Electric Power Systems Research 122 (2015) 159–167

Contents lists available at ScienceDirect

Electric Power Systems Research

journal homepage: www.elsevier.com/locate/epsr

Review of hierarchical control in DC microgrids

C.N. Papadimitriou ∗ , E.I. Zountouridou, N.D. Hatziargyriou
School of Electrical and Computer Engineering, National Technical University of Athens (NTUA), University Campus, Zografou, Athens 15780, Greece

a r t i c l e i n f o a b s t r a c t

Article history: DC microgrids (DC MGs) are characterized by attractive features such as high system efficiency, high
Received 13 March 2014 power quality, reduced cost, and less complex control. The hierarchical control is extensively proposed
Received in revised form 1 October 2014 by researchers for DC MGs. This paper reviews and classifies different primary and secondary control
Accepted 8 January 2015
techniques applied to DC MGs. The load sharing mechanisms employed in primary control are distin-
guished in passive methods and active methods. The different methods for secondary control are also
Keywords:
categorized. Their key points and their limitations together with solutions that have been proposed by
DC microgrid
the research community are presented and critically assessed.
Primary control
Secondary control © 2015 Elsevier B.V. All rights reserved.
Lоad sharing mechanisms
Energy management
Hierarchical control

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
2. Hierarchical control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
3. Primary control—Load sharing mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
3.1. Passive load sharing or the droop concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
3.1.1. Poor voltage regulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
3.1.2. Circulating currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
3.2. Active load sharing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
4. Upper level: Secondary control and power management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
4.1. Centralized control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
4.2. Decentralized control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
4.3. Hybrid power control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

1. Introduction MGs is the electric power supply of isolated systems like vehicles,
space crafts, data centers, telecom systems, while they have been
Microgrids (MG) are a novel form of distribution systems, which proposed for rural areas and islands [2–4].
belong to the wider concept of Smartgrids. The Microgrid can be The DGs are interconnected via an AC link forming an AC MG,
considered as a small-scale electricity grid, which operates in low or via a DC link forming a DC MG. While a lot of work has been
or medium voltage networks. It consists of distributed generation done in the operation and control of AC MGs, DC MGs have started
(DG) units, such as renewable energy generators and combined attracting attention recently, due to their potential advantages over
heat and power units, along with storage devices and controllable AC MGs, such as:
loads (e.g. air conditioners) [1]. Their unique characteristic is that (1) The incorporated DGs can be easier coordinated, as their control
they can be islanded, especially in case of faults, increasing the is based on DC voltage without the need for synchronization.
supply reliability. Currently, the most common application of DC (2) The corresponding primary control is notably less complex as
the reactive power flow control is absent. Yet, the DC link can
suffer from harmonic content.
∗ Corresponding author. Tel.: +30 210 772 4378; fax: +30 210 772 3659. (3) As the DC electronic domestic loads dominate today, unnec-
E-mail address: chpapadi@mail.ntua.gr (C.N. Papadimitriou). essary AC/DC power conversions are avoided as most DGs

http://dx.doi.org/10.1016/j.epsr.2015.01.006
0378-7796/© 2015 Elsevier B.V. All rights reserved.
160 C.N. Papadimitriou et al. / Electric Power Systems Research 122 (2015) 159–167
Energy Storage Green power generation
Wind
Energy Storage Turbine
Device
3 -ph
DC AC
DC
DC
DC Bus
DC DC DC
DC DC DC
TV &
Lighting Refrigerator
Computer
Dc Loads
Fig. 1. The single line diagram of a typical microgrid structure.
generate DC outputs. This has a direct effect on system cost
and losses. Also, the converters used for the DC microsources
interface, are mostly transformer-less reducing further the size
and cost of the system.
(4) DC protection in general is difficult due to no zero crossing to
interrupt on. But the DC system does not experience high fault
currents as the contribution to faults by the converters of the
power electronic interfaced load or DGs is limited [2,5,6].
Fig. 1 shows a typical DC microgrid configuration with a com-
mon DC bus. Note that the DC MG topology may differ from
radial single feeder configuration to two-pole or ring configura-
tion. In these topologies either unipolar or bipolar configurations
can be implemented. Bipolar configurations can provide more volt-
age level options in comparison with unipolar connections. With
respect to the voltage levels, they can differ in accordance with
the operating requirements of each system. For example, 380 V is a
typical voltage level for data centers, while 20,230,325 V are typical
voltage levels for house installations. Other levels could be 1500 V,
±750 V, ±230 V, ±170 V etc.
The interface terminals within a MG, as shown in Fig. 1, can
be mainly categorized into four types: generation (determinis-
tic or non deterministic), load, energy storage system (ESS), and
upstream grid connection using voltage-source converters (VSCs).
These terminals have to be parallelized with the appropriate inter-
faces in order to form the MG. Paralleling the DC sources presents a
number of challenges: The first challenge concerns the stability of
the system that needs a proper converters design. Effective control
of the DC bus voltage by the DGs is important, as electronic loads are
sensitive to voltage deviations. Another important issue concerns
effective load sharing among DGs, as the load should be shared
“equally” or depending on DGs’ ratings or costs. In paralleling DGs,
the role of the source output impedance is very important. This
impedance has a considerable impact on the interaction between
the source and the load [7]. This interaction and the way it affects
load sharing among DGs will be analyzed later (Fig. 1).
The control that is applied in DC microgrids, should ideally
respond to all the aforementioned challenges. An important char-
acteristic is that DGs should be capable to support the peer to peer
scenario (local control) of operation by exercising autonomous con-
trol, especially at primary level. This feature provides modularity
and improved reliability of the system. In practice, a compromise
Ac Grid
PV panel
Bidirectional
DC AC Grid -Interfacing
DC Converter
DC
DC DC
DC DC
Air
EV charging
conditioner
station
& fan
has to be made among load sharing, modularity and autonomous
control.
In [8], a review of DC microgrid control is presented, focusing
however mostly on storage devices for MGs. Four typical control
architectures are considered: droop control, hierarchical control,
fuzzy control and multi-agent based control. This categorization
seems rather untargeted, as the hierarchical control may include
droop control, while it does not exclude intelligent control, such
as fuzzy control. The droop control is a load sharing mechanism
that can be used extensively in many different control architec-
tures. Multi Agent System (MAS) control is mostly intelligent and
can incorporate fuzzy logic.
This paper provides a review and a classification of the differ-
ent control methods applied to DC MGs, especially for primary
and upper level of the hierarchical control. In general, the con-
trol can be divided in centralized, decentralized and hybrid. The
main drawback of the control – centralized or not – that is based
on communication channels is the poor reliability in case of links
failure, whereas the communication free control – decentralized
control – suffers from poor voltage control. The hybrid control tries
to combine the advantages of the aforementioned controls. Their
limitations and ways to overcome them are also discussed.
The rest of the paper is structured as follows: Section 2 presents
the general hierarchical control of a DC microgrid and the catego-
rization of the power management strategies. Section 3 focuses on
the primary level and the different load sharing mechanisms, while
it addresses appropriate solutions to overcome their limitations.
Section 4 reviews the different power management strategies and
Section 5 concludes the paper.
2. Hierarchical control
The hierarchical control, as in AC MGs, applies for DC MGs too
and can be divided in three levels [9,10] (also in accordance to ISA-
95) (Fig. 2).
Tertiary control: import/export power
Secondary control:
restoration and synchronization
Primary control: droop
Fig. 2. Hierarchical control of microgrids [9].
 C.N. Papadimitriou et al. / Electric Power Systems Research 122 (2015) 159–167
• Primary control: This control deals with the load sharing among
the DGs. The DC–DC power converters of the DGs are responsible
for this mechanism.
• Secondary control: This control is responsible for voltage fluctu-
ations regulation. It is also responsible for the synchronization
process to re-connect seamlessly the microgrid to the upper grid.
• Tertiary control: It sets the power flow between the DC MG and the
upper grid. It is also known as energy management system and it
communicates with the distribution system operator (DSO). The
DSO or even the transmission system operator (TSO) might decide
the schedule of power exchange with the MG. (Fig. 2).
In the following, secondary and tertiary controls are discussed
together.
3. Primary control—Load sharing mechanisms
There are two types of methods to achieve power sharing and
control of the output voltage level by the DGs: (i) passive control
methods and (ii) active load sharing. It is also possible to apply
hybrid control methods combining the best features of the previous
methods.
3.1. Passive load sharing or the droop concept
The basic principle that allows synchronously rotating AC gen-
erators to change their power output in response to a change in
the system load, without an explicit communication network, is
the frequency and voltage variation at the machine terminals. Nor-
mally, frequency is linked to active power, and voltage is linked
to reactive power. Standard rotating generator systems inherently
support these droops (natural synchronizing torque) [1]. Similar
droops are also emulated at the DGs inverters for power sharing in
an AC microgrid.
The droop concept applied at a DC microgrid is slightly different,
as the frequency and reactive power are absent and thus, the active
power is linked directly to the DC voltage. The droop characteristic
of a converter in a DC microgrid can be a linear function between V
and I (commonly used) or between P and V (Fig. 3).
This droop concept can be easily applied at the DG power con-
verters offering independent control and modularity. Load sharing
is achieved directly without the need of communication. The limi-
tations of the droop control method are analyzed next. It should be
noted that most research papers attempt to overcome one limita-
tion at a time, depending on their study application.
3.1.1. Poor voltage regulation
The conventional droop concept has an inherent trade-off
between voltage regulation and current sharing, when the voltage
is controlled by several converters.
The sharing accuracy can be affected considerably by the DC
voltage error, especially when the converters have different char-
acteristics.
Let’s assume two parallel, DC converters of equal power ratings.
In Fig. 4, the unequal load sharing due to an error (V1o–V2o) in
their nominal voltages is shown. From the V–I characteristics, it can
be seen that voltage regulation is more accurate for a small droop
Fig. 3. Droop characteristic.
161
inverter, but load sharing, that should be equally divided, is poor.
For an inverter with large droop, load sharing is tighter, while the
voltage regulation is poor. Obviously, the opposite holds for the P–V
droop characteristic. So, for a large DC gain, the voltage regulation
is tighter, whereas the load sharing is poorer. For a smaller DC gain,
the load sharing is tighter while the voltage regulation is poorer
[7,10,11] (Fig. 4).
A compromise between load sharing and accurate voltage reg-
ulation is therefore needed, when the droop concept is employed.
This can be a real problem with storage devices in the microgrid,
as the challenge of balancing the energy storage needs also to be
taken into account. In order to overcome this problem, novel control
strategies have been proposed.
In [7], the voltage control and the load sharing between the
parallel sources are achieved through Gain-Scheduling Control.
The gain K (slope of the droop characteristic) changes dynami-
cally, as it follows the gain scheduling curve, when the load power
changes and the voltage need to be controlled. Load sharing is main-
tained at an acceptable level and eventually the system is more
robust. In [11] a sophisticated voltage regulation technique based
on Fuzzy Control and Gain-Scheduling Control is proposed. The
control accomplishes good voltage regulation, good load sharing
and good energy balance for the storage devices, simultaneously.
As previously, the gain-scheduling technique, changes the gain
K according to a linear function of K with respect to the output
power of the sources. This function is derived from simulations and
achieves better voltage regulation and better load sharing, simul-
taneously. Fuzzy control is employed to balance the stored energy
by changing the DC bus voltage reference.
3.1.2. Circulating currents
Application of the droop concept can create circulating currents
among the DGs when the power converters are treated as voltage
sources.
In order to suppress these circulating currents, two solutions
are proposed: use of (a) series resistor, (b) virtual output resistance
(adaptive voltage positioning (AVP)).
According to the first method, a resistor is placed in series with
the DG output to provide a voltage drop in the output. The resistor
value is set via a potentiometer so that the voltage drop of the out-
put of all paralleled DGs, are made almost identical. Obviously, this
method is impractical in real systems, since it results high power
losses in the series resistor, if the drop in output voltage is large
[12].
The second method is better applicable in the DC MG con-
cept and has common philosophy with the virtual or fictitious
impedance method in AC MGs [2,5,12]. Specifically, the DC MG
droop control is based on subtracting part of the converter out-
put current proportional to a virtual resistance (VR), Rdi , from the
voltage reference at no load. This is illustrated in the two-nodes DC
microgrid depicted in Fig. 5, where each converter is simplified by
its Thevenin equivalent model. For DG1:
∗ ∗
Vdc1 = Vdc − idc1 Rd1 (1)
Vdc1 is the voltage reference for voltage loop, idc1 is the module
output current, Rd1 is the virtual output resistance (VR), and Vdc ∗
is the output voltage reference at no load. If we consider the line
162 C.N. Papadimitriou et al. / Electric Power Systems Research 122 (2015) 159–167

Fig. 4. The droop concept limitations (a) V–I characteristic [10] (b) P–V characteristic.

impedances Rline1 and Rline2 , the following equations are derived At no load, the voltage deviation is zero. When the inverter is
(Fig. 5): connected, the deviation value varies with the load current. It also
∗ −i depends on the value of the virtual resistance. To guarantee that
Vload = Vdc dc1 Rd1 − idc1 Rline1
(2) the voltage deviation does not exceed its maximum acceptable
∗ −i
Vload = Vdc value, as a stable voltage is of crucial importance in DC MGs, the
dc2 Rd2 − idc2 Rline2
value of the virtual resistance should be limited according to the
These expressions then yield the following: following(derived from (6)):
 
idc1 R Rline2 − Rd2 /Rd1 Rline1 Vdci max
= d2 + (3) Rdi ≤ (7)
idc2 Rd1 Rd1 + Rline1 idcfli

In the droop-controlled DC microgrid, the DC output current
of each converter is set to be inversely proportional to its virtual
resistance. Thus, the current sharing error can be eliminated if the
second term of Eq. (3) is zero. In this case, the following expression
is derived:
Rd1 R
= line1
Rd2 Rline2
In case of small DC MGs, the line impedances are quite small
(order of 0.2 m), while a larger virtual resistance can be selected
(order of 0.2 ). So, as Rd1 > >Rline1 and Rd2 > >Rline2 , the line
impedances can be neglected and the following applies (from (2)):
idc1 R + Rline2 R = Vdc idci − Rdi idci → Pdci = Vdc idci − Rdi idci
= d2 ≈ d2
idc2 Rd1 + Rline1 Rd1
If the system is larger, though, Eq. (5) cannot be satisfied, while
for higher virtual resistances, the system stability might be jeopar-
dized (see (7)) [10,13].
In [14], the circulating currents are suppressed by applying a
power strategy of bidirectional VSCs implemented into two Syn-
chronous Reference Frames (SRF): positive SRF and negative SRF.
It should be noted, that the droop control loop has an inher-
ent load-dependent voltage deviation. The voltage deviation can be
found from (1) as:
where idcfli is the output current of the i-converter at full load
[9,13]. A small value of the droop gain/virtual resistance is used
to restrict large variation in system voltage and the above equa-
tion is an inherent droop concept limitation. In order to overcome
the voltage deviation and succeed good load sharing, some authors
have proposed multiplication of measured voltage deviation to a
(4) value reciprocal to virtual resistance (VR) Rdi [15].
In the following, the relation between the droop coefficient K of
the P–V droop characteristic and the VR of the V–I droop character-
istic is analyzed for the system of Fig. 5. From Eq. (1):
Vdci = Rdi idci → Vdci = Vdc − Rdi idci → Vdci idci
2 2
(8)
(5)
From the P–V droop characteristic, the following holds:
Pdci = KVdci (9)
From (1), (8) and (9) the following is derived:
2
KRdi idci = Vdc idci − Rdi idci → KRdi = Vdc − Rdi idci → Rdi (K + idci ) = Vdc
Vdc
K= − idci (10)
Rdi

∗ In practice, the DC loads are far from constant resistances and

Vdci = −idci Rdi (i = 1, 2. . .)
Fig. 5. A simplified dc-microgrid with two nodes [13].
(6)
usually are converter-based constant power loads (CPL). Some
researchers prove that CPL can significantly affect system stabil-
ity and transient behavior, especially in large-scale DC networks
[6,16–20]. The main concern is what happens with the system sta-
bility, while droop controlled sources co-exist with CPLs in the MG.
Ref. [21] defines a stable operating point when a CPL is present. Note
that these power electronic loads (CPLs) together with the rectifiers
that interface the DC MG with the AC system can create harmonic
pollution in the DC link. The simpler filter applicable would be a
capacitor on the DC side of inverters and rectifiers [22].
A novel variant of the droop control is called the voltage-based
power–voltage control method (VbPV) [23]. The VbPV control
method provides the flattest and approximately nominal voltage
profile, which is necessary for the operation of sensitive loads,
 C.N. Papadimitriou et al. / Electric Power Systems Research 122 (2015) 159–167
V terminal
V threshold
V MaxPower
Kd
V Fault
0 I max I Source
Fig. 6. V–I characteristic of each source [22].
regardless of the size and structure of the DC MGs. The method
also preserves the voltage drops from increasing in case of long
lines in the DC MGs and suppresses the circulating currents. The
voltage–current characteristic, as shown in Fig. 6, is assigned to
each source (Fig. 6).
Where Vterminal is the terminal voltage of the DG and Isource is the
injected current by the DG. Kd is the droop coefficient of the charac-
teristic (the same as Rdi ) and the rest of parameters are explained in
the following. The DG generates power when its terminal voltage
becomes less than the Vthreshold . As the terminal voltage decreases,
the generated power increases until its maximum generated power
(Pmax ) at VMaxPower . The injected current, Isource , is calculated by:
(Vterminal − Vthreshold )
Isource =
Kd
The injected current remains constant and equals Imax , as long
as the terminal voltage is between VMaxPower and VFault . When the
terminal voltage becomes less than VFault , the DG is switched off.
The value of VFault depends on the DC MG characteristics such as
the lower limit of the permitted voltage and the beginning voltage
of the voltage collapse. Obviously, VFault of all the sources in the
DC MG is the same. Analysis about the choice of the parameters is
carried out in [23], while numerical simulations for a realistic DC
MG are performed proving the good performance of this control
method.
3.2. Active load sharing
The active load sharing control methods can be classified into
four different types and are derived from control schemes of
parallel-connected DC–DC converters: (i) centralized control, (ii)
master-slave control (MS), (iii) average load sharing and (iv) circu-
lar chain control (3C) [24] Although these control schemes achieve
both good output-voltage regulation and equal current sharing,
they need communication links among the modules. In the follow-
ing, the four methods are briefly discussed.
Fig. 7. Active load sharing methods: (a) centralized control (b) master–slave control (c) Average load sharing (d) 3C control.
163
Centralized control (Fig. 7a): A central control board (CCB) is nec-
essary in this scheme in order to set the reference current for each
module. The measured load current is driven in the CCB where is
divided by the number of the modules in parallel (N), forming the
reference current (ij∗ ) of each module. Then the reference current is
subtracted from the current of each module. The error is processed
through a current control loop (CL). An outer control loop in the
centralized control adjusts the load voltage. The main drawback of
this method, apart the central controller, is the need to measure
the total load current, so the application of this scheme in a large
distribution system is difficult.
Master–slave control (Fig. 7b): One inverter (master) regulates
the voltage and sets the current references of the other units
(slaves). So, the master inverter operates in voltage control mode
and the rest of the units in current control mode. The main draw-
back of this method is the single point failure and the requirement
of a supervisory control. The system is also difficult to expand, fail-
ing to satisfy the plug and play functionalities. The method is further
categorized depending on the role of the master: (i) dedicated (ii)
rotary and (iii) high-crest current.
Average load sharing (Fig. 7c): A single wire is used, 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 convert-
ers with different power ratings by suppressing the circulating
currents. The average current of all the modules is the reference
(11)
for each individual one. This control scheme is more reliable as no
Master–Slave philosophy is present. In addition, the approach is
highly modular and expandable.
Circular chain control (Fig. 7d): In this scheme, the current ref-
erence of each module is taken from the other module, forming
a control ring. Obviously, in order to form the circular chain the
current reference of the first unit is obtained from that of the last
unit.
An interesting variant of the circular chain control is the current
limitation control. In this case, the master–slave logic is present. The
voltage is controlled by the master module (inverter under voltage
control) and the slave modules share the load current (inverters
under current control). The circular chain in this case is formed
only by the slaves modules and the master module is exempted. The
current command of the slave is generated by its previous module
and limited in amplitude forming the circular chain. Note that every
module can become the master (Fig. 7).
4. Upper level: Secondary control and power management
The key point of power management in DC MGs is to main-
tain the power balance between energy sources, storage devices
164 C.N. Papadimitriou et al. / Electric Power Systems Research 122 (2015) 159–167
Hierarchical
Control
Centralized Hybrid
Decentralized
Communication No
link
available
communication
Power Management
Strategies in Dc Microgrids
Fig. 8. Power management strategies classification.
and loads at any time, which is represented by the stable DC bus
voltage. The power management scheme dictates in which mode
the DC microgrid operates, in order to optimize its performance. In
interconnected mode, power exchange with the upstream grid is
included. If the power is not enough for the loads, the DC MG will
absorb power from the upper grid. If the power generated within
the MG is more than needed, it will transfer power to the upper grid.
Keeping this in mind, many research papers present strategies for
grid-connected, islanded and transition operation [3,8,25,26].
Most energy management strategies are based on Hierarchical
Control. These management strategies can be divided into three
major categories – depending on the way the secondary control
is implemented – centralized, decentralized and hybrid. Fig. 8
shows this classification. The next subsections analyze each cat-
egory (Fig. 8).
4.1. Centralized control
In this method a centralized controller is employed in order to
overcome the voltage deviation caused by the primary controllers
and to achieve power balance among the DGs and loads in the DC
MG. This method can achieve optimal control, but it requires some
form of real-time communication.
Ref. [9] describes a centralized controller that restores the volt-
age level by using low bandwidth communication channels (LBC).
The voltage level in the MG bus is compared with the voltage ref-
erence, and the error ıV, processed through a compensator, is sent
to all the modules via the LBC to restore the output voltage. So, Eq.
(1) for each DG becomes:
∗
Vdci = Vdc ∗
+ ıV − idci Rdi where i = 1, 2. . .
In [2], a centralized control, called supervisory, is applied in
order to avoid voltage deviations. An interesting approach is pre-
sented by changing the virtual resistance (VR) values, Rdi , online, in
a DC MG with two batteries connected to the main bus. The virtual
resistance of their droop control is adapted online by the supervi-
sory control; so that the batteries of the islanded system have the
same SOC at all times and therefore preserve equally their cycle life.
The value of Rdi corresponds to the current SOC and capacity of the
battery i. Higher Rdi will cause lower charge/discharge rate and vice
versa. Therefore, when batteries are charging, higher Rdi are given
to a battery with higher SOC. On the other hand, when discharging,
higher Rdi is given to a battery with lower SOC. A symmetric func-
tion for computing charge and discharge VRs is proposed, taking
into account the batteries’ SOCs and their rate of change.
Application of Centralized Control is reported in [26], where
an intelligent multi-layer supervision subsystem is suggested to
achieve power balance in a DC MG of a tertiary building. The super-
vision system dictates power setpoints for the DGs and decides load
shedding actions according to the end-user demands, forecasts of
PV production and load, energy cost management, etc.
4.2. Decentralized control
In this control strategy, the DGs are autonomously controlled
using local real-time feedbacks. The methods for autonomously
controlled MGs can be further categorized into (1) with commu-
nication (2) no-communication.
Microgrid with communication: Communication between DGs is
required, but the operational decisions are taken in a decentralized
way, usually at the DGs level. That is the substantial difference with
the centralized secondary control.
In [13], a LBC is used for the exchange of the DC voltage and cur-
rent information of the converters, and all of the calculations and
operations are realized locally. The aim of the proposed strategy is
to solve the two main problems produced by droop primary con-
trol: poor current sharing and voltage deviation. Two signals are
formed locally and affect the droop equation of the primary level
control. The error between the voltage reference and the calculated
average value of the DC voltage is driven through a compensator
and forms the first signal. It is responsible for the DC bus voltage
restoration. The error between the current reference and the calcu-
lated average value of the current is driven through a compensator
and forms the second signal, which is responsible for the precise
current sharing. In [10], a decentralized approach is presented to
address the primary control voltage deviation. Droop V–I character-
istic is shifted along the voltage axis by addition of a small voltage
signal u. This shift control is termed as digital average current
sharing (DACS) control. The signal’s value is determined through
the utilization of a LBC as follows: The controller of each source
communicates with the controller of other sources and sends the
magnitude of its current (in per unit). Using this information, the
individual source controller determines the average value of the
current supplied by all the sources. This current multiplied by a
shift gain forms the voltage signal u. Distributed control based
on average current sharing (ACS) is also presented in [27]. The local
controllers communicate with each other using a common bus of
information. The measured value of each source current is con-
verted to voltage signal, via a resistance Rj which is connected to
the ACS bus (analog). If resistances of all modules are equal, the
voltage appearing on the bus corresponds to the currents average
value. This signal is added to the droop equation. The scheme offers
equal load sharing among sources and tight voltage regulation of
the DC bus. The main drawback is that the current sharing bus has to
be distributed within the DC MG region along with power lines. This
may inject significant external noise in the bus. In [28] a decentral-
(12) ized secondary control that is based on power-line signaling (PLS)
is proposed. In this case, the power network serves as a communi-
cation channel where the exchanging messages come in the form
of PLS signals directly injected from the converters’ primary control
loops. The localized controller in each source receives the PLS sig-
nal, extracts the needed information and broadcasts by giving rise
to the respective voltage references of the primary controllers. This
way, the sources change mode of operation whereas the DC voltage
bus can deviate from its nominal value but in a certain range. This
deviation can be optionally canceled by secondary control action
without affecting system’s proper operation. The main drawbacks
of this scheme are the slow communication through PLS and the
possible mismatch in the electromagnetic compatibility (EMC) with
the electronic devices in the MG.
Agent based control is predominantly decentralized. Multi-
Agent Systems (MAS) are composed of intelligent entities which
can be software or hardware units with only local knowledge and
limited abilities, but can be able to interact with each other to
achieve a global target [1]. Agents can work with the help of con-
ventional control strategy, as well as advanced control methods,
e.g. artificial intelligent based techniques and expert systems. Espe-
cially, DC MG present a unique challenge to agent modeling due
 C.N. Papadimitriou et al. / Electric Power Systems Research 122 (2015) 159–167
Fig. 9. Power management strategies (a) centralized control (b) decentralized control with communication (c) decentralized control without communication (d) hybrid
power control.
to the effect of the very fast transients that occur in the distribu-
tion lines, while the collaborative decision making process among
agents requires some time [29]. In [30], a DC-MG lab is presented
that provides a flexible platform for different control schemes with
the help of Rockwell’s agent based distributed control software. In
[29], a small scale DC MG is used to implement and test several
intelligent distributed control concepts. These are implemented
using a combination of real-time control and intelligent agents
that communicate with each other. In this distributed information
sharing process, each agent is responsible for solving a portion of
the problem, such as load priority, voltage balancing and battery
charging.
In [31], a DC MG containing three constant power loads (CPL)
is introduced. As already mentioned in Section 3.1, such a system
could become unstable under some conditions. Local stabilizing
agents on CPLs are implemented and are designed in an appro-
priate way to ensure the system stability, even during faults, such
as loss of one of the stabilizing agents and failure of one or several
loads.
No-communication microgrid: In this category, all operations can
be performed without communications. The method mostly used
is the DC Bus Signaling (DBS) technique.
This method deviates from the concept of accurate power shar-
ing with minimal voltage deviations on the bus. Significant voltage
deviations from nominal are permitted, since it is assumed that the
system is power electronics based and the source and load inter-
faces can be designed to operate satisfactorily within a broader
specified voltage range. The DC bus voltage is used as an infor-
mation carrier to distinguish different modes of operation for the
DGs. In this case, voltage thresholds that dictate the different modes
of operation and priority of the devices and loads have to be set
[32]. The main advantage is that the transition between different
165
modes and the corresponding changes of control methods for con-
verters can be achieved without additional communication links.
The benefits are cost reduction and reliability enhancement [33].
The voltage thresholds have to be fairly distinctive, but also tight
enough in order not to destabilize the system. For DC networks
with longer transmission cables and considering possible DC mea-
surement errors, the DC voltages at different terminals can be
slightly different. So, care must be taken when selecting the volt-
age thresholds to ensure normal load and generation variation will
not cause wrong transitions. One should take also into account the
different control methods used for different terminals [21,34]. The
thresholds for the DGs converters are calculated beginning with the
highest threshold. Each successive threshold is calculated to ensure
that when the sources assigned to the previous threshold are online,
voltage drop in the system, caused by distribution line resistance
and voltage droop, do not prematurely activate sources assigned
to the next threshold. In general, the first voltage threshold, V0 , is
set to the nominal operating voltage of the system. Each successive
voltage threshold, Vn, is calculated by subtracting the voltage drop
and a margin of error from the preceding threshold. For load shed-
ding, the shutdown thresholds must be calculated such that the
shutdown priority of the loads remains unaffected by the unequal
propagation of the DC bus voltage due to long cables throughout the
system. In this case, the shutdown threshold for the lowest priority
load is calculated first.
Refs. [33,35,36] present different power management strategies
for different DC MGs focusing on extreme conditions. In [33], the
DC bus signaling is employed for controlling a DC MG based on
modular photovoltaic generation. The paper deals with extreme
conditions in islanding mode, such as fully charged or fully dis-
charged batteries. Ref. [35] designs a power sharing strategy based
on DBS for a laboratory DC MG including PV, batteries and fuel cell.
166 C.N. Papadimitriou et al. / Electric Power Systems Research 122 (2015) 159–167

Limitations to overcome
Voltage Accurate Circulating Proposed
control power Currents solutions
sharing
fair good Fair suppressing Gain scheduling
technique[7]
good good Good Gain scheduling
suppressing technique and
Fuzzy control[11]
good - Good control strategy

(communication free)
suppressing implemented into

Passive load sharing

two Synchronous
Reference Frames
(SRF):
positive SRF and
Primary Control

negative SRF [14]

- good - multiplication of
measured voltage
deviation to a value
reciprocal to virtual
resistance [15]
good - Good The voltage-based
suppressing power-voltage
control method
(VbPV)[22]
excellent excellent Excellent (i) centralized
communicat
Active load

suppressing control, (ii) master-

sharing
(with

slave (MS), (iii)

ion)

average load
sharing and (iv)
3C[23]
good good Good Supervisory control
suppressing with adaptive
Centralized

droop[2]
control

good - Good Voltage restoration

suppressing level enabled [9]
excellent excellent Excellent intelligent multi-
suppressing layer supervision
subsystem[25]
good good - Improved droop
control[13]
Decentralized control with

- good - DACS[10]
Secondary Control

communication

good good - ACS[26]

excellent excellent - PLS[27]

excellent excellent Excellent Multi- agent based

suppressing control systems[28-
30]
good good -
communication
control without
Decentralized

DBS[4,21, 32-35]

good good - N o vel hierarchical

Hybr

c o ntr o l[38]
id

Fig. 10. Synopsis of references on DC microgrids control.

The designed controller takes effective control to limit the influ-
ence of diverse load. Ref. [36] presents a novel adaptive DBS for a
DC MG containing batteries. The batteries swap modes of opera-
tion (droop mode and constant current charging mode), through
their I–V characteristics, to avoid that the bus voltage reaches its
maximum or minimum value. Another use of the DBS for power
management involves load shedding. Ref. [4] proposes load shed-
ding based on predefined load priority levels, during abnormal or
island conditions. Load shedding is activated by insufficient power
generation or insufficient energy storage. Ref. [21] also presents
a control strategy based on DBS that involves load shedding and
generation curtailment.
The main drawback of the DBS methods is that the number of
sources and storages within the system is restricted by the num-
ber of voltage levels, which cannot be divided unlimitedly due to
the DC bus voltage tolerance. Also, adding sources with higher pri-
ority involves changing the states of all other sources with a lower
priority, which restrains “plug and play” operation. Moreover, volt-
age level at different locations varies due to resistive drop across
the interconnecting cables causing erroneous operation transitions
 C.N. Papadimitriou et al. / Electric Power Systems Research 122 (2015) 159–167
of DGs. For this reason, most of the works reported in this area
describe applications of DBS to superconducting DC MG or small
MG with very short electrical lines [10,32].
To address the limitation of the varying voltage level due to line
resistances, a small AC signal over the DC signal is injected in a
method reported in [37,38]. The frequency of the AC signal acts
as a means of communication. This method is prone to noise on
power cables. Further, it requires circuits for accurate injection and
detection of the AC signal. This limits the viability of the scheme.
4.3. Hybrid power control
Hybrid control architectures combine the aforementioned
strategies at different levels, so as to achieve better results. In [39] a
novel hierarchical control is implemented. The first level is based on
the DBS technique and is activated when the communication links
fail. The second level is activated under normal operation where
the Communication link provides full observability over the DC MG
including real time bus voltage, power flow and operation status of
converters. Fig. 9 depicts the general power management strate-
gies that are presented in Section 4, while Fig. 10 summarizes the
references at both primary and secondary/tertiary level (Fig. 9).
5. Conclusion
This paper reviews and classifies the primary and secondary
control methods applied to DC MGs.
The hierarchical control, as known in AC MGs, is also applied to
DC MGs. It is divided into three main categories – depending on the
way the secondary control is implemented – centralized, decentral-
ized and hybrid. The decentralized control is further divided into
techniques requiring communication and without communication.
The above control methods are implemented by different load
sharing mechanisms, such as passive methods (droop control,
VbPV) or active methods and different control techniques such
as DBS, fuzzy, MAS etc. Their key points and their limitations are
presented and critically assessed.
References
[1] N. Hatziargyriou, Microgrids: Architectures and Control, Wiley, IEEE Press,
Athens, 2014.
[2] T. Dragicevic, J.M. Guerrero, J.C. Vasquez, D. Skrlec, Supervisory control of an
adaptive-droop regulated DC microgrid with battery management capability,
IEEE Trans. Power Electron. 29 (2) (2014) 695–706.
[3] D. Salomonsson, L. Soder, A. Sannino, An adaptive control system for a DC
microgrid for data centers, IEEE Trans. Ind. Appl. 44 (6) (2008) 1910–1917.
[4] Lie Xu, Dong Chen, Control and operation of a DC microgrid with variable gener-
ation and energy storage, IEEE Trans. Power Delivery 26 (4) (2011) 2513–2522.
[5] Y. Ito, Y. Zhongqing, H. Akagi, DC micro-grid based distribution power genera-
tion system, in: Power Electronics and Motion Control Conference, vol. 3, 2004,
pp. 1740–1745.
[6] A.A.A. Radwan, Y.A.-R.I. Mohamed, Linear active stabilization of converter-
dominated DC microgrids, IEEE Trans. Smart Grid 3 (1) (2012) 203–216.
[7] Zhihong Ye, D. Boroyevich, Kun Xing, F.C. Lee, Design of parallel sources in
DC distributed power systems by using gain-scheduling technique, in: Power
Electronics Specialists Conference. vol. 1, 1999, pp. 161–165.
[8] Z.H. Jian, Z.Y. He, J. Jia, Y. Xie, A review of control strategies for DC micro-
grid, in: Fourth International Conference on Intelligent Control and Information
Processing (ICICIP), 2013, pp. 666–671.
[9] J.M. Guerrero, J.C. Vasquez, J. Matas, L.G. de Vicuña, M. Castilla, Hierarchical
control of droop-controlled AC and DC microgrids—a general approach toward
standardization, IEEE Trans. Ind. Electron. 58 (1) (2011) 158–172.
[10] S. Anand, B.G. Fernandes, M. Guerrero, Distributed control to ensure pro-
portional load sharing and improve voltage regulation in low voltage DC
microgrids, IEEE Trans. Power Electron. 28 (4) (2013) 1900–1913.
[11] H. Kakigano, A. Nishino, T. Ise, Distribution voltage control for DC microgrid
with fuzzy control and gain-scheduling control, in: IEEE Eighth International
Conference on Power Electronics and ECCE Asia (ICPE & ECCE), 2011, pp.
256–263.
167
[12] Gulin Marko, Control of a DC Microgrid, 2012, https://www.fer.unizg.hr/
download/repository/KDI Marko Gulin.pdf.
[13] Xiaonan Lu, J.M. Guerrero, Kai Sun, J.C. Vasquez, An improved droop control
method for DC microgrids based on low bandwidth communication with DC
bus voltage restoration and enhanced current sharing accuracy, IEEE Trans.
Power Electron. 29 (4) (2014) 1800–1812.
[14] M. Kumar, S.N. Singh, S.C. Srivastava, Design and control of smart DC micro-
grid for integration of renewable energy sources, in: Power and Energy Society
General Meeting, 2012, pp. 1–7.
[15] P. Karlsson, J. Svensson, DC bus voltage control for a distributed power system,
IEEE Trans. Power Electron. 18 (6) (2003) 1405–1412.
[16] R.S. Balog, Autonomous Local Control in Distributed DC Power Systems, Dept.
Elect. Comput. Eng., Univ. Illinois at Urbana-Champaign, Champaign, IL, 2006
(Ph.D. Dissertation).
[17] S.D. Sudhoff, S.F. Glover, P.T. Lamm, D.H. Schmucker, D.E. Delisle, Admittance
space stability analysis of power electronic systems, IEEE Trans. Aerosp. Elec-
tron. Syst. 36 (3) (2000) 965–973.
[18] C.H. Rivetta, A. Emadi, G.A. Williamson, R. Jayabalan, B. Fahimi, Analysis
and control of a buck DC–DC converter operating with constant power
load in sea and undersea vehicles, IEEE Trans. Ind. Appl. 42 (2) (2006)
559–572.
[19] A. Griffo, J. Wang,D. Howe, Large signal stability analysis of DC power systems
with constant power loads, in: Vehicle Power and Propulsion Conference, 2008.
VPPC ’08. IEEE, 3–5 Sept. 2008, pp. 1–6.
[20] C.N. Onwuchekwa, A. Kwasinski, Dynamic behavior of singlephase full-wave
uncontrolled rectifiers with instantaneous constant power loads, in: Energy
Conversion Congress and Exposition (ECCE) IEEE, 2011, pp. 3472–3479.
[21] D. Chen, L. Xu, L. Yao, DC voltage variation based autonomous control of DC
microgrids, IEEE Trans. Power Delivery 28 (2) (2013) 637–648.
[22] Andrey Lana, Kaipia Tero, Partanen Jarmo, Investigation into harmonics of LVDC
power distribution network using EMTDC/PSCAD software, in: International
Conference on Renewable Energies and Power Quality (ICREPQ’11), 2010, pp.
1–6.
[23] R. Asad, A. Kazemi, A novel decentralized voltage control method for direct
current microgrids with sensitive loads, Int. Trans. Electr. Energy Syst. (2013),
http://dx.doi.org/10.1002/etep.1833.
[24] J.M. Guerrero, Lijun Hang, J. Uceda, Control of distributed uninterruptible power
supply systems, IEEE Trans. Ind. Electron. 55 (8) (2008) 2845–2859.
[25] H. Kakigano, Y. Miura, T. Ise, R. Uchida, DC micro-grid for super high quality
distribution-system configuration and control of distributed generations and
energy storage devices, in: Power Electronics Specialists Conference PESC’06,
2006, pp. 1–7.
[26] Baochao Wang, M. Sechilariu, F. Locment, Intelligent DC microgrid with smart
grid communications: control strategy consideration and design, IEEE Trans.
Smart Grid 3 (4) (2012) 2148–2156.
[27] W. Qiu, Z. Liang, Practical design considerations of current sharing control for
parallel VRM applications, in: 20th Applied Power Electronics Conference and
Exposition. vol. 1, 2005, pp. 281–286.
[28] T. Dragicevic, J.M. Guerrero, J.C. Vasquez, A distributed control strategy for coor-
dination of an autonomous LVDC microgrid based on power-line signaling, IEEE
Trans. Ind. Electron. 61 (7) (2014) 3313–3326.
[29] F.P. Maturana, R.J. Staron, D.L. Carnahan, K.A. Loparo, Distributed control con-
cepts for future power grids, in: Energytech IEEE, 2013, pp. 1–6.
[30] D.H. Moore, J.M. Murray, F.P. Maturana, T. Wendel, K.A. Loparo, Agent-based
control of a DC microgrid, in: Energytech IEEE, 2013, pp. 1–6.
[31] P. Magne, B. Nahid-Mobarakeh, S. Pierfederici, A design method for a fault-
tolerant multi-agent stabilizing system for DC microgrids with constant power
loads, in: Transportation Electrification Conference and Expo (ITEC) IEEE, 2012,
pp. 1–6.
[32] J. Bryan, R. Duke, S. Round, Decentralized generator scheduling in a nanogrid
using DC bus signaling, in: IEEE Power Engineering Society General Meeting.
vol. 1, 2004, pp. 977–982.
[33] Zhang Li, Tianjin Wu, Xing Yan, Kai Sun, J.M. Gurrero, Power control of DC micro-
grid using DC bus signaling, in: 26th Applied Power Electronics Conference and
Exposition (APEC) IEEE, 2011, pp. 1926–1932.
[34] J.K. Schonberger, Distributed Control of a Nanogrid Using DC Bus Signalling,
University of Canterbury, Christchurch, New Zealand, 2005 (Ph.D. Dissertation).
[35] Xiaofeng Sun, Zhizhen Lian, Baocheng Wang, Xin Li, A hybrid renewable DC
microgrid voltage control, in: IEEE Sixth International Power Electronics and
Motion Control Conference, IPEMC, 2009, pp. 725–729.
[36] Xunwei Yu, A. Huang, R. Burgos, Jun Li, Yu Du, A fully autonomous power man-
agement strategy for DC microgrid bus voltages, in: 28th IEEE Applied Power
Electronics Conference and Exposition (APEC), 2013, pp. 2876–2881.
[37] D.J. Perreault, R.L. Selders, J.G. Kassakian, Frequency-based current-sharing
techniques for paralleled power converters, IEEE Trans. Power Electron. 13 (4)
(1998) 626–634.
[38] A. Tuladhar, K. Jin, A novel control technique to operate DC/DC converters in
parallel with no control interconnections, in: 29th Annual IEEE Power Electron-
ics Specialists Conference. vol. 1, 1998, pp. 892–898.
[39] Jin Chi, Peng Wang, Jianfang Xiao, Tang Yi, Hoong Choo Fook, Implementation
of hierarchical control in DC microgrids, IEEE Trans. Ind. Electron. 61 (8) (2014)
4032–4042.