You are on page 1of 6

DQ-axis Current-based Droop Controller

Ibrahim Alsaleh1 and Lingling Fan2

� � �
Abstract— During an islanding operation of a microgrid
comprised of multiple distributed generation (DG) units serving �

a load, DG units interfaced with voltage-sourced converters
(VSC) are regarded as the only energy sources. Hence, DG units
should be controlled to form the frequency and voltage, pick up �

the full load, and respond to any load changes. Conventional


P − f /Q − V droop controller enables a communication-less �
load sharing among DGs while maintaining close deviations of
frequency and voltage. Utilizing the dq-frame power flow anal-
ysis, this paper presents an equivalent primary droop controller
id −f /iq −V based on the direct and quadrature components of … �
the feeder currents, which are already existent in each voltage
controller for disturbance rejection. More emphasis is placed on
the phenomenon of inaccurate reactive power sharing caused by
mismatched feeder impedances. Analysis is presented in terms
of the droop equation, and compensation methods are examined Fig. 1: Microgrid structure.
with focus on computation and communication requirements,
and PCC voltage profile. For validation of the controller’s
response, a testbed model exhibiting the dynamics of a two-
DG microgrid is developed and simulated in Matlab/Simulink. a phyical inductance. A thourough analysis of the virtual
impedance concept is presented in [2]. In [3], a simple static
I. INTRODUCTION virtual inductance based on the fundamental frequency is
proposed for power decoupling.
The gradual reliance of renewable energy sources in
power production has re-shaped the modern picture of the Another problem is encountered when feeders have mis-
utility grid. The power generation is no longer exclusive matched impedances, provoking inaccurate reactive power
to power plants, thanks to the distributed generation (DG) sharing among DG units. This issue has been garnering a
units which are integrated into the distribution network. In lot of attention, and numerous improvements on the con-
fact, the high penetration of DG units has led to dividing the ventional droop control have been developed to increase
grid into microgrids, constituting the futuristic smart grid the adaptability of DG units to the feeder impedance. [4]
[4]. The microgrid technology potentially facilitates control- proposed a strategy that injects a small active power distur-
lablity over a number of paralleled DG units connected to bance in the system. Although results show accurate power
a point of common coupling with the grid, and can operate sharing, processing the disturbance signal entails complex
autonomously realizing UPS features [1]. implementation, especially in a noisy environment [5]. In
Load sharing can be achieved with or without com- [3], the effect of approximated voltage drop across the feedr
munication. Schemes that employ communication require reactance, X/Vref , is estimated and incorporated into the
high-bandwidth communication channels, which reduce the droop coefficient. The estimation takes place during grid-
reliablity and microgrid expandability as dispersed DG units connected operation. By using data exchanged via commu-
would rely on one another. The P −f /Q−V droop controller; nication between DG local controllers, the strategy proposed
on the other hand, achieves the desired communication-less in [5] tunes the adaptive virtual impedance without any
operation, and has broadly been adopted [1],[3]-[6]. estimation of the feeder impedances.
The conventional droop controller counts on the notion When the d-axis terminal voltage is aligned with its
that microgrids have high X/R ratios, while the feeder space vector, PDG and QDG are proportional to the dq-
impedance in low-voltage distribution networks is typically axis currents. In this paper, we explore a droop controller
resistive in practice. This inroduces coupling between the based on the dq-axis currents, and show that it can achieve
active and reactive powers. A few solutions have been accurate load sharing. Unlike the conventional droop, the
proposed to dominate the inductive component using vir- droop coefficient in the modified droop is directly related
tual impedance, and thereby avoiding any installations of to the feeder reactance, since currents multiplied by gains
are deemed virtual voltage drops. Therefore, the adaptive
1 I. Alsaleh is with the Dept. of Electrical Engineering, University
virtual impedance is employed to correct the reactive power
of South Florida, Tampa FL 33620, USA; and the Dept. of Electrical
Engineering, University of Hail, Hail, KSA (e-mail: ialsaleh@mail.usf.edu)
sharing. With knowledge of the PCC voltage phasor obtained
2 L. Fan is with the Faculty of Electrical Engineering, University of South by a single-direction communication signal from the central
Florida, Tampa FL 33620, USA (e-mail: linglingfan@usf.edu) controller shown in Fig. 1, the virtual impedance is realized.
� ��

��

DG , DG �, �


� ∠� 0 �� ∠0° B. Power Sharing Analysis
� � �
𝑉𝑡the
First, (𝑡)power flow in Fig. 2a at the sending and receiving
ends, and across the feeder is expressed in the dq frame.

� �
After that, the power sharing in Fig. 3 will be examined
��
with particular regard to feeder impedances, assuming X1 =
X + δX and X2 = X.
(a) All phasors in Fig. 2a are transformed to the dq frame
with respect to the terminal voltage phase angle, θt . Thus the
𝐼𝑚 terminal voltage at the fundamental frequency is expressed
𝜔(𝑡) as follows
𝑖𝑑 𝑉𝑡
𝛾𝑜 𝑅𝑒
𝑉𝑡𝑑 Vtd + j0 = (VLd + jVLq ) + (Rk + jXk )(ikd + jikq ) (4a)
𝑖𝑞 𝑖
Vtd = VLd + Rk ikd − Xk ikq (4b)

(b) To obtain the powers, (4b) is multiplied by the scaled


conjugated feeder current 23 i∗kdq = 32 (ikd − jikq )
Fig. 2: (a) One-line diagram of a single-loaded DG unit. (b)
Space vectors and the dq components in the complex plane. ( 3 )Vtd ikd +j ( −3 )Vtd ikq = ( 3 )VLd ikd +j ( −3 )VLd ikq
| 2 {z } | 2 {z } | 2 {z } | 2 {z }
PDG k QDG k PL QL

II. S TEADY S TATE A NALYSIS + ( 32 )ikd (Rk ikd − Xk ikq )


| {z } (5)
A. id − f /iq − V Droop Controller Ploss k

During islands, local controllers must be self-aware so that + j ( 32 )ikq (Xikq − Rk ikd )
| {z }
DG units pick up their share of the load with proportion Qloss k
to their capacities. The P − f /Q − V droop controller is
1) Without Droop: From (5), it is inferred that powers at
conventionally adopted, which is based on a predominant
the DG terminals are influenced by the feeder resistance and
inductive behavior of the feeder achieved by physical or static
reactance. For a droopless power sharing of equally-rated
virtual inductnace. Therefore, when (θz ≈ 90◦ ) and a small
DG units in Fig. 3, QDG1 <QDG2 and PDG1 <PDG2 , which
phase angle, θt , are considered, the power flow equations
is dangerous as one unit could be forced to supply more than
between the two buses in Fig. 2a are expressed as
its capacity.
Vt VL XPDG In order to accurately share the load without droop con-
PDG = sin θt ⇒ θt ≈ (1a)
X Vt VL trollers, and in proportion to DG units’ capacities, assuming
Vt (Vt − VL ) XQDG their output voltage phasor Vt ∠θt are exactly the same, the
QDG = ⇒ (Vt − VL ) ≈ (1b)
X Vt following constraints must be satisfied
PDG and QDG are proportional to the phase angle and volt- a) Even Sharing: When all DG units have the same
age difference; respectively. The droop control should output capacities, resistances and reactances of both feeders must
the reference value for the terminal (capacitor) voltage phasor be unified.
Vt ∠θt . The formed phase angle, θt , is used to transform the
R1 = R2 X1 = X2 (6)
terminal voltages, filter currents and feeder currents to the
dq frame, resulting in [Vtd 0]T , [if d if q ]T and [id iq ]T , b) Uneven Sharing: When DG units have different
where quadrature currents are negative because their space capacities, resistances and reactances must be inversely pro-
vectors lag the voltage space vector. Fig. 2b illustrates the portional to ikd and ikq ; respectively.
space vectors of the terminal voltage and feeder current, and
their corresponding dq components in the complex plane. R1 i1d = R2 i2d X1 i1q = X2 i2q (7)
As a result, the dq-frame active and reactive powers derived
from S = 32 Vtd (id − jiq ) are Put simply, voltage drops across the feeders must be unified.
2) With Droop: Despite the impedance’s influence on
3 3 active powers, the id − f droop should promise accurate
PDG = Vtd id QDG = − Vtd iq (2)
2 2 active power sharing because the frequency is a global
id and (−iq ) track the DG active and reactive powers; re- quantity in the microgrid, hence (id = i1d = i2d ). On
spectively. The conventional droop can therefore be replaced the other hand, the iq − V droop achieves accurate reactive
with current sharing id − f /iq − V as follows power sharing if and only if feeders have matched voltage
drops, satisfying (6)-(7). Otherwise, DG units will generate
ω = ω ∗ − mid (3a)
disparate reactive powers (i1q 6= i2q ) as shown in Fig. 4a,
Vtd = Vd∗ + niq (3b) and compensation is therefore required.
� ��

��

DG 1 ,
�,
DG 1

� 1 ∠� 1 �� ∠0° QDG1
=
n + X2
(11)
� � �1 � + ��
1 QDG2 n + X1
In contrast, if the conventional droop were employed with
� 1 the voltage drop effect approximated as 2(XQDG +RPDG )
3Vd∗ , the
� � ratio would be
��
2X2
QDG1 n+ 3Vd∗
DG 2 , DG 2
= 2X1
(12)
QDG2 n+
� �
� 2 ∠� 2 3Vd∗
�2 2 � The compensation methods based on the constraint in (10)
are as follows
𝑍𝑜1 𝑍1 𝑖1 � 𝑉𝑠 ∠0 2𝑖2𝑍𝑜1 𝑍2 𝑍1 𝑍𝑜21)𝑖 Large
1 𝑉𝑠 ∠0Droop
𝑖2 𝑍2 𝑍𝑜2By enlarging n in (10), the
Coefficient:
� � feeder reactances will be outweighed, and the ratio will
𝑃2𝑚𝑎𝑥 𝑉𝑖𝑛𝑣 1 𝑃1 , 𝑄1 𝑃2 , 𝑄2 𝑉𝑖𝑛𝑣 1 𝑃1 , 𝑄1
approximately 𝑃be
2 , 𝑄2unity.

𝜔 i1q n+X
𝜔𝑚𝑖𝑛 𝑡 = 2𝑠 DG units conected to a load with X1 >X2 .
Fig. 3: Two nX = ≈1 (13)
i2q n + X + δX
However, n is bounded by the voltage deviation. With the
𝜔 𝑉 𝑉 voltage deviation being fixed at 5%, it can be concluded that
𝜔∗ 𝑉𝑡1,2 𝐺𝐶 𝑉𝑡1,2 𝐺𝐶 n depends primarily on the reactive power rating. Therefore,
𝑉𝐿 𝐺𝐶 𝑉𝐿 𝐺𝐶 for small reactive power ratings, iq is small and n can be set
𝑚1
𝜔𝐷𝐸𝑅1
𝑉𝑡1
𝑛1
𝑉𝑡1 𝑛1
large to limit the error. Although difficult to achieve accuracy,
𝑚2
𝜔𝐷𝐸𝑅2 = 𝜔𝑚𝑖𝑛 n partially compensates for the mismatched impedance.
𝑉𝑡2 𝑉𝑡2
𝑛2 𝑛2
𝑉𝐿 𝑉𝐿 2) Adaptive Mismatch Virtual Reactance: The adaptive
𝑃𝑖𝑛𝑗 1 𝑃𝑖𝑛 𝑗2

𝑃
virtual reactance, δXv , is calculated (X1 − X2 = δXv ),
𝑃 𝑃𝐷𝐸𝑅2−𝑖1𝑞 −𝑖2𝑞
−𝑖𝑞 −𝑖𝑞
𝑃𝐷𝐸𝑅1 −𝑖1𝑞,2𝑞 and incorporated into n. This suggests that different droop
(a) (b) coefficients should be assigned to result in a unity ratio as
Fig. 4: iq − V droop with (a) uncompensated slopes (b) i1q n2 + X n∗ + X
= = ∗ = 1 (14)
compensated slopes with adaptive virtual reactance. i2q n1 + X + δX (n − δXv ) + X + δX
However, the method requires communication between DG
local controllers to exchange feeder reactances and calculate
C. Correction of Reactive Power Sharing the mismatch.
For simplification, even sharing case is considered with 3) Adaptive Feeder Virtual Impedance: It is based on
two equally-rated DG units using the id − f /iq − V droop compensating for the entire feeder impedance. With knowl-
controller, where an accurate reactive power sharing is equiv- edge of the PCC voltage as shown in Fig. 1, the virtual
alent to having matched i1q and i2q . impedance is realized utilizing separate real and imaginary
equations in (4a) as follows.
With effective terminal voltage regulation, (Vtd = Vd∗ +
ikq
niq ) is substituted into (4), yielding Vd − VLd − ikd VLq
Rvk = Rk = i2
kq
Vd∗ + ni1q = VLd + R1 i1d − X1 i1q (8a) ikd + ikd (15)
Vd∗ + ni2q = VLd + R2 i2d − X2 i2q (8b) −VLq − Rk ikq
Xvk = Xk =
ikd
The quadrature-current error is By incorporating the virtual voltage drops, (8) becomes
i1d
X2 i2q − X1 i1q + (R1 − R2 )id Vd∗ + (n∗ + Rv1 − Xv1 ) i1q = VLd + R1 i1d − X1 i1q (16a)
i1q − i2q = (9) i1q
n | {z }
n1
Assuming inductive feeder impedance, and (R1 − R2 = 0), i2d
Vd∗ ∗
+ (n + Rv2 − Xv2 ) i2q = VLd + R2 i2d − X2 i2q (16b)
the following ratio is obtained i2q
| {z }
n2
i1q n + X2 n+X
= = (10) This method does not require further communication. After
i2q n + X1 n + X + δX
realizing the virtual impedance, the signal carrying the PCC
Therefore, for accurate reactive power sharing, the ratio voltage phasor can be disrupted. Moreover, increasing the
should be constrained to unity. (10) can also be expressed terminal voltage by the amount of the voltage drop improves
in terms of reactive powers, with the approximation QDG ≈ the PCC voltage. The 5% voltage deviation will be allocated
−3 ∗
2 iq Vd for the PCC voltage rather than the terminal voltage.
Droop Controller �∗ DG1
�1 + � 1 �
_

�1 Voltage/Current Controllers RLC Plant �1



� � ��
�1� + + � + � ∗ _ + _
1 �� _
1
+ _ + + _
�� (�) + _ +
�� (�)
+ _
÷ Limiter ×
+ �� + �� + + �
�1
+
�1
�� �
� � ��� � �
� dq
2 abc
� � ��� � �
�1 �1�
� 1�
dq _ + ��∗ � + _ 1 ��� _ 1
0 +
�� (�) + + _
�� (�)
+
÷ Limiter ×
+ _ �� + �� + _ �
� 1�
� abc + +

���
�1� �1�


�1 �2 ��
�1
1 �1
� Feeder-Load
�2
�� Equations �2

�� �� ��1

�2
DG2

Fig. 5: Block diagram of a two-DG microgrid with linear loads.

III. S IMULATION : T WO -DG M ICROGRID


if d `i (s) ωB
A mathematical model is built in Matlab/Simulink to Gi = ∗ = = (18)
id 1 + `i (s) s + ωB
depict the dynamics of a microgrid with two DG units
connected to a load through mismatched feeders. b) Outer Voltage Controller: The parameters of
A. Testbed Modeling Kv (s) are tuned to design a stable open-loop gain,
Kv (s)Gi (s)(Cf s)−1 , which has double integrators. A des-
1) RLC Plant and Averaged VSC Models: As shown in ignated stable phase margin, typically selected within the
Fig. 5, each DG-unit block consists of an averaged VSC and limit (45◦ 6 θpm 6 60◦ ), can be fulfilled by adopting the
RLC plant models. For their simple tracking requiremnts, Symmetrical Optimum method [7].
the PI controllers are used to regulate the inductor current
and capacitor (terminal) voltage at prespecified operating 1 − sin θpm √
α= kvp = C
�� f αωB kvi = αωB kvp (19)
points in the dq frame. The RLC plant is also expressed in 1 + sin θpm
the dq frame, and Laplace transformed to obtain a consoli- √ 2
dated MIMO framework. The PI parameter tuning is well- Vtd αωB (s + αωB )
Gv (s) = = √ 2 √ (20)
established in [6]-[7], and presented as follows Vd s + ωB s2 + αωB
3 3
s + α3 ωB
a) Inner Current Controller: The proportional and in-
Therefore, upon choosing ωB and θpm , parameters of both
tegral gains of Ki (s) are tuned to cancel the low-frequency
Ki (s) and Kv (s) can be designed. Moreover, udq and i∗dq
RL pole and set the current transfer function bandwidth, ωB .
are modified with feedforward loops to reject the disturbance
kpi s + kii of the capacitor voltage and feeder current, and to decouple
kpi = Lf ωB kii = Rf ωB `i (s) = (17)
s(Lf s + Rf ) the dq current and voltage loops as shown in Fig. 5.
c) Droop Controller: The id − f /iq − V controller is reactance is introduced to F eeder 1 at t = 5s, inaccurate
added as the outmost loop to provide the operating point Vd , power sharing is triggered as analyzed from (5).
and improved by the adaptive feeder virtual impedance. 350

Active Power (W)


2) Feeder-load Network Model: Assuming linear loads 270
contribute larger than non-linear loads, harmonic powers are
omitted, and a parallel resistive-inductive load is adopted. 175 P DG1

With ~ikabc and ~iLabc identified as the feeder and load state P DG2

variables and V~tkabc as inputs, the network’s dynamic model 80


2 3 5 7

is expressed as a state space matrix Time (s)


(a)
180

Reactive Power (VAR)


 

~i
 −RT 1 −RL RL ~i1abc

d ~1abc   −R
L1 L1
−RT 2
L1
RL ~
135
i2abc =  L2L L2  i2abc
 
L2
dt ~ RL RL −RL ~iLabc QDG1
iLabc LL LL LL
75
QDG2
1 
L1 0 
~
 30
1  Vt1abc
2 3 5 7
+ 0 L2 ~t2abc
Time (s)
V (b)
0 0
where RT k = Rk + RL , k=1,2. Fig. 6: Load sharing without droop.
The matrix is computed using Matlab embedded function, 2) With id − f /iq − V Droop: n and m are tuned such
where first-order derivatives are integrated and injected as that ∆f = 0.05Hz, and ∆V = 5V. While an accurate
inputs with the DG terminal voltages. The load and feeder active power sharing is ensured in Fig. 7a, the error between
mismatch inductance are injected using step functions to QDG1 and QDG2 in Fig. 7bis minimized because of the
emulate load change and reactive power inaccuracy. q n. Also, it is noted from Fig. 7c that the PCC voltage,
large
2 + V 2 , is deteriorated due to the feeder voltage drops,
VLd Lq
TABLE I: Simulation Parameters which can adversely impact voltage-dependent loads.
350
Active Power (W)

Quantity Value 270


DG terminal voltage 110V (rms)
Nominal frequency 60Hz 175 P DG1

Rf = 10mΩ P DG2

RLC filter Lf = 10mH 80


Cf = 20µF 2 3 5 7
Time (s)
R1,2 = 5mΩ
Feeder parameters L1,2 = 5mH (a)
180
Reactive Power (VAR)

δL1 = 2mH
ωB = 2000 rad/s 135
Controller parameters θpm = 53◦
m = 0.15
QDG1
n=5 75
QDG2

30
2 3 5 7
Time (s)
B. Operation Scenario
(b)
Considering the following scenario, the testbed model in 100
Voltage Magnitude (V)

V t1
Fig. 5 is validated with two DG units that should evenly V t2
share 540W and 270 VAR. The simulation parameters are 90 VL

listed in Table I. 85
1) The microgrid is initially grid-tied, and the PCC volt- 80
age is provided. Each DG supplies 175W and 75VAR. 2 3
Time (s)
5 7

2) At t = 3 s, the microgrid is islanded. Each DG unit (c)


should properly pick up its share of the full load, which 60.01
60
is 270W and 135 VAR. f1
Frequency (Hz)

f2
3) At t = 5 s, a mismatch inductance, δL1 , is instilled
into F eeder 1.
59.95
Conducted case studies and simulation analysis are pre- 59.94
sented as follows 2 3
Time (s)
5 7

1) Without Droop: The microgrid is islanded without a (d)


droop controller. Fig. 6 shows an accurate power sharing with
matched feeder impedances. However, when the mismatch Fig. 7: Load sharing with id − f /iq − V droop.
3) Adaptive Mismatch Virtual Reactance: At t = 7s, the with and without the droop was analyzed and constraints on
mismatch virtual reactance, ωδLv1 , is incorporated into n1 feeder impedance were addressed for accurate power sharing.
as in (14). Consequently, the error between QDG1 and QDG2 Since the active power sharing is not a concern with the
is further minimized as shown in Fig. 8a. Therefore, by as- droop controller, compensation methods were examined to
signing different droop coefficients, the constraints of having accomplish accurate reactive power sharing via modification
unified voltage drops in (6) and a unity ratio in (14) are on the iq − V droop coefficient. A testbed model of a
satisfied. However, acquisition of δLv1 can be cumbersome, microgrid with two DG units was built in Matlab/Simulink,
especially with a large number of DG units connected to which could also be extended to include k-DG units. It was
mismatched feeder impedances, and may require additional shown that a large coefficient could be assigned for the iq −V
communication links. Furthermore, the PCC voltage is still droop minimizing the error between reactive powers to an
deteriorated in Fig. 8b. acceptable extent and with a 5% voltage deviation. Two
other compensation methods based on the adaptive virtual
180
Reactive Power (VAR)

impedance are also shown in the simulation results which


135 further minimized the error. Although both of which exhibit
similar efficacy, the adoption of the adaptive feeder virtual
QDG1
75
QDG2
impedance offers the advantages of less communication and
30 computation requirements, and maintains the load voltage at
2 3 5 7 9
Time (s)
the PCC bus within the designed voltage deviation.
(a)
100
R EFERENCES
Voltage Magnitude (V)

V t1 [1] J. M. Guerrero, L. Hang and J. Uceda, “Control of distributed


V t2 uninterruptible power supply systems,”IEEE Trans. Ind. Electron., vol.
VL
90 55, no. 8, pp. 2845-2859, Aug. 2008.
[2] X. Wang, Y. W. Li, F. Blaabjerg and P. C. Loh, “Virtual-impedance-
85
based control for voltage-source and current-source converters,”IEEE
80 Trans. on Power Electron., vol. 30, no. 12, pp. 7019-7037, Dec. 2015.
2 3 5 7 9
[3] Y. W. Li and C. N. Kao, “An accurate power control strategy for
Time (s)
power-electronics-interfaced distributed generation units operating in
(b) a low-voltage multibus microgrid,”IEEE Trans. on Power Electron.,
vol. 24, no. 12, pp. 2977-2988, Dec. 2009.
Fig. 8: iq − V droop with mismatch virtual reactance. [4] J. He and Y. W. Li, “An enhanced microgrid load demand sharing
strategy,”IEEE Trans. on Power Electron., vol. 27, no. 9, pp. 3984-
4) Adaptive Feeder Virtual impedance: Fig. 9a shows 3995, Sept. 2012.
an accurate reactive power accomplished alternatively by [5] H. Mahmood, D. Michaelson and J. Jiang, “Accurate reactive
incorporating the entire feeder impedance into the iq − V power sharing in an islanded microgrid using adaptive virtual
impedances,”IEEE Trans. on Power Electron., vol. 30, no. 3, pp. 1605-
droop as in (16). With this method activated at t = 7s, not 1617, March 2015.
only is a near zero error between QDG1 and QDG2 obtained, [6] L. Fan, “Frequency and voltage control in a microgrid,”in Control and
but the PCC voltage is also improved and kept at the designed Dynamics in Power Systems and Microgrids, CRC Press, 2017.
[7] A. Yazdani and R. Iravani, “Controlled-frequency vsc system,”in
voltage deviation. For this, more powers will be exerted. Voltage-Sourced Converters in Power Systems: Modeling, Control, and
Applications, Hoboken, NJ: John Wiley & Sons, Inc., 2010.
180
Reactive Power (VAR)

135

QDG1
75
QDG2

30
2 3 5 7 9
Time (s)
(a)
100
Voltage Magnitude (V)

V t1
V t2

90 VL

85

80
2 3 5 7 9
Time (s)
(b)

Fig. 9: iq − V droop with feeder virtual impedance.

IV. CONCLUSIONS
This paper presents a modified droop controller based
on the dq-axis feeder current sharing. The power sharing

You might also like