Electrical Power and Energy Systems 104 (2019) 603–614

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

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

Load frequency control strategy via fractional-order controller and reduced- T

order modeling
⁎
Sahaj Saxena
Electrical and Instrumentation Engineering Department, Thapar Institute of Engineering and Technology, Patiala 147004, India

A R T I C LE I N FO A B S T R A C T

Keywords: This paper proposes a simple approach to design fractional-order (FO) controller via internal model control
Fractional-order filter (IMC) technique for load frequency control (LFC) problem in power systems. The proposed scheme utilizes the
Internal model control concept of CRONE principle, model-order reduction and FO filter in IMC framework to derive a robust controller.
Load frequency control Initially, the scheme is applied to single-area power system and then extended to two-area interconnected
PID tuning
system. The turbines considered are non-reheated, reheated and hydro type; and physical constraints of turbine
Reduced model
and governor are also taken into account to validate the applicability in more realistic environment. Simulation
results show that it can bring improved disturbance rejection performance in nominal condition as well as in
presence of uncertainties and constraints in plant parameters.

1. Introduction using optimal, robust, adaptive and intelligent control perspectives

[5–7].
Power system control is one of the most challenging task in control These days FO control scheme has received great attention among
engineering because the total generated power should balance the total the control practitioners due to improved control performance espe-
load in presence of numerous electrical machines such as generating cially for the systems working in uncertain environment, and exact
units, protection devices, controller loops and power transmission lines modeling of complex systems [8,9]. The FO system and control schemes
that generally spread in large geographical areas. Essentially, there are generally better than their integer-order (IO) counterparts. As a
would be performance deterioration in the form of frequency fluctua- result, a few fractional-order PID (FO-PID) control methodologies have
tions, voltage instability, constant but unexpected load change, opera- also been introduced for LFC problem. The first FO-PID scheme was
tional limits, rotor angle instability, economy in operation, and physical presented by Alomoush [10] in which LFC has been considered as a
and environmental disturbances. Therefore, these discrepancies must constrained optimization problem for two-area power systems. The
be eliminated for satisfactory operation of power system. integral error criterion particularly ITAE was selected as an optimiza-
Among the various power system control strategies [1], LFC deals tion function to evaluate PID parameters. In [11], the stability
with the regulation of frequency fluctuations,i.e., frequency should re- boundary locus method was employed to search the stabilizing FO-PID
main nearly constant in all control areas. In short, the LFC adjusts the parameters of a hybrid single-area power systems. Later on, nature
load reference point against the variation of the load changes in order inspired evolutionary and soft computing schemes (like, the non-
to keep the system frequency and tie-line power as closed to the pre- dominated sorting genetic algorithm-II, firefly algorithm, imperialist
scribed values as possible. The main objectives of LFC are to: 1) competitive algorithm) are introduced to design optimal FO-PID con-
maintain zero steady state error for frequency and tie-line power de- trollers for multi-area power systems [12–14]. In these intelligent op-
viations, 2) reject sudden load disturbance, 3) attain optimal transient timization schemes, the multi-objective functions are framed using in-
behavior under prescribed overshoot, settling time and error tolerance, tegral square indices such as ITAE , ISE , ISDCO (integral of the
4) provide robust performance in presence of modeling uncertainties squared deviation in controller output) and other figure of merits to
and nonlinearities, 5) establish better security margin of system in sense tune PID parameters. Although the aforementioned LFC schemes have
of stable frequency regulation, and less computing power [2–4]. Thus, shown their effectiveness and dominance over classical approach, there
LFC can be treated as an objective optimization and robust control are deficiencies due to heavy computational burden, premature con-
problem. In view of this various LFC strategies have been developed vergence during optimization process and sluggish disturbance

⁎
Corresponding author.
E-mail address: [email protected].

https://doi.org/10.1016/j.ijepes.2018.07.005
Received 27 November 2017; Received in revised form 23 May 2018; Accepted 5 July 2018
Available online 26 July 2018
0142-0615/ © 2018 Elsevier Ltd. All rights reserved.
S. Saxena Electrical Power and Energy Systems 104 (2019) 603–614

attenuation.
IMC technique is a control strategy that has been successfully used
for few decades [15–20]. It is observed that simplicity, robustness, sub-
optimality and wide area applicability are some features that have
popularized IMC among control scientists and practitioners. After the
introduction of IMC scheme in FO systems and control for last
2–3 years, the FO-PID got a new way for its synthesis and tuning (See
[21] and the references therein). Moreover, in literature on one hand, Fig. 1. Block diagram of single-area power system.
CRONE (abbreviation of “Commande Robuste d’Ordre Non Entier”
which means “non integer order robust control”) principle is highly
d 1 1
popular for designing FO controller [22] and on the other hand IMC is ΔPG (t ) = − ΔPG (t ) + ΔXG (t )
dt TT TT (2)
famous control scheme for designing IO controller. Fortunately, the
pioneer work of Maâmar & Rachid [23] bridges both control schemes to d 1 1 1
ΔXG (t ) = − Δf (t )− ΔXG (t ) + u (t )
build FO controller. Through this method, the controller acquires the dt RTG TG TG (3)
FO-PID form via IMC methodology and tuning scheme is evolved using
CRONE principle. In terms of transfer function model, the governor is
Motivated by the celebrated work of [23], the FO controller design 1
PG (s ) = ,
scheme is proposed in this paper which make use of reduced-order TG s + 1 (4)
modeling to acquire the dominant features of the higher-order plant.
the non-reheated turbine is
Also to the best of author’s knowledge, such LFC scheme is missing in
power control research. Therefore in this paper, a FO-PID based on IMC 1
PT (s ) =
and CRONE schemes is proposed for frequency regulation of single and TT s + 1 (5)
multi-area power systems. The proposed design requires only frequency
and the load and machine is
domain specifications particularly phase margin and gain crossover
frequency as a prerequisite. The main advantages of this work is that: KP
PP (s ) =
(i) the proposed scheme exhibits robustness as the controller para- TP s + 1 (6)
meters, tuned with the help of gain and phase margin specifications,
Now using (4)–(6), the whole plant can be written as
works well when parametric uncertainties are present in power plant,
(ii) the controller is optimal as it minimizes the integral error indices, Δf (s ) PG (s ) PT (s ) PP (s ) KP
= P (s ) = =
and (iii) for executing LFC, substantial improvements are observed in u (s ) 1 + PG (s ) PT (s ) PP (s )/ R a3 s 3 + a2 s 2 + a1 s + a0 (7)
the performance using the proposed method in comparison to the re-
where
cently developed methods.
a3 = TG TT TP , a2 = TG TT + TP TT + TP TG,
KP
2. Description of LFC model a1 = TG + TT + TP , a0 = 1 + R (8)

Electric power systems are complex non-linear dynamical systems As LFC is a disturbance rejection problem, so our aim is to find a control
consisting of numerous generators and loads. However for modeling law: u (s ) = −C (s )Δf (s ) such that lim t →∞Δf (t ) = 0, for all ΔPd .
purpose, all the generators are lumped into single equivalent generator Remark 1. Nonlinearities (backlash and wind-up problems) in the
and likewise for loads. Since, power systems are exposed to small load speed control are normally neglected except for rate limiter and the
changes, the system can be adequately represented by its linear model limits on valve position. All damping torque to prime-mover, generator
[2,33]. The basic power system notations are presented in Table 1. and the HVDC system are also assumed to be negligible.

2.1. Single-area power system

3. Design tools
The block diagram of a single-area power system supplying power
to single service area through single generator is shown in Fig. 1. The In this section, we put forward few prerequisites to present our
dynamics of this plant which consists of governor, non-reheated tur- proposed work. Throughout the paper, the real and natural numbers are
bine, and load and machine can be written as symbolized by  and  , respectively. Further + denotes the real po-
sitive numbers. For any signal x (t ) , its Laplace transform is denoted by
d 1 K X (s ) . A stable continuous-time, linear time-invariant finite dimensional
Δf (t ) = − Δf (t ) + P (ΔPG (t )−ΔPd (t ))
dt TP TP (1) single-input single-output system described by a rational proper
transfer function G (s ) is considered whose order is denoted with ρ (G ) .
Table 1 A stable system G (s ) is a minimum-phase system if the zeros of the
Nomenclature of Basic Power Systems Terms. system are stable, i.e., roots of the numerator polynomial are in left-half
Δf (t ) Incremental change in frequency (Hz) of complex s-plane.
ΔPd (t ) Load disturbance (p.u. MW)
ΔPG (t ) Incremental change in generator output (p.u. MW) 3.1. Fractional-order system
ΔXG (t ) Incremental change in governor valve position (p.u. MW)
KP Electric system gain
Here, a brief exposition of FO operators and their properties are
TP Load and machine time constant (s)
TT Non-reheated turbine time constant (s) given. Fractional calculus is actually the generalization of IO integra-
Tr Reheated turbine time constant (s) tion and differentiation to any arbitrary real number. It is an old con-
Tw Hydro turbine time constant (s) cept in mathematics however in control engineering it has witnessed a
TG Governor time constant (s) remarkable progress from last one decade after the introduction of FO
c Percentage of the power generated in the reheat portion
controllers [8,9,24]. Now, we introduce the notion of generalized FO
R Speed regulation due to governor action (Hz/p.u. MW)
Bi Frequency bias (p.u. MW/Hz) operators. The continuous integro-differential operator of order α ∈ 
(often denoted by x Dtα , where x and t denote the limits of the operation)

604
S. Saxena Electrical Power and Energy Systems 104 (2019) 603–614

1
is defined as whose magnitude A (ω) = ωp
in dB is given by

⎧ dα /dt α
α > 0, M (ω) = −20plog(ω) (10)
α
x Dt = 1 α = 0,
⎨ t and the phase is
∫ (dτ )−α α < 0.
⎩ x
1 pπ
ϕ (ω) = arg ⎡ p j−p ⎤ = −
In this paper, we define the FO system using the differential equation of ⎣ω ⎦ 2 (11)
the form
From (10) and (11), it is clear that the magnitude of a FO integrator in
m l
μ the frequency domain drops at a rate of 20p dB/dec and its phase is
∑ ai Dt i y (t ) = ∑ bi Dtνi u (t )
−pπ/2 throughout the domain. Whereas the IO integrator yields fixed
i=1 i=1
drop at a rate of −20 dB/dec in magnitude and −π/2 in phase response.
where μm > μm − 1 > … > μ1 > 0 and νl > νl − 1 > … > ν1 > 0 are strictly This may hinders the stability and robustness of the closed-loop system.
positive real numbers and (ai , bi ) ∈ 2 . On interpreting this equation in Thus the FO integrator introduces new degrees of flexibility that sim-
the popular Caputo sense (see Definition 1) and applying the Laplace plifies the design of high performance controller.
transform for zero initial condition, the transfer function can be ob-
tained as
3.2. IMC technique [15,16]
l
∑ bi s νi IMC is a model predictive based control technique which utilizes an
i=1
G (s ) = m additional plant model to predict the output and rectify the error be-
∑ a i s μi tween desired and actual output. Fig. 2 shows the structure of the IMC
∼
i=1 controller in which P (s ) is a plant and its model is P (s ) . The IMC
∼
controller Q (s ) is composed of inverse of P (s ) cascaded with IMC filter
where am ≠ 0 and μm > νl is assumed so that G (s ) is strictly proper.1
F (s ) . The output of the plant is yo for input yi and disturbance is d. The
Definition 1. The Caputo definition of FO derivative of order α of a output is formulated as
continuous function f : + →  is defined as: ∼−1 ∼−1
P (s ) P (s ) F (s ) 1−F (s ) P (s ) P (s )
1 t f (n) (τ ) Yo (s ) = ∼−1 Yi (s ) + ∼−1 D (s )
α
aDt f (t ) =
Γ(n−α )
∫a (t −τ )α − n + 1
dτ , ∀ n−1 < α < n 1 + F (s ) P (s )ΔP (s ) 1 + F (s ) P (s )ΔP (s ) (12)
∼
where ΔP (s ) = P (s )−P (s ) is the plant-model mismatch. If the model of
where f (n) (t ) is the nth derivative of f (t ) with respect to t , n ∈  and ∼
plant is an exact representation of the real plant, i.e., P (s ) = P (s ) , and
Γ(•) is Gamma function.2 The Laplace transform of this derivative is
F (s ) = 1 then from (12) we get Yo (s ) = Yi (s ) for all D (s ) . Thus, perfect
given by
tracking and disturbance rejection can be achieved in this ideal case.
n
t However, such control strategy cannot be directly implemented in the
∫0 e−st aDtα f (t )dt = s αF (s )− ∑ s α − k − 1f (k ) (0) case where the model of the plant is strictly proper or non-minimum
k=0
phase.
dn
Remark 2. When α = n ∈  then aDtα f (t ) = dtn
f (t ) . Generally in real time scenario, plant-model mismatching is present
and the uncertainty in the plant increases with frequency. At this stage,
Remark 3. In literature, various definitions of fractional calculus are
robustness against plant-model mismatch can be improved by means of
presented but Caputo definition is highly popular particularly in ∼
filter F (s ) . This filter is designed to add poles to P (s ) and is chosen such
engineering. In Caputo definition, the initial conditions are of integer-
that the closed-loop system retains its asymptotic tracking properties
order (i.e., derivative of constant is zero) which make them easier to
(i.e., zero offset at steady state for asymptotically constant inputs and
interpret because the IO derivatives of involved variables have well-
step type disturbances). It is usually a low-pass filter of the type
established physical meanings and can be easily obtained by
experimental means. 1
F (s ) =
(1 + λs )n (13)
Like the IO system where the building blocks of system are in-
tegrators and differentiators, the FO system also constitutes FO in- where λ is the filter parameter that fixes the bandwidth of the closed-
∼
tegrators and differentiators as their basic elements. loop system and ρ (F ) = n is chosen according to the order of P (s ) .
Using this approach, Q (s ) parameters are linked in a unique straight-
Definition 2. The transfer function of FO integrator is defined as ∼
forward manner to P (s ) parameters. In (13), λ is now the only para-
1 meter to be tuned to influence the speed of response of the closed-loop
G (s ) = , p ∈ (0, 1).
sp (9) system. This λ is also detuned to maintain the robustness in presence of
plant-model uncertainties. Therefore a trade-off is imposed for sacrifi-
cing performance to attain robustness which is inherent to any control
For p = 1, G (s ) is a simple pure integrator. As p tends towards 0, system.
the effect of integration operation eliminates because s 0 = 1. The IMC structure is complex for practical implementation, and it is
Remark 4. In control theory, the addition of pure integrator retards the usually rearranged into its equivalent conventional feedback control
∼
speed of response but here the FO integrator relaxes this constraint. structure as shown in Fig. 3. The relation between Q (s ) and P (s ) of
Fig. 2 and C (s ) of Fig. 3 is given by
Putting s = jω in (9), the spectral transfer function is obtained as
Q (s )
1 C (s ) = ∼
G (jω) = 1−Q (s ) P (s ) (14)
pπ pπ
ω p ⎡cos( 2 ) + jsin( 2 ) ⎤
⎣ ⎦
Remark 5. The main advantage of the IMC technique is the stability of
the closed-loop system. As the IMC structure is internally stable (i.e.,
1
A strictly proper transfer function satisfies G (s ) → 0 as s → ∞. P (s ) and Q (s ) are stable), therefore its equivalent conventional
2 ∞
The Gamma function is defined by Γ(p) = ∫0 t p − 1e−t dt , R (p) > 0 feedback control structure is also stable.

605
S. Saxena Electrical Power and Energy Systems 104 (2019) 603–614

ωgc = τ −1/ γ (15)

3
is a gain crossover frequency. Thus in open-loop, it has constant slope
γπ
of −20γ dB/s and phase curve is horizontal line at − 2 . The Nyquist
γπ
curve is a straight line through the origin with argument − 2 . Let us
now consider a unity feedback system with Po (s ) ≔ P (s ) C (s ) = L (s )
inserted in the forward path as shown in Fig. 3 for which the closed-
loop transfer function can be written as

Po (s ) 1
Fig. 2. Schematic diagram of the IMC. T (s ) ≔ =
1 + Po (s ) 1 + τs γ (16)
π
Substituting s = jω and j = ej2, the spectral transfer function of T (s ) is
1
T (jω) =
( 1 + τωγ cos
γπ
2 ) + jτω sin
γ γπ
2 (17)

The amplitude and phase of T (jω) can be evaluated as

Fig. 3. Schematic diagram of the conventional feedback control structure. 1
|T (jω)|dB = 20log γπ
1 + 2τωγ cos 2
+ (τωγ )2 (18)
Remark 6. At many instances, the actual plant exhibits non-minimum
phase characteristics (RHP zeros or delay). In such cases, IMC and
procedure is applied by segregating the plant into its minimum phase
γ γπ
and non-minimum phase elements, i.e., ⎡ τω sin 2 ⎤
arg [T (jω)] = −arctan ⎢
1 + τωγ cos 2 ⎥
γπ
P (s ) = P−(s ) P+(s ) ⎣ ⎦ (19)

where P−(s ) is minimum phase part and P+(s ) is non-minimum phase From (18) and (19), it is obvious that at ω = 0, T (jω) = 0 and
part. And the minimum phase element is used to design the controller. arg[T (jω)] = 0 . The asymptotic behavior of T (jω) at ω → ∞ is
γπ
T (jω) ≈ −20γ log(τω) and arg[T (jω)] ≈ − 2 . See appendix A for eva-
luation. Hence, the frequency response resembles with the low-pass
3.3. Reduced-order modeling filter.
Definition 4. The frequency at which the maximum value of the
Reduced-order modeling is a tool which simplifies the high-order
spectral transfer function (known as resonance peak Mr ) is attained is
complex real plant into its adequate low-order model such that the
called resonance frequency ωr .
important features of the original system are retained in the reduced
model. This tool reduces the computational effort for analyzing the The resonance peak Mr at resonance frequency ωr is given by the
complex dynamics of the real plant by removing any redundant in- formula
formation. Thus, it helps in designing and developing the controller 1
with less effort and at cheaper cost [25–28]. In control system, it pur- 1 1 γπ τ
Mr = γπ , ωr = cos
sues the following definition. sin 2 τ 2

Definition 3. Let B (s ): u ↦ y be the original system with ρ (B ) = v , Refer appendix B for proof. The natural frequency and damping ratio
then the reduced-order modeling is a technique to find a reduced-order are given by
∼ ∼
model B (s ): u ↦ ∼ y with ρ (B ) = w so that w < v and for the same input
u (t ) ∈ L2 , ∼
y (t ) ≈ y (t ) . 1 1
ωp = sinπ ⎛⎜1− ⎞⎟
τ ⎝ γ⎠
The reduced-order modeling scheme must satisfies the following
properties: (i) it targets to minimize the infinity norm approximation and
∼ ∼
error defined by E (s ) = || B (s )−B (s ) ||∞ = supω ⩾ 0 |B (ω)−B (ω)|,
for all ω ∈  ; (ii) system properties, such as stability, are preserved; π
ζ = −cos
and (iii) the procedure is computationally efficient. In this paper, we γ
follow the Routh approximation based reduced-order modeling method
respectively. See appendix C for derivation. Also, the phase margin4 is
[29]. In this method, the Routh table for the original plant is developed,
and then the reduced model is constructed in such a way that the γ
ϕ = π ⎛1− ⎞
coefficients of its Routh table matches up to a given order with that of ⎝ 2⎠ (20)
the original plant. The detailed procedure is provided in the later sec-
tion. Now it is clear that when the system parameter τ varies while keeping
the γ fixed, only the rise time (i.e., natural frequency) and speed of the
response (i.e., resonance frequency) varies while ensuring the constant
3.4. CRONE principle resonance peak and phase margin and thus correspondingly a constant
damping ratio and overshoot in time domain. Therefore, we can shape
The CRONE principle relies on the concept of robustness in order to the output response close to the desired response by varying the re-
maintain time and frequency domain performance measures (iso- ference tuning parameters (γ , ωgc ).
damping property, stability margin) using complex fractional integra-
tion [30,31]. It actually includes the concept of Bode’s ideal transfer
3
function. In CRONE principle, the open-loop transfer function L (s ) is A gain crossover frequency, ωgc ∈ [0, ∞) , for L (s ) , is a frequency at which
1
the transfer function of a FO integrator as L (s ) = τs γ , γ ∈ (1, 2) and |L (jωgc )| = 1.
4
τ > 0 , where The phase margin for T (s ) is defined by ϕ = arg[L (jωgc )] + π .

606
S. Saxena Electrical Power and Energy Systems 104 (2019) 603–614

4. Proposed FO-PID controller design scheme controller is

∼−1
We now present our proposed controller design scheme. G (s ) F (s )
C (s ) =
1−F (s )
4.1. Controller framework The closed-loop transfer function T (s ) ≔
C (s ) G (s )
, now becomes
1 + C (s ) G (s )
∼−1
Let us consider an all-pole plant to be controlled as G (s ) G (s ) F (s )
T (s ) = ∼−1
K 1 + (G (s ) G (s )−1) F (s ) (27)
P (s ) = , K>0
D (s ) To evaluate T (s ) , two cases are considered.
∼
where D (s ) = dn s n + dn − 1 s n − 1 + …+d1 s + d 0 , and {di}i = 1,2, … , n ∈  such Case 1: When G (s ) = G (s ) , i.e., the case of perfect control, then from
that P (s ) is stable. The general form of second-order reduced-model can (27) we have
be written as T (s ) = F (s ) (28)
∼ k ∼
P (s ) = ∼ ∼ ∼, k>0 Case 2: When G (s ) ≈ G (s ) , i.e., the case of good control (Lemma 1),
2
d2 s + d1 s + d 0 (21) then from (27) we get
∼
where {di }i = 1,2,3 ∈ +. To design IMC based controller, a FO filter of T (s ) ≈ F (s ) (29)
form
Eqs. (28) and (29) imply that the closed-loop transfer function of the
∼ 1 system to be controlled using IMC technique is exactly or approximately
F (s ) = , 0 < ψ < 1; λ ⩾ 0
1 + λs ψ + 1 (22) the transfer function of the filter used. □
is chosen in place of filter of form (13). The IMC controller is de- ∼
If F (s ) described in (22) is used to design IMC controller, then from
termined as ∼
Theorem 1, we get T (s ) = F (s ) . Now, this FO filter can be treated as a
∼ ∼ ∼ reference closed-loop model as per the CRONE principle. On comparing
d s 2 + d1 s + d 0
Q (s ) = 2 . (22) and (16), we get τ = λ and γ = ψ + 1 and substituting these values
k (1 + λs ψ + 1)
in (15) and (20), we get
Using (14), the equivalent conventional feedback controller is
π −ϕm ψ+1
∼ ∼ ∼ ψ= −1; λ = 1/ ωgc
1 d s 2 + d1 s + d 0 π /2 (30)
C (s ) = ψ 2
s kλs
Thus, with the help of desired ϕm and ωgc the proposed controller can be
which can be further rearranged as tuned.
1 1 Remark 8. In general practice, the bandwidth of the overall control
C (s ) = kc ⎛1 +
⎜ + τd s ⎞ ⎛ ψ ⎞
⎟

⎝ τi s ⎠⎝ s ⎠ (23) system is considered greater than that of the plant whereas it is opposite
∼ ∼ ∼ ∼ ∼ in the case where plant has non-minimum phase characteristics or input
where kc = d1/(kλ ), τi = d1/ d 0, τd = d2/ d1.
constraints.
Remark 7. Eq. (23) represents C (s ) as a FO-PID controller which is a
combination of conventional IO type PID controller and FO integrator
4.3. Rejection capacity and stability of control system
1/ s ψ .
.
4.2. Tuning of controller
Corollary 1. Under the assumption that there is almost zero plant
model mismatching, the disturbance response for step type input is zero
In the proposed controller, only two parameters (λ, ψ ) are unknown.
when FO filter of form (22) is used.
To determine these tuning parameters, we present few immediate re-
sults derived from discussions in Section 3. Proof. The immediate result of Theorem 1 gives T (s ) = F (s ) . The
transfer function from disturbance to output is the sensitivity function
Lemma 1. A good IMC based control is obtained if the reduced-order model
∼ S (s ) and is given by
P (jω) approximates the magnitude of real plant P (jω) within achievable
bandwidth. Y (s ) λs ψ + 1
= S (s ) ≔ 1−T (s ) =
D (s ) 1 + λs ψ + 1
Proof. From (12), we can say that a good control (i.e.,
1
Yo (s ) ≈ Yi (s ) for all D (s ) ) is obtained when Therefore, for a unit step input D (s ) = s , the output Y (s ) is
∼−1 1 λs ψ + 1
|P (jω)ΔP (jω)| ≈ 0 (24) Y (s ) =
s 1 + λs ψ + 1
and
∼−1 On applying final value theorem of signal processing theory, we get
|P (jω) P (jω)| ≈ 1 (25)
limt →∞y (t ) = lims → 0 sY (s ) = 0
within achievable bandwidth. Therefore, (24) and (25) imply that
□
∼
|P (jω)| ≈ |P (jω)| (26)
Thus it is clear that the proposed controller has capability to reject
Theorem 1. For a minimum-phase system, the closed-loop transfer function the disturbance. Now we examine the stability of the closed-loop
derived using IMC technique is exactly or approximately the transfer system. The stability analysis methodology for FO system is different
function of the IMC filter used. from that of the IO system. Here, the stability is defined using extended
Matignon’s Theorem as stated below.
Proof. Consider a minimum-phase plant G (s ) and its approximated
∼
model as G (s ) . The IMC controller with filter F (s ) can be obtained as Theorem 2. If pi ’s are the roots of a characteristic equation
∼−1 m
Q (s ) = G (s ) F (s ) . The corresponding conventional feedback Δ(s ) = 1 + ∑i = 1 ai s μi , then the system is bounded-input, bounded-output

607
S. Saxena Electrical Power and Energy Systems 104 (2019) 603–614

stable if and only if called area control error (ACE) is utilized in controller as a feedback
μπ variable. For two-area system, ACEi , i = 1, 2 are defined as
|arg (pi )| >
2 (31) ACE1 (s ) = ΔPT (s ) + B1 Δf1 (s )
provided
ACE2 (s ) = −ΔPT (s ) + B2 Δf2 (s )
0<μ<2 (32)
T
where ΔPT = 12 s
(Δf1 −Δf2 ) is the tie-line power deviation (p.u.MW) from
Proof. The proof could be performed in a way as given in [32].
the scheduled tie-line exchange power, and T12 is tie-line synchronizing
From Theorem 1, the denominator of F (s ) in (22) acts as a char- coefficient (p.u.MW/radian) between area 1 and 2. For two-area power
acteristic equation: system, decentralized controllers C1 (s ) and C2 (s ) can be synthesized
y assuming ΔPT (s ) = 0 which implies T12 = 0 . In this case, the transfer
Δ(s ) = 1 + λs x (33) function of ith control area is given by
y y 1
where = 1 + ψ; x , y ∈  and 1 <
x x
< 2 . Let σ = sx then (33) becomes PG, i (s ) PT , i (s ) PP, i (s )
Pi (s ) = Bi
Δ(σ ) = 1 + τs y whose roots are 1 + PG, i (s ) PT , i (s ) PP, i (s )/Ri (35)
1 j 1 +y2l π where PG, i (s ), PT , i (s ), PP, i (s ) are the transfer functions of governor, the
{pl }l = 0,1, … , (y − 1) = e
τ (34) turbine and the generator for ith area. Lastly, our aim is to develop
According to Theorem 2, for stability of (33), it is required to prove that decentralized regulation law which takes the form:
1 1 + 2l π
|arg(pi ) = τ e j y π | > 2x . From (32), it can be said that 0 < μ < 2 or ui (s ) = −Ci (s ) ACEi (s ) so that limt →∞ACEi (t ) = 0, for all ΔPd, i .
1 y 1 1
0< x
< 2 . Since x < 2 or y < 2x ⇒ y
> 2x
. Therefore, it is evident that
(1 + 2l) π π
5.2. Reduced model identification
y
> 2x . Thus, the closed-loop FO system is stable.
Power systems are highly large interconnected network of power
5. FO-PID controller design for LFC apparatus. Even a single-area power system plant transfer function (7)
containing single generator is of third-order. Therefore for fast and cost
We now apply the proposed scheme to solve LFC problem. efficient planning, operations and control, reduced-order models of
power systems are necessary. Here, Routh approximation model re-
5.1. Two-area power system duction scheme [29] is applied to obtain second-order model of power
system.
∼
The LFC design can be extended to multi-area interconnected power For a single-area plant (7), let the reduced model is P (s ) with
∼
systems. Without the loss of generality, the LFC problem for a two-area ρ (P ) = 2 . To apply Routh approximation method [29], we first re-
power system is presented in this paper. Fig. 4(a) depicts the simplified ciprocate P (s ) using relation
diagram of this two-area system and schematic diagram of ith -area is
1 1
shown in Fig. 4(b) [34]. In multi-area system, not only the frequency P ̂(s ) = P⎛ ⎞
s ⎝s⎠ (36)
deviation but also the tie-line power must return to its scheduled value
during load fluctuations in any area. Therefore, a composite measure, which gives

KP s 2
P ̂(s ) =
a0 s3 + a1 s 2 + a2 s + a3 (37)

Now, expand P ̂(s ) in the following canonical form:

P (̂ s ) = β1 E1 (s ) + β2 E1 (s ) E2 (s ) + β3 E1 (s ) E2 (s ) E3 (s ) (38)

where {βi}i = 1,2,3 are constants, E1 (s ) = 1 + α1 s, E2 (s ) and E3 (s ) are

functions of α2 and α3. Since we are interested in calculating the reduced
model, therefore as described in [29] the second-order reduced model is
given by

∼̂ α2 β1 s + β2
P (s ) =
α2 α1 s 2 + α2 s + 1 (39)

where α1,2 can be calculated with the help of α -table as shown in Table 2
and β1,2 are given by β1 = KP / a1 and β2 = 0 . On substituting the values
of α1,2 and β1,2 in (39) and further using relation (36), we get the re-
∼
duced model P (s ) as:
∼ a1 KP
P (s ) =
(a1 a2−a0 a3) s 2 + a12 s + a0 a1 (40)

Table 2
α−table.

a0 a2

a0 a1 a3
α1 =
a1
a12 a1 a2 − a0 a3
Fig. 4. (a) Block diagram of two-area power system and (b) the functional block α2 = a1
a1 a2 − a0 a3
diagram of ith control area in LFC framework.

608
S. Saxena Electrical Power and Energy Systems 104 (2019) 603–614

Remark 9. In case of multi-area power system, we know that only Bi is 3

an additional variable which is to be multiplied with each control area Original
Reduced
plant model (refer (35)). Therefore, same reduced model multiplied by 2.5
Bi is obtained for each control area.
2

Amplitude
5.3. Controller synthesis
1.5
For single-area power system, the model in (40) resembles with
(21), therefore one can write 1
∼ ∼ ∼
k = a1 KP , d2 = a1 a2−a0 a3, d1 = a12 , d 0 = a0 a1 (41)
0.5
On substituting these values of (41) in (23), the proposed FO-PID
controller can be obtained as 0
0 1 2 3 4
a a 1 a a −a a 1 Time (seconds)
C (s ) = 1 ⎛⎜1 + ⎛ 0 ⎞ + ⎜⎛ 1 2 2 0 3 ⎟⎞ s ⎞⎟ ⎛ ψ ⎞
⎜ ⎟

λKP ⎝ a
⎝ ⎠
1 s ⎝ a1 ⎠ ⎠ ⎝ s ⎠ (42)
The same procedure is applied for multi-area power system. 0

Magnitude (dB)
Therefore, without the loss of generality, for ith control area, the de-
centralized controller Ci (s ) is given by í50

a1, i ⎛ a 0, i ⎞ 1 a1, i a2, i−a 0, i a3, i ⎞ ⎞ 1

Ci (s ) = 1 + ⎜⎛ + ⎜⎛ ⎟ s⎟ ⎛ ⎞ í100
λBi KP, i ⎜
⎟

⎝ a1, i ⎠ s ⎝ a1,2 i ψi
⎠ ⎠⎝ s ⎠ (43)
⎝
í150 Original
0 Reduced
6. Simulation tests Phase (deg)
í0
For executing the proposed scheme for single and two-area config-
uration, simulations are carried out using Intel® CORE™ i7 processor í10
through MATLAB® and Simulink® (using FOMCON toolbox available
athttp://fomcon.net/fomcon-toolbox/download/). The fractional deri- í20 í1 0 1 2 3
10 10 10 10 10
vative has been implemented by the Oustaloup recursive filter ap-
Frequency (rad/s)
proximation choosing a frequency range of [10−3, 103] rad/s and order of
filter N = 5. See appendix D for more information. The nominal para-
meters of power system plant are taken as [35]: Fig. 5. (a) Step and (b) frequency responses of original and reduced-order
KP = 120, TP = 20, TT = 0.3, TG = 0.08, R = 2.4 (44) model.

6.1. LFC design for single-area system KP = [60, 170], TP = [10, 40], TG = [0.04, 0.1], R = [1.2, 3.6] (48)

Using this controller, disturbance rejection performance is obtained for

On substituting values of (44) in (8), the original plant in (7) is
nominal plant and perturbed plants (lower and upper bounds). The
given by
main advantage of the proposed controller is that the controller para-
250 meters need not to be changed even though there exist variations in the
P (s ) =
s 3 + 15.88s 2 + 42.46s + 106.2 (45) system parameters. To examine the efficiency of the proposed scheme,
and its reduced-model from (40) becomes the frequency deviation responses of the power system using the pro-
posed scheme is compared with the schemes recently developed by Tan
∼ 18.38
P (s ) = [35], and Anwar and Pan [36] as shown in Fig. 6. It is observed that the
s 2 + 3.173s + 7.94 (46)
proposed controller nullifies the change in frequency rapidly with least
The time and frequency domain responses of the original plant and the variations in its magnitude when compared with the LFC schemes de-
reduced model as shown in Fig. 5 confirms the resemblance of reduced signed by Tan [35] and Anwar and Pan [36]. Thus it can be claimed
model with the original one. that proposed scheme gives better disturbance rejection performance
In order to achieve good disturbance rejection performance, the with least settling time and overshoot for nominal as well as upper and
bandwidth of closed-loop control system should be considerably larger lower bounds of perturbed system.
than the plant to be controlled. Keeping this fact in mind, we select To measure the optimality of the proposed scheme, various per-
ωgc = 15 rad/s for the closed-loop control system whereas for plant in formance measures defined in the form of integral error criterion
∞ ∞ ∞
closed-loop configuration without controller, ωgc = 6.52 rad/s. Further, (ISE = ∫0 Δf (t )2dt ; IAE = ∫0 |Δf (t )| dt; ITAE = ∫0 t |Δf (t )| dt ; IE )
ϕm = π/3 is generally selected as a standard phase margin for tuning the ∞
= ∫0 Δf (t ) dt
controller. Now on applying the proposed method, the FO-PID con-
are calculated in Table 3. Indirectly, these performance measures de-
troller is obtained as
note several characteristics like settling time, overshoot, speed of re-
2.5026 1 sponse, disturbance rejection, etc. They can be treated as objective
C (s ) = 6.2926 ⎛1 + + 0.3146s ⎞ ⎛ 0.333 ⎞
⎝ s ⎠⎝ s ⎠ (47) function to investigate optimality of the controller. From Table 3, it is
clear that the values of these objective functions are least for the pro-
To show the performance of the proposed controller, a step load
posed scheme in comparison to other schemes. Thus, the LFC system
ΔPD = 0.01 p.u. is applied at t = 1 s, and ± 50% uncertainty is also
using proposed scheme is optimal in nature.
added in all the parameters of the power plants to observe the ro-
bustness of the controller, i.e.,

609
S. Saxena Electrical Power and Energy Systems 104 (2019) 603–614

í3
x 10 parameters of each area are same as given in (44). Therefore, the open-
2 loop plant model (P1 (s ) and P2 (s ) ) of both the areas are represented by
(45) which require B1,2 as a multiplier. For LFC design,
0 B1 = B2 = 0.425, T12 = 0.3770 and T21 = 0.4398 are taken. With the same
controller tuning settings as given in single-area LFC design, we keep
í2
ωgc = 15 rad/s and ϕm = π/3. Using these controller setting values and
í4 (43), the proposed decentralized controller Ci (s ), i = 1, 2 for each
Δ f (Hz)

control area is obtained as

í
2.5026 1
í Ci (s ) = 14.8061 ⎛1 + + 0.3146s ⎞ ⎛ 0.333 ⎞
⎝ s ⎠⎝ s ⎠ (49)
í10 Proposed To investigate the efficiency and robustness of the controller, the
Tan
parameters of area 2 are perturbed from their nominal values. From
í12 Anwar & Pan
(48), the perturbed values (lower bound) are taken as
í14
0 1 2 3 4 5 KP,2 = 60, TP,2 = 10, TT ,2 = 0.15, TG,2 = 0.04, R2 = 1.2.
Time (s)
The analysis is carried out by keeping the parameters of the area 1 in
nominal state. Now, the step loads of ΔPd,1 (s ) = 0.01 at t = 1 s and
ΔPd,2 (s ) = 0.01 at t = 10 s are applied to area 1 and area 2, respectively.
í3 The frequency and tie-line power deviations of the system using pro-
x 10
posed controller, Tan [37], and Padhan and Majhi [38] schemes are
0 shown in Figs. 7 and 8, respectively. It is observed that the frequency
deviations in both areas settle to zero in minimum time with least
overshoot by the proposed controller in comparison to the responses
í2 obtained by Tan, and Padhan and Majhi techniques. The similar im-
proved performance in comparison to other approaches is also obtained
Δ f (Hz)

in case of tie-line power as depicted in Fig. 8. Thus, the proposed

í4 controller rejects the load fluctuations and tolerate the uncertainties of
the plant parameters efficiently.

í Proposed 6.3. LFC design for two-area system having reheated and hydro turbines
Tan
Anwar & Pan
To extend the applicability of the proposed control scheme in LFC,
í we considered the two-area system in which the area 1 consists of a
0 2 4   10
reheated turbine of the form
Time (s)
cTr s + 1
PT (s ) =
(Tr s + 1)(TT s + 1) (50)
í3
x 10
5 and area 2 consists of hydro turbine of the form
1−Tw s
PT (s ) =
1 + 0.5Tw s (51)
0
Remark 10. Eq. (51) denotes the non-minimum phase characteristics
due to presence of RHP zero.
Δ f (Hz)

í5
For area 1, the model parameters are taken as
KP = 120, TP = 20, TT = 0.3, TG = 0.08, R = 2.4, Tr = 4.2, c = 0.35
í10 (52)
Proposed
The plant model now becomes
Tan
Anwar & Pan 87.5s + 59.52
í15 P (s ) =
s 4 + 16.12s 346.24s 2 + 48.65s + 25.3 (53)
0 1 2 3 4 5  7
Time (s) On applying the Routh approximation based reduced-order modeling
method [29], we get
1.572
Fig. 6. Frequency deviation for (a) nominal, (b) lower and (c) upper models. P (s ) =
s 2 + 1.285s + 0.6683 (54)
The PID parameters of Tan [35] are: kc = 0.4036, ki = 0.6356, kd = 0.1832 and
Anwar & Pan [36] are: kc = 1.52, ki = 2.50, kd = 0.27 . With the specifications ωgc = 5 rad/s and ϕm = π/3, the proposed
scheme using (43) yields
6.2. LFC design for two-area system 0.6683 1
C1 (s ) = 12.7923 ⎛1.285 + + s ⎞ ⎛ 0.333 ⎞
⎝ s ⎠⎝ s ⎠ (55)
We extend our proposed scheme to design decentralized PID tuning
for interconnected two-area power system. As mentioned in (35), one For area 2, the hydro turbine power system with following parameters
just needs to multiply the plant model by Bi to design a controller. For are considered:
simplicity, the two areas are assumed to be identical and the nominal KP = 1, TP = 6, Tw = 4, TG = 0.2. (56)

610
S. Saxena Electrical Power and Energy Systems 104 (2019) 603–614

Table 3
Performance (× 10−5 ).
Nominal Lower Upper

Method ISE IAE ITAE IE ISE IAE ITAE IE ISE IAE ITAE IE
Proposed 61.6 249.4 446.8 22.6 13.3 118 243.8 23 1.2 406 786 23
Tan [35] 1355 981.8 1663 981.8 677.8 1524 4106 1524 21.1 2146 4842 1529
Anwar & Pan [36] 261.6 426.1 609.6 400 109.5 400 703.8 400 4.7 876 1756 400

í3 í3
x 10 x 10
1
4

2 0.5
0

Δ P (MW)
í2 0
Δ f1 (Hz)

í4

1
í0.5
í6
Proposed Proposed
í8 Tan Tan
í1
Padhan & Majhi Padhan & Majhi
í10

í12 í1.5
0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18
Time (s) Time (s)

í3 í3
x 10 x 10

0 1.5
Proposed
Tan
í2 1 Padhan & Majhi
Δ P (MW)

í4
Δ f (Hz)

0.5
2
2

í6
0
Proposed
Tan
í8 Padhan & Majhi í0.5

í10
í1
0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18
Time (s) Time (s)

Fig. 8. Tie-line power deviation for (a) area 1 and (b) area 2.
Fig. 7. Frequency deviation for (a) area 1 and (b) area 2. The parameters of PID
from Tan [37] are: kc = 1.5692, ki = 2.3966, kd = 0.5259 , and Padhan & Majhi
[38] are: kc = 1.9822, ki = 0.5242, kd = 0.1756 . 1 1
C2 (s ) = 0.0635 ⎛8.2 + + 13.6s ⎞ ⎛ 0.5 ⎞
⎝ s ⎠⎝ s ⎠ (60)
The plant model without droop characteristics is The proposed controllers are applied to the actual LFC system having
reheated and hydro turbines in area 1 and area 2, respectively. The
−4s + 1
P (s ) = responses of frequency and tie-line power deviation of step load of
(0.2s + 1)(2s + 1)(6s + 1) (57)
ΔPd, i = 0.01 (i = 1, 2 ) occurring at t = 1 s and t = 30 s, respectively, in
Eq. (57) can be factorized in non-minimum phase part: P+ (s ) = −4s + 1 area 1 and 2, are presented in Figs. 9 and 10, which shows that the
and fluctuations in frequency and tie-line power tend to zero.

1 6.4. LFC design in presence of GRC and GDB

P− (s ) = .
(0.2s + 1)(2s + 1)(6s + 1) (58)
We consider a more realistic condition of power system where the
Using (40), the reduced-model concept is applied to P−(s ) which gives
physical constraints such as generation rate constraint (GRC) in turbine
8.2 and governor dead band (GDB) exist (see Fig. 11). To test the utility of
P−(s ) = the proposed controller, the study is extended to the case described in
111.52s 2 + 67.24s + 8.2 (59)
Section 5.1. A generation rate limitation of 0.1 p.u. per minute is con-
Using the specification ωgc = 0.09 rad/s and ϕm = π/4 , the proposed sidered here, i.e., ΔPG ⩽ 0.1 p. u. /min = 0.0017 p. u. /s [37] and the
controller becomes GDB width considered is 0.036 Hz [40]. Using the specification

611
S. Saxena Electrical Power and Energy Systems 104 (2019) 603–614

Fig. 11. (a) Turbine model with GRC and (b) dead-band in governor control
loop.

0.1

f (Hz)
-0.1

Fig. 9. Frequency deviation for (a) area 1 (reheated turbine) and (b) area 2 -0.2
(hydro turbine). 0 50 100 150
Time (s)
10-3 Fig. 12. Response in presence of GRC and GDB.
3

2 2.5026 1
(MW)

C (s ) = 1.726 × 10−4 ⎛1 + + 0.3146s ⎞ ⎛ 0.5 ⎞

⎝ s ⎠⎝ s ⎠ (61)
1
T1

The response of the frequency deviation is depicted in Fig. 12 which

P

0 shows that the proposed scheme can also work well in presence of GRC
and GDB constraints.
-1

0 50 100 150 7. Conclusion and future work

Time (s)
(a) This paper proposes a simple analytical PID load-frequency con-
troller to improve power system performance via fractional-order IMC
10-3 tuning and reduced-order modeling scheme. The computer simulations
2
have been conducted for single and multi-area power systems con-
sisting different types of turbines and physical limitations. The pro-
PT2 (MW)

0 posed scheme brings good disturbance rejection and eliminates mod-

eling error and parametric uncertainties.
The merits of the proposed scheme can be accounted in terms of
-2 simplicity in the design algorithm because the conventional fractional-
order control schemes require complex mathematical manipulations or
-4 optimization techniques to evaluate the tuning parameters. The other
0 50 100 150 advantage is that the controller evolved through the design process is
Time (s) PID controller followed by a fractional-order integrator. As far as LFC is
(b) concerned, this paper will serve a valuable resource for further research
when other physical constraints such as crossover elements in a thermal
Fig. 10. Tie-line power deviation for (a) area 1 (reheated turbine) and (b) area unit and communication delays are present in power systems.
2 (hydro turbine). Moreover, this work will encourage the researchers to investigate the
efficient reduced-order modeling algorithms for better dynamic per-
ωgc = 0.01 rad/s and ϕm = π/4 , the proposed scheme yields formance of power systems.

Appendix A. Asymptotic behavior of T (jω)

Eq. (18) can be rewritten as

612
S. Saxena Electrical Power and Energy Systems 104 (2019) 603–614

1
|T (jω)|dB= 20log γπ
1 + 2τωγ cos + (τωγ )2
2
1 1
= × 20log γπ
2 1 + 2τωγ cos + (τωγ )2
2

(
= −10log 1 + 2τωγ cos 2 + (τωγ )2
γπ
)
γπ
⎛ cos ⎞
= −10log ⎜ (τωγ )2 ⎡ γ 2 + + 1⎤ ⎟
1
2 τωγ2
⎢ (τω ) ⎥
⎝ ⎣ ⎦⎠
γπ
⎛ cos ⎞
= −20log ⎜ (τωγ ) ⎡ γ 2 + 2 τωγ2 + 1⎤ ⎟
1
⎢ (τω ) ⎥
⎝ ⎣ ⎦⎠ (A.1)
Therefore, the asymptotic behavior of (A.1) is obtained as
lim ω →∞|T (jω)|dB ≈ −20γ log(τω)

Similarly (19) can be produced as

γ γπ
τω sin
arg [T (jω)]= −arctan ⎡ 2 ⎤
⎢ 1 + τωγ cos γπ ⎥
⎣ 2 ⎦
γπ
sin
= −arctan ⎡ 2 ⎤
⎢ 1γ + cos γπ ⎥
⎣ τω 2 ⎦ (A.2)
and therefore for (A.2), we have
γπ
lim ω →∞ arg [T (jω)] ≈ −
2

Appendix B. Derivation of ωr and Mr

From Definition 4, Mr = max |T (jω)|ω = ωr and it is maximum when the denominator term of (17), i.e.,
γπ γπ
d (ω) = ⎛1 + τωγ cos ⎞ + jτωγ sin
⎝ 2 ⎠ 2 (B.1)
is minimum. Now differentiate (B.1) with respect to ω , and equate to zero, i.e.,
d ⎡ γπ γπ
⎛1 + τωγ cos ⎞ + jτωγ sin ⎤ = 0
dω ⎣ ⎝ 2 ⎠ 2 ⎦
1 γπ frac1τ 1
we get ωr = τ
cos 2
. On substituting ωr in (18), we get Mr = γπ .
sin
2

Appendix C. Derivation of ωp and ζ

From (16), the characteristic equation is

1 + τs γ = 0
1 π
The two poles are given by s1,2 = τ e ±j γ . The poles are complex and conjugate and form a center angle 2θ with respect to imaginary jω axis, where

( ) 1
θ = π 1− γ . Now, from the information of the poles, i.e., through the modulus 1
τ
and θ , we can obtain

1 1 1 1 π
ωp = sinθ = sinπ ⎜⎛1− ⎟⎞ = sin
τ τ ⎝ γ ⎠ τ γ

and

1 π
ζ = cosθ = cosπ ⎜⎛1− ⎟⎞ = −cos
⎝ γ ⎠ γ

Appendix D. Integer-order approximation of FO transfer function

In simulation, the Oustaloup method [39] is used to find integer-order approximations of FO transfer function in which
N
s+ω ∼
i
sμ = K ∏
i=1
s + ωi (D.1)
∼ = ω ω (2i − 1 − α )/ N , ω = ω ω (2i − 1 + α )/ N , K = ω α and ω =
where ω ωh
. Note that the number of poles and zeros (N) of approximating transfer function
i l r i l r h r
ωl
and the frequency range ([ωl , ωh]) must be selected before evaluating (D.1). When μ > 1 then it can be written in the form s μ = s[μ] s δ where [μ] is
greatest integer and then the term s δ is replaced by approximation transfer function (D.1).

613
S. Saxena Electrical Power and Energy Systems 104 (2019) 603–614

Appendix E. Supplementary data

Supplementary data associated with this article can be found, in the online version, at https://doi.org/10.1016/j.ijepes.2018.07.005.

References control. Arab. J. Sci. Eng. 2016;41(9):3473–89.

[20] Saxena S, Hote YV. Internal model control based PID tuning using first-order filter.
Int J Control Autom Syst (in press).
[1] Kundur P, et al. Definition and classification of power system stability IEEE/CIGRE [21] Muresana CI, et al. Tuning algorithms for fractional order internal model controllers
joint task force on stability terms and definitions. IEEE Trans. Power Syst. for time delay processes. Int. J. Control 2016;89(3):579–93.
2004;19(3):1387–401. [22] Lanusse P, Malti R, Melchior P. CRONE control system design toolbox for the
[2] Bevrani H. Robust power system frequency control. Springer; 2009. control engineering community: tutorial and case study. Philos. Trans. R. Soc. A
[3] Saxena S, Hote YV. Load frequency control in power systems via internal model 2013;371.
control scheme and model-order reduction. IEEE Trans. Power Syst. [23] Maâmar B, Rachid M. IMC-PID-fractional-order filter controller design for integer
2013;28(3):2749–57. order systems. ISA Trans. 2014;53:1620–8.
[4] Saxena S, Hote YV. Decentralized PID load frequency control for perturbed multi- [24] Podlubny I. Fractional differential equations. San Diego: Academic Press; 1999.
area power systems. Int. J. Electric Power Energy Syst. 2016;81:405–15. [25] Fortuna L, Nunnari G, Gallo A. Model order reduction techniques with applications
[5] Ibraheem Kumar P, Kothari DP. Recent philosophies of automatic generation con- in electrical engineering. London: Springer-Verlag; 1992.
trol strategies in power systems. IEEE Trans. Power Syst. 2005;20(1):346–57. [26] Benner P, Mehrmann V, Sorensen DC. Dimension reduction of large-scale systems.
[6] Shayeghi H, Shayanfar HA, Jalili A. Load frequency control strategies: a state-of-the- Springer; 2005.
art survey for the researcher. Energy Convers. Manage. 2009;50:344–53. [27] Schilders WH, Van der Vorst HA, Rommes J. Model order reduction: theory, re-
[7] Pandey SK, Mohanty SR, Kishor N. A literature survey on load-frequency control for search aspects and applications. Springer; 2008.
conventional and distribution generation power systems. Renew. Sustain. Energy [28] Fortuna L, Frasca M. Optimal and robust control: advanced topics with MATLAB.
Rev. 2013;25:318–34. CRC Press; 2012.
[8] Padula F, Visioli A. Advances in robust fractional control. Springer; 2015. [29] Hutton MF, Friedland B. Routh approximations for reducing order of linear, time-
[9] Monje CA, Chen YQ, Vinagre BM, Xue D, Feliu V. Fractional-order systems and invariant systems. IEEE Trans. Autom. Control 1975;AC-20:329–37.
controls: fundamentals and applications. Springer; 2010. [30] Oustaloup A. La commande CRONE. Paris: Hermès Editions; 1991.
[10] Alomoush MI. Load frequency control and automatic generation control using [31] Sabatier J, et al. CRONE Control: principles and extension to time-variant plants
fractional-order controllers. Electr. Eng. 2010;91:357–68. with asymptotically constant coefficients. Nonlinear Dyn. 2002;29:363–85.
[11] Sondhi S, Hote YV. Fractional order PID controller for load frequency control. [32] Malti R, Moreau X, Khemane F, Oustaloup A. Stability and resonance conditions of
Energy Convers. Manage. 2014;85:343–53. elementary fractional transfer functions. Automatica 2011;47(11):2462–7.
[12] Debbarma S, Saikia LC, Sinha N. Automatic generation control using two degree of [33] Kundur P. Power systems stability and control. Mc-Graw Hill; 1994.
freedom fractional order PID controller. Int. J. Electr. Power Energy Syst. [34] Tan W. Load frequency control: problems and solutions. In: Proc. 30th Chinese
2014;58:120–9. Control Conference, Yantai, China, July 22–24; 2011.
[13] Pan I, Das S. Fractional-order load-frequency control of interconnected power [35] Tan W. Unified tuning of PID load frequency controller for power systems via IMC.
systems using chaotic multi-objective optimization. Appl. Soft Comput. IEEE Trans. Power Syst. 2010;25(1):341–50.
2015;29:328–44. [36] Anwar MN, Pan S. A new PID load frequency controller design method in frequency
[14] Taher SA, Fini MH, Aliabadi SF. Fractional order PID controller design for LFC in domain through direct synthesis approach. Electr. Power Energy Syst.
electric power systems using imperialist competitive algorithm. Ain Shams Eng. J. 2015;67:560–9.
2014;5:121–35. [37] Tan W. Tuning of PID load frequency controller for power systems. Energy Convers.
[15] Morari M, Zafiriou E. Robust process control. NJ, Prentice-Hall: Englewood Cliffs; Manage. 2009;50:1465–72.
1989. [38] Padhan DG, Majhi S. A new control scheme for PID load frequency controller of
[16] Saxena S, Hote YV. Advances in internal model control technique: a review and single-area and multi area power systems. ISA Trans. 2013;52:242–51.
future prospects. IETE Tech. Rev. 2012;29(6):461–72. [39] Oustaloup A, Levron F, Mathieu B, Nanot FM. Frequency band complex noninteger
[17] Gaona DC, Goytia ELM, Lara OA. Fault ride-through improvement of DFIG-WT by differentiator: characterization and synthesis. IEEE Trans. Circuits Syst. I: Fundam.
integrating a two-degrees-of-freedom internal model control. IEEE Trans. Industr. Theory Appl. 2000;47(1):25–39.
Electron. 2013;60(3):1133–45. [40] IEEE Standard 122-1991. Recommended practice for functional and performance
[18] Yazdanian M, Sani AM. Internal model-based current control of the RL filter-based characteristics of control systems for steam turbine-generator units; 1992.
voltage-sourced converter. IEEE Trans. Energy Convers. 2014;29(4):873–81.
[19] Saxena S, Hote YV. Simple approach to design PID controller via internal model

614