IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO.

6, NOVEMBER/DECEMBER 2005

1707

Microgrid Power Quality Enhancement

Using a Three-Phase Four-Wire
Grid-Interfacing Compensator
Yunwei Li, Student Member, IEEE, D. Mahinda Vilathgamuwa, Senior Member, IEEE, and
Poh Chiang Loh, Member, IEEE

AbstractThis paper presents a three-phase four-wire grid-interfacing power quality compensator for microgrid applications.
The compensator is proposed for use with each individual distributed generation (DG) system in the microgrid and consists
of two four-phase-leg inverters (a shunt and a series), optimally
controlled to achieve an enhancement of both the quality of
power within the microgrid and the quality of currents flowing
between the microgrid and the utility system. During utility grid
voltage unbalance, the four-phase-leg compensator can compensate for all the unwanted positive-, negative-, and zero-sequence
voltagecurrent components found within the unbalanced utility.
Specifically, the shunt four-leg inverter is controlled to ensure
balanced voltages within the microgrid and to regulate power
sharing among the parallel-connected DG systems. The series
inverter is controlled complementarily to inject negative- and
zero-sequence voltages in series to balance the line currents, while
generating zero real and reactive power. During utility voltage
sags, the series inverter can also be controlled using a newly
proposed fluxcharge current-limiting algorithm to limit the flow
of large fault currents between the micro- and utility grids. The
performance of the proposed compensator has been verified in
simulations and experimentally using a laboratory prototype.
Index TermsFault current limitation, four-phase-leg inverter, microgrid, power quality compensator, sequence voltages/currents.

I. I NTRODUCTION

ICROGRIDS can generally be viewed as a cluster of

microgenerators connected to the mains utility grid,
usually through some voltage-source-inverter (VSI)-based interfaces. Concerning the interfacing of a microgrid to the utility
system, an important area of study is to investigate the impact
of unbalanced utility grid voltages and utility voltage sags,
which are two most common utility voltage quality problems,
on the overall system performance. As a common practice, if
the utility grid voltages are seriously unbalanced, a separation
device, connected between the microgrid and the mains grid to
provide isolation in the event of mains faults as in Fig. 1(a),

Paper IPCSD-05-064, presented at the 2004 Industry Applications Society

Annual Meeting, Seattle, WA, October 37, and approved for publication in the
IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS by the Industrial Power
Converter Committee of the IEEE Industry Applications Society. Manuscript
submitted for review October 15, 2004 and released for publication August 26,
2005.
The authors are with the Center for Advanced Power Electronics, School
of Electrical and Electronic Engineering, Nanyang Technological University,
Singapore 639798 (e-mail: emahinda@ntu.edu.sg).
Digital Object Identifier 10.1109/TIA.2005.858262

will open and isolate the microgrid. However, when the utility
voltages are not so seriously unbalanced, the separation device
will remain closed, subjecting the microgrid to sustained unbalanced voltages at the point of common coupling (PCC), if
no compensating action is taken. Such an unbalance in voltages
can cause increased losses in motor loads and abnormal operation of sensitive equipment.
An obvious solution is to balance the voltages within the
microgrid using some voltage regulation techniques. However,
large unbalanced currents can flow between the unbalanced
utility grid and microgrid due to the very low line impedance interfacing both grids, if only the microgrid voltages are regulated
[1]. This flow of large currents can overstress semiconductor
devices within the interfacing inverters and the distribution
lines, and is expected to worsen during utility voltage sags when
the voltage differences between the utility grid and the microgrid increase. For low voltage distribution, where microgrids
are usually constructed with a four-wire configuration to supply
both single-phase and three-phase loads, the problem is further
complicated by the flow of zero-sequence currents through the
line and neutral conductors.
To mitigate the above-mentioned complications, this paper
proposes a grid-interfacing power quality compensator for
three-phase four-wire microgrid applications. The proposed
compensator is to be used with each individual distributed
generator (DG), and it consists of two optimally controlled
four-phase-leg inverters (a shunt and a series as in Fig. 1).
Operating together, the two four-leg inverters can compensate
for all the unwanted positive-, negative-, and zero-sequence
voltages/currents within the system, enhancing both the quality
of power within the microgrid and the quality of current flowing
between the microgrid and the utility. During utility voltage
sags, the compensator can also be controlled to limit the flow of
large fault currents using a newly proposed fluxcharge currentlimiting algorithm. The proposed system has been tested in
simulations and experimentally using a laboratory hardware
prototype. Lastly, to assist readers in identifying objectives of
the paper, Table I, summarizing the compensator functionalities
and control features, is included.
II. T HREE -P HASE F OUR -W IRE G RID -I NTERFACING
P OWER Q UALITY C OMPENSATOR
Fig. 1 shows the general layout of the proposed grid-interfacing power quality compensator. The compensator consists

0093-9994/$20.00 2005 IEEE

1708

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO. 6, NOVEMBER/DECEMBER 2005

Fig. 1. Proposed microgrid compensator. (a) Overall system structure. (b) Inverter topology.
TABLE I
POWER COMPENSATOR FEATURES AND CONTROL STRUCTURES

of two four-phase-leg inverters, namely inverter A (shunt) and

inverter B (series). The main functions of inverter A are to
maintain a set of balanced sensitive load voltages within the
microgrid even under unbalanced load and grid voltage conditions, generate and dispatch power, share the power demand
optimally with the other parallel-connected DG systems when
the microgrid islands, and synchronize the microgrid with the
utility system at the instant of connection. In addition, being
connected directly to the microsource, inverter A must have
sufficiently high power capacity with rated voltage and headroom for current higher than the rated value (in fault conditions)
in order to continuously condition the routine energy supplied
by the microsource (similar to most shunt inverter applications
on power systems). On the contrary, the main functions of
inverter B are to maintain a set of balanced line currents by introducing negative- and zero-sequence voltages to compensate
for the grid voltage unbalance, and to limit the flow of large
fault currents during utility voltage sags. This second control

task would require inverter B to carry relatively large current

with a high compensating voltage appearing across it, implying
that inverter B must have high power rating, too.
It is commented that the operational principles of the
compensator presented here are different from those of a
unified power quality conditioner (UPQC), which is usually
constructed using a shunt and a series three-phase-leg inverter.
For a UPQC, the series three-leg inverter injects voltages to
maintain a set of balanced distortion-free voltages at the load
terminals, while the shunt three-leg inverter injects harmonic
compensating currents into the alternating current (ac) system
to shape the supply currents drawn by the UPQC-conditioned
loads as balanced sinusoids. In addition, the UPQC is usually
designed to function with zero real power flow through it during
steady state to minimize the size of the direct current (dc)
link energy storage capacitor [2]. These control objectives of a
UPQC are obviously different from those of the proposed fourphase-leg compensator, where inverter A is used for voltage

LI et al.: MICROGRID POWER QUALITY ENHANCEMENT USING A GRID-INTERFACING COMPENSATOR

Fig. 2. Control scheme for shunt inverter A. (a) Overall control structure.
(b) Voltage control algorithm.

regulation and power control, and inverter B is used for line

current balancing and fault current limiting, as described earlier.
Therefore, for the proposed compensator, the development of
an appropriate control algorithm is challenging since existing
UPQC control algorithms are not directly applicable, and most
references on four-phase-leg compensators in the literature are
for either a shunt or a series four-leg inverter, but not for both
used simultaneously. This control development is presented
next in the remaining sections of the paper.
III. C ONTROL OF S HUNT I NVERTER A
A. Description of Control Algorithm
As shown in Fig. 2(a), the control system of shunt inverter
A contains a voltagecurrent regulation block, and external real
and reactive power control blocks. The detailed design of the
power control blocks has already been presented by Li et al. [3]
and is therefore not duplicated here. Instead, this paper focuses
more on the design of the voltagecurrent regulation block of
shunt inverter A, after giving an introductory description of the
power control blocks for the sake of completeness.
In brief, the power blocks control real and reactive power
flow, and facilitate power sharing between the paralleled DG
systems when a utility fault occurs and the microgrid islands.
This sharing of power is achieved with no physical communication link between the DG systems by introducing artificial real
power versus supply frequency and reactive power versus
voltage magnitude droop characteristics into the power loops.
Outputs of the power controllers are the reference frequency ,
phase , and the positive-sequence voltage magnitude E. These

1709

variables are subsequently transformed to stationary 0

frame quantities {V , V , V0 }, used as reference voltages for
the voltagecurrent regulation block.
The external power blocks also incorporate synchronization
algorithms for ensuring smooth and safe reconnection of the
micro- and utility grids when the fault is cleared. Synchronization can be achieved by aligning the voltage phasors at
the microgrid and utility ends of the separation device whose
location is indicated in Fig. 1(a), and it can conveniently be implemented by adding two separate proportional and integral (PI)
synchronization regulators to the external real and reactive
power control algorithms [see Fig. 2(a)]. Inputs to these PI
regulators are the magnitude E and phase errors of the
two voltage phasors at both ends of the separation device, and
their outputs are fed to the real and reactive power loops to make
E 0 and 0, ensuring close tracking of the voltage
phasors at both ends of the separation device. As mentioned
earlier, details of these power control algorithms can be found
in an earlier publication [3].
Fig. 2(b) shows the voltagecurrent regulation scheme for
inverter A, which contains an inner filter inductor current control loop and an outer load voltage control loop. For the outer
voltage loop, the theoretical analysis is first presented in the
synchronous dq0 reference frames to assist in understanding the control principles. The developed voltage controllers are
subsequently converted back to the stationary frame to allow for
easier physical implementation.
As illustrated in Fig. 2(b), the three-phase load voltages are
first transformed from the stationary abc frame to the stationary 0 frame using the following quad transformation
matrices [4]


2
T0 Uabcn
(1)
U0 =
3
U0 = [U U U0 ]T
(2)
Uabcn = [Uan Ubn Ucn Un ]T

21
0
1
12

3
T0 = 0
23
0
2
1
1
1

23 2
2 2
2 2
2 2

(3)
(4)

where Un is a placeholder quantity, usually chosen to be 0

[4]. The 0 voltages are subsequently transformed to
the forward-rotating positive-sequence synchronous frame
(indicated with superscript +) and reverse-rotating negativesequence frame (indicated with superscript ) using the
following transformation matrices:

+
= Ud+ Uq+
Udq0

cos(t)
+
= sin(t)
Tdq0
0


Udq0
= Ud Uq

cos(t)

Tdq0
= sin(t)
0

U0+

+
= Tdq0
U0

sin(t) 0
cos(t) 0
0
1

T

U0 = Tdq0
U0

sin(t) 0
cos(t) 0 .
0
1

(5)
(6)
(7)
(8)

1710

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO. 6, NOVEMBER/DECEMBER 2005

Note that the first two rows of T0 , Tdq0

, and Tdq0
resemble the commonly used three-phase abc to
to dq transformation. Equations (1)(8) can therefore be
viewed simply as an extension from the three-phase abc
to to dq transformation with only an independent
zero-sequence row added. It would therefore be convenient to
control the dq voltages independently from the zero-sequence
voltages, as indicated in Fig. 2(b) with the zero-sequence control path drawn external to the dotted box enclosing the dq
control paths.
To achieve zero steady-state errors when controlling the
positive- and negative-sequence dq voltages, two PI controllers are used along the positive- and negative-sequence dq
control paths. Practically, the implementation of these dq controllers can be computationally inefficient due to the numerous
coordinate frame transformations needed [6]. A more efficient
form of the voltage controllers can be derived by transforming
both the positive- and negative-sequence PI controllers to the
stationary frame using either the state-space or frequencydomain technique [5], [6]. Performing this inverse transformation with kp+ = kp = kp and ki+ = ki = ki , the frame
voltage controllers can be expressed as

is
2kp + s22k+
2
0

0
is
2kp + s22k+
2

(9)

Effectively, these voltage ( P + resonant) controllers introduce an infinite gain at the positive (50 Hz) and negative
(50 Hz) fundamental frequencies to force the positive- and
negative-sequence voltage errors to zero [see Fig. 3(a) for an
example bode plot of (9) using a positive resonant frequency
of 50 Hz], and can conveniently be implemented in the stationary frame with minimum computational requirements
[6]. The same P + resonant compensator with the resonant
frequency set at 50 Hz can also be used for controlling the zerosequence voltage, again with zero steady-state error, as shown
in Fig. 2(b). Note that for this work, the resonant frequency is
set by inverter A real power versus supply frequency droop
controller and not the utility grid frequency. A mechanism to
track the grid frequency for determining resonant condition (or
transformation phase for synchronous PI implementation) is
therefore not required.
The outputs of the individual voltage controllers when added
together give the demanded reference currents I , I , and
I0 for the inner filter inductor current loop. This inner loop
is implemented using only proportional controllers with peak
current limiting in the stationary 0 frame as any steadystate error in this loop would not affect the outer voltage loop
accuracy substantially. The outputs of the current controllers
are then transformed back to the abc frame and sent to
a four-phase-leg pulsewidth-modulation (PWM) modulator for
switching shunt inverter A.
B. Closed-Loop Transfer Functions
As far as the dynamics of the outer voltage loop is concerned,
the inner current loop is considered to be having fast dynamic
response. The inner loop can therefore be represented as a

Fig. 3. Bode plots of P + resonant compensators using (a) (9) and (b) (15)
with kP = 1, ki = 20, = 314 rad/s, and cut = 10 rad/s.

constant gain and the closed-loop transfer functions of inverter

A control scheme can be derived in the frame as
V =

Cs3

2kp s2 + 2ki s + 2 2 kp
V
+ 2kp s2 + ( 2 C + 2ki )s + 2 2 kp

(10)

and along the zero-sequence axis as

V =

kp0 s2 + 2ki0 s + 2 kp0

V .
Cs3 + kp0 s2 + ( 2 C + 2ki0 )s + 2 kp0 0

(11)

Both (10) and (11) are third-order transfer functions, whose

transient and tracking performance can be improved by increasing the proportional gains {kp , kp0 } and resonant gains {ki , ki0 },
respectively, at the expense of stability. For this work, a tradeoff
is obtained by choosing 2kp = kp0 = 0.1 and ki = ki0 = 100
(other system parameters are listed in Table II), which gives
the same bode plot in Fig. 4 for both (10) and (11). This figure
generally indicates good fundamental compensation and rapid
high-frequency attenuation of the controller with an infinite

LI et al.: MICROGRID POWER QUALITY ENHANCEMENT USING A GRID-INTERFACING COMPENSATOR

1711

TABLE II
SYSTEM PARAMETERS

Fig. 4.

Bode plot of closed voltage loop of inverter A.

gain margin, 61.9 phase margin, 3.1-ms step response time,

and 22.5% overshoot.
IV. C ONTROL OF S ERIES I NVERTER B D URING
U TILITY V OLTAGE U NBALANCE
A. Description of Control Algorithm
The circuit connection of series inverter B is shown in
Fig. 1(a), where a four-leg inverter is again adopted for the
control of zero-sequence component. The main function of the
series inverter is to maintain a set of balanced line currents;
i.e., to force the negative- and zero-sequence currents to zero in
the three-phase four-wire system. Note that this inverter is not
designed to control the positive-sequence currents, which are
already (directly) regulated by shunt inverter A (in the power
control loop).
Fig. 5(a) shows the control scheme for series inverter B,
where an inner voltage loop is shown embedded within an
outer current loop. The outer current loop (also referred to as
the reference voltage generator) functions to generate reference
voltages for the inner voltage loop using the negative- and
zero-sequence line currents as inputs. As seen, the measured

Fig. 5. Control scheme for series inverter B. (a) General representation.

(b) Outer current loop in the negative synchronous frame. (c) Inner
voltage loop.

line currents are first transformed to the negative synchronous

dq0 reference frame using (1) and (7), and regulated with
zero reference values using PI controllers for the negativesequence dq currents and a resonant controller for the zerosequence component [see Fig. 5(b)]. (Note that the
resonant controllers described in Section III compensate for
both positive- and negative-sequence components and therefore
cannot be used here for solely compensating the negativesequence component. In addition, note the use of a nonideal
resonant controller along the zero-sequence control path, which
is particularly needed for this work due to experimental constraints described in the next subsection.)
Before feeding into the PI controllers, the current signals
should be filtered to remove positive-sequence currents, which
appear as ac signals at twice the fundamental frequency, leaving
only dc negative-sequence currents for compensation. This
filtering is needed for enhancing the robustness of the control
loop and is performed by averaging the dq currents over half a
fundamental cycle in the negative synchronous frame [8]. Quite
obviously, a degradation associated with this half fundamental

1712

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO. 6, NOVEMBER/DECEMBER 2005

cycle filtering would be the overall poorer controller transient

response. Also shown in Fig. 5(b) are measures taken to decouple the d- and q-control paths (indicated by dotted lines
in the figure) to arrive at three decoupled d-, q-, and 0-axis
control paths, which can be tuned independently. The outputs
of this outer reference voltage generator, consisting only of
negative- and zero-sequence components, are then transformed
back to the 0 frame and fed into the inner voltage
control loop.
The block representation of the inner voltage loop is shown
in Fig. 5(c), where P + resonant controllers are used along the
0 control paths. The use of P + resonant controllers in
the inner control loop will force the filtered capacitor voltages
VC of the series inverter to track the demanded negativeand zero-sequence reference voltages VC with zero steadystate errors and eliminate any positive-sequence component
since its reference value is kept at zero for the series inverter.
The series inverter therefore injects only negative- and zerosequence voltages into the system to maintain a set of balanced
line currents with no real and reactive power generation (or
absorption) in the steady state.
Note that an inner filter inductor current loop can be added
to the voltage control loop to give a better dynamic response.
However, due to the slow response of the outer reference voltage generation loop, a single inner voltage loop is considered
sufficient, and this loop can reasonably be represented by a
unity gain when designing the control scheme.
B. Closed-Loop Transfer Functions
As a common practice, the design of the proposed control
algorithm for the series inverter begins with the inner voltage
loop in the 0 frame. Analyzing Fig. 5(c) with Kinv =
2/Vdc , the closed-loop transfer function of the inner voltage
loop can be derived as in (12), shown at the bottom of the
page, where RLine and LLine , which normally represent line
resistance and inductance, are here lumped together with the
series transformer winding parameters for convenience. When
performing zero-sequence analysis, neutral line parameters are
also lumped into RLine and LLine .
To maintain stability, the Rouths stability criteria of (13) and
(14) must be met for the inner voltage loop
2ki L2Line
> 0
RLine
2kp RLine + 2ki LLine + RLine + Lf Cf RLine 2




2
+ L2f Cf RLine 2
2ki Lf Cf RLine

> 0.
2k L2Line
Lf Ri Line
Lf

(13)

(14)

To confirm the robustness of the inverter B control scheme, a

typical distribution system where the line reactance is either
the same or smaller than the line resistance [7] is considered. With the series transformer leakage impedance lumped
together with the line impedance for simplicity, and even with
an onerous condition of XLine = (2f )LLine = 2RLine for
this distribution system, the stability condition (13) becomes
Lf (2ki RLine / 2 f 2 ) > 0, which can easily be met. Consequently, (14) can also be met since the term in (14) is
now minimized by having a relatively larger denominator than
numerator (the numerator is usually small due to the presence
of a microfarad capacitive term Cf ).
Once the design of the inner voltage loop is completed, the
next step is to design the outer reference voltage generator
in the negative synchronous reference frame. In the negative
synchronous dq frame, positive-sequence currents, which
appear as ac signals at twice the fundamental frequency, should
be removed using half fundamental cycle filtering [8] to
enhance the robustness of the control loop. Another consideration is to identify an alternative nonideal resonant controller
[see Fig. 5(b)] for use along the zero-sequence control path,
whose transfer function is expressed as [6]

kp0 +

2ki0 cut s
s2 + 2cut s + 2

(15)

where cut represents the low-frequency cutoff. An example

bode plot of (15) is given in Fig. 3(b), which obviously shows
a wider resonant peak with a large finite gain, as compared to
that of (9) given in Fig. 3(a). Equation (15) is therefore less
sensitive to frequency fluctuation, and the steady-state error
would be still kept relatively small. This nonideal controller
should generally be used in cases with resonant frequency
variations, and it is particularly applicable to the experimental
series inverter implemented in Section VIII. In that prototype,
the inverter neutral phase-leg is asynchronously switched by
an external analog comparator and triangular signal generator due to the shortage of synchronized PWM channels on
the dSPACE DS1103 controller card. This causes slight frequency variation in the neutral line current waveform, whose
effect is nullified by the 0 transformation matrix in
(4) along the outer negative-sequence PI control path. However, along the 0-axis control path, its effect is not nullified,
and, therefore, the nonideal P + resonant compensator with a
wider resonant bandwidth is needed for reducing the system
sensitivity level.
Incorporating these fine-tuning schemes, the final closedloop transfer functions of the proposed control algorithm can



GVcl = kp LLine s3 + (2kp RLine + 2ki LLine )s2 + (2kp 2 LLine + 2ki RLine )s + 2kp 2 RLine

Lf Cf LLine s5 + Lf Cf RLine s4 + (Lf + LLine + 2kp LLine + Lf Cf LLine 2 )s3 + (2kp RLine + 2ki LLine



+ RLine + Lf Cf RLine 2 )s2 + (2kp + 1) 2 LLine + 2 Lf + 2ki RLine s + (2kp + 1) 2 RLine

(12)

LI et al.: MICROGRID POWER QUALITY ENHANCEMENT USING A GRID-INTERFACING COMPENSATOR

1713

Fig. 7. Fluxcharge control scheme for series inverter B.

Fig. 6.

Open-loop Bode plot of zero-sequence control path.

be written as (16) in the negative dq frame and (17) (shown at

the bottom of the page) along the 0-axis
GIcl =

kp s + ki
.
LLine s2 + (RLine + kp )s + ki

(16)

Using (16) and setting kp = 1.5 and ki = 80, the outer

control loop gain margin, phase margin, and settling time are
specified as infinity, 114 , and 0.2 s, respectively, with the long
settling time inherently caused by the half fundamental cycle
response time needed for outer loop filtering. For the 0-axis,
setting kp0 = 16, ki0 = 200, and cut = 10 rad/s, and using the
open-loop bode plot in Fig. 6 for the zero-sequence control path
with the neutral line impedance included (see Table II for the
neutral impedance parameters), the gain and phase margins are
set to infinity and 89.7 , respectively. Together, these selected
control parameters ensure the stability of series inverter B.

V. C ONTROL OF C OMPENSATOR D URING U TILITY

V OLTAGE S AGS (F AULT C URRENT L IMITATION )
The control schemes, presented in the earlier two sections,
regulate the compensator well during normal operating conditions, but not during utility voltage sags. Referring to Fig. 7,
sag drops below its nominal value, and
at the start of a sag, V
shunt inverter A should now disable its external power control
algorithms and set its voltage references ({V , V } in Fig. 2) to
some appropriate values according to the load sensitivity level.
As an example, consider the (ideal) case where {V , V } and,
D , are kept
hence, the shunt inverter terminal voltage phasor V
constant at their nominal values with their phase-angles locked

GIcl

at the presag values to prevent the tripping of sensitive loads.

D constant, however, gives rise to a large voltage
Maintaining V
sag ) across series inverter B and the distribution

phasor (VD V
line, resulting in large fault currents flowing along the line. If
these fault currents are allowed to flow through the line until
clearance of the sag or opening of the electromechanical circuit
breaker, there is a risk of damaging semiconductor devices in
the inverters.
A possible ride-through scheme for the above-mentioned
complications would be to control series inverter B to act as
a virtual current-limiting inductor L0 immediately after the
sensing of a voltage sag. Sensing of voltage sag is indirectly
done here by measuring the local three-phase line currents since
in actual cases, the PCC can be located far away from the
power compensator, and, therefore, its voltages are inaccessible
to the compensator. Upon the instantaneous current value of a
phase exceeding a specified threshold IT , inverter B is invoked
to inject a large virtual inductance L0 in series with the line
impedance {RLine , LLine } to limit the current flowing along the
sag . By keeping V
E V
sag , smooth
E V
line, maintaining V
recovery from a voltage sag can also be ensured by simply
sag back to its nominal
E , which will rise with V
sensing V
value. The fault current limiting action of inverter B can then
be inhibited and the power algorithms of inverter A can be
reenabled.
One way of controlling inverter B as a virtual inductor
is to use the flux-model control concept reported in [9] and
[10]. In passing, it is commented that the predictive fluxmodel control presented in [9] and [10] is highly sensitive to
system parameter variations, which is a common feature for
most predictive schemes. An alternative robust fluxchargemodel control, with an outer flux-model and an inner chargemodel loop, is therefore proposed in this paper for controlling
inverter B, as shown in Fig. 7. The control variable used for the
outer flux
 model is the inverter filtered terminal flux, defined
as = VC dt, where VC is the filter capacitor voltage of
the series inverter. This flux variable is compared against a



kp0 s2 + 2cut kp0 + ki0 s + 2 kp0

 (17)


=
LLine s3 + 2cut LLine + RLine + kp0 s2 + 2 LLine + 2cut RLine + 2cut kp0 + 2cut ki0 )s + 2 (RLine + kp0

1714

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO. 6, NOVEMBER/DECEMBER 2005

Fig. 8. Bode plots of (a) closed charge loop and (b) open flux loop.

reference flux given by ref = L0 iLine , where iLine is the line

current and the negative sign is to indicate that iLine is drawn
flowing out of the series inverter in this paper. The flux error
is then fed to the flux regulator, implemented using the P +
resonant compensator given in (9).
It can be shown through classical control analysis that this
single flux model cannot damp out the resonant peak of the
inductancecapacitance (LC) filter connected to the output of
inverter B (analytical details are not shown here due to space
limitation). To stabilize the system, an inner charge model is
therefore
 added to force the filter capacitor charge, defined as
Q = iC dt = Cf VC , where iC and Cf are the current through
and capacitance of the filter capacitor, to track the reference
charge output Qref of the flux regulator.
Intuitively, the presented fluxcharge model can be derived
from the voltagecurrent control presented in Section III-A
for shunt inverter A by integrating the voltage and current
variables. The minor difference is that the capacitor current is
to be integrated for the fluxcharge model rather than the filter
inductor current. Integrating the capacitor current would also

mean measuring the capacitor voltage. Therefore, the system

was implemented by sensing the capacitor voltage, and
 Q can
then be calculated directly from VC (noting that Q = iC dt =
Cf VC ) without requiring additional current sensors. Although
sharing common control principles, the fluxcharge control
does have an implementation advantage over voltagecurrent
control. The additional integration in fluxcharge control would
allow its outer loop flux reference to be easily computed by
multiplying L0 and iLine without differentiation (differentiation
of iLine is needed for voltagecurrent control). The resulting
flux reference is therefore more accurately calculated (without
phase delay and noise amplification associated with differentiation), and the fluxcharge-controlled series inverter would
emulate virtual inductor L0 more closely.
The calculated charge error is then fed to the charge regulator
with a designed transfer function of kTd s/(1 + (Td s/N )).
Note from this expression that a derivative term is found in
its numerator. Intuitively, this derivative term neutralizes the
effects of voltage and current integrations at the inputs of
the fluxcharge model, resulting in the proposed algorithm
having the same regulation performance as the multiloop
voltagecurrent feedback control. This similarity in performance is confirmed by plotting the closed-loop response of
the charge model in Fig. 8(a) with kTd = 1000 and Td /N =
0.00002. As anticipated, the resulting bode plots have bandpass characteristics with a near-unity gain at the fundamental
frequency, similar to those of an inner capacitor current loop
(example bode plots of an inner capacitor current loop can be
found in [11]).
Lastly, to confirm the stability of the proposed fluxchargemodel control, its open-loop Bode plots are plotted in Fig. 8(b)
with 2kp = 0.08 and 2ki = 160. The plotted figure clearly
shows the system exhibiting good performance at the fundamental frequency with all high-frequency switching components effectively attenuated. (Closed-loop response of the
fluxcharge-model control is very similar to that of shunt
inverter A in Fig. 4 and is therefore not explicitly shown.)
VI. S IMULATION R ESULTS
Simulation results using Matlab/Simulink are provided to
verify the effectiveness of the proposed system. The system
parameters used for simulation are given in Table II. These
represent typical distribution system parameters as given in
[7]. Fig. 9(a) shows the voltage profile used for simulation
where the grid voltages are initially balanced. From t = 1.2 s
onwards, negative sequence voltages of 0.1 per unit (p.u.) and
zero-sequence voltages of 0.1 p.u. are added to the original
balanced voltages.
With the proposed grid-interfacing compensator added,
Fig. 9(b) and (c) shows that the sensitive load voltages in
the microgrid are kept well balanced even under unbalanced
utility voltage conditions. This clearly verifies the performance
of shunt inverter A in maintaining good power quality within
the microgrid. Fig. 10(a) and (b) shows the extremely unbalanced line currents under grid voltage unbalance and without
the functioning of series inverter B, which can cause system
complications, as described in Section I. By operating series

LI et al.: MICROGRID POWER QUALITY ENHANCEMENT USING A GRID-INTERFACING COMPENSATOR

Fig. 9. Simulated waveforms of (a) mains grid voltages, (b) abc frame
sensitive load voltages in the microgrid, and (c) 0 frame sensitive load
voltages in the microgrid.

1715

Fig. 11. Simulated filtered voltages generated by series inverter B. (a) abc
frame. (b) 0 frame (zoomed-in view in the steady state).

Fig. 12. Simulated line currents with series inverter B. (a) abc frame.
(b) Negative dq0 frame.
Fig. 10. Simulated line currents without series inverter B. (a) abc frame.
(b) Negative dq0 frame.

inverter B, these unbalanced line currents can be mitigated,

as illustrated in Figs. 11 and 12. Fig. 11(a) shows the filtered
voltages [VCan , VCbn , and VCcn ; see Fig. 1(a)] generated by
series inverter B during the transient period of inverter B activation (from 1.15 to 1.5 s), while Fig. 11(b) shows the zoomedin view of the voltages in steady state (from 1.9 s onwards),
which clearly show the presence of only negative- and zerosequence components. The performance of series inverter B
is further verified by observing the line currents shown in
Fig. 12(a) and (b), where the initial unbalanced line currents
become balanced again after 0.2 s later. This 0.2-s time lapse
is anticipated due to the half fundamental cycle averaging
process used in inverter B outer control loop, as described in
Section IV-B.
Fig. 13(a)(d) shows the power generated by the two inverters. As illustrated in these figures, shunt inverter A dispatches a
constant real power of 300 W and a reactive power of 160 Var
in steady state, while series inverter B constantly maintains zero
real and reactive power output. The zero power output of series
inverter B is expected since it injects only negative- and zerosequence voltages into a system where only positive-sequence
line currents flow.

Fig. 13. Simulated waveforms of (a) real power supplied by shunt inverter A,
(b) reactive power supplied by shunt inverter A, (c) real power supplied by
series inverter B, and (d) reactive power supplied by series inverter B.

The current-limiting action of the compensator is also confirmed by simulating an unbalanced utility voltage sag (VGa =
32 44.7 , VGb = 66.4 170 , and VGc = 54.5 38.6 )

1716

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO. 6, NOVEMBER/DECEMBER 2005

Fig. 14. Simulated waveforms of (a) mains grid voltages and (b) abc
frame sensitive load voltages in the microgrid.

Fig. 16. Simulated waveforms of (a) real power supplied by shunt inverter A,
(b) reactive power supplied by shunt inverter A, (c) real power supplied by
series inverter B, and (d) reactive power supplied by series inverter B.

real and reactive power absorbed by series inverter B. As

confirmed in Fig. 16(c), series inverter B absorbs no real active
power since it acts as a fictitious inductor during the sag period.
VII. E XPERIMENTAL R ESULTS

Fig. 15. Simulated waveforms of (a) abc frame voltages injected by series
inverter B and (b) abc frame line currents.

at t = 1.2 s in Fig. 14(a). From the load viewpoint, shunt

inverter A should ideally maintain the microgrid voltages at
their nominal values (power control algorithms of inverter A
are disabled during the sag, as mentioned earlier). This ideal
case, however, does not give the worst scenario for performance
evaluation of the compensator since it would mean a step
voltage disturbance only on the utility side of the compensator.
An alternative case where the shunt inverter regulates its voltages at 90% of nominal is therefore assumed here to test the
compensator performance when voltage steps are introduced on
both its utility grid and microgrid sides while still drawing high
line currents. Fig. 14(b) shows the corresponding simulated
waveforms for inverter A, where, as anticipated, the microgrid
voltages are maintained at 90% of their desired nominal values
during the sag.
With the proposed fluxcharge algorithm implemented, the
series inverter now acts like a large virtual inductor connected
in series with the distribution feeder, limiting the line currents
to a specified peak value of 6 A. The voltages across this
inserted virtual inductor and the limited line currents are shown
in Fig. 15(a) and (b), respectively. Lastly, Fig. 16(a) and (b)
shows the increase in real and reactive power supplied by shunt
inverter A during the sag, while Fig. 16(c) and (d) shows the

To verify the performance of the proposed compensator

experimentally, a hardware prototype that has the same system
parameters as in the simulation has been built in the laboratory
(Table II). For the experimental system, a stiff programmable ac
source connected to a resistanceinductance (RL) load is used
to represent the utility grid and is connected to an emulated microgrid through a three-phase back-to-back silicon-controlled
rectifier (SCR) isolation switch. The microgrid consists of the
proposed compensator [a shunt four-leg insulated gate bipolar
transistor (IGBT) inverter and a series four-leg IGBT inverter
with an injection transformer] and a connected RL load. The
compensator is controlled using a dSPACE DS1103 controller
card with a relatively low sampling frequency fsa of 5 kHz
used due to the long computational time needed for controlling two inverters simultaneously and the use of Simulink
auto-code generation without optimization. (Note that the inverter PWM pulses are generated by embedded peripherals
in the slave TMS320F240 processor mounted on the DS1103
card. The inverter switching frequency fsw can thus be set at
10 kHz even though the sampling frequency of the main
DS1103 processor is only 5 kHz.) According to digital control theory, an immediate impact of using a lower sampling
frequency would then be the poorer experimental response as
compared to its simulated counterpart (this has been verified
in simulation, but the simulated waveforms with distortions are
not shown here due to space limitation).
To ensure proper starting of the experimental system, the
utility grid and the microgrid are powered up separately. After
synchronizing the voltages at both ends of the SCR switch,
the switch is closed to connect the micro- and utility grids,
allowing the system to transit smoothly into the grid-connected
mode of operation. Fig. 17(a)(c) shows the utility voltages and

LI et al.: MICROGRID POWER QUALITY ENHANCEMENT USING A GRID-INTERFACING COMPENSATOR

Fig. 17. Experimental waveforms of (a) mains grid voltages, (b) abc
frame sensitive load voltages in the microgrid, and (c) 0 frame sensitive
load voltages in the microgrid.

Fig. 18. Experimental filtered voltages generated by series inverter B.

(a) abc frame. (b) 0 frame (zoomed-in view in the steady state).

microgrid load voltages, which are initially balanced. At t =

3.775 s, the utility voltages become unbalanced with 0.1-p.u.
negative-sequence and 0.1-p.u. zero-sequence voltage components added. Despite this unbalance in utility voltages, the load
voltages in the microgrid are kept balanced by controlling shunt
inverter A. Similarly, to balance the currents flowing between
the micro- and utility grids, series inverter B is operated with
its injected negative- and zero-sequence voltages shown in
Fig. 18(a) and (b).
As compared to their simulated counterparts in Fig. 11, the
experimental waveforms in Fig. 18(b) are observed to
be oscillatory, whereas the 0-axis waveform is (nearly) undistorted. As mentioned earlier, the relatively low experimental
sampling frequency (fsa = 5 kHz = 0.5fsw ) used can cause
oscillatory response in the system. Along the control
path for series inverter B (see Section IV-A), this oscillatory
distortion is averaged out by the half fundamental cycle filter
and is therefore not eliminated by the PI controllers. On the
other hand, along the 0-axis control path where no filtering is
performed, this distortion is directly compensated and therefore
has lesser effect on the system.

1717

Fig. 19. Experimental line currents with series inverter B. (a) abc frame.
(b) Negative dq0 frame.

Fig. 20. Experimental waveforms of (a) real power supplied by shunt inverter
A, (b) reactive power supplied by shunt inverter A, (c) real power supplied by
series inverter B, and (d) reactive power supplied by series inverter B.

The effectiveness of series inverter B in balancing the line

currents by suppressing negative- and zero-sequence current
components is shown in Fig. 19(a) and (b). These experimental
waveforms are less damped as compared to their simulated
counterparts in Fig. 12, and the cause is identified in simulation
to be the less accurate half fundamental cycle averaging associated with a low experimental sampling rate. Fig. 20(a)(d)
shows the real and reactive power supplied by the shunt and
series inverters. These figures clearly show the shunt inverter
supplying its specified dispatched power of 300 W and 160 var,
and the series inverter injecting zero real and reactive power, as
anticipated.
The performance of the compensator during a voltage sag
is also verified experimentally by programming the ac source
to emulate a voltage sag (similar to that used for simulation)
at t = 0.807 s, as shown in Fig. 21(a). Fig. 21(b) shows the
microgrid load voltages kept constant at 90% of the nominal
value during the transition from normal to sag condition, while
Fig. 22(a) and (b) shows the voltages across and limited line
currents through series inverter B, which now acts as a large

1718

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO. 6, NOVEMBER/DECEMBER 2005

Fig. 21. Experimental waveforms of (a) mains grid voltages and (b) abc
frame sensitive load voltages in the microgrid.
Fig. 23. Experimental waveforms of (a) real power supplied by shunt inverter
A, (b) reactive power supplied by shunt inverter A, (c) real power supplied by
series inverter B, and (d) reactive power supplied by series inverter B.

balance the line currents, generating zero real and reactive

power during compensation. During utility voltage sags, the
series inverter can also be controlled to limit the flow of large
fault currents using the proposed fluxcharge control algorithm.
The proposed system has been tested in simulations and experimentally using a laboratory prototype with satisfactory results
obtained for confirming the effectiveness of the compensator.
R EFERENCES
Fig. 22. Experimental waveforms of (a) abc frame voltages injected by
series inverter B and (b) abc frame line currents.

virtual inductor. These waveforms again exhibit some transient

oscillations, which can be eliminated by setting fsa = fsw or
2fsw (commonly adopted for optimized digital implementation), as verified in the simulation. Lastly, the corresponding
increase in real and reactive power supplied by shunt inverter
A during the sag are shown in Fig. 23(a) and (b), while the real
and reactive power absorbed by series inverter B are shown in
Fig. 23(c) and (d). Fig. 23(c) again shows the zero real power
flow to inverter B.
VIII. C ONCLUSION
This paper presents a three-phase four-wire grid-interfacing
power quality compensator for compensating the two most
common voltage quality problems, namely voltage unbalance
and voltage sag, in a microgrid. During grid voltage unbalance,
the proposed compensator, using a shunt and a series fourphase-leg inverter, can enhance both the quality of power within
the microgrid and the quality of currents flowing between the
microgrid and utility system. Functionally, the shunt four-leg
inverter is controlled to maintain a set of balanced distortionfree voltages within the microgrid and to regulate power sharing
among the parallel-connected distributed generation (DG) systems. To complement, the series four-leg inverter is controlled
to inject negative- and zero-sequence voltages in series to

[1] M. Prodanovic and T. C. Green, Control and filter design of three-phase

inverters for high power quality grid connection, IEEE Trans. Power
Electron., vol. 18, no. 1, pp. 373380, Jan. 2003.
[2] A. Ghosh and G. Ledwich, Power Quality Enhancement Using Custom
Power Devices. Boston, MA: Kluwer, 2002.
[3] Y. W. Li, D. M. Vilathgamuwa, and P. C. Loh, Design, analysis and realtime testing of controllers for multi-bus micro-grid system, IEEE Trans.
Power Electron., vol. 19, no. 5, pp. 11951204, Sep. 2004.
[4] M. J. Ryan, R. W. De Doncker, and R. D. Lorenz, Decoupled control of
a four-leg inverter via a new 4 4 transformation matrix, IEEE Trans.
Power Electron., vol. 16, no. 5, pp. 694701, Sep. 2001.
[5] C. B. Jacobina, M. B. de R. Correa, T. M. Oliveiro, A. M. N. Lima, and
E. R. C. da Silva, Current control of unbalanced electrical systems,
IEEE Trans. Ind. Electron., vol. 48, no. 3, pp. 517525, Jun. 2001.
[6] D. N. Zmood, D. G. Holmes, and G. H. Bode, Frequency-domain analysis of three-phase linear current regulators, IEEE Trans. Ind. Appl.,
vol. 37, no. 2, pp. 601610, Mar./Apr. 2001.
[7] P. Heine, A. Lehtonen, and E. Lakervi, Voltage sag analysis taken into
account in distribution network design, in Proc. IEEE Power Tech. Conf.,
Porto, Portugal, 2001.
[8] H. Awad and J. Svensson, Compensation of unbalanced voltage dips
using vector-controlled static series compensator with LC-filter, in
Conf. Record IEEE Industry Applications Society (IAS) Annu. Meeting,
Pittsburgh, PA, 2002, pp. 904910.
[9] H. Funato, T. Ishikawa, and K. Kamiyama, Transient response of three
phase variable inductance realized by variable active-passive reactance
(VAPAR), in Proc. IEEE Applied Power Electronics Conf. (APEC),
Anaheim, CA, 2001, vol. 2, pp. 12811286.
[10] H. Funato, K. Kamiyama, and A. Kawamura, Transient performance of
power circuit including virtual inductance realized by fully digital controlled variable active-passive reactance (VAPAR), in Proc. IEEE Power
Electronics Specialists Conf. (PESC), Galway, Ireland, 2000, vol. 3,
pp. 11951200.
[11] N. M. Abdel-Rahim and J. E. Quaicoe, Analysis and design of a multiple feedback loop control strategy for single-phase voltage-source UPS
inverters, IEEE Trans. Power Electron., vol. 11, no. 4, pp. 532541,
Jul. 1996.

LI et al.: MICROGRID POWER QUALITY ENHANCEMENT USING A GRID-INTERFACING COMPENSATOR

Yunwei Li (S04) received the B.Eng. degree in electrical engineering from Tianjin University, Tianjin,
China, in 2002. He is currently working toward the
Ph.D. degree in the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore.
From February to July 2005, he was with the
Institute of Energy Technology, Aalborg University,
Aalborg, Denmark, as a Visiting Scholar.
Mr. Li is a member of the IEEE Industry Applications Society (IAS).

D. Mahinda Vilathgamuwa (S90M93SM99)

received the B.Sc. and Ph.D. degrees in electrical engineering from the University of Moratuwa,
Katubedda Moratuwa, Sri Lanka, and Cambridge
University, Cambridge, U.K., in 1985 and 1993,
respectively.
He joined the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore, in 1993, as a Lecturer, where he is
now an Associate Professor. His research interests
are power electronic converters, electrical drives, and
power quality. He has published more than 80 research papers in refereed
journals and conference proceedings.
Dr. Vilathgamuwa is the Co-Chairman of The Power Electronics and Drives
Systems Conference 2005 (PEDS05).

1719

Poh Chiang Loh (S01M03) received the B.Eng.

(Hons.) and M.Eng. degrees from the National University of Singapore, Singapore, in 1998 and 2000,
respectively, and the Ph.D. degree from Monash University, Monash, Australia, in 2002, all in electrical
engineering.
During the summer of 2001, he was a Visiting
Scholar at the Wisconsin Electric Machine and Power Electronics Consortium, University of Wisconsin,
Madison, where he worked on the synchronized
implementation of cascaded multilevel inverters and
reduced common-mode carrier-based and hysteresis control strategies for
multilevel inverters. From 2002 to 2003, he was a Project Engineer at the
Defence Science and Technology Agency, Singapore, managing major defense
infrastructure projects and exploring new technology for intelligent defense
applications. Since 2003, he has been an Assistant Professor at Nanyang
Technological University, Singapore.