Professional Documents
Culture Documents
https://doi.org/10.1007/s12652-022-03751-x
ORIGINAL RESEARCH
Abstract
Nonlinear structure, vague loading characteristics and parameter uncertainties of the interconnected power system (IPS) have
recently given birth to various controllers that better deal with automatic generation control (AGC). AGC plays a key role
in ensuring the balance of generation and load demand in IPSs. If this balance is lost, then the system faces large frequency
deviations. Thus, this work proposes a new fuzzy 1 + proportional + derivative-proportional + integral (F1PD-PI) controller to
escalate AGC performance of different IPSs integrated with renewable energy sources (RES) including wind, solar and fuel
cells. Inspiration for the proposed controller is unique and comes from combining the merits of fuzzy, 1PD and PI controllers.
Salp swarm algorithm (SSA) is utilized to optimize the proposed controller’s gains as well as fuzzy membership functions.
The effectiveness and contribution of the advocated approach are demonstrated on a two-area reheat thermal system and a
two-area multi-source hydro-thermal system by realizing an extensive comparison study with the state-of-the-art variants.
The results substantiate that SSA optimized F1PD-PI controller has better performance than its competing peers in terms of
minimum settling time, peak undershoot, peak overshoot and error-integrating performance criterion of the system responses.
Nonlinearities from governor dead band and generation rate constraint are also studied, which verifies the performance of
the control strategy in tackling nonlinearities. Additionally, the robustness of the controller is affirmed against parameter
uncertainties and load disturbances. Finally, the stability of our proposal is checked using eigenanalysis.
Keywords Interconnected power system · Automatic generation control · Renewable energy · Fuzzy logic controller ·
Governor dead band · Generation rate constraint · Salp swarm algorithm
13
Vol.:(0123456789)
E. Çelik
growing population have turned the modern IPS out to be vehicles. As a result of the above review, the contribution
a huge and intricate network. On the other hand, the power of RES on different AGC systems is found to be salient.
demand on IPS is continually changing and this causes Yet to our knowledge, there is still some research gap to be
undesirable fluctuations in output voltage and frequency of bridged in AGC of IPSs that exploit the usefulness of RES
the alternator over the whole network, where all the gen- like STGs, WTGs and AE-FC units all together. This entails
erating units must perform synchronously with a common further investigation as launched in this work.
frequency. Frequency variations of different control areas The performance of an AGC mechanism depends
also result in transient deviations in the tie-line power flows. closely on three elements: (i) controller structure (ii) opti-
Preserving frequency and tie-line power flows within nomi- mization technique opted for efficient gain scheduling and
nal ranges is possible only by matching the power genera- (iii) objective function used in the optimization algorithm
tion with demanded power at all times. Such equilibrium in for assessing the quality of obtained solutions (Çelik et al.
IPSs is maintained by a strategy called automatic genera- 2021a). The relevant literature has witnessed an enor-
tion control (AGC). AGC allows the generators of control mous volume of researches particularly on new controller
areas to set their output generations in accordance with the designs. Some of them are firefly algorithm (FA) opti-
load demands. This way, steady state errors of the system mized PI controller (Abd-Elazim and Ali 2018), improved
responses are ensured at zero and the IPS’s synchronism is stochastic fractal search (ISFS) algorithm (Çelik 2020)/
sustained during normal and abnormal situations. In view differential search algorithm (DSA) (Guha et al. 2017a)/
of the above discussion, AGC needs for a successful control differential evolution (DE) (Hota and Mohanty 2016)/
strategy to effectively suppress the effect of random load symbiotic organisms search (SOS) algorithm (Yılmaz
variations. It is to be noted that simplicity and easy con- et al. 2021; Hasanien and El-Fergany 2017) optimized
troller commissioning are usually ignored in the literature PID controllers, quasi-oppositional SOS (QOSOS) (Guha
works, yet these properties are of great importance from the et al. 2017b)/backtracking search algorithm (BSA) (Guha
view point of practical implementation. et al. 2018)/hybrid FA-pattern search (hFA-PS) (Sahu
Fossil fuels currently meet the significant portion of et al. 2015)/grey wolf optimization (GWO) (Guha et al.
the world’s total energy demand. They are finite resources 2016a) optimized PI/PID controllers, DE optimized I/PI/
depleting day by day and hazardous to the environment. As PID controllers (Mohanty et al. 2014), jaya algorithm (JA)
per the U.S. Energy Information Administration’s note, 76% optimized PID with first-order low-pass filter (Singh et al.
of U.S. greenhouse gas emissions came from burning the 2017), hybrid stochastic fractal search and local unimodal
fossil fuels in 2016. As such, the negative effects of fossil sampling (hSFS-LUS) algorithm based multistage PDF
fuels started to make the obligatory transition away from a plus (1 + PI) controller (Sivalingam et al. 2017), hybrid
destructive fossil fuel towards using an alternative energy stochastic fractal search and pattern search technique
resource like renewable energy. As the name implies, renew- (hSFS-PS) based cascade PI-PD controller (Padhy and
able energy sources (RES) are renewable, clean and free in Panda 2017), sine–cosine algorithm (SCA) optimized dou-
nature, which give birth to fascinating attraction to benefit ble derivate filter based PD-PID cascade controller (Sata-
from them in generating electricity. Especially, solar and pathy et al. 2018), biogeography based optimised three
wind energy are broadly available in nature and easier to be degree of freedom (3DOF) PID controller (Rahman et al.
accessed compared to other RES. Thus, one of the current 2016), evolutionary algorithm designed integral minus PD
attentions of researchers is centered on the integration of (IPD) controller (Kler et al. 2019), and dragonfly search
solar and wind into the modern IPSs. In a study by Rahman algorithm (DSA) optimized (1 + PD)-PID controller (Çelik
et al. (2016), solar thermal system considering the influence et al. 2021a). The above-counted classical controllers are
of a dish-stirling is analyzed in AGC of a two-area reheat developed around a certain condition and thus they may
thermal system (RTS). An AGC research on a two-area RTS perform poorly in a changeable operating environment
supported by RES such as wind turbine generators (WTGs), as frequently experienced in power systems. To avert the
solar thermal generators (STGs) and aqua electrolyser (AE)- shortcomings of classical controllers, there are also other
fuel cells (FCs) units is carried out in Kler et al. (2019). advanced control schemes grounded on sliding modes
The effect of AE-FC units on AGC capability of two-area (Dahiya et al. 2019), H-infinity (Rosaline and Somarajan
non-reheat thermal system, PV-RTS and multi-source hydro- 2019), fractional calculus theory (Tasnin et al. 2018; Tas-
thermal system (MSHTS) is also demonstrated in Arya nin and Saikia 2018; Arya and Kumar 2017a; Çelik 2021;
(2019). Besides, the outcomes of STG and geo thermal Arya et al. 2021a,b), artificial neural network approach
(GT) systems have been investigated in deregulated systems (Saikia et al. 2011), non-linear disturbance observer
with 2 and 3 areas (Tasnin et al. 2018; Tasnin and Saikia (Sharma and Mishra 2019), disturbance observer aided
2018). Guangdong power grid is explored in Xi et al. (2020) optimized fractional order 3DOF tilt-integral-derivative
by benefitting from thermal, hydro, wind, PV and electric (FO3DOF TID) (Guha et al. 2021). However, many of
13
Performance analysis of SSA optimized fuzzy 1PD‑PI controller on AGC of renewable energy assisted…
these controllers suffer from complex control laws and In summary, the main objectives of this paper can be out-
heavy computational burden. At this point, fuzzy logic lined as step-down:
controller (FLC) as an intellectual approach can improve
the current situation by achieving a better balance between i. To design a new fuzzy 1 + proportional + deriva-
performance amelioration and design complexity. tive-proportional + integral (F1PD-PI) controller to
FLC is a good choice when dealing with complex and ameliorate AGC performance of two-area RTS and
nonlinear control systems that have vague or unknown MSHTS.
mathematical models as it intends to stimulate the human ii. To apply SSA concept to search for optimal controller
knowledge in the form of linguistic rules (Çelik and parameters.
Öztürk 2019). In the case of IPS, there are several works iii. To examine the outcomes of RES penetration from
proved successful using FLC (Arya and Kumar 2017b; wind, solar and fuel cells.
Rajesh et al. 2019; Sahu et al. 2014; Debnath et al. 2017). iv. To authenticate the potential and contribution of the
The performance of FLC depends greatly on its adjust- proposed approach comparing with the state-of-the-art
able design parameters such as input/output scaling fac- (SOTA) variants for the similar systems.
tors (IOSFs), control rules (CRs) and membership func- v. To challenge the advocated controller with nonline-
tions (MFs. Classically, they are chosen by trial or expert’s arities resulting from governor dead band (GDB) and
guidance if it is available. However, in either case, the generation rate constraint (GRC).
optimal design cannot be guaranteed and much time is vi. To evaluate the system reaction against different limits
devoted to the design process (Öztürk and Çelik 2012; of GRC.
Çelik and Öztürk 2021). To solve this problem, various vii. To investigate the controller robustness with respect
optimization techniques are applied to automate the FLC to parameter uncertainties and load disturbances.
design in different kinds of IPSs by viewing the problem viii. To check the controller stability using eigenanalysis.
from the perspective of optimization. As the simultaneous
optimization of the whole FLC parameters is computation- In view of our literature knowledge, it seems there are two
ally expensive, the generic trend in the related works is to main innovations in this work differentiating the proposed
optimize IOSFs only while other tunable parameters are approach from the SOTA. Firstly, to date, no attempt has been
remained fixed. In the fuzzy PI/PID (FPI/FPID) control- made to design a F1PD-PI controller in the control engineer-
ler, BFOA technique is employed to tune IOSFs (Arya ing area. We test the performance of this controller in AGC of
and Kumar 2017b). Hybrid improved FA-PS (hIFA-PS) two-area RTS and MSHTS with and without RES penetration.
(Rajesh et al. 2019) and hybrid DE-PSO (hDE-PSO) (Sahu Secondly, an effective way of optimizing membership func-
et al. 2014) are proposed to successfully optimize IOSFs tions is presented and utilized to improve the controller perfor-
of FPID controller for diverse IPSs. Since the derivative mance further. All the relevant works in the literature focus on
action of FPID controller is sensitive to noise in area optimizing the controller gains only, leaving the membership
control error, a first-order low-pass filter is considered in functions unoptimized.
the derivative part to obtain FPID controller with filter The article layout is as follows: Sect. 2 presents TF block
(FPIDN) that eliminates the unwanted action of the pre- diagrams of IPS models under consideration. The structure of
vious version. In Debnath et al. (2017), JA protocol is F1PD-PI controller is proposed in Sect. 3. The current state of
implemented to optimize the IOSFs of FPIDN controller metaheuristic algorithms is reviewed in Sect. 4. After introduc-
and it is revealed that FPIDN performs better than FPID. ing to SSA in Sect. 5, the controller optimization is carried
There are also other recent researches that combine FLC out in Sect. 6. Section 7 is devoted to extensive simulation
and fractional order (FO) theory, such as imperialist com- results obtained and their assessment with respect to the exist-
petitive algorithm (ICA) based FPIDN-FOPIDN (Arya ing results. In Sect. 8, nonlinear effect of GDB and GRC on
et al. 2021a)/cascade fuzzy-FOIDN (C-IλDμN) (Arya et al. the system performance is analyzed which is followed by the
2021b) structured controllers. Consequently, we realize robustness analysis in Sect. 9. The stability analysis and the
that many FLC configurations hybridized with classical computational complexity are presented in Sects. 10 and 11.
and FO controllers exist in the field and various algorithms Finally, the paper is concluded in Sect. 12.
are utilized to procure the optimal controller gains and
IOSFs of the FLC. Nonetheless, a systematic attempt to
optimize MFs is currently open in the relevant literature. 2 Studied IPS models
Moreover, in addition to the inconveniences identified for
the earlier control schemes, the system responses offered IPS models investigated in this paper are two-area RTS and
by them are still in need of being improved further, as will two-area MSHTS depicted in Fig. 1. In the systems depicted
be demonstrated extensively in this article. in Fig. 1a and b, both generation areas are of equal sizes. A
13
E. Çelik
Area 1
1
ΔPD1
Governor-1 Reheat turbine-1 Power system-1
Controller-1
ACE1 ΔPref1 1 ΔPg1 ΔF1
SSA: F1PD-PI
+ 1 + 1
ΔPgRES
RES ΔPtie
1
ΔPD2
Area 2
(a)
Area 1
(b)
Solar Thermal power system
Solar part-1 Thermal part-1
ΔPgSTG
Psol
Solar part-2 Thermal part-2
ΔPgRES
Kn
Wind power system Wind turbine-1
Fuel Cell power system
1−Kn Fuel cell-1
Wind turbine-2
ΔPgWTG ΔPgFC
PW
Fuel cell-2
Aqua
Wind turbine-3 Electrolyzer
(c)
13
Performance analysis of SSA optimized fuzzy 1PD‑PI controller on AGC of renewable energy assisted…
single reheat steam turbine exists in each area of RTS, and is assumed as load disturbance subjected to the whole sys-
a non-reheat steam turbine and a hydro turbine are included tem. The outer loop equation can be given by Eq. 1.
in each area of MSHTS. Besides, area-1 of both systems
is equipped with RES based generating unit consisting of
Y(s) = G1 (s)u1 (s) (1)
2 STGs, 3 WTGs, 1 AE and 2 FCs as shown in Fig. 1c.
The STG has solar part and thermal part. The solar part has where u1 (s) = y2 (s) − d1 (s) . Assuming d1 (s) = 0 , then the
solar energy collectors that capture and focus sunlight onto outer process input u1 (s) becomes the output from the inner
a receiver. A special fluid is heated and circulated in the process y2 (s).
receiver to produce steam. The steam is then transformed The inner loop includes the inner or supply process
into mechanical energy in a turbine, which drives a generator G2 (s) and its equation can be given by Eq. 2.
to produce electrical energy. In the AE, hydrogen energy is
acquired by using some of the power from STGs and WTGs y2 (s) = G2 (s)u2 (s) (2)
(specified by Kn ) to be fed to FCs that convert hydrogen into
electricity via a couple of chemical redox reactions.
where u2 (s) is the inner process input. The inner loop is capa-
Every generating unit of the systems has its own speed
ble of making an immediate effort to confine the effect of
governor system, turbine and generator. For an area, there are
inner process gain variations arisen from set point changes
three inputs and two outputs. The inputs indicated with blue
or disturbances (Arya et al. 2021b).
color are controller output ΔPref , load disturbance ΔPD and
tie-line power error ΔPtie . The output of RES, ΔPgRES , is also
inputting to area-1. The outputs highlighted by brown color
3.1 F1PD‑PI controller
are area frequency error ΔF and area control error (ACE). The
test models and data for the system parameters are taken from
The main role of a controller developed for AGC is to quickly
the work by Arya et al. (2021b). At the end of the paper, the
drive frequency, generation and tie-line power deviations
relevant parameters values are given and nomenclature is also
toward zero under step load disturbance (SLD), parameter
provided to describe several parameters utilized.
uncertainty and other unknown disturbances. In this case, an
area control error (ACE) is generated and the controller pro-
duces a control signal by acting on ACE to accommodate new
3 Cascade control
working condition. In our literature survey presented previously,
many controllers tuned by several optimization techniques are
Cascade control conception includes the control of two sub-
investigated thoroughly considering their advantages and dis-
sequent processes, where the inner process or the output
advantages. However, the cascade combination of fuzzy, 1PD
of the first feeds the outer or second process in order. The
and PI controllers (F1PD-PI controller) has not been witnessed
fundamental cascade control system consists of two control
to date. Therefore, the performance analysis of this controller
loops (inner and outer) as seen in Fig. 2 (Arya et al. 2021a).
on AGC of different IPSs is worthy studying and thus we have
The inner loop is responsible for alleviating the impact of
addressed it in this paper.
supply disturbance or any internal disturbance on the outer
The proposed controller design is depicted in Fig. 3. As
process whereas the outer loop manages to control the qual-
seen, two controllers are chosen and named C1 (s) and C2 (s) as
ity of process final output. Such a control system is famous
the outer controller and the inner controller, respectively. Here,
for its swift rejection of disturbance before it spreads over
F1PD controller is set as outer controller and PI as inner con-
other parts of the plant (Çelik 2021).
troller. The F1PD controller receives ACE signal as input while
In Fig. 2, the outer loop carries the process output Y(s) ,
the PI controller produces the output control signal or plant
which is controlled by primary control to achieve a given
input signal ΔPref . u is the output of FLC, which is obtained
reference signal R(s) . G1 (s) is the outer process and d1 (s)
by fuzzy approximation using ACE and the time derivative of
ACE. Since the ACE and its time derivative are observed to be
Secondary inherently in the range [− 1, 1], no scaling factors are used for
controller the inputs. The output scaling factors of FLC correspond to the
gains of 1PD and PI controllers.
Primary In accordance with Figs. 1a and b, ACE1 and ACE2 can
Inner loop
controller be defined by
Outer loop
ACE1 = −BΔF 1 − ΔPtie
ACE2 = −BΔF 2 + ΔPtie (3)
Fig. 2 Cascade control conception
13
E. Çelik
u + 1 + ΔPref
ACE Kp1 + + Ki +
Δu + s
Δt
Mux Δu
Kd ΔPtie
Δt
Table 1 Rule table relating ACE and ACE derivative to u • If ACE is Negative Big and ACE derivative is Negative
ACE derivative Big, this signifies that the plant input signal ΔPref is so big
NB NM NS ZE PS PM PB
at present, thus u should be Negative Big.
• If ACE is Positive Big and ACE derivative is Positive Big,
NB NB NB NB NM NM NS ZE
NM NB NM NM NM NS ZE PS this signifies that the plant input signal ΔPref is so small at
NS NB NM NS NS ZE PS PM present, thus u should be Positive Big.
A
C ZE NM NM NS ZE PS PM PM • If ACE is Zero and ACE derivative is Zero, this signifies
E PS NM NS ZE PS PS PM PB that the plant input signal ΔPref is in its required value at
PM NS ZE PS PM PM PM PB present, thus u should be Zero.
PB ZE PS PM PM PB PB PB • If ACE and ACE derivative fulfil the other cases, this signi-
fies that the plant input signal ΔPref is neither too large nor
too small; thus u should be a transition value to the Nega-
where B is the frequency bias, ΔF i is area frequency error/
tive Big or Positive Big.
deviation and ΔPtie is tie-line power error/deviation. Thus,
the overall control rule relating ΔPref to ACE can be repre-
sented as in Eq. 4.
As the fuzzy inference method, the Mamdani’s min–max
( ) approach is employed using the following model:
({ } ) K
ΔPref = u + uKp1 + usKd − ΔF − ΔPtie × Kp2 + i
s Rn = An × Bn (6)
(4)
where Kp1 , Kd , Kp2 and Ki are proportional, derivative, pro- where Rn shows the nth fuzzy relation between two antecedent
portional and integral gains, respectively. The optimal values linguistic terms withn = 1, 2, .., K . Then, the membership degree
of these gains are procured employing SSA. of each relation is computed by the minimum (^) operator:
In FLC, due to its simplicity, we use an interval type-1 ( ) ( ) ( )
fuzzy system, where the membership is a crisp value. 7 MFs 𝜇Rn x1 , x2 = 𝜇An x1 ∧ 𝜇Bn x2 (7)
are chosen for the two inputs and one output. The linguistic
values are defined as NB: negative big, NM: negative medium, Notice that the minimum is computed for all possible pairs
NS: negative small, ZE: zero, PS: positive small, PM: positive of x1 and x2 , which is equal to K = 4.
medium and PB: positive big. All the MFs are in triangular The fuzzy relation R representing the entire model can be
form, lying over the interval [− 1, 1]. Thus, degree of member- obtained by combining the K individual rule’s relations as
ship for each input is calculated by the following rule: follows:
( ) ( )
R = ∪Kn=1 Ri ,
𝜇Aj xi = fAj xi
i i
(5) [ ]
that is, 𝜇R (x1 , x2 ) = max 𝜇An (x1 ) ∧ 𝜇Bn (x2 ) (8)
⏟⏟⏟
1≤n≤K
where xi is the ith input, Ai ( i = 1, 2 and j = 1, 2, .., 7 ) is the
j
jth MF for the ith input and f is the mathematical function In the defuzzification process the crisp output u is
for triangular shape. obtained from the output of FLC using the center of gravity
Because 7 MFs are considered for each input, the total num- (centroid) technique in the following:
ber of possible rules is 49, which are constructed as in Table 1.
These rules are declared based on the expert knowledge and
extensive simulations made in this work. The following rules
describe the function of Table 1.
13
Performance analysis of SSA optimized fuzzy 1PD‑PI controller on AGC of renewable energy assisted…
13
E. Çelik
( )
5 Salp swarm algorithm j
xi = lbj + r ubj − lbj (10)
dimension.
The pseudocode of SSA is illustrated in Algorithm 1. It
Follower
salp displays that SSA technique starts seeking the global opti-
Leader mum by initiating n search agents with d dimensions ran-
salp domly in the search space. It then calculates the fitness of
Direction of
each salp, devises the one with the best fitness and assigns
movement the position of the best salp to the variable F as the food
source that has been the most valuable information obtained
so far about the global optimum. After updating the param-
eter c1 using Eq. 13, each dimension of the leading salp is
calculated by Eq. 12 and the positions of the follower salps
Fig. 5 The shape of salp chain
13
Performance analysis of SSA optimized fuzzy 1PD‑PI controller on AGC of renewable energy assisted…
by Eq. 14. All the above-described steps except initialization The number of gains to be tuned for the proposed F1PD-PI
are iterated till the satisfaction of a termination criterion, and controller is 4, which are Kp1 , Kd , Kp2 and Ki . Optimization of
in the end, SSA reports F as the final solution. the MFs is explained next.
One triangular MF is normally declared by at least three
Algorithm 1: The salp swarm algorithm (SSA) parameters. This in turn forms a complex optimization problem
Initialize the salp population [ ] considering and in our case, which demands to tune seven triangular MFs for each
while < do variable of FLC. To reduce the number of tunable parameters and
Evaluate the fitness of each candidate solution (salp) expedite the optimization, the following assumptions are made
Set as the best solution in the present paper (Öztürk and Çelik 2012):
Update the value of 1 using Eq. 13
for each salp ( , = 1,2, . . , ) do • The arrangement of MFs is symmetrical about the y-axis.
if ( == 1) then
• The peak values of MFs at two edges are − 1 and + 1,
Update the position of the leading salp using Eq. 12
else respectively.
Update the position of the follower salp using Eq. 14 • A crisp input can fire at most two MFs, and the sum of the
end if associated membership degrees must be 1.
end for
Update the salps based on the upper and lower bounds of Thanks to the above assumptions, the center points of input
variables
and output MFs can be easily characterized using the follow-
= +1
end while ing formula:
return the best solution ( ) ( )cp
i
(16)
n
where n = m−12
, m is the number of MFs and i = 1, 2, .., n.
In Eq. 16, the only tunable parameter is cp . As one illus-
6 Controller optimization trative example, Fig. 6 shows the shape of seven MFs when
cp = 0.5 and cp = 1.75 whereas they appear as in Fig. 4 when
The goal of a controller optimization is to procure the opti- cp = 1 . As seen, if cp < 1 , the center points of MFs move
mal controller parameters that satisfy the given control away from the μ-axis and otherwise they are gathered closely
objectives. These control objectives are generally encapsu- around the μ-axis.
lated in an objective function ( J ) designed for the system As we have two input and one output variable for the
and its value is minimized by an optimization technique FLC, three more parameters, namely cp1 , cp2 and cp3 , are
considering the system constraints. At the end of the opti-
mization process, minimum value of J results in optimal
controller parameters. Here, it is worth stressing that the µ
attained controller parameters are expected to be optimal or
near-optimal, yet they may not be the true optimal. NB NM NS ZE PS PM PB
1
formulated in Eq. 15, ISE integrates the square of the three ACE (pu)
ACE derivative (pu)
error terms over time and it penalizes large errors more than u (pu)
smaller errors. Thereby, ISE has the merit of eliminating large -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
NB NM NS ZE PS PM PB
tsim 1
( 2 )
J = ISE = ΔF 1 + ΔF 22 + ΔP2tie dt (15)
∫ 0.5
0
ACE (pu)
ACE derivative (pu)
u (pu)
(b)
13
E. Çelik
required to be optimized. Thus, F1PD-PI controller has a Later, according to the calculated fitness values, the salps
total of 7 parameters. The optimization problem that we con- are updated in the algorithm for the next iterations. Any
sider can be formulated in the following, where the bounds salp violating the constraints in Eq. 17 is brought back on
of each controller parameter act as problem constraints. the boundaries. The process of going forward and back
Minimize J , subject to: between the Simulink model and the SSA routine is iterated
min max
till Miter = 50.
⎧ Kp1 ≤K
p1
≤ Kp1 The optimization process is repeated 30 times, and F1PD-
⎪
⎪
Kdmin ≤ K d ≤ Kdmax PI controller parameters with and without RES offering
⎪ min
Kp2 ≤ K ≤ Kp2 max
minimum value of J over 30 runs are assumed optimum
⎪ p2
and reported in Table 2 for both RTS and MSHTS. In any
⎨ Kimin ≤ K i ≤ Kimax (17)
⎪ cmin ≤ c ≤ cmax of the following tables, better results are bolded and a quite
p1 p1
⎪ p1
narrow tolerance band of ± 0.00005 is considered for meas-
⎪ cmin ≤c ≤ cmax
p2 p2 p2 uring settling time.
⎪ cmin ≤ cmax
⎩ p3
≤c
p3 p3
7.2 Two‑area reheat thermal system
where the superscripts min and max speak for the minimum The performance assessment of the proposed F1PD-PI
and maximum values of the respective controller param- controller is initially made on a two-area RTS assisted by
eter. Following the termination of SSA, the set of controller RES in area-1 as shown in Fig. 1a. The system dynamic
parameters yielding the smallest value of J is assumed opti- responses such as area frequency deviations ( ΔF1 , ΔF2 )
mal and they are utilized to collect the results of the systems. and tie-line power deviation ( ΔPtie ) are acquired by
our proposal and presented in Fig. 8 when a SLD of
0.01puMW is applied to area-1 at t = 0. Meanwhile, at the
7 Simulation results same instant, it is assumed STGs and WTGs are subject
to a SLD of 0.001puMW and 0.002puMW, respectively.
7.1 SSA implementation for the controller synthesis In the zoomed views, the dashed lines are displayed to
clearly show the tolerance band. To rigorously check the
The general block diagram of the proposed control scheme value and contribution of SSA tuned F1PD-PI controller,
is presented in Fig. 7. The studied IPS models shown in several comparisons with the SOTA like bacterial swarm
Fig. 1, as well as the F1PD-PI controller recommended in optimization (BSO) optimized FOPID (Özdemir et al.
Fig. 3, are modelled in Matlab/Simulink platform. Program 2015), SOS optimized PIDN (Hasanien and El-Fergany
codes of SSA routine are written using Matlab/Script editor 2017), hDE-PSO optimized FPID (Sahu et al. 2014),
which is linked to the Simulink model in order to assess the SCA optimized C-PDN-PIDN (Satapathy et al. 2018), JA
quality of candidate solutions in SSA. At the beginning of optimized FPIDN (Debnath et al. 2017), dragonfly search
the algorithm, 50 salps are generated randomly; each salp algorithm (DSA) optimized FOPI-FOPD (Çelik 2021),
has seven controller parameters, i.e. Kp1 , Kd , Kp2 , Ki , cp1 , improved SCA (ISCA) optimized adaptive fuzzy based
cp2 and cp3 . Then, each salp is fed into the Simulink model PID (AFPID) (Rajesh and Dash 2019), chaotic crow search
to calculate the associated fitness value J after simulating algorithm (CCSA) optimized hybrid fuzzy proportional
and acquiring the system responses of ΔF1 , ΔF2 and ΔPtie . derivative–tilt integral derivative (FPD-TID) (Khokhar
et al. 2021) and ICA optimized C-I λD μN (Arya et al.
13
Performance analysis of SSA optimized fuzzy 1PD‑PI controller on AGC of renewable energy assisted…
Tolerance Tolerance
Band Band
2021b) controllers are established. Inspecting the results in area-1, the thermal power generation is ΔPgThermal = 0.00425
Fig. 8 points out that F1PD-PI controller offers marvelous puMW. Thus, the total power generated in area-1 is
performance over other reported approaches in obtaining ΔPg1 = ΔPgRES + ΔPgThermal = 0.01 puMW, which is equal
fast responses with negligible Ts ∕Us ∕Os in ΔF1 , ΔF2 and to the power demand ΔPD1 as necessitated. Meanwhile,
ΔPtie signals. The closest competitor to F1PD-PI control- owing to ΔPD2 = 0 puMW, the generated power in area-2
ler is FPD-TID controller, which is followed by the AFPID is ΔPg2 = 0 puMW.
controller. However, this controller results in some salient
steady state error comparing with other controllers. It is 7.3 Two‑area multi‑source hydro‑thermal system
also found that C-IλDμN controller gains the fourth rank.
To evaluate the system dynamic responses quantitatively, To investigate the potential of the F1PD-PI controller
time-domain characteristics of the responses such as set- further, the study is forwarded by considering a two-area
tling time Ts , undershoot Us and overshoot Os are chosen MSHTS. The TF block diagram of this system is displayed
as comparison criteria among the approaches. Numerical in Fig. 1b. Two generating units based on non-reheat ther-
values of Ts , Us , Os and J are measured from Fig. 8 and mal and mechanical governor driven hydro are available in
noted in Table 3. As per the results reported, SSA optimized area-1 and area-2. Additionally, RES contributes collec-
F1PD-PI controller has superior performance over its pub- tively to the total power generation of area-1. The simulation
lished peers in terms of smaller values of J ( = 1.77E − 10 ), results showing ΔF1 , ΔF2 and ΔPtie responses are obtained
Ts ( ΔF1 = 0.01 s, ΔF2 = ΔPtie = 0 s), Us ( ΔF1 = −0.0001 by the advocated approach under a SLD of 0.015 puMW to
Hz, ΔF2 = −2.06E − 6 Hz, ΔPtie = −8.56E − 7 puMW). area-1 at t = 0 and demonstrated in Fig. 10. For compari-
Moreover, thanks to the inclusion of RES generating pow- son, numerous results offered by the earlier studies such as
ers simultaneously in critical moments by STGs, WTGs and quasi-oppositional differential search algorithm (QODSA)
AE-FC units, the system performance are further improved (Guha et al. 2016b)/GWO (Guha et al. 2016a)/hFA-PS (Sahu
by the proposed controller via further amelioration in J et al. 2015) optimized PID controller, BFOA optimized FPI/
( = 7.14E − 11 ), Ts ( ΔF1 = 0.008 s) and Us ( ΔF1 = −0.00007 FPID/FOPID (Arya and Kumar 2017a,b), DSA optimized
Hz, ΔF2 = −1.94E − 6 Hz, ΔPtie = −7.98E − 7 puMW). FOPI-FOPD (Çelik 2021), ISCA optimized AFPID (Rajesh
The percentage improvement in J contributed by RES is and Dash 2019), CCSA optimized FPD-TID (Khokhar et al.
calculated as 59.6%. Considering the value of J , our pro- 2021) and ICA optimized FPIDN/C-IλDμN (Arya et al.
posal takes the first rank for this system, which is followed 2021b) are also presented in Fig. 10. As we can see, SSA
by FPD-TID, AFPID, C-IλDμN and so on. optimized F1PD-PI controller is able to accomplish the best
Figure 9 displays the generated power by each gen- transient response by rejecting the given SLD very fast com-
eration unit in area-1 and area-2. As discussed in the paring with all other published approaches. The second-best
introduction part of this article, frequency regulation is controller in this case is again FPD-TID controller, which is
accomplished by matching the total generated power to followed by AFPID controller. Nonetheless, similar to the
demanded load. As seen, the power generated by RES is previous case, AFPID controller is not good at eliminating
ΔPgRES = 0.00575 puMW, which is contributed by STGs the steady state error in the responses. The performance of
( ΔPgSTG = 0.0036 puMW), W TGs ( ΔPgWTG = 0.006 C-IλDμN controller comes in the fourth place, outperforming
puMW) and FCs ( ΔPgFC = 1.29E − 7 puMW). In the rest of published controllers in the literature.
13
E. Çelik
7.14E − 11
1.77E − 10
6.61E − 9
2.38E − 8
3.79E − 6
7.44E − 7
2.57E − 6
2.05E − 6
1.76E − 5
6.57E − 6
1.07E − 7
3.31E − 8
and ΔPtie responses are comparatively reported in Table 4.
Note that n/a means not applicable. In this case, settling
times of ΔF1 and ΔF2 obtained by AFPID controller are
J
enter into the tolerance band over the whole simulation time.
3.62E − 6
2.94E − 5
2.30E − 5
9.15E − 4
Critical observation of Table 4 points out that SSA tuned
ΔPtie
0
0
0
0
0
0
0
0
regard to smaller values of J ( = 8.31E − 10 ), Ts ( ΔF1 = 0.23
s, ΔF2 = 0.13 s, ΔPtie = 0 s), Us ( ΔF1 = −0.0015 Hz,
1.42E − 6
8.54E − 6
7.25E − 5
4.85E − 5
2.65E − 3
0
0
0
0
0
Ts ( ΔF1 = 0.15 s, ΔF2 = 0 s), Us ( ΔF1 = −0.0014 Hz,
ΔF2 = −4.32E − 5 Hz, ΔPtie = −1.19E − 5 puMW) and Os
1.10E − 4
1.99E − 4
8.36E − 5
8.92E − 5
2.37E − 4
3.56E − 3
|Os |(Hz)
0
0
0
0
0.000130
0.000034
0.00013
0.00007
0.00007
0.00010
0.0010
0.0005
ΔF1
| |
3.73
1.83
2.39
3.83
3.75
6.41
0.63
0
0
0
1.32
0.86
ΔF2
0
0
0.84
0.11
9.53
1.09
3.99
7.38
6.60
0.01
8.76
0.55
0.48
ΔF1
13
Performance analysis of SSA optimized fuzzy 1PD‑PI controller on AGC of renewable energy assisted…
Fig. 9 Power generations
responses in RTS a ΔPgArea−1 . b
ΔPgArea−2
(a) (b)
Tolerance
Band Tolerance
Band
QODSA: PID (Guha et al. 2016b) 4.48 5.53 4.59 0.0154 0.0078 0.00274 2.50E − 4 0 0 1.50E − 4
GWO: PID (Guha et al. 2016a) 3.53 6.01 4.51 0.0149 0.0069 0.00260 0 0 0 1.47E − 4
hFA-PS: PID (Sahu et al. 2015) 5.75 7.12 5.75 0.0134 0.0066 0.00220 0 0 0 1.17E − 4
BFOA: FOPID (Arya and Kumar 2017a) 3.10 4.52 3.81 0.0115 0.0046 0.0015 4.51E − 4 0 0 5.22E − 5
BFOA: FPI (Arya and Kumar 2017b) 2.62 5.00 4.25 0.0093 0.0029 0.00108 0 0 0 1.60E − 5
BFOA: FPID (Arya and Kumar 2017a,b) 2.75 3.23 2.66 0.0065 0.0019 0.00079 2.59E − 4 0 0 1.60E − 5
DSA: FOPI-POPD (Çelik 2021) 7.45 8.45 3.60 0.0029 0.0007 0.00029 8.11E − 5 6.08E − 5 2.71E − 5 2.94E − 6
ISCA: AFPID (Rajesh and Dash 2019) n/a n/a 1.15 0.0016 0.0001 4.81E − 5 3.25E − 4 0 0 3.02E − 7
CCSA: FPD-TID (Khokhar et al. 2021) 1.06 1.69 0.93 0.0013 0.0001 5.02E − 5 0 0 0 1.01E − 7
ICA: FPIDN (Arya et al. 2021b) 1.75 3.79 3.05 0.0047 0.0011 0.00043 0 0 0 6.01E − 6
ICA: C-IλDμN (Arya et al. 2021b) 3.19 1.92 1.44 0.0027 0.0004 0.00015 0 0 0 9.41E − 7
ICA: C-IλDμN (RES) (Arya et al. 2021b) 1.61 1.43 0.96 0.0020 0.0002 0.00009 0 0 0 3.68E − 7
SSA: F1PD-PI 0.23 0.13 0 0.0015 8.71E − 5 2.13E − 5 3.03E − 4 0 0 8.31E − 8
SSA: F1PD-PI (RES) 0.15 0 0 0.0014 4.32E − 5 1.19E − 5 0 0 0 2.94E − 8
13
E. Çelik
Fig. 11 Power generations
responses in MSHTS a
ΔPgArea−1 . b ΔPgArea−2
(a) (b)
Fig. 12 Responses of two-area
MSHTS under challenging envi-
ronment conditions a variable
PW ∕Psol . b ΔF1 , ΔF2 and ΔPtie .
c ΔPgArea−1 . d ΔPgArea−2
(a) (b)
(c) (d)
8 Analysis of governor dead band rate constraint (GRC) are employed in two-area RTS with-
and generation rate constraint out RES penetration. GDB causes the system to behave
oscillatory because the valve position due to the effect of
To reveal the capability of the F1PD-PI controller in GDB does not change during a number of speed variations
dealing with more realistic AGC systems, nonlinearity (Çelik 2021). To study the dynamics of GDB nonlinearity,
sources like governor dead band (GDB) and generation the governor TF of each area of RTS shown in Fig. 1a is
replaced by the TF stated below:
13
Performance analysis of SSA optimized fuzzy 1PD‑PI controller on AGC of renewable energy assisted…
Δu +δ 1
at time and SSA based optimization procedure described
Δt
ΔPg before is reinitiated to tune the controller parameters by
-δ s
considering a SLD of 0.01puMW in area-1 at t = 0. Optimal
controller parameters and the ISE value for tsim = 20 s are
Fig. 13 Open loop model of GRC presented in Table 5, where a comparison with the nominal
case (without GDB and GRC) is also provided. As we can
−(0.2∕𝜋)s + 0.8 see in Table 5, the value of J is increased comparing with
Gs (s) =
sTg + 1
, (18) the nominal case owing to the occurrence of GDB and GRC,
meaning that the frequency regulation is deteriorated.
The respective dynamic responses in comparison with
where Tg = 0.2 s. the nominal response are depicted in Fig. 14. The presence
Another important source of nonlinearity in IPSs comes of GDB, particularly GRC, in two-area RTS leads to larger
from generation rate constraint (GRC) that imposes limit on deviations in Ts ∕Us ∕Os than that of the nominal response.
the rate of change of the power generation due to the limita- Nonetheless, ΔF1 , ΔF2 and ΔPtie signals of the nonlinear
tions of thermal and mechanical actions as they require some RTS settle again to the desired 0 steady state value with
finite time to be completed. The inclusion of GRC enables some overshoot and no sustained oscillations. In vision of
the generated power to vary only at a certain rate, thus it is this, the proposed controller can be recognized capable to
a suitable means for a realistic AGC study. The open loop tackle the introduced nonlinearities satisfactorily without
model of GRC displayed in Fig. 13 is employed in each area loss of system stability. It can be also disclosed that the
of RTS with ±𝛿 being the limits of GRC. system performance improves as the limit of GRC is set to
GDB and GRC of ± 0.02 puMW/s, ± 0.0033 puMW/s higher value.
and ± 0.002 puMW/s are considered in two-area RTS one
Fig. 14 Responses of two-area RTS in presence of GDB and GRC nonlinearities a ΔF1 . b ΔF2 . c ΔPtie
13
E. Çelik
9 Robustness to system parameters considered here is more challenging since its duration and
variation and load disturbances magnitude change continuously and abruptly, which are
faced frequently in practice. As per the ΔF1 , ΔF2 and ΔPtie
Finally, the robustness of the proposed F1PD-PI controller signals in Fig. 16b, the F1PD-PI controller is still able to
to parameter variations and load disturbances is evaluated force the responses toward zero without unwanted oscil-
in this section. The aim of robustness test is to check how lations, affirming the robustness of F1PD-PI controller to
healthy the controller parameters are against a change in load disturbances.
operating condition and decide whether to reset them to Consequently, the robustness analysis concludes that
accommodate new conditions. Two-area RTS is chosen as the proposed controller with the gains tuned at nomi-
test system and the controller parameters remain the same nal conditions performs robustly with respect to param-
as in Table 2. Initially, the system parameters such as Tg , eter variation and load disturbance and its gains do not
Tt,Tr and Kr are assumed to deviate from their nominal val- require to be reset for the considered changes in working
ues (provided in Appendix) in the scale of + 20 to − 20%. condition.
The resulting ΔF1 , ΔF2 and ΔPtie responses of RTS under
nominal and varying parameters are displayed in Fig. 15.
As seen, parameter changes considered do affect the sys- 10 Stability analysis
tem responses inconsiderably as they are found to be very
close to the nominal response. While improving the dynamic responses of an AGC sys-
In the second exercise, the robustness property of tem, it is mandatory to have a sufficient degree of stability.
the proposed approach is analyzed when a SLD shown There are several approaches to make stability analysis
in Fig. 16a is applied to both areas of RTS. The SLD of a system and eigenanalysis is one of them. In order
to show whether the proposed controller is stable or not,
Fig. 16 Responses of two-area
RTS under load disturbance
(a) (b)
13
Performance analysis of SSA optimized fuzzy 1PD‑PI controller on AGC of renewable energy assisted…
the eigen values of RTS and MSHTS with SSA optimized 11 Computational complexity
F1PD-PI controller are obtained and provided in Table 6,
when the controller parameters are set as in Table 2, i.e. In this section, finally, some controllers utilized in simula-
those used in Sects. 7.2 and 7.3. It is obvious from Table 6 tions are compared in terms of computational complexity/
that all the poles of both systems with and without RES are cost and the number of tunable parameters they require.
located at the left-half of the s-plane. Therefore, it can be For this goal, two-area MSHTS is allowed to be controlled
concluded that SSA optimized F1PD-PI controller is able by each controller while the simulation time is tsim = 10 s
to control the studied systems stably. and the sampling time is h = 0.001 s. The runtime of the
simulation program is measured for each controller under
identical conditions and this information is considered as
an indicator to apprise the readership of computational
cost that the controllers demand. Table 7 shows the num-
ber of tunable parameters required for each controller
Table 6 Eigenvalues of the systems controlled by SSA optimized
F1PD-PI controller
and the mean runtime in seconds calculated over 30 runs
with each controller. As we can see, F1PD-PI controller
Two-area RTS Two-area MSHTS requires the tuning of 7 parameters, which is smaller than
Without RES With RES Without RES With RES that for AFPID, FPD-TID and C-I λD μN controllers, but
higher than that for PID, FOPID, FPI, FPID and FOPI-
− 19.2 − 100 ± 100i − 100 ± 100i − 16
FOPD controllers. Thus, we can remark that the tuning
− 2.20 − 20.2 − 18 − 16.3
of F1PD-PI controller parameters has moderate effort for
− 0.99 ± 11i − 2.27 − 18.4 − 1.89 ± 6.85i
engineering commissioning. On the other hand, Table 7
− 19.8 − 1.27 ± 12i − 1.29 ± 9.02i − 0.0382
divulges that the simpler the controller, the lower its
− 0.202 − 20.8 − 0.65 ± 8.91i − 0.0382
runtime or computational cost. It is also noticed that an
− 0.720 − 0.201 − 0.0382 − 1.24 ± 6.64i
important portion of the total runtime for a given FLC-
− 0.200 − 0.773 − 0.0382 − 1.61 ± 0.5i
based controller is allocated to FLC. Eventually, it can be
− 1.37 ± 10.9i − 0.200 − 2.20 ± 0.244i − 2.20 ± 0.57i
concluded that all the FLC-based controllers have more
− 0.757 − 0.89 ± 11.9i − 1.98 ± 0.556i − 1.84
or less the same runtime around 55 s, signifying that the
− 0.25 − 0.801 − 2.54 − 2.09 ± 0.63i
complexity of these controllers is similar and higher than
− 0.25 − 0.250 − 2.07 ± 0.625i − 0.25
other controllers.
− 0.250 − 100 ± 100i − 0.25
− 2 − 100 ± 100i − 2
− 3.33 − 3.33
12 Conclusions
− 3.33 − 3.33
− 0.667 − 0.667
In this paper, a new F1PD-PI controller is introduced for
− 0.667 − 0.667
AGC of two-area IPSs that are integrated with RES. The
− 0.667 − 0.667
proposed control scheme connects FLC, 1PD and PI con-
− 0.556 − 0.556
trollers in cascade. SSA is utilized to optimize the gains
− 0.556 − 0.556
13
E. Çelik
13
Performance analysis of SSA optimized fuzzy 1PD‑PI controller on AGC of renewable energy assisted…
Çelik E, Öztürk N, Arya Y, Ocak C (2021a) (1+PD)-PID cascade Kaur A, Jain S, Goel S (2020) Sandpiper optimization algorithm: a
controller design for performance betterment of load frequency novel approach for solving real-life engineering problems. Appl
control in diverse electric power systems. Neural Comput Appl Intell 50:582–619
33:15433–15456 Khokhar B, Dahlya S, Singh Parmar KP (2021) A novel hybrid fuzzy
Çelik E, Öztürk N, Arya Y (2021b) Advancement of the search process PD-TID controller for load frequency control of a standalone
of salp swarm algorithm for global optimization problems. Expert microgrid. Arab J Sci Eng 46:1053–1065
Syst Appl 182:115292 Khorami E, Babaei FM, Azadeh A (2021) Optimal diagnosis of
Dahiya P, Sharma V, Naresh R (2019) Optimal sliding mode control for COVID-19 based on convolutional neural network and red fox
frequency regulation in deregulated power systems with DFIG- optimization algorithm. Comput Intell Neurosci. https://doi.org/
based wind turbine and TCSC–SMES. Neural Comput Appl 10.1155/2021/4454507
31:3039–3056 Kler D, Kumar V, Rana KPS (2019) Optimal integral minus propor-
Debnath MK, Sinha S, Mallick RK (2017) Application of fuzzy-PIDF tional derivative controller design by evolutionary algorithm for
controller for automatic generation control using jaya algorithm. thermal-renewable energy-hybrid power systems. IET Renew
Int J Pure Appl Math 114(9):51–61 Power Gener 13(11):2000–2012
Eberhart RC, Kennedy J (1995) A new optimizer using particle swarm Majdar RS, Ghassemian H (2021) Spectral-spatial classification
theory. In: Proceedings of the sixth international symposium on fusion for hyperspectral images in the probabilistic framework
micro machine and human science, pp 39–43 via arithmetic optimization algorithm. Int J Image Data Fusion.
Ekinci S, Hekimoglu B (2018) Parameter optimization of power system https://doi.org/10.1080/19479832.2021.2001051
stabilizer via salp swarm algorithm. In: 5th International confer- Mirjalili S, Gandomi AH, Mirjalili SZ, Saremi S, Faris H, Mirjalili
ence on electrical and electronic engineering, pp 143–147 SM (2017) Salp swarm algorithm: a bio-inspired optimizer for
El-Fergany AA (2018) Extracting optimal parameters of PEM fuel cells engineering design problems. Adv Eng Softw 114:163–191
using salp swarm optimizer. Renew Energy 119:641–648 Mohammedzadeh A, Rathinasamy S (2020) Energy management in
Elkasem AHA, Khamies M, Magdy G, Taha IBM, Kamel S (2021) photovoltaic battery hybrid systems: a novel type-2 fuzzy con-
Frequency stability of AC/DC interconnected power systems with trol. Int J Hydrog Energy 45:20970–20982
wind energy using arithmetic optimization algorithm-based fuzzy- Mohammedzadeh A, Sabzalian MH, Zhang W (2020) An interval
PID controller. Sustainability 13:12095 type-3 fuzzy system and a new online fractional-order learn-
Goldberg DE, Holland JH (1988) Genetic algorithms and machine ing algorithm: theory and practice. IEEE Trans Fuzzy Syst
learning. Mach Learn 3:95–99 28(9):1940–1950
Guha D, Roy PK, Banerjee S (2016a) Load frequency control of inter- Mohanty B, Panda S, Hota PK (2014) Controller parameters tuning
connected power system using grey wolf optimization. Swarm of differential evolution algorithm and its application to load
Evol Comput 27:97–115 frequency control of multi-source power system. Int J Electr
Guha D, Roy PK, Banerjee S (2016b) Quasi-oppositional differential Power Energy Syst 54:77–85
search algorithm applied to load frequency control. Int J Eng Sci Mosavi A, Qasem SN, Shokri M, Band SS, Mohammedzadeh A
Technol 19(4):1635–1654 (2020) Fractional-order fuzzy control approach for photovoltaic/
Guha D, Roy PK, Banerjee S (2017a) Study of differential search algo- battery systems under unknown dynamics, variable irradiation
rithm based automatic generation control of an interconnected and temperature. Electronics 9(255):1–19
thermal-thermal system with governor dead-band. Appl Soft Naik A, Satapathy SC (2021) Past present future: a new human-
Comput 52:160–175 based algorithm for stochastic optimization. Soft Comput
Guha D, Roy P, Banerjee S (2017b) Quasi-oppositional symbiotic 25:12915–12976
organism search algorithm applied to load frequency control. Özdemir MT, Öztürk D, Eke İ, Çelik V, Lee KY (2015) Tuning of
Swarm Evol Comput 33:46–67 optimal classical and fractional order PID parameters for auto-
Guha D, Roy PK, Banerjee S (2018) Application of backtracking search matic generation control based on the bacterial swarm optimiza-
algorithm in load frequency control of multi-area interconnected tion. IFAC-Papers OnLine 48(30):501–506
power system. Ain Shams Eng J 9:257–276 Öztürk N, Çelik E (2012) Speed control of permanent magnet syn-
Guha D, Roy PK, Banerjee S (2021) Disturbance observer aided chronous motors using fuzzy controller based on genetic algo-
optimised fractional-order three-degree-of-freedom tilt-integral- rithms. Int J Electr Power Energy Syst 43(1):889–898
derivative controller for load frequency control of power systems. Padhy S, Panda S (2017) A hybrid stochastic fractal search and
IET Gener Transm Distrib 15(4):716–736 pattern search technique based cascade PI-PD controller for
Hasanien HM, El-Fergany A (2017) Symbiotic organisms search automatic generation control of multi-source power systems in
algorithm for automatic generation control of interconnected presence of plug in electric vehicles. CAAI Trans Intell Technol
power systems including wind farms. IET Gener Transm Distrib 2:12–25
11(7):1692–1700 Polap D (2020) An adaptive genetic algorithm as a supporting mecha-
Hayyolalam V, Pourhaji Kazem AA (2020) Black widow optimization nism for microscopy image analysis in a cascade of convolution
algorithm: a novel meta-heuristic approach for solving engineer- neural networks. Appl Soft Comput 97(B):106824
ing optimization problems. Eng Appl Artif Intell 87:103249 Polap D, Woźniak M (2021) Red fox optimization algorithm. Expert
Hota PK, Mohanty B (2016) Automatic generation control of multi Syst Appl 166:114107
source power generation under deregulated environment. Int J Rahman A, Saikia LC, Sinha N (2016) AGC of dish-Stirling solar ther-
Electr Power Energy Syst 75:205–214 mal integrated thermal system with biogeography based optimised
Hussien AG, Hassanien AE, Houssein EH (2017) Swarming behav- three degree of freedom PID controller. IET Renew Power Gener
iour of salps algorithm for predicting chemical compound 10(8):1161–1170
activities. In: IEEE international 8th conference on intelligent Rajesh KS, Dash SS (2019) Load frequency control of autonomous
computing and information systems, pp 315–320 power system using adaptive fuzzy based PID controller opti-
Ismael S, Aleem S, Abdelaziz A, Zobaa A (2018) Practical consid- mized on improved sine cosine algorithm. J Ambient Intell Hum
erations for optimal conductor reinforcement and hosting capac- Comput 10:2361–2373
ity enhancement in radial distribution systems. IEEE Access Rajesh KS, Dash SS, Rajagopal R (2019) Hybrid improved firefly-
6:27268–27277 pattern search optimized fuzzy aided PID controller for automatic
13
E. Çelik
generation control of power systems with multi-type generations. interconnected power system using jaya algorithm. Eng Appl
Swarm Evol Comput 44:200–211 Artif Intell 60(2):35–44
Ravikumar S, Kavitha D (2021) IOT based autonomous car driver Sivalingam R, Chinnamuthu S, Dash SS (2017) A hybrid stochastic
scheme based on ANFIS and black widow optimization. J Ambient fractal search and local unimodal sampling based multistage PDF
Intell Hum Comput. https://doi.org/10.1007/s12652-020-02725-1 plus (1+PI) controller for automatic generation control of power
Reddy YVK, Reddy MD (2018) Solving economic load dispatch systems. J Frankl Inst 354:4762–4783
problem with multiple fuels using teaching learning based opti- Tasnin W, Saikia LC (2018) Maiden application of an sine-cosine algo-
mization and salp swarm algorithm. J Intell Syst Theory Appl rithm optimised FO cascade controller in automatic generation
1(1):5–15 control of multi-area thermal system incorporating dish-stirling
Rosaline AD, Somarajan UK (2019) Structured H-infinity controller solar and geothermal power plants. IET Renew Power Gener
for an uncertain deregulated power system. IEEE Trans Ind Appl 12(5):585–597
55(1):892–906 Tasnin W, Saikia LC, Raju M (2018) Deregulated AGC of multi-area
Sabzalian MH, Mohammedzadeh A, Lin S, Zhang W (2019) Robust system incorporating dish-stirling solar thermal and geothermal
fuzzy control for fractional-order systems with estimated fraction- power plants using fractional order cascade controller. Int J Electr
order. Nonlinear Dyn 98:2375–2385 Power Energy Syst 101:60–74
Sahu BK, Pati S, Panda S (2014) Hybrid differential evolution particle Wolpert DH, Macready WG (1997) No free lunch theorems for opti-
swarm optimization optimised fuzzy proportional integral deriva- mization. IEEE Trans Evol Comput 1(1):67–82
tive controller for automatic generation control of interconnected Xi L, Zhou L, Liu L, Duan D, Xu Y, Yang L, Wang S (2020) A deep
power system. IET Gener Transm Distrib 8(11):1789–1800 reinforcement learning algorithm for the order optimization allo-
Sahu RK, Panda S, Padhan S (2015) A hybrid firefly algorithm and cation of total power in the interconnected power grids. CSEE J
pattern search technique for automatic generation control of multi Power Energy Syst 6(3):712–723
area power systems. Int J Electr Power Energy Syst 64:9–23 Yılmaz ZY, Bal G, Çelik E, Öztürk N, Güvenç U, Arya Y (2021)
Saikia LC, Mishra S, Sinha N, Nanda J (2011) Automatic generation A new objective function design for optimization of second-
control of a multi area hydrothermal system using reinforced ary controllers in load frequency control. J Fac Eng Archit Gaz
learning neural network controller. Int J Electr Power Energy Syst 36(4):2053–2067
33(4):1101–1108 Zhiheng W, Jianhua L (2021) Flamingo search algorithm: a new swarm
Satapathy P, Debnath MK, Mohanty MK (2018) Design of PD-PID intelligence optimization algorithm. IEEE Access 9:88564–88582
controller with double derivative filter for frequency regulation.
In: IEEE international conference on power electronics, intelligent Publisher's Note Springer Nature remains neutral with regard to
control and energy systems, pp 1142–1147 jurisdictional claims in published maps and institutional affiliations.
Sharma D, Mishra S (2019) Non-linear disturbance observer-based
improved frequency and tie-line power control of modern
interconnected power systems. IET Gener Transm Distrib
13(16):3564–3573
Singh SP, Prakash T, Singh VP, Babu MG (2017) Analytic hierar-
chy process based automatic generation control of multi-area
13