You are on page 1of 12

1076 JOURNAL OF MODERN POWER SYSTEMS AND CLEAN ENERGY, VOL. 9, NO.

5, September 2021

Virtual Inertia Estimation Method of DFIG-based


Wind Farm with Additional Frequency Control
Pengwei Chen, Member, IEEE, Chenchen Qi, and Xin Chen, Member, IEEE

Abstract—
—With the increasing penetration of wind power, us‐ [7]. Against this background and demand, various methodolo‐
ing wind turbines to participate in the frequency regulation to gies have been explored in the literature or practical applica‐
support power system has become a clear consensus. To accu‐
tions, which fall into three main categories: rotor speed con‐
rately quantify the inertia provided by the doubly-fed induction
generator (DFIG) based wind farm, the frequency response trol, pitch angle control, and inertia control [8], [9]. In the
model of DFIG with additional frequency control is established, rotor speed control, the operating point is set to be retarded
and then by using Routh approximation, the explicit expression or advanced from the optimum operation conditions to
of the virtual moment of inertia is derived for the DFIG grid- achieve the active power reserve, which is usually employed
connected system. To further enhance the availability of the ex‐ at low wind speed [10]. The pitch angle control is also a de-
pression, an estimation method is proposed based on the matrix
pencil method and the least squares algorithm for estimating loading technique, but its response speed participating in pri‐
the virtual moment of inertia provided by the wind farm. Final‐ mary frequency regulation is still limited because of the me‐
ly, numerical results tested by a DFIG grid-connected system chanical constraints [11]. By adding an auxiliary signal to
and a modified IEEE 30-bus system verify the derived expres‐ the active power reference value, the inertia control can pro‐
sion of the virtual moment of inertia and the proposed estima‐ vide temporary active power support by releasing active
tion method.
power reserve under a variable wind speed condition
Index Terms——Doubly-fed induction generator (DFIG), addi‐ [10]-[12].
tional frequency control, virtual moment of inertia, matrix pen‐ The conventional inertia in the power system is generally
cil method, least squares algorithm.
defined as the total amount of kinetic energy stored in all ro‐
tating grid-connected generators and motors. Although the
virtual inertia of DFIG is also provided by releasing the re‐
I. INTRODUCTION
served active power, the inertia is time-varying and hidden

D UE to cleanliness, safety, and adequacy, wind power


generation has experienced rapid development in the
past decades [1]-[3]. Doubly-fed induction generator (DFIG)
in the complex control process. From the perspective of sys‐
tem planning and operation, the quantification of virtual iner‐
tia is valuable.
is one of the most commercialized wind turbines (WTs) with In terms of energy, an equivalent time constant of virtual
good performance in integrating intermittent wind power, inertia is defined for DFIG in [13], consisting of actual
but its converter control makes the rotor speed decoupled changed kinetic energy, energy loss, and captured wind pow‐
with the system frequency [3]. With the continuous growth er. Such a time constant of virtual inertia is decided by the
of DFIG-based wind farms, the total equivalent rotary inertia integration of output active power and the control time of ro‐
of the system is passively decreased, and the high penetra‐ tor speed, which is hard to calculate accurately when the pa‐
tion itself also generates additional disturbance because of rameters are unknown. In [14] and [15], based on the as‐
the time-varying output. Meanwhile, these factors will chal‐ sumption about the effective inertia of rotating components,
lenge the system frequency stability [4], [5]. the equivalent inertia constant is acquired from the mechani‐
To reduce the negative impact of high penetration of wind cal energy captured by WTs. This type of method ignores
power, using WTs to participate in the frequency regulation the impact of DFIG control on the inertia, and thus the accu‐
has become a clear consensus among many countries [6], racy of the derived equivalent time constant cannot be guar‐
anteed. Taking the additional frequency control into consider‐
Manuscript received: December 28, 2020; accepted: April 22, 2021. Date of ation, a definition of virtual inertia constant based on the ra‐
CrossCheck: April 22, 2021. Date of online publication: June 7, 2021. tio between the changes of system angular frequency and ro‐
This work was supported in part by the National Science Foundation of Chi‐
na (No. 51877015) and the Science and Technology Foundation of State Grid
tor angular velocity is also defined in [16], but the ratio is
Corporation of China (No. SGTYHT/19-JS-215). difficult to determine as the dynamics of the converter has
This article is distributed under the terms of the Creative Commons Attribu‐ to be considered. Different from focusing on the inertia con‐
tion 4.0 International License (http://creativecommons.org/licenses/by/4.0/).
P. Chen (corresponding author), C. Qi, and X. Chen are with the Jiangsu Key
stant that relies on the conversion of the moment of inertia,
Laboratory of New Energy Generation and Power Conversion, College of Auto‐ rated capacity, and rated rotor angular velocity, an expres‐
mation Engineering, Nanjing University of Aeronautics and Astronautics, Nan‐ sion of the equivalent moment of inertia described by trans‐
jing, China (e-mail: chenpw2019@nuaa.edu.cn; qichenchen@nuaa.edu.cn; chen.
xin@nuaa.edu.cn). fer function is derived in [17] by using the change of kinetic
DOI: 10.35833/MPCE.2020.000908 energy. In essence, such a moment of inertia is still a kind
CHEN et al.: VIRTUAL INERTIA ESTIMATION METHOD OF DFIG-BASED WIND FARM WITH ADDITIONAL FREQUENCY CONTROL 1077

of dynamic process, which is significantly different from the 3) From the perspective of the application, an estimation
accepted deterministic definition. If the inertia is measured method based on the matrix pencil method and the least
in this way, the function intuitively indicating the system in‐ squares algorithm is proposed for identifying the frequency
ertia will not exist. response characteristics of the wind farm with additional fre‐
As for a wind farm, its size may vary from a small num‐ quency control. Combining the derived expression and the
ber to several hundred WTs covering an extensive area. No frequency response, the basic framework to efficiently esti‐
matter what description or definition is adopted, the total vir‐ mate the virtual moment of inertia is obtained, and thus the
tual inertia can be accumulated by that of each WT in theory complex processing of the parametric derivation and re‐
[18], [19]. However, each WT has its own operating point, sponse integration for the wind farm is avoided, which
of which operation data such as rotor velocity are usually makes the real-time application of virtual inertia estimation
difficult to be observed completely. Such an accumulation possible.
processing is inefficient and even inaccessible. Another ap‐ The remainder of this paper is organized as follows. Sec‐
proach to obtain the total virtual inertia is to utilize the estab‐ tion II introduces the inertia of the power system with DFIG
lished multi-machine dynamic equivalent model of wind and the principle of approximate decoupling between DFIG
farm [20], but this two-stage processing essentially transfers and system angular frequency. Section III summarizes the
the workload of inertia accumulation to multi-machine repre‐ frequency response modeling process of DFIG with addition‐
sentation as well as introduces unpredictable errors. Al‐ al frequency control as well as the basic expression of the
though the equivalent inertia of an entire system can be virtual moment of inertia. Section IV presents the identifica‐
roughly estimated based on the power imbalance and change tion method of the frequency response characteristic on a
rate of frequency [21], or other monitoring data, the part pro‐ wind farm and the main process of calculating its virtual mo‐
vided by the wind farm still remains to be determined, and ment of inertia. The case studies are discussed in Section V,
the estimation also relies on the integrated system with the followed by Section VI, which concludes the paper.
wind farm. The direct estimation of the virtual moment of in‐
ertia is more valuable for the planning and operation of the II. INERTIA OF POWER SYSTEM WITH DFIG
wind farm itself, which is efficient to determine the impact
on the characteristics of system frequency no matter what A. Inertia of Synchronous Generator
changes occur in the system. Regarding the estimation for Inertia is an inherent property of any physical object. The
the virtual inertia of devices, some techniques such as distur‐ inertia of a synchronous generator usually refers to the rota‐
bance observer and recursive least squares algorithm have tional inertia, which reflects the degree of resistance to the
been explored for the grid-connected inverter with virtual change of rotor motion state. To describe the strength of re‐
synchronous generator control or droop control, the drive sistance, the moment of inertia that determines the torque
system based on the permanent magnet synchronous ma‐ needed for a desired angular acceleration about a rotational
chine, and so on [22]-[24]. It is noted that the dynamics of axis is frequently employed, which is defined as:

the wind farm with additional frequency control are more
complicated than that of a single inverter, and the definition J g = r 2 dm (1)
of virtual inertia supported by the wind farm is not clear due where r and m are the rotation radius and the mass of the ro‐
to its non-synchronous nature, which also results in the inap‐ tor shaft, respectively.
plicability of the existing estimation techniques. The synchronous generator can be considered as a rotating
To efficiently analyze and reveal the impact of high pene‐ rigid body. According to the electric machine theory, the rela‐
tration of wind power on the characteristics of system fre‐ tionship between the mechanical angular velocity ω mg and
quency, a method of estimating the virtual inertia supported the torque exerting on the rotor can be expressed as:
by the additional frequency control of the DFIG-based wind
dω mg
farm is investigated in this paper. The main contributions are Jg = T mg - T eg - D g ω mg (2)
as follows. dt
1) Taking the inertia of power system as a reference, the where T mg, T eg, and D g are the mechanical torque, electro‐
response characteristic of DFIG with conventional vector magnetic torque, and damping coefficient of a generator, re‐
control is investigated, and the principle of approximate de‐ spectively.
coupling between DFIG and system frequency is further dis‐ Substitute T mg = P mg ω mg T eg = P eg ω mg and ω mg = ω 1 n pg
cussed. into (2) with P mg, P eg, ω 1, and n pg being the mechanical pow‐
2) To demonstrate the origin of the virtual moment of iner‐ er, electromagnetic power, system angular frequency, and
tia directly, the frequency response model of DFIG with ad‐ pole pairs, respectively. By linearization, the motion equa‐
ditional frequency control is established, and then by using tion concerning the system angular frequency can be de‐
Routh approximation, the expression of the virtual moment scribed as:
of inertia is derived. The derivation process is based on the J g ω 10 dDω 1 D g ω 10
DFIG grid-connected system with an equivalent generator, = DP mg - DP eg - Dω 1 (3)
n pg
2
dt n 2pg
but the final derived expression is independent of the back‐
ground system, which essentially completes the transforma‐ where ω 10 is the steady-state value of system angular fre‐
tion of DFIG from non-synchronous to synchronous and de‐ quency.
termines the basic idea of inertia estimation for a wind farm. Thus, the transfer function between the system angular fre‐
1078 JOURNAL OF MODERN POWER SYSTEMS AND CLEAN ENERGY, VOL. 9, NO. 5, September 2021

quency and active power can be given by: maximum power point tracking, the maximum power cap‐
Dω 1 n 2pg tured by the WT can be represented as P opt = k w ω 3md N g,
= (4) where N g is the transformation ratio of the gearbox, k w =
DP mg - DP eg J g ω 10 s + D g ω 10
0.5ρS w (R w λ opt )3 C p max, and R w, λ opt, and C p max are the wind
It can be observed that the change of the angular frequen‐
cy caused by the active power imbalance is a first-order pro‐ wheel radius, optimal tip speed ratio, and maximum value of
cess, where the moment of inertia and the damping coeffi‐ coefficient of power, respectively [14]. To retain the margin
cient jointly determine the dynamic. In particular, the higher of active power, the reference value P s* can be set by:
the moment of inertia is, the slower the change rate of sys‐ kw
P s* = kω 3md k < (7)
tem angular frequency will be. Ng
Remark 1: in term of a multi-machine system, the rotor When the copper loss in both stator and rotor is neglect‐
speed of each synchronous generator is coupled to the sys‐ ed, we can obtain the approximation of the grid-connected
tem angular frequency, and thus the total moment of inertia power of DFIG as:
in such a system can be accumulated by all participating ma‐
n pd ω md
chines, where it needs to transform all the moments of iner‐ Pw = Ps (8)
tia to the same pole pairs. From this perspective, the esti‐ ω1
mate to the moment of inertia of each machine that actively Assume that the background system can be represented by
or passively participates in frequency regulation is the basis an equivalent synchronous generator and the DFIG is con‐
of evaluating system inertia. nected to the generator. Combining the model of synchro‐
B. Inertia of DFIG nous generator, the angular frequency dynamic of such a
DFIG grid-connected system described by small-signal equa‐
Using the motor convention, the electromechanical tran‐ tions is shown in Fig. 1.
sient process of DFIG can be expressed as [25]:
DFIG
ì dω md
ïJ d = T ed - T md - D d ω md npd/ω10 ∆Ps *
1 ∆Ps 3kω 2 ∆ωmd
+ 2
npdPs0 /ω10
ï dt Jds+Dd+β αs+1 md0 +
+ Jds+Dd+β
ï
ïT = - n pd P
í ed ω1 s (5) npdPs0 + + + npdωmd0
+ npdωmd0Ps0
ï ω10 ω10
+
 2
ï ω10
ïT = - ρS w C p v
3 ∆PW
ï md 2ω md ∆PL Motion equation
î ∆ω1ref Regulator-throttle ∆Pmg
 + 2
npg ∆ω1
where ω md is the angular velocity of rotor; J d and D d are the + + actuator + + Jgω10s+Dgω10

total moment of inertia and damping coefficient converted
onto the motor axis, respectively; T ed and T md are the electro‐ Fig. 1. Block diagram of angular frequency dynamic in DFIG grid-con‐
nected system.
magnetic torque and mechanical torque, respectively; P s is
the active power of the stator; and ρ, S w, v, C p, and n pd are
the air density, swept area of WT blades, wind speed, coeffi‐ Since the above DFIG model only involves the electrome‐
cient of power, and pole pairs of DFIG, respectively. chanical transient process, its small-signal equations can be
In the vector control strategy of DFIG, the control of the easily derived by linearizing (5)-(8), which will not be listed
rotor-side converter can be simplified as a first-order dynam‐ in this paper due to the limited space. It is worth noting that
ic process by Routh approximation. The time constant is the expression of T md is nonlinear, and after the linearization,
shown in (6), but it is worth noting that we can also obtain we can obtain the coefficient β with respect to Dω md, namely
different expressions of the time constant by using other or‐ DT md = βDω md, as:
der-reduction methods. To make the statement easier to fol‐ ρS w Cv 30 ω md0 - 2P md0
low and more general, the detailed derivation on this part is β=- (9)
2ω 2md0
presented in Appendix A.
where v 0 is the steady-state wind speed; ω md0 is the steady-
ì Ps 1 state angular velocity of rotor; P md0 is the steady-state value
ï *=
ï sP αs +1 of the mechanical power; and C is obtained by linearizing
í (6) the fit expression of C p and thus making DC p = CDω md.
ï 3KL m u sd0 k p1 + 2L s K
ïα = 3KL m u sd0 k i1 As depicted in Fig. 1, if the control loop is reduced to the
î unity gain, namely DP s* = DP s, the relationship between Dω md
where P s and P s* are the active power at the stator side and and Dω 1 can be formed as:
its reference value, respectively; K is the gain of pulse width n pd P s0
modulation; L m is the mutual inductance between the coaxial Dω md = Dω 1 (10)
equivalent stator and rotor winding; L s is equivalent to the J d ω 210 s + 3kn pd ω 10 ω 2md0 + βω 210 + D d ω 210
self-inductance of the stator winding; u sd0 is the steady-state where P s0 is the steady-state active power at the stator side.
d-axis voltage of stator; and k p1 and k i1 are the proportional We can observe that the gain of (10) is smaller than 1 so
and integral parameters of the outer control loop, respectively. that the changing range of Dω md is much smaller than that of
Given that DFIG is usually controlled by the scheme of Dω 1 under disturbance, and an assumption of Dω md » 0 can
CHEN et al.: VIRTUAL INERTIA ESTIMATION METHOD OF DFIG-BASED WIND FARM WITH ADDITIONAL FREQUENCY CONTROL 1079

be easily obtained, which means the rotor speed is decou‐ lar frequency.
pled from the system angular frequency under the control of Thus, the reference value of active power at the stator
rotor-side converter. Considering the above inference, the lin‐ side is modified as:
earization of (8) can be expressed as: P s* = kω 3md - P id (13)
n pd ω 10 ω md0 DP s + n pd ω 10 P s0 Dω md - n pd ω md P s0 Dω 1 Combined with the block diagram shown in Fig. 1, the
DP w = =
ω 210 system angular frequency dynamic considering the additional
3kn pd ω 3md0 frequency control can be updated as shown in Fig. 2. The ad‐
Dω md » 0 (11) ditional frequency control promotes the change of reference
ω 10
value of active power at the stator side when the frequency
Remark 2: under the vector control strategy, the DFIG disturbance occurs, thus constituting the response to the fre‐
grid-connected system is equivalent to a separate constant quency variety. Such progress can rapidly release the kinetic
power source. Although such a source is usually affected by energy of the rotor of DFIG to compensate for the imbal‐
system voltage and wind speed, the kinetic energy of the ro‐ ance of active power.
tor is insensible when the frequency disturbance occurs, To determine the virtual moment of inertia provided by
which means the system inertia is mainly determined by the the above process, the frequency response is necessary. Ac‐
synchronous generators. DFIG itself provides little inertia cording to the simplified block diagram shown in Fig. 2, the
support, and such a characteristic will not change significant‐ relationship between Dω 1 and DP w can be obtained as:
ly with the increase of DFIG penetration. g1 + g2 + g3 + g4 g0
DP w = Dω 1 (14)
g0
III. VIRTUAL MOMENT OF INERTIA SUPPORTED BY DFIG
When DFIG is configured with additional frequency con‐ ì 3kω 2md0 n pd
trol, the grid-connected power is directly affected by the sys‐ ïg 0 = 1+
ï ω 10 (αs + 1)(J d s + D d + β)
tem frequency, which makes DFIG actively participating in ïï
ïï 3kω 3md0 n 2pd P s0
the primary frequency regulation possible. To determine the ïïg 1 = 3
virtual moment of inertia supported by DFIG, the frequency ïï ω 10 (αs + 1)(J d s + D d + β)
response model of DFIG with additional frequency control is ï
ï n 2pd P s02 (αs + 1)
established below, and the method to estimate the virtual mo‐ íg 2 = - 3 (15)
ï ω 10 (αs + 1)(J d s + D d + β)
ment of inertia is also proposed. ï
ïï n pd ω md0 (k df s + k pf )
A. Frequency Response of DFIG ïïg 3 = - (1+ T s)(αs + 1)ω
ïï l 10
The additional frequency control considered in this paper ïï
consists of two loops, i.e., the virtual inertia control and the n ω
ïg = - pd md0 s0 P
ï 4
droop control [9]-[11], [13], as shown in Fig. 2. î ω 210
As the time constant T l in the low-pass filter is small, if
DFIG Additional frequency control
npd/ω10
the corresponding part is neglected, then (14) can be simpli‐
kdf s+kpf
Jds+Dd+β ∆Pid 1+Tl s
fied as:
DP w b 2 s2 + b 1 s + b 0
= G pω (s) = (16)
npdPs0 1

2
3kωmd0
∆ωmd + 2
npdPs0 /ω10 Dω 1 a 2 s2 + a 1 s + a 0
+ +
ω10 ∆Ps αs+1 + + Jds+Dd+β
ìa 2 = J d α
+ npdωmd0Ps0 ï
+ + + npdωmd0
+ ïa 1 = αD d + αβ + J d
ω10  2
ω10 ï
∆PW ïa 0 = D d + β
∆PL Motion equation ï
 + 2
npg ï k df J d n pd ω md0 P s0 n pd ω md0 P s0 a 2
∆ω1ref Regulator-throttle ∆Pmg ∆ω1 ïb 2 = - -
+ +

actuator + +
Jgω10s+Dgω10 ï ω 10 ω 210
ï
ï n pd ω md0 (k df D d + k df β + k pf J d )
Fig. 2. Block diagram of system angular frequency dynamic in DFIG grid- ïb 1 = - ω 10
-
connected system with additional frequency control. í (17)
ï
ï n ω
pd md0 s0 1P a n 2
P
pd s0
2
α
ï -
When the system frequency disturbance occurs, the refer‐ ï ω 210 ω 310
ence value of additional active power generated by the addi‐ ï
ï 3kω 2md0 n pd + n pd ω md0 k pf a 0
tional frequency controller can be given by: ïb 0 = - -
ï ω 10
k df sω 1
P id = + k (ω - ω 1ref ) (12) ï n pd ω md0 P s0 a 0 3kω 3md0 n 2pd P s0 + n 2pd P s02
T l s + 1 pf 1 ï -
ï
where T l is the time constant of a low-pass filter; k df and k pf î ω 210 ω 310
are the virtual inertia coefficient and droop coefficient, re‐ Through the above simplification, the frequency response
spectively; and ω 1ref is the setting value of the system angu‐ of DFIG with additional frequency control is reduced to a
1080 JOURNAL OF MODERN POWER SYSTEMS AND CLEAN ENERGY, VOL. 9, NO. 5, September 2021

second-order transfer function, as shown in Fig. 3(a). 1


Ĝ ω (s) =
c1 c0 (21)
s+
DFIG ∆Pw b2s2+b1s+b0 a0 a0
a2s2+a1s+a0
If Ĝ ω (s) is used to replace G ω (s), the angular frequency
∆PL Motion equation dynamic shown in Fig. 3(a) can be further simplified, as
∆ω1ref Regulator-throttle ∆Pmg
 + 2
npg ∆ω1 shown in Fig. 3(b). Obviously, c 1 a 0 contains the total mo‐
+ + actuator + + Jgω10s+Dgω10 ment of inertia J total from both generator and DFIG, and we

can easily obtain:
(a)
D g a 1 b 1 n 2pg
∆PL Motion equation J total = J g + - (22)
 2
npg ∆ω1
a0 a 0 ω 10
∆ω1ref Regulator-throttle ∆Pmg
+ + actuator + + (Jeq+Jg)ω10s+Dgω10 More specifically, J total is made up of three terms, where

the first term is the inherent moment of inertia provided by
(b)
the synchronous generator. The second term is related to the
Fig. 3. Simplified block diagram of angular frequency dynamic in DFIG damping coefficient in both generator and DFIG, which is
grid-connected system with additional frequency control. (a) Original block
diagram. (b) Simplified block diagram with effect of DFIG represented by
much smaller than J g and can be ignored. Therefore, the vir‐
virtual moment of inertia. tual moment of inertia J eq supported by DFIG can be given
by:
b 1 n 2pg
B. Estimation for Virtual Moment of Inertia J eq » - (23)
a 0 ω 10
Combining the grid-connected power of DFIG and the mo‐
tion equation of synchronous generator, the response of the Remark 3: the virtual moment of inertia obtained by the
angular frequency dynamic to the load variation can be ob‐ model order reduction is almost independent of the back‐
tained as: ground system, and it mainly relies on the modeling of the
frequency response G pω (s). For other types of additional fre‐
J g ω 10 dDω 1 D g ω 10
= DP mg - DP L + DP w - Dω 1 (18) quency control schemes, the expression of J eq can also be ob‐
n pg
2
dt n 2pg tained in the same way, and there are only some differences
where DP L is the load disturbance. in coefficients a 0 and b 1. Although we can rigorously derive
Substituting (16) into (18), the transfer function between the exact expression of a single DFIG with all parameters
the power imbalance and the system angular frequency can known, such a modeling process is inefficient and even inac‐
be expressed as: cessible for a wind farm due to its size. Therefore, the effi‐
cient estimate to the frequency response of the wind farm
Dω 1 a 2 s2 + a 1 s + a 0 with additional frequency control becomes the key point to
= G ω (s) = (19)
DP mg - DP L c 3 s3 + c 2 s2 + c 1 s + c 0 the practical application of the proposed estimation method.

ì D g ω 10 a 0 IV. ESTIMATION METHOD FOR VIRTUAL MOMENT OF


ïc 0 = - b0
ï n 2pg INERTIA
ïï
ïï J g ω 10 a 0 D g ω 10 a 1 For a wind farm, its complete frequency response is made
ïc 1 = n 2
+
n 2
- b1
up of the characteristics of all WTs. To avoid the difficult in‐
ï pg pg
í (20) tegration of theoretical expressions, an estimation method us‐
ï J g ω 10 a 1 D g ω 10 a 2
ïc 2 = + - b2 ing the simulation data or actual monitoring data as an alter‐
ïï n 2pg n 2pg native is introduced below.
ïï
ïc = J g ω 10 a 2 A. Data Processing
ï 3 n 2pg
î The wind farm is a typical system based on power elec‐
It can be observed that G ω (s) is a high-order transfer func‐ tronics, which determines the presence of both broadband
tion, which contains the electromechanical transient process harmonics and disturbances in the sampling data of active
of synchronous generator and the active regulation process power. Different from the low-frequency oscillation related
of DFIG. The virtual moment of inertia provided by the addi‐ to system inertia, harmonics can be regarded as a kind of da‐
ta noise, and thus the sampling data sequence of grid-con‐
tional frequency control of DFIG is hidden in such a com‐
nected power can be formulated as:
plex transfer function.
y(k) = x(k) + σ(k) » ∑p h z hk + σ(k) k = 12N
M
To directly demonstrate the virtual moment of inertia just (24)
like (4), the model order reduction is employed to obtain a h= 1

first-order dynamic process. To ensure the approximate accu‐ where x(k) is the real data of the grid-connected power; σ(k)
jθ (α + j2πf )Dt
racy in the low-frequency band, we choose the Routh ap‐ is the data noise; p h = A h e , z h = e
h h h
, and A h, θ h, α h, and
proximation to implement model order reduction [26], and f h are the amplitude, initial phase, damping factor, and angu‐
the reduced result is: lar frequency of component h, respectively; and Dt, M, and
CHEN et al.: VIRTUAL INERTIA ESTIMATION METHOD OF DFIG-BASED WIND FARM WITH ADDITIONAL FREQUENCY CONTROL 1081

N are the sample interval, number of dominant low-frequen‐ B. Estimation Process of Virtual Moment of Inertia
cy components, and number of sample points, respectively. The sampling data sequence is discrete so that there is a
To ensure the identification accuracy of the moment of in‐ gap between it and the s-domain model shown in (16). Thus,
ertia, the data noise should be separated from the dominant the backward difference formula s = (1- z -1 ) Dt is used to re‐
low-frequency components. Therefore, the matrix pencil alize the z-transformation. Then, we can obtain the transfer
method [27], [28], as a signal reconstruction algorithm with function in the z-domain as:
good performance, is employed, of which the specific usage
n2 z2 + n1 z + n0
is as follows. G pω (z) = 2 (31)
For the sampling data sequence, the Hankel matrix Y is z + m1 z + m0
constructed firstly in the form of where m 1 m 0 n 2 n 1 n 0 are the parameters that need to be
é y(1) y(2)  y(L + 1) ù identified.
ê ú Once the G pω (z) is obtained, G pω (s) is automatically deter‐
ê y(2) y(3)  y(L + 2)ú
Y = êê úú (25) mined. Through the above transformation, the detailed esti‐
ê    ú mation process based on the least squares algorithm can be
ê ú
ëy(N - L) y(N - L + 1)  y(N) û summarized in Fig. 4, which contains 5 steps.
where L is the pencil parameter, which is usually chosen Start
from N 3 to N 2 for efficient noise filtering.
Data collection Inertia estimation
Then, the singular-value decomposition is carried out to Input sequences and use least
Set sample parameters and
achieve denoising. Considering the singular value σ h of Y as construct sequences squares algorithm to calculate θ
M, thus we can obtain:
σh σh + 1 Normalize sequences and Obtain Gpω(z) and transfer
< 10-γ £ (26) judge sequence state Gpω(z) to Gpω(s)
σ max σ max
Select and reconstruct sequences Based on Gpω(s), estimate
where γ is the threshold value; and σ max is the maximum sin‐ by using matrix pencil method moment of inertia of wind farm Jeq
gular value. The ratio smaller than 10-γ is considered as the
noise-singular value. N Is selection Y End
After the number of dominant components M is deter‐ finished?
mined, we can use the filtered right-unitary matrix V that
Fig. 4. Flow chart of estimation process based on least squares algorithm.
contains only M dominant right-singular vectors to construct
the matrix pair {V 1 V 2}, which satisfies
V 1H +V 2H - z h I = 0 (27) Step 1: data preparation. Set the sampling interval Dt and
sampling period T, then the length of all data sequences is
where the superscripts“+”and“H”represent the operations
obtained as N = T Dt. The simulations or monitoring data
of Moore-Penrose pseudoinverse and conjugate transpose, re‐ are used to construct the average wind speed sequence u v,
spectively; V 1 is obtained by removing the last row of V; V 2 system angular frequency sequence u ω, and grid-connected
is obtained by removing the first row of V; z h is the general‐ power sequence y, where u v is the mean result of all moni‐
ized eigenvalue; and I is the unit matrix. toring points at each sampling interval.
Once z h and M are known, the residues p h in (24) can be Step 2: sequence state judgment. Let x represent the nor‐
solved from the following least squares problem. malized result of u v or u ω, if "x(k)k = 12N, meets
é y(1) ù é 1 1  1 ù é p1 ù (x(k) - x̂ )2 £ d max, then x can be considered as the stationary
ê ú ê ê ú
êê y(2) úú ê z 1 z 2  z M úú ê p 2 ú sequence denoted by xˉ; otherwise, x is the non-stationary se‐
êê ú = ê ú (28)
 ú êê    ú êê  úú quence denoted by x͂ . x̂ and d max are the average of the data
ú
ê ú
ëy(N)û ëz 1
N -1
z 2N - 1  z MN - 1 û êëp M úû sequence and the threshold value of judgment, respectively.
Step 3: sequence selection. If both u v and u ω can be classi‐
Thus, the data sequence of active power can be recon‐ fied as the state group {uˉ v u͂ ω}, the matrix pencil method is
structed as: used to reconstruct the corresponding sequence of grid-con‐
y(t) = ∑2A h e h cos(2πf h t + θ h ) t = 0DtNDt
M
α t nected power and the low-frequency component y p is ob‐
(29)
h= 1 tained, and then go to Step 4; otherwise, return to Step 1.
Step 4: frequency response identification. Let u͂ pω - u pω0
ìA h = | p h | and y p - y p0 be the input and output sequences, where u pω0
ï
ïï and y p0 are the initial states of u͂ pω and y p, the accuracy of pa‐
ïïθ = tan -1 Im( p h )
ïï h Re( p h ) rameter vector θ =[m 1 m 0 n 2 n 1 n 0 ]T can be measured by the
ï following error sum of squares [29].
ïα = ln | z h |
í (30)
J = ∑(y(k) - φ k θ)2
N

ïï h Dt (32)
ïï k=1
ïï 1 Im(z h ) where φ k =[-y p (k -1)+y p0 -y p (k -2)+y p0 u͂ pω (k)-u͂ pω0 u͂ pω (k -
ïf h = 2πDt tan Re(z )
-1

î h 1) - u͂ pω0 u͂ pω (k - 2) - u͂ pω0 ].


1082 JOURNAL OF MODERN POWER SYSTEMS AND CLEAN ENERGY, VOL. 9, NO. 5, September 2021

To minimize the error, θ should satisfy ment of inertia and the rated capacity of the generator are
¶J set to be 6500 kg ⋅ m2 and 100 MVA, respectively, and the
= -Φ T y + Φ T Φθ = 0 (33) load is 25 MW. The frequency disturbance is set by putting
¶θ
into or reducing the load of the background system.
where y =[y(1)y(1)y(N)]; and Φ T =[φ T1 φ T2 φ TN ]. If Φ is
Setting the frequency disturbance by changing the load (4
full rank, then θ can be solved by the least squares algo‐
MW) in the background system, the dynamic responses of
rithm:
angular frequency in the DFIG grid-connected system with
θ = (Φ T Φ)-1 Φ T y (34) or without the additional frequency control are shown in
Step 5: moment of inertia estimation. When G pω (z) is ob‐ Fig. 5. To illustrate the DFIG without the additional frequen‐
tained, G pω (z) is transfered to G pω (s) using the principle as: cy control participating in the frequency regulation, the

{
DFIG is removed from the grid-connected system and only
G pω (s) = G pω (z)
the background system is reserved (labeled as DFIG re‐
1 (35) moved). When the DFIG is not configured with the addition‐
z=
1- sDt al frequency control, its frequency response to load distur‐
Then, according to the expression in (23), the moment of bance is a close resemblance to that of DFIG removed,
inertia supported by a wind farm can be finally estimated. which demonstrates the system angular frequency is mostly
Remark 4: under the extreme condition, if no sequence that dependent on the inertia of the equivalent synchronous gener‐
meets the selection criteria, i.e., all u v and u ω are classified as ator. After the additional frequency control is added, DFIG
the state group {u͂ v u͂ ω}, the component of grid-connected pow‐ can release the kinetic energy of the DFIG rotor and even
er corresponding to the wind speed disturbance can be re‐ potential unused wind energy to provide the system with in‐
moved, and then the proposed method can be used to estimate ertia support, and thus its angular velocity of rotor and grid-
the virtual moment of inertia for a DFIG-based wind farm. connected power are no longer stable.
DFIG removed
315.0 314.5 DFIG with additional
V. CASE STUDIES DFIG without additional
frequency control

ω1 rad·s-1)
ω1 rad·s-1)

314.5 frequency control


To illustrate the performance of the proposed method, a 314.0 DFIG without additional
DFIG grid-connected system and a modified IEEE 30-bus 314.0 frequency control
system are both used. The sampling data used to estimate 313.5 313.5
0 2 4 6 8 10 0 2 4 6 8 10
the virtual moment of inertia are captured by the simulation t (s) t (s)
using MATLAB/Simulink on an Intel Core i7-9700 3.00 (a) (b)
GHz machine. 116 DFIG with additional 2.5
ωmd rad·s-1)

frequency control DFIG with additional

Pw (MW)
A. DFIG Grid-connected System 115 2.0 frequency control

As the virtual moment of inertia provided by a single 114 DFIG without additional 1.5 DFIG without
frequency control additional frequency control
DFIG can be deduced directly, the DFIG grid-connected sys‐ 113 1.0
0 2 4 6 8 10 0 2 4 6 8 10
tem is employed to facilitate the accuracy comparison and a t (s) t (s)
clear demonstration, of which simulation parameters are giv‐ (c) (d)
en in Table I. Fig. 5. Dynamic responses of DFIG grid-connected system with or with‐
out additional frequency control. (a) System angular frequency with DFIG
TABLE I removed. (b) System angular frequency. (c) Angular velocity of rotor. (d)
PARAMETERS OF DFIG GRID-CONNECTED SYSTEM Grid-connected power.

Parameter Value
Given that all parameters of DFIG are known, the value
Rated capacity 2 MVA
of the virtual moment of inertia supported by the DFIG with
Rated voltage of stator 575 V
additional frequency control can be directly obtained by
Total moment of inertia Jd 580 kg∙m2 (23). For k pf = k df = 3´ 105, J eq=1143.23 kg ⋅ m2, and for k pf =
Resistance of stator Rs 23 mΩ k df = 2 ´ 105, J eq=758.25 kg ⋅ m2. To demonstrate the accuracy
Resistance of rotor Rr 16 mΩ of the derived results, Jeq is added into the moment of inertia
Leakage inductance of stator Lls 0.18 mH in the equivalent synchronous generator and the additional
Leakage inductance of rotor Llr 0.16 mH frequency control is removed from the original DFIG grid-
Mutual inductance Lm 2.9 mH connected system. The modified system is referred to as the
Proportional-integral (PI) coefficient of inner loop [5, 40] benchmark system of inertia. If the derived virtual moment
PI coefficient of outer loop [0.2, 20] of inertia is accurate enough, then the benchmark system of
Virtual inertia coefficient kdf 0.3´ 106 inertia will possess the same inertia as the original system in
theory as well as the frequency response to load disturbance.
Droop coefficient kpf 0.3´ 106
The comparison between the DFIG grid-connected system
and benchmark system of inertia using the derived results is
The background system is made up of an equivalent syn‐ shown in Fig. 6. The error of the system angular frequency
chronous generator, a load, and a 3-bus network. The mo‐ e ω is the difference between two trajectories at the corre‐
1
CHEN et al.: VIRTUAL INERTIA ESTIMATION METHOD OF DFIG-BASED WIND FARM WITH ADDITIONAL FREQUENCY CONTROL 1083

sponding sampling time. It can be observed that the errors grid-connected system and the benchmark system of inertia
are all located in the range of - 0.02-0.02, which proves the using the estimated result is shown in Fig. 8. It can be ob‐
accuracy of the derived virtual moments of inertia. served that the estimated J eq is accurate enough to be applied
Benchmark system
under different conditions of frequency disturbance.
315.0 of inertia 20

distribution (%)
Benchmark system
DFIG grid-connected

Probability
ω1 rad·s-1)

15 315.0 of inertia 20

distribution (%)
314.5 system
DFIG grid-connected

Probability
10

ω1 rad·s-1)
system 15
314.0 314.5
5 10
313.5 0 314.0 5
0 2 4 6 8 10 -0.02 -0.01 0 0.01 0.02
t (s) eω1 (rad·s-1) 0
313.5 -0.02 -0.01 0 0.01 0.02
(a) (b) 0 2 4 6 8 10
Benchmark system t (s) eω1 (rad·s-1)
315.0 20
distribution (%)
of inertia (a) (b)
Probability

DFIG grid-connected 15 Benchmark system


ω1 rad·s-1)

314.5 315.0 20

distribution (%)
system of inertia
10

Probability
DFIG grid-connected 15

ω1 rad·s-1)
314.0 5 314.5 system
10
0
313.5 -0.02 -0.01 0 0.01 0.02 314.0 5
0 2 4 6 8 10 eω1 (rad·s-1)
t (s) 0
(c) (d) 313.5 -0.02 -0.01 0 0.01 0.02
0 2 4 6 8 10
t (s) eω1 (rad·s-1)
Fig. 6. Comparison between DFIG grid-connected system and benchmark (c) (d)
system of inertia using derived results. (a) System angular frequency with
k pf = k df = 3´ 105 and ∆P L = 3 MW. (b) Error distribution of system angular Fig. 8. Comparison between DFIG grid-connected system and benchmark
frequency with k pf = k df = 3´ 105 by taking result of benchmark system of in‐ system of inertia using estimated result. (a) System angular frequency with
ertia as a reference. (c) System angular frequency with k pf = k df = 2 ´ 105 and ∆P L = 4 MW. (b) Error distribution of system angular frequency with ∆P L =
∆P L = 3 MW. (d) Error distribution of system angular frequency with k pf = 4 MW. (c) System angular frequency with ∆P L = 3 MW. (b) Error distribu‐
k df = 2 ´ 105 by taking result of benchmark system of inertia as a reference. tion of system angular frequency with ∆P L = 3 MW.

To further illustrate the performance of the proposed meth‐ Considering the effect of the parameters used in the addi‐
od, we take the DFIG grid-connected system with k pf = k df = tional frequency control, we further change the values of k df
2 ´ 105 as an application object. The sampling data sequences and k pf from 2×105 to 6×105. After the estimation, the corre‐
of the DFIG grid-connected system obtained by the electro‐ sponding estimated results of virtual moment of inertia are
magnetic transient simulation are shown in Fig. 7, where the summarized in Table II. In addition to the moment of inertia
sampling interval Dt and the sampling period T are set to be J d in DFIG, the virtual moment of inertia mainly depends on
2 ms and 10 s, respectively. the parameters of additional frequency control, and the pro‐
posed method is of high sensitivity with changes of control
9.5
parameters.
Wind speed

9.0
(m·s-1)

8.5 TABLE II
8.0 ESTIMATED RESULTS OF VIRTUAL MOMENT OF INTERTIA WITH DIFFERENT
7.5
0 2 4 6 8 10 12 14 16 18 20 PARAMETERS
t (s)
(a) Jeq
314.5 kdf
ω1 rad·s-1)

k pf = 2 × 105 k pf = 4 × 105 k pf = 6 × 105


314.0 2×105 787.07 1023.04 1259.25
313.5 4×105 1303.07 1593.72 1775.92
0 2 4 6 8 10 12 14 16 18 20
t (s) 6×105 1819.06 2055.95 2292.16
(b)
2.5
B. Modified IEEE 30-bus System with Wind Farm
Pw (MW)

2.0
To illustrate the performance of the proposed method ap‐
1.5
plied to a wind farm, a modified IEEE 30-bus system with a
1.0
0 2 4 6 8 10 12 14 16 18 20 wind farm shown in Fig. 9 is taken as the test system. Limit‐
t (s) ed by the electromagnetic transient simulation capability of
(c)
MATLAB/Simulink, the wind farm consists of three DFIGs,
Fig. 7. Sampling data sequences of DFIG grid-connected system. (a) Wind where the capacity of each DFIG is 10 MVA and their con‐
speed. (b) System angular frequency. (c) Grid-connected power.
verters employ the average model.
Setting the frequency disturbance by changing the load
The data marked by red lines are selected as the input, (25 MW) in the modified IEEE 30-bus system, the dynamic
and eventually, the virtual moment of inertia is estimated to responses of the system with or without the additional fre‐
be J eq = 787.07 kg·m 2. The comparison between the DFIG quency control are shown in Fig. 10. Regarding Fig. 10(c)
1084 JOURNAL OF MODERN POWER SYSTEMS AND CLEAN ENERGY, VOL. 9, NO. 5, September 2021

and (d), the load at each bus is cut in half, and the location moment of inertia into the generator at the slack bus. The
of the wind farm is changed to Bus 13, while the generators comparison between the modified IEEE 30-bus system and
at Buses 5 and 13 are removed. the corresponding benchmark system of inertia in the total
participation mode is shown in Fig. 12. It can be observed
1 2 5 7 8 that although the background system is extended and the
WT1 number of WTs is increased, the proposed method is still of
3 4 6 28 high accuracy.
WT2
Wind farm
9.5

Wind speed
WT3 9.0 WT1 WT2 WT3
13 12 11 9

(m·s-1)
29 8.5
27
8.0
7.5
15 16 17 25 0 2 4 6 8 10 12 14 16 18 20
14 10 t (s)
30 26
(a)
23 18 21 22 314.8
19 20 24

ω1 rad·s-1)
314.0
Fig. 9. Modified IEEE 30-bus system with a wind farm.
313.2
314.8 0 2 4 6 8 10 12 14 16 18 20
26 t (s)
ω1 rad·s-1)

314.4 24 (b)
Pw (MW)

22 28
314.0

Pw (MW)
20 24
313.6
18 20
313.2 16
0 2 4 6 8 10 0 2 4 6 8 10 16
t (s) t (s) 0 2 4 6 8 10 12 14 16 18 20
(a) t (s)
(b)
(c)
314.9 26 Fig. 11. Sampling data sequences of DFIG-based wind farm in total partic‐
ipation mode. (a) Wind speeds of three WTs. (b) System angular frequency.
ω1 rad·s-1)

314.3 24
Pw (MW)

22 (c) Grid-connected power.


313.7 20
Benchmark system
18
313.1 of inertia
315.2 16

distribution (%)
0 2 4 6 8 10 16 Modified IEEE
0
4 6 8 2 10
Probability
t (s) 314.8
ω1 rad·s-1)

t (s) 30-bus system 12


(c) (d) 314.4 8
Wind farm without additional frequency control 314.0
4
Wind farm with additional frequency control 313.6
0
313.2 -0.02 -0.01 0 0.01 0.02
Fig. 10. Dynamic responses of modified IEEE 30-bus system with or with‐ 0 2 4 6 8 10 eω1 (rad·s-1)
out additional frequency control. (a) System angular frequency. (b) Grid-con‐ t (s)
(a) (b)
nected power of wind farm. (c) System angular frequency with load at each
Benchmark system
bus cut in half. (d) Grid-connected power of wind farm with load at each 315.2 16
distribution (%)

of inertia
bus cut in half.
Probability

12
ω1 rad·s-1)

314.8 Modified IEEE


30-bus system 8
314.4
Apply the proposed method to estimate the virtual mo‐ 314.0
4
ment of inertia provided by these three equivalent WTs at 0
313.6 -0.02 -0.01 0 0.01 0.02
different wind speeds. Similar to Fig. 7, the sampling data 0 2 4 6 8 10
eω1 (rad·s-1)
t (s)
sequences of wind speeds individually measured by three (c) (d)
equivalent WTs are shown in Fig. 11(a), whereas the system
Fig. 12. Comparison between modified IEEE 30-bus system and corre‐
angular frequency and grid-connected power are shown in sponding benchmark system of inertia in total participation mode. (a) Sys‐
Fig. 11(b) and (c). The operation mode with all WTs partici‐ tem angular frequency with DP L = 25 MW. (b) Error distribution of system
pating in the DFIG-based wind farm is labeled as total par‐ angular frequency with DP L = 25 MW. (c) System angular frequency with
ticipation mode. After the judgment, the data segments DP L = 20 MW. (d) Error distribution of system angular frequency with DP L =
25 MW.
marked by red lines shown in Fig. 11 are determined as the
input of the estimation process, where the sampling interval
and sampling period are the same as those in Section V-A. To verify the estimation method in different operation
The virtual moment of inertia is estimated to be modes of the wind farm, two equivalent WTs are set to par‐
J eq = 3814.38 kg·m 2. The benchmark system of inertia corre‐ ticipate in the frequency regulation and the other one is not.
sponding to this operation mode is constructed in a similar Such an operation mode is labeled as partial participation
way to Section V-A, namely removing the additional fre‐ mode. The verification of the estimated virtual moment of in‐
quency control from the wind farm and adding the estimated ertia by using its benchmark system of inertia is shown in
CHEN et al.: VIRTUAL INERTIA ESTIMATION METHOD OF DFIG-BASED WIND FARM WITH ADDITIONAL FREQUENCY CONTROL 1085

Fig. 13. In the partial participation mode, since the accuracy the time cost of data collection will decrease significantly,
performance is the same as that of the total participation whereas the time cost on inertia estimation still maintains a
mode, it is not explained in detail any longer. satisfactory level. Regarding the sampling interval Δt = 2 ms
By setting different parameters of the additional frequency adopted above, the total time cost is less than the length of
control, Table III presents the estimated results of the virtual the data sequence of 10 s, which means the proposed meth‐
moment of inertia provided by the DFIG-based wind farm in od can be used online to monitor the virtual moment of iner‐
total participation mode, whereas Table IV summarizes the tia in real-time. It is worth noting that the derived expres‐
estimated results in the partial participation mode. Combined sion essentially completes the transformation of DFIG or
with Fig. 13, it can be observed that the proposed method is DFIG-based wind farm from non-synchronous to synchro‐
sensitive to the changes of control parameters and operation nous, which determines the dominant component of the virtu‐
mode. al moment of inertia and becomes the key point in the esti‐
mation process. In addition to the above academic contribu‐
TABLE III
ESTIMATED RESULTS OF VIRTUAL MOMENT OF INERTIA WITH DIFFERENT
tions, the proposed method presents a framework and an im‐
PARAMETERS IN TOTAL PARTICIPATION MODE plementation scheme with the satisfactory performance.
From the perspective of substitutability, more techniques can
Jeq be introduced to improve the efficiency and accuracy.
kdf
k pf = 6.0 × 105 k pf = 1.2 × 106 k pf = 1.8 × 106 TABLE V
6.0×105 3814.38 4414.01 5013.59 AVERAGE TIME COST OF MAIN STEPS IN AN EFFECTIVE ESTIMATION
1.2×106 7700.05 8299.74 8899.35 PROCESS
1.8×10 6
11585.89 12185.50 12785.11
Average time cost (s)
Main step
TABLE IV Δt = 1 ms Δt = 2 ms Δt = 3 ms
ESTIMATED RESULTS OF VIRTUAL MOMENT OF INERTIA WITH DIFFERENT Data collection (Steps 1-3) 21.1092 2.1523 0.6944
PARAMRTERS IN PARTIAL PARTICIPATION MODE Inertia estimation (Steps 4, 5) 0.0633 0.0570 0.0470
Total 21.1725 2.1823 0.7414
Jeq
kdf
k pf = 6.0 × 105 k pf = 1.2 × 106 k pf = 1.8 × 106
6.0×105 2888.41 3288.37 3688.33
VI. CONCLUSION
1.2×106 5478.18 5878.29 6278.25 Starting from the derivation of the frequency response
1.8×10 6
8068.25 8468.21 8868.17 model of DFIG, this paper proposes an estimation method to
calculate the virtual moment of inertia provided by the
Benchmark system DFIG-based wind farm with additional frequency control.
315.0
of inertia
16 The case studies tested by the DFIG grid-connected system
distribution (%)

Modified IEEE
and modified IEEE 30-bus system demonstrate the effective‐
Probability

30-bus system
ω1 rad·s-1)

314.6 12
8 ness of the derived expression and the proposed method.
314.2
4 Some conclusions are derived below.
313.8
0
1) The derived expression of the virtual moment of inertia
313.4
0 2 4 6 8 10
-0.02 -0.01 0 0.01 0.02 can directly represents the inertial characteristic of WT with
eω1 (rad·s-1)
t (s) additional frequency control, which is easy to combine with
(a) (b)
Benchmark system
the existing moment of inertia in the power system and thus
315.0 of inertia 16 determine the overall frequency response of the power system.
distribution (%)

Modified IEEE 2) The estimation method can efficiently calculate the vir‐
Probability

12
ω1 rad·s-1)

314.6 30-bus system


8 tual moment of inertia provided by the wind farm, whether
314.2
4
the WTs are fully involved in the frequency regulation or
313.8
0
not. Such an identification way avoids the complex process‐
313.4
0 2 4 6 8 10
-0.02 -0.01 0 0.01 0.02 ing of the parametric derivation and frequency response inte‐
t (s) eω1 (rad·s-1) gration, which is very suitable for the power system with
(c) (d) high penetration of wind power.
Fig. 13. Comparison between modified IEEE 30-bus system and corre‐
sponding benchmark system of inertia in partial participation mode. (a) Sys‐ APPENDIX A
tem angular frequency with DP L = 25 MW. (b) Error distribution of system
angular frequency with DP L = 25 MW. (c) System angular frequency with Taking the stator voltage-oriented vector control as exam‐
DP L = 20 MW. (d) Error distribution of system angular frequency with DP L = ple, the increment of output active power at the stator side can
20 MW.
be expressed as:
Lm
Table V presents the average time cost of the main steps 1.5G 1 (s)G i (s)u sd0
Ls
in an effective estimation process, where the estimation DP s = DP s* (A1)
times is 10. The total time cost mainly depends on the pro‐ Lm
1+ 1.5G 1 (s)G i (s)u sd0
cess of data collection. As the sampling interval increases, Ls
1086 JOURNAL OF MODERN POWER SYSTEMS AND CLEAN ENERGY, VOL. 9, NO. 5, September 2021

where G 1 (s) = k p1 + k i1 s is the PI regulator of the outer power vol. 8, no. 3, pp. 499-510, May 2020.
[2] P. Mukherjee and V. V. Rao, “Superconducting magnetic energy stor‐
loop, and k p1 and k i1 are the corresponding proportional and in‐ age for stabilizing grid integrated with wind power generation sys‐
tegral parameters, respectively; and G i (s) is the transfer func‐ tems,” Journal of Modern Power Systems and Clean Energy, vol. 7,
no. 2, pp. 400-411, Mar. 2019.
tion of the inner current loop, i.e., [3] L. Hao, J. Ji, D. Xie et al., “Scenario-based unit commitment optimi‐
G 2 (s)G pwm (s)G c (s) zation for power system with large-scale wind power participating in
G i (s) = (A2) primary frequency regulation,” Journal of Modern Power Systems and
1+ G 2 (s)G pwm (s)G c (s) Clean Energy, vol. 8, no. 6, pp. 1259-1267, Nov. 2020.
[4] D. Ochoa and S. Martinez, “Fast-frequency response provided by
where G 2 (s) = k p2 + k i2 s is the PI regulator of the inner current DFIG-wind turbines and its impact on the grid,” IEEE Transactions
loop, and k p2 and k i2 are the corresponding proportional and in‐ on Power Systems, vol. 32, no. 5, pp. 4002-4011, Sept. 2017.
[5] W. Yan, X. Wang, W. Gao et al., “Electro-mechanical modeling of
tegral parameters, respectively; G pwm (s) = K (1+ T δ s), and T δ is wind turbine and energy storage systems with enhanced inertial re‐
the switching cycle; and G c (s) = 1 (σL r s + R r ), and R r and L r sponse,” Journal of Modern Power Systems and Clean Energy, vol. 8,
no. 5, pp. 820-830, Sept. 2020.
are the resistance and inductance at the rotor side, respectively, [6] S. Wang and K. Tomsovic, “A novel active power control framework
σ = 1- L 2m (L s L r ). for wind turbine generators to improve frequency response,” IEEE
Transactions on Power Systems, vol. 33, no. 6, pp. 6579-6589, Nov.
Since G pwm (s) is usually of large bandwidth, in the low-fre‐ 2018.
quency band, it can be assumed as the gain of K, and then [7] Z. Wu, W. Gao, T. Gao et al., “State-of-the-art review on frequency re‐
sponse of wind power plants in power systems,” Journal of Modern
G i (s) is approximated as: Power Systems and Clean Energy, vol. 6, no. 1, pp. 1-16, Jan. 2018.

Ĝ i (s) =
( )
k p2 K s +
k i2
k p2
[8] A. Žertek, G. Verbič, and M. Pantoš, “Optimised control approach for
frequency-control contribution of variable speed wind turbines,” IET
Renewable Power Generation, vol. 6, no. 1, pp. 17-23, Jan. 2012.

( )
(A3) [9] H. T. Ma and B. H. Chowdhury, “Working towards frequency regula‐
Rr tion with wind plants: combined control approaches,” IET Renewable
sσL r s + Power Generation, vol. 4, no. 4, pp. 308-316, Jul. 2010.
σL r [10] L. Chang-Chien, W. Lin, and Y. Yin, “Enhancing frequency response
control by DFIGs in the high wind penetrated power systems,” IEEE
Note that due to the plant pole at s = - R r (σL r ), which is Transactions on Power Systems, vol. 26, no. 2, pp. 710-718, May
fairly close to the origin, the magnitude and phase of the loop 2011.
[11] Z. Zhang, Y. Sun, and G. Lin, “Coordinated frequency regulation by
gain start to drop from a relatively low frequency. When the doubly fed induction generator-based wind power plants,” IET Renew‐
plant pole is first canceled by the compensator zero, Ĝ i (s) be‐ able Power Generation, vol. 6, no. 1, pp. 38-47, Jan. 2012.
comes: [12] Y. Fu, Y. Wang, and X. Zhang, “Integrated wind turbine controller
with virtual inertia and primary frequency responses for grid dynamic
k p2 K frequency support,” IET Renewable Power Generation, vol. 11, no. 8,
Ĝ ′i (s) = (A4) pp. 1129-1137, Jun. 2017.
σL r s + k p2 K
[13] X. Tian, W. Wang, Y. Chi et al., “Virtual inertia optimisation control
Replacing G i (s) in (A1) with Ĝ ′i (s), we can obtain: of DFIG and assessment of equivalent inertia time constant of power
grid,” IET Renewable Power Generation, vol. 12, no. 15, pp. 1733-
DP s B1 s + B2 1740, Nov. 2018.
( s ) = G p (s) = (A5) [14] L. Wu and D. Infield, “Towards an assessment of power system fre‐
DP *
s A 0 s2 + A 1 s + A 2 quency support from wind plant – modeling aggregate inertial re‐
sponse,” IEEE Transactions on Power System, vol. 28, no. 3, pp.
ìA 0 = 2L s σL r 2283-2291, Aug. 2013.
ï [15] L. Wu and D. Infield, “Power system frequency management challeng‐
ïA 1 = 3L m Ku sd0 k p2 k p1 + 2k p2 L s K es–a new approach to assessing the potential of wind capacity to aid
ï
íA 2 = 3L m Ku sd0 k p2 k i1
system frequency stability,” IET Renewable Power Generation, vol. 8,
(A6) no. 7, pp. 733-739, Sept. 2014.
ï
ïB 1 = 3L m Ku sd0 k p2 k p1 [16] S. Li, C. Deng, Z. Long et al., “Calculation of equivalent virtual iner‐
ïB = 3L Ku k k tia time constant of wind farm,” Automation of Electric Power Sys‐
î 2 m sd0 p2 i1 tems, vol. 40, no. 7, pp. 22-29, Apr. 2016.
[17] H. Li, X. Zhang, Y. Wang et al., “Virtual inertia control of DFIG-
By using Routh approximation, two types of the first-order based wind turbines based on the optimal power tracking,” Proceed‐
transfer function close to G p (s) in the low-frequency band are ings of the CESS, vol. 32, no. 7, pp. 32-39, Mar. 2012.
obtained. [18] A. B. T. Attya and T. Hartkopf, “Control and quantification of kinetic
energy released by wind farms during power system frequency drops,”
ì B1 IET Renewable Power Generation, vol. 7, no. 3, pp. 210-224, May
ïG′p (s) = 2013.
ï A 0 + A1
s [19] T. Littler, B. Fox, and D. Flynn, “Measurement-based estimation of
í (A7) wind farm inertia,” in Proceedings of 2005 IEEE Russia Power Tech,
ï ̂ B2
ï G ′ (s) = St. Petersburg, Russia, Jun. 2005, pp. 1-5.
î
p
A1 s + A2 [20] Y. Wang, C. Lu, L. Zhu et al., “Comprehensive modeling and parame‐
ter identification of wind farms based on wide-area measurement sys‐
After comparing their pole positions and selecting the one tems,” Journal of Modern Power Systems and Clean Energy, vol. 4,
close to the imaginary axis, the expression of the time constant no. 3, pp. 383-393, Jul. 2016.
[21] A. Fernández-Guillamón, A. Vigueras-Rodríguez, and Á. Molina-
is finally determined. García, “Analysis of power system inertia estimation in high wind
power plant integration scenarios,” IET Renewable Power Generation,
REFERENCES vol. 13, no. 15, pp. 2807-2816, Feb. 2019.
[22] Z. Zeng and W. Shao, “Estimation of inertia and damping for virtual
[1] J. Radosavljević, N. Arsić, M. Milovanović et al., “Optimal placement synchronous generator based on linearized model,” Automation of
and sizing of renewable distributed generation using hybrid metaheuris‐ Electric Power Systems, vol. 41, no. 10, pp. 37-43, May 2017.
tic algorithm,” Journal of Modern Power Systems and Clean Energy, [23] Q. Luo, J. Su, Z. Lin et al., “Parameter identification method for virtu‐
CHEN et al.: VIRTUAL INERTIA ESTIMATION METHOD OF DFIG-BASED WIND FARM WITH ADDITIONAL FREQUENCY CONTROL 1087

al synchronous generators based on recursive least squares algorithm,” Pengwei Chen received the B.S. and Ph.D. degrees in electric power engi‐
Automation of Electric Power Systems, vol. 43, no. 1, pp. 215-221, neering from North China Electric Power University, Beijing, China, in
Jan. 2019. 2014 and 2019, respectively. From 2017 to 2018, he was a Visiting Scholar
[24] K. Liu and Z. Zhu, “Fast determination of moment of inertia of perma‐ at the Future Renewable Electric Energy Delivery and Management
nent magnet synchronous machine drives for design of speed loop reg‐ (FREEDM) Systems Center, North Carolina State University, Raleigh, USA.
ulator,” IEEE Transactions on Control Systems Technology, vol. 25, He was the recipient of the Outstanding Graduate Award of Beijing, China.
no. 5, pp. 1816-1824, Sept. 2017. He is presently a Lecturer at Nanjing University of Aeronautics and Astro‐
[25] K. V. Vidyanandan and N. Senroy, “Primary frequency regulation by nautics, Nanjing, China. His research interests include DC distribution net‐
work simulation, stability, and power quality control.
deloaded wind turbines using variable droop,” IEEE Transactions on
Power Systems, vol. 28, no. 2, pp. 837-846, May 2013.
[26] C. S. Hsieh and C. Hwang, “Model reduction of continuous-time sys‐ Chenchen Qi received the B. S. degree in electrical engineering from the
Nanjing University of Aeronautics and Astronautics, Nanjing, China, in
tems using a modified Routh approximation method,” IEE Proceed‐
2019, where he is currently working toward the master’s degree in electri‐
ings D–Control Theory and Applications, vol. 136, no. 4, pp. 151-156,
cal engineering. His research interests include new energy generation tech‐
Jul. 1989. nology, DC distribution network model, and control.
[27] T. K. Sarkar and O. Pereira, “Using the matrix pencil method to esti‐
mate the parameters of a sum of complex exponentials,” IEEE Anten‐
Xin Chen received the B. S. and Ph. D. degrees in electrical engineering
nas and Propagation, vol. 37, no. 1, pp. 48-55, Feb. 1995. from the Nanjing University of Aeronautics and Astronautics, Nanjing, Chi‐
[28] N. Yilmazer, J. Koh, and T. K. Sarkar, “Utilization of a unitary trans‐ na, in 1996 and 2001, respectively. From 2001 to 2003, he was a Chief En‐
form for efficient computation in the matrix pencil method to find the gineer with the Power Division of ZTE Corporation, Nanjing, China. From
direction of arrival,” IEEE Transactions on Antennas and Propaga‐ 2010 to 2011, he was a Visiting Scholar with the Rensselaer Polytechnic In‐
tion, vol. 54, no. 1, pp. 175-181, Jan. 2006. stitute, Troy, USA. He is currently a Professor at the Nanjing University of
[29] G. Vandersteen, “On the use of compensated total least squares in sys‐ Aeronautics and Astronautics. His research interests include power electron‐
tem identification,” IEEE Transactions on Automatic Control, vol. 43, ics and energy conversion, with a focus on modeling, control, stability, as
no. 10, pp. 1436-1441, Oct. 1998. well as applications in renewable energy system.

You might also like