Direct AC Voltage Control for Grid-Forming Inverters

Taoufik Qorai, Chuanyue Li, Ko Oue, Francois Gruson, Fréderic Colas, Xavier
Guillaud

To cite this version:

Taoufik Qorai, Chuanyue Li, Ko Oue, Francois Gruson, Fréderic Colas, et al.. Direct AC Voltage
Control for Grid-Forming Inverters. Journal of Power Electronics, 2019, pp.198-211. �hal-02458108�

HAL Id: hal-02458108

https://hal.archives-ouvertes.fr/hal-02458108
Submitted on 28 Jan 2020

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 ORIGINAL ARTICLE

Direct AC voltage control for grid‑forming inverters

Taoufik Qoria1 · Chuanyue Li1 · Ko Oue1 · Francois Gruson1 · Frederic Colas1 · Xavier Guillaud1

Received: 13 April 2019 / Revised: 22 July 2019 / Accepted: 12 August 2019

© The Korean Institute of Power Electronics 2019

Abstract
Grid-forming inverters usually use inner cascaded controllers to regulate output AC voltage and converter output current.
However, at the power transmission system level where the power inverter bandwidth is limited, i.e., low switching frequency,
it is difficult to tune controller parameters to achieve the desired performances because of control loop interactions. In this
paper, a direct AC voltage control-based state-feedback control is applied. Its control gains are tuned using a linear quadratic
regulator. In addition, a sensitivity analysis is proposed to choose the right cost factors that allow the system to achieve the
imposed specifications. Conventionally, a system based on direct AC voltage control has no restriction on the inverter cur-
rent. Hence, in this paper, a threshold virtual impedance has been added to the state-feedback control in order to protect the
inverter against overcurrent. The robustness of the proposed control is assessed for different short-circuit ratios using small-
signal stability analysis. Then, it is checked in different grid topologies using time domain simulations. An experimental test
bench is developed in order to validate the proposed control.

Keywords Power transmission system · 2-Level voltage source inverter · Grid-forming control · State-feedback control ·
Small-signal stability analysis · Current limitation · Transient power coupling

1 Introduction operation. As a result, electrical power can no longer be

provided to loads. Therefore, the present operation mode
An ever-increasing number of renewable energy production is dramatically changed, while the grid stability still has to
systems and HVDC systems are being connected to the grid. be ensured with the same level of reliability as today, or
This has led to an increase in power electronic devices in the better. To operate autonomously, the control law should be
power system. Nowadays, synchronous generators (SGs) are changed. Power inverters need to change from following the
dominating the electrical grid, establishing a stable volt- grid to leading the grid behavior [1–4]. This capability is
age and frequency that allow for voltage source inverters known as the “grid-forming” concept, where power inverters
(VSIs) to be synchronized at the point of common coupling are able to generate an AC voltage with a given amplitude
(PCC) through a phase-locked loop (PLL) and injecting the and frequency at the PCC.
power to the grid. These inverters are characterized as “grid- The inner control is usually ensured by cascaded PI con-
following” VSIs and behave as current sources. However, trollers. Conventionally, PI controllers are independently
since the installation of these generating units is rapidly tuned by considering a sufficient frequency separation
increasing, some synchronous areas might occasionally between the control loops, i.e., to avoid interactions, where
operate without synchronous machines. In such conditions, the current loop dynamics is limited by the inverter band-
and since the power inverter-based grid-following concept width. This control structure is favored due to the ease of
is not able to form an instantaneous AC voltage [1], the its implementation and its design. In addition to the con-
system may lose synchronism, which can lead to unstable ventional PI controllers, many other control strategies have
been used for grid-following and grid-forming. Predictive
control [5], table-based control [6], fuzzy logic control [7],
* Taoufik Qoria repetitive control [8] and neural network-based control [9]
[email protected] have been proposed for grid-following. For the grid-form-
1
Univ. Lille, Centrale Lille, Arts et Metiers ParisTech,
ing concept, some control techniques have been proposed
HEI, EA2697, L2EP, Laboratoire d’Electrotechnique de such as sliding mode control for the inner current loop and
Puissance, 59000 Lille, France

13
Vol.:(0123456789)
a mixed H2 ∕H∞ for the AC voltage loop [10]. This con- of the developed control to operate stably under normal and
trol technique aims to design a robust controller to achieve abnormal scenarios that can occur in power transmission
high performances. However, this technique is complex and systems. Finally, the developed control is validated on an
requires very high computations. In a recent paper, a frac- experimental test bench.
tional order controller was proposed for grid-forming invert- This paper is organized as follows. Section II presents the
ers. It was motivated by its flexibility and its high number of system modeling of a 2-level VSI and recalls the conven-
degrees of freedom. Among the control strategies proposed tional structure of inner cascaded PI controllers. In Section
for grid-forming inverters, a few studies have discussed the III, the proposed direct AC voltage control embedding the
control design in high-power applications [10, 11] where the current limitation algorithm is presented, analyzed and vali-
switching frequency is lower than 5 kHz. Such condition can dated using time domain simulations. Section IV performs
result in a slow dynamics [11], restrained stability regions some tests in different grid topologies in order to demon-
and interactions between control loops, which can lead to an strate the effectiveness of the developed control. In Section
unstable system [12]. Therefore, controller parameter tuning V, the developed control is validated in an experimental test
in such conditions is still a challenge. bench. Finally, some conclusions are drawn in Section VI.
This paper focuses mainly on VSI grid-forming control
tuning for high-power application. It is an extension of the
work done in [13] and refers to work done in [11]. In the 2 Grid‑forming based on the conventional
latter, the authors proposed an algorithm that deduces the cascaded control structure
cascaded PI controller gains based on eigenvalues location.
This method improves the system stability. However, the AC 2.1 System modeling
voltage response time is still very large; the system is poorly
damped and presents a strong transient coupling between the The system illustrated in Fig. 1 consists of a three-phase
AC voltage and the active power. Moreover, because of the 2-level voltage source inverter represented by a switching
control loop interactions, it is not possible to achieve desired model. It is supplied by a DC voltage source that is assumed
performances. In this paper, the cascaded control structure is to be a DC storage and/or primary source, i.e., PV, wind tur-
replaced with a direct AC voltage regulator based on state- bine, etc., and connected to an AC system through an Lf Cf Lc
feedback control, which aims to enhance the AC voltage filter. The AC system is model by an equivalent AC voltage
dynamics and allows for achieving desired performances. source in series with its equivalent impedance Zg = Rg + jXj,
Employing a system state-space model, the controller gains i.e., the AC system is assumed to be a stiff symmetrical AC
are designed based on a linear quadratic regulator (LQR). system.
Usually, LQR cost factors are designed manually [13–15]. Following the notations in Fig. 1, the state variables are
However, in this paper, a sensitivity analysis is proposed to the VSI output current is through the filter inductor Lf , the
choose the LQR cost factors. The controller tuning considers AC voltage eg across the filter capacitor Cf , and the grid
both the AC voltage response time and the transient coupling current ig through the equivalent transformer inductance Lc.
between the AC voltage and the active power. To assess the For the system analysis, only the average modeling of the
robustness of the developed control, an AC grid impedance system is used. Then, a switching model is used to validate
variation is performed. the proposed control by time domain simulations.
The major disadvantage of the direct AC voltage control The state variables are represented in the d − q frame
is its inability to handle the VSI output current, which can using a Park transform. The following is a state-space model
lead to system damage in case of overcurrent. Indeed, when of a three-phase system at the device level:
compared to synchronous machines that can support up to
seven times over its rated current, power inverters can only
cope with few percent of overcurrent (20–40%). Therefore,
inverters have to be protected against extreme events such
as short circuits and other events that can induce a small
overcurrent such as phase shift, connection of large loads
and tripping of a line protect power converter against over-
current [16]. Since the proposed control does not have a
servo-current control to limit current transients, a parallel
threshold virtual impedance (TVI) [2] is adopted and com-
bined with the proposed AC voltage control.
Several test cases are performed based on single and Fig. 1  Power electronic converter connected to an AC system via an
multi-inverter systems. The aim is to check the effectiveness LCL filter
 

{
ẋ = [A]x + [B]u + [F]w feedforward decoupling Cf 𝜔egdq and compensation igdq . The
y = [C]x (1) inner current control is also is ensured by two propor-
tional–integral (PI) controllers considering the feedforward
where decoupling Lf 𝜔isdq and compensation egdq.
xT = isd isq egd egq igd igq kpv and kiv are the proportional gain and integral gain for
� � � � � � the AC voltage control, respectively. Meanwhile, kpc and kic
uT = vmd vmq , wT = vpccd vpccq , yT = egd egq are the proportional gain and integral gain for the VSI cur-
� 𝜔b � � �
0 0000 001000 rent control, respectively.
T Lf
B = 𝜔b
0 L 0000
, C=
000100 The current control loop generates the modulated voltage
f to the linearization stage that delivers the modulation signals
⎡−
Rf 𝜔b
𝜔𝜔b − Lb
𝜔
0 0 0 ⎤ to the switching stage of the inverter. The control angle 𝜃VSC
⎢ Lf
Rf 𝜔b
f
𝜔
⎥ and the AC voltage reference e∗g are provided by the pri-
⎢ −𝜔𝜔b − 0 − Lb 0 0 ⎥ dq

⎢ 𝜔b Lf f
𝜔b ⎥ mary control based on the droop control P − 𝜔 and Eg − Q
⎢ 0 0 𝜔𝜔b − C 0 ⎥ droop.
𝐀 = ⎢ Cf 𝜔b f
𝜔 ⎥
⎢ 0 Cf
−𝜔𝜔b 0 0 − Cb ⎥ The primary control is expressed by the following
f
⎢ 0 0
𝜔b
0 −
Rc𝜔b
𝜔𝜔 ⎥ equations:
⎢ Lc Lc b ⎥
⎢ 0 𝜔b Rc𝜔b ⎥ mp 𝜔c
⎣ 0 0 L
−𝜔𝜔b − L ⎦ 𝜔VSC − 𝜔set = (pmes − p∗ ) (3)
� 𝜔b � c c
𝜔c + s
0 0 0 0 −L 0
FT = c
𝜔 [ ( ) ]
0 0 0 0 0 − Lb
1
e∗gd ∗
c
(2) − Eset = nq qmes −q (4)
1 + TQ s
All of the variables are expressed in per unit. The vectors
x , u and w are the system state variables, the system inputs where 𝜔VSC, mp and nq are the VSI output frequency, active
supposed to be the modulated voltage vmdq and the system droop gain and reactive droop gain, respectively. The low-
disturbances supposed to be the AC voltage at the PCC vpccdq . pass filter in the reactive power droop aims to filter the
measurement noises. Meanwhile, the low-pass filter used
The output y is the AC voltage egdq . In addition, 𝜔b and 𝜔
in the active power droop aims to simultaneously filter the
are the base value for the angular frequency in rad∕s and the
measurement noises and emulate the inertia effect of the
nominal frequency in per unit, respectively.
synchronous machine [17–19].
Conventionally, the controllers in the cascaded structure
2.2 Conventional control structure
are independently tuned by setting a lower response time
for the inner current loop, i.e., fastest eigenvalues, which
Figure 2 presents the grid-forming cascaded control struc-
is limited by the first-order transfer function approximating
ture. It consists of an inner cascaded AC voltage and a VSI
the PWM effect [20] in (5), and a higher response time for
output current control represented in the synchronous refer-
the outer loops:
ence frame (SRF). The inner voltage control is ensured by
two proportional–integral (PI) controllers considering the

Fig. 2  Conventional control structure of a grid-forming VSI [12]

 ∞

∫ (11)
J= (̄xT Q̄x + uT Ru)dt
0

where Q and R are positive-definite/positive-semidefinite

Hermitian matrices. The first term on the right side of (11)
Fig. 3  Block diagram of direct AC voltage control
x̄ T Q̄x is related to the convergence speed of each state vari-
able. The second term uT Ru accounts for the expenditure of
the control signals energy.
1 The optimum gain matrix K̄ is expressed as follows:
TPWM = 1 (5)
1+ 2fs𝜔
s K̄ = R−1 B̄ T P (12)

where fs𝜔 denotes the switching frequency. where P is the solution of the RICCATI function [22]:
The conventional controller tuning for grid-forming ̄ −1 B̄ T P = 0
Ā T P + PĀ + Q − PBR (13)
inverter-based cascaded structure shows its effectiveness in
terms of ensuring stable operation in the standalone mode, The existence of the matrix P implies the system stability.
but suffers from instability issues in the grid-connected One of the most common (initial) LQR tuning approaches
mode following the analysis in [12, 21]. The instability is is to consider all of the cost factors equally [23], i.e.,
mainly caused by the inner current control loop [12]. Thus, Rk = [I]2×2 and Qi = [I]8×8 . This initial parametrization
the next section presents an alternative control structure is used to check the system stability and the AC voltage
based on direct AC voltage control. dynamics. Thus, the choice of cost factors is improved
thanks to the parametric sensitivity analysis.

3.1 LQR cost factors design

3 Direct AC voltage control for grid‑forming
inverters Based on the parameters listed in Table 1 and the ini-
tial parameterization of the cost factors Qi = [I]8×8 and
The direct AC voltage control depicted in Fig. 3 consists Rk = [I]2×2, the eigenvalues of the linear system (see appen-
of a state-feedback controller and an integral compensator. dix) listed in Table 2 have a negative real part, which results
K and Ki are the pole placement gains ( 2 × 6 ) and the in stable system operation.
servo-controller gains (2 × 2), respectively. In order to check the inner AC voltage loop dynamics,
The vector 𝜁 denotes the error derivative between the AC time domain simulations are shown in Fig. 4. These simula-
voltage references r and the measured output voltage egdq: tions are performed in MATLAB/SimPowerSystem.
[ ]T
r = e∗gd e∗gq (6)

𝜁̇ = r − y = r − Cx (7) Table 1  System parameters Parameter Value

The augmented controlled system matrices can be writ- Pn 1 GW

ten as: cos 𝜑 0.95
[ ] [ ] [ ] [ ] [ ] fn 50 Hz
ẋ A 0 x B 0 F Uac 320 kV
𝜁̇
=
−C 0 𝜁
+
0
u +
I
r +
0
w (8)
Rf, Rc 0.005 p.u
Lf, Lc 0.15 p.u
[ ]
y = C 0 [x 𝜁]T (9) Cf 0.066 p.u
mp 2%
[ ] Eset 1 p.u
u = −Kx + Ki 𝜁 = − K −Ki [x 𝜁 ]T (10) Rg 0.005 p.u
Lg 0.05 p.u
where I denotes the identity
[ matrix.
] 𝜔c 31.4 rad/s
The control gains K̄ = −KKi are tuned based on linear
nq 1e−4 p.u
quadratic control. This provides a control gains matrix based
fsw 4 kHz
on the minimization of the performance index J in (11):
 

Table 2  System eigenvalues Eigenvalue Location parameters. Thus, these sensitivities can identify the param-
eters that strongly impact specified eigenvalues (i.e., 𝜆10−11)
𝜆1−2 −758.65 ± 5507.4i [24].
𝜆3−4 −758.92 ± 4879.2i Since the aim is to improve the dynamic of the states
𝜆5−6 −12.356 ± 311.23i xi , only the cost factors Qi are tuned. The sensitivity sℜ−ℑ
𝜆7−8 −14.652 ± 39.50i of the eigenvalues 𝜆i with respect to the parameter Qi is
𝜆9 −31.416 + 0i expressed by the following expressions:
𝜆10 −0.36445 + 0i
𝜆11 −0.49623 + 0i 𝜕ℜ𝜆i 𝜕ℑ𝜆i
sℜ = , sℑ = . (16)
𝜕Qi 𝜕Qi

The real and imaginary parts of the sensitivities are

associated with the derivatives of the pole location along
the ℜ and ℑ axes, respectively.
Figure 5 shows that only Q7−8 have a large impact on the
eigenvalues 𝜆10−11. Thus, the AC voltage dynamics can be
improved by tuning these cost factors.
The choice of the cost factors is made according to the
following specifications:

• Ensuring a reasonable voltage response and a low over-

shoot.

Fig. 4  AC voltage dynamics for Q = [I]8×8 and R = [I]2×2

A voltage step of Δe∗g = 0.03pu is applied at t = 1s in

dq

both the d− axis and the q− axis.

The obtained results in Fig. 4 show that both the d − q
AC voltage components have a slow first-order response,
where Trd−axis = 8.1s and Trq−axis = 6s:
where Tr denotes the response time.
The dynamics of the d -axis AC voltage corresponds
(a)
exactly to the eigenvalue 𝜆10 = −0.36445 + 0i:
√( )
Tr5% = 3∕ ℜ2 + ℑ2 = 8.1682s. (14)

The q-axis AC voltage dynamics corresponds exactly to

the eigenvalue 𝜆11 = −0.49623 + 0i:
√( )
Tr5% = 3∕ ℜ2 + ℑ2 = 6.0456s. (15)

In order to improve the AC voltage dynamics, it is impor-

tant to find the link between the eigenvalues and the system
parameters to target the parameters that impact the AC volt-
age dynamics. This link is found using a parametric sensi- (b)
tivities tool. The parameter sensitivities of 𝜆i are defined
as the derivative of eigenvalues with respect to the control Fig. 5  Parameter sensitivity analysis, eigenvalues corresponding to
the AC voltage: a d-axis; b q-axis
 Fig. 7  AC voltage change and its impact on the active power

Table 3  Controller parameters

Parameter Value
[ ]
K 0.72 0 1.02 0 −0.73 0
0 0.7197 1.2e − 3 1.014 −0.0004 −0.72
[ ]
KI −38.62 −2.88
0.744 −9.9722

Table 4  System eigenvalues Eigenvalue Location

Fig. 6  Impact of the cost factors Q7−8 on the AC voltage response 𝜆1−2 −758.32 ± 5511i
time
𝜆3−4 −758.94 ± 4882i
𝜆5−6 −12.385 ± 312.97i
• Ensuring a transient power decoupling. This criterion is 𝜆7−8 −14.864 ± 21.38i
not considered in most voltage controller tuning methods 𝜆9 −31.41 + 0i
[10–12, 21]. 𝜆10 −13.982 + 0i
𝜆11 −5.3441 + 0i
A variation of the cost factors Q7−8 results in a varia-
tion of 𝜆10−11. This, in turn, leads to a modification of the
AC voltage response time as depicted in Fig. 6, where it From Figs. 6 and 7, the response time of the AC voltage
is ascertainable that the response time of the AC voltage is set to Tr5% = 200ms. This specification is reasonable for
loop decreases when Q7−8 increase. This demonstrates the power transmission systems applications. This specification
capability and flexibility of the proposed method to achieve corresponds to Q7 = 1500 and Q8 = 100.
different response times for the AC voltage in comparison Based on the defined response time and cost factors, the
with cascaded PI controllers structures [11, 12]. control parameters and the new system eigenvalues are listed
The time domain simulations in Fig. 7 show the impact in Tables 3 and 4, respectively.
of the AC voltage dynamics on the active power. It can Compared to Table 2, Table 5 shows that only 𝜆10−11,
be seen that the increase in the AC voltage rise time (rise which correspond to the voltage loop dynamics, are modi-
time = response time for the first-order response) results in fied while the rest of eigenvalues have not been impacted.
a high transient coupling with the active power. Therefore,
it is important to choose the cost factors in such a way as to 3.2 Comparison with conventional methods
ensure both a low-power transient coupling and a reasonable
AC voltage convergence dynamics. In order to show the relevance of the proposed method, a
comparison with the method in [11] has been performed.
 

Table 5  System parameters Parameter Value the proposed method, where the developed method has a
smooth AC voltage change without overshoot. Moreover,
Pn 1 MW the specified AC voltage dynamics result in a negligible cou-
cos 𝜑 1 pling with the active power.
fn 50 Hz
Uac 690 V 3.3 Control robustness against grid impedance
Rf, Rc 0.003 p.u variation
Lf, Lc 0.1 p.u
Cf 0.2 p.u The robustness of the grid-forming inverter against topologi-
Lg, Rg 0 p.u cal changes, which are modeled as a variation of the grid
impedance and defined by the short-circuit ratio in (17), is
a very important criterion in transmission power systems:
Table 6  Controller parameters
1
Parameter Value SCR = . (17)
Xgpu
[ ]
K 0.73 0 0.52 −0.001 −0.082 −0.001
0 0.73 0 0.53 0.014 −0.073 In Fig. 9, an SCR variation from 20 to 1, i.e., from a
[ ] strong to a very weak grid, is performed for P0 = 1p.u .
KI −20 −1.829
45.72 −0.8091 This shows that the system maintains a stable operation
over a wide SCR range for a critical operating point (i.e.,
SCR ≥ 1.2).

3.4 Grid‑forming inverters protection

against overcurrent

Unlike grid-following inverters, which behave as current

sources, grid-forming inverters behave as voltage sources.
Thus, they are more sensitive to the external disturbances
such as short circuits, tripping lines and heavy-load con-
nections. These disturbances may induce overcurrent that
can damage semiconductor components. Hence, a current

Fig. 8  Comparison with the conventional method. Simulation results

following an AC voltage change

Fig. 9  Impact of the SCR on system stability

The same analysis done in this paper has been applied to the
system parameters used in [11]. These results are listed in
the following table and used only for comparison.
Based on the system parameters in Table 5, the controller
gains deduced from the developed method in this paper are
listed in Table 6.
The performed test case in Fig. 8 consists of applying a
voltage change of − 10% at t = 1 s.
When compared to the conventional method, the obtained
results in Fig. 8 clearly show the improvements brought by Fig. 10  Threshold virtual impedance principle
limitation has to be implemented. Since direct AC voltage In the following, the TVI parameters are set to
control does not possess a current servo-controller to limit Imax = 1.2 p.u, kpR = 1.31 p.u and 𝜎X∕R = 3.
current transients, the threshold virtual impedance (TVI)
VI

depicted in Fig. 10 is adopted.

As illustrated in Fig. 10, the TVI operation principle is 4 Simulation results
divided into three main phases.
In this section, several case studies are used to evaluate
• Overcurrent detection and activation, i.e., activation the developed control. The case studies cover several pro-
only when the current exceeds In = 1 p.u. totypical scenarios based on single-inverter and multi-
• Virtual impedance calculation. inverter systems.
• AC voltage drop calculation.
4.1 Single inverter connected to a grid
The expressions of XVI and RVI are given in (18a) and
(18b), respectively. In addition to the AC voltage dynamics, which are already
{ demonstrated in Fig. 8, this subsection focuses on the system
kpR 𝜎X∕R 𝛿I if 𝛿I > 0
if 𝛿I ≤ 0
XVI = VI
(18a) behavior in the case of a fault.
0
4.1.1 Symmetrical three‑phase short circuit
RVI = XVI ∕𝜎X∕R (18b)
A symmetrical three-phase bolted fault is applied to the
where 𝛿I = Is − In .kpR and 𝜎X∕R are defined as the virtual system at the PCC level as depicted in Fig. 12. At t = 1s, a
VI
impedance proportional gain and the virtual impedance
ratio, respectively.
The parameter kpR is tuned to limit the current magni-
VI
tude to a suitable level Imax during overcurrent, while 𝜎X∕R
ensures a good system dynamics during overcurrent. The
tuning method of these parameters is explained in [25].
When the virtual impedance is activated, new AC volt-
age references are given by (19) and (20):
( )
∗ 1 ∗
egd = Eset + qmes − q np − 𝛿egd (19)
1 + TQ s
Fig. 12  Single inverter subjected to a three-phase short-circuit
e∗gq = −𝛿egq . (20)

The effect of the virtual impedance on the system

depends mainly on the voltage loop dynamics. Therefore,
a direct feedforward on the modulated voltage vmdq can be
added [2]. A block diagram of the direct AC voltage con-
trol embedded in the TVI is depicted in Fig. 11.

Fig. 11  Block diagram of the direct AC voltage control embedded in

the current limitation algorithm Fig. 13  System behavior subjected to a three-phase short circuit
150ms fault occurred. Simulation results are gathered and
presented in Fig. 13.
As can be seen from Fig. 13, the AC voltage drops to zero
during the fault, which leads to an increase in the VSI output
current. The output current amplitude Is reaches 1.65 p.u in
the first 10 ms. Then, it is limited to its maximum allowable
value Imax = 1.2 p.u during the rest of the fault duration.
Once the fault is cleared, the system recovers stably to its
operating point within 300 ms.

4.1.2 Phase shift

Continuing with the system shown in Fig. 12, a phase shift

of 30° is applied to the system at t = 1s as shown by the
grid voltage vg in Fig. 14. During the phase shift, the current
increase is limited to Imax . Then, the system recovers to its Fig. 15  Three-bus system-based grid-forming inverters
operating point in a stable manner.

4.2 Three‑bus system case study Simulation results for the following test cases show the
local AC signals of each power inverter, i.e., AC voltage
In future power systems, grid-forming inverters of different across the capacitors, output current of the LCL filter, active
ratings will be interconnected via transmission lines with power and reactive power.
diverse topologies. To study the behavior of a system and the
performances of the developed control in such a scenario, 4.2.1 P‑Load change
the simple meshed grid of 320 kV illustrated in Fig. 15 is
used. The studied system consists of overhead lines, i.e., Initially, the system is at no load Pload = 0W . Then, a load
L1 = 25 km , L2 = 4L1 and L3 = 5L1, a three-phase resis- change of Pload = 1125MW is applied at t = 0.5s.
tive load, and a three-bus system as shown in Fig. 15. The As shown in Fig. 16, the load change leads to a voltage
switching effect of the power inverters is taken into account. transient in all of the VSIs. The AC voltage drop during the
The LCL filter design of each inverter depends on its rated load connection depends on the electrical distance between
power as indicated in the downstairs of each inverter in the VSIs and the load, i.e., since VSI2 is far from the load
Fig. 15 and the per-unit values listed in Table 1. location, the transient AC voltage drop is more important
than VSI1 and VSI3. Based on the proposed control, the AC
voltage has an overshoot of 0.7% after the load connection
and recovers stability to its nominal value within 80ms.

Fig. 14  System behavior subjected to a phase shift of 30° Fig. 16  System dynamics following a load change
 4.2.3 Symmetrical three‑phase short circuit followed
by line tripping

A 400-ms symmetrical three-phase short circuit has

occurred between two lines as depicted in Fig. 18. Then, it
is cleared by tripping the lines using the switch K1. In this
test case a Pload of 1125MW is assumed to be connected.
As shown in Fig. 18, when a fault occurs, the AC volt-
ages drop by 50% for VSI2 and 60% for VSI1−3. As a result,
it should be noticed that the output current Ig reaches 1.65
p.u in first 30 ms. Then, the output currents of all of the
inverters are limited to 1.2 p.u. Once the fault is cleared,
the system recovers stably to its operating point within
50 ms.
Fig. 17  System dynamics following an AC voltage change

5 Experimental results

The developed control is validated on a small-scale experi-

mental bench like the one illustrated in Fig. 19(a). The
power inverter is supplied by an ideal DC voltage source
of 600 V and connected to a 300 V ph–ph AC grid through
an LCL filter as presented in this paper and shown in

Fig. 18  System behavior during a symmetrical three-phase short cir-

cuit followed by a tripping line

4.2.2 AC voltage amplitude change

Initially, a resistive load of Pload = 1125MW is connected

and the power inverters are driven to there nominal AC volt-
age EgVSC123 = 1pu . Then, a -10% AC voltage amplitude is
applied on VSI1 at t = 2s as shown in Fig. 17.
The AC voltage dynamics of VSI1 corresponds to the
one defined in the specifications, i.e., Tr = 200ms. The AC
voltage change results in a reactive power change with
the same dynamics, as well as a low transient coupling
with the active power as expected. For VSI2 and VSI3, this
change is seen as an external disturbance leading to an
AC voltage transient due to the reactive power exchange
between the VSIs. At t = 3.25s , the AC voltage of VSI1
recovers its nominal value with same dynamics.

Fig. 19  Experimental bench: a mockup presentation; b functional

schema
Table 7  System parameters Parameter Value

Pn 5.62 kW
cos 𝜑 0.9
fsw 10 kHz
Lf 0.15 p.u = 10.91 mH
Cf 0.066 p.u = 9.19 μF
Lc 0.17 p.u = 12.91 mH

Fig. 21  Resistive load change

Fig. 20  AC voltage change and its impact on active power Fig. 22  AC voltage change

Fig. 19(b). The switching frequency of the converter is 5.2 Standalone mode

fsw = 10kHz , and a DSPACE1005 is used as the controller
with a time step of 40 μs. Grid-forming inverters should remain stable regardless of
The mockup parameters are listed in Table 7. The the grid topology and based on the same control law. This
parameter of the LCL filter approximatively corresponds subsection aims to show the effectiveness of the developed
to those used in the paper in per unit, i.e., Table 1. The control in operating in the standalone mode.
control parameters are those listed in Table 3. The power inverter is now connected to a variable resis-
Two possible experimentations have been completed tive load of Pload = 4kW . In Fig. 21, load changes of 0.2 p.u
based on the developed control, i.e., the grid-connected and 0.4 p.u have been applied. The results show that the VSI
mode where the VSI is connected to the laboratory power feeds the load in a stable manner and represents a very weak
system ( Kg is closed), and the autonomous mode where the AC voltage transient.
VSI is connected to a three-phase variable resistive load ( KL In Fig. 22, an AC voltage amplitude change of 10%
is closed). occurs. The obtained results show that the AC voltage
dynamics respect the response time specifications in the
5.1 Grid‑connected mode autonomous mode as in the grid-connected mode.
These simple test cases aim to show the flexibility of the
Similar to the test case shown in Fig. 8, the test case shown developed control and its ability to operate in various grid
in Fig. 20 first brings the system to the steady state with an topologies, as well as its ability to achieve higher dynamic
AC voltage of Eg = 1.1 p.u before it is stepped back to 1.0 performances than the conventional cascaded control.
p.u.
The experimental results in Fig. 20 support the analysis
done in this paper where the AC voltage dynamics respect 6 Conclusions
the imposed specifications, i.e., Tr5% = 200ms and the tran-
sient coupling between the AC voltage and the active power This paper proposes a direct AC voltage control structure
is very small. and its controller tuning for grid-forming inverters. Despite
the fact that the state-feedback control technique is consid-
ered as a conventional one, this paper demonstrates that its
application to the grid-forming inverters in high-power system ⎡ 0 0 − Cb
𝜔
0 0⎤
applications has many advantages when compared to the cas- ⎢ Rc 𝜔b
f ⎥
caded control structure, especially in terms of dynamics. The ⎢ 0 − 𝜔𝜔b 0 0⎥
⎢ 𝜔 Lc
⎥
proposed tuning method for controller gains has allowed for ⎢ Lb −𝜔0 𝜔b −
Rc𝜔b
0 0⎥
achieving the desired AC voltage response time, while taking ⎢ c Lc
⎥
Aaug22 =⎢ 0 0 1 0 0⎥
into account the coupling issues with the active power. ⎢ 1 0 0 0 0⎥
The fundamental issue with using direct AC voltage control ⎢ 0 0 0 0 0⎥
is its inability to protect inverters against overcurrent. Thus, ⎢ ⎥
⎢ 𝜔c Igq 0 𝜔c Egd 0 𝜔c Egd 0 0 0⎥
this paper combines AC voltage control with threshold virtual ⎢𝜔 I
⎣ c gd 0 𝜔c Egq 0 −𝜔c Egd 0 0 0 ⎥⎦
impedance. The choice of this solution is motivated by the ease
of its implementation. Moreover, it does not have any impact
on the system behavior under normal operation. The effective- ⎡ 0 0 0⎤
ness of the proposed control is demonstrated in a number of ⎢ 0 ⎥⎥
Vpccd0 sin (𝛿0 )𝜔b
⎢ −𝜔b Igq0
test cases by time domain simulations and experiments. Lc
⎢ Vpccd0 cos (𝛿0 )𝜔b ⎥
⎢ Lc
𝜔b Igd0 0⎥
Acknowledgements This project has received funding from the Euro-
pean Union’s Horizon 2020 research and innovation program under
Aaug23 =⎢ 0 0 0⎥
⎢ ⎥
grant agreement No 691800. This paper reflects only the author’s ⎢ 0 0 0⎥
views, and the European Commission is not responsible for any use ⎢ 0 −mp 0⎥
that may be made of the information it contains. ⎢ 0 −𝜔c 0 ⎥⎥
⎢
⎣ −𝜔c 0 0⎦
Appendix

Augmented state-space control matrices:

[ ] References
Δisd Δisq Δegd Δegq Δigd Δigq Δ𝜁d Δ𝜁q …
xT =
Δ𝛿Δpmes Δqmes 1. Erlich, I., et al.: New control of wind turbines ensuring stable and
secure operation following islanding of wind farms. IEEE Trans.
Energy Convers. 32(3), 1263–1271 (2017)
⎡ − Rf 𝜔b + 𝜔b K11 𝜔𝜔 + 𝜔b K12 − 𝜔b + 𝜔b K13 ⎤ 2. Denis, G., Prevost, T., Debry, M., Xavier, F., Guillaud, X., Menze,
⎢ Lf Lf b Lf Lf Lf ⎥ A.: The Migrate project: the challenges of operating a transmis-
Aaug11 ⎢
= −𝜔𝜔b + 𝜔b K21
−
Rfpu 𝜔b
+
𝜔b K21
+
𝜔b K21 ⎥ sion grid with only inverter-based generation. A grid-forming
⎢ Lf Lf Lf Lf ⎥ control improvement with transient current-limiting control. IET
⎢ 𝜔b
0 0 ⎥ Renew. Power Gener. 12(5), 523–529 (2018)
⎣ C ⎦
3. Erlich, I.: Control challenges in power systems dominated by con-
f

verter interfaced generation and transmission technologies. In:

NEIS 2017; Conference on Sustainable Energy Supply and Energy
⎡ 𝜔b K14 𝜔b K15 𝜔b K16 𝜔 K 𝜔 K
− bL I11 − bL I12 ⎤ Storage Systems, pp. 1–8 (2017)
⎢ 𝜔b 𝜔b K21 𝜔b K21
Lf Lf Lf
𝜔b K21
f
𝜔b KI21
f
𝜔b KI21 ⎥ 4. Rocabert, J., Luna, A., Blaabjerg, F., Rodríguez, P.: Control of
⎢−L + L + − − ⎥
L
⎢ f 𝜔𝜔 f − 𝜔f b
Lf Lf Lf
⎥ power converters in AC microgrids. IEEE Trans. Power Electron.
Aaug12 =⎢ b Cf
0 0 0 ⎥ 27(11), 4734–4749 (2012)
⎢ 𝜔 ⎥ 5. Judewicz, M.G., González, S.A., Fischer, J.R., Martínez, J.F., Car-
0 0 − Cb 0 0
⎢ f ⎥ rica, D.O.: Inverter-side current control of grid-connected voltage
⎢ 0
Rc𝜔b
− L 𝜔𝜔b 0 0 ⎥ source inverters with LCL filter based on generalized predictive
⎣ c
⎦ control. In: IEEE J. Emerg. Sel. Top. Power Electron., vol. 6, no.
4, pp. 1732–1743 (2018)
6. Hamed, H.A., Abdou, A.F., Acharya, S., Moursi, M.S.E., EL-
Aaug13 = [0]3×3 Kholy, E.E.: A novel dynamic switching table based direct power
control strategy for grid connected converters. In: IEEE Trans.
Energy Convers., vol. 33, no. 3, pp. 1086–1097 (2018)
⎡ 0 0 0000 0 0 ⎤ 7. Hannan, M.A., Ghani, Z.A., Mohamed, A., Uddin, M.N.: Real-
⎢ 𝜔b 0 0 0 0 0 0 0 ⎥ time testing of a fuzzy-logic-controller-based grid-connected
ATaug = ⎢ Cf ⎥ photovoltaic inverter system. IEEE Trans. Ind. Appl. 51(6),
⎢ −𝜔𝜔b ⎥
21 𝜔b
0 1 0 0 𝜔 I −𝜔 I 4775–4784 (2015)
⎣ Lc c g d 0 c gq ⎦
0
8. Chen, D., Zhang, J., Qian, Z.: An improved repetitive control
scheme for grid-connected inverter with frequency-adaptive capa-
bility. IEEE Trans. Ind. Electron. 60(2), 814–823 (2013)
 9. Fu, X., Li, S.: Control of single-phase grid-connected converters 25. Paquette, A.D., Divan, D.M.: Virtual impedance current limiting
with LCL filters using recurrent neural network and conventional for inverters in microgrids with synchronous generators. IEEE
control methods. IEEE Trans. Power Electron. 31(7), 5354–5364 Trans. Ind. Appl. 51(2), 1630–1638 (2015)
(2016)
10. Li, Z., Zang, C., Zeng, P., Yu, H., Li, S., Bian, J.: Control of a grid-
forming inverter based on sliding-mode and mixed $H_2/H_ınfty Taoufik Qoria received his M.S.
$ control. IEEE Trans. Ind. Electron. 64(5), 3862–3872 (2017) degree in Electrical Engineering
11. D’Arco, S., Suul, J.A., Fosso, O.B.: Automatic tuning of cascaded for Sustainable Development
controllers for power converters using eigenvalue parametric sen- from Lille 1 University of Sci-
sitivities. IEEE Trans. Ind. Appl. 51(2), 1743–1753 (2015) ence and Technology, Villeneuve
12. Qoria, T, Gruson, F. Colas, F., Guillaud, X., Debry, M., Prevost, d’Ascq, France, in 2016. Since
T.: Tuning of cascaded controllers for robust grid-forming voltage 2016, he has been working on
source converter. In: 2018 Power Systems Computation Confer- the modeling and control of the
ence (PSCC), pp. 1–7 (2018) power inverters in HVDC and
13. Qoria, T., et al.: Tuning of AC voltage-controlled VSC based HVAC applications. He is pres-
linear quadratic regulation. In: IEEE Milan PowerTech, pp 1–6. ently working toward his Ph.D.
Milan, Italy (2019) degree at ENSAM ParisTech in
14. Vinifa, R., Kavitha, A.: Linear quadratic regulator based current the Laboratory of Electrical
control of grid connected inverter for renewable energy appli- Engineering and the Power Elec-
cations. In: 2016 International Conference on Energy Efficient tronics of Lille (L2EP), Lille,
Technologies for Sustainability (ICEETS), pp. 106–111 (2016) France, which is funded by the
15. Shao, R., Yang, S., Wei, R., Chang, L.: Advanced current control European project H2020 MIGRATE. His current research interests
based on linear quadratic regulators for 3-phase grid-connected include the massive integration of power electronic devices in power
inverters. In: 2015 IEEE 6th International Symposium on Power transmission systems.
Electronics for Distributed Generation Systems (PEDG), pp. 1–5
(2015)
16. Denis, G.: From grid-following to grid-forming: The new strategy
to build 100% power-electronics interfaced transmission system Chuanyue Li (S’16) received his
with enhanced transient behavior. Ph.D thesis, Ecole centrale de B.S. degree in Electrical and
Lille (2017) Electronic Engineering from
17. Qoria, T., Gruson, F., Colas, F., Denis, G., Prevost, T., Guillaud, both Cardiff University, Cardiff,
X.: Inertia effect and load sharing capability of grid forming con- WLS, UK, and the North China
verters connected to a transmission grid. In: 15th IET Interna- Electric Power University, Bei-
tional Conference on AC and DC Power Transmission (ACDC jing, China, in 2013; and his
2019), pp. 1–6 (2019) Ph.D. degree in Electrical and
18. D’Arco, S., Suul, J.A.: Equivalence of virtual synchronous Electronic Engineering from
machines and frequency-droops for converter-based microGrids. Cardiff University, in 2017. He is
IEEE Trans. Smart Grid 5(1), 394–395 (2014) a Postdoctoral Scholar at the
19. Liu, J., Miura, Y., Ise, T.: Comparison of dynamic characteris- Laboratory of Electrical Engi-
tics between virtual synchronous generator and droop control in neering and Power Electronics
inverter-based distributed generators. IEEE Trans. Power Elec- (L2EP), Lille, France. His cur-
tron. 31(5), 3600–3611 (2016) rent research interests include
20. Bajracharya, C., Molinas, M., Suul, J.A., Undeland, T.M.: Under- HVDC control and protection,
standing of tuning techniques of converter controllers for VSC- and power electronics.
HVDC. In: Proc. of Nordic Workshop on Power and Industrial
Electronics, NORPIE 2008, Espoo, Finland, 9–11 June (2008) Ko Oue was born in Hiroshima,
21. Denis, G., Prevost, T., Panciatici, P., Kestelyn, X., Colas, F., Guil- Japan. He received his B.S. and
laud, X.: Improving robustness against grid stiffness, with internal M.S. degrees in Electrical Engi-
control of an AC voltage-controlled VSC. In: 2016 IEEE Power neering from Doshisha Univer-
and Energy Society General Meeting (PESGM), pp. 1–5 (2016) sity, Kyoto, Japan, in 2017 and
22. Kemin, Z., John, C. D., Keith, G.: Robust and optimal control. 2019, respectively. He received
ISBN: 0134565673, 9780134565675 (1996) his M.S. degree from Ecole Cen-
23. Sirisha, C.: Design and implementation: linear quadratic regu- trale of Lille, Villeneuve-d’Ascq,
lator. LAP Lambert Academic Publishing. ISBN: 3846556432, France, in 2018.
9783846556436 (2011)
24. Danielsen, S., Fosso, O.B., Toftevaag, T.: Use of participation
factors and parameter sensitivities in study and improvement of
low-frequency stability between electrical rail vehicle and power
supply. In: 2009 13th European Conference on Power Electronics
and Applications, pp. 1–10 (2009)