See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/341671702

Robust Sliding Mode and Mixed H2/H∞ Output Feedback Primary Control of
AC Microgrids

Article in IEEE Systems Journal · June 2021

DOI: 10.1109/JSYST.2020.2999553

CITATIONS READS
26 701

4 authors:

Mohammad Raeispour Hajar Atrianfar

Amirkabir University of Technology Amirkabir University of Technology
10 PUBLICATIONS 106 CITATIONS 42 PUBLICATIONS 333 CITATIONS

SEE PROFILE SEE PROFILE

Hamid Reza Baghaee Gevork B. Gharehpetian

Amirkabir University of Technology Amirkabir University of Technology
177 PUBLICATIONS 3,994 CITATIONS 1,048 PUBLICATIONS 18,521 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

A Novel Optimization algorithm, meta-heuristic method for real parameter optimization View project

Impact of converting gas consumed loads into electric loads on distribution systems for resiliency improvement View project

All content following this page was uploaded by Mohammad Raeispour on 03 September 2020.

The user has requested enhancement of the downloaded file.

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

IEEE SYSTEMS JOURNAL 1

Robust Sliding Mode and Mixed H2/H∞ Output

Feedback Primary Control of AC Microgrids
Mohammad Raeispour , Hajar Atrianfar , Member, IEEE, Hamid Reza Baghaee , Member, IEEE,
and Gevork B. Gharehpetian , Member, IEEE

Abstract—In this article, a unified, robust hierarchical control Current control loop parameters
structure is proposed for operating networked multi distributed
generation grid-forming power converters in islanded alternating
λ, β1 , β2 Positive constant weight coefficients related to
current (ac) microgrids (MGs). The primary control level stabilizes sliding surface.
the frequency and the voltage of an ac MG through a cascaded p, q Positive odd integers related to sliding surface.
structure, consisting of a droop controller, a mixed H2 /H∞ -based γ1 , γ2 Upper bound of sum of the uncertainties and distur-
voltage controller, and an adaptive backstepping integral non- bances.
singular fast terminal sliding mode control (ABINFTSMC) for γ̃1 , γ̃2 Upper bound estimation of sum of the uncertainties
current loop. In the inner control loop, an adaptive robust nonlinear and disturbances.
controller is designed for regulating and tracking of the current
reference in the presence of unknown bounded uncertainties and
external disturbances. The outer loop controller design problem is Voltage-control loop parameters
formulated by a set of linear matrix inequalities and then solved as
a multiobjective optimization problem, by using a fuzzy decision- G, Gd Plant transfer function and transfer function from
making tool (FDMT) to provide the best tradeoff solution. Besides, the disturbance to the output.
the conventional distributed protocol has been used in the sec- So Output sensitivity function.
ondary layer to compensate for the deviation caused by the primary W̄d Wr Weight functions to control signal.
layer. Finally, the effectiveness of the proposed control scheme is Trz1 , Trz3 Transfer function from reference input to the
evaluated through offline time-domain simulation studies based on output z1 , z3 .
different scenarios in the MATLAB/Simulink environment for a Tdz2 Transfer function from disturbance signal to the
test MG system. output z2 .
Index Terms—Distributed generation, microgrid (MG), mixed
H2 /H∞ control, multiobjective optimization, robustness, sliding
mode control. Multiobjective optimization parameters
Fi ith normalized objective function.
NOMENCLATURE j Number of objectives that should be minimized.
nobj Total number of objective functions.
Distributed generation parameters Fin , μi Value of the ith objective function in the nth Pareto
optimal solution.
Vinom , ωinom Voltage and frequency nominal reference from
the secondary layer.
ωi , ω Operating frequency of the units and frequency Secondary control layer parameters
of the common reference frame.
Vodq , Iodq Direct and quadrature components of the output cv , cω , cP , cQ Positive constant gains to tune voltage,
voltage and current. frequency, active, and reactive power.
Q
Vtdq , Itdq Direct and quadrature components of the uvi , uω P
i , ui , ui Secondary control signals of voltage, fre-
terminal voltage and current. quency, active, and reactive power.
Vref , ωref Voltage and frequency references from the
secondary layer. I. INTRODUCTION
Pi , Q i Low-pass filtered values of output active and
reactive power. A. Motivation and Literature Review
P Q
mi , ni Droop coefficients for active power and reactive ICROGRIDS (MGs) refer to a cluster of interconnected
power.
Rf , Lf , Cf Nominal values of resistance, inductance and
M small-scale (almost low-voltage) renewable/distributed
energy resources (RERs/DERs) such as distributed generations
capacitance of the grid filter. (DGs), energy storage systems, and loads. The main goal of
MGs is locally solving energy issues and enhance flexibility by
Manuscript received January 7, 2020; revised March 30, 2020; accepted May providing an efficient way of supplying and consuming energy
27, 2020. (Corresponding author: Hajar Atrianfar.) (energy management). The basic and the most commonly used
The authors are with the Department of Electrical Engineering, Amirkabir
University of Technology, Tehran 15914, Iran (e-mail: [email protected];
control structure for MGs is the hierarchical control, includ-
[email protected]; [email protected]; [email protected]). ing primary control (droop control and primary stabilization),
Digital Object Identifier 10.1109/JSYST.2020.2999553 secondary control (restoration of voltage and frequency), and

1937-9234 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 06:54:31 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE SYSTEMS JOURNAL

tertiary control (optimal operation) [1], [2]. In the primary on a single objective such as the voltage, current, or power
layer, a current controller is more perfect; meanwhile, it per- control, which is a restricted contribution. Some other robust
mits precise restrictions and protections versus overcurrent and nonlinear control strategies for MGs have been reported in [18]
detracts the harmonic distortion. Besides, the voltage controller and [19]. SMC, as another robust method can be combined
ensures the regulation performance of voltage reference, and the with other techniques, due to this, adaptive SMC [20], fuzzy
current loop confirms fast dynamic compensation for system SMC [21], and fractional order SMC [22] have been discussed
disturbances [3]. The PI-based controllers in the primary layer, in MG control.
on account of competence and efficiency, have been extensively Next, the secondary control level will restore the fluctuations
employed in the balanced operation of MGs [3]. Besides, the caused by the primary layer to regulate voltage and frequency
parameters of the cascaded PI controllers are difficult to tune. of the islanded MG [23]. Multiagent systems have drawn sub-
They need to exact model of the system to achieve the intimate stantial attention from various scientific communities with a
performance of low total harmonic distortion (THD) and fast rather diverse background, such as biology and control the-
dynamic response and usually have a narrow stable operating ory [24], [25]. Distributed control has been widely discussed
range when the switching frequency is low. LC grid filters are for secondary layer of MGs, for example some aspect such as
generally adopted to alleviate the switching ripples. However, time delays [26], noise [27], [28], fault [29], [30], and uncertain
if it is not appropriately controlled, the inherent resonances time-delayed communications [31] distributed SMC-based pro-
of the LC filters can lead to possible oscillatory and unstable tocol [32] with considering the uncertainty have been addressed
behavior [4]. However, when the MG is faced with disturbances in the literature.
or perturbations, the PI-based approach is not favorable. Conse-
quently, many advanced control approaches, such as predictive B. Contributions and Article Organization
control [5], [6], intelligent control, and neural-network-based
control [7]–[9] have been applied for the current and voltage- The main focus of this article is devoted to the primary layer,
loop control of inverters. However, there are some disadvantages including the cascade structure of inner (current), an outer (volt-
associated with these methods, such as the predictive control age), and droop control loops. To improve the performance of
needs exact system parameters to reach the desired performance MG, and enhance the robustness and efficiency of MG, this study
and probably concerns for their lack of enough robustness. The proposes a novel robust hierarchical control strategy. Besides,
neural-network-based control requirements intricate training we design a robust primary control of MG with two different
procedure. For intelligent control, a learning process is required, kinds of the controller. To control the current, a robust adaptive
and the learning rate parameters are difficult to tune. So, the backstepping integral nonsingular fast terminal sliding mode
advanced robust control plan contains sliding mode control control (ABINFTSMC) strategy is exploited to reject the distur-
(SMC), and H∞ can have superior performance in MGs. The H2 bances and guarantee the robustness of the MG. In the voltage
control has been extensively applied to converse with the tran- loop, by utilizing the robust mixed H2 /H∞ controller, sufficient
sient performance of a control system; however, the H2 control condition is presented based on a MOO problem. Unlike existing
may not necessarily be robust respect to changes in the system the reported control strategies, uncertainty in the parameters are
parameter variation and cannot handle the disturbances [10]. considered. Moreover, the fuzzy decision-making tool (FDMT)
Besides, robust control schemes based on H∞ have been pro- is used to find the best tradeoff between solution. To overcome
posed to mitigate different MG control problems [11]–[13]. the fluctuations of voltage and frequency stabilization caused by
In [14], by using H∞ and repetitive control approach, a control the primary layer, a distributed consensus-based protocol is used
protocol has been suggested for compensation of the distorted in the secondary control layer to produce the reference signal for
load currents; this strategy leads to low THD compared with the primary control loops.
the traditional proportional resonant, PI, and predictive deadbeat The rest of this article is organized as follows. Section II
controllers, for the output current. Also, a two degree of freedom describes the primary layer and the controller design procedure.
master–slave H∞ method has been proposed to address the same The current, voltage, and power controllers are designed in this
problem [15]. In this method, just the voltage is controlled based section. In Section III, a distributed consensus-based protocol is
on H∞ control, and active and reactive powers are controlled used in the secondary control layer. In Section IV, the verifica-
conventionally. Later, in [16], this method has been extended for tion of the performance of the proposed control strategy based on
voltage regulation in MGs, including unbalanced and nonlinear offline digital time-domain simulations in MATLAB/Simulink
loads to reduce the voltage THD and the voltage unbalance software environment and comparison with other reported tech-
factor for sensitive loads. The H∞ control contracts with the niques are provided. Finally, Section V concludes this article.
uncertainties of the system parameters at the expenditure of
scarifying other performance of the system, as H∞ control II. DESIGN OF PRIMARY CONTROLLER
has the issue of conservatism. Finally, a decentralize mixed
H2 /H∞ control has been proposed in [10] for multi-DER MG A. Design AINFTSMC for Current Loop
based on multiobjective optimization (MOO). Unlike the work Uncertainties in grid filter and external disturbances are
in[15], which needs two controllers for voltage regulation and the main sources of instability and lead to a deflection
load current rejection, in [10], only one controller has been of the set-point tracking in the control layer. So, to overcome
used in the primary layer of a hierarchical control structure to the challenges of the traditional PI-based primary control mech-
enhance its performance, not only for nonlinear and unbalanced anism, a robust SMC-based structure has been proposed for
loads but also for small- and large-signal disturbances. A robust current control. The proposed primary control structure of three-
algorithm based on a mixed H2 /H∞ control strategy has been phase grid-forming power converters with an output LC filter
presented in [17] to improve power-sharing among dispatchable and output connector is shown in Fig. 1. It includes a novel
electronically coupled DERs. All of these methods have focused ABINFTSMC-based current-control loop, mixed H2 /H∞ -based

Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 06:54:31 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

RAEISPOUR et al.: ROBUST SLIDING MODE AND MIXED H2 /H∞ OUTPUT FEEDBACK PRIMARY CONTROL OF AC MICROGRIDS 3

Fig. 1. Proposed primary control structure for three-phase grid-forming power converters with the output LC filter.

voltage-control loop, and droop control loop. In this part, the error vector, we can write
ABINFTSMC strategy is considered in the inner loop of the 
primary layer to improve the robust performance and robust ėid = I˙td − I˙td,ref
(3)
stability in faced with destabilizing factors. Moreover, we ex- ėiq = I˙tq − I˙tq,ref .
pect this method gives the advantages of constant switching
frequency, low THD, robustness, and fast transient response to By substituting (1) into (3), next
the current control. To obtain the mathematical model of MG ⎧
⎨ ėid = −I˙td,ref − Rf Itd + ωItq + 1
Vtd + D1
based on Kirchhoff voltage and current laws, the state-space Lf Lf
(4)
model of the dynamic equation in the synchronous reference ⎩ ėiq = −I˙tq,ref − Rf
− ωItd + 1
Vtq + D2
Lf Itq Lf
frame (dq coordinate) can be reached as follows [3]:
⎧ where D1 and D2 are the sum of the uncertainty and disturbance,
⎨ I˙td = − Rf +ΔR Itd + ωItq + 1 1
Lf +ΔL Vtd − Lf +ΔL Vod
and we assume that theses values are time-varying and bounded.
Lf +ΔL
Theorem II.1: Consider system (4), the SMC law (5) under
⎩ I˙tq = −ωItd − Rf +ΔR Itq + 1 1
Lf +ΔL Lf +ΔL Vtq − Lf +ΔL Voq the sliding surface (2) is asymptotically stable
 1 1 ⎧  
V̇od = ωVoq + Cf +ΔC Itd − Cf +ΔC Iod q
[− ]
⎪ Itd − I˙td,ref − Lf β2 p (eid
Rf
⎪ Vtd = −Lf ωItq −
⎪
⎨ Lf
1 1 
V̇oq = −ωVod + Cf +ΔC Itq − Cf +ΔC Ioq [λ] q

⎪ +β1 eid + (Λ1 + Λ2 )SId + (1 + Λ1 Λ2 ) SId dt + udisd )[ p ]

(1) ⎪
⎪ 
⎩
⎧ udisd = − (ε + γ̂1 )sign(ṠId + Λ1 S Id
) dt
q
−[ p ]
[6pt]
where Δ denotes the uncertainty of the grid filter. Defin- ⎪
⎪
Rf ˙
Vtq = −Lf ωItd − L Itq − Itq,ref − Lf β2 (eiq
⎪
⎨
ing the current error vectors as eid = Itd − Itd,ref and
f
[λ]  q
eiq = Itq − Itq,ref where Itd,ref and Itq,ref are the reference cur- ⎪ +β e
1 iq + (Λ 1 + Λ 2 )S Iq + (1 + Λ1 Λ2 ) SIq dt + udisq )[ p ]
⎪
⎪ 
rents generated by voltage loop. By combining the fast terminal ⎩
udisq = − (ε + γ̂2 )sign(ṠIq + Λ1 SIq ) dt.
sliding mode (FTSM) and nonsingular sliding mode surface and
(5)
using the results in [33], we define the integral nonsingular fast
terminal sliding surface as Proof: Each of the proposed control laws has two parts,
⎧ namely continuous and discontinuous parts. The continuous part
 p
⎨ SId = (eid + β1 e[λ] + β2 ė[ q ] ) dt leads the system trajectories to go to the sliding surface, and
id id discontinuous part eventuate the trajectories stay on the sliding
 (2)
⎩ [λ] [p] surface. So, we consider the sliding surface (2), and compute the
SIq = (eiq + β1 eiq + β2 ėiqq ) dt
first and second derivative of the sliding surface as follows:
 [p]
where β1 and β2 are the positive constant weight coefficients to [λ]
ṠId = eid + β1 eid + β2 ėidq
adjust the convergence speed, p and q are positive odd integers [λ] [p] (6)
that p > q , 1 < pq < 2, and λ > pq . By differentiating of the S̈Id = dt
d
(eid + β1 eid + β2 ėidq ).

Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 06:54:31 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE SYSTEMS JOURNAL

Thus, the second-order state-space model of the system and define the discontinuous part as
dynamics based on the selected sliding surface can be reached
as u̇disd = −(ε + γ̂1 ) sgn(ϑ2 ) (15)

ṠId = χId we have in the next
[p] (7)
χ̇Id = dt
d [λ]
(eid + β1 eid + β2 ėidq ). V̇2 ≤ −Λ1 ϑ21 − Λ2 ϑ22 − (ε + γ̂) |ϑ2 | + |ϑ2 | γ. (16)

To find an effective control input of system (3), a backstep- Now, in the final step, we consider the Lyapunov candidate V3 =
ping design procedure is employed. The following change of V2 + 12 γ̃12 , and by time differentiating of V3 , we have
coordinates is introduced:
 V̇3 = V̇2 + γ̃ γ̃˙
ϑ1 (t) = SId
ϑ2 (t) = χId − σ1
(8) ≤ −Λ1 |ϑ1 |2 − Λ2 |ϑ2 |2 + γ̃ |ϑ2 | − γ̃ |ϑ2 | + ε |ϑ2 |
where σ1 is the virtual control. Define a positive-definite Lya- ≤ −Λ1 |ϑ1 |2 − Λ2 |ϑ2 |2 (17)
punov function V1 = 12 ϑ21 . By time differentiating of V1 , and
selecting the virtual control σ1 (t) = −Λ1 ϑ1 (t), where Λ1 > 0, which implies that ϑ1 (t) and ϑ2 (t) will converge to zero in a
we have finite time. Therefore, it can be concluded that the proposed
backstepping integral nonsingular FTSM system is asymptot-
V̇1 = ϑ1 ϑ̇1 ically stable despite the existence of uncertainties. Hence, by
substituting ϑ1 and ϑ2 based on the relation (8), the sliding
= ϑ1 (ϑ2 + σ1 )
surface and the virtual control σ1 , the control input can be
= ϑ1 (ϑ2 − Λ1 ϑ1 ). (9) reached as (5). A similar design procedure can be applied for
the q-axis of the terminal voltage, and due to lack of the space,
It is obvious that V̇1 = −Λ1 |ϑ1 |2 once ϑ2 = 0, and conse- we omitted it. 
quently, ϑ1 will be asymptotically stable.
Next, we consider the positive-definite Lyapunov candidate B. Mixed H2 /H∞ Control for Voltage Loop
V2 = V1 + 12 ϑ22 . By time differentiating of V2 , and selecting the
virtual control σ1 (t) = −Λ1 ϑ1 (t) where Λ1 > 0, we have In the outer loop of the primary layer, another robust con-
troller is designed based on mixed H2 /H∞ control. We consider
V̇2 = V̇1 + ϑ2 ϑ̇2 the voltage error vector as: [evod , evoq ]T = [Vod,ref , Voq,ref ]T −
= −Λ1 ϑ21 + ϑ1 ϑ2 + ϑ2 (χ̇Id − σ̇1 ). (10) [Vod , Voq ]T . To achieve the dynamic model of the system, with-
out loss of generality, we consider some simplifying assumption.
Hence, by selecting the control input as We suppose the load current and its derivative are bounded
 and also consider the voltage reference as a step signal with
Rf q
[− p ]
Vtd = − Lf ωItq − Itd − I˙td,ref − Lf β 2 (eid zero derivative. According to (1), we can write the augmented
Lf state-space dynamic model as follows:
[λ]
+ β1 eid + (Λ2 ϑ2 + ϑ1 )dt − σ1 + udisd )[ p ] (11)
q
ẋ = Ax + (B + ΔB)u + wd
y = Cs x (18)
[p]
where Λ2 > 0, and by assuming that | dt d
β2 D1q | ≤ γ1 and ap-
plying the proposed controller in (11) into (10), we obtain where x = [evod⎡, evoq , V̇od , V̇oq ]T⎤, u = [⎡I˙td , I˙tq ]T , d ⎤
=
0 0 −1 0 0 0
V̇2 ≤ −Λ1 ϑ21 + ϑ1 ϑ2 + ϑ2 (−Λ2 ϑ2 − ϑ1 + u̇disd ) + |ϑ2 | γ1 . [I˙od , I˙oq ]T , A = ⎣ 00 00 00 −1 ⎦
ω , B = 1/Cf
⎣0 0 ⎦,
0
(12) 0 0 −ω 0 0 1/Cf
⎡ ⎤
By selecting u̇disd = −(ε + γ1 ) sgn(ϑ2 ), where ε is a positive 0 0
⎢0 0 ⎥  
small constant, we have w = ⎣ − C +ΔC 1
0 ⎦, Cs = 10 01 00 00 , and
f
Cf
V̇2 ≤ −Λ1 |ϑ1 |2 − Λ2 |ϑ2 |2 − (ε + γ1 ) |ϑ2 | + γ1 |ϑ2 | 0
⎡
− C+ΔC
⎤
0 0
≤ −Λ1 |ϑ1 |2 − Λ2 |ϑ2 |2 . (13) ⎢0 0 ⎥
ΔB = ⎣ − C (CΔC 0 ⎦, which A⊂
f +ΔC)
− Cf (CΔC
f
Then, V̇1 and V̇2 are negative semidefinite, which implies that ϑ1 0 f +ΔC)
and ϑ2 will converge to zero in a finite time. The design proce-
4×4 , B ⊂ 4×2 , Cs ⊂ 2×4 , w ⊂ 4×2 , and ΔB ⊂ 4×2
dure is based on the assumption that the upper bound value γ1 of
[p] are the system matrix. Therefore, we can derive the output
the unknown function | dt d
β2 D1q | can be obtained in advance. feedback control law Itd and Itq of system (1) as follows:
However, it is difficult to estimate the value of the uncertainties
in advance. In order to resolve this limitation, a simple adaptive [Itd , Itq ]T = udt. (19)
law is used to estimate the upper bound of unknown uncertain
terms. Consider the estimation error γ̃1 (t) = γ1 − γ̂1 (t) where Thus, the current reference vector for the inner loop current
γ̂1 (t) is used to approximate the upper bound value γ1 . Take the control can be calculated as follows:
adaptive law as follows:
[Itd,ref , Itq,ref ]T = [Itd , Itq ]T = udt. (20)
γ̂˙ 1 = |ϑ2 | (14)

Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 06:54:31 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

RAEISPOUR et al.: ROBUST SLIDING MODE AND MIXED H2 /H∞ OUTPUT FEEDBACK PRIMARY CONTROL OF AC MICROGRIDS 5

Fig. 2. Virtual block diagram of the voltage-control loop.

The transfer function of the above-mentioned system can be

derived as

G = Cs (sI − A)−1 B. (21)

Also, the uncertainties of parameters are formulated as ad-

ditive uncertainty Δ and consequently the open-loop transfer Fig. 3. LFT representation of the augmented closed-loop system.
function can be written as G = Gnom + Wr Δ. The layout of
virtual block diagram of the proposed voltage-loop control
3) To avoid saturation of the actuators, when Δ = 0
demonstrated in Fig. 2. The specifications and objectives under

consideration include the H∞ performance and H2 performance. zu L2
The H∞ performance is convenient to enforce robustness to min sup = min Trzu ∞ = min Wu KSo ∞ .
K r∈L2 rL2
model uncertainty and to express frequency-domain specifica-
(24)
tions such as bandwidth, low-frequency gain, and roll-off. The
H2 performance has two applications. It is useful to handle 4) To achieve robust stability for additive uncertainty, when
stochastic aspects such as measurement noise and random dis- Δ = 0
turbance. In addition to the stochastic application, it is used to  
avoid saturation of desired outputs in the closed-loop system. qL2
min sup = min Tqp ∞ = min Wr KSo ∞
In the paper, we focus on this second application. Rather than Δ=0 p∈L2 pL
2
bounding the output energy, it may be desirable in some cases
(25)
to keep the peak amplitude of the output below a certain level,
e.g., to avoid actuator saturations. To sum up, the objectives of where So = (I + KG)−1 denote the output sensitivity function,
the mixed H2 /H∞ controller can be listed as follows. We , Wu , Wd , W̄d , and Wr represent the weight functions. One
1) To minimize the effect of the disturbance signal on the of the key contributions of the proposed design procedure is the
output of the system (disturbance rejection), when Δ = 0 appropriate selection of the weight functions. Often for tracking
 and low-frequency components, low-pass filter is utilized, and
zd L2 for the case of disturbance rejection, depending on the distur-
min sup = min Tdzd ∞
K d∈L2 dL2 bance frequency, appropriate weight function is used. To design
the uncertainty weight function Wr based on the frequency
= min Wd GSo Gd ∞ . (22) response of the nominal system and uncertain system, we can
reach the upper bound of uncertainty and choose the weight
Tdzd ∞ is the largest gain across frequency in the singular
function. Due to the lack of space, we avoid further explanations.
value norm. The H∞ norm measures the system input–
The interested readers can see [34] and [35] for more details.
output gain for finite energy or finite root-mean-square
Here, we select the weight function as follows, and the related
input signals and gives the system gain when both the input
singular values are depicted in Fig. 4
and the output are measured in the energy or L2 norm.
 0.25s+50 
Therefore, min Tdzd ∞ aims to minimize the effect of the 0
disturbance signal d with bounded energy on the energy We = s+0.5
0 0.25s+50
s+0.5
of the output signal zd .
 
2) To minimize the amplitude of the tracking error based on 0.9 s+1000
s+25
0
H2 norm (proper reference tracking), when Δ = 0 Wu =
0 0.9 s+1000
s+25

  
ze L∞ 0.1 s+0.005
s+0.5
0
min sup = min Trze 2 = min We So 2 . W̄d =
r∈L2 rL2 0 0.1 s+0.005
K s+0.5

(23)  394383 
2 +43.96s+394383 0
Wd = s
394383
Trze 2 measures the peak amplitude of the tracking error 0 s2 +43.96s+394383
signal ze (t) over all unit-energy reference inputs r(t).  
0.19s2 +40.54s+1540
Moreover, if the reference signal is energy bounded but not 2 +0.87s+0.02 0
a unit-energy signal, Trze 2 measures the system input– Wr = s
0.19s2 +40.54s+1540 . (26)
0 s2 +0.87s+0.02
output energy-to-amplitude gain. Therefore min Trze 2 ,
aims to minimize the peak amplitude of the tracking error To solve the above-mentioned optimization problem, linear
in response to energy bounded reference inputs r(t). fractional transformation (LFT) tools are used to shape the

Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 06:54:31 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE SYSTEMS JOURNAL

⎡    ⎤
A + BDk C BCk Bd + BDk Dd
= ⎣ Bk C Ak B k Dd ⎦
[ Cz2 + Dz2 Dk C Dz2 Ck ] [Ddz2 + Dz2 Dk Dd ]
(29)
 
 B̂3
Trz3 =
Ĉ3 D̂3
⎡    ⎤
A + BDk C BCk Br + BDk Dr
= ⎣ Bk C Ak B k Dr ⎦.
[ Cz3 + Dz3 Dk C Dz3 Ck ] [Drz3 + Dz3 Dk Dr ]
(30)

Therefore, the MOO problem can be written as

Object 1 min(γ∞ = Tdz2 ∞ )

K

Fig. 4. Singular values of the weighting functions. Object 2 min(γ2 = Trz1 2 )

K

Subject to Trz3 ∞ < 1. (31)

closed-loop characteristics of the system. Fig. 3 displays the Lemma 1 (C. Scherer-H2 Performance, see [36]): The
LFT scheme of the proposed system. The following transfer closed-loop H2 norm of Trz1 is less than γ2 if there exists
function can describe the open-loop system shown in Fig. 3: positive symmetric matrices P2 and Q such that the following
⎡ ⎤ LMIs are feasible:
⎡ ⎤ We −We GW̄d −We G
z1   ⎧ 
⎢0 Wd GW̄d Wd G ⎥ r ÂP2 + P2 ÂT P2 B̂1
⎢ z2 ⎥ ⎢    ⎥ ⎪
⎪
⎨ <0
⎣z ⎦ = ⎢ 0 Wr W̄d Wr ⎥ d . T
−I
3 ⎣ 0 0 Wu ⎦ u  B̂1 P2T  (32)
y1 ⎪
⎪ P2 Ĉ1
I −GW̄d −G ⎩ > 0, T r(Q) < γ2 , D̂1 = 0.
Ĉ1 Q
Using the state-space realization of G, W̄d , We , Wd , Wu , Wr ,
the open-loop state-space model P , and controller K = Lemma 2 (C. Scherer-H∞ Performance, see [36]): The
A B closed-loop H∞ norm of Tdz2 is less than γ∞ , if there exists
[ k k ] can be obtained as follows: positive symmetric matrix P∞ such that the following LMI is
Ck Dk
feasible:
⎡ ⎤ ⎡ ⎤
A Br Bd B
⎢ Cz1 Drz1 Ddz1 Dz1 ⎥ ÂP∞ + P∞ ÂT P∞ B̂2 Ĉ2T
⎢ ⎥ ⎣ −γ∞ I D̂2T ⎦ < 0. (33)
P : ⎢ Cz2 Drz2 Ddz2 Dz2 ⎥
⎣C D ⎦ ∗ −γ∞ I
z3 rz3 Ddz3 Dz3
C Dr Dd 0
 Based on the above-mentioned lemmas, the MOO problem is
K : ψ̇ = Ak ψ + Bk y1 (27) converted to a convex optimization that leads to a LMI problem.
u = Ck ψ + Dk y1 . Motivated by [36], we consider P = P∞ = P2 as follows, where
P ≥ 0 and
Considering [x, ψ]T as the states vector, [r, d]T as the exogenous    
input, and [z1 , z2 , z3 ]T as the output signals related to the per- Y N −1 X M
P = P = . (34)
formance of the control system, the closed-loop state-space set NT ∗ MT ∗
can be reached. Hence, the state-space realization of Trz1 , Tdz2 ,
and Trz3 are derived as follows: We select the following change of variable:
  ⎧
 B̂1
Trz1 = ⎪
⎪ Ã = N Ak M T + N Bk CX + Y BCk M T
Ĉ1 D̂1 ⎪
⎪
⎪
⎪ +Y (A + BDk C)X
⎡    ⎤ ⎨
A + BDk C BCk Br + BDk Dr (35)
= ⎣ Bk C Ak B k Dr ⎦ ⎪
⎪ B̃ = N Bk + Y BDk
⎪
⎪
[ z1
C + D D C D z1 k ] [Drz1 + Dz1 Dk Dr ]
C ⎪
⎪ C̃ = Ck M T + Dk CX
z1 k ⎩
(28) D̃ = Dk .
 
 B̂2 Consequently, the MOO problem (31) is converted to an LMI
Tdz2 =
Ĉ2 D̂2 set shown at the bottom of the next page. After solving, the

Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 06:54:31 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

RAEISPOUR et al.: ROBUST SLIDING MODE AND MIXED H2 /H∞ OUTPUT FEEDBACK PRIMARY CONTROL OF AC MICROGRIDS 7

controller parameters can be derived as all possible values of the objective functions. In this article,
⎧ an FDMT is used to find an optimal solution wherein a linear
⎪
⎪ Ak = N −1 (Ã − N Bk CX − Y BCk M T membership function (μi ) is defined for each objective function
⎪
⎪
⎪
⎪ −Y (A + BDk C)X)M −T as follows:
⎨
⎧
⎪ Bk = N −1 (B̃ − Y BDk )
(36) ⎨1 Fin ≤ min(Fi )
⎪
⎪ max(F )−F n
⎪
⎪ C = (C̃ − Dk CX)M −T μni = max(Fi )−min(F
i i
i)
min(Fi ) ≤ Fin ≤ max(Fi )
⎪
⎩ k ⎩
Dk = D̃.
i=1,...,j
0 Fin ≥ max(Fi ).
(38)
The MOO problem, unlike the single-objective optimization,
has a different definition for the optimum point. Thus, in the For the objective functions that should be maximized, member-
conventional mixed H2 /H∞ problem, selecting the appropriate ship function (μi ) is defined as [10]
value between objectives is the critical point. To choose the best ⎧
tradeoff between objectives, we utilize the FDMT described in ⎨1 n Fin ≤ min(Fi )
Fi −min(Fi )
the following section. μin
= max(Fi )−min(Fi ) min(Fi ) ≤ Fin ≤ max(Fi )
i=j+1,...,nobj ⎩
0 Fin ≥ max(Fi ).
C. Multiobjective Optimization (39)
MOO is a specific filed of multiple-criteria/multiattribute The membership functions are used to evaluate the optimal de-
decision making (concerned with mathematical optimization gree of the Pareto optimal solutions. The most preferred solution
problems involving more than one objective to be optimized can be selected as
simultaneously). A general minimization problem including d  p 
n
variables and M objective functions can be stated mathemati- i=1 ωi .μi
μopt,I = max sup M p (40)
cally as x = [x1 , x2 , . . ., xd ]T n
n=1 i=1 ωi .μi
Minimize
 : f (x) = [f1 (x), f2 (x), . . ., fM (x)]T 
p
gj (x) ≤ 0, i = 1, 2, . . . , J (37) ωi ≥ 0 ωi = 1
s.t :
hk (x) = 0, i = 1, 2, . . ., K i=1

where d is the dimension of the decision variable space and and ωi is the weight value associated with the ith objective
fi (x) is the ith objective function, gj (x) is the jth inequality function, and M is the number of Pareto optimal solutions.
constraint, and hk (x) is the kth equality constraint [10], [37]. The lower values of the membership function indicate more
The constraints given in (39) define feasible region Ω and any degrees of achievement of the objective function. The solution
point x in Ω defines a feasible solution. The vector f (x) is with the minimum membership function can be considered as the
a function that maps the set Ω into the set Λ and represents best compromise solution. Another criterion to choose the best

min γ∞
min γ2
⎡ T ⎤
AX + XAT + B C̃ + (B C̃) ∗ ∗ ∗
⎢ T T ⎥
⎢ Ã + (A + B D̃C) A Y + Y A + B̃C + (B̃C)
T
∗ ∗ ⎥
⎢ T  T ⎥<0
⎢ ⎥
⎣ Bd + B D̃Dd Y Bd + B̃Dd −γ∞ I ∗ ⎦
Cz2 X + Dz2 C̃ Cz2 + Dz2 D̃C Ddz2 + Dz2 D̃Dd −γ∞ I
⎡ T ⎤
AX + XAT + B C̃ + (B C̃) ∗ ∗ ∗
⎢ T T ⎥
⎢ Ã + (A + B D̃C) A Y + Y A + B̃C + (B̃C)
T
∗ ∗ ⎥
⎢ T  T ⎥<0
⎢ ⎥
⎣ Br + B D̃Dr Y Br + B̃Dr −I ∗ ⎦
Cz3 X + Dz3 C̃ Cz3 + Dz3 D̃C Drz3 + Dz3 D̃Dr −I
⎡ T ⎤
AX + XAT + B C̃ + (B C̃) ∗ ∗
⎢ T T ⎥
⎢ Ã + (A + B D̃C) AT Y + Y A + B̃C + (B̃C) ∗ ⎥<0
⎣ T  T ⎦
Br + B D̃Dr Y Br + B̃Dr −I
⎡ ⎤
X ∗ ∗
⎣I Y ∗ ⎦ > 0, T r(Q) < γ2 , Drz1 + Dz1 D̃Dr = 0.
Cz1 X + Dz1 C̃ Cz1 + Dz1 D̃C Q

Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 06:54:31 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE SYSTEMS JOURNAL

compromise solution has been proposed in [10]. The distance of

each nondominated solution (k) from the origin in the fuzzified
coordinate is calculated as follows:
M 1/2

n 2
μopt,II =
k
(μi ) . (41)
i=1

D. Droop Control
Droop control is realized by simulating the droop charac-
teristic of generators in a traditional grid and controlling the
output voltage and frequency of the voltage-source converter
(VSC) according to variation of the output power. The in-
verter parallel operations have employed the droop control
idea because of the power system inertia, and the frequency
decreases with an increase in load. In the MG operation, RERs
are directly connected through the power-electronic converters.
Therefore, the system inertia has been neglected. Introducing
a droop control technique will imitate the inertia of VSC as
within the synchronous type of wind turbine generator system
and also the manner of a synchronous generator, which di-
minishes the frequency when the active power increases [38].
So, the contribution of the primary layer in controlling the
voltage and frequency is achieved through the following droop
equations:
⎧
⎨ Vodi = Vinom − nQ i Qi Fig. 5. Proposed robust hierarchical control structure.
Voqi = 0 (42)
⎩
ωi = ωinom − mP P
i i .
However, the power-sharing is affected by the line impedance.
To deal with this problem, a virtual output-impedance loop is
always needed, but we do not discuss in this article and just
focus on active power control.

III. DISTRIBUTED SECONDARY CONTROL

The proposed hierarchical control structure, including a ro-
bust primary and cooperative consensus-based, distributed sec-
ondary controller, is illustrated in Fig. 5. The distributed sec-
ondary control has been wholly discussed in [39]. In this article,
we imitate this method instead of the decentralized approach
that has many limitations because of practical operation. Mo-
tivated by [39], we consider N DGs that communicate with
each other through the prescribed sponsored links, the protocol
Fig. 6. Single-line diagram of the test MG system.
(43) asymptotically solves the consensus problem for voltage,
frequency, active powers, and reactive power, respectively,

uvi = cv aij (Vodj − Vodi ) + bi (Vref − Vodi ) IV. SIMULATION RESULTS
j∈Ni To evaluate the effectiveness of the proposed method, the
 multi-DG MG is simulated in MATLAB/SimPowerSystem as
i = cω
uω aij (ωj − ωi ) + bi (ωref − ωi ) shown in Fig. 6, by implementing detailed power converter
j∈Ni switching modules with a switching frequency of 2.5 kHz. MG
 control system consists of the primary layer, the power sharing
i = cP
uP aij (mP j Pj − mPi Pi ) controller, and the agent-based secondary layer. The parameters
j∈Ni are presented in Table I.
 Assume that the maximum perturbation of the grid fil-
uQ ter parameters ΔR, ΔL, ΔC are ±30%. The inner-loop con-
i = cQ aij (nQj Qj − nQi Qi ) (43)
j∈Ni
troller parameters are adjusted as β1 = 10, β2 = 5, λ = 1.5,
p = 9, q = 7, Λ1 = 10, and Λ2 = 5. By solving the LMI ob-
for i = 1, . . ., N and aij represent the interaction connection tained for voltage control, the mixed H2 /H∞ controller param-
between ith and jth DG. eters are easily obtained, which we avoided writing due to the

Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 06:54:31 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

RAEISPOUR et al.: ROBUST SLIDING MODE AND MIXED H2 /H∞ OUTPUT FEEDBACK PRIMARY CONTROL OF AC MICROGRIDS 9

TABLE I
PARAMETERS OF THE MG SYSTEM

lack of space in the article. Moreover, the secondary control

parameter are chosen as cv = cω = cP = cQ = 5. Based on
the results of the previously published works [10], [14], [16],
[17], [22], such a detailed simulation mechanism is very close
to the actual measured experimental results, which makes the
detailed simulation strategy appropriate as a replacement of
the corresponding hardware experiment especially when the
experimental hardware system is very expensive or impossible
to build. We have not seen such a detailed study of integrating
primary- and secondary-level controls using a detailed power
converter switching model, particularly with the agent-based
control configuration. The simulations were run on a laptop
with Intel Core i7-5500 U CPU at 2.40 GHz and 8.00-GB
memory. The following scenarios have been considered to show
the effectiveness of the proposed method.

A. Scenario 1: Step Load Change

In the first study, the MG works in normal conditions with
the proposed robust inner and outer controller. At t = 0.9s, the
heavy load is added to DG#2 and disconnected at t = 1.2 s,
as can be observed in Fig. 7, DG units well cover the load
increment. Fig. 7(a) and (b) shows the three-phase output voltage
and terminal current of DG#2, which have smooth behavior
to these changes. Due to increasing the load of DG#2, the
generated power of DG #2 is increased, consequently owing to
the natural behavior of the droop control, the frequency of DGs
is decreased. Similar changes can be observed in the voltage
and reactive power. The dynamic responses of the whole system
illustrated in Fig. 7(c)–(f), which shows the robustness and
stability of the system and fast response to return to normal
condition. Fig. 7. Performance evaluation of MG in step load change. (a) Three-phase
voltage of DG#2. (b) Three-phase current of DG#2. (c) Active powers of DGs.
B. Scenario 2: Three-Phase Short-Circuit Fault (d) Reactive powers of DGs. (e) Voltage of DGs. (f) Frequency of DGs.

In the next scenario, a three-phase ground fault occurs in the

middle of the line#1 (see Fig. 6). The fault happens at t = 1 s, to return to normal condition after the faults. It is observed
and is cleared at t = 1.2 s. During the time interval t = 1 − 1.2 s, that the proposed method leads to having higher performance
the controllable components of MG greatly has been effected as opposed to the hierarchical PI-based scheme of [3]. Note that
by three-phase ground fault and the MG has became unstable, fault-tolerant control is not the contribution of this article and
voltage and frequency have been deviated from nominal setpoint will be the subject of the future research of the authors. As shown
and power-sharing is missed. It can be observed that after fault, in Fig. 8, the smooth transition during fault clearing is achieved,
the voltage becomes stable, and power-sharing maintain without which proves the fault ride-through capability of the proposed
any problem. The inner robust controller, lead to fast response strategy.

Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 06:54:31 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE SYSTEMS JOURNAL

Fig. 8. Performance evaluation of MG in three-phase short circuit fault. (a) Fig. 9. Performance evaluation of MG for P&amp;P functionality. (a) Three-
Three-phase voltage of DG#1. (b) Three-phase current of DG#1. (c) Active phase voltage of DG#4. (b) Three-phase current of DG#4. (c) Active powers
powers of DGs. (d) Reactive powers of DGs. (e) Voltage of DGs. (f) Frequency of DGs. (d) Reactive powers of DGs. (e) Voltage of DGs. (f) Frequency of DGs.
of DGs.

have been successfully restored even after the disconnection of

one DG from MG. Due to its own local primary control, DG#4’s
C. Scenario 3: Plug and Play (P&P) Functionality
frequency and voltage are also finally restored to the reference
In this case, the (P&P) capability of the robust proposed values quickly. Fig. 9(c) and (d) shows active and reactive
method is investigated. At t = 1.4 s, DG#4 is disconnected from powers of each DG unit, after disconnection of DG#4 from
MG and reconnected to the MG at t = 1.8 s, and results illus- the MG, other DGs feed the entire load by injecting appropriate
trated in Fig. 9. As seen in the results, the frequency and voltage active powers relating to their output ratios. Consequently, by

Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 06:54:31 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

RAEISPOUR et al.: ROBUST SLIDING MODE AND MIXED H2 /H∞ OUTPUT FEEDBACK PRIMARY CONTROL OF AC MICROGRIDS 11

Fig. 10. Performance evaluation of MG with nonlinear load; three-phase Fig. 11. Performance evaluation of MG with nonlinear load; three-phase
voltage of DG#2. (a) Proposed method. (b) Robust H∞ method presented current of DG#2. (a) Proposed method. (b) Robust H∞ method presented
in [16]. (c) Cascaded PI-based method presented in [3]. (d) Instantaneous THDs in [16]. (c) Cascaded PI-based method presented in [3]. (d) Instantaneous THDs
for voltage of DG#2. for current of DG#2.

The THD value of voltage and current is less than 5%, which
replugging the DG#4 at t = 1.8 s, other DGs return to normal
is acceptable according to the IEEE Standard 519 [40], but the
conditions and share the power associated with the droop factors.
proposed ABINFTSMC controller in the current loop leads to
Finally, one can say, in addition to providing robust stability, this
have less THD, higher robustness, and faster transient response.
approach realizes the (P&P) functionality of the MG.

D. Scenario 4: Inclusion of Nonlinear Load V. CONCLUSION

In this part of the analysis, the performance of the MG system In this article, a unified, robust hierarchical control was pro-
concerning the inclusion of a highly nonlinear load is inves- posed for a multi-DG islanded MG based on ABINFTSMC
tigated. At t = 0.5 s, the nominal load#2 is suddenly discon- and mixed H2 /H∞ control strategies for current and voltage
nected and a three-phase six-pulse diode-bridge rectifier is con- controllers of the primary layer, respectively. The theoretical
nected to the DG#2. The rectifier output terminals are connected concepts, including mathematical modeling of MG, basic lem-
to an RL load whose parameters are P = 10kW and QL = mas, and design procedure of both ABINFTSMC and mixed
100var. Moreover, to show the superiority of the proposed H2 /H∞ controllers, was presented. Finally, the performance
method over other reported techniques, we compare the of the controller was evaluated for small- and large-signal dis-
proposed control strategy with the robust H∞ method presented turbances and nonlinear and unbalanced load based on offline
in [16] and conventional cascaded PI-based control in [3], results time-domain simulations in the MATLAB/Simulink environ-
are demonstrated in Fig. 10 for voltage and Fig. 11 for current, ment and comparison with other reported control techniques. As
respectively. As can be seen, the proposed method except for was observed and indicated by the results, by using the proposed
some transients, successfully regulates the output and has a control scheme, the MG maintained its stability when faced with
good robust performance in comparison with other approaches. small- and large-signal disturbances.

Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 06:54:31 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

12 IEEE SYSTEMS JOURNAL

REFERENCES [20] U. K. Kalla, B. Singh, S. S. Murthy, C. Jain, and K. Kant, “Adaptive sliding
mode control of standalone single-phase microgrid using hydro, wind, and
[1] J. M. Guerrero, J. C. Vasquez, J. Matas, L. G. de Vicuna, and M. Castilla, solar PV array-based generation,” IEEE Trans. Smart Grid, vol. 9, no. 6,
“Hierarchical control of droop-controlled AC and DC microgrids—A pp. 6806–6814, Nov. 2018.
general approach toward standardization,” IEEE Trans. Ind. Electron., [21] D. Xu, Y. Dai, C. Yang, and X. Yan, “Adaptive fuzzy sliding mode
vol. 58, no. 1, pp. 158–172, Jan. 2011. command-filtered backstepping control for islanded PV microgrid with
[2] J. M. Guerrero, M. Chandorkar, T.-L. Lee, and P. C. Loh, “Advanced energy storage system,” J. Franklin Inst., vol. 356, no. 4, pp. 1880–1898,
control architectures for intelligent microgrids—Part I: Decentralized 2019.
and hierarchical control,” IEEE Trans. Ind. Electron., vol. 60, no. 4, [22] M. B. Delghavi, S. Shoja-Majidabad, and A. Yazdani, “Fractional-order
pp. 1254–1262, Apr. 2013. sliding-mode control of islanded distributed energy resource systems,”
[3] S. D’Arco, J. A. Suul, and O. B. Fosso, “Automatic tuning of cascaded IEEE Trans. Sustain. Energy, vol. 7, no. 4, pp. 1482–1491, Oct. 2016.
controllers for power converters using eigenvalue parametric sensitivities,” [23] A. Bidram and A. Davoudi, “Hierarchical structure of micro-
IEEE Trans. Ind. Appl., vol. 51, no. 2, pp. 1743–1753, Mar./Apr. 2014. grids control system,” IEEE Trans. Smart Grid, vol. 3, no. 4,
[4] J. He and Y. W. Li, “Generalized closed-loop control schemes with pp. 1963–1976, Dec. 2012.
embedded virtual impedances for voltage source converters with LC or [24] J. Choi, Z. Hakimi, J. Sampson, and V. Narayanan, “Byzantine-tolerant
LCL filters,” IEEE Trans. Power Electron., vol. 27, no. 4, pp. 1850–1861, inference in distributed deep intelligent system: Challenges and opportuni-
Apr. 2012. ties,” IEEE J. Emerg. Sel. Topics Circuits Syst., vol. 9, no. 3, pp. 509–519,
[5] J. Rodriguez et al., “Predictive current control of a voltage source inverter,” Sep. 2019.
IEEE Trans. Ind. Electron., vol. 54, no. 1, pp. 495–503, Feb. 2007. [25] H. Atrianfar, “Sampled-time containment control of high-order
[6] V. Yaramasu, M. Rivera, M. Narimani, B. Wu, and J. Rodriguez, “Model continuous-time mass under heterogenuous time-varying delays and
predictive approach for a simple and effective load voltage control of four- switching topologies: A scrambling matrix approach,” Neurocomputing,
leg inverter with an output LC filter,” IEEE Trans. Ind. Electron., vol. 61, vol. 395, pp. 24–38, 2020.
no. 10, pp. 5259–5270, Oct. 2014. [26] M. Raeispour, H. Atrianfar, H. R. Baghaee, and G. B. Gharehpetian,
[7] A. Bouafia, F. Krim, and J.-P. Gaubert, “Fuzzy-logic-based switching state “Distributed LMI-based control of heterogeneous microgrids considering
selection for direct power control of three-phase PWM rectifier,” IEEE fixed time-delays and switching topologies,” IET Ren. Power Gener.,
Trans. Ind. Electron., vol. 56, no. 6, pp. 1984–1992, Jun. 2009. pp. 1–11, Mar. 2020, doi: 10.1049/iet-rpg.2019.1113.
[8] X. Fu and S. Li, “Control of single-phase grid-connected converters [27] N. M. Dehkordi, H. R. Baghaee, N. Sadati, and J. M. Guerrero, “Dis-
with LCL filters using recurrent neural network and conventional control tributed noise-resilient secondary voltage and frequency control for is-
methods,” IEEE Trans. Power Electron., vol. 31, no. 7, pp. 5354–5364, landed microgrids,” IEEE Trans. Smart Grid, vol. 10, no. 4, pp. 3780–3790,
Jul. 2016. Jul. 2019.
[9] J. Jung, V. Leu, D. Dang, T. Do, F. Mwasilu, and H. Choi, “Intelligent [28] A. Afshari, M. Karrari, H. R. Baghaee, and G. B. Gharehpetian, “Resilient
voltage control strategy for three-phase ups inverters with output LC filter,” cooperative control of AC microgrids considering relative state-dependent
Int. J. Electron., vol. 102, no. 8, pp. 1267–1288, 2015. noises and Communication time-delays,” IET Ren. Power Gener., vol. 14,
[10] H. R. Baghaee, M. Mirsalim, G. B. Gharehpetian, and H. A. Talebi, no. 2, pp. 1321–1331, Jun. 2020, doi: 10.1049/iet-rpg.2019.1180.
“A decentralized robust mixed H2 /H∞ voltage control scheme to im- [29] A. Afshari, M. Karrari, H. R. Baghaee, G. B. Gharehpetian, and S. Karrari,
prove small/large-signal stability and FRT capability of islanded multi-der “Cooperative fault-tolerant control of microgrids under switching commu-
microgrid considering load disturbances,” IEEE Syst. J., vol. 12, no. 3, nication topology,” IEEE Trans. Smart Grid, vol. 11, no. 3, pp. 1866–1879,
pp. 2610–2621, Sep. 2018. May 2020.
[11] H. Karimi, E. J. Davison, and R. Iravani, “Multivariable servomechanism [30] A. Afshari, M. Karrari, H. R. Baghaee, and G. B. Gharehpetian, “Dis-
controller for autonomous operation of a distributed generation unit: tributed fault-tolerant voltage/frequency synchronization in autonomous
Design and performance evaluation,” IEEE Trans. Power Syst., vol. 25, AC microgrids,” IEEE Trans. Power Syst., to be published, doi:
no. 2, pp. 853–865, May 2009. 10.1109/TPWRS.2020.2975115.
[12] A. Fathi, Q. Shafiee, and H. Bevrani, “Robust frequency control of mi- [31] M. Raeispour, H. Atrianfar, H. R. Baghaee, and G. B. Gharehpetian,
crogrids using an extended virtual synchronous generator,” IEEE Trans. “Resilient H∞ consensus-based control of autonomous AC microgrids
Power Syst., vol. 33, no. 6, pp. 6289–6297, Nov. 2018. with uncertain time-delayed communications,” IEEE Trans. Smart Grid,
[13] H. R. Baghaee, M. Mirsalim, G. B. Gharehpetian, and H. A. Talebi, “A gen- to be published, doi: 10.1109/TSG.2020.2992646.
eralized descriptor-system robust H∞ control of autonomous microgrids [32] A. Pilloni, A. Pisano, and E. Usai, “Robust finite-time frequency and volt-
to improve small and large signal stability considering communication age restoration of inverter-based microgrids via sliding-mode cooperative
delays and load nonlinearities,” Int. J. Elect. Power Energy Syst., vol. 92, control,” IEEE Trans. Ind. Electron., vol. 65, no. 1, pp. 907–917, Jan. 2018.
pp. 63–82, 2017. [33] M. Van, M. Mavrovouniotis, and S. S. Ge, “An adaptive backstepping
[14] T. Hornik and Q.-C. Zhong, “A current-control strategy for voltage- nonsingular fast terminal sliding mode control for robust fault tolerant
source inverters in microgrids based on H2 /H∞ and repetitive con- control of robot manipulators,” IEEE Trans. Syst., Man, Cybern.: Syst.,
trol,” IEEE Trans. Power Electron., vol. 26, no. 3, pp. 943–952, vol. 49, no. 7, pp. 1448–1458, Jul. 2019.
Mar. 2011. [34] S. Skogestad and I. Postlethwaite, Multivariable Feedback Control: Anal-
[15] M. Babazadeh and H. Karimi, “A robust two-degree-of-freedom control ysis and Design, vol. 2. New York, NY, USA: Wiley, 2007.
strategy for an islanded microgrid,” IEEE Trans. Power Del., vol. 28, no. 3, [35] R. Beaven, M. Wright, and D. Seaward, “Weighting function selection in
pp. 1339–1347, Jul. 2013. the h∞ design process,” Control Eng. Pract., vol. 4, no. 5, pp. 625–633,
[16] M. Hamzeh, S. Emamian, H. Karimi, and J. Mahseredjian, “Robust control 1996.
of an islanded microgrid under unbalanced and nonlinear load conditions,” [36] C. Scherer, P. Gahinet, and M. Chilali, “Multiobjective output-feedback
IEEE J. Emerg. Sel. Topics Power Electron., vol. 4, no. 2, pp. 512–520, control via LMI optimization,” IEEE Trans. Autom. Control, vol. 42, no. 7,
Jun. 2016. pp. 896–911, Jul. 1997.
[17] S. Gholami, S. Saha, and M. Aldeen, “Robust multiobjective control [37] P. Ngatchou, A. Zarei, and A. El-Sharkawi, “Pareto multi objective opti-
method for power sharing among distributed energy resources in islanded mization,” in Proc. 13th Int. Conf., Intell. Syst. App. Power Syst., 2005,
microgrids with unbalanced and nonlinear loads,” Int. J. Elect. Power pp. 84–91.
Energy Syst., vol. 94, pp. 321–338, 2018. [38] S. de Haan and K. Visscher, “Virtual synchronous machines for frequency
[18] M. Cucuzzella, G. P. Incremona, and A. Ferrara, “Decentralized sliding stabilisation in future grids with a significant share of decentralized gen-
mode control of islanded AC microgrids with arbitrary topology,” IEEE eration,” in Proc. CIRED Seminar, 2008, pp. 23–24.
Trans. Ind. Electron., vol. 64, no. 8, pp. 6706–6713, Aug. 2017. [39] A. Bidram, A. Davoudi, F. L. Lewis, and Z. Qu, “Secondary control
[19] H. R. Baghaee, M. Mirsalim, G. B. Gharehpetian, and H. A. Talebi, “A of microgrids based on distributed cooperative control of multi-agent
decentralized power management and sliding mode control strategy for systems,” IET Gener., Transmiss. Distrib., vol. 7, no. 8, pp. 822–831, 2013.
hybrid AC/DC microgrids including renewable energy resources,” IEEE [40] IEEE Recommended Practices and Requirements for Harmonic Control
Trans. Ind. Informat., to be published. in Electrical Power Systems, IEEE Std 519-2014, 1993.

Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 06:54:31 UTC from IEEE Xplore. Restrictions apply.
View publication stats