which changes the pitch of the rotor in
value of GRC considered in this work is 0.0005 p.u.MW/s [37].
order to keep the power produced by the wind energy conversion
On the other hand, the dead band in thermal steam units is system constant. As shown in Fig. 2(a) three WTPG are considered
induced by backlash in the linkage that connects the servo pis- and represented by first order lag-based transfer function as [40]:
tons to the camshaft. Therefore, in this study, the limiting value
of dead band [38] is set at 0.05%. The nonlinearity associated ∆PWTPG KWTG
GWTG (s) = = (2)
with the reheat thermal system, Gas, hydro, and RES generation ∆Pwind 1 + sTWTG
unit is mathematically linearized around certain operating points, where KWTG and TWTG denote the gain and time constant of WTPG.
resulting in a first-order transfer function as depicted in Fig. 2.
2.3. Aqua electrolyzer and fuel cell
2.1. Solar thermal power generation
The STPG/WTG system with an abundantly clean power gener-
Solar thermal power plants use the sun as a heat source to ation can meet the load demand. However, the STPG/WTG output
generate electricity. The concentrating solar power is essential power is not constant and varies with the intensity of solar
for reaching high enough temperatures for a power plant to radiation and wind speed [41]. Moreover, these fluctuations result
operate effectively. A thermodynamic cycle which converts the in variations in frequency. To overcome these issues, AE were
useful heat received from concentrated solar thermal power into used to absorb and produce hydrogen from the rapidly varying
electricity [39]. The solar collectors and working fluid are the two output power of STPG/WTG. The hydrogen produced is kept in a
most important components of the STPG system. Solar collectors hydrogen tank and utilized to power fuel cells. The total output
with parabolic troughs are commonly used in the solar field to power from STPG, WTG and FC is supplied to meet the load
collect solar energy, which is focused onto a tube filled with demand. Fig. 3 depicts the energy exchange process of AE and FC.
working fluid (oil, air, or water) and circulates. When the working The linearized transfer function model of AE and FC is expressed
fluid is heated, it is used by the power cycle to generate high- as [40]:
pressure steam in a boiler and then to generate electricity by KAE ∆PAE
expanding it in a turbo-generator. For small-signal analysis, the GAE (s) = = (3)
1 + sTAE (∆PWTG + ∆PSTPG )(1 − Kn )
5
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

Fig. 2. Transfer function model of (a) Renewable Energy Sources (b) Thermal system (c) Hydro Power system (d) Gas Power System.

where KAE and TAE represents the gain and time constant of AE,
respectively. Kn = (P P+tP ) is presumed as 60%.
WTG STPG

KFC
GFC (s) = (4)
1 + sTFC
where KFC and TFC denote the gain and time constant of FC,
respectively.

3. Concept of FOPID controller

The fractional-order calculus is based on the extension of dif-

ferentiation and integration to any order, which can be rational,
irrational, or complex. Over the last few decades, fractional-order Fig. 3. Processes of energy exchange in AE-FC [21].
calculus plays a significant role in designing various controllers
by expanding the fractional integrals and derivatives to fractional
orders to achieve robustness and superior performance. FOPID are [26]: attaining a good response to a minimum phase system,
controller is the generalization of traditional PID controller which controlling of nonlinear system without linearization, and less
has two additional tuning parameters offering great flexibility to sensitive to system parameter variations. Fig. 4 depicts the pro-
the plant to be controlled. The unique features of FOPID controller posed IASO-FOPID controller approach for ALFC of power system.
6
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

Fig. 4. Closed-loop feedback control of Power system with IASO-FOPID controller.

The transfer function and control output of FOPID controller are others simply by attraction, ensuring that the algorithm has a
given as [21]: better exploitation capability. Throughout the iterative process,
constraint force between each atom and the atom with the best
TFOPID (s) = Kp + Ki s−λ + Kd sµ (5)
fitness always exists. For a more detailed discussion of ASO, one
−λ µ
U(s) = (Kp + Ki s + Kd s ) × ACE (6) can refer to [21].
where Kp , Ki, and Kd are the proportional, integral, and deriva-
tive gains of the controller, λ and µ represent the integral and 3.2. Chaotic maps
differential operators, which take the values of arbitrary real
number. In this study, the n-integer toolbox is used to design the Chaos is described as ‘‘randomness’’ caused by simple deter-
FOPID controller in MATLAB/Simulink environment. In toolbox, ministic systems in mathematics. Chaos theory is highly depen-
the CRONE approximation method which utilizes a recursive dent on the initial conditions of the problem. The small change in
distribution of N poles and N zeros is applied to approximate
the initial value results in significant diverse future behaviours,
the fractional-order differentiator/integrator. In this work, the
such as ergodicity, steadily fixed points, and bifurcations [42].
ITAE is chosen as an objective function to tune parameters of
Moreover, a chaotic motion can go through each state in a par-
FOPID controller by minimizing the ACE as it provides superior
ticular region by its own periodicity, and each state can only be
performance by reducing the settling time and undershoots of
obtained once. It is evident from the literature [43] that chaos
the plant to be controlled than other steady state indices. The
shares some of the common properties of many metaheuristic
objective function can be defined as [24],
algorithms; therefore, chaotic search is integrated into the pro-
∫ T
⏐∆fi + ∆Ptie,i j ⏐ .tdt ∵ i, j = 1, 2, 3 & j ̸= i (7)
[⏐ ⏐] cess of heuristic algorithms to improve the convergence rate and
J = ITAE =
0
pre-empt from being confined to local optima. To achieve this, the
chaotic maps are incorporated in the procedures of metaheuristic
Subjected to the following boundary constraints:
algorithms to generate chaotic sequences. The most widely used
Kp min ≤ Kpi ≤ Kp max chaotic maps are 1D chaotic maps due to their simple structure
Kd min ≤ Kdi ≤ Kd max and easy implementation. The various 1D chaotic maps are circle
map, cubic map, Gauss map, iterative chaotic map with infinite
µmin ≤ µi ≤ µmax (8) collapses (ICMIC), logistic map, sine map, tent map, Chebyshev
Ki min
≤ Kii ≤ Ki max map, piecewise map, and singer map. Out of all these, the logistic
map, sine map, and tent map are the most popular 1D chaotic
λ min
≤ λi ≤ λmax maps used in optimization techniques to improve the conver-
gence speed and search capability [28]. Hence, in this research
3.1. Atom search optimization these three maps are adopted to improve the performance of
tuning the parameters of FOPID controller than other mapping.
ASO is a recent population-based metaheuristic optimization The chaotic maps studied in this research work are detailed as
algorithm inspired by physical theory of molecular dynamics, follows:
which describes the structure of matter and how atoms inter-
act at the micro level. The interaction force from the potential 3.2.1. Logistic map
energy expressed by Lennard-Jones potential and the geometric The logistic map is widely used second-order, simple, nonlin-
constraint force between two atoms are used by ASO to perform ear, and dynamic polynomial map. This map’s equation generally
a search. The interaction force between atoms has two main appears in biological population of nonlinear dynamics exhibit-
properties. The first is the repulsion to compression, which repels ing chaotic behaviour. Mathematically the logistic map can be
the atom when crowdedness is within a certain range. The second
expressed as [44]:
is the attraction between atoms. Constraint forces are employed
to keep the bond length and angle between any two atoms at xk+1 = rxk (1 − xk ) (9)
the desired levels. During early iterations of ASO, an individual
atom interacts with other atoms through attraction or repul- where k defines the number of iterations, xk signifies the chaotic
sion, and repulsion can prevent over accumulation of atoms and variable, and r is the control parameter lies in the interval (0, 4).
premature convergence of the algorithm improving exploration In most of the literature work, the value of r is chosen between
ability across the whole search space. As iterations progress, the 1 to 4 to generate the chaotic sequence [44]. Since, the value
repulsion steadily weakens and the attraction gradually strength- of r between 1 to 3 gives poor chaotic behaviour as inferred
ens, indicating that exploration is decreasing, and exploitation is from Fig. 5. But, the chaotic sequence is exhibited in the range
increasing. During the final iterations, each atom interacts with of 3≤r≤4 with certain selection of parameters.
7
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

Fig. 7. Bifurcation diagram of Tent map.

Fig. 5. Bifurcation diagram of Logistic map.

3.3. IASO for tuning of FOPID controller

The steps for tuning the FOPID controller using IASO are de-
tailed as follows:
Step 1: Initialize the gain and fractional order terms randomly,
and its positions are described as Xi = (xi1 , . . . xi d , . . . .xi n ), where
i denotes an atom in the population (i=1,2, . . . N), and d denotes
the dimension (d=1,2, . . . D) of the search space.
Step 2: Define the values of initial parameters such as depth
weight (α ), multiplier weight (β ), and a maximum number of
iterations (T) and initialize the number of atoms (N), its posi-
tion, and velocity. For the given parameters, calculate the fitness
function defined in Eq. (7) for each atom.
Step 3: Calculate the mass mi (t) of ith atom using the equation
defined below,
Fig. 6. Bifurcation diagram of Sine map.
Mi (t) −
Fiti (t)−Fitbest (t)
mi (t) = ∑N , Mi (t) = e Fitworst (t)−Fitbest (t) (12)
j=1 Mi (t)

3.2.2. Sine map where Fiti (t) is the ith atom fitness value at tth iteration and
Sine map is one of the simple 1D chaotic maps that generates Fitbest (t) and Fitworst (t) are the atom best and worst fitness value
similar chaotic sequences as logistic map. However, the mathe- at tth iteration which can be defined as
matical expression of the sine map is totally different from the
logistic map and is given as [44]: Fitbest (t) = min Fiti (t) (13)
i∈{1,2,...,N}

xk+1 = rsin (π xk ) (10) Fitworst (t) = max Fiti (t) (14)

i∈{1,2,...,N}
Eq. (10) describes the sine function, which maps the input
angle in the range of [0,4], where r is the bifurcation parameter, xk Step-4: To intensify the exploration phase in the early stage of
and xk+1 denotes the current and next iterative sequence between iterations, each atom must interact with as many neighbourhood
0 and 1, respectively. The sine map shows satisfactory chaotic atoms with better fitness values as possible, where the K neigh-
behaviour when r ∈ [0.7, 4]. Moreover, it is substantially revealed bour atom performs to improve the exploitation during the final
from Fig. 6 that there is some correlation between the sine and iterations. Thus, each atom must interact with few atoms with
logistic maps at the initial stage. better fitness values as possible as its K neighbours. As a result,
K decreases with progress in iteration number corresponding to
3.2.3. Tent map
function of time as follows.
A tent map is a discrete-time, piecewise, linear chaotic map √
with some unique attributes like simple shape, chaotic orbits, etc. t
K (t) = N − (N − 2) × (15)
The mathematical formulation of the tent map is given by [33], T
⎧ xk
⎨r 2 xi < 2 Step 5: Compute the total interaction force Fid (t) exerted by other
xk+1 = 0 < r ≤ 4; xk ∈ [0, 1] (11)
⎩r (1 − xk ) x > 2 atoms on the ith atom in the dth dimension space that can be
i
2 expressed as sum of the forces with random weights as follows.
The dynamic behaviour of the tent map depends on the con-
∑
Fdi (t) = randj Fdij (t) (16)
trol parameter r which varying from predictable to chaotic. The
j∈Kbest
bifurcation diagram of the tent map is illustrated in Fig. 7. The
tent map has distinctive advantages of higher iterative speed where randj is a random number ranging between 0 and 1.
than logistic map, and the probability density function is uniform The constraint force Gdi (t) is expressed as a force-controlled by
compared with the logistic and sine map [45]. the best atom on each atom. Therefore, the constraint force can
8
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

Fig. 8. Proposed IASO Algorithm to optimize FOPID controller for ALFC.

be calculated as In case of IASO, the chaotic sequences created by the chaotic

map (logistic, sine, and tent) replace the random parameter rand di
Gdi (t) = λ (t) xdbest (t) − xdi (t)
( )
(17)
of ASO which improves the flexibility of the atom in each iteration
−20t
where λ is the Lagrangian multiplier expressed as λ (t) = β e T , and convergence rate. The velocity updating equation of IASO is
xdbest (t) defines the best atom found at tth iteration in dth position, defined in Eq. (20), and the corresponding position of each atom
xdi (t) depicts the best atom attained at tth iteration in ith position. is updated at iteration t is formulated as given in Eq. (21).
Step 6: Determine the acceleration of each atom through interac-
vdi (t + 1) = xk vdi (t) + adi (t) (20)
tion and constraint force of atom
( as below,)3
Fdi (t) Gdi (t) t−1 −20t xdi (t + 1) = xdi (t) + vdi (t + 1) (21)
ai (t) = d
d
+ d = −α 1 − e T
mi [(t) mi (t) ] T
)13
∑ randj 2 × hij (t) − h7ij Step 7: Check maximum number of iterations is reached and
(

(18) obtain the optimized controller parameters as xbest . If not, repeat

mi (t) from step 2 through 7 until the stopping criterion is reached.
j∈Kbest
( ) − xi (t)
( )
xdj t d
−20t x
d
(t) − xdi (t) As a result, the optimized FOPID control parameters are ob-
 + β e T best
Xi (t) , Xj (t) mi (t) tained by reducing the ACE of interconnected HPS by reaching

2
the global minima with the atoms’ location providing the best
In case of ASO algorithm, the velocity of each atom is calcu-
optimal solution. The initialized parameters of IASO are T = 100,
lated using Eq. (19) defined as,
N = 50, α = 50, and β = 0.2. The above steps are summarized
vdi (t + 1) = randdi vdi (t) + adi (t) (19) in the form of flowchart for optimizing the parameters of FOPID
In Eq. (19), and vdi (t) adi (t)
denote the current velocity and controller as depicted in Fig. 8. The boundaries for tuning the
acceleration vectors of ith atom in dth dimension in search space, parameters of FOPID controller such as kp , ki and kd [kpi , min , kpi ,
respectively, where vdi (t+1) signifies the new velocity values, and max , kii , min , kii , max , kdi , min , kdi , max ] are bounded between [−3, 3],
rand di generates a random number in the range (0,1). λ and µ is selected between [0, 1].
9
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

Table 2
Benchmark functions.
Function name Type Formulae Dimension(d) Range Minimum value
d
∑
Sphere F1 US f(x) = x2i 30 [−100,100] 0
i=1

d− 1
∑
Rosenbrock F2 UN f(x) = [100(xi+1 − x2i )2 + (xi − 1)2 ] 30 [−30,30] 0
i=1

d
∑
Rastrigin F3 MS f(x) = [x2i − 10 cos(2π xi ) + (xi − 1)2 ] 30 [−5.12,5.12] 0
i=1

d (
x1 (b2i + bi x2 )
∑ )
Kowalik F4 MN f(x) = ai − 4 [−5,5] 0.0003
b2i + bi x3 + x4
i=1

5
⏐(xi − ai )(xi − ai )T + ci ⏐−1
∑ ⏐ ⏐
Shekel 5 F5 MN f(x) = 4 [0,10] −10.1532
i=1

4. Results and discussion Table 3

Statistical results of benchmark functions.

In this section, initially the performance of the proposed IASO Function Index TM-CASO SM-CASO LM-CASO ASO

is verified by comparing with the standard ASO algorithm in Best 4.43e−24 5.32e−23 0.00015 270.97
F1 SD 1.85e−23 5.74e−22 7.01e−05 290.69
terms of statistical results by considering five classical benchmark
Mean 8.40e−21 3.75e−22 3.14e−05 600.84
functions extensively reported in the literature. Furthermore, the
Best 86.75 84.01 452.18 22815.96
proposed IASO tuned FOPID controller is analysed for multi-area
F2 SD 30.84 190.99 127.31 22438.31
multi-source HPS. The model is developed in MATLAB/Simulink Mean 45.22 155.33 165.42 34152.12
and the simulation results are obtained. The results of IASO (logis- Best 66.66 67.65 56.71 91.76
tic map based chaotic atom search optimization (LM-CASO), sine F3 SD 3.82 16.13 17.68 34.58
map based chaotic atom search optimization (SM-CASO) and tent Mean 62.07 63.67 64.08 118.33
map based chaotic atom search optimization (TM-CASO)) tuned Best 0 0 0 7.61
FOPID controller are compared with the standard ASO technique. F4 SD 0 0.0033 0.0033 17.22
The following section discuss in detail about the results obtained Mean 0 0.0014 0.0015 39.60
for various benchmark functions and numerous cases of HPS Best 0.001 0.0013 0.0014 0.0016
model. F5 SD 0.0002 0.00036 0.0004 0.005
Mean 0.0011 0.0011 0.0013 0.0054

4.1. Validation of benchmark functions

Table 4
To validate the performance of the proposed IASO algorithm Controller gains for two-area HPS.

with ASO algorithm, five classical benchmark functions with di- Proposed control methods to Case I
tune FOPID controller kp kd µ ki λ
verse set of attributes are considered [46]. Table 2 depicts the
details of benchmark functions used for validation of various TM-CASO −3.3573 −1.2357 0.9895 −4.9961 0.8804
form chaotic based ASO. The functions F1 and F2 are unimodal SM-CASO −3.0573 −1.1610 0.9567 −3.9889 0.8727
LM-CASO −2.8095 −0.9178 0.9825 −2.9397 0.8864
benchmark function with a single global best which are denoted
ASO −1.9698 −1.5232 0.7980 −2.9807 0.8775
by the letter U. Further, if the unimodal function is separable
or non-separable, the S and N are written after the designation
of U as shown in Table 2. The algorithm exploitation ability is
evaluated by the unimodal benchmark functions, indicating that 4.2. Time domain analysis of interconnected system
good results require an algorithm with a strong local search
capability. F3-F5 depicts the multimodal functions. In contrast In this section, the proposed IASO is further validated for
to unimodal functions, multimodal functions have more than the ALFC application of HPS by optimizing the parameters of
one global best and are used to evaluate the algorithm explo- FOPID controller in each area of interconnected system. The tuned
ration ability. The functions in Table 2 that have the label M values of FOPID controller using standard and chaotic based ASO
are multimodal functions with two types: separable and non- for two- and three- area power system is given in Tables 4 and
separable [47]. The statistical results of the algorithms on the 5, respectively. The effectiveness of the proposed method is ex-
classical benchmark functions are presented in Table 3. These amined in various cases of power system operation as described
results are compared in terms of mean and standard deviation below
(SD) of the best values achieved. Each algorithm was run 30
Case I: Two-area HPS
times independently and the results were recorded. The various
In this case, a two-area HPS model that comprises reheat
benchmark functions are tested using each chaotic and standard
thermal system incorporated with BD, GDB, GRC, and RES like
ASO technique by considering the atom population of 50 and
WTPG and STPG combined with energy storage systems of AE
number of iterations as 100, respectively. The statistical results
and FC were considered. A step load change of 50 MW and 30
inferred from Table 3 shows that chaotic maps based ASO have
MW is applied in area-1 and area-2 of interconnected system,
significantly improved not only the exploration phase but also the
respectively. The total power generation of HPS can be expressed
exploitation phase. For both unimodal and multimodal functions,
as:
TM-CASO shows lower means and standard deviations when
compared with other chaos embedded ASO and standard ASO. Ps = PThermal + PWind + PSolar − PAE + PFC (22)
10
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

Table 5
Controller gains for three area multi-source power system.
Proposed control methods Case II
to tune FOPID controller kp kd µ ki λ
Area-1 −0.9574 −0.8309 0.9880 −2.9986 0.8110
TM-CASO Area-2 0.4931 −2.5689 0.6257 −2.3836 0.8621
Area-3 −1.2824 −1.9583 0.8430 −2.4521 0.9028
Area-1 −0.8773 −0.8744 0.9720 −2.9989 0.8237
SM-CASO Area-2 0.4927 −2.5338 0.5024 −1.9549 0.8211
Area-3 −1.8421 −2.1095 0.7579 −1.6505 0.9889
Area-1 −1.6068 −1.0242 0.9874 −2.9949 0.8064
LM-CASO Area-2 0.7536 −2.7315 0.6697 −1.7803 0.9721
Area-3 −1.8873 −1.5059 0.7985 −2.4186 0.9065
Area-1 −1.5826 −1.0834 0.9868 −2.9907 0.7937
ASO Area-2 0.8381 −2.5749 0.6792 −2.6118 0.8254
Area-3 −1.6945 −2.3021 0.6861 −2.5195 0.9748

Fig. 9. Transient performance response of two-area HPS. Fig. 10. Transient performance response of three-area multisource power
system.

The system dynamic responses (frequency deviation and tie-

line power variations) are illustrated in Fig. 9. It is clear from the comprises reheat thermal system with nonlinearities like BD,
response that the FOPID controller tuned by TM-CASO damped GRC, and GDB in all areas, hydro generating unit in area-1 and
out the frequency deviations more rapidly than other methods. In area-3, and gas turbine power plant in all areas of interconnected
terms of transient specifications as given in Table 6, the proposed
system. An increased SLP of 60 MW in area-1, 30 MW in area-
method (TM-CASO) exhibits minimal ST (4.96 s, 6.4329 s) and P-
2, and 50 MW in area-3 were applied, respectively. The dynamic
U (0.0395, 0.0251) in area-1 and area-2, respectively. Further, the
tie-line power tend to be smooth and stable for the case of IASO. response of the system with optimal values of FOPID controller
In addition, the other form of IASO such as SM-CASO (6.1219 s, using TM-CASO is significantly better than LM-CASO, SM-CASO
6.81 s) and LM-CASO (6.7148 s, 7.4139 s) tuned FOPID controller and ASO tuned FOPID controller as shown in Fig. 10. The pro-
also have minimum ST than ASO (7.29 s, 7.8381 s) adapted posed method (TM-CASO) has minimum ST with less undershoots
FOPID controller. As the chaotic map embedded in ASO algorithm compared with other methods presented. The work done by the
generate the search agents more dynamically which improves controller is evaluated in terms of CE and the results exhibit that
the exploration and exploitation capabilities of standard ASO. The the TM-CASO optimized FOPID controller achieve low CE even for
transient performance indices like ST, SE, and P-U are improved
the large-scale system of three-area compared with standard ASO
by 18% to 32%, 41% to 43%, and 31% to 57%, respectively using
TM-CASO than ASO method. On the other hand, the steady-state and other chaotic forms (SM-CASO and LM-CASO). The response
performance indices like ISE, ITSE, IAE, and ITAE are significantly of the standard ASO suffers from transient oscillation due to
enhanced by a proportion of 22% to 25%, 25% to 40%, 24% to 28%, poor exploration and exploitation ability. Table 7 provides the
and 41% to 44%, respectively by the proffered TM-CASO based performance indices of IAE, ITAE, ISE, ITSE, CE, ST, SE and P-
FOPID controller than other methods presented in Table 6. U which are enhanced by 7% to 20%, 16% to 25%, 24% to 34%,
Case II: Three area power system 35% to 43%, 17% to 22%, 27% to 46%, 7% to 29% and 10% to 15%,
The robustness of the proposed control technique have been respectively for TM-CASO tuned FOPID controller than standard
studied for three-area interconnected multi-source power system ASO method.
11
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

Table 6
Performance Indices of two-area HPS.
Case Performance ∆f1 ∆f2
indices TM-CASO SM-CASO LM-CASO ASO TM-CASO SM-CASO LM-CASO ASO
IAE 0.0333 0.0341 0.0391 0.0445 0.0298 0.0317 0.0372 0.03914
ITAE 0.068 0.0842 0.1034 0.1212 0.067 0.0827 0.1013 0.1135
ISE 0.0006 0.0007 0.0008 0.0008 0.0004 0.0004 0.0006 0.0006
ITSE 0.0003 0.0003 0.0004 0.0005 0.0002 0.0002 0.0004 0.0004
Case I
CE 0.0092 0.0114 0.0147 0.0151 0.0089 0.0114 0.1096 0.0150
ST (s) 4.96 6.1219 6.7148 7.29 6.4329 6.81 7.4139 7.8381
SE (×10−5 ) 1.324 1.862 2.175 2.246 1.33 1.909 1.391 2.315
|P-U| 0.0395 0.0403 0.0420 0.0419 0.0251 0.0265 0.0287 0.0361

Table 7
Performance Indices of three area multi-source power system.
Case Performance TM-CASO SM-CASO LM-CASO ASO
indices
IAE 0.0864 0.1021 0.0914 0.1046
ITAE 0.283 0.3417 0.2928 0.3504
ISE 0.0025 0.0032 0.0027 0.0033
ITSE 0.002 0.0029 0.0023 0.0031
∆f1
CE 0.0381 0.0431 0.0452 0.0484
ST (s) 7.25 9.5957 9.7215 9.9842
SE (×10−5 ) 4.75 6.635 6.615 6.691
|P-U| 0.0606 0.0649 0.0648 0.0697
IAE 0.0879 0.1093 0.0947 0.1098
ITAE 0.2952 0.3787 0.3261 0.3916
ISE 0.0021 0.0028 0.0024 0.0032
Case II ITSE 0.0025 0.0040 0.0303 0.0044
∆f2
CE 0.0439 0.0488 0.0430 0.0529 Fig. 11. Solar Power Profile.
ST (s) 6.76 11.2885 11.4213 12.5658
SE (×10−5 ) 4.96 6.309 5.822 7.032
|P-U| 0.0466 0.0532 0.0510 0.0545
IAE 0.0891 0.0935 0.1074 0.1096
ITAE 0.3059 0.3146 0.3421 0.373
ISE 0.0022 0.0022 0.0024 0.0031
ITSE 0.0024 0.0026 0.0039 0.0041
∆f3
CE 0.0420 0.0487 0.0430 0.0540
ST (s) 7.12 11.4580 10.2467 11.9152
SE (×10−5 ) 4.92 6.08 6.286 6.786
|P-U| 0.0505 0.0513 0.0531 0.0560

4.2.1. Robustness analysis of proposed control strategy

To validate the robustness of the proposed TM-CASO opti-
mized FOPID controller, the HPS model in case I has been tested
with real time variation in STPG and wind power. Furthermore,
to test the adaptability of the proposed technique the two-area
HPS as in case I has been considered by varying system parame-
ters like inertia constant (H) and system loading (which reflects
frequency dependent system loading).
4.2.1.1. Real time variation of RES. In this case, the variation in
STPG unit output power is considered for real-time monthly- Fig. 12. Dynamic response of HPS of random solar power variation.
average irradiance data at Universiti Putra Malaysia from Subang
meteorological centre in Malaysia [48]. The irradiance values are
taken from January to December 2014 for 200-time slots of 0.5 to other techniques presented. Thus, overall it is seen that among
s as given in Fig. 11. The developed HPS model is tested for the chaotic map that used to improve the performance of the
this real-time solar power variations and the dynamic response ASO method of tuning FOPID controller the proposed TM-CASO
is recorded as given in Fig. 12 for IASO and ASO method. It is performs remarkably to regulate the system frequency and tie-
inferred that the TM-CASO tuned FOPID controller shows better line power to steady state under step load perturbation and
and smoother response with minimum overshoot and undershoot real-time variation in RES.
than other tuning methods. Similarly, the wind power varia-
tion for change in wind speed is obtained in real-time at the 4.2.1.2. Performance under uncertain variation. In this case, the
Malaysian Meteorological Department in Kuala Terengganu [36] system parameter H and loading conditions are varied from their
with latitude= 5◦ 23’ N and longitude=103◦ 06’ E (considering el- nominal values. In order to prove the effectiveness of the con-
evation=5.2 m) for a whole year, and converted into seconds to troller, the tuned values under nominal conditions as given in
assess the reliability of the controller, as depicted in Fig. 13. The Table 4 are considered. Fig. 15 shows the frequency and tie line
frequency response and tie-line power deviation for this variation power response obtained for change of H by ±25% from its nomi-
in wind power is shown in Fig. 14. It is observed that the TM- nal value (H=5). From this figure it is observed, the efficacy of the
CASO has minimum frequency oscillations with less SE compared controller tuned using TM-CASO is optimum for wide variation
12
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

Fig. 13. Wind Power Fluctuation.

Fig. 15. Sensitivity Analysis of TM-CASO tuned FOPID control for change in
inertia (H) of the system.

Fig. 14. Dynamic response of HPS of random wind power variation.

of H. Furthermore, changes in system loading of ±25% from the

nominal value (50%) also observed. Fig. 16 depicts the dynamic
response of frequency and tie line power variations for changes in
system loading. It can be noted that the system adjusts to changes
in system loading more quickly, without any delay in response.
Fig. 16. Sensitivity Analysis of TM-CASO tuned FOPID control for change in
Hence, it can be concluded that the controller tuning parameters system loading.
need not to be re-tuned for parameter changes as the proposed
control strategy is adaptive for any change in system parameter
or uncertainty.
reliable in producing superior results with minimum settling time
by avoiding the premature convergence due to its diversification
4.3. Computational cost of chaotic sequences compared to the LM-CASO and SM-CASO,
where the chaotic sequence is appeared for minimum time span.
To investigate the computational complexity of the proposed Thus, the exploration and exploitation capability of the ASO is
IASO technique compared to the standard ASO method, this sec- significantly improved by the tent map compared to other chaotic
tion presents the convergence performance of the proposed algo- maps.
rithm (TM-CASO, LM-CASO and SM-CASO) for tuning the FOPID
controller of two-area HPS model as discussed in Case I. Fig. 17 4.4. Comparative analysis
shows the convergence curve which depicting the minimization
of ITAE of ACE for HPS model considered. It is inferred that the To further validate the superior performance of the TM-CASO
various IASO method converges faster compared to the standard algorithm, a comparison between the proposed TM-CASO and
ASO. However, the TM-CASO algorithm is more faster than the several well established meta-heuristic algorithms widely used in
LM-CASO and SM-CASO method of tuning the controller. Also, the literature such as PSO, GWO and MFO are used to optimize
from the results obtained it is seen that the TM-CASO is more the parameters of FOPID controller for two-area HPS model as
13
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

Fig. 17. Convergence curve of the algorithms.

Table 8
Gain values of different optimization method.
Control kp kd µ ki λ Execution Fig. 18. Transient performance response of different algorithms for HPS.
methods time (s)
TM-CASO −3.3573 −1.2357 0.9895 −4.9961 0.8804 1066.92
PSO −0.3927 −2.0409 0.4863 0.5920 0.5680 2258.64
GWO −1.9698 −1.5232 0.7980 −2.9807 0.8775 2615.39
MFO −0.9937 −0.8958 0.8171 −0.7579 0.8730 3736.13

discussed in Case I. All algorithms were simulated in MATLAB

language with system specifications of Intel core i5 11th genera-
tion processor and 8 GB RAM. For fair comparison, the maximum
number of iterations and size of population is set to 100 and
50, respectively for all algorithms. Fig. 18 depicts the dynamic
response of two-area HPS compared with different algorithms.
It is inferred from the figure that, the proposed TM-CASO opti-
mized FOPID controller displays better frequency response with
minimum settling time and undershoot as compared to other
algorithms. Similarly in the case of execution time PSO takes
2258.64 s, GWO takes 2615.39 s, MFO takes 3736.13 s, but TM-
CASO takes only 1066.92 s for optimizing the gain values of
the FOPID controller. The gain values of the controller optimized
Fig. 19. Convergence curves of different algorithms for HPS.
using different algorithms and their execution time are given
in Table 8. Table 9 shows transient specifications of frequency
response for all the algorithms and their percentage improvement
has been calculated considering MFO as benchmark. Fig. 19 por- mathematical formulation of rational oustaloup filter is described
trays the convergence characteristics of these algorithms which as [35]:
shows that the proposed TM-CASO convergence curve is obvi- N
ously better than other algorithms and provides a better global
∏ s + ωi′
G(s) = k (23)
optimal value. Thus, it can be concluded from the results that s + ωi
i=−N
TM-CASO outperforms than PSO, GWO and MFO algorithms. 1
) i+N+2N2+(11 +q)
ωh
(
ωi =
′
(24)
5. Approximation of fractional order operators and stability ωb
analysis 1
) i+N+2N2+(11 −q)
ωh
(
For practical implementation of fractional-order controllers, ωi = (25)
ωb
fractional-order operators are approximated to integer order. In
literature, many methods like oustaloup, Carlson, Matsuda, and where N is defined as an order of approximation, ωh and ωb
continued fraction expansion [35] are considered as rational con- are the higher and lower bound of the selected frequency band,
tinuous approximations in the frequency domain. In this study, ωi′ and ωi are the zeros and poles of the filter by dividing the
the most commonly used oustaloup approximation approach frequency range into 2N+1 intervals.
that relies on recursive distribution of zeros and poles in a To analyse the frequency stability of the power system, the
selected frequency band has been used to convert the non-integer two area HPS model considered in case I with the proposed TM-
order transfer function to integer order transfer function. The CASO tuned FOPID controller is considered. As the performance
14
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

Table 9
Gain values of different algorithms.
Optimization method Signal ST (s) % |P-U| % ITAE
Improvement Improvement
ST in |P-U|
TM-CASO ∆f1 4.96 46.01 0.0395 12.02 0.068
∆f2 6.81 51.21 0.0265 11.07 0.067
PSO ∆f1 6.59 41.78 0.0398 11.35 0.0725
∆f2 7.23 48.21 0.0289 3.02 0.069
GWO ∆f1 8.21 27.47 0.0412 8.24 0.0856
∆f2 9.48 32.09 0.029 2.6 0.0821
MFO ∆f1 11.32 – 0.0449 – 0.091
∆f2 13.96 – 0.0298 – 0.089

6. Conclusion

In this work, an improved form of ASO algorithm (TM-CASO,

SM-CASO and LM-CASO) is proposed to tune the parameters of
FOPID controller for multi-area multi-source HPS that consisting
of both conventional and renewable energy sources. Initially, to
establish the competence of TM-CASO, SM-CASO and LM-CASO
over ASO, five classical benchmark functions have been evaluated.
The output results in terms of mean, best and standard deviation
values clearly substantiate the competence of TM-CASO over
SM-CASO, LM-CASO and ASO algorithms. Further, to prove the
efficacy of chaotic versions of ASO they are used to tune the
FOPID controller for two and three-area power system and the
performance is compared with standard ASO by minimizing ITAE
of ACE. The results of frequency deviation and tie-line power
variation obtained for different cases reveal that the TM-CASO
method outperforms significantly compared to other form of IASO
and standard ASO. This claim has been verified by measuring the
steady state and transient performance indices, it is seen that
the TM-CASO has improvement of 24% to 95%, LM-CASO enhance
the performance by 10% to 80% and SM-CASO improves by an
amount of 5% to 73%, respectively. Further, it is seen that the work
done by the controller measured in terms of controller effort
showing lower effort is required to handle the system to retain
Fig. 20. Bode stability analysis.
steady state frequency even for the three-area system by the
TM-CASO compared to other methods presented. The robustness
of the controller has been tested by considering ±25% variation
in system parameters and its real-time applicability has been
of TM-CASO is predominantly remarkable for several cases con- validated for uncertain variation in solar thermal and wind power
sidered compared with other chaotic methods (SM-CASO and variations as per the Metrological data obtained from Malaysia.
LM-CASO) and standard ASO technique of tuning FOPID con- The results obtained in this case are remarkable with minimum
troller. By applying the oustaloup approximation method for ω settling time and undershoots for the case of TM-CSO than other
∈ [10−2 ,102 ] and N=4 to convert the fractional-order transfer forms presented. In addition, the closed loop stability of the
function (FOTF) to the integer-order transfer function. The order system is studied through oustaloup method of approximation
of transfer function after approximation reaches 48 and it is with Hankel model reduction. The frequency response analysis
difficult to analyse the system stability for high order system. shows that the system is stable for the tuned value of controller.
Hence, the higher-order plant transfer function is reduced to a Although, for a wide operating range, the controller gains need to
second-order transfer function using the Hankel norm method. be retuned for the proposed method. However, these limitations
The detailed procedure for model order reduction using Hankel can be solved by incorporating the proposed algorithm with an
approach for HPS is discussed in previous study of authors [49]. adaptive neural network will be the future scope of this research.
The reduced lower order transfer function obtained using Hankel
method can be expressed as follows: Declaration of competing interest

−1.49s2 − 2.514s − 4.38 The authors declare that they have no known competing finan-
CLTF(case I−A) = (26) cial interests or personal relationships that could have appeared
s2 − 0.00442s − 0.0432
to influence the work reported in this paper.
Fig. 20 represents the frequency response plot of two-area HPS
model considered with gain margin of 3.46 dB and gain cross Acknowledgements
over frequency of infinity. The bode analysis response shows that
the closed-loop system is stable for the proposed TM-CASO tuned The authors gratefully acknowledge Advanced Lightning,
FOPID controller. Similarly, the performance of other cases can power and energy research, Universiti Putra Malaysia for
also be obtained, as the power system is interconnected system providing research fund under UPM, Malaysia Grant No. GP-
and hence it is enough to analyse any one of the cases considered. GPB/2021/9706100 to carry out this research.
15
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

Appendix [19] M.A. Sobhy, A.Y. Abdelaziz, H.M. Hasanien, M. Ezzat, Marine predators
algorithm for load frequency control of modern interconnected power
Synchronizing coefficients: T12 = T21 = 0.08674 systems including renewable energy sources and energy storage units, Ain
Shams Eng. J. (2021).
p.u.MW/rad.Hz, Power system block gain and time constant: [20] I. Pan, S. Das, Fractional order fuzzy control of hybrid power system with
Kpi = 200 Hz/(p.u. MW), Tpi = 20 s, Reheat gain and time renewable generation using chaotic PSO, ISA Trans. 62 (2016) 19–29.
constant: Kri = 0.5, Tri = 10 s, Governor time constant: Tgi = [21] I. Andrew Xavier Raj, N.I. Abdul Wahab, M.G. Umamaheswari, M.A.M.
0.08 s, Steam turbine time constant: Tti = 0.3 s, Frequency bias Radzi, N.B. Sulaiman, V. Veerasamy, S.C. Prasanna, R. Ramachandran, A
Matignon’s theorem based stability analysis of hybrid power system for
coefficient: Bi = 0.425 p.u. MW/Hz, Governor speed regulation:
automatic load frequency control using atom search optimized FOPID
Ri = 2.4 p.u. MW/Hz, Change in step load perturbation with controller, IEEE Access 8 (2020) 168751-168772.
frequency: Di = 8.33 × 10-3 p.u. MW/Hz, System inertia: Hi = 5 [22] S. Padhy, S, Panda, Application of a simplified grey wolf optimization
s. Solar collector gain and time constant: KS = 1.8, TS = 1.8 technique for adaptive fuzzy PID controller design for frequency regulation
s, Solar thermal steam turbine gain and time constant: KT = 1, of a distributed power generation system, Prot. Control Mod. Power Syst.
6 (2021) 1–6.
TT = 0.3 s, Wind turbine gain and time constant: KWTG = 1, [23] N.C. Patel, B.K. Sahu, D.P. Bagarty, P. Das, M.K. Debnath, A novel application
TWTG = 1.5 s, Aqua electrolyzer gain and time constant: KAE = of ALO-based fractional order fuzzy PID controller for AGC of power system
0.002, TAE = 0.5 s, Fuel cell gain and time constant: KFC = 0.01, with diverse sources of generation, Int. J. Electr. Eng. 58 (2021) 465–487.
TFC = 4 s, area capacity ratio: a12 = −1, constant for renewable [24] R. Ramachandran, J. Satheesh Kumar, B. Madasamy, V. Veerasamy, A hybrid
MFO-GHNN tuned self-adaptive FOPID controller for ALFC of renewable
power sharing Kn = 0.6 and loading = 50%. Hydro turbine model
energy integrated hybrid power system, IET Ren Power Gener. 15 (2021)
parameters: TH = 28.75 s, TRS = 1 s, TRH = 0.3, Tw = 3.06 1582–1595.
s, KH = 0.5747 and Gas Turbine model parameters: Cg = 1, [25] I. Podlubny, Fractional-order systems and PIλ Dµ controllers, IEEE Trans.
bg = 0.05 s, Xc = 0.6 s, Yc = 1 s, TCR = 0.01 s, TF = 0.23 s, Autom. Control. 44 (1999) 208–214.
TCD = 0.2 s, KG = 0.1304. [26] A. A. Tepljakov, B.B. Alagoz, C. Yeroglu, E. Gonzalez, S.H. HosseinNia, E.
Petlenkov, FOPID controllers and their industrial applications: a survey of
recent results, IFAC-PapersOnLine 51 (2018) 25–30.
References [27] D. Yang, G. Li, G. Cheng, On the efficiency of chaos optimization algorithms
for global optimization, Chaos Solit. Fractals 34 (2007) 1366–1375.
[1] I. Olle, Elgerd Electric Energy Systems Theory – An Introduction, second [28] M. Premkumar, P. Jangir, R. Sowmya, R.M. Elavarasan, B.S. Kumar, En-
ed., Tata McGraw-Hill, New York, 2007. hanced chaotic JAYA algorithm for parameter estimation of photovoltaic
[2] H. Bevrani, Robust Power System Frequency Control, Springer, New York, cell/modules, ISA Trans. (2021).
2014. [29] R. Tang, S. Fong, R.K. Wong, K.K. Wong, Dynamic group optimization
[3] Y. Arya, Agc performance enrichment of multi-source hydrothermal gas algorithm with embedded chaos, IEEE Access 25 (2018) 22728–22743.
power systems using new optimized FOFPID controller and redox flow [30] B. Khokhar, S. Dahiya, K.S. Parmar, A novel hybrid fuzzy PD-TID controller
batteries, Energy 127 (2017) 704–715. for load frequency control of a standalone microgrid, Arab J. Sci. Eng. 46
[4] Y.V. Hote, S.Jain, PID controller design for load frequency control: Past, (2021) 1053–1065.
present and future challenges, IFAC-Pap. Online 51 (2018) 604–609. [31] B. Khokhar, S. Dahiya, K.S. Parmar, Load frequency control of a microgrid
[5] Y. Arya, N. Kumar, S.K. Gupta, Load frequency control of a four-area power employing a 2D Sine logistic map based chaotic sine cosine algorithm,
system using linear quadratic regulator, Int. J. Energy Sci. 2 (2012) 69–76. Appl. Soft Comput. 7 (2021) 107564.
[6] S.K. Pradhan, D.K. Das, H∞ load frequency control design based on delay [32] W. Zhao, L. Wang, Z. Zhang, Atom search optimization and its application
discretization approach for interconnected power systems with time delay, to solve a hydrogeologic parameter estimation problem, Knowl. -Based
J. Modern. Power Syst. Clean Energy (2020). Syst. 163 (2019) 283–304.
[7] J. Guo, Application of a novel adaptive sliding mode control method to the [33] P. Sun, H. Liu, Y. Zhang, Tu. L., Q. Meng, An intensify atom search
load frequency control, Eur. J. Control 57 (2021) 172–178. optimization for engineering design problems, Appl. Math. Model. 89
[8] M.A. Mohamed, A.A. Diab, H. Rezk, T. Jin, A novel adaptive model predictive (2021) 837–859.
controller for load frequency control of power systems integrated with [34] F.B. Demir, T. Tuncer, A.F. Kocamaz, A chaotic optimization method based
DFIG wind turbines, Neural Comput Appl. 32 (2020) 7171–7181. on logistic-sine map for numerical function optimization, Neural Comput.
[9] S. Prakash, S.K. Sinha, Simulation based neuro-fuzzy hybrid intelligent PI Appl. 32 (2020) 14227–14239.
control approach in four-area load frequency control of interconnected [35] F.N. Deniz, B.B. Alagoz, N. Tan, D.P. Atherton, An integer order approx-
power system, Appl. Soft Comput. 23 (2014) 152–164. imation method based on stability boundary locus for fractional order
[10] R. Ramachandran, B. Madasamy, V. Veerasamy, L. Saravanan, Load derivative/integrator operators, ISA Trans. 62 (2016) 154–163.
frequency control of a dynamic interconnected power system using gen- [36] A. Ali Kadhem, N.I. Abdul Wahab, A.N. Abdalla, Wind energy generation
eralized hopfield neural network-based self-adaptive PID controller, IET assessment at specific sites in a Peninsula in Malaysia based on reliability
Gener. Transm. Distrib. 12 (2018) 5713–5722. indices, Processes 7 (2019) 399.
[11] A. Fathy, A.M. Kassem, A.Y. Abdelaziz, Optimal design of fuzzy PID [37] J. Morsali, K. Zare, M.T. Hagh, Appropriate generation rate constraint
controller for deregulated LFC of multi-area power system via mine blast (GRC) modeling method for reheat thermal units to obtain optimal load
algorithm, Neural Comput. Appl. 32 (2020) 4531–4551. frequency controller (LFC), in: 5th Conference on CTPP, IEEE, 2014, pp.
[12] I. Pan, S. Das, Fractional-order load-frequency control of interconnected 29–34.
power systems using chaotic multi-objective optimization, Appl. Soft [38] S.C. Tripathy, R. Balasubramanian, P.S. Chandramohan Nair, Effect of su-
Comput. 29 (2015) 328–344. perconducting magnetic energy storage on automatic generation control
[13] Z. Bingul, O. Karahan, Comparison of PID and FOPID controllers tuned by considering governor deadband and boiler dynamics, IEEE Trans. Power
PSO and ABC algorithms for unstable and integrating systems with time Syst. 7 (1992) 1266–1273.
delay, Optim. Control Appl. Methods 39 (2018) 1431–1450. [39] A. Boretti, Integration of solar thermal and photovoltaic, wind, and battery
[14] V. Veerasamy, N.I. Abdul Wahab, R. Ramachandran, A. Vinayagam, M.L. energy storage through AI in NEOM city, Energy AI 3 (2021) 100038.
Othman, H. Hizam, J. Satheeshkumar, Automatic load frequency control [40] D. Kler, V. Kumar, K.P. Rana, Optimal integral minus proportional
of a multi-area dynamic interconnected power system using a hybrid derivative controller design by evolutionary algorithm for thermal-
PSO-GSA-tuned PID controller, Sustainability 24 (2019) 6908. renewable energy-hybrid power systems, IET Ren. Power Gener. 13 (2019)
[15] R.K. Sahu, T.S. Gorripotu, S. Panda, Automatic generation control of 2000–2012.
multi-area power systems with diverse energy sources using teaching [41] D.J. Lee, L. Wang, Small-signal stability analysis of an autonomous hy-
learning-based optimization algorithm, Int. J. Eng. Sci. Technol. 19 (2016) brid renewable energy power generation/energy storage system part I:
113–134. Time-domain simulations, IEEE Trans. Energy Convers. 23 (2008) 311–320.
[16] H.M. Hasanien, Whale optimisation algorithm for automatic generation [42] A. Kaveh, Advances in Metaheuristic Algorithms for Optimal Design of
control of interconnected modern power systems including renewable Structures, Springer International Publishing, Switzerland, 2014.
energy sources, IET Gener. Transm. Distrib. 12 (2018) 607–614. [43] R. Tang, S. Fong, N. Dey, Metaheuristics and chaos theory, Chaos Theory
[17] D. Guha, P.K.Roy, S. Banerjee, Quasi-oppositional JAYA optimized 2-degree- (2018) 182–196.
of-freedom PID controller for load-frequency control of interconnected [44] Y. Zhou, L. Bao, C.P. Chen, A new 1D chaotic system for image encryption,
power systems, Int. J. Simul. Model. 15 (2020) 1–23. Signal Process. 97 (2014) 172–182.
[18] D. Guha, P.K. Roy, S. Banerjee, Multi-verse optimisation: a novel method [45] J. Sun, Y. Li, Social cognitive optimization with tent map for combined heat
for solution of load frequency control problem in power system, IET Gener. and power economic dispatch, Int. Trans. Electr. Energy Syst. 29 (2020)
Transm. Distrib. 11 (2017) 3601–3611. e2660.

16
A.X.R. Irudayaraj, N.I. Abdul Wahab, M. Premkumar et al. Applied Soft Computing 119 (2022) 108574

[46] D. Yousri, A.M. AbdelAty, L.A. Said, A.S. Elwakil, B. Maundy, A.G. Rad- [48] S.M. Lurwan, N. Mariun, H. Hizam, M.A.M. Radzi, A. Zakaria, Predicting
wan, Chaotic flower pollination and grey wolf algorithms for parameter power output of photovoltaic systems with solar radiation model, IEEE
extraction of bio-impedance models, Appl. Soft Comput. 75 (2019) Int. Conf. Power Energy (2014) 304–308.
750–774. [49] V. Veerasamy, N.I.A. Wahab, R. Ramachandran, M.L. Othman, H. Hizam,
[47] S. Barshandeh, M. Haghzadeh, A new hybrid chaotic atom search optimiza- A.X. Irudayaraj, J.M. Guerrero, J. Satheesh Kumar, A Hankel matrix based
tion based on tree-seed algorithm and Levy flight for solving optimization reduced order model for stability analysis of hybrid power system using
problems, Eng. Comput. 37 (2021) 3079–3122. PSO-GSA optimized cascade PI-PD controller for automatic load frequency
control, IEEE Access 8 (2020) 71422–71446.

17