Analytical Model of A Grid-Forming Inverter

Ratik Mittal, Student Member, IEEE, Zhixin Miao, Senior Member, IEEE

Abstract—Recently, inverter-based resources are being inte- eling. The developed model is validated in the time-domain
grated into the power grid at a rapid rate. Most of these resources with the EMT model for weak grid operations. The linear
are grid-following inverters, where weak grid operation becomes state-space model is also extracted from the analytical model
an issue. The research focus has now shifted towards grid forming
inverters, which emulate synchronous generators. Thus proper to perform eigenvalue analysis. Furthermore, laboratory-scale
modeling and analysis are required to understand the grid hardware experimentation was also conducted to replicate the
forming inverters. This paper presents the nonlinear analytical EMT studies and analytical model results when operating
model of a grid forming inverter when operating in grid- under weak grid conditions.
connected mode in dq -domain. The developed model is validated The rest of the paper is structured as follows. Section
in the time domain with the Electromagnetic Transients (EMT)
test bed when the system operates under weak grid conditions. II describes the system topology. Section III illustrates the
2022 IEEE Power & Energy Society General Meeting (PESGM) | 978-1-6654-0823-3/22/$31.00 ©2022 IEEE | DOI: 10.1109/PESGM48719.2022.9916676

The analytical model is also used to perform the eigenvalues analytical model building. Section IV presents the results and
analysis. The EMT model is simulated in MATLAB/Simscape analysis. Section V outlines the hardware implementation.
and the analytical model is simulated in MATLAB/Simulink. Section VI concludes the paper.
Hardware experimentation is also performed for further bench-
marking with the help of laboratory scale hardware setup. II. S YSTEM T OPOLOGY
Index Terms—Grid forming inverter, weak grid, analytical
model, eigenvalue The schematic of the GFM presented in Fig. 1. The DC
terminal of the three-phase VSC is connected to a DC voltage
I. I NTRODUCTION source, and the converter is connected to the grid through a
transmission line, represented by Rg , and Lg . A RLC choke
T HE MOTIVATION to achieve the goal of clean energy
has led to rapid advances towards power electronics-
based sources, known as inverter-based resources (IBRs). IBRs
filter is connected at the terminal of the VSC with Rf , Lf , and
Cf as the resistance, inductance, and capacitance respectively.
A purely resistive load RLoad is also connected at the PCC
are quickly replacing the traditional synchronous generators
bus. The control structure of the GFM is implemented in dq
based power grids. In fact, in some parts of the world like
frame. The following notation are used: icdq , vdq and iLdq are
Tasmanian, Ireland, South Australia met more than 90% their
the current flowing through the filter inductor (Lf ), voltage at
energy demand by renewable energy sources [1], [2], [3].
the PCC bus and the load current in dq frame respectively. P
Predominantly these IBRs are grid following inverters
and Q are the real and reactive power at the PCC bus.
(GFLs), which imitate the characteristics of a current source.
The inner loop is PI controller-based current control struc-
Recently researchers have shifted their focus on a different
ture, where the filter inductor current (icdq ) as the control
type of IBRs, known as “grid forming inverter” (GFM) [4].
variable. The outer loop regulates the voltage at the PCC bus
Theoretically, the primary function of the GFMs is to act as
(vdq ). Similar to the inner loop, outer loop also implements
a controllable voltage source. This characteristic property of
a PI controller. Two droop control methods are implemented:
GFMs makes them a suitable replacement for synchronous
(1) Q − V droop and (2) P − f droop. Q − V droop generates
generators in low inertia systems. Furthermore, GFMs can also
the d-axis reference signal (vdref ) for the outer loop. On the
be used to black start a power grid [5], [6].
other hand, the P −f droop regulates the PCC bus’s real power
Therefore, a proper modeling approach is required to study
and provides the desired angle (ωt) for the control. Parameters
GFM inverters. The first approach uses EMT simulations
used in this paper are listed in TABLE. I.
packages like MATLAB/Simscape, PSCAD, EMTP, etc. These
packages are suitable for performing time-domain simulations. III. dq- DOMAIN A NALYTICAL M ODEL
They are a kind of virtual lab for experiments. The second
In this paper, the analytical model of the GFM presented in
approach is nonlinear analytical model building. The analytical
Fig. 1 is developed in dq -domain [11], and is based on per unit
modeling is a widely adopted technique in research community
system. The block diagram of the analytical model is presented
[7], [8], [9], [10]. It can provide many more insights into the
in Fig. 2. As per Fig. 2, the system is modeled with four main
system dynamics, by providing linear state-space models for
blocks: 1) droop control, 2) inner loop, 3) outer loop, 4) circuit
further analysis.
dynamics, and 5) frame conversion. It is noted that, while the
In this paper, we develop the dq-domain analytical model of
signals at the circuit level are based on grid frame (superscript
the GFM when operating in grid-connected mode. The analyt-
“g”), the signals associated with the control structure are based
ical model includes the controller dynamics and circuit mod-
on the reference frame or reference angle provided by the the
R.Mittal and Z.Miao are with the Department of Electrical Engineering, P − f droop. Hence, the control structure is in the P − f
University of South Florida, Tampa, FL, 33630 USA. Email: [email protected] droop frame of reference (superscript “c”). Furthermore, there

978-1-6654-0823-3/22/$31.00 ©2022 IEEE

Authorized licensed use limited to: Technical University of Denmark DTU Library. Downloaded on June 01,2024 at 08:14:32 UTC from IEEE Xplore. Restrictions apply.
 YW 5I/I LR Y 5J/J YJ A. Circuit Dynamics
9'& &'&
LF L&I &I L/ 5/RDG LJ The circuit dynamics includes the grid current iggdq , the
current through the choke filter igcdq and the PCC bus voltage
3:0 g
Y LF L/
*HQHUDWLRQ I Y LF vdq . The model of the circuit dynamics is developed in the
ǻ ș 3UHI grid dq frame that rotates at a speed of “ω0 ”. The differential
P$%& $%& ȦW 3


3I'URRS
ȦW  ǻ ș GT 3 4 equations for the circuit dynamics are presented as follows:
.


ȦW  YGUHI 4 FDOFXODWLRQ

49'URRS
YW$%& YGT LFGT L/GT
4UHI
$%& ǻ ș diggd Rgpu g 1 g 1 g
9UHI =− i + v − v + ω0 iggq


GT YG L/G dt Lgpu gd Lgpu d Lgpu gd

LFG YG
YWG LFGUHI
diggq Rgpu g 1 g 1 g
YGUHI =− i + v − v + ω0 iggd
YWT ,QQHU
LFTUHI
2XWHU dt Lgpu gq Lgpu q Lgpu gq
&XUUHQW/RRS 9ROWDJH/RRS YTUHI 
digcd Rf pu g 1 g 1 g
LGT YT
=− icd + vtd − v + ω0 igcq
. 9EDVH✕9'& dt Lf pu Lf pu Lf pu d
YT L/T (1)
digcq Rf pu g 1 g 1 g
=− icq + vtq − v + ω0 igcd
Fig. 1: System topology of grid connected system with grid forming control. dt Lf pu Lf pu Lf pu q
dvdg 1
= (ig − igod ) + ω0 vqg
TABLE I: Parameters used for grid forming converter. Note: The PI controller
dt Cf pu cd
gains are based on per unit system. dvqg 1
= (ig − igoq ) + ω0 vdg
Description Parameter Value dt Cf pu cq
Power Base Sb 50 VA
Voltage Base Vbase 20 V (L-L rms)
Here Rf pu , Lf pu , and Cf pu are the per unit values of the
Nominal Frequency f0 60 Hz RLC components of the choke filter, and similarly Rgpu
Grid Voltage Vg 11.5 V (L-N, rms) and Lgpu are the per-unit values of the transmission line
DC Voltage VDC 40 V parameters. Also, igod and igoq are the dq components of the
DC Link Capacitor CDC 260 μF g g
Choke Filter L f , Rf , Cf 1.5 mH, 70 mΩ, 47 μF current io (marked in Fig. 1) in grid frame, and vtd and vtq
Transmission Line Rg , L g 0.2 Ω, 7.5mH are the dq components of the terminal voltage vt . The per unit
Load RLoad 20 W expressions for the real power and the reactive power at the
Switching Frequency fSW 5 kHz PCC bus are:
Inner Loop: PI gains kpi , kii 1, 5
Outer Loop: PI gains kpv , kiv 3, 50
Droop Parameters m, n 0.08, 0.1 P = vdg igcd + vqg igcq and Q = −vdg igcq + vqg igcd (2)

And the load current in dq frame (grid frame) is given by:

vdg vqg
are 11 state variables in the analytical model. x1 is associated igLd = and igLq = (3)
RLoadpu RLoadpu
with the P − f droop, x2 -x5 are related to the outer loop and
the inner loop, iggdq , igdq and vdq
g
are the state-variables related Hence, (1) - (3) completes the analytical model of the circuit
to the circuit dynamics. dynamics.

2XWHU/RRS%ORFN ,QQHU/RRS%ORFN )& YJJG YJJT )&

LF/G YFG
NSY NSL 
YGUHI  [[  LFFGUHI [[  YFWG GTF YJWG LJFGT LFFGT
NLY œ   NLL GTJ
 

œ





    &LUFXLW
YFG LFFG YJGT YFGT
%F 

;I 0RGHO 

F
YFT %F L FT ;I LJJGT LFJGT
    


 [[ LFFTUHI  [[ YFWT YJWT

YTUHI  NSY œœ   NSL
 

œœ


  LJ/GT GTF LF/GT

NLY
 GTJ
NLL
LF/TT  F
Y T
ǻ ș 3 4 ǻ ș
ȦW




 ǻ ș
I


ȦW I 3
ʌ œ  [  3 UHI
œ ʌ  P 
ȦW
)& )UDPH&RQYHUVLRQ%ORFN
YGUHI Q  4


'URRS&RQWURO%ORFN 9UHI 4UHI

Fig. 2: Block diagram of the analytical model for GFM. “FC” stands for “frame conversion”.

978-1-6654-0823-3/22/$31.00 ©2022 IEEE

Authorized licensed use limited to: Technical University of Denmark DTU Library. Downloaded on June 01,2024 at 08:14:32 UTC from IEEE Xplore. Restrictions apply.
 TABLE II: Steady-state values obtained from initialization procedure when
B. Frame conversion Pref = 2.22 pu and Qref = 0 .
As discussed earlier, the control system is operating in Measurement Value
a different reference rotating at a speed of “ω” which is Q 0.7882 pu
provided by the P − f droop.The control structure regulates Vmag 0.9212 pu
Δθ 67.6752°
the converter current ic , and the voltage at the PCC bus v is
dq frame. Since the variables from the circuit level model and
the control level model are in a different frame of reference,
a proper transformation is required to capture the analytical IV. A NALYSIS AND S IMULATION -BASED VALIDATIONS
model accurately. For the transformation, the concept of space The EMT test bed and the analytical model are used to study
vector is used [12]. For example, the space vector for the PCC the VSC system under weak grid conditions (Xg = 0.35 pu).
bus voltage V , is presented as: The EMT test-bed is implemented in MATLAB/Simscape, and
→
− the analytical model using MATLAB/Simulink environment.
V = (vdc + j vqc )ejωt = (vdg + j vqg ) ejω0 t (4) First, we present the time-domain results from the EMT test-
bed and the analytical model. The analytical model is then
Hence: used to extract the eigenvalues for further analysis.
(vdc + j vqc ) = (vdg + j vqg ) e−jΔθ . (5)
A. Time-domain results
From (5) the signals in grid frame of reference can be Pref is increased in steps, and the system response is
transformed to the P − f droop frame of reference, and vice- recorded for analysis. The time domain results are presented
verse, and Δθ = ωt - ω0 t. in Fig. 3. It is observed that as the reference real power is in-
creased the system undergoes oscillations and the frequency of
C. Analytical Model Initialization the oscillations is 3-Hz. Also, the system is marginally stable
As discussed earlier, there are 11 state variables in the when Pref = 2.23 pu. The results from the developed analytical
analytical model. A proper initialization procedure is needed model match well with the EMT simulations. Moreover, when
to develop the analytical model accurately. Pref = 2.22 pu the steady state values match very well with
The complex power at the PCC bus is given: S = V Ic ∗ , the results tabulated in Table. II. Please note: Qref = 0 for the
where V is the PCC bus voltage phasor and Ic is the current simulation results.
phasor through the filter inductor. The real power is: P = B. Eigenvalue Analysis
real(S), and the reactive power is: Q = imag(Q). The PCC
bus voltage phasor is defined as V = Vmag θ and the current The linearized state-space model is obtained from the de-
Ic is obtained as: veloped dq-domain analytical model. This is done with the
help of MATLAB’s “linmod” command, which computes the
V − Vg V linear state-space model. The eigenvalues are then computed
Ig = , ILoad =
Rgpu + i Xgpu RLoad,pu from the obtained state-space model, by using the “A” matrix.
(6)
V The eigenvalue loci when Pref is varied from 2.0 pu to 2.25
ICf = , Ic = Ig + ILoad + ICf
−i XCpu pu is presented in Fig. 4. It can be observed from Fig. 4 that,
when Pref is increased, the eigenvalue associated with the 3-
Here, Ig is the current in the transmission line, ICf is the Hz component moves to the right-half plane (RHP), with the
current through the filter capacitor Cf , and Vg = 1. Since,
the GFM is in grid connected mode the system is always at marginal condition as Pref = 2.23 pu. The eigenvalue trajectory
the nominal frequency (f0 ). The steady-state computing can is in coherence with the time-domain results, presented in Fig.
be formulated as an optimization problem and solved using 3.
YALMIP [13]. Part of the code isis shown below: Fig. 5 presents the mode shape of all the state-variables.
Vmag = sdpvar(1,1); The mode shape is obtained from the right eigenvector (V )
theta = sdpvar(1,1); corresponding to a particular mode (for this paper 3-Hz mode).
P = sdpvar(1,1);
Q = sdpvar(1,1); Mode shape provides the information on how a particular
assign(Vmag,1); mode is reflected in the system. It is observed from Fig. 5 that
assign(P, 1); 3-Hz oscillation mode is due to the state-variable associated
assign(Q, 0);
assign(theta,0); with the q axis grid current iggq . Furthermore, the state variables
Constraints = [P==Pref; vdg and iggq (red line) have a phase difference of about 180
Vmag-1 == n*(Qref-Q)]; degrees and hence oscillate against each other. Additionally,
Objective = 0;
state variables igcq and igcq (blue dashed line) have a phase
Here Pref and Qref are the desired operating conditions for the difference of about 90 degrees.
system. The optimization problem solves for the value of Vmag Moreover, from the linear state-space model we can also
and θ. Given Vmag and θ, the state variables are calculated for determine the mode under investigation is observable in which
initialization. For Pref = 2.22 pu and Qref = 0, steady-state measurement. In this paper the measurements are P , Q, |V |
values for Vmag θ, and Q are listed in Table. II. and Δθ. Here |V | is the magnitude of the PCC bus voltage,

978-1-6654-0823-3/22/$31.00 ©2022 IEEE

Authorized licensed use limited to: Technical University of Denmark DTU Library. Downloaded on June 01,2024 at 08:14:32 UTC from IEEE Xplore. Restrictions apply.
 P P P P
ref ref
Pref= 2.23pu Pref= 2.24pu

P (pu)
P (pu)
2.24 2.24 Pref= 2.23pu
2.22 P = 2.24pu 2.22
ref
2.2 2.2
31 32 33 34 35 36 37 38 39 44 45 46 47 48 49 50 51 52
0.9 0.9

Q (pu)
Q (pu)

0.86 Q = 0.7882 pu Q = 0.7882 pu

0.82 0.8
Q = 0.041 pu
31 32 33 34 35 36 37 38 39 44 45 46 47 48 49 50 51 52
0.94 0.94

|V| (pu)
|V| = 0.0148 pu
|V|(pu)

0.93
0.92 0.92
0.91 |V| = 0.9212 pu |V| = 0.9212 pu
0.9 0.9
31 32 33 34 35 36 37 38 39 44 45 46 47 48 49 50 51 52
60.1

Frequency
60.1
Frequency

60 60

(Hz)
(Hz)

59.9 59.9
31 32 33 34 35 36 37 38 39 44 45 46 47 48 49 50 51 52
72

(deg)
(deg)

72 = 67.7o o
70 70 = 67.5
68 ( )=2
0 68
31 32 33 34 35 36 37 38 39 44 45 46 47 48 49 50 51 52
Time (s) Time (s)

(a) (b)
Fig. 3: Side-by-side comparison of results obtained from the EMT test-bed and developed analytical model. (a) Time domain results from the EMT test-bed.
(b) Time domain results from analytical model.

with the results tabulated in Table. II.

5
P = 2 pu TABLE III: Observability of the 3-Hz mode in the measurements.
ref
4
P = 2.25 pu
ref
3 Measurement Value
2 P
ref
= 2.23pu P 0.1404 −94.08°
Q 0.3050 −30.08°
1 |V | 0.1027 −157.43°
Imag(Hz)

0 Δθ 0.227 −5.24°
-1

-2

-3

-4

-5
-10 -8 -6 -4 -2 0 2 4
Real
Fig. 4: Eigenvalue loci when Pref is varied. Here we can see the frequency
of oscillations is about 3-Hz, and the marginal condition is at Pref = 2.23pu.

and Δθ is angle from the P − f droop. Table. III lists the

observability of the 3-Hz in the measurements. It can be
seen that the 3-Hz oscillation mode is most observable in the
reactive power measurement (Q).
From the time domain responses presented in Fig. 3, we Fig. 5: Mode shape obtained from the right eigen vector corresponding to the
can observe that the ripple magnitude (in % with respect to 3-Hz oscillation mode.
steady state value) of the 3-Hz oscillation for measurement
Q is about 5.1%, for |V | = 1.16%, P = 0.9%, and Δθ =
2.89%. The conclusion can be made that Q measurement has V. H ARDWARE I MPLEMENTATION
the largest presence of 3 Hz oscillations in its response. Laboratory-scale hardware implementation was done to
Additionally, on comparison of the response of Q and |V | implement GFM presented in Fig. 1 under the weak grid. The
at t = 45s, when Pref is increased from 2.22 pu to 2.23 pu, it is hardware testbed consists of three Imperix- PEB8024 power
observed that the phase shift between the two waveform is ≈ modules. The individual power modules are built with SiC
120°, with Q leading |V |. These observations match closely MOSFETs and onboard current sensors and are connected to

978-1-6654-0823-3/22/$31.00 ©2022 IEEE

Authorized licensed use limited to: Technical University of Denmark DTU Library. Downloaded on June 01,2024 at 08:14:32 UTC from IEEE Xplore. Restrictions apply.
 &KURPD*ULG 7UDQV 3&&%XV 5/&)LOWHU ,PSHUL[3(% The real-time results obtained from the hardware test-bed
6LPXODWRU /LQH ij%RDUGV
are presented in Fig. 7, where it is observed that as Pref is
increased, the system experiences low frequency oscillations
when connected to a weak grid and the frequency of oscilla-
tions is ≈ 3-Hz. The results from the EMT studies, analytical
model, and the hardware experimentation are a very good
match.
5HVLVWLYH/RDG 9$%&
,F$%&
&RQWUROOHU
,/RDG )URQW6LGH VI. C ONCLUSION
230HDVXUHPHQW8QLW 3(%
&XUUHQW
6HQVRU
9ROWDJH
6HQVRU
This paper presents the analytical model of the GFM when
operating in grid-connected mode. The analytical model is
23 developed in dq frame and includes the control dynamics net-
3RZHU,QWHUIDFH
3:0287 work model. First, the developed analytical model is validated
with the EMT model in the time domain. Eigenvalue analysis
Fig. 6: Laboratory-scale hardware test bed for implementing GFM.
is also conducted for further validations of the developed
analytical model. Laboratory-scale hardware experimentation
was also performed for additional validations.
form a three-phase voltage source converter. A DC voltage
source supplies the DC-bus voltage. The power grid source ACKNOWLEDGEMENT
is emulated by Chroma Regenerative Grid Simulator 61845, The authors wish to acknowledge Prof. Lingling Fan for
which is controlled using LabVIEW. Additional voltage and providing suggestions to improve this paper.
current sensors unit OP8662 from OPAL-RT is used to mea-
R EFERENCES
sure the PCC bus voltage and load current. Opal RT’s OP4500
acts as a real-time controller. The OP4500 acquires all the [1] H. Holttinen, A. Groom, E. Kennedy, D. Woodfin, L. Barroso, A. Orths,
K. Ogimoto, C. Wang, R. Moreno, K. Parks, and T. Ackermann,
measured analog signals from different sensor units, and the “Variable renewable energy integration: Status around the world,” IEEE
control algorithm is implemented in OP4500 with the help of Power and Energy Magazine, vol. 19, no. 6, pp. 86–96, 2021.
RT-Lab. Opal RT’s power interface acts as a junction between [2] Y. Lin, J. H. Eto, B. B. Johnson, J. D. Flicker, R. H. Lasseter, H. N.
Villegas Pico, G.-S. Seo, B. J. Pierre, and A. Ellis, “Research roadmap
OP4500 and the VSC system. The PWM pulses to the VSC on grid-forming inverters,” National Renewable Energy Lab.(NREL),
switches are relayed via the power interface using an optic Golden, CO (United States), Tech. Rep., 2020.
fiber cable. Fig. 6 presents the hardware testbed for the GFM. [3] B. Kroposki, B. Johnson, Y. Zhang, V. Gevorgian, P. Denholm, B.-M.
Hodge, and B. Hannegan, “Achieving a 100% renewable grid: Operating
The parameters used in the experimentation are listed in Table. electric power systems with extremely high levels of variable renewable
I. energy,” IEEE Power and Energy Magazine, vol. 15, no. 2, pp. 61–73,
2017.
[4] R. Lasseter, “Microgrids,” in 2002 IEEE Power Engineering Society
Winter Meeting. Conference Proceedings (Cat. No.02CH37309), vol. 1,
2002, pp. 305–308 vol.1.
P =2.23 pu
112 ref [5] A. Jain, J. N. Sakamuri, and N. A. Cutululis, “Grid-forming control
P (W)

strategies for black start by offshore wind power plants,” Wind Energy
P =2.24 pu Pref Science, vol. 5, no. 4, pp. 1297–1313, 2020. [Online]. Available:
111 ref P
https://wes.copernicus.org/articles/5/1297/2020/
435 440 445 450
[6] H. Jain, G.-S. Seo, E. Lockhart, V. Gevorgian, and B. Kroposki,
16
“Blackstart of power grids with inverter-based resources,” in 2020 IEEE
vd (V)

Power Energy Society General Meeting (PESGM), 2020, pp. 1–5.

15.5
[7] B. Wen, D. Boroyevich, R. Burgos, P. Mattavelli, and Z. Shen, “Analysis
of d-q small-signal impedance of grid-tied inverters,” IEEE Transactions
15
435 440 445 450 on Power Electronics, vol. 31, no. 1, pp. 675–687, 2016.
[8] L. Fan, “Modeling type-4 wind in weak grids,” IEEE Transactions on
7.2
Sustainable Energy, vol. 10, no. 2, pp. 853–864, 2019.
icd (A)

[9] F. Doost Mohammadi, H. Keshtkar Vanashi, and A. Feliachi, “State-

7.1 space modeling, analysis, and distributed secondary frequency control of
7 isolated microgrids,” IEEE Transactions on Energy Conversion, vol. 33,
435 440 445 450 no. 1, pp. 155–165, 2018.
Frequency (Hz)

[10] A. Adib, B. Mirafzal, X. Wang, and F. Blaabjerg, “On stability of voltage

60.2 source inverters in weak grids,” IEEE Access, vol. 6, pp. 4427–4439,
60 2018.
59.8 [11] A. Yazdani and R. Iravani, Voltage-sourced converters in power systems:
modeling, control, and applications. John Wiley & Sons, 2010.
435 440 445 450
[12] L. Fan, “Modeling type-4 wind in weak grids,” IEEE Transactions on
Time (s)
Sustainable Energy, vol. 10, no. 2, pp. 853–864, 2019.
[13] J. Löfberg, “Yalmip : A toolbox for modeling and optimization in
Fig. 7: Time-domain responses of real power P (W), d-axis PCC bus voltage matlab,” in In Proceedings of the CACSD Conference, Taipei, Taiwan,
vd (V), d-axis choke filter current icd (A) and frequency (Hz) from the 2004.
hardware test-bed, when Pref is increased. In the hardware test-bed, the
switching frequency of the SiC switches is 5 kHz.

978-1-6654-0823-3/22/$31.00 ©2022 IEEE

Authorized licensed use limited to: Technical University of Denmark DTU Library. Downloaded on June 01,2024 at 08:14:32 UTC from IEEE Xplore. Restrictions apply.