IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 39, NO.

4, JULY 2024 5981

Load Frequency Control in Microgrids: A Robust

Bi-Objective Virtual Dynamics Technique
Ali Rafiee , Yazdan Batmani , Ali Mehrizi-Sani , Senior Member, IEEE, and Hassan Bevrani , Fellow, IEEE

Abstract—In this paper, a new bi-objective control technique is electrical power; and as a result, the overload is supplied by the
proposed for the load frequency control in microgrids. The first spinning reserve generation capacity. The spinning reserve is the
objective is to set the system frequency to its desired value despite total amount of generation available from all units synchronized
uncertainties in the system inertia and damping caused mainly
by the penetration of renewable energy sources. The employed (i.e., spinning) on the system, minus the present load and losses
strategy to address this objective is based on the virtual inertia being supplied [1], [2]. To secure safe system operation and a
concept. The second objective is to design an automatic mechanism possibility of activating the primary control, the system operator
for charging/discharging of the battery storage units. By achieving must have an adequate spinning reserve at its disposal [3], [4],
this objective, it is possible to attain a compromise between the main [5].
mechanical units and the backup units to participate in supplying
the load. The strategy to address this objective is to estimate the In modern power grids, with increasing penetration of dis-
load change and then design a controller to track this signal by the tributed generation (DG) sources, the moments of inertia and
mechanical units. A multivariable observer-based proportional- inertia constant of the whole system decrease; and consequently,
integral (PI) controller is designed to simultaneously achieve both the system sensitivity to disturbances, caused by load changes
objectives. Applying the proposed method not only compensates and deviations in the DGs production, increases [3], [6]. Another
for the adverse effects of the uncertainties and disturbances, but
it also decreases the needed capacity of energy storage resources. challenge in modern power grids is the variability of the inertia
Simulation results show that the proposed method can improve since the power generated by DGs is highly dependent on
the microgrid load frequency characteristic while robust stability environmental and climatic conditions. Therefore, conventional
and performance are guaranteed with the limited energy storage control methods used in traditional power grids may not attain
capacity. the desirable performance of a modern power grid such as
Index Terms—Backup system control, load frequency control, microgrids.
microgrid, multivariable virtual inertia control. The utilization of energy storage devices (ESDs) and inverter-
interfaced distributed generators (IIDGs) is a viable option for
enhancing the system frequency, and the virtual inertia con-
I. INTRODUCTION
trol represents an alternative methodology for attaining this
N AC power systems with synchronous generating units,
I the quality of the frequency and active power control are
affected by the rotor moment of inertia. Frequency stability
objective. Numerous technologies exist for the implementation
of an energy storage system, such as batteries, flywheels, su-
perconducting magnetic energy storage, and supercapacitors.
and control are essential for the safe operation of power grids, Among the available candidates for energy storage units in
which are closely related to the balance of active power between energy storage systems, supercapacitors have the advantages
generation and consumption. If the electrical load increases of high power density and long lifetime [7], hence the focus
suddenly, the electrical power output exceeds the mechanical of this paper is on supercapacitors, although the same concepts
input power. This power shortage is compensated by the kinetic can be applied to other energy storage systems as well. The
energy of the rotating system. A decrease in the kinetic energy concept of virtual inertia refers to the immediate injection of
causes the turbine to decelerate resulting in a reduction of the additional active power into the power grid in the case of an
system frequency. When the governor senses these changes, the active power imbalance leading to an increase in the frequency,
steam inlet valve opens to balance the mechanical power and the which is made possible by a suitable control algorithm [3], [8].
The general approach for developing virtual inertia control is to
Manuscript received 29 June 2023; revised 13 October 2023; accepted 19 focus on the synchronous generators dynamics as represented
November 2023. Date of publication 28 November 2023; date of current ver- by an oscillation equation, and applying the concept of a vir-
sion 20 June 2024. Paper no. TPWRS-01005-2023. (Corresponding author:
Yazdan Batmani.)
tual synchronous generator (VSG) [8]. Additional dynamics
Ali Rafiee, Yazdan Batmani, and Hassan Bevrani are with the Department of synchronous generators, including primary and secondary
of Electrical Engineering, University of Kurdistan, Sanandaj 66177-15175, Iran control loop dynamics, rotor damper wiring dynamics, and
(e-mail: [email protected]; [email protected]; [email protected]).
Ali Mehrizi-Sani is with the Bradley Department of Electrical and Com-
stator wiring dynamics, have been suggested in previous lit-
puter Engineering, Virginia Tech, Blacksburg, VA 24061-0131 USA (e-mail: erature [9], [10]. Recently, numerous studies are conducted in
[email protected]). the field of virtual inertia control (for example, robust control
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TPWRS.2023.3336692.
methods [11], [12], [13], adaptive control methods [14], [15],
Digital Object Identifier 10.1109/TPWRS.2023.3336692 integral control method [16], optimal control methods [17],

0885-8950 © 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on January 23,2025 at 12:09:45 UTC from IEEE Xplore. Restrictions apply.
 5982 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 39, NO. 4, JULY 2024

[18], fuzzy-logic and neural network methods [19], [20], and

predictive controllers [21], [22]). These methods improve the
frequency characteristics of the system. Since they have only
focused on a single objective, i.e., the frequency regulation,
other aspects of the problem are not addressed in their design.
For instance, in the robust control methods [11], [12], [13], the
frequency sensitivity to the system inertia constant is reduced
by designing controllers with relatively large steady-state gains;
and as a result, in the case of an active power balance disturbance
leading to an increase in the frequency, almost only the virtual Fig. 1. Block diagram of the one-area power grid consisting of the conven-
inertia source is triggered to increase the generate active power tional virtual inertia control unit.
and the mechanical units are used with limited contribution.
In this paper, a bi-objective load frequency control (LFC) while its numerator does not considerably change; and hence,
problem is defined in which (i) the frequency regulation is ad- Hsys is reduced. As a well-known index for assessing the system
dressed and (ii) the mechanical units are successfully controlled frequency, the maximum rate of change of frequency (RoCoF)
to assist in contributing active power balancing when the virtual 
can be computed as RoCoF = −ΔP0 fn /(2 N i=1 Si Hi ) [3]
control loop is active. To solve this extended LFC problem sys- where ΔP0 is the change in the active power balance defined as a
tematically, a multivariable control strategy is proposed based loss of the power generation or an increase in the power demand;
on a high-gain proportional-integral (PI) controller. An adaptive fn is the nominal system frequency. Equipping the system with
set-point for the mechanical units is provided by a state observer, some additional inertia, either naturally or virtually, can decrease
and a PI controller is tuned to enable these units to participate in the adverse effects of any inertia reduction on the frequency
overload supplement. As the utilized controller is a two-input characteristics. To effectively inject active power into the grid
two-output (TITO) PI controller, both the above mentioned via ESDs or IIDGs, virtual inertia methods must operate rapidly.
objectives are achieved. Effectiveness of the proposed control One approach to achieving this is through the implementation of
strategy is investigated by simulating some realistic scenarios. virtual inertia control, which involves mimicking the standard
The features of the proposed TITO structure are as follows:
r The robust stability and performance of the closed-loop swing equation of an SG. This equation can be expressed in a
specific format, as illustrated below [11]:
system are fulfilled even in the presence of parametric un-
certainties and external disturbances due to the penetration dΔf
ΔPVI = KVI + DVI Δf, (1)
of the IIDGs and varying climatic conditions. dt
r An automatic mechanism for charging/discharging of the
where ΔPVI represents the quantity of extra power that is
ESDs is established without inserting any additional con- injected into the grid to attain power balance, in accordance
trol loops. with the virtual inertia control algorithm; KVI and DVI are the
r The required capacity of the ESDs can be considerably
virtual inertia constant and the virtual damping coefficients,
reduced due to forcing the mechanical units to participate respectively. The amount of kinetic energy reduction due to
in the change in the active power balance. penetration of IIDGs plays an important role to determine these
r As the order of the proposed observer-based TITO control
gains, and robust virtual inertia control methods are widely
strategy is fixed, the overall closed-loop system is simple. employed due to severe uncertainty in this key factor [10], [11],
The other parts of the paper are organized as follows. In [13]. In these methods, robust stability and performance of the
Section II, the system is described, and the control problem is system are maintained by reducing their sensitivity to the amount
defined. In Section III, the proposed TITO controller is designed. of kinetic energy reduction. As depicted in Fig. 1, the system
In Section IV, a state observer is designed to estimate the load consists of synchronously oscillating diesel generators and/or
change. Section V examines the performance of the proposed micro-gas turbines, in addition to IIDGs. Here, H and D are
control strategy using the MATLAB/SIMULINK software. Fi- the system inertia and damping coefficient, respectively; τt and
nally, Section VI concludes the paper. τg are the turbine and governor time constants, respectively; R
is the governor droop constant; Ki , and β are the secondary
II. PROBLEM STATEMENT integral gain and the bias factor, respectively; and ΔP0 is the
external disturbance. The output power granted by the IIDGs
To address the LFC problem in a modern power grid with is regarded as external disturbances, and accounted for in ΔP0 ,
synchronous generators, the overall behavior of all generators and the influence of the penetration coefficient of these sources is
is examined. This entails considering the coherent response of considered as a system uncertainty since the values of H and D
all generators to variations in the system load and expressing it are uncertain [10], [11]. The closed-loop sensitivity is defined as
in terms of an equivalent generator. The equivalent generator S(s) = 1/(1 + L(s)), where L(s) = G(s)C(s) is the loop gain
has an inertia constant, shown with Hsys and computed as
 N consists of the designed virtual inertia controller C(s), and G(s)
Hsys = N i=1 Si Hi / i=1 Si [3] where N is the number of is defined as follows:
generators; Si and Hi are the power rating and inertia constant
Δf −s(sτg +1)(sτt +1)
of the ith generator, respectively. Increasing the penetration of G(s) = = . (2)
IIDGs increases the denominator of the above equation for Hsys , ΔP0 s(2Hs+D)(sτg +1)(sτt +1)+Ki + Rs
Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on January 23,2025 at 12:09:45 UTC from IEEE Xplore. Restrictions apply.
 RAFIEE et al.: LOAD FREQUENCY CONTROL IN MICROGRIDS: A ROBUST BI-OBJECTIVE VIRTUAL DYNAMICS TECHNIQUE 5983

The transfer function of the virtual inertial control loop has the
following general form that can encompass all the dynamics of
the controller and power converter.
bm sm + bm−1 sm−1 + · · · + b1 s + b0
C(s) = (3)
sn + an−1 sn−1 + · · · + a1 s + a0
As previously mentioned, a robust virtual inertia controller
aims to decrease the sensitivity of the closed-loop system S(s),
which is typically achieved by setting a high DC gain value for
the transfer function, denoted as C(0). However, the following
analysis indicates that using a high DC gain value for shaping
the virtual inertia control loop may limit the system’s ability Fig. 2. Interface topology used to regulate the power flow between the ESD
and microgrid.
to effectively utilize its spinning reserve generation capacity.
To explore this issue, let us compute the transfer function from
ΔP0 (s) to the mechanical power change ΔPm (s) as follows:
ΔPm (s) 1 s+βRKi
= × . (4)
ΔP0 (s) sR(1+sτg )(1+sτt ) (2Hs+D)+C(s)
The steady-state participation of the spinning reserve generation
units can be determined by applying the final value theorem
based on (4) as follows:
Fig. 3. Bidirectional DC–DC topology used in the ESSC.
ΔPm βKi
= . (5)
ΔP0 D + C(0)
emergency control are out of the scope of this paper and are not
Using (5) and the fact that C(0) is large, it can be concluded analyzed here.
that the spinning reserve generation units have a minor role in
responding to active power balance changes. For example, when III. PROPOSED TITO CONTROLLER
applying the robust controllers designed in [11], since C(0) =
By delivering short and instantaneous energy injections, this
1.82 × 103 and using (5), the ratio of the spinning reserve
paper focuses on enhancing the effectiveness of ESDs, specif-
generation power change to the load change is 6.44 × 10−5 .
ically supercapacitors, to improve the frequency performance
Therefore, almost all the overload must be supplied by the ESD,
of microgrids comprised of small local thermal power plants
and we need a high-capacity storage source. From a practical
or diesel generators and IIDGs. The interface topology used to
point of view, however, the ESDs have limited capacity. If the
regulate the power flow between the ESD and the microgrid
overload remains for long periods of time or when successive
consists of an energy storage side converter (ESSC) connected
overloads present even with short inter intervals, the power of
to the microgrid via a DC-link and a grid side converter (GSC)
the ESDs constantly decrease in the absence of the spinning
is shown in Fig. 2. The presence of the DC-link permits a
reserve generation contribution. In this situation, the frequency
degree of decoupling between the ESSC and GSC, allowing
regulation is not successfully achieved; and consequently, the
each controller design to be independently conducted.
overall system performance decreases.
The ESSC controller regulates the exchange of energy be-
Based on the above discussion, the following two main ob-
tween the ESD and the DC-link. The bidirectional DC/DC
jectives should be simultaneously addressed by a proper control
configuration, used to implement the ESSC, is depicted in Fig. 3.
structure:
r Objective #1: The system frequency should be regulated In this case, VDC is always higher than VESD , but energy can flow
in either direction. This converter can operate in the buck mode,
to its desired value despite uncertainty in the total inertia
which supplies energy to the storage device. The converter can
constant and damping coefficient as well as external dis-
also operate in the boost mode, which draws energy from the
turbances caused by the load changes and DGs variations.
r Objective #2: The spinning reserve generation units storage device. The following unified model can be utilized to
mathematically characterize these two operational modes:
should be excited to enable them to effectively contribute
in supplying the overload. VESD VDC
I˙L = − + Sw ,
Remark 1: In the case of a severe fault or a significant loss LESD LESD
of generation, the frequency may continue to decline until the IL
system experiences a frequency collapse. In such cases, it is V̇ESD = , (6)
CESD
necessary to activate the atypical operations of tertiary control,
emergency control (e.g., underfrequency load shedding), and where Sw1 = Sw , Sw2 = S̄w . The overline symbol is used to
protection mechanisms to reduce the likelihood of cascading represent the logical opposite state, i.e., if Sw = 1, then S̄w = 0.
failures, load or grid/network separation events, and additional In the power control mode, the control developed in [23] ensures
generation events [1], [3], [5]. However, the tertiary control and that the power transferred to/from the grid is controlled to track

Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on January 23,2025 at 12:09:45 UTC from IEEE Xplore. Restrictions apply.
 5984 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 39, NO. 4, JULY 2024

Fig. 4. Conventional active power control loop.

∗
the power reference PESD using the control IL such that the
voltage of the capacitor bank does not exceed its acceptable
range VESDmin < VESD < VESDmax . In this situation, the closed loop
dynamics of the ESSC system is obtained as follows [23]:
IL3
I˙L = − ∗ ,
PESD CESD
∗
IL PESD
V̇ESD = = , (7) Fig. 5. Low inertia microgrid.
CESD VESD CESD
where VESD and CESD are the voltage and capacity of the super- ∗
capacitors, respectively. The GSC controller regulates VDC re- where PESD is the reference power for the ESD; τGSC is the time
garding the output power flow to maintain the DC-link’s energy constant of the grid side converter, that is displayed with τVI to
balance. The GSC control typically employs a phase locked-loop integrate this topic.
(PLL) that generates the grid frequency and phase-angle for As depicted in Fig. 5, the low inertia microgrid can consist
the dq-frame current control, as well as the outer voltage-loop of local thermal power plants or diesel generators and two types
and the inner current-loop controllers. More details on the GSC of IIDGs systems. The first type is grid-imposed frequency
control can be found in [7], [24]. However, the power tracking IIDGs such as photovoltaic (PV) and wind turbine generator
model of the GSC is used in the power-frequency synthesis and (WTG) systems, in which the operating frequency is prede-
analysis. The typical active power control loop of the GSC is termined and imposed by the AC grid [29]. The presence of
shown in Fig. 4 [25] where ωf is the cut-off angular frequency these resources decreases the system’s moment of inertia. The
of the low-pass filter; ω and δ are the angular frequency and second type of IIDGs is controlled-frequency IIDGs such as fuel
power angle, respectively; P and P ∗ are the active power cell (FC) systems, flywheel energy storage systems (FESS), and
and its reference, respectively. The gains Kp = ΔωMAX /Pn and battery energy storage systems (BESS), in which the operating
Kpd = 3Vn2 /X represent the frequency droop coefficient and the frequency is not determined by the grid but rather by the IIDGs
amplification from the change in δ to the change in the active system itself [29]. The presence of controlled-frequency IIDGs
power, respectively. ΔωMAX is the maximum allowable angular increases the system inertia, and consequently, the microgrid
frequency deviation; Pn is the nominal active power; X = ωLs performance [10]. These sources can adjust their power output
is the impedance of the line inductor Ls . The transfer function in response to changes in the microgrid frequency. As the pro-
from P ∗ to P can be derived from Fig. 4 as follows: posed controller is robust against uncertainties, it is expected
that the controller can maintain its performance in the absence
P 1 and presence of both of these resources (see Scenario 1). More
∗
= , (8)
P τp (τω s + 1)s + 1 precisely, the following two types of uncertainty can be handled
where τp = X/3Kp Vn2 = X pu /ΔωMAX and τω = 1/ωf . The by the controller:
r Input disturbance ΔP0 : It refers to changes in the active
time constant τp is proportional to the per-unit output inductance
of the GSC, and the time constant τω is equal to the inverse of power balance, which can be defined as a decrease in
the low-pass filter’s cut-off frequency. ωf is typically one decade power generation or an increase in power demand. This
below 50 Hz, and ΔωMAX  1 [25], [26], [27]. As a result, the disturbance is depicted in Fig. 5 as the total effects of the
condition 0 < τω  τp often holds, and (8) can be simplified as first type of IIDGs and load changes.
r Unknown dynamics of the system: This uncertainty arises
follows:
due to ignoring the dynamics of the second type of IIDGs.
P 1
= . (9) The proposed controller does not use the positive impacts
P∗ τp s + 1 of these resources in improving the frequency restoration
Based on (7) and (9), in the power tracking mode, the ESD of the microgrid.
dynamics can be characterised by [28] The constant of inertia and the damping coefficient depend
∗
PESD on the penetration of IIDGs; and thus, both of the above sources
V̇ESD = , of uncertainties change these two parameters.
VESD CESD
To address both of the defined objectives in Section II, we
∗
PESD − PESD propose a novel disturbance observer-based control method in
ṖESD = , (10)
τGSC which the whole system is considered as a TITO system. Fig. 6

Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on January 23,2025 at 12:09:45 UTC from IEEE Xplore. Restrictions apply.
 RAFIEE et al.: LOAD FREQUENCY CONTROL IN MICROGRIDS: A ROBUST BI-OBJECTIVE VIRTUAL DYNAMICS TECHNIQUE 5985

TABLE I
USED PARAMETERS IN THE STUDIED MICROGRID

the system must be functional controllable [30]. By computing

Fig. 6. Block diagram of proposed observer-based control. G(s) = C(sI − A)−1 B based on (11), it can be concluded that
G(s) is full rank; and thus, both the system outputs can be
shows the block diagram of the proposed structure. It is assumed independently controlled.
that the synchronous generator and the ESDs are located in the Definition 1: Consider the state-space representation
same place. The duty of the TITO controller is to simultaneously
gain the following issues: ẋ(t) = Ax(t) + Bu(t),
r Set the system frequency to its desired value (Objective 1).
y(t) = Cx(t), (12)
r Track the change in the active power balance by the thermal
unit (Objective 2). where x(t) ∈ Rn is the system state; u(t) ∈ Rm is the control
By defining x1 (t) = Δf (t), x2 (t) = ΔPm (t), x3 (t) = ΔPg input; y(t) ∈ Rq is the system output; A ∈ Rn×n , B ∈ Rn×m ,
(t), x4 (t) = ΔPESD (t), X1 (t) = [x1 (t), x2 (t)]T , and X2 (t) = and C ∈ Rq×n are constant matrices. The Markov parameter
[x3 (t), x4 (t)]T , the following state-space representation is sequence of (12) is shown as {h(i)} (i = 0, 1, . . . ) and defined
achieved: as {0q×n , CB, CAB, . . . , CAi−1 B, . . . } [31].
        
Ẋ1 A11 A12 X1 02×2 − 2H
1 Definition 2: The state-space representation (12) is called
= + u+ ΔP0 , regular if h(1) = CB is full rank. Otherwise, it is irregular.
Ẋ2 A21 A22 X2 B2 03×1
         Using the nominal parameters in Table I, the state-space
A B Bd representation (11) is irregular. Moreover, by computing the
  controllability and observability matrices [32], one can find that
X1 both matrices are full rank; and hence, the triple (A, B, C) is
y = I2 02×2 (11)
   X2 controllable and observable. By designing a TITO controller
C such that the closed-loop system is stable and has the ability to
where 0i×j and Ii stand for the i × j zero matrix and i × i track reference signals by the system outputs, both Objectives
identity matrix, respectively, and ΔP0 is the change in the active 1 and 2 are satisfied. The employed observer is designed to
power balance defined as a loss of the power generation or an estimate the change in the active power balance, ΔP̂0 is a step
increase in the power demand, i.e., ΔP0 = ΔPLoad − ΔPPV − signal. The desired values of the first and second outputs, i.e., Δf
ΔPW . The constant matrices in (11) are and ΔPm , are zero and ΔP̂0 , respectively. The TITO controller
      must be able to stabilize the whole system and eliminate the
D −1
− 2H 1
0 2H1
Rτg 0 tracking error. Other important issues that must be considered
A11 = 2H , A12 = 1 , A21 = , in the controller design procedure are the robustness of the
0 − τ1t τt 0 0 0
    system and the simplicity of the controller. In the following,
− τ1g 0 1
τg 0 a TITO high-gain PI controller is designed since it has all these
A22 = , B2 = . properties. The block diagram of the controller is shown in
0 − τ1VI 0 τ1VI
Fig. 7 where r(t) ∈ R2 is the reference signal; KP ∈ R2×2 and
To achieve the above control objectives simultaneously, the KI ∈ R2×2 are the proportional and integral gains of the PI
system outputs in (11) must be controlled independently. An controller; g  0 is the tuning gain of the controller; M ∈ R2×2
m-input l-output MIMO system with the matrix transfer func- is the measurement matrix used to construct the extended output
tion G(s) is called functional controllable if the normal rank w(t) ∈ R2 . The problem is to find proper values of KP , KI , g,
of G(s) is l [30]. In order to control all outputs independently, and M such that the control objectives are fulfilled.

Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on January 23,2025 at 12:09:45 UTC from IEEE Xplore. Restrictions apply.
 5986 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 39, NO. 4, JULY 2024

r A11 − A12 F−1 F1 : This set of observable but uncontrol-

2
lable slow dynamics is the solutions of |λI2 − A11 +
A12 F−12 F1 | = 0. These poles are also the zeros of the
system and usually decrease the speed of the closed-loop
system response.
r −K−1 KI : This set of uncontrollable and unobservable
P
slow dynamics is the solutions of |λI2 + K−1 P KI | = 0.
Fig. 7. Block diagram of the TITO high-gain PI controller. Selecting KI = αKP , where α is a user-defined positive
constant, can stabilize this set.
r −gF2 B2 KP : This set of controllable and observable fast
Defining e(t) = r(t) − w(t), ż(t) = e(t), the control input
u(t) ∈ R2 and the extended output w(t) can be written as dynamics is the solutions of |λI2 + gF2 B2 KP | = 0. Se-
lecting KP = (F2 B2 )−1Σ , it is possible to place these
u(t) = g (KP e(t) + KI z(t)) , poles at −gσi (i = 1, 2) where Σ = diag{σ1 , σ2 } and σi
(i = 1, 2) are the user-defined positive scalars.
w(t) = y(t) + MẊ1 (t). (13)
Using (15) and (16), shown at the bottom of this page and
Ẋ1 (t) asymptotically tends to zero provided that the closed- when g is large enough, the closed-loop matrix transfer function
loop system is stable and the system output tracks a step ref- is as follows:
erence signal. Therefore, using (13), w(t) and y(t) have the −1
same steady-state values. Using (11), w(t) can be rewritten as Γf (s) + Γs (s) = sI2 − A11 +A12 F−1
Γ(s) =Γ 2 F1 A12 F−1
2
follows:
    Since the second and last sets of the closed-loop poles are
X1 (t) X1 (t) placed in the predefined locations in the left half-plane, the
w(t) = I2 02×2 + M A11 A12
X2 (t) X2 (t) above transfer function is stable. If the measurement matrix
  M is selected such that F2 is invertible, the eigenvalues of
X1 (t) A11 − A12 F−1 2 F1 are placed in the left half-plane [26], [33].
= F1 F 2 , (14)
X2 (t) Using the nominal parameters in Table I, the constant matrices
in (11) are as follows:
where F1 = I2 + MA11 and F2 = MA12 . Augmenting (14)
   
and ż(t) = e(t) leads to closed-loop state-space representation −0.09 6.02 0 6.02
(15) [26], [33]. A11 = , A12 = ,
0 −2.70 2.70 0
To simplify the stability analysis of the closed-loop system, a
similarity transformation is applied to achieve a block diagonal      
−5.49 0 14.28 0 14.28 0
state matrix. The equivalent state-space representation (16) is A21= , A22= , B2 = .
attained for the closed-loop system where xs (t) and xf (t) are the 0 0 0 −1 0 1
slow and fast dynamics of the system. Since the state matrix in (17)
(16) is block diagonal, the closed-loop poles are the set of eigen-
values of the diagonal elements −K−1 −1 Selecting M as follows leads to F1 presented below:
P KI , A11 − A12 F2 F1 ,
and −gB2 KP F2 . To place the eigenvalues of these elements    
in some pre-determined locations in the s-plane, the following 0.25 0.05 0.975 1.370
M= , F1 = ,
issues should be considered [26], [33]: 0.05 0.25 −0.005 0.625

⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
ż(t) 0 −F1 −F2 z(t) I2
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎣ 1 ⎦
ẋ (t) = ⎣ 0 A11 A12 ⎦ ⎣x1 (t)⎦ + ⎣ 0 ⎦ r(t) (15)
ẋ2 (t) gB2 KI A21 − gB2 KP F1 A22 − gB2 KP F2 x2 (t) gB2 KP
⎡ ⎤
z(t)
⎢ ⎥
y(t) = 0 I2 0 ⎣x1 (t)⎦
x2 (t)
⎡ ⎤ ⎡ ⎤
  A11 − A12 F−1   A12 F−1
2 F1 0 0 2
ẋs (t) ⎢ ⎥ xs (t) ⎢ ⎥
=⎣ 0 −K−1
P KI 0 ⎦ +⎣ 0 ⎦ r(t)
ẋf (t) xf (t)
0 0 −gB2 KP F2 gB2 KP
 
xs (t)
y(t) = I2 0 0 (16)
xf (t)

Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on January 23,2025 at 12:09:45 UTC from IEEE Xplore. Restrictions apply.
 RAFIEE et al.: LOAD FREQUENCY CONTROL IN MICROGRIDS: A ROBUST BI-OBJECTIVE VIRTUAL DYNAMICS TECHNIQUE 5987

and we have the following full rank solution for F2 : locations in the left-half-plane (LHP) of the s-plane [32]. In
  this situation, the stability of the estimation error is guaranteed
0.135 1.506 ˙
since ẽ(t) = (A − LC )ẽ(t) is stable; and hence, ẽ(t) asymp-
F2 = .
0.675 0.301 totically tends to zero. Paying attention to the dominant poles
of the closed-loop system with the designed TITO PI controller,
For the above values, the eigenvalues of A11 − A12 F−1
2 F1 L is selected such that the first four observer poles are placed at
are −5 and −3.33. By tuning σ1 = 10σ2 = 10 and α = 50, the
−95, −85, −40, and −20. The last pole of the observer is also
controller gains are computed as follows:
    located at −1. The obtained observer gain is as follows:
−0.215 0.107 −10.791 5.396  T
KP = , KI = . 129.248 −3.034 −17.034 −0.001 −593.893
6.917 −0.138 345.833 −6.917 L 1=
−4.303 93.667 162.536 0.024 82.485
The other eigenvalues of the system, i.e., the second and last
parts, are placed in −50, −50, −100, and −1000; and therefore, Remark 2: It is also possible to select the last pole of the
the asymptotic stability of the closed-loop system is guaranteed. observer far away from the imagery axis. In this situation, the
The second objective, i.e., tracking a step-kind reference signal, convergence rate of the obtained observer and as a result its
is also guaranteed due to the presence of non-zero integral action frequency bandwidth are increased. Adverse effects of noises
in the control structure. on the closed-loop system, however, may be intensified.

IV. ESTIMATION OF THE ACTIVE POWER BALANCE V. SIMULATION RESULTS

As mentioned earlier, the TITO high-gain PI controller is In this section, the performance of the proposed TITO con-
designed such that the system output tracks specific reference troller is investigated using simulations of a microgrid test
signals. While the first output, i.e., Δf , must be set to zero, system comprising 15 MW of domestic loads, 12 MW of thermal
the second output ΔPm must track changes in the active power power plant, 7 MW of wind farm, 4 MW of solar farm, 2 MW
balance. Therefore, the designed TITO controller can be imple- fuel cell energy storage system, and 3 MW of electrical and
mented in practice provided that ΔP0 is available. Nevertheless, mechanical energy storage system. The system base is 20 MW.
due to its nature, ΔP0 is unknown to the controller. In this Assume that the PVs of the solar farm and wind farm turbines
subsection, a dynamic observer is employed to compute an are connected to the DC sides of the IIDGs as a DC power
estimation of the load change, shown as d(t). source. The real and reactive power control of the IIDGs in a
Since the load may change slowly; and therefore, it is assumed given situation could be either current or voltage control mode,
˙ = 0. Augmenting this dynamic with (11) leads to the
that d(t) as discussed in detail in [26], [29], [34], and it is concluded that
following state-space representation: by effectively regulating the voltage on the DC side, the response
    of the output power to the input reference power can be modeled
A Bd B as a first-order transfer function, with the time constant τ as a
Ẋ (t) = X (t) + u(t),
01×4 0 01×2 variable that can be adjusted to meet design requirements. The
      microgrid parameters are presented in Table I.
A B 1) Effects of Unmodeled Dynamics: As previously stated
y(t) = C 02×1 X (t), (18) and for the aim of simplicity, the dynamics of FC, BESS,
   and FESS are not considered in the design procedure of the
C proposed controller. It is assumed that the load is increased
0.1 pu at t = 0.1 s. The frequency and ESD power variations
where X (t) = [x (t), d(t)]T ∈ R5 is the augmented state
T
in the presence and absence of the aforementioned sources are
vector. If the pair (A , C ) is observable, a full-order Lu-
depicted in Fig. 8(a) and (b), respectively. As can be seen, these
enberger observer can be used to estimate the load change
sources have a positive impact on the system’s performance due
d(t) by measuring Δf (t) and ΔPm (t). To check the state
to their negative feedback control loops. However, these effects
observability of this pair, the observability matrix O =
are minor, and we can conclude that the ESD controlled by the
[C T , (CA )T , (C A 2 )T , (C A 3 )T , (C A 4 )T ] is computed,
proposed method provides almost all required power to achieve
and since it is full rank, (A , C ) is fully state observable. Now,
power balancing.
let X̂ (t) shows the estimation of the augmented state X (t).
1) Base Scenario: In this scenario, a simulation is done to
Selecting the Luenberger observer
  investigate the fundamental properties of the proposed control
˙
X̂ (t) = A X̂ (t) + B u(t) + L y(t) − C X̂ (t) , (19) structure. For the aim of comparison, two single-input-single-
output (SISO) robust controllers, designed using the quantitative
the problem is to find the observer gain L ∈ R5×2 such that feedback theory (QFT) [10] and the H∞ method [11], are
the estimation error ẽ(t) = X (t) − X̂ (t) asymptotically con- considered. For the nominal values of the system parameters
verges to zero. This aim is attained if A − LC is a Hurwitz and when the load changes ±0.1 pu at t = 0.1 s, Fig. 9(a)
matrix, i.e., all of its eigenvalues have strictly negative real parts. and (b) depict the system frequency deviation. The frequency
Since (A , C ) is observable, it is possible to find L such that nadir and the maximal RoCoF as two important indices for
all the eigenvalues of A − LC are placed in predetermined evaluating the performance of these controllers are presented in

Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on January 23,2025 at 12:09:45 UTC from IEEE Xplore. Restrictions apply.
 5988 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 39, NO. 4, JULY 2024

Fig. 10. Basic scenario: (a) Change in active power balance and estimation of
Fig. 8. Effects of FC, BESS, and FESS (a) Δf (t), (b) ΔPESD (t). its value, (b) ΔPESD (t) and ΔPm (t) in TITO control.

Fig. 11. Basic scenario: (a) ΔPESD (t), (b) ΔPm (t) in SISO control.
Fig. 9. Comparison results: (a) ΔP0 = 0.1 pu, (b) ΔP0 = −0.1 pu.

from the ESD in the short-term and instantaneous power.

TABLE II
FIRST OBJECTIVE PERFORMANCE INDEXES OF THE STUDIED MG UNDER THE The ESD only provides short-term power, and the thermal
QFT, H∞ AND THE PROPOSED TITO CONTROLLERS unit provides the active power needed to create power
balance, which is the second objective of the proposed
technique.
r As depicted in Fig. 11(a), and (b), using a robust SISO
controller, such as the QFT or the H∞ controller, the
virtual inertial control can only accomplish the frequency
Table II. Fig. 10(a) depicts the change in the active power balance regulation. To cope with uncertainties in the parameters,
resulting from a change in the local load and its estimation by large steady-state gains must be selected for these SISO
the observer of the TITO controller. Fig. 10(b) represent the controllers. Consequently, nearly all the power required to
mechanical power change and the ESD power change for the establish the power balance is supplied by the ESD over an
proposed method. The ESD power change and the mechanical extended period of time, necessitating that the ESD utilizes
power change for the QFT and H∞ controllers are shown in a primary power source, such as a PV or WT, to perpetually
Fig. 11(a) and (b), respectively. From the obtained results, the inject power into the grid.
following issues should be highlighted: 3) Uncertainties in H and D: As the actual values of the
r From Table II and Fig. 9, using the proposed TITO con- inertia constant and the damping coefficient may be different
troller improves the load-frequency characteristics, which from their nominal values, it is vital to examine the robustness
is Objective #1 of this paper. In comparison with the QFT of the proposed TITO controller against uncertainties in these
(H∞ ) controller, the proposed method reduces the fre- parameters. For a set of 20 pairs of H and D selected randomly
quency nadir and the maximum RoCoF by 10.5% (17.8%) in the ranges of ±80% of their nominal values, Fig. 12(a) shows
and 17.5% (21.4%), respectively. the system frequency deviations when the load increases by 0.1
r As seen from Fig. 10(a), the designed observer accurately pu at t = 0.1 s. The mechanical power change and the ESD
estimates the changes in the active power balance. power change are depicted in Fig. 12(b). From these figures, it
r From Fig. 10(b), using the proposed control structure, the can be concluded that the proposed control structure is robust
required active power to create the power balance is derived against parametric uncertainties.

Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on January 23,2025 at 12:09:45 UTC from IEEE Xplore. Restrictions apply.
 RAFIEE et al.: LOAD FREQUENCY CONTROL IN MICROGRIDS: A ROBUST BI-OBJECTIVE VIRTUAL DYNAMICS TECHNIQUE 5989

Fig. 14. Effects of DER trip: (a) Δf (t), (b) ΔPm (t) and ΔPESD (t).
Fig. 12. Uncertainties in H and D: (a) Δf (t), (b) ΔPm (t) and ΔPESD (t).

Fig. 13. ESD charge and discharge modes: (a) Δf (t), (b) ΔPm (t) and
ΔPESD (t).

4) ESD Charge and Discharge Modes: In this scenario, it is

assumed that an overload 0.1 pu is demanded during the time
interval t = [1, 4] s. The obtained results are depicted in Fig.
13(a) and (b). Fig. 13(a) shows that the proposed TITO con-
troller successfully maintains the frequency deviations. Unlike
conventional SISO controllers, as can be seen from Fig. 13(b),
the most important property of the proposed control structure is Fig. 15. Effects of the PV system and the wind turbine generator: (a) ΔP0 (t)
its ability to excite the thermal unit in supplying the overload in and ΔP̂0 (t), (b) ẽ(t), (c) Δf (t), and (d) ΔPESD (t) and ΔPm (t).
a proper period of time. As expected, the ESD mainly supplies
the demanded overload at the beginning of the simulation due close to the thermal units, and the control system employs local
to its fast dynamic. Nevertheless, the control structure excites information without any communication platforms.
the main unit (the thermal unit) after a while; and therefore, 5) Effects of DER Trip: In this scenario, it is assumed that
the ESD role is decreasing. When the load comes back to its the microgrid provides a steady-state local load of 12 MW (0.8
nominal value, the thermal unit starts to reduce its speed, and of nominal load). To supply this load, the thermal power plant,
the energy saved in its mechanical components flows into the wind farm, solar farm, and fuel cell system contribute 4 MW, 4
ESD that leads to charge it. In the SISO control approach, the MW, 3 MW, and 1 MW, respectively. It is presumed that half of
ESDs require a primary power source, such as a PV or WT, the wind farm’s capacity is cut off from the grid at t = 1 s and
to continually inject power into the grid. Due to operational the fuel cell system is unplugged at t = 5 s. Fig. 14(a) shows
constraints and climatic conditions, they may be located at a that the frequency deviations are effectively eliminated by the
distance from the thermal unit. In this case, the control system proposed TITO controller. Fig. 14(b) shows that this controller
requires communication links to transmit information, whereas can use efficiently the spinning reserve of the system to establish
in the proposed method, the energy required to recharge the the power balance.
ESDs is directly supplied by the thermal unit, that is provided 6) Uncertainties in ΔPLoad (t), ΔPW (t), and ΔPPV (t): In
using bidirectional converters. since there is no requirement this scenario, the system disturbance is assumed to be equal
for a primary power source, it is possible to install the ESDs to local load variations and the production range of the solar
Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on January 23,2025 at 12:09:45 UTC from IEEE Xplore. Restrictions apply.
 5990 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 39, NO. 4, JULY 2024

and wind farms. The purpose of this scenario is to verify the [12] J. Ma, Z. Song, Y. Zhang, Y. Zhao, and J. S. Thorp, “Robust stochastic
effectiveness of the suggested estimate technique for precisely stability analysis method of DFIG integration on power system consid-
ering virtual inertia control,” IEEE Trans. Power Syst., vol. 32, no. 5,
determining the power imbalance when faced with entirely ran- pp. 4069–4079, Sep. 2017.
dom disturbances and noise. Fig. 15(a) illustrates the changes in [13] T. Kerdphol, F. S. Rahman, M. Watanabe, and Y. Mitani, “Robust virtual
the power balance within this state, along with the corresponding inertia control of a low inertia microgrid considering frequency measure-
ment effects,” IEEE Access, vol. 7, pp. 57550–57560, 2019.
estimation of these changes. Fig. 15(b) illustrates the estimation [14] X. Hou, Y. Sun, X. Zhang, J. Lu, P. Wang, and J. M. Guerrero, “Improve-
error, providing verification of the appropriate operation of the ment of frequency regulation in VSG-based AC microgrid via adaptive vir-
estimator. Fig. 15(c) illustrates the variations in frequency, and tual inertia,” IEEE Trans. Power Electron., vol. 35, no. 2, pp. 1589–1602,
Feb. 2020.
Fig. 15(d) depicts the power exchange between the storage [15] F. Wang, L. Zhang, X. Feng, and H. Guo, “An adaptive control strategy
source and the thermal unit within the microgrid, with the for virtual synchronous generator,” IEEE Trans. Ind. Appl., vol. 54, no. 5,
objective of achieving power balance. pp. 5124–5133, Sep./Oct. 2018.
[16] K. Jiang, H. Su, H. Lin, K. He, H. Zeng, and Y. Che, “A practical secondary
frequency control strategy for virtual synchronous generator,” IEEE Trans.
VI. CONCLUSION Smart Grid, vol. 11, no. 3, pp. 2734–2736, May 2020.
[17] M. M. Manaz and C.-N. Lu, “Optimal switched mode control to synthesize
In this paper, by defining a bi-objective control problem and dynamic frequency response from virtual synchronous machine in islanded
introducing a two-input-two-output (TITO) control strategy, a microgrid operation,” IFAC-PapersOnLine, vol. 51, no. 28, pp. 610–615,
Jan. 2018.
novel unified solution has been proposed for the load frequency [18] N. Xu, Y. Wang, M. Li, W. Wang, N. Wang, and J. Li, “An optimal control
control problem in modern power systems. The first control method of virtual angular acceleration to improve transient response based
objective is the system’s ability to restore its frequency to the on virtual synchronous generator,” in Proc. IEEE 3rd Int. Future Energy
Electron. Conf. Asia, 2017, pp. 1559–1563.
desired value. The second objective is to design an automatic [19] K.-H. Tan, F.-J. Lin, C.-M. Shih, and C.-N. Kuo, “Intelligent control
mechanism to address the need for cooperation between the main of microgrid with virtual inertia using recurrent probabilistic wavelet
power source unit and the ESD in supplying loads. These objec- fuzzy neural network,” IEEE Trans. Power Electron., vol. 35, no. 7,
pp. 7451–7464, Jul. 2020.
tives have been robustly attained by employing the proposed [20] F. Yao, J. Zhao, X. Li, L. Mao, and K. Qu, “RBF neural network based
TITO high-gain PI controller with two important properties. virtual synchronous generator control with improved frequency stability,”
The first one is that the primary, secondary, and virtual inertia IEEE Trans. Ind. Inform., vol. 17, no. 6, pp. 4014–4024, Jun. 2021.
[21] C. Zheng, T. Dragicevic, and F. Blaabjerg, “Model predictive control based
control loops have been simultaneously designed. The second virtual inertia emulator for an islanded ac microgrid,” IEEE Trans. Ind.
property is the ability of the method to create an automatic charg- Electron., vol. 68, no. 8, pp. 7167–7177, Aug. 2021.
ing/discharging control mechanism for the ESD. According to [22] T. Kerdphol, F. S. Rahman, Y. Mitani, K. Hongesombut, and S. Küfeoğlu,
“Virtual inertia control-based model predictive control for microgrid
the obtained results, the proposed TITO control strategy has frequency stabilization considering high renewable energy integration,”
decreased the required capacity of the backup unit. Through Sustainability, vol. 9, no. 5, Mar. 2017, Art. no. 773.
some simulated scenarios, the effectiveness of the proposed [23] F. A. Inthamoussou, J. Pegueroles-Queralt, and F. D. Bianchi, “Control of
a supercapacitor energy storage system for microgrid applications,” IEEE
control strategy has been proved. Trans. Energy Convers., vol. 28, no. 3, pp. 690–697, Sep. 2013.
[24] J. Fang, H. Li, Y. Tang, and F. Blaabjerg, “Distributed power system virtual
REFERENCES inertia implemented by grid-connected power converters,” IEEE Trans.
Power Electron., vol. 33, no. 10, pp. 8488–8499, Oct. 2018.
[1] H. Bevrani, Robust Power System Frequency Control. Berlin, Germany: [25] X. Guo, Z. Lu, B. Wang, X. Sun, L. Wang, and J. M. Guerrero, “Dynamic
Springer, 2014. phasors-based modeling and stability analysis of droop-controlled invert-
[2] A. J. Wood, B. F. Wollenberg, and G. B. Sheblé, Power Generation, ers for microgrid applications,” IEEE Trans. Smart Grid, vol. 5, no. 6,
Operation, and Control. Hoboken, NJ, USA: Wiley, 2013. pp. 2980–2987, Nov. 2014.
[3] J. Machowski, Z. Lubosny, J. W. Bialek, and J. R. Bumby, Power System [26] A. Rafiee, Y. Batmani, A. Mehrizi-Sani, T. Kato, and H. Bevrani,
Dynamics: Stability and Control. Hoboken, NJ, USA: Wiley, 2020. “Zero-level control design in AC microgrids: A robust multivariable
[4] A. Ahmadi-Khatir, M. Bozorg, and R. Cherkaoui, “Probabilistic spinning approach,” IEEE Trans. Power Electron., vol. 38, no. 7, pp. 8394–8405,
reserve provision model in multi-control zone power system,” IEEE Trans. Jul. 2023.
Power Syst., vol. 28, no. 3, pp. 2819–2829, Aug. 2013. [27] Q.-C. Zhong, Power Electronics-Enabled Autonomous Power Systems:
[5] P. Kundur, Power System Stability. New York, NY, USA: CRC Press, 2007. Next Generation Smart Grids. Hoboken, NJ, USA: Wiley, 2020.
[6] S. Zhang, Y. Mishra, and M. Shahidehpour, “Fuzzy-logic based frequency [28] J. Pegueroles-Queralt, F. D. Bianchi, and O. Gomis-Bellmunt, “A power
controller for wind farms augmented with energy storage systems,” IEEE smoothing system based on supercapacitors for renewable distributed
Trans. Power Syst., vol. 31, no. 2, pp. 1595–1603, Mar. 2016. generation,” IEEE Trans. Ind. Electron., vol. 62, no. 1, pp. 343–350,
[7] J. Fang, H. Li, Y. Tang, and F. Blaabjerg, “On the inertia of future Jan. 2015.
more-electronics power systems,” IEEE Trans. Emerg. Sel. Topics Power [29] A. Yazdani and R. Iravani, Voltage-Sourced Converters in Power Systems:
Electron., vol. 7, no. 4, pp. 2130–2146, Dec. 2019. Modeling, Control, and Applications. Hoboken, NJ, USA: Wiley, 2010.
[8] X. Meng, J. Liu, and Z. Liu, “A generalized droop control for grid- [30] S. Skogestad and I. Postlethwaite, Multivariable Feedback Control: Anal-
supporting inverter based on comparison between traditional droop control ysis and Design. New York: Wiley, vol. 2, 2007.
and virtual synchronous generator control,” IEEE Trans. Power Electron., [31] Y. Wan, T. Keviczky, and M. Verhaegen, “Fault estimation filter design
vol. 34, no. 6, pp. 5416–5438, Jun. 2019. with guaranteed stability using Markov parameters,” IEEE Trans. Autom.
[9] J. Liu, Y. Miura, H. Bevrani, and T. Ise, “A unified modeling method of Control, vol. 63, no. 4, pp. 1132–1139, Apr. 2018.
virtual synchronous generator for multi-operation-mode analyses,” IEEE [32] R. C. Dorf and R. H. Bishop, Modern Control Systems. London, U.K.:
Trans. Emerg. Sel. Topics Power Electron., vol. 9, no. 2, pp. 2394–2409, Pearson, 2011.
Apr. 2021. [33] D. B. Ridgely, S. S. Banda, and J. J. D’Azzo, “Decoupling of high-gain
[10] A. Rafiee, Y. Batmani, F. Ahmadi, and H. Bevrani, “Robust load-frequency multivariable tracking systems,” J. Guid., Control, Dyn., vol. 8, no. 1,
control in islanded microgrids: Virtual synchronous generator concept and pp. 44–49, May 2012.
quantitative feedback theory,” IEEE Trans. Power Syst., vol. 36, no. 6, [34] A. Rafiee, Y. Batmani, H. Bevrani, and T. Kato, “Robust MIMO controller
pp. 5408–5416, Nov. 2021. design for VSC-based microgrids: Sequential loop closing concept and
[11] T. Kerdphol, F. S. Rahman, M. Watanabe, and Y. Mitani, Virtual Inertia quantitative feedback theory,” IEEE Trans. Smart Grid, vol. 13, no. 1,
Synthesis and Control. Berlin, Germany: Springer, 2021. pp. 129–138, Jan. 2022.

Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on January 23,2025 at 12:09:45 UTC from IEEE Xplore. Restrictions apply.