IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 55, NO. 8, AUGUST 2008 2845
Control of Distributed Uninterruptible
Power Supply Systems
Josep M. Guerrero, Senior Member, IEEE, Lijun Hang, and Javier Uceda, Fellow, IEEE
Abstract—In the last years, the use of distributed uninterrupt-
ible power supply (UPS) systems has been growing into the mar-
ket, becoming an alternative to large conventional UPS systems.
In addition, with the increasing interest in renewable energy
integration and distributed generation, distributed UPS systems
can be a suitable solution for storage energy in microgrids. This
paper depicts the most important control schemes for the parallel
operation of UPS systems. Active load-sharing techniques and
droop control approaches are described. The recent improvements
and variants of these control techniques are presented.
Index Terms—Droop method, load sharing, microgrids, parallel
connection, uninterruptible power supply (UPS).
I. INTRODUCTION
D ISTRIBUTED generation (DG) is an emerging conceptto decentralize the management of electricity production.
However, DG makes no sense without distributed storage en-
ergy systems. Thus, the parallel operation is a special fea-
ture of high-performance uninterruptible power supply (UPS)
systems [1]–[21]. The parallel connection of UPS inverters is
a challenging problem that is more complex than paralleling
dc sources, since every inverter must properly share the load
while staying synchronized. In theory, if the output voltage of
every inverter has the same amplitude, frequency, and phase,
the current load could equally be distributed. However, due
to the physical differences between the inverters and the line
impedance mismatches, the load will not properly be shared.
This fact will lead to a circulating current among the inverters
that can damage or overload them.
The fast development of digital signal processors has brought
about an increase in control techniques for the parallel opera-
tion of UPS inverters. These control schemes can be classified
into two main groups with regard to the use of control wire
interconnections [7]. The first one is based on active load-
sharing techniques, and the major part of them is derived from
control schemes of parallel-connected dc–dc converters, such
as centralized [22], [23], master–slave (MS) [24]–[32], average
load sharing (ALS) [33]–[41], and circular chain control (3C)
[42], [43]. Although these control schemes achieve both good
output-voltage regulation and equal current sharing, they need
Manuscript received February 16, 2008; revised April 7, 2008. Published
July 30, 2008 (projected).
J. M. Guerrero is with the Department of Automatic Control and Com-
puter Engineering, Technical University of Catalonia, 08036 Barcelona, Spain
(e-mail: josep.m.guerrero@upc.edu).
L. Hang is with the College of Electrical Engineering, Zhejiang University,
Hangzhou 310027, China.
J. Uceda is with the División de Ingeniería Electrónica, Universidad Politéc-
nica de Madrid, 28006 Madrid, Spain.
Digital Object Identifier 10.1109/TIE.2008.924173
critical intercommunication lines among modules that could
reduce the system reliability and expandability.
The second kind of control scheme for the parallel op-
eration of inverters is mainly based on the droop method
[44]–[74]. This technique consists of adjusting the output-
voltage frequency and amplitude in function of the active and
reactive power delivered by the inverter. The droop method
achieves higher reliability and flexibility in the physical lo-
cation of the modules, since it uses only local power mea-
surements. Nevertheless, the conventional droop method shows
several drawbacks that limit its application area, such as slow
transient response, tradeoff between the power-sharing accu-
racy and the frequency and voltage deviations, unbalance har-
monic current sharing, and high dependency on the inverter
output impedance. In addition, the line impedance is unknown,
which can result in reactive power unbalances. This problem
can be overcome by injecting high-frequency signals through
the power lines or by adding external data communication
signals. These communication systems, typically digital, must
not be critical and robust. This way, controller area networks,
power line communications, or wireless (radio frequency links)
are often implemented [75]–[78].
In this paper, a review of the control schemes for the parallel
operation of UPS systems and the trends of these systems in
DG systems and microgrids are presented. Although the control
of standalone UPS inverters was widely studied, it will not
be shown in this paper [79]–[86]. This paper is organized as
follows. Section II describes the configuration types of distrib-
uted UPS systems. Section III analyzes the circulating current
problem derived from the parallel operation of UPS inverters.
Section IV depicts the active load-sharing techniques, including
centralized control, MS control, ALS, and 3C. Section V pro-
vides the description of the conventional droop control method,
including a power flow analysis. Then, a generalization of the
droop method, the virtual impedance loop approach, and the
multiloop droop control techniques are described. Finally, in
Section VI, a comparison between the control techniques and
the conclusions is provided.
II. CONFIGURATIONS OF DISTRIBUTED UPS SYSTEMS
Distributed UPS systems support UPS units and critical loads
flexibly located in an interconnected electrical power network.
In order to add reliability and expandability to the system,
redundant and parallel UPS systems are usually integrated into
the power system. There are two major types of distributed UPS
systems (see Fig. 1), i.e., online and line-interactive distributed
systems [47].
0278-0046/$25.00 © 2008 IEEE
2846 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 55, NO. 8, AUGUST 2008
Fig. 1. Distributed UPS system configurations. (a) Online. (b) Line interactive.
The distributed UPS systems are highly reliable because of
redundancy. It is an advantage to achieve the N + 1 or N + X
redundancy in these systems, where N UPS units supply the
load, and 1 or X additional units stay in reserve. They are also
highly flexible to increase the capacity of the system when more
power is needed, by simply adding more UPS units [21].
A. Redundancy N + 1 or N + X UPS Units
The redundancy concept consists of having one (N + 1)
or more UPS units (N + X) in reserve, and, if some of the
rest of the N modules are damaged or disconnected, this/these
modules can automatically be connected to supply the functions
of that unit. The redundancy can reduce the single point failure.
A parallel redundant system can typically provide up to 99.99%
availability, which means that the system does not operate for
less than 1 h/year. In addition to having extra UPS modules,
the parallel redundant system needs to give the operator some
measure of system-level functionality. The simplest redundant
UPS system is the 1 + 1 parallel redundant, which consists of
using a centralized UPS with one reserve module [19].
B. Parallel Operation of UPS Systems
The proper parallel operation of the N modules that config-
ure the distributed UPS system is crucial. Generally speaking,
a paralleled UPS system must achieve the following features
[5], [6]: 1) the same output-voltage amplitude, frequency, and
phase; 2) equal current sharing between the units; 3) flexibility
to increase the number of units; and 4) plug and play operation
at any time, also known as hot-swap operation capability. The
parallel operation of UPS has a number of advantages, includ-
ing thermal management, reliability, redundancy, modularity,
maintainability, and size reduction.
III. CIRCULATING CURRENT ANALYSIS
The output currents of each UPS should be equal or at
least proportional to its nominal power rating. The difference
Fig. 2. Circulating current concept.
between those currents provokes circulating currents among the
UPS units. The circulating current (ic) is particularly dangerous
at no-load or light-load conditions, since one or several modules
can absorb active power operating in rectifier mode, as shown
Fig. 2. This current increases the dc-link voltage level, which
can result in damage to the dc-link capacitors or in a shutdown
due to overload [24].
An analysis of the circulating current can be done by using
the equivalent circuit of two UPS units connected in parallel,
sharing a common load. The analysis presented in this section
will be made using phasors, being only valid under sinusoidal
conditions. Following Fig. 3, we can define the circulating
apparent power as
∆S ∆= S1 − S2 (1)
and, consequently, the active and reactive circulating powers are
∆P ∆=P1 − P2 (2)
∆Q ∆=Q1 −Q2. (3)
Assuming that L1 � Zo1 + ZL1, L2 � Zo2 + ZL2, and that
the total output impedance XT (L1 + L2) is mainly inductive,
GUERRERO et al.: CONTROL OF DISTRIBUTED UNINTERRUPTIBLE POWER SUPPLY SYSTEMS 2847
Fig. 3. Equivalent circuit of two UPSs connected in parallel, taking into
account the output impedance of the inverters (Zo1 and Zo2) and the power
line impedances (�ZL1 = rL1 + jωL1 and �ZL2 = rL2 + jωL2).
then (2) and (3) can be simplified as
∆P ∼= E1E2
XT
sin∆φ ∼= E1E2
XT
∆φ (4)
∆Q ∼= V
XT
∆E (5)
where ∆φ = φ1 − φ2, and ∆E = E1 − E2.
Thus, these equations can be expressed in function of the
currents instead of the power, being the active and reactive
circulating currents, as
∆iP ∼= E1E2
V XT
∆φ (6)
∆iQ ∼= ∆E
XT
. (7)
In conclusion, assuming an inductive output impedance, the
active and reactive powers or currents can be controlled by
adjusting the phase and amplitude of the output voltage.
In order to try to avoid the circulating current, there exist a
number of control strategies that can be classified in active load-
sharing and droop control techniques, depending on the use or
not of communication links between UPS units.
IV. ACTIVE LOAD SHARING
The first kind of control scheme, named the active load-
sharing technique, needs intercommunication links. Although
these links limit the flexibility of the UPS system and degrade
its redundancy, both tight current sharing and low-output-
voltage total harmonic distortion (THD) can be achieved. The
following section provides a review of the existing techniques
for paralleling inverters available in the literature. The active
load-sharing techniques can be classified into four different
types, i.e., centralized control [22], [23], MS [24]–[32], ALS
[33]–[41], and 3C [42], [43]. Using these techniques, we will
generate the current or power reference of each module, which
is easy to scale according to its nominal power rating.
A. Centralized Control
This control technique, also known as concentrated control,
is depicted in Fig. 4. It consists of dividing the total load current
iL by the number of modules N , so that this value becomes the
current reference (i∗j) of each module j [22], [23]
i∗j =
iL
N
, for j = 1, . . . , N. (8)
Fig. 4. Block diagram of a centralized controller for paralleled UPS system.
The current reference value is subtracted by the current of each
module, obtaining the current error ∆Ij , which is processed
through a current control loop.
An outer control loop in the centralized control adjusts the
load voltage. This system is normally used in common UPS
equipment with several output inverters connected in parallel.
Using this approach, it is necessary to measure the total load
current iL, so it cannot be used in a large distributed system.
Consequently, a central control board is necessary.
The control implementation can follow two philosophies.
The first one is expressed by (8), and the second is to calculate
the current error ∆i = i∗j − ij and to decompose it in direct
current error ∆ip and in quadrature current error ∆iq . Finally,
∆ip and ∆iq can be used to adjust the phase and amplitude
of the output-voltage reference of each UPS unit. The other
possibility is to use ∆i and the output voltage to calculate ∆P
and ∆Q instead of ∆ip and ∆iq , as shown in Fig. 5 [6].
B. MS
In this technique, the master module regulates the load volt-
age. Hence, the master current iM fixes the current references
of the rest of the modules (slaves) as
i∗S = iM , for S = 2, . . . , N. (9)
Consequently, as shown in Fig. 6, the master acts as a voltage
source inverter (VSI), whereas the slave works as a current
source inverter (CSI) [30]. In this configuration, if the master
unit fails, another module will take the role of master in order
to avoid the overall failure of the system. There exist different
variants of this control scheme, depending on the role of the
master.
1) Dedicated: the master is one fix module.
2) Rotary: the master is arbitrarily chosen.
3) High-crest current: the master can be fixed by the module
that brings the maximum rms or crest current. Unitrode
ICs, such as UC3902 or UC3907, are used to parallel
dc/dc converters, implementing the MS strategy in which
2848 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 55, NO. 8, AUGUST 2008
Fig. 5. Block diagram of a centralized controller based on active and reactive power deviation decomposition.
Fig. 6. MS control strategy. (a) Block diagram of the system. (b) Equivalent
circuit of the parallel UPS system controlled through the MS strategy.
the module that brings the maximum current automati-
cally becomes the master.
This strategy can also be implemented by using average
active and reactive powers as
P ∗S =PM , for S = 2, . . . , N (10)
Q∗S =QM , for S = 2, . . . , N. (11)
In this particular case, and by using the highest P and Q
average values, a scheme similar to the high-crest current
control strategy can be obtained. This control strategy is shown
in Fig. 7, and in that case, the master UPS of P and Q can be
different modules [28]. MS control is often adopted when using
different UPS units mounted into a rack.
C. ALS
This is a true democratic control scheme in which every
module tracks the average current done by all the active mod-
ules [33]–[41]. This scheme, shown in Fig. 8(a), is simple to
implement by using a single wire, which contains the average
current information computed by a resistor connected to the
current sensor of every single module. In addition, adjusting
the resistor to a proper value, we can parallel converters with
different power rating. This control technique starts from an
idea applied to parallel dc/dc converters by using current-
sharing resistors connected to a common information bus. The
current of all modules is averaged by means of a common
current bus. The average current of all the modules is the
reference for each individual one. This control scheme is highly
reliable due to the real democratic conception, in which no
MS philosophy is present. In addition, the approach is highly
modular and expandable, making it interesting for industrial
UPS systems. In general, this scheme is the most robust and
useful of the aforementioned controllers. A variant of this
technique is the current weighting distribution control [39]. The
current reference of each module can be expressed as
i∗k =
1
N
N∑
j=1
ij , for k = 1, . . . , N. (12)
This control approach can be performed by using an inner
or an outer current loop. The problem with using an outer loop
GUERRERO et al.: CONTROL OF DISTRIBUTED UNINTERRUPTIBLE POWER SUPPLY SYSTEMS 2849
Fig. 7. Auto-MS (highest crest P/Q master) power-sharing control scheme.
is that due to that, the voltage loop has a narrow bandwidth,
and in order to avoid instabilities, the current loop needs a
compensator. As a consequence of the bandwidth reduction of
this loop, the current dynamics is very slow, provoking poor
current sharing during transients.
As usual, another possibility is to use active and reactive
power information instead of the current. Thus, we use active
and reactive power to adjust the phase and amplitude of each
module. Fig. 8(b) shows the block diagram of the average
power-sharing technique [41]. Using this technique, each UPS
unit controls the active and reactive power flow in order to
match the average active and reactive powers of the system
by adjusting the phase and the amplitude of its own inner
output-voltage reference. The active and reactive power can be
obtained through the direct and reactive component decomposi-
tion of the output current. The average active and reactive power
references of each module can be expressed as
P ∗k =
1
N
N∑
j=1
Pj (13)
Q∗k =
1
N
N∑
j=1
Qj , for k = 1, . . . , N. (14)
An earlier work uses this method to achieve the power
sharing between two UPS modules [40]. This control scheme
can be extended to more units by using the active and reactive
average power-sharing buses. Notice that this technique does
not require any master or slave unit, and only low-bandwidth
digital communications are required to achieve good P and Q
sharing. Nevertheless, it only acts over the fundamental com-
ponent of the output current, misleading the harmonic content.
Hence, unbalances between the power stages and the power
lines can produce large circulating harmonic current between
the units.
D. 3C
This control scheme, shown in Fig. 9, consists of the current
reference of each module taken from the aforementioned mod-
ule, forming a control ring [42]. Note that the current reference
of the first unit is obtained from that of the last unit to form
a circular chain connection. This strategy can be expressed
through
i∗1 = iN (15)
i∗k = ik−1, for k = 2, . . . , N. (16)
The approach can be interesting for distributed power sys-
tems based on ac power rings due to the distribution of power
lines [54]. Fig. 10 illustrates the example of a distributed ring-
forming UPS system. There are two lines in order to achieve
bidirectional communication and to increase system reliability.
2850 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 55, NO. 8, AUGUST 2008
Fig. 8. ALS control schemes. (a) Average current sharing. (b) Average power sharing.
GUERRERO et al.: CONTROL OF DISTRIBUTED UNINTERRUPTIBLE POWER SUPPLY SYSTEMS 2851
Fig. 9. Block diagram of the current chain control (3C).
Fig. 10. Communication links of the current chain control (3C).
The current limitation control is a variant of the 3C. In
this case, the load voltage is controlled by the master module,
whereas the slave modules are only for sharing the load current.
Except for the master module, the current command of the slave
is generated by its previous module and limited in amplitude
[43]. In this scheme, any module can be the master (dedicated,
rotating, or high-crest current). The connection of all control
circuits can form a circular chain connection such that every
module may become the master.
Fig. 11. Equivalent circuit of a UPS inverter connected to a common ac bus.
V. DROOP CONTROL METHOD
The second kind of control scheme, named the droop control
method, is able to avoid critical communication links. The
absence of critical communications between the modules im-
proves the reliability without restricting the physical location
of the modules [44]–[74].
In the literature, the droop method is also called independent,
autonomous, or wireless control. The droop method is based
on a well-known concept in large-scale power systems, which
consists of drooping the frequency of the ac generator when its
output power increases. In the case of parallel-connected UPS
inverters, the active and reactive powers supplied to the ac bus
are sensed and averaged, and the resulting signals are used to
adjust the frequency and amplitude of the UPS inverter output-
voltage reference. The droop method achieves higher reliability
and flexibility in the physical location of the modules since it
only uses local power measurements.
A. Active and Reactive Power Droop Control
Traditionally, the inverter output impedance is considered to
be inductive due to the high inductive