IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 32, NO.

2, JUNE 2017 675

Fully Distributed Cooperative Secondary Frequency

and Voltage Control of Islanded Microgrids
Nima Mahdian Dehkordi, Nasser Sadati, Member, IEEE, and Mohsen Hamzeh, Member, IEEE

Abstract—This paper proposes a new distributed cooperative As the primary control, the well-known droop control method
secondary control for both frequency and voltage restoration of [3]–[6] can maintain the stability of a microgrid voltage and
an islanded microgrid with droop-controlled, inverter-based dis- frequency. Despite application of the primary control, the volt-
tributed generations (DGs). Existing distributed methods com-
monly design secondary control based on the minimum real part ages and frequency of the microgrid can still deviate from their
of the nonzero Laplacian matrix eigenvalues related to the micro- reference values. Therefore, to compensate for the deviations
grid communication graph, which, however, is global information. caused by the droop control, employing an additional control
In contrast to the existing distributed methods, in this paper we level, called the secondary control, is necessary for restoring the
design a fully distributed adaptive control based on the dynamic microgrid frequency and voltage magnitudes. Several studies
model of DG units and on information from neighboring units.
Therefore, the proposed control scheme increases the system reli- have addressed the secondary control of islanded microgrids.
ability, decreases its sensitivity to failures, and eliminates the need Existing secondary control methods consist of centralized and
for a central processing unit. The fully distributed controllers re- distributed structures. The traditional secondary control meth-
store the islanded microgrid frequency and voltage magnitudes to ods use a centralized structure that requires all the information
their reference values for all DG units irrespective of paramet-
of the individual droop-controlled DG units and a central com-
ric uncertainties and disturbances while providing accurate real
power sharing. Furthermore, the proposed method considers the puting and communication unit. These requirements reduce the
coupling between the islanded microgrid frequency and voltages. overall system reliability and increase its sensitivity to failures,
Finally, we have conducted comprehensive simulation studies in which can lead to a single point of failure [4], [7], [8]. To
the MATLAB/SimPowerSystems toolbox to verify the proposed overcome the aforementioned issues, several recent studies in-
control strategy performance. spired by the decentralized control concept have presented the
Index Terms—Adaptive control, distributed cooperative sec- distributed cooperative secondary control as a promising ap-
ondary control, feedback linearization, microgrid, multi-agent sys-
tem, power sharing.
proach [9]–[24]. The distributed cooperative secondary control
uses local rather than global information. Furthermore, this ap-
I. INTRODUCTION proach has several advantages over a central control structure,
including improving the system reliability, decreasing its sen-
ISTRIBUTED generation (DG), in the form of mecha-
D nisms such as wind turbines (WTs), photovoltaics (PVs),
and micro-turbines (MTs), is the building block of the emerging
sitivity to failures, and eliminating the need for a central com-
puting and communication unit. Moreover, it provides a robust
secondary control framework that works properly irrespective
microgrid paradigm. A microgrid is a group of DGs, local loads,
of time-varying, restricted, and unreliable communication net-
and energy storage units that can work in both islanded and grid-
works [11], [12], [16], [20]–[22]. A centralized control structure
connected operating modes. Microgrids offer more efficiency
requires a complex communication network with two-way com-
and reliability than conventional power grids. Under normal op-
munication links, which reduces overall system reliability and
eration conditions, the microgrid works in grid-connected mode,
increases its sensitivity to failures. However, the communication
and the main grid imposes the microgrid frequency and volt-
network of the distributed controllers is sparse (i.e., each agent
ages. In this case, the legacy grid provides constant frequency
communicates with local neighbors rather than all neighbors).
and voltages for the microgrid. Unpredictable disturbances or
Unlike a fully connected network, this sparse network reduces
preplanned scheduling may cause islanding, disconnecting the
communication infrastructure costs and makes it scalable and
microgrid from the main grid. Consequently, the pre-islanding
reliable [11], [12], [16], [20]–[22].
control strategy can make the grid unstable [1], [2].
Recently, researchers have recommended various types of dis-
tributed secondary control algorithms for microgrids. However,
Manuscript received August 1, 2016; revised October 17, 2016 and December
9, 2016; accepted December 9, 2016. Date of publication December 13, 2016; existing methods have the following drawbacks:
date of current version May 18, 2017. Paper no. TEC-00646-2016. 1) Many control strategies designs are based on the small-
N. Mahdian and N. Sadati are with the Department of Electrical Engi- signal model. Despite their simplicity, small-signal-based
neering, Sharif University of Technology, Tehran 113659363, Iran (e-mail:
[email protected]; [email protected]). controllers cannot guarantee global stability, a necessary
M. Hamzeh is with the Department of Electrical Engineering, Shahid Beheshti requirement in complex networks [10], [12], [14], [20],
University, Tehran 1983963113, Iran (e-mail: [email protected]). [23].
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. 2) Several presented models suffer from incomplete plant dy-
Digital Object Identifier 10.1109/TEC.2016.2638858 namics because they ignore the inner controllers’ impact

0885-8969 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

Authorized licensed use limited to: Indian Institute of Technology- Goa. Downloaded on June 01,2020 at 05:43:19 UTC from IEEE Xplore. Restrictions apply.
676 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 32, NO. 2, JUNE 2017

on the secondary control [11], [12], [14], thus affecting

both the stability and performance of high-power convert-
ers. For example, [12] presents a tracking synchronization
controller for a linear first-order system.
3) Controllers presented in the existing studies are com-
monly designed based on the minimum real part of the
nonzero Laplacian matrix eigenvalues related to the mi-
crogrid communication graph, which, however is global
information [9]–[17], [20]–[23].
4) Key drawbacks of the previous methods are that they can-
not guarantee robust stability when facing parametric un-
certainties, unmodeled dynamics, or disturbances. Unlike
[13], [24] proposes an adaptive distributed cooperative
Fig. 1. Schematic diagram of an inverter-based DG with its primary and
secondary voltage control to compensate for parametric secondary controllers.
uncertainties; however, this method is not robust with re-
spect to disturbances.
To overcome the aforementioned difficulties, this paper pro- global information. However, the proposed method, based
poses a new, fully distributed cooperative secondary control for on the DG unit dynamics and on information from neigh-
the frequency and voltage restoration of an islanded microgrid. boring units, is fully distributed.
DG units in a microgrid play the role of agents that act as one Finally, we verify the effectiveness of the proposed strategy
team toward a common synchronization object and follow coop- using the MATLAB/SimPowerSystems toolbox.
erative decisions. Distributed cooperative control of multi-agent
systems consists of two structures: the tracking synchronization
problem and the regulator synchronization problem (known as II. LARGE-SIGNAL DYNAMICAL MODEL OF
leaderless consensus problem). The cooperative secondary con- MICROGRID SYSTEM
trol of microgrids is a tracking synchronization problem, in
which all DGs synchronize to a leader node that acts as a pre- Fig. 1 shows the configuration of a typical inverter-based DG
specified reference generator [12]–[16]. DG units presumably in islanded mode operation. It includes a DC energy source (e.g.,
communicate with one another through a sparse communica- solar cells, battery systems, or fuel cells), the voltage source
tion network. First, the feedback linearization method trans- converter (VSC), the series LCL filter, the output connector,
forms the nonlinear dynamics of DG units to linear dynamics, and the current, voltage, and power control loops. The control
and it then transforms the secondary control into a second- loops use a simple PI controller to adjust the VSC frequency and
order tracking synchronization problem. This paper employs a voltage. The nonlinear model of each DG can be presented in the
weighted directed graph to model the communication network synchronously rotating d−q reference frame. We assume that
topology among DG units. Next, fully distributed controllers re- one DG reference frame is designated as the common reference
store the islanded microgrid frequency and voltage magnitudes frame and the other DG unit models have been converted to
to their reference values irrespective of parametric uncertainties the common reference frame with the rotating frequency of
and disturbances. We note that for frequency restoration, the ωcom . We can obtain coordination of the primary controller by
control method should also meet the power sharing property. considering the dynamics of the real and reactive power droop
We can summarize the main features of the proposed method techniques. It provides the reference points for the ith DG unit’s
as follows: voltage control loop and the operating frequency ωi for the VSC.
1) To the best of the authors’ knowledge, this paper is the The ith DG unit’s real and reactive power are low-pass filtered
first to propose a new fully distributed cooperative sec- with the cutoff frequency of ωci . Equation (1) represents the
ondary control for both voltage and frequency restoration droop controller of the ith DG:
of an islanded microgrid based on consideration of a com-
plete nonlinear system model, irrespective of parametric ωi = ωni − mP i Pi
uncertainties and disturbances.
2) In the case of frequency restoration, the distributed vo∗ di = Vni − nQ i Qi
consensus-based control demonstrates the accuracy of vo∗ qi = 0, (1)
power sharing.
3) Easy implementation of the proposed protocols on digital
hardware devices under both uncertain and fixed commu- where Pi and Qi represent the filtered values of the real and
nication networks facilitates fulfillment of the plug-and- reactive power of the ith DG unit, respectively. mP i and nQ i
play capability of intermittent DG units. represent the static droop gains. Vni and ωni are the reference
4) Existing distributed control methods use the minimum real voltage and frequency setpoints, respectively. We represent the
part of the nonzero Laplacian matrix eigenvalues related to complete nonlinear model of the ith DG, illustrated in Fig. 1,
the microgrid communication network, which, however is as the following multi-input multi-output (MIMO) nonlinear

Authorized licensed use limited to: Indian Institute of Technology- Goa. Downloaded on June 01,2020 at 05:43:19 UTC from IEEE Xplore. Restrictions apply.
DEHKORDI et al.: FULLY DISTRIBUTED COOPERATIVE SECONDARY FREQUENCY AND VOLTAGE CONTROL OF ISLANDED MICROGRIDS 677

system: From (1) and (3), the voltage magnitude synchronization of

DG units, vo,m agi , is equal to synchronizing the d component of
ẋi = fi (xi ) + gi1 (xi )ui1 + gi2 (xi )ui2 + ki (xi )Di output voltages vo di . We assume that DG units can communi-
yi1 = vo di = hi1 (xi ) cate with one another through the prescribed communication di-
rected graph G. Therefore, the secondary voltage control selects
yi2 = ωi = ωni − mP i Pi = hi2 (xi ) + di ui2 , (2) ui1 in (2) so that yi1 −→ vref , ∀i. Motivated by the state-space
form of the open loop system (2), we use the partial feedback
where xi = [αi Pi Qi iLdi iLqi vo di vo qi io di io qi ]T and αi is the linearizing scheme [27] to transform the MIMO nonlinear state-
DG unit reference frame angle with respect to the common space form of the ith DG (2) to the linearized system as follows:
reference frame. iLqi , iLdi , io qi , io di , vo qi , and vo di repre-
sent the quadratic and direct components of iL i , ioi , and ÿi1 = L2f i hi1 + Lg i 1 Lf i 1 hi1 ui1
voi in Fig. 1. yi = [yi1 , yi2 ]T = [vo di , ωi ]T , ui = [ui1 , ui2 ]T = μ̇i1 = Wi1 (yi1 , μi1 ), ∀i, (4)
[Vni , ωni ]T , and Di = [ωcom vb di vqdi ]T are the outputs, inputs,
and known disturbances, respectively. Ref. [13] presents more ∂h(x)
where Lf (x) h(x) = f (x) is the Lie derivative [27] of
details about fi (xi ), gi1 (xi ), gi2 (xi ), and ki (xi ). The purpose ∂x n
of the secondary control is to select Vni and ωni , so vo di and ωi h(x) along f (x). μi1 ∈ R i1
represents the set of internal dy-
are regulated to the desired nominal values. namics that are asymptotically stable [13]. Therefore, we may
write (4) as follows:

III. INTRODUCTION TO GRAPH THEORY ẏi1 = yi1,2

The microgrid is commonly known as a multi-agent cooper- ẏi1,2 = fi (xi ) + gi (xi )ui1 , ∀i. (5)
ative system. In the microgrid, DG units and transmission lines We define vi = fi (xi ) + gi (xi )ui1 so that we have
play the roles of agents and edges, respectively. Each line in the
microgrid interconnects a pair of two nodes, and each node is ẏi1 = yi1,2
connected to a DG. Each DG unit can communicate with other ẏi1,2 = vi , ∀i (6)
DGs through a sparse communication network. In this paper,
we model the communication topology among DG units us- or in matrix form,
ing a weighted directed graph G = (VG , EG , AG ) with a set of ẏi = Ayi + Bvi , ∀i, (7)
N agents VG = {1, 2, ..., N }, a set of edges EG ⊂ VG × VG ,  
and a weighted adjacency matrix AG = (aij ≥ 0) ∈ RN ×N . where yi = [yi1 yi1,2 ]T , A =
01
, and B = [0 1]T . The dy-
(i, j) ∈ EG ⇔ aj i = 1; otherwise, aj i = 0 and aii = 0 for 00
all i ∈ VG because of (i, i) ∈ / EG . The neighbors of node namic of the reference generator (the leader) can be shown in
i are represented by Ni = {j ∈ VG : (i, j) ∈ EG , j ∈ / i}. For matrix form as follows:
the multi-agent system compromising the leader with N fol- ẏ0 = Ay0 , (8)
lowers, the adjacency matrix of the leader is described as
g = [g1 , g2 , ..., gN ]T ∈ RN . If agent i is a neighbor of the leader, where y0 = [vref v̇ref ]T .
gi = 1; otherwise, gi = 0. Lemma 4.1. Let graph G comprise a directed spanning tree
The elements of Laplacian
 matrix L = [Lij ] ∈ RN ×N are (equivalently, the leader has directed paths to all followers),
represented as Lii = j = i aij and Lij = −aij [25]. We note under the following protocol [28]:
that the corresponding directed graph edges of the communica- 
N
tion network imply the communication links among DG units. vi = cK̃ aij (xi − xj ), i = 1, ..., N (9)
Moreover, for a directed graph, if node i is a neighbor of node j, j =1
then node j receives information from node i but not necessarily
1
vice versa [26]. A directed graph contains a directed spanning with K̃ = −B T P −1 and c ≥ , where λi ,
mini=1,...,N Re(λi )
tree if a root node exists that has a direct path from that node to
i = 1, ..., N , represents the Laplacian matrix eigenvalue and P
all other nodes in the graph.
is a solution to the following linear matrix inequality (LMI):

IV. FULLY DISTRIBUTED ADAPTIVE SECONDARY AP + P AT − 2BB T < 0; (10)

VOLTAGE CONTROL all yi in (7) synchronize to y0 .
In this section, a new control strategy based on a fully dis- As the above mentioned lemma shows, to ensure stabil-
tributed cooperative control selects proper control inputs Vni in ity, the design of consensus protocol usually needs the in-
(1) to synchronize the DG units voltag magnitude , vo,m agi , to formation of mini=1,...,N Re(λi ): that is, the minimum real
the nominal voltage vref . Based on d−q transformation, the DG part of the nonzero Laplacian matrix eigenvalues. Note that
unit output voltage magnitude is mini=1,...,N Re(λi ) is global information (i.e., each DG must
have all the communication graph information to calculate

it). As a result, the controllers represented by Lemma 4.1 are
vo,m agi = vo2 di + vo2 qi . (3)
not fully distributed. Because of this shortcoming, inspired by

Authorized licensed use limited to: Indian Institute of Technology- Goa. Downloaded on June 01,2020 at 05:43:19 UTC from IEEE Xplore. Restrictions apply.
678 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 32, NO. 2, JUNE 2017

[29], we have designed fully distributed controllers for sys- Therefore, V will definitely be positive with respect to ei and
tem (7), whose communication graph G contains a directed c̃i . The time-derivative of V is
spanning tree. Note that just one DG receives the voltage set-
N N  T −1
ċi qi e i P e i
point; therefore, the voltage local neighborhood tracking error is V̇ = ci qi ρi (ei )eTi P −1 ėi + ρi (s)ds
defined as i=1 i=1
2 0

ei = (yi − yk ) + gi (yi − y0 ) (11) λ̂0 
N

k ∈N c i + (ci − α)eTi P −1 BB T P −1 ei . (16)

12 i=1
in which gi is called the pinning gain. Nci represents the commu-
nication neighborhood set of the ith protocol. Moreover, each By using (14) and GD L1 + LT1 GD ≥ λ̂0 I, we obtain
DG has access to its neighbors. Motivated by [29], we suggest 
N
the following distributed adaptive control: ci qi ρi (ei )eTi P −1 ėi = eT (Ĉ ρ̂GD ⊗ P −1 )ė
vi = ci ρi (eTi P −1 ei )Kei
i=1

1 T
ċi = eTi Γei , i = 1, 2, ..., N, (12) = e [Ĉ ρ̂GD ⊗ (P −1 A + AT P −1 )
2
in which P > 0 is a solution to the LMI (10) ; ci represents the − Ĉ ρ̂(GD L1 + LT1 GD )Ĉ ρ̂ ⊗ P −1 BB T P −1 ]e
time-varying coupling weight related to the ith DG with ci (0) ≥
1 T
1 ; Γ ∈ Rn ×n and K ∈ Rp×n represent the gain matrices to ≤ e [Ĉ ρ̂GD ⊗ (P −1 A + AT P −1 )
be determined; ρi (.) represents a strictly incremental function 2
to be designed, thus satisfying ρi (s) ≥ 1 for s > 0; and ei is − λ̂0 Ĉ 2 ρ̂2 ⊗ P −1 BB T P −1 ]e. (17)
defined in (11).
Because ρi represents a strictly incremental function and sat-
Let e = [eT1 , · · · , eTn ]T . Therefore, we define the global error
isfies ρi (s) ≥ 1 for s > 0, from [29] we have
vector e as
⎡ ⎤ N  e Ti P −1 e i
y1 − y0
⎢ ⎥ ċi qi ρi (s)ds
e = (L1 ⊗ In )⎣ ..
. ⎦, (13) i=1 0

yN − y0 N  3 
qi 2
≤ + λ̂ ρ2
eTi P −1 BB T P −1 ei . (18)
where In , ⊗, and e represent the unit matrix of size n, Kro- 3λ̂ 2 3
0 i
i=1 0
necker product, and consensus error, respectively. Substituting
for vi from (12) into (7) and performing some mathematical qi3
Select α > α̂ + maxi=1,··· ,N (2 ), where α̂ is to be deter-
manipulations yields λ̂30
 mined. Substituting (17) and (18) into (16) and using the fact
ė = In ⊗ A + L1 Ĉ ρ̂(e) ⊗ BK e that ρi ≥ 1 and ci ≥ 1, i = 1, ..., N yields
1 T
ċi = eTi Γei , (14) V̇ ≤ e [Ĉ ρ̂GD ⊗ (P −1 A + AT P −1 )]e
2
where ρ̂(e) = diag(ρ1 (e), ..., ρn (e)) and Ĉ= diag(c1 , ..., cn ).
λ̂0  2 2
N
The following lemma is necessary for designing the auxiliary − (c ρ + α̂)eTi P −1 BB T P −1 ei
controls vi . 12 i=1 i i
Lemma IV.2. There exists a positive diagonal matrix 1 T
GD such that GD L1 + LT1 GD > 0. GD is given by e [Ĉ ρ̂GD ⊗ (P −1 A + AT P −1 )
≤
2
diag(q1 , ..., qN ), where q = [q1 , ..., qN ]T = (LT1 )−1 1 and 1
1√
represents a column vector of appropriate dimension with its − α̂λ̂0 Ĉ ρ̂ ⊗ P −1 BB T P −1 ]e. (19)
elements equal to one [26]. 3

In what follows, we employ the stability conditions from [29]. We define ẽ = ( Ĉ ρ̂GD ⊗ I)e and select α̂ such that
To stabilize (14), by considering the positive definite Lyapunov √
function as α̂λ̂0 G−1
D ≥ 6I. Therefore, we have

 T −1 1  
N
ci qi e i P e i λ̂0  2
N
V̇ ≤ ẽT In ⊗ (P −1 A + AT P −1 − 2P −1 BB T P −1 ) ẽ ≤ 0.
V = ρi (s)ds + c̃ , (15) 2
i=1
2 0 24 i=1 i (20)
in which c̃i =ci − α, α denotes a positive scalar to be designed, Hence, P −1 A + AT P −1 − 2P −1 BB T P −1 < 0; therefore,
λ̂0 represents the minimum eigenvalue of GD L1 + LT1 GD , and V̇ ≤ 0. By virtue of (20), ẽ converges exponentially to zero−that
GD = diag(q1 , ..., qN ) is determined as in Lemma 4.2 such is, the consensus error e converges exponentially to zero−thus
that GD L1 + LT1 GD > 0. Because ci (0) ≥ 1, we may conclude solving the problem of the leader-follower consensus. Moreover,
from (14) that ci (t) ≥ 1 for t > 0. Moreover, ρi (.) represents a the consensus protocol (21) is designed by solving the algebraic
strictly incremental function that satisfies ρi (s) ≥ 1 for s > 0. Ricatti equation [20]: AT P + P A + Q − P BR−1 B T P = 0,

Authorized licensed use limited to: Indian Institute of Technology- Goa. Downloaded on June 01,2020 at 05:43:19 UTC from IEEE Xplore. Restrictions apply.
DEHKORDI et al.: FULLY DISTRIBUTED COOPERATIVE SECONDARY FREQUENCY AND VOLTAGE CONTROL OF ISLANDED MICROGRIDS 679

where R > 0 (positive definite) and Q ≥ 0 (positive semi-  according to the algebraic Ricatti equation, Kω =
where,
definite). Further, the control parameters in (21) are chosen as Qω
− and Qω and Rω are positive scalars. According to
Γ = P BR−1 B T P , K = −R−1 B T P , and ρi = (1 + eTi Qei )3 . Rω
As a result, the continuous protocol (25) and (27), we can write ωni as

1
Vni = ui1 = (vi − fi (xi )) ωn i = (uω i + mP i Ṗi )dt, i = 1, 2, ..., N. (28)
gi (xi )
vi = ci (1 + eTi Qei )3 Kei We note that in the presence of the primary control, the DG
output powers meet
ċi = eTi Γei , i = 1, 2, ..., N
Pj mpi
0 = AT P + P A + Q − P BR−1 B T P = , ∀i, j ∈ N. (29)
Pi mpj
K = −R−1 B T P When the secondary frequency control is activated, the power
−1 T of DGs must also satisfy (29). Therefore, we consider an extra
Γ = P BR B P (21)
distributed cooperative term for mP i Ṗi in (28). We must design
can guarantee that the direct term of the DG voltage vo di tracks the auxiliary controls upi in (30) to ensure the accuracy of real
the reference value vref . Fig. 2 shows the schematic diagram power sharing (i.e., mpk Pk −→ mpi Pi , ∀i):
of the fully distributed adaptive secondary voltage control. The
control parameters K, Γ, R, and Q can adjust the convergence mP i Ṗi = up i . (30)
rate of the DG voltage. The auxiliary controls upi are chosen as in (27), as
V. FULLY DISTRIBUTED ADAPTIVE SECONDARY up i = cp i (1 + QP e2p i )3 KP ep i
FREQUENCY CONTROL
ċp i = QP e2p i , i = 1, 2, ..., N, (31)
In this section, a fully distributed cooperative control selects 
proper control inputs ωni in (1) to synchronize the frequency of in which ep i = k ∈N c i (mP i Pi − mP k Pk ) represents the con-

DG units, ωi , to the reference value , ωref . We assume that DG QP
sensus tracking error, KP = − , and QP and RP are
units can communicate with one another through the prescribed RP
communication network-directed graph G. Therefore, the sec- positive control parameters. According to (28) and (30), we can
ondary frequency control selects ui2 in (2) so that yi2 −→ ωref . write ωni as
As in the previous section, we can obtain the partial feedback 
linearization for the system (2) as ωn i = (uω i + up i )dt, i = 1, 2, ..., N. (32)
ẏi2 = Lf i hi2 + Lg i hi2 ui2 + di u̇i2 Therefore, the control inputs ωni solve the frequency synchro-
μ̇i2 = Wi2 (yi2 , μi2 ), (22) nization problem while guaranteeing the accuracy of real power
sharing. As (32) demonstrates, the secondary control ωn i con-
where μi2 ∈ Rn i 2 represents the set of internal dynamics that tains two parts. The first part of the controller, uω i , leads to the
are asymptotically stable [13]. Equation (22) can be written as steady-state track of the nominal frequency (i.e., ωi −→ ωref ),
follows: whereas the second part, up i , ensures real power sharing ac-
ẏi2 = ω̇i = ω̇ni − mP i Ṗi . (23) curacy (i.e., mpj Pj −→ mpi Pi ). Unlike the conventional dis-
tributed methods, the proposed method is fully distributed (i.e.,
Now we define the auxiliary control uω i as the design of ωni is based on its own and its neighbors’ infor-
uω i = ω̇ni − mP i Ṗi , (24) mation).
Fig. 3 shows the block diagram of the fully distributed co-
where (24) determines the feedback linearization approach lead- operative secondary frequency control. The control parameters
ing to the following linear systems: Qω , QP , Rω , and RP can adjust the convergence rate of the DG
frequency and filtered power.
ω̇i = uω i , i = 1, 2, ..., N. (25)
Now we define the consensus frequency tracking error as VI. SIMULATION RESULTS

eω i = (ωi − ωk ) + gi (ωi − ωref ). (26) To evaluate the effectiveness of the proposed cooperative con-
k ∈N c i trol method, we have simulated a 380 V, 50 Hz islanded micro-
grid, as Fig. 4 depicts, under various scenarios in the MAT-
As in the previous section, the following proposed control
LAB/SimPowerSystems software environment. The islanded
method provides the synchronization:
microgrid test consists of five inverter-based DG units and six
uω i = cω i (1 + Qω e2ω i )3 Kω eω i loads.
The loads are modeled as series RL branches. Table I provides
ċω i = Qω e2ω i , i = 1, 2, ..., N, (27) the DG units, loads, and lines parameters of the understudy

Authorized licensed use limited to: Indian Institute of Technology- Goa. Downloaded on June 01,2020 at 05:43:19 UTC from IEEE Xplore. Restrictions apply.
680 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 32, NO. 2, JUNE 2017

Fig. 2. Schematic diagram of the fully distributed adaptive secondary voltage control.

Fig. 3. Schematic diagram of the fully distributed adaptive secondary frequency control.

TABLE I
PARAMETERS OF THE MICROGRID SYSTEM

DG # 1&2 DG # 3&4&5

DGs mP 4.7 × 10−5 mP 6.25 × 10−5

nQ 0.65 × 10−3 nQ 0.75 × 10−3
Zc .03 + j 0.22 Ω Zc .03 + j 0.11 Ω
Lf 1 , Lf 2 1.35, 0.27 mH Lf 1 , Lf 2 1.35, 0.27 mH
Rf 1 , Rf 2 0.1, .05 Ω Rf 1 , Rf 2 0.1, .05 Ω
Cf 47 μF Cf 47 μF
KP V 0.1 KP V 0.05
KIV 420 KIV 390
KP C 15 KP C 10.5
KIC 20000 KIC 16000
Lines ZL i n e 1 , ZL i n e 4 0.06 + j 0.025 Ω ZL i n e 3 0.05 + j 0.07 Ω
Zl i n e 2 0.05 + j 0.07 Ω Zl i n e 5 0.1 + j 0.15 Ω
Loads Load # 1 R = 80 Ω, L = 128 mH Load # 3 , Load # 5 R = 100 Ω, L = 128 mH
Load # 2 R = 120 Ω, L = 190 mH Load # 4 R = 100 Ω, L = 190 mH

Fig. 5. Topology of the communication graph.

system in detail. In this table, KPV , KIV , KPC , and KIC are
the corresponding current and voltage controller parameters.
We assume that DG units communicate with one another
through the directed graph (Fig. 5). For both the frequency
and voltage restoration problems, we consider the voltage and
frequency references to be the leader node outputs.
Moreover, only DG #1 is connected to the leader node with
the pinning gain g = 1. Note that to verify the robustness of the
Fig. 4. Block diagram of the islanded microgrid. proposed controllers with respect to the parameters, we have

Authorized licensed use limited to: Indian Institute of Technology- Goa. Downloaded on June 01,2020 at 05:43:19 UTC from IEEE Xplore. Restrictions apply.
DEHKORDI et al.: FULLY DISTRIBUTED COOPERATIVE SECONDARY FREQUENCY AND VOLTAGE CONTROL OF ISLANDED MICROGRIDS 681

Fig. 6. The real and reactive power of the microgrid loads.

performed the simulations with 50% uncertainty in the filter

parameters, as Table I shows. For the corresponding
 secondary

100 0
controller parameters, we have chosen Q = , R=
0 1
0.01, Qω = 5, Rω = 0.01, Qp = 100, and Rp = 0.01. The
simulation scenario proceeds as follows:
1) t = 0.0 s (simulation initialization period). Only the pri- Fig. 7. DGs (a) frequency, (b) output real power, and (c) real power ratio.
mary control is activated, and Load #6 is out of the circuit.
2) t = 1.5 s. The proposed cooperative secondary control is
applied.
3) t = 2.5 s. Load #2 is increased.
4) t = 3.5 s. Load #3 is disconnected.
5) t = 4.5 s. Switch S in Fig. 4 is opened.
Fig. 6 depicts the real and reactive power components of the
microgrid loads.
First, at the beginning of the simulation, the microgrid is dis-
connected from the grid and goes through the islanded mode
condition. During the first 1.5 s of the operation, the microgrid
works under only the primary controller. As Figs. 7(a) and 8(a)
show, especially during the first 1.5 s of operation, the DGs ter-
minal voltage magnitude, vo,m agi , and the microgrid frequency
deviate from their nominal values by their droop controller.
Therefore, the microgrid frequency and voltages must be re-
stored in the secondary control layer of each inverter-based DG.
To show the effectiveness of the proposed distributed coop-
erative control, we apply the secondary control at t = 1.5 s. As
Figs. 7(a) and 8(a) show, the cooperative control can remove Fig. 8. (a,b) DGs output voltage magnitude and reactive power.
the droop-controller-produced deviations, thus rapidly restor-
ing both frequency and voltages to their nominal values (vref = We then investigate the robust performance of the proposed
ωref
380 V, fref = = 50 Hz). Fig. 7(c) depicts the real powers cooperative distributed secondary control in the presence of load
2π
multiplied by the real power droop coefficients (i.e., mP i Pi ) changes. After the previous case study at t = 2.5 s, the islanded
for DG units. As we can see, the frequency controller fulfills microgrid experiences an overload accident. An RL load with R
(29). In other words, after applying the cooperative secondary = 120 Ω and L = 47.6 mH is connected in parallel to Load #2.
control, the real power sharing is satisfied according to the rated Figs. 7(a) and 8(a) show that the proposed secondary control
power of the DG units. demonstrates good tracking and robust performance against the

Authorized licensed use limited to: Indian Institute of Technology- Goa. Downloaded on June 01,2020 at 05:43:19 UTC from IEEE Xplore. Restrictions apply.
682 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 32, NO. 2, JUNE 2017

Fig. 10. Performance comparison between the proposed secondary voltage

control and the one presented in [13].
Fig. 9. (a,b) The microgrid frequency and voltage magnitude of Bus 2.

restoration method can effectively compensate for the microgrid

connection of the RL load and accurately restores the microgid frequency and voltages deviations (which the primary controller
voltages and frequency. Fig. 7(c) depicts the accuracy of the imposes) with uncertain parameters.
proposed strategy for sharing the real power among all five DG
units.
To highlight the robust performance of the proposed method A. Comparing the Proposed Method with the Methods in [12]
against load variations at t = 3.5 s, Load #3 is suddenly dis- and [13]
connected from the islanded microgrid. The results confirm that As Lemma 4.1 stated previously, the general form of dis-
the distributed cooperative secondary control can restore the tributed control methods in the literature, especially in [12]–
microgrid frequency and voltages after a short transient time, [14], is of the form given in (9). According to Lemma
as Figs. 7(a) and 8(a) show. Moreover, the effectiveness of the 4.1, consensus protocol design usually needs the informa-
proposed strategy for sharing the real power among all five DG tion of mini=1,...,N Re(λi ): that is, the minimum real part
units is satisfied. of the nonzero Laplacian matrix eigenvalues. We note that
To exhibit the plug-and-play capability of the intermittent mini=1,...,N Re(λi ) is global information (i.e., each DG must
DG units, the islanded microgrid is perturbed when the DG have all the communication graph information to calculate it).
units are disconnected: at t = 4.5 s, switch S is opened and Therefore, when the microgrid size increases, the global com-
DG #5 is disconnected from the islanded test microgrid. As munication graph structure changes. Consequently, the lower
Figs. 7(a) and 8(a) show, the distributed cooperative controllers bound on the coupling gain c is changed, leading to the degra-
robustly regulate the microgrid frequency and voltages after the dation of protocol performance. However, the proposed proto-
disconnection of DG #5. Once DG #5 is disconnected from cols are sparse and fully distributed so that they are scalable
the microgrid, the remaining DG units generate extra power to to large-scale microgrid systems. The computation complexity
compensate for the amount of power that DG #5 has already of the proposed protocols is much lower than that of exist-
produced. As a result, the power ratios of the remaining DGs ing methods and is independent of the total number of invert-
increase, whereas those of DG #5 drop to zero, as shown in ers in the microgrid (except the number of neighbors of each
Fig. 7(c). DG). Therefore, the proposed protocols rapidly reach consen-
The output real power of the DG units is depicted in Fig. 7(b). sus and show more accurate robust performance for controller
As this figure shows, the real power sharing among DG units is activation, structure reconfiguration, and load changes than do
accurately performed according to the droop coefficients of DG asymptotic controllers.
units in various load changes and microgrid reconfigurations, As previously stated in the introduction, few studies solve
whereas the proposed secondary control restores the islanded both voltage and frequency restoration problems. We compare
microgrid frequency and voltages. Fig. 8(b) shows the reactive the proposed distributed voltage controller (21) with the one pre-
power of DGs. Because of the transmission line impedance, we sented in [13] and the proposed distributed frequency controller
observe little difference in the reactive power of DGs. Moreover, (32) with the one presented in [12]. Figs. 10 and 11 show the
Fig. 9 shows the microgrid frequency and voltage magnitude of simulation results when the distributed voltage control in [13]
Bus 2. As this figure shows, the proposed frequency and voltage and the distributed frequency control in [12] are used to restore

Authorized licensed use limited to: Indian Institute of Technology- Goa. Downloaded on June 01,2020 at 05:43:19 UTC from IEEE Xplore. Restrictions apply.
DEHKORDI et al.: FULLY DISTRIBUTED COOPERATIVE SECONDARY FREQUENCY AND VOLTAGE CONTROL OF ISLANDED MICROGRIDS 683

 10 0

Fig. 12. DGs output voltage magnitude with (a) Q = and R =
 100 0
 0 1
0.01 and (b) Q = 0 1 and R = 0.1.

quadratic energy function:

 t
 
Ji = (yi − y0 )T Q(yi − y0 ) + viT Rvi dt, (33)
0
where R > 0 and Q ≥ 0 represent weighting matrices over the
inputs and states, respectively, balancing the optimization be-
tween the convergence speed and the required control effort.
The larger Q and/or smaller R result in smaller (yi − y0 ), which
leads to small Ji and improves the dynamic response of the
Fig. 11. Performance comparison between the proposed secondary frequency DG. To demonstrate the control parameters effect on the con-
control and the one presented in [12]. (a) and (b) DG #1 frequency and real
power. vergence rate of the secondary voltage control, we consider
two
 different
 cases. In the first case, Q and R are set as Q =
10 0
and R = 0.01. The (1,1)−th entry of matrix Q that
the output voltage magnitudes and frequency of the microgrid. 0 1
The control gains are set the same in both our proposed method directly affects the control of vo di is selected to be smaller than
and the ones presented in [12] and [13]. For the sake of sim- in the previous study, whereas the (2,2)−th entry of matrix
plicity, we depict only the frequency and voltage responses of Q directly affects the control of v̇o di . Moreover, matrix Q can
DG #1. The results of Figs. 10 and 11 verify the following ad- be non-diagonal in general; however, in this study the conver-
vantages. First, unlike the distributed protocols in [12] and [13], gence rates of vo di and v̇o di can be independently determined
our proposed protocols rapidly reach consensus and show more using (1,1)−th and (2,2)−th entry of matrix Q, respectively. As
accurate robust performance for controller activation, structure Fig. 12(a) shows, with a larger Q, the output voltage magnitudes
reconfiguration, and load changes. Second, by increasing pro- track the reference value more slowly than does the case study
tocol gains, the conventional methods may obtain the same per- in Fig. 8(a) for controller activation, structure reconfiguration,
formance as our method, but our proposed method yields better  loadchanges. In the latter case, Q and R are set as Q =
and
disturbance rejection properties in structure reconfiguration and 100 0
and R = 0.1.
load changes. Finally, because the computation complexity of 0 1
the existing distributed methods is higher than that of the one, As Fig. 12(b) shows, with a larger R, the proposed method is
the existing methods may excite the high-frequency dynamics. slower than the case studied in Fig. 8(a) for controller activation,
structure reconfiguration, and load changes.

B. Control Parameters Effect on the Dynamic Response VII. CONCLUSION

The parameters of the proposed protocols have a direct impact This paper presents a fully distributed cooperative sec-
on the dynamic response of the DG units. The optimal control ondary control for both frequency and voltage restoration of an
parameters are calculated such that they minimize the following islanded microgrid with accurate real power sharing among

Authorized licensed use limited to: Indian Institute of Technology- Goa. Downloaded on June 01,2020 at 05:43:19 UTC from IEEE Xplore. Restrictions apply.
684 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 32, NO. 2, JUNE 2017

DG units. The microgrid is considered a multi-agent system [11] F. Guo, C. Wen, J. Mao, and Y.-D. Song, “Distributed secondary volt-
in which DG units communicate with one another through a age and frequency restoration control of droop-controlled inverter-based
microgrids,” IEEE Trans. Ind. Electron., vol. 62, no. 7, pp. 4355–4364,
directed sparse communication network. In contrast to exist- Jul. 2015.
ing distributed methods, a distributed adaptive method whose [12] A. Bidram, A. Davoudi, F. Lewis, and Z. Qu, “Secondary control of mi-
design is based on the DG unit dynamic model and on the in- crogrids based on distributed cooperative control of multi-agent systems,”
IET Gener., Transm. Distrib., vol. 7, no. 8, pp. 822–831, Aug. 2013.
formation of neighboring units is fully distributed. Therefore, [13] A. Bidram, A. Davoudi, F. Lewis, and J. Guerrero, “Distributed coopera-
the proposed secondary control increases the system reliability tive secondary control of microgrids using feedback linearization,” IEEE
and eliminates the need for a central computing and communi- Trans. Power Syst., vol. 28, no. 3, pp. 3462–3470, Aug. 2013.
[14] A. Bidram, A. Davoudi, and F. Lewis, “A multiobjective distributed con-
cation unit. Not only does the controller consider the coupling trol framework for islanded AC microgrids,” IEEE Trans. Ind. Informat.,
between the frequency and voltage control loop, but it is also vol. 10, no. 3, pp. 1785–1798, Aug. 2014.
robust with respect to parametric uncertainties, unmodeled dy- [15] Q. Shafiee, J. Guerrero, and J. Vasquez, “Distributed secondary control
for islanded microgrids -A novel approach,” IEEE Trans. Power Electron.,
namics, and disturbances in the microgrid system. We note that vol. 29, no. 2, pp. 1018–1031, Feb. 2014.
using a new fully distributed cooperative secondary control for [16] J. W. Simpson-Porco, F. Drfler, and F. Bullo, “Synchronization and power
both frequency and voltage restoration of an islanded microgrid sharing for droop-controlled inverters in islanded microgrids,” Automat-
ica, vol. 49, no. 9, pp. 2603–2611, Sep. 2013.
system has not before been proposed or investigated before. To [17] H. Cai, G. Hu, F. Lewis, and A. Davoudi, “A distributed feedforward
evaluate the performance of the proposed method, we carried out approach to cooperative control of AC microgrids,” IEEE Trans. Power
various simulation studies in the MATLAB/SimPowerSystems Syst., vol. 31, no. 5, pp. 4057–4067, Sep. 2016.
[18] C. Ahumada, R. Crdenas, D. Sez, and J. M. Guerrero, “Secondary control
toolbox. The simulation results verify that the presented strat- strategies for frequency restoration in islanded microgrids with consider-
egy: ation of communication delays,” IEEE Trans. Smart Grid, vol. 7, no. 3,
1) robustly restores the frequency and voltages of the over- pp. 1430–1441, May 2016.
[19] V. Nasirian, Q. Shafiee, J. Guerrero, F. Lewis, and A. Davoudi, “Droop-
all microgrid to their reference values in the presence of free distributed control for AC microgrids,” IEEE Trans. Power Electron.,
parametric uncertainties, unmodeled dynamics, and dis- vol. 31, no. 2, pp. 1600–1617, Feb. 2016.
turbances in the microgrid system, in a fully distributed [20] T. Morstyn, B. Hredzak, and V. Agelidis, “Distributed cooperative con-
trol of microgrid storage,” IEEE Trans. Power Syst., vol. 30, no. 5,
manner while maintaining the accuracy of real power shar- pp. 2780–2789, Sep. 2015.
ing; and [21] F. Chen, M. Chen, Q. Li, K. Meng, J. M. Guerrero, and D. Abbott,
2) enables the system with plug-and-play capability (i.e., the “Multiagent-based reactive power sharing and control model for islanded
microgrids,” IEEE Trans. Sustain. Energy, vol. 7, no. 3, pp. 1232–1244,
possibility of adding/removing a DG unit, line, and/or Jul. 2016.
load without distorting the microgrid stability). [22] M. A. Mahmud, M. J. Hossain, H. R. Pota, and A. M. T. Oo, “Robust
nonlinear distributed controller design for active and reactive power shar-
ing in islanded microgrids,” IEEE Trans. Energy Convers., vol. 29, no. 4,
REFERENCES pp. 893–903, Dec. 2014.
[23] A. Bidram, A. Davoudi, and F. Lewis, “Finite-time frequency synchroniza-
[1] S.-H. Ko, S. R. Lee, H. Dehbonei, and C. V. Nayar, “Application of tion in microgrids,” in Proc. IEEE Energy Convers. Congr. and Exposition,
voltage- and current-controlled voltage source inverters for distributed Sep. 2014, pp. 2648–2654.
generation systems,” IEEE Trans. Energy Convers., vol. 21, no. 3, [24] A. Bidram, A. Davoudi, F. L. Lewis, and S. S. Ge, “Distributed adap-
pp. 782–792, Sep. 2006. tive voltage control of inverter-based microgrids,” IEEE Trans. Energy
[2] J. Rocabert, A. Luna, F. Blaabjerg, and P. Rodrguez, “Control of power Convers., vol. 29, no. 4, pp. 862–872, Dec. 2014.
converters in AC microgrids,” IEEE Trans. Energy Convers., vol. 27, [25] V. Nasirian, A. Davoudi, F. L. Lewis, and J. M. Guerrero, “Distributed
no. 11, pp. 4734–4749, Nov. 2012. adaptive droop control for DC distribution systems,” IEEE Trans. Energy
[3] J. Peas Lopes, C. Moreira, and A. Madureira, “Defining control strategies Convers., vol. 29, no. 4, pp. 944–956, Dec. 2014.
for microgrids islanded operation,” IEEE Trans. Power Syst., vol. 21, [26] Z. Qu, Cooperative Control of Dynamical Systems: Applications to Au-
no. 2, pp. 916–924, May 2006. tonomous Vehicles. New York, NY, USA: Springer, 2009.
[4] A. Bidram and A. Davoudi, “Hierarchical structure of microgrids control [27] H. K. Khalil, Nonlinear Systems, 3rd ed. Upper Saddle River, NJ, USA:
system,” IEEE Trans. Smart Grid, vol. 3, no. 4, pp. 1963–1976, Dec. 2012. Prentice-Hall, 2002.
[5] D. Wu, F. Tang, T. Dragicevic, J. C. Vasquez, and J. M. Guerrero, “Au- [28] H. Zhang, F. Lewis, and A. Das, “Optimal design for synchronization of
tonomous active power control for islanded AC microgrids with pho- cooperative systems: State feedback, observer and output feedback,” IEEE
tovoltaic generation and energy storage system,” IEEE Trans. Energy Trans. Autom. Control, vol. 56, no. 8, pp. 1948–1952, Aug. 2011.
Convers., vol. 29, no. 4, pp. 882–892, Dec. 2014. [29] Z. Li, G. Wen, Z. Duan, and W. Ren, “Designing fully distributed consen-
[6] Y. Han, P. Shen, X. Zhao, and J. M. Guerrero, “An enhanced power sharing sus protocols for linear multi-agent systems with directed graphs,” IEEE
scheme for voltage unbalance and harmonics compensation in an islanded Trans. Autom. Control, vol. 60, no. 4, pp. 1152–1157, Apr. 2015.
microgrid,” IEEE Trans. Energy Convers., vol. 31, no. 3, pp. 1037–1050,
Sept. 2016.
[7] J. Guerrero, J. Vasquez, J. Matas, L. de Vicua, and M. Castilla, “Hier-
archical control of droop-controlled AC and DC microgrids -A general
approach toward standardization,” IEEE Trans. Ind. Electron., vol. 58, Nima Mahdian Dehkordi was born in Shahrekord,
no. 1, pp. 158–172, Jan. 2011. Iran, on September 20, 1988. He received the B.Sc
[8] B. Marinescu and H. Bourles, “Robust predictive control for the flexi- degree in electrical engineering from the University
ble coordinated secondary voltage control of large-scale power systems,” of Shahrekord, Shahrekord, Iran, in 2010 and the
IEEE Trans. Power Syst., vol. 14, no. 4, pp. 1262–1268, Nov. 1999. M.Sc. degree in electrical engineering from the Sharif
[9] J. Schiffer, T. Seel, J. Raisch, and T. Sezi, “Voltage stability and re- University of Technology, Tehran, Iran, in 2012. He
active power sharing in inverter-based microgrids with consensus-based is currently working toward the Ph.D. degree in
distributed voltage control,” IEEE Trans. Control Syst. Technol., vol. 24, electrical engineering at Sharif University of Tech-
no. 1, pp. 96–109, Jan. 2016. nology, Tehran, Iran. His research interests include
[10] J. W. Simpson-Porco, Q. Shafiee, F. Dfler, J. C. Vasquez, J. M. Guerrero, control systems, applications of control theory in
and F. Bullo, “Secondary frequency and voltage control of islanded mi- power electronics, microgrid control, distributed and
crogrids via distributed averaging,” IEEE Trans. Ind. Electron., vol. 62, cooperative control, nonlinear control, and network control.
no. 11, pp. 7025–7038, Nov. 2015.

Authorized licensed use limited to: Indian Institute of Technology- Goa. Downloaded on June 01,2020 at 05:43:19 UTC from IEEE Xplore. Restrictions apply.
DEHKORDI et al.: FULLY DISTRIBUTED COOPERATIVE SECONDARY FREQUENCY AND VOLTAGE CONTROL OF ISLANDED MICROGRIDS 685

Nasser Sadati (M’95) is a Full Professor with the De- Mohsen Hamzeh (S’09–M’13) received the B.Sc
partment of Electrical Engineering, Sharif University and M.Sc degrees from the University of Tehran,
of Technology, Tehran, Iran, where he is the Direc- Tehran, Iran, in 2006 and 2008, respectively, and
tor of the Intelligent Systems Laboratory, and the the Ph.D. degree from Sharif University of Tech-
Founder of research on intelligent control of large- nology, Tehran, Iran, in 2012, all in electrical en-
scale systems. He has published two control books gineering. Since 2010, he has been the Senior Re-
in Persian and nearly 300 technical papers in peer- search Engineer with the SGP Company, Tehran,
reviewed journals and conference proceedings. His Iran. In 2013, he joined the Department of Electrical
book published by Wiley-IEEE Press (2012) is on hy- and Computer Engineering, Shahid Beheshti Univer-
brid control and motion planning of dynamical legged sity, Tehran, Iran, where he is currently an Assistant
locomotion. His research interests include intelligent Professor. His research interests include renewable
control, large-scale distributed control systems, applications of control theory in energies, microgrid control and applications of power electronics in power dis-
power systems, microgrid control, control of biped locomotion, multi-modeling, tribution systems.
and robust control.
He is a Technical Committee Member on distributed intelligent systems of
the IEEE Systems, Man, and Cybernetics Society.

Authorized licensed use limited to: Indian Institute of Technology- Goa. Downloaded on June 01,2020 at 05:43:19 UTC from IEEE Xplore. Restrictions apply.