This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2843527, IEEE
Transactions on Smart Grid

Reliability Analysis of a Decentralized Microgrid

Control Architecture
Abedalsalam Bani-Ahmed IEEE Member, Mohammad Rashidi, Adel Nasiri IEEE SeniorMember, and
Hossein Hosseini, IEEE SeniorMember


Abstract— Reliability enhancement of microgrids is challenged
by environmental and operational failures. Centrally controlled
microgrids are susceptible to failures at high probability due to a
single-point-of-failure, e.g. the central controller. True
decentralization of microgrid architecture entails elimination of
the central controller, attaining a parallel configuration for the
system. In this paper, decentralized microgrid control
architecture is proposed as a solution for reliability degradation
over the time, and analyzes the reliability aspects of centralized
and decentralized control architectures for microgrids. Degree of
importance of a single controller in centralized and decentralized
architectures is determined and validated by Markov Chain
Models (MCM). Results confirm that higher reliability is
achieved when true decentralization of control architecture is
adopted. Challenges of implementing a true decentralized control
architecture are discussed. Hardware-In-the-Loop simulation
results for microgrid controller failure scenarios for both Figure 1. Residential Microgrid location on the power grid map.
architectures are presented and discussed.
for renewables. Microgrids also support market participation
Index Terms—Decentralized control, distributed control, HIL,
Markov Chain, microgrid, reliability, smart grid.
of smaller power sources that can be aggregated to provide
power necessary to meet the target goals for Distributed
I. INTRODUCTION Energy Resources (DER) [9]. As the electricity grid continues
to modernize, DER such as storage and advanced renewable
T he usage of the terms smart grid and microgrid are
growing immensely, feeding from the multi-disciplinary
pool of research and visions for the electrical grid. The basic
technologies can help facilitate the transition to a smarter grid
islanding capabilities. Microgrids also support management of
critical and non-critical loads for available generation. Other
concept of Smart Grid is to add monitoring and microgrid requirements involve secure operations, deploying
communication to existing partially traditional grid [1]. It also cyber secure communications network that guarantee
adds control in a manner that moves this traditional grid into a distributed and resilient supervisory control architecture.
two-way power and information flow entity. This will allow Studying microgrid reliability is a challenge due to the
new forms of generations and energy storage to connect to the variety of power sources that can be included [22]. Generally,
system and participate in many mainstream and ancillary the evaluation of microgrid reliability must consider the load
functions of the grid. demand, which influences the microgrid architecture at the
Acting as one of the drives of Smart Grid advancement, design stage [2]. Other aspects such as protection schemes are
microgrids are a localized grouping of electricity sources and considered as a microgrid reliability enhancement mechanism
loads that normally operate connected to and synchronous [3] [4]. The advancement of power electronics research efforts
with the traditional centralized grid (macrogrid), but can and control strategies for microgrid inverters, and hybrid AC-
disconnect and function autonomously as physical and/or DC microgrids had invigorated power systems researchers in
economic conditions dictate [2]. Figure 1 shows a generic general to adopt state of the art technologies in designing
illustration of the microgrid concept within the electric power reliable microgrid systems [5]. Communication-assisted
grid map. The distribution network in the power grid supports control techniques drove the improvement of microgrid
residential and industrial areas providing utility services where reliability arising cyber-security concerns [13].
microgrids are deployed in order to support local power The focus of this paper is microgrid reliability enhancement
demand and respond to ancillary services requests. and analysis by decentralizing the control architecture.
Typical microgrid requirements involve grid connection Although the scope of the paper does not involve a specific
capabilities, and optimization of economic operation, and control mechanism, section II reviews state of the art
support integration of high penetration and energy harvesting microgrid control strategies including decentralized controls.
Microgrid reliability analyses are discussed and conducted

1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2843527, IEEE
Transactions on Smart Grid

providing a quantitative evidence of the reliability microgrid operation [18]. The decentralized control is a
improvement in decentralized microgrids as opposed to the variant on the master/ slave control. A local decentralized
centrally controlled microgrids in Section III. Section VI controller controls the reference voltage and influences the
discusses experimental results of a microgrid during normal output current of the units. The voltage magnitude, frequency
operations and the impact of controller failure in centralized and power sharing are locally controlled. Hence, in
and decentralized control concepts. decentralized control, only low bandwidth communication is
required to exchange local updates with the peer controllers
Operator
Logging
[12].
HMI MG Decentralized Controller
PCC Client Server Microgrid control hierarchy as illustrated in Fig. 3, has three
HIGH SPEED COMMUNICATION INFRASTRUCTURE
levels of controls where each level satisfies certain
Utility Grid requirements and roles maintaining power reliability, quality,
and economical concerns [12]. Primary control level
comprises two control loops for voltage and current, playing
an essential role in stabilizing the voltage and frequency. The
Backup
Generators Energy
voltage/angle or active/reactive power commands are provided
Renewables Renwewables
Storage EV Charging
from the secondary control and voltage and current references
Figure 2. Decentralized Microgrid control system architecture with
communication infrastructure. are generated and provided to the source.
Although the primary control level is responsible of
frequency regulation, some deviation may occur due to the
II. MICROGRID CONTROL STRATEGIES load power demand fluctuations or intermittency of
Control strategies for microgrids have been developed for renewables. In microgrid systems, energy storage enables the
the past decade targeting the seamless transients and microgrid microgrid to compensate for frequency deviations for short
operations. Control of microgrids is generally more complex terms, the role of secondary control comes at a slower
than traditional power systems due to limited energy storage response to frequency fluctuations in comparison the primary
capacity and lack of inertia, fast dynamics and short response control ensuring power quality. The control system
time of inverter-based distributed resources, and a high degree continuously monitors the microgrid frequency and voltage in
of parametric and topological uncertainties [11]. These issues real time, and dispatches microgrid components such as
transform into more complex challenges when the system energy storage or backup generators for active and reactive
reliability is at risk, and a robust control architecture becomes power and updates the primary levels at each power source
essential as smart grid functionality is enabled [1]. That with the appropriate power command [15].
includes the intelligent interconnection and integration of Local frequency control [13] and voltage regulation [14] at
DERs, demand response, and consequently achieving net the primary control level are the major drives for decentralized
metering. controls of microgrids. Other controlled variables including
Control methods can be categorized based on the control active and reactive power are managed by the Energy
Management System (EMS) at the secondary and tertiary
architecture design, communication infrastructure, load
control levels [15]. The variation of decentralized primary
sharing strategy. Centralized and distributed (decentralized)
control techniques for different microgrid components as grid-
control methods differ in many aspects, the major aspect forming and grid-feeding sources relies on the behavior of the
involves data concentration power sharing command sources. component and the controllability of microgrid variable at the
Generally, if the DGs are capable of generating commands source terminal. As microgrid topologies varies, the control
locally, regardless of the presence of communication with methods consider inverter-based power sources [17] only, or a
other DGs, this is considered to be a distributed control.
Centralized microgrid control consists of the central
controller that monitors and controls all DER units and local
loads [16]. Decentralized microgrid control is virtually
centralized, that is, the central controller is physically
distributed over a decentralized infrastructure. As shown in
Fig. 2, the architecture suggests deploying multiple local
controllers to achieve seamless transients during the operation
control scheme. Resource sharing must be guaranteed, where
every controller shares the status of its own DER with peer
controllers in real-time manner. This requires a naming
scheme that guarantees unique identification of each controller
and its local DER. Each controller must have a general
awareness of the whole system status, especially for the inputs
to the microgrid control algorithm running in each controller.
This is a key requirement to protect the integrity of the system, Figure 3. A typical microgrid control hierarchy.
otherwise, inconsistent algorithm outputs and control
commands may arise, which can lead to disturbances in combination of AC and DC sources [9] [18]. Methods have

1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2843527, IEEE
Transactions on Smart Grid

been proposed enabling real-time management of microgrids A. Reliability Measurements

involving energy storage units over a decentralized secondary 1) Degree of Importance
control architecture [12], [16]. In this paper, the architecture The importance of a component indicates the impact of the
under investigation applies the Virtual Droop Control method component’s failure on the system failure [19]. In the early
[21]. stages of system development, the components life
A decentralized secondary level coordination is essential distribution or reliabilities are assumed to be equal.
when parallelization of DERs in microgrids is achieved and A system with n components, considering a state 𝑥𝑖 of
the purpose of parallelization is maintained. Feedback to the component i is defined by:
control loop can be local measurements of the power source,
1 if 𝑖 is functioning
or status updates from other components over the 𝑥𝑖 = { 𝑖 ∈ (1, 𝑛) (1)
0 if 𝑖 is 𝐧𝐨𝐭 functioning
communication infrastructure as shown in Fig. 2, or both.
Although the system is connected at the network level, the A deterministic binary function 𝜑 of the system state, with x
scalability is possible when we overcome the challenge of as the function vector input 𝑥 = (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) is defined as
dynamic configuration of the control algorithm, and true follows:
decentralization is applied to the control architecture. The 1 if the system is functioning
𝜑(𝑥) = { (2)
tertiary control level manages the bidirectional power flow 0 if the system is not functioning
between the microgrid and the grid at the point of common
In our case, it is possible to calculate the structural
coupling (PCC). This level also ensures optimal economical importance of a component i in structure φ using [19] (3).
operation of the microgrid (Fig. 3). 1
Scalability is an advantage of decentralized architecture, 𝐼𝜑 (𝑖) = ∫(ℎ(1𝑖 , 𝑃) − ℎ(0𝑖 , 𝑃))𝑑𝑝 (3)
when the microgrid grows in terms of the number of power 0
components without disrupting the operation or re-engineering Where ℎ(1𝑖 , 𝑃) is the probability that the system operates
the control algorithm. This advantage influences the plug-and- with no failures, and ℎ(0𝑖 , 𝑃) is the probability that the system
play capability of the system. Scalability also outcomes a fault fails when component i fails. As defined in the function of
tolerant system, where it maintains availability and operates at component reliabilities
the minimum level of reliability. This demands a recovery (1𝑖 , 𝑃) = (𝑥1 , … 𝑥𝑖−1 , 1, … , 𝑥𝑛 ) (4)
algorithm as a part of the energy management system and (0𝑖 , 𝑃) = (𝑥1 , … 𝑥𝑖−1 , 0, … , 𝑥𝑛 ) (5)
certain level of redundancy to boost the reliability of the
microgrid system. Figure 4 shows the reliability block diagram of a possible
microgrid configuration. Each block represents one possible
III. MICROGRID CONTROL RELIABILITY ANALYSIS component or a subsystem with a pre-defined failure rate. A
working system remains while a continuous line from left to
Microgrids can be deployed for various purposes in an
right is maintained.
island or grid-connected structure. For example, a microgrid
Higher reliability of a system is proportional to the degree
intended to operate in two modes (grid-connected and
of parallelization of the reliability model [22]. In microgrids,
islanded) can be dispatchable, serving the purpose of
the controller is a vital component to maintaining operation.
supporting the distribution system. Distant microgrids away
As shown in Fig. 4, The red portion of the diagram represents
from the grid usually serve the purpose of continuously and
a controller as in series block to the system. Failure of the
independently supporting local loads. Loads can be
controller breaks the line and the system is declared in failure
categorized into critical and non-critical, and their
state.
characteristics can vary from static to dynamic behaviors.
Decentralization of the control architecture transforms the
Regardless of the type, microgrids under any disturbance or
series-parallel reliability block diagram in Fig. 4 into a parallel
fault condition have different behavior and performance, while
system [22], which decreases the degree of importance of a
supporting critical loads. The reliability analysis of microgrids
controller in the architecture. The degree of importance of a
is performed here based on three objectives: (i) Supporting
controller is calculated using (3) for three cases: 1) Centralized
critical loads, with the assumption of partially shedable loads,
controller architecture. 2) Redundant control architecture with
(ii) Microgrid bus voltage regulation, and (iii) Microgrid bus
two controllers. 3) True decentralized control architecture.
frequency regulation.
Equations that govern the importance of three cases from
equations (3-5) are
Case 1) Centralized:
𝑅𝐶𝑒𝑛 (𝑡) = 𝑅𝐶𝑀𝐺𝐶 (𝑡) ∗ 𝑅𝑃𝑆 (𝑡)
𝑚 )𝑛 )
𝑅𝐶𝑒𝑛 (𝑡) = 𝑃𝑐𝑜𝑛 (1 − (1 − 𝑃𝑐𝑜𝑚
(6)
ℎ(1𝐶𝑐𝑜𝑛 , 𝑃) = 1 − (1 − 𝑃 4 )5
ℎ(0𝑐𝑜𝑛 , 𝑃) = 0 }
Case 2) Redundant:
𝑅𝑅𝐸𝐷 (𝑡) = 𝑅𝑅𝑀𝐺𝐶 (𝑡) ∗ 𝑅𝑃𝑆 (𝑡)
2 𝑚 )𝑛 )
𝑅𝑅𝐸𝐷 (𝑡) = 1 − (1 − 𝑃𝑐𝑜𝑛 ) ∗ (1 − (1 − 𝑃𝑐𝑜𝑚
4 5
(7)
ℎ(1𝑅𝑐𝑜𝑛 , 𝑃) = 1 − (1 − 𝑃𝑐𝑜𝑚 )
4 )5
Figure 4. Reliability block diagram of a microgrid architecture. ℎ(0𝑅𝑐𝑜𝑛 , 𝑃) = (1 − (1 − 𝑃𝑐𝑜𝑛 )) ∗ 1 − (1 − 𝑃𝑐𝑜𝑚 }

1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2843527, IEEE
Transactions on Smart Grid

Case 3) Decentralized: time interval ∆𝑡, and does not have a memory characteristic.
𝑅𝐷𝐸𝐶 (𝑡) = 𝑅𝑃𝑆 (𝑡) For a system of n states, a probability transition matrix is
𝑚 )𝑛
𝑅𝐷𝐸𝐶 (𝑡) = (1 − (1 − 𝑃𝑐𝑜𝑚 ) defined as
4 4 3 (8) 𝑃11 (∆𝑡) 𝑃12 (∆𝑡) … 𝑃1𝑛 (∆𝑡)
ℎ(1𝐷𝑐𝑜𝑛 , 𝑃) = 1 − ((1 − 𝑃𝑐𝑜𝑚 ) ∗ (1 − 𝑃𝑐𝑜𝑚 ))
4 4 𝑃21 (∆𝑡) 𝑃22 (∆𝑡) … 𝑃2𝑛 (∆𝑡)
ℎ(0𝐷𝑐𝑜𝑛 , 𝑃) = 1 − (1 − 𝑃𝑐𝑜𝑚 ) } 𝑷(∆𝑡) = [ ] (10)
⋮ ⋮ ⋱ ⋮
𝑃𝑛1 (∆𝑡) 𝑃𝑛2 (∆𝑡) ⋯ 𝑃𝑛𝑛 (∆𝑡)
Where 𝑅𝐶𝑒𝑛 (𝑡), 𝑅𝑅𝑐𝑜𝑛 (𝑡), 𝑅𝐷𝑐𝑜𝑛 (𝑡) are the total system
reliability for the three cases, respectively. Where
𝑅𝐶𝑀𝐺𝐶 (𝑡), 𝑅𝑅𝑀𝐺𝐶 (𝑡), 𝑅𝐷𝑀𝐺𝐶 (𝑡) are the controller reliability for 𝑃𝑖𝑗 (∆𝑡) ≥ 0 𝑖, 𝑗 ∈ [1, 𝑛] (11)
each case. 𝑅𝑃𝑆 (𝑡) is the reliability of the system not including 𝑛

the controller (parallel section). 𝑃𝑐𝑜𝑛 , 𝑃𝑐𝑜𝑚 are the failure ∑ 𝑃𝑖𝑗 (∆𝑡) = 1, 𝑖 ∈ [1, 𝑛] (12)
probabilities of a controller and any other component, 𝑗=1

respectively. Equation (10) can be written as (13) due to homogeneous

Figure 5 shows the calculation results of importance property.
calculations in a microgrid control architecture for the three 𝑃11 𝑃12 … 𝑃1𝑛
𝑃21 𝑃22 … 𝑃2𝑛
cases. Assuming four components in a parallel branch, and 𝑃=[ ] (13)
⋮ ⋮ ⋱ ⋮
varying the number of possible DGs in a microgrid. Scaling up 𝑃𝑛1 𝑃𝑛2 ⋯ 𝑃𝑛𝑛
the microgrid, the importance of a controller increases in the Markov reliability models can be simulated based on failure
centralized architecture even with a redundant controller. rates λ of system components instead of probability of failure
However, due to parallelization in decentralized architecture, P, forming a transition matrix M. If the system is repairable,
the importance of each controller decreases as the microgrid
repair rates μ are included to the transition matrix [20].
Simulation of the reliability model results with a predicted
reliability of the system. An example of such technique is
proposed in the next section.

B. Reliably of Decentralized Control Architectures

Figure 7 shows the proposed microgrid decentralized
control architecture. By eliminating the centralized controller
of a conventional architectures, the system transforms into
certain number of parallel branches (subsystems). For this
study purposes, each branch is assumed to have four
components: the local decentralized controller, and two
Figure 5. Controller Degree of Importance in three Centralized, communication lines and the distributed generation (DG) unit.
Redundant, and Decentralized Architectures. As an example, the DG in the expanded branch illustrates a
PV system.
expands in numbers. A review of literature was performed to identify the failure
2) Markov Chain Analysis rates for each component. The reliability of a controller is
Markov Chain Reliability Model (MRM) uses a stochastic governed by the quality of the material and the possible
process which may describe complex behavior of a stochastic protection mechanisms. Software wise, the decentralized
system. MRM is being used to model a system with several algorithm is more complex and more susceptible to logic
states and transitions between states. A Markov reliability errors if the software engineering level was not adequate. Due
model contains a series of the possible states in the system and to the various factors, the possibility of establishing a firm
uses possible failure rates and repair rates between those comparison between centralized and decentralized controllers’
states. One of the advantages of using Markov Chain failure rates was irrelevant, So, we followed the literature by
Modeling is the capability of computing the steady state using the equal failure rates for the corresponding components
probabilities of all system states, and estimating probability of
rarely occurring events and failures. The practical values of
failure rates used in our simulations can be found in [20].
Markov Chain modeling works with systems that are
relatively small in terms of number of states. Larger systems
(i.e power systems) can have thousands of states, and requires
additional techniques in order to achieve faster computation
without losing the accuracy of the model.
Given X(t) a random variable in Markov process, the
transition probability function between two states i, j is
denoted as
𝑃𝑖𝑗 (∆𝑡) = 𝑃[𝑋(𝑡 + ∆𝑡) = 𝑗 |𝑋(𝑡) = 𝑖] (9)
Figure 6. Markov model and state transition diagram for a parallel
The transition from state i to j depends on the transition branch.

1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2843527, IEEE
Transactions on Smart Grid

of both architecture..
Markov reliability model for the system in Fig. 7 results
with transition matrix representing 241 states, for n=5 DGs in
centralized architecture, and 240 in decentralized architecture.
Due to large number of states. Lumping technique is used to
simplify the transition matrix for the Microgrid System [20].
Reliability of each branch is evaluated using Markov
modeling. Two cases are considered, repairable and non-
repairable. In a non-repairable system, failure of any
component is considered permanent. A repairable system is a
practical case in power systems, where a failed component is
repaired or replaced after failure is discovered. Markov chain
simulation predicts the steady state reliability of the system. A
repairable system converges to certain reliability with time, on
the contrary of a non-repairable system where the reliability
curves converges to zero, depending on the length of the
simulation interval intended for analysis.

Table 1 BRANCH STATES AND THEIR CORRESPONDING

RELIABILITY EQUATIONS.
State DER CL2 LC CL1 System
R1(t) R2(t) R3(t) R4(t) State C P(t) Figure 7. Microgrid Decentralized Control Architecture (Green).
0 Up Up Up Up Up R1(t). R2(t). R3(t). R4(t) Eliminated centralized controller (Red). Example branch components
1 Down Up Up Up Down (1-R1(t)). R2(t). R3(t). R4(t) (Subsystem).
2 Up Down Up Up Down R1(t). (1-R2(t)). R3(t).R4(t)
3 Up Up Down Up Down R1(t). R2(t).(1-R3(t)). R4(t) 𝐴
4 Up Up Up Down Down R1(t). R2(t). R3(t).(1-R4(t))
𝐵
Table 1 represents a branch states of Fig (6). The total 𝑀𝐶𝑒𝑛𝑡𝑟𝑎𝑙𝑖𝑧𝑒𝑑 = (16)
reliability of the branch P(t) depends on the previous state. 𝐶
The reliability equation for each state follows the reliability
equations in [22]. For each branch, the transition matrices as [𝐷 ]
non-repairable and repairable cases as shown in Fig. 6 are Where A and B are 20Χ64 matrices. A and B represent the
depicted in equations (14) and (15), respectively. For a branch acceptable states and critically acceptable states respectively
that comprises of 4 components in series, any component (total of 20 states). At these states, the centralized controller is
failure will cause a failure in the whole branch due to the high in working state, while in C and D (44Χ64 matrices), the
dependency of the component to each other. The branch controller is down and the microgrid system is considered
moves from an UP state to an intermediate state at different down or unstable and requires shutting down (total 44 states).
failure probabilities (rates), which represents a DOWN state of In case of decentralization of controls, the transition matrix is
the whole branch. Each branch will be considered a subsystem reduced to 50% in terms of number of states since a single
of a microgrid, with a failure rate determined using equation point of failure has been eliminated which is depicted as the
(14) or (15), depending on the reparability of each component. red portion of the block diagram in Fig.4. The transition
∗ 𝜆1 𝜆2 𝜆3 𝜆4
matrix for this case is defined as
0 0 0 0 0
𝑀𝑁𝑜_𝑅𝑒𝑝𝑎𝑖𝑟 = 0 (14) 𝐴
0 0 0 0
𝑀𝐷𝑒𝑐𝑒𝑛𝑡𝑟𝑎𝑙𝑖𝑧𝑒𝑑 = [ ]
0 0 0 0 0 (17)
[0 0 0 0 0] 𝐵

∗ 𝜆1 𝜆2 𝜆3 𝜆4 Where A and B are 20x32 and 12x32 matrices,

𝜇1 −𝜇1 0 0 0 respectively. The failure states follow the same description of
𝑀𝑅𝑒𝑝𝑎𝑖𝑟 = 𝜇2 0 −𝜇2 0 0 (15) the centralized transition matrix.
𝜇3 0 0 −𝜇3 0 Markov Chain simulation is conducted using MATLAB©.
[𝜇4 0 0 0 −𝜇4 ] equivalent failure rates for each branch is calculated for the
The asterisk value is the negative summation of the rest of equivalent fault tree according to the rates in [20]. The main
the row. Similarly, given a microgrid with 5 DGs, transition purpose of such analysis is to identify the improvement of the
matrices are implemented. Using lumping technique, the overall microgrid system reliability moving from centralized
number of states are reduced, since the microgrid are now to decentralized architecture. Another purpose is to study the
consisting of 5 subsystems in addition to the controller (in impact of a single controller on the overall system in both
case of centralized). Equations (16) and (17) shows the architectures. The probability distribution vector (18) is
transition matrices for both cases. obtained using the transition matrix.
𝑃(𝑡) = [𝑃0 (𝑡) … 𝑃𝑛 (𝑡)] (18)
𝑃̇ (𝑡) = 𝑃(𝑡) ∙ 𝑀 (19)

1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2843527, IEEE
Transactions on Smart Grid

IV. EXPERIMENTAL RESULTS

Start
A. Decentralized Architecture Implementation
For testing purposes, a Hardware-In-the-Loop (HIL)
Initialize system state platform [12] were developed to study microgrid operations
matrix X and C
with real physical communication layer and applying the
decentralized architecture shown in Fig. 7. A dedicated
workstation running microgrid simulation model. The
j=1, t=0
workstation is equipped with multi-Ethernet ports, binding the
model with a dedicated Ethernet port serves the purpose of
= ̇ ( ). avoiding impractical network congestion with other network
= +
related traffic, i.e. Internet. PSCAD is an ideal candidate for
our platform. The simulator is widely used for multi-phase
( )= ∗
power systems and control networks in time domain, and
mainly dedicated to the study of transients of power system,
No
j > iterations

Yes

End

Figure 10 Markov Chain Simulation flowchart

Where X is the initial state of the system

𝑋 = [1, 0, … . , 0𝑛−1 ], n=32 or 64 (decentralized, centralized).
C represents the states where the microgrid system is in a
healthy state 𝐶 = [1, 1, … . , 0,1, 𝐶𝑛−1 ]. Initially, the iteration
counter is set to j=0, and the simulation loops until the number
of iterations is reached. The number of iteration is determined
using the following equation:

𝑇𝑜𝑡𝑎𝑙 𝑠𝑖𝑚𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑡𝑖𝑚𝑒

𝐼𝑡𝑒𝑟𝑎𝑡𝑖𝑜𝑛𝑠 = (20)
∆𝑡

The results of the Markov reliability simulations are

illustrated in Figures 9-10. Figure 9 shows the reliability
curves of the overall microgrid system for the two
architectures: centralized and decentralized. The reliability
Figure 8. Microgrid system reliability curve assuming no repairs
function R(t) is the probability that an item does not fail in the
time interval (0, t]. In centralized case, the overall microgrid
reliability decreases with time and goes below 50% at 2.5
years, in comparison to approximately 90% with decentralized
architecture. However, for a practical case where the system is
repairable; the reliability of the microgrid converges to 56% in
12 years with a centralized architecture compared to a 94% for
the same time period in decentralized architecture.
Four improvements of controller failure rates are included in
simulations results in Figures 8 and 9, reflecting 20% decrease
in failure rate of single controller. The failure rate is reapplied
to the transaction matrix M for each improvement. The failure
rate is calculated using the following equation
𝑞+1 𝑞
𝜆𝑐 = 0.8𝜆𝑐 (20)
Where 𝜆𝑐 is the new failure rate of the controller, and 𝜆𝑞𝑐 is the
𝑞+1

previous failure rate.

Validating the results in Fig. 5, the degree of importance of
a single controller on the overall system reliability is larger in
the case of centralized architecture. Generally, scaling up the
microgrid (increasing the number of DGs), the overall
reliability of the microgrid is improved when the architecture
is decentralized, unlike the centralized architecture where the
reliability decreases. Figure 9. Microgrid system reliability curves (repairable system).

1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2843527, IEEE
Transactions on Smart Grid

which is one of the future aspects to study using the proposed

platform. Accurate model interaction between power system
components and loads with various control topologies is also a
preferred feature in simulation that is available in PSCAD.
Microgrid controls in this platform are developed using the
real-time module of the CompactRIO from National
Instruments. Its capability to run in real-time interface mode
serves the purpose of the platform.
B. Microgrid Normal Operation
For the proof-of-concept purposes, one scenario for normal
operations is considered. The microgrid operates in island
mode, energy storage regulates the voltage and frequency of
the bus during the transition period. Figure 11 shows the
active power curves captured over 130 seconds period. The
system simulation starts with fully charge Energy Storage
(ES). Natural Gas generator 1 (NG1), Natural Gas generator 2
(NG2) are off and their breakers are open. The nature of the
loads varies with time starting with 60 KW and increasing. ES
provides the power to the loads for 16 seconds until the
decentralized controller at the ES unit detects 60% SOC on Figure 11. Normal operation of decentralized microgrid control
remaining on the battery. As a decentralized system and with architecture.
general awareness condition is maintained; Decentralized
Microgrid Controller (DMGC) of NG1 commands NG1 to
start and synchronize with the bus, and commands the breaker
after 6 seconds providing 190KW (at full capacity). Since the
load demand is greater than the capacity of NG1, DMGC of
NG2 detects the issue and connects at t = 26s. The DMGC if
ES detects that NG1 and NG2 are active, and switches to
charging mode.
C. Controller Failure Impact
One of the advantages of using HIL platform is the
capability of configuring and injecting failures at the hardware
and/or software level. Since the scope of this paper is the
failure of the controller; failing a controller is performed by
powering down the controller, or resetting the controller
manually. Two test cases have been performed in order to
validate the impact of failures in the control system: 1) failing
the central controller in a centralized architecture. 2) failing Figure 12. Impact of failing microgrid central controller on bus voltage
and frequency.
one decentralized controller. The conducted test cases do not
represent the most extreme cases, but they were chosen as they
occur during transient periods. These cases may have low
probabilities to occur, but the reliability analysis in section IV
were simulated for over 10-year period.
Figure 12 illustrates a scenario where the central controller
of the microgrid fails during islanded mode. For this case, at
t=9s controller commands NG1 and NG2 to connect as the
SOC of the ES is low (as in subsection B), during the
transition the controller fails, at t=11s. As a result of this
failure, bus voltage and frequency become unstable, and
converge to undesired levels. This scenario forces the
microgrid to shut down.
Decentralization of a control system comes with additional
overhead algorithm in response to failures. Generally, a rule
based decision making algorithm is executed in this case in
order to support the objectives of microgrid operations
(frequency, voltage, critical loads). The responses of the Figure 13. Impact of failing one decentralized controller on microgrid
decentralized controllers should insure fast transition after the operation.
failure occur.

1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2843527, IEEE
Transactions on Smart Grid

Figure 13 illustrates the case of failure of one decentralized [4] X. Xu, J. Mitra, T. Wang and L. Mu, "An Evaluation Strategy for
Microgrid Reliability Considering the Effects of Protection System," in
controller, the failing controller for this case is the NG1,
IEEE Transactions on Power Delivery, vol. 31, no. 5, pp. 1989-1997,
which could be one of the extremist cases since the generator Oct. 2016.
may be responsible regulating the bus voltage/frequency at the [5] P. Baboli, "Flexible and overall reliability analysis of hybrid AC–DC
time of failure. At t=47.5, the controller of NG1 fails while microgrid among various distributed energy resource expansion
scenarios," in IET Generation, Transmission & Distribution, vol. 10, no.
both generators are running and the ES is in charge mode.
16, pp. 3978-3984, 12 8 2016.
Two controllers can respond fast to this change, ES controller [6] S. Wang, Z. Li, L. Wu, M. Shahidehpour and Z. Li, “New Metrics for
can command ES to take over, or the PCC controller can Assessing the Reliability and Economics of Microgrids in Distribution
command emergency grid connection. For this case, PCC System,” in IEEE Transactions on Power Systems, vol. 28, no. 3, pp.
2852-2861, Aug. 2013.
responded since the SOC of the battery is critically low.
[7] D. E. Olivares et al., "Trends in Microgrid Control," in IEEE
NG1and NG2 are shut down. One other case where microgrid Transactions on Smart Grid, vol. 5, no. 4, pp. 1905-1919, July 2014.
needs to remain in islanded mode, NG2 can remain running [8] Z. Li, Y. Yuan and F. Li, “Evaluating the reliability of islanded
but load shedding can be performed until the failure clears [2]. microgrid in an emergency mode,” 45th International Universities
Power Engineering Conference UPEC2010, Cardiff, Wales, 2010, pp. 1-
In order to prove the scalability of a decentralized control
5.
architecture, NG1 controller returns to the system at t=66s and [9] A. Solanki, A. Nasiri, V. Bhavaraju, T. Abdullah, D. Yu, “A New
the reconnection operation is performed. The return of NG1 is Control Method for Microgrid Power Management”, International
reported to the rest of the controllers in order to return to Conference on Renewable Energy Research and Applications. Oct.
2013, Madrid.
normal operation mode.
[10] A. Bani-Ahmed, L. Weber, A. Nasiri and H. Hosseini, "Microgrid
communications: State of the art and future trends," 2014 International
V. CONCLUSION Conference on Renewable Energy Research and Application (ICRERA),
Milwaukee, WI, 2014, pp. 780-785.
Microgrids are considered a solution to reliability [11] A. Bani-Ahmed and A. Nasiri, "Development of real-time hardware-in-
enhancement of distribution systems. As a vital investment, the-loop platform for complex microgrids," 2015 International
reliability of microgrids becomes a challenge when various Conference on Renewable Energy Research and Applications
(ICRERA), pp. 994-998, Palermo, 2015.
technologies are integrated. This paper investigated the [12] Guerrero J M, Chandorkar M, Lee T, Loh, P C. “Advanced Control
reliability measures of microgrid control architectures. Architectures for Intelligent Microgrids—Part I: Decentralized and
Reliability models of centralized and decentralized Hierarchical Control. Industrial Electronics”, IEEE Transactions on
Industrial Electronics, vol.60, no.4, pp.1254-1262.
architectures were analyzed and demonstrated using HIL [13] W. Liu, W. Gu, W. Sheng, X. Meng, Z. Wu and W. Chen,
platform. Results show that decentralization of microgrid "Decentralized Multi-Agent System-Based Cooperative Frequency
control architecture improves the reliability as the single point Control for Autonomous Microgrids With Communication Constraints,"
in IEEE Transactions on Sustainable Energy, vol. 5, no. 2, pp. 446-456,
of failure is eliminated in centralized architecture. Challenges April 2014.
arise when decentralized architectures are adopted, controller [14] F. Shahnia, R. P. S. Chandrasena, S. Rajakaruna and A. Ghosh, "Primary
to controller synchronization and certain levels of intelligence control level of parallel distributed energy resources converters in
is needed to achieve optimized performance and protection. system of multiple interconnected autonomous microgrids within self-
healing networks," in IET Generation, Transmission & Distribution, vol.
Cyber-security concerns emerge as communication links are 8, no. 2, pp. 203-222, February 2014.
essential for optimal performance. Interdisciplinary Research [15] W.Su, J. Wang, “Energy Management Systems in Microgrid
on decentralization is necessary comprising research fields of Operations” in The Electricity Journal, , vol. 25, no. 8, October 2012.
[16] T. Wang, D. O’Neill and H. Kamath, "Dynamic Control and
software, hardware, and communications. Future efforts Optimization of Distributed Energy Resources in a Microgrid," in IEEE
include diagnostic algorithms and resiliency of the control Transactions on Smart Grid, vol. 6, no. 6, pp. 2884-2894, Nov. 2015.
architecture is still needed, considering the primary control [17] M. I. Azim, M. J. Hossain, F. H. M. R. Griffith and H. R. Pota, "An
improved droop control scheme for islanded microgrids," 2015 5th
level. Australian Control Conference (AUCC), , 2015, pp. 225-229.
[18] J. M. Guerrero, P. C. Loh, T. L. Lee and M. Chandorkar, "Advanced
ACKNOWLEDGEMENT Control Architectures for Intelligent Microgrids—Part II: Power
Quality, Energy Storage, and AC/DC Microgrids," in IEEE Transactions
on Industrial Electronics, vol. 60, no. 4, pp. 1263-1270, April 2013.
Financial support for this project was partially provided by [19] G. Levitin, L. Xing, H. Ben-Haim and Y. Dai, "Reliability of Series-
the NSF I/UCRC center on Grid-connected Advanced Power Parallel Systems With Random Failure Propagation Time," in IEEE
Electronics Systems (GRAPES) under project number GR-17- Transactions on Reliability, vol. 62, no. 3, pp. 637-647, Sept. 2013.
[20] G. Bolch et al., Queueing Networks and Markov Chains, New York,
10. NY: Wiley Interscience. 1998.
[21] A. Bani-Ahmed, A. Nasiri and H. Hosseini, "Design and development of
REFERENCES a true decentralized control architecture for microgrid," 2016 IEEE
Energy Conversion Congress and Exposition (ECCE), Milwaukee, WI,
[1] S. Bi; Y. J. A. Zhang, "Graph-based Cyber Security Analysis of State 2016, pp. 1-5.
Estimation in Smart Power Grid," in IEEE Communications Magazine, [22] Y.F. Li, R. Peng, “Availability modeling and optimization of dynamic
vol.99, pp.2-9, February 2017 multi-state series–parallel systems with random reconfiguration” ,
[2] J. Zhang, L. Guan and X. Wang, “Impacts of island load shedding and Reliability Engineering & System Safety. v.27, pp 47-57, 2014,
restoration strategies on reliability of microgrid in distribution system,”
2016 IEEE PES Asia-Pacific Power and Energy Engineering
Conference (APPEEC), Xi'an, China, 2016, pp. 1594-1598.
[3] X. Xu; T. Wang; L. Mu; J. Mitra, "Predictive Analysis of Microgrid
Reliability Using a Probabilistic Model of Protection System
Operation," in IEEE Transactions on Power Systems. vol, no.99, pp.1-1

1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2843527, IEEE
Transactions on Smart Grid

Abedalsalam Bani-Ahmed (M’11) was Mohammad Rashidi was born in Parsian,

born in Sakib, Jordan. He received his Iran, in 1986. He received B.S. and M.S.
B.Sc degree in Engineering Technology degrees from Amirkabir University of
from Yarmouk University, Irbid, Jordan Technology (Tehran Polytechnic), Tehran,
in 2007, and his M.Sc degree in Computer Iran, in 2009 and 2012, respectively, and the
Engineering from Jordan University of PhD degree from the University of
Science & Technology, Irbid, Jordan, in Wisconsin Milwaukee, Milwaukee,
2011, and the PhD degree in Electrical Wisconsin, in 2017, all in electrical
Engineering from the University of Wisconsin - Milwaukee, engineering-Power.
Milwaukee, Wisconsin, in 2017. Dr. Rashidi is presently a Lead Power Conversion Engineer
He worked as research assistant at the Center for with the Corporate Research and Technology Group, Eaton
Sustainable Electrical Energy Systems (UW-Milwaukee) from Corporation, Menomonee Falls, WI. He is also serving as
2012 to 2017, and a Power Systems Engineer at Eaton chapter chair of IEEE PES chapter at Milwaukee section. He
Corporation in 2017. Dr. Bani-Ahmed is currently a Lead has published over 10 conference and journal papers.
Engineer at Eaton Corporation – Corporate Research and His current research interests include modeling and assessing
Technology. His research interests in Smart Grids, Distributed of microgrids, and design and implementation of power
Energy Resources Management Systems (DERMS), IIoT electronic converters for grid connected DERs.
communications infrastructure, protocols, cyber-security and
standards. He has published numerous technical journal and
conference papers, and white papers related topics. Hoessein Hosseini is a professor of
Dr. Bani-Ahmed is currently a serving in IEEE Smart Grid Electrical Engineering and Computer
R&D committee, IEEE Smart Grid Operations committee, Science at University of Wisconsin-
IEEE Smart Cities R&D committee, IEEE Smart Cities Milwaukee (UWM). He received his PhD
Education committee, IEEE Smart Cities Publication degree in Electrical and Computer
committee. He also serves as a reviewer to multiple IEEE PES Engineering from University of Iowa in
and IAS transactions, and various international Smart Grid 1982. Dr. Hosseini’s expertise is in the
journals. areas of Computer Networks, Computer
Architecture, Fault-Tolerance, Distributed and Parallel
Adel Nasiri (SM’06) was born in Sari, Computing. He is the founder and director of the Computer
Iran, in 1974. He received B.S. and M.S. Networks Laboratory and serves on the advisory board of the
degrees from Sharif University of Connected System Institute (CSI) at UWM. Dr. Hosseini has
Technology, Tehran, Iran, in 1996 and published over 130 research papers in refereed journals and
1998, respectively, and the PhD degree conference proceedings. One of his co-authored papers has
from Illinois Institute of Technology, won the Best Paper Award. He has published two book
Chicago, Illinois, in 2004, all in electrical chapters. He is the co-recipient of a patent in the field of
engineering. Computer Networks. He has received funding from NSF and
He worked for Moshanir Power Engineering Company, from industry in support of his research and education. He has
1998 to 2001. He also worked for ForHealth Technologies, supervised and graduated over ten PhD and 60 MS students.
Inc., Daytona Beach, Florida, from 2004 to 2005. Dr. Nasiri is He has served on the editorial board of a journal and on the
presently a professor and Director of Center for Sustainable program committee of several international conference
Electrical Energy Systems in the College of Engineering and proceedings. He regularly reviews research papers for various
Applied Sciences at the University of Wisconsin–Milwaukee. journals and conference proceedings and textbooks for book
His research interests are renewable energy interface, energy publishers. Dr. Hosseini has played a leading role in the
storage, and microgrids. Dr. Nasiri has published numerous growth and development of Electrical Engineering and
technical journal and conference papers on related topics. He Computer Science Programs at UWM. His efforts include the
also has seven patent disclosures. He is a co-author of the development of the BS degree program in Computer
book “Uninterruptible Power Supplies and Active Filters,” Engineering program. Dr. Hosseini has extensive
CRC Press, Boca Raton, FL. administrative experiences. He served as the Computer
Dr. Nasiri is currently an Editor of IEEE Transactions on Science chair (department co-chair) from 2002 to 2012. He
Smart Grid, Paper Review Chair for IEEE Transactions on has served on important committees such as College of
Industry Applications, an Editor of Power Components and Engineering and Applied Science Strategic Planning
Systems, and Associate Editor of the International Journal of Committee, College Academic Planning Committee (APC),
Power Electronics. He was the general Chair of 2012 IEEE Division of Natural Sciences Executive Committee, and
Symposium on Sensorless Electric Drives, 2014 International UWM Senate.
Conference on Renewable Energy Research and Applications .
(ICRERA 2014), and 2014 IEEE Power Electronics and
Machines for Wind and Water Applications (PEMWA 2014).

1949-3053 (c) 2018 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.