Professional Documents
Culture Documents
Using EVs As Distributed Energy Resources For Critical Load Restoration in Resilient Power Distribution Systems
Using EVs As Distributed Energy Resources For Critical Load Restoration in Resilient Power Distribution Systems
Research Article
Abstract: Owing to the important role of the power system in modern societies, its resilience against natural disasters has
become a top priority for power system operators and planners. Moreover, in recent years, due to the increasing number and
severity of weather-related events, several operational solutions have been proposed to increase the resilience of power
systems. For instance, in the distribution part of power systems, dynamic micro-grid (MG) formation can enhance resilience by
means of distributed energy resources (DERs) when the main grid is unavailable. Following this concept, the capabilities of
electric vehicles (EVs) such as vehicle-to-grid make it possible to use the energy stored inside them to restore critical loads.
Therefore, the aggregation of EVs in the distribution system can be seen as a DER from the system operator viewpoint, and
operators can benefit from this limited energy resource in MG formation. This study uses two types of EV aggregation for this
purpose, i.e. a public parking lot and a residential parking. To improve system performance when using these resources, two
steps should be taken: (i) employing a master–slave control technique in each MG and (ii) using a demand-side management
programme.
IET Gener. Transm. Distrib., 2020, Vol. 14 Iss. 18, pp. 3750-3761 3750
© The Institution of Engineering and Technology 2020
ργi, t binary variable indicating employing status of load islands. In [8], critical loads in distribution feeders and local loads
control step γ in node i were restored using MGs. Dynamic MG formation for load
pik, t active power injected in node i restoration from several DERs is used in [1, 5, 9–11]. The effect of
reactive power injected in node i tie switches on increasing restored loads was studied in [1, 5, 10,
qik, t
11]. The master–slave control technique was deployed in [1, 10] to
V ik, t voltage of node i use more than one resource in each formed MG and increase the
i, t
PSup supplied active power in node i number of restored loads. In [1], different resilience dimensions
i, t supplied reactive power in node i and associated indexes of each were also introduced.
QSup
Owing to increasing environmental concerns, the use of electric
pSt . Geni active power generation of the slave unit vehicles (EVs) is now the subject of intense interest. Moreover, the
qSt . Geni reactive power generation of the slave unit in node i aggregation of large numbers of EVs with capabilities of
active power injected in node i along path θ bidirectional power transfer has special advantages for the power
pik, ,θt
system. Vehicle-to-grid (V2G) and grid-to-vehicle are two
qik, ,θt reactive power injected in node i along path θ capabilities of EVs that enable them to participate in power system
k , i, t
Pcommon common active power generated/consumed in node i frequency regulation service [12–14], perform local voltage control
k, i, t common reactive power generated/consumed in node [15], and enhance power system reliability [16, 17]. During the
Qcommon
i outage, this ability can help to restore more critical loads by
voltage of node i along path θ employing EVs’ stored energy. The novel aspects employed in this
V ik, ,θt
study and other references are shown in Table 1.
m, t m, t
nch , nche binary variable indicating the charging status of EV Despite several technical studies on MG formation using
m, in normal or emergency condition
n
different resources such as diesel generators and deep-cycle
m, t
ndch m, t
, ndch binary variable indicating the discharging status of EV batteries, to the best of our knowledge, none of them has focused
m, in normal or emergency condition
n e
on the energy stored in EVs and their energy management
m, t m, t
Pch , Pche charging power of EV m, in a normal or emergency immediately after the incident. In the present study, we consider
condition
n
each EV parking as a distributed energy resource (PLs and
m, t
Pdch m, t
, Pdch discharging power of EV m, in a normal or emergency residential parking (RP) are considered). A master–slave control
condition
n e
technique and demand response programme are also implemented
SOCm
n
,t SOC of EV m, in a normal condition in this study. When considering EVs in resilience studies, many
m, t
SOCe SOC of EV m, in an emergency condition questions need to be answered, such as how can EV owners
participate in resilience improvement programmes after disasters?
How much battery capacity do they have before events? Also, what
1 Introduction are the incentives for EV owners? These questions are
In recent years, severe natural phenomena such as hurricanes, fundamentals in resilience studies and are beyond the scope of the
floods, and earthquakes have increased throughout the world and current study. The authors will answer these questions in another
led to an increase in the number of sustained power outages [1]. work in the near future. This study tries to evaluate the techno-
For instance, in 2012, Hurricane Sandy damaged a large portion of economic benefits of the EVs’ participation in MG formation. The
the power system, causing a power outage nearly 7.5 million main contributions of this study can be summarised as follows:
customers on the east coast of the United States for several weeks
[2]. In such incidents, the distribution part of the power system is • A resilient MG-forming model using PLs and RP as DERs is
more vulnerable than other parts. In another case, in 1998, about proposed to restore critical loads.
80% of therestoration costs due to hurricanes and tropical storms • A comprehensive formulation is proposed to employ load
were attributed to distribution and 20% to the transmission part of control capabilities and master–slave technique for effective
the power system [3]. To reduce the social and economic losses management of the system in critical situations.
caused by these incidents, improving the resilience of power • A novel method for load restoration in mesh networks is
systems has become an important concern for operators and presented.
planners. According to the classification in [4], four essential steps • Mixed-integer linear programming (MILP) is employed to solve
are possible: anticipate & prepare, resist & absorb, response & the proposed model.
adopt, and recovery. The last step is followed in the present study,
as well. The remainder of this paper is organised as follows: in Section 2,
With natural disasters, substations may be damaged, causing the stochastic modelling of EVs is performed as distributed energy
main grid to be disconnected from the distribution network and storage and comprehensive analysis of parking lots’ (PLs’)
unable to supply power. Hence, employing traditional approaches available energy to participate in load restoration in various
to restore the distribution system is not feasible [5]. However, in conditions is carried out. MG formation and EVs energy
recent years, due to the emergence of the smart grid and, as a management are formulated in Section 3. Simulation results are
result, the enhanced level of connectivity between information and presented in Section 4. The applicability of the model is discussed
communications, new approaches have been presented for the in Section 5. Finally, Section 6 concludes the paper.
restoration of power distribution systems. Among these
approaches, the use of distributed energy resources (DERs) and the
formation of micro-grids (MGs) are common. In the literature, 2 EVs available energy
several papers have reviewed these approaches. In [6, 7], heuristic In general, EV charging/discharging is classified under two modes,
and exhaustive search algorithms were used to get the optimal centralised charging/discharging (utilised in PLs) and distributed
Table 1 Comparison of employing the novel aspects in this study and other references
References [12–14] [15] [16, 17] [5] [10] [1] This paper
EVs capabilities voltage control ✓ ✓
frequency control ✓ ✓
emergency back up ✓ ✓
MG formation features using tie switches ✓ ✓ ✓ ✓
employing the master–slave control technique ✓ ✓ ✓
using load control ✓ ✓
using EVs ✓
IET Gener. Transm. Distrib., 2020, Vol. 14 Iss. 18, pp. 3750-3761 3751
© The Institution of Engineering and Technology 2020
Table 2 Probability distribution of the stochastic parameters
EVs probabilistic parameter PDF of EVs in PL PDF of EVs in RP
car arrival time (tarr) Weibull (α = 0.9831, β = 16.8) truncated Gaussian distribution (TGD) (mean = 19, St.dev = 2, min = 16,
max = 1)
car departure time (tdep) Weibull (α = 80, β = 4.665) TGD (mean = 7, St.dev = 2, min = 5, max = 12)
initial SOC normal (μ = 50%, σ = 15%) TGD (mean = 75, St.dev = 25, min = 25, max = 95)
total number of arriving/departing EVs normal (μ = 20 cars, σ = 5 cars) normal (μ = 25 cars, σ = 5 cars)
from 7 to 23:59
Fig. 1 Schematic demonstration of random sampling from reduced 2.2 Parking available energy in each scenario
scenarios
In the previous section, after the generation of early scenarios, the
amount of available parking energy is obtained for each scenario.
For this purpose, first, the charging scheduling of EVs is
determined [17]. So, we assumed that the charging scheduling
presented in [17] is implemented by each aggregator. The objective
of this scheduling is to minimise the total user cost, which includes
two terms in each hour: (i) the cost of energy consumed for
charging and (ii) the reward payment for discharging battery
energy. Once the problem's objective function is specified, the
charge/discharge behaviour of each EV can be modelled at
different times based on constraints (1) to (8). For each scenario,
the energy profiles of all EVs in 24 h are identified using the
charge scheduling formulation. Then, the amount of energy stored
in each battery at any time is determined. Here, the aggregators
must determine the amount of energy available to each EV in
accordance with a contract signed with each EV user specifying the
emergency SOC at departure. To do this, the aggregators consider
Fig. 2 Energy profile of EVs for various scenarios the amount of surplus energy over the final emergency SOC as the
available energy for each EV using (9). It is of note that in every
charging (utilised in residential garages). Regardless of the mode of scenario, for EVs entering parking after the incident, their initial
operation, providing various services to the grid requires EVs to be SOC is used in (9) to calculate the energy available at the time of
organised by an aggregator. Each aggregator is an interface the accident. Fig. 2 shows various situations that may occur for
between the power system operator and the availableEVs [18]. If each EV in our consideration. In this figure, EVs’ amount of stored
an incident occurs and the main grid fails to supply the distribution energy of EV at the start of restoration (trs) in the first and 10,000th
system, according to requests from the control centre, aggregators scenarios is higher than the initial SOC. In comparison, in the
must provide information such as the amount of available energy second scenario, because EVs enter after the incident, their amount
and power of vehicles participating in the V2G emergency of stored energy is equal to their initial SOC. Finally, in the third
programme. For critical load restoration, the control centreuses this scenario, because EVs depart before the incident, the amount of
information and other real-time data such as resource availability, energy available to them for restoration is 0
location of damaged remote-control switches, load forecasting
data, and distribution lines damaged due to the incident. MG tm . dep
3752 IET Gener. Transm. Distrib., 2020, Vol. 14 Iss. 18, pp. 3750-3761
© The Institution of Engineering and Technology 2020
SOCtnarr = SOCinitial (7)
t
SOCndep = SOCfinal (8)
IET Gener. Transm. Distrib., 2020, Vol. 14 Iss. 18, pp. 3750-3761 3753
© The Institution of Engineering and Technology 2020
with regard to its switch status (si) can participate in restoration
with one master unit. If si = 1, the slave unit can participate in
restoration and vice versa. Constraint (14) indicates that if the line
between nodes i and j is damaged due to the incident, neither node
can be in one MG. Constraint (15) shows that if the remote control
switch between nodes i and j is damaged due to the incident, the
restoration status of nodes i and j must be the same
∑ αik = li ∀i ∈ Ncritical_load (12) Fig. 4 Illustrative example of node restoration in each structure (radial
k ∈ Mmaster
and mesh)
∑ αik = si ∀i ∈ Nslave (13) 3.2 Power flow constraints
k ∈ Mmaster
For each formed MG, the amount of active and reactive power
αik + αkj ≤ 1 ∀i ∈ N, j ∈ Nki , children, k ∈ Nmaster (14) flowing in the lines, as well as the amplitude of the voltages in the
bus, shall be specified. For this reason, an appropriate power flow
αik − αkj ≤ 0 ∀i ∈ N, j ∈ Nki , children, k ∈ Nmaster (15) model must be provided. In this study, the simplified linearisation
of DistFlow equations was used for power flow modeling [24]. The
rationale for choosing this model in our study is the lower power
3.1.2 Radial connectivity constraint: Constraint (16) guarantees consumption of the network after contingency and consequently,
that if node i, which does not belong to any loop/mesh, is restored lower losses in distribution lines than normal conditions. The type
by unit k, then its parent node must be restored by the same unit of power flow modelling is different for radial and mesh parts.
Load control for the active part of loads is considered at eight
αkj − αik ≤ 0 levels. To completely interrupt each load, the associated variable of
(16) load switch (li) must be set to 0. In this model, load control is
∀i ∈ N, j ∈ Nki , children, i, j ∉ Nloop, k ∈ Nmaster
applied to the reactive part of the load in accordance with the
power factor [1].
3.1.3 Mesh connectivity constraint: These constraints defined
for load restoration in potential loop or mesh. Constraint (17)
indicates that each node belonging to loop/mesh and restored by 3.2.1 Power flow constraints for radial parts: Constraints (20)–
unit k must be restored only through one path from unit k. Not (23) are defined for each node in a distribution system that does not
considered in this constraint are nodes in loop/mesh on which DER belong to any loop/mesh, therefore, the active/reactive power
is installed (e.g. node c) or nodes in loop/mesh that can be restored injected to each node is the sum of net consumed power of that
by DER are out of the loop before other nodes in the loop (e.g. node and power consumed by its children nodes. Thus, according
nodes a and e). Constraint (18) guarantees that if each node to the master–slave control principles, based on which the
belongs to loop/mesh and if among different paths, path θ is modelling is performed, the generation/consumption power of the
slave unit is added to the power flow equations of the node with
selected for its restoration by unit k, the variable αik must be set to 1 negative/positive sign as a constant term. Further details on the
for all nodes in path θ. The number of restoration paths and set of master–slave control technique as well as how to use it in the
nodes for each restoration path can be determined by applying problem modelling are presented in the Appendix. Using constraint
‘Yen's k-shortest path’ algorithm [23]. For example, in Fig. 4, two (24), the voltage of each node in the radial part will be obtained.
restoration paths are available to restore node {g}. The first path The voltage of each DER is set to the reference value (V 0k).
consists of {h, a, b, c} nodes and the second one consists of {f, e, d,
c} nodes. These sets and number of restoration paths are obtained
by applying the source node {c} and sink node {g} to the 3.2.2 Power flow constraints for loops: Constraints (25)–(27)
mentioned algorithm. Constraint (19) shows that some nodes in the express the power flow in nodes belonging to loop/mesh. To obtain
loop may restore using two paths, e.g. nodes {a, e} in Fig. 4. In the power and voltage at the loop nodes, we first need to write power
case of these nodes, we cannot use constraint (17). For these nodes, flow along each path and then select one of the powers and
voltages by using constraints (28)–(30). In constraints (25)–(26),
variable βik, θ depends on the same variable but in the children node
there are two types of children nodes for each node: (i) children
along path θ. For example, in Fig. 4, βe2, 1 depends on βd2 , 1 and βe2, 2
nodes along path θ, which is denoted by set Nki,,θchildren; for instance,
depends on β2f , 2 in Fig. 4, node {a} is the children node of node {b} along path one.
(ii) Children node, which is common in each path and is
αik = ∑ βik, θ ∀i ∈ Nloop, i ∉ Nloop_DER, k ∈ Nmaster
(17)
θ
θ∈ ℵk, i
i, t
pik, t = αik × PSup + ∑ pkj , t − si ⋅ αik × pSi, t. Gen (20)
αkj j∈ i
Nk,
βik, θ ≤ ∑ i, θ
children
i, θ
j ∈ Nk, parent
Nk, parent (18)
∀ i, j ∈ Nloop, θ ∈ ℵθk, i, k ∈ Nmaster
i, t
qik, t = αik × QSup + ∑ qkj , t − si ⋅ αik × qSi, t. Gen (21)
i
j ∈ Nk, children
βik, θ = βkj, θ B
(19) itγ
i, t i, t
∑ ργi t × Pcontrol (22)
, , ,
∀i ∈ Nloop_DER, θ ∈ ℵθk, i, j ∈ Nki,,θchildren, k ∈ Nmaster PSup = Pcritical −
γ=1
3754 IET Gener. Transm. Distrib., 2020, Vol. 14 Iss. 18, pp. 3750-3761
© The Institution of Engineering and Technology 2020
B
itγ
3.3 Operational constraints
i, t
QSup i, t
= Qcritical − ∑ ρit γ × Pcontrol
,
, ,
× tan φi
γ=1 (23) These constraints represent operational limits in each formed MG,
such as voltage limits and energy limits of EVs. Since the use of
∀i ∉ Nloop, j ∈ Nki , children, k ∈ Nmaster , t ∈ T energy stored in the PLs due to changes in the amount of available
energy and power at different hours requires planning to control the
ri j ptj + xi jqtj charging/discharging of the EVs, it is necessary to define the
V ik, t − V kj , t =
V 0k (24) appropriate constraints for optimal utilisation of PLs and other
resources.
∀ i, j ∉ Nloop, j ∈ Nki , children, k ∈ Nmaster , t ∈ T Constraint (35) indicates that all voltages must be within the
allowed range. In this research, the maximum permissible variation
k , i, t
pik, ,θt = Pcommon + pkj,,θt × βkj, θ for the voltage is 5%. Constraints (36) and (37) guarantee that
(25) active/reactive power generation of DG units is within the allowed
∀i ∈ Nloop, θ ∈ ℵθk, i, j ∈ Nki,,θchildren, k ∈ Nmaster , t ∈ T range.
Owing to the failure and repair rate of the DGs, the power
k , i, t
qik, ,θt = Qcommon + qkj,,θt × βkj, θ electronic instruments, and the availability of accessing the EVs,
(26) the two energy resources are always different in terms of
∀i ∈ Nloop, θ ∈ ℵθk, i, j ∈ Nki,,θchildren, k ∈ Nmaster , t ∈ T availability in the MG formation problem. To model this
difference, the availability factor should be defined for each of
ri j ptj, θ + xi jqtj, θ these resources. In general, the problem should be modelled in
V ik, ,θt − V kj,,θt = such a way that resources with a high availability factor are used
V 0k (27) more than other resources in the restoration stage.
∀ i, j ∈ Nloop, j ∈ Nki,,θchildren, θ ∈ ℵθk, i, k ∈ Nmaster, t ∈ T This availability factor is defined for the master resources and is
shown in constraints (36) and (37). This factor is determined for
DERs based on their lifetime performance.
pik, t = ∑ βik, θ × pik, ,θt − τik, p Constraint (38) guarantees that the energy generation of each
θ∈ ℵθk, i (28) DG in the restoration time does not exceed the amount of DG fuel
reserve. Constraint (39) guarantees that at each time, at most one
∀i ∈ Nloop, θ ∈ ℵθk, i, k ∈ Nmaster, t ∈ T step of the load control block is used.
Constraints (41)–(44) indicate a relation between network
variables such as active/reactive power generation of the nodes in
qik, t = ∑ βik, θ × qik, ,θt − τik, q which the PL is located and the total power generation by all EVs
θ∈ ℵθk, i
(29) in that PL. For charging and discharging control of each EV in the
∀i ∈ Nloop, k ∈ Nmaster, t ∈ T PL at the emergency time, binary variables nch n, t
e
and ndch
n, t
e
are used,
respectively. These variables are set to 1 for charging/discharging
and 0 for not charging/discharging. In this study, it is assumed that
each EV charger has a power correction factor to regulate reactive
V ik, t = ∑ βik, θ × V ik, ,θt
power. Similarly, due to the non-linear relationship between the
θ ∈ ℵθk, i (30)
active, reactive, and apparent power of each battery in (40), to
∀i ∈ Nloop, ∀i ∉ NloopDER, k ∈ Nmaster , t ∈ T avoid using non-linear equations, we assume that the apparent
power of each EV is such that it can generate reactive power up to
50% of its active power (see constraints (42) and (44)) [26, 27].
τik, p = ∑ βik, θ − 1 × Pcommon
k, i, t Here, constraints (41) and (42) are for the master units, and
θ∈ ℵθk, i
(31) constraints (43) and (44) are for the slave units
∀i ∈ Nloop_DER, k ∈ Nmaster , t ∈ T
k , i, t
Qcommon i, t
= αik × QSup + ∑ qyk, t − si ⋅ αik × qSi, t. Gen
i
y ∈ Nk, (34)
τik, q = ∑ βik, θ − 1 × Qcommon
k , i, t common
θ ∈ ℵθk, i
(32) ∀i ∈ Nloop−DER, k ∈ Nmaster , t ∈ T
∀i ∈ Nloop_DER, k ∈ Nmaster , t ∈ T
0.95 × V 0k ≤ V ik, t ≤ 1.05 × V 0k ∀i ∈ N, k ∈ Nmaster , t ∈ T (35)
k , i, t
Pcommon i, t
= αik × PSup + ∑ pyk, t − si ⋅ αik × pSi, t. Gen 0 ≤ pkk, t ≤ Akt × pkmax DG
∀k ∈ Nmaster ,t ∈ T (36)
y∈ i
Nk, common
(33)
∀i ∈ Nloop_DER, k ∈ Nmaster , t ∈ T
DG
0 ≤ qkk, t ≤ Akt × qkmax ∀k ∈ Nmaster ,t ∈ T (37)
t = tre
denoted by set Nki , common; e.g. in Fig. 4, node {x} is the common DG
∑ pkk t ≤ ,
T × CAPK ∀k ∈ Nmaster (38)
children node of node {f}. Constraints (31) and (32) express that if t = trs
each node in the loop belongs to NloopDER, the nodes may be
restored through several paths. Therefore, if powers determined ∑ ρit γ ≤ 1 ∀i ∈ Ncritical_load, t ∈ T
from (25) and (26) are applied to (28) and (29), the common parts
, (39)
γ∈B
(i.e. injected power to common children nodes, load consumed in
that node, slave generation in that node) are computed several
(40)
max, accessable max 2
max 2
times and for each path. To solve this problem, the additional parts Qdch = Sdch − Pdch
must be eliminated. Variables τik, p and τik, q are used in (28) and (29) m, t
to eliminate additional parts of active and reactive power, Pdch
respectively. The common active/reactive power generated/
pkk, t = ∑ m, t
(ndch e
ηdch
e m, t m, t
− nch P × ηch)
e che
m∈ k
MEV (41)
consumed in each node determined by (33) and (34). Here, non-
PL
linear terms, given their non-linearity, are linearised by ‘Big M’ ∀k ∈ Nmaster , t∈ T
method [25].
IET Gener. Transm. Distrib., 2020, Vol. 14 Iss. 18, pp. 3750-3761 3755
© The Institution of Engineering and Technology 2020
m, t
Pdch pkk, t + ∑ αik × pSi, t. Gen = ∑ i, t
αik × PSup
qkk, t ≤ 0.5 × ∑ m, t
ndch e
ηdch
e m, t m, t
− nch P × ηch
e che i ∈ Nslave i ∈ Ncritical_load (54)
k
m ∈ MEV (42)
∀k ∈ Nmaster , t ∈ T
PL
∀k ∈ Nmaster , t∈ T
qkk, t + ∑ αik × qSi, t. Gen = ∑ i, t
αik × QSup
m, t
Pdch i ∈ Nslave i ∈ Ncritical_load (55)
pSi, t. Gen = ∑ m, t
ndch dch
e m, t m, t
− nch P × ηch
e che ∀k ∈ Nmaster , t ∈ T
(43)
e
i
m ∈ MEV
η
∀i ∈ Nslave, t ∈ T To evaluate the system performance after a power outage, the
‘resilience index’ has been used. This metric which is introduced in
m, t
Pdch [1], is the ratio of the supplied energy of the loads during the study
qSi, t. Gen ≤ 0.5 × ∑ m, t
ndch
ηdch
e m, t m, t
− nch P × ηch
e che
period to the total energy of the system loads and calculated using
(44)
e
m∈ i
MEV constraint (56)
∀i ∈ Nslave, t ∈ T i, t
∑t ∈ T ∑i ∈ Ncritical_load PSup × li
Resilience index = (56)
Constraints (45)–(47) indicate operational limits of charging/ i, t
∑t ∈ T ∑i ∈ Ncritical_load Pcritical × li
discharging power and SOC of each EV in restoration. The
availability factor for all EVs is also shown in constraints (45) and
3.4 Objective function
(46)
In general, in most of the resilience schemes two approaches are
m, t
0 ≤ Pche
m, max
≤ Ami, t × Pch m, t
⋅ nch e
used to determine the objective function.
(45)
k i PL
∀m ∈ MEV ∪ MEV, k ∈ Nmaster , i ∈ Nslave, t ∈ T (i) Technical approach: In this approach, maximum load
restoration and minimum use of demand response control blocks
m, t
0 ≤ Pdch m, max
≤ Ami, t × Pdch m, t
⋅ ndch are selected as objective functions. According to (57), after solving
e
the problem and maximising the supplied load, both of the
e
(46)
PL
mentioned objective functions are optimised. It is assumed that by
k i
∀m ∈ MEV ∪ MEV , k ∈ Nmaster , i ∈ Nslave, t ∈ T
providing appropriate incentive schemes by the network operator,
SOCm, min ≤ SOCm,t
e ≤ SOC
m, max economic aspects such as minimising the generation cost of the
(47) DGs and maximising the EV owner's profit are not a top priority. In
k i PL
∀m ∈ MEV ∪ MEV , k ∈ Nmaster , i ∈ Nslave, t ∈ T this approach, maximising the amount of supplied load is the top
priority
Constraints (48)–(50) describe the relation of SOC at each time slot
and determine the initial and final desired SOC. Constraint (51) Max: Obj F = ∑ ∑ i, t
PSup × li (57)
guarantees that charging and discharging does not occur t ∈ T i ∈ Ncritical_load
simultaneously (ii) Economic approach: In this approach, the system operator aims
to minimise the overall cost of the system and also maximise the
m, t
Pdch EV owner's profit when participating in the CLs restoration.
SOCm
e
, t+1
= SOCm ,t m, t ch
e + Pche η −
e
Δt (48) Therefore, the objective function consists of three terms; the first
ηdch
term is the CLs interruption cost, the second term is the generation
m, arr m, arr cost of the DGs, and the third term is the EV owner's cost
SOCm
e
, trs + Δt
= SOCm
n
, trs + Δt
(49)
SOCe
m, tm, dep
≥ SOCm, min_desired Min: Obj F = ∑ ∑ i, t
(Pcritical i, t
− PSup ) × IC t
CL, i
(50) t ∈ T i ∈ Ncritical_load
k i PL
∀m ∈ MEV ∪ MEV , k ∈ Nmaster , i ∈ Nslave, t ∈ T + ∑ pkk, t × GC t
DG, k
(58)
DG
k∈ Nmaster
m, t m, t
nche
+ ndch e
≤1
(51)
k
∀m ∈ MEV i
∪ MEV PL
, k ∈ Nmaster , i ∈ Nslave, t ∈ T +∑ ∑ m, t m, t
(nche che
m, t m, t
P × Ct − ndch P × Rt )
e dche
i ∈ N m ∈ MEV
i
Table 4 DG parameters
DG Node Pmax, kW Qmax, kVar CAP, kWh
1 44 129.52 89.02 919.12
2 60 101.63 152.89 550.23
3 86 92.63 54.56 895.29
Table 6 Location and the maximum amount of modified outage duration in all cases is assumed to be 11 h. According to
critical loads consumption in the selected scenario [31], most outages last <24 h. Therefore, 11 h is a logical duration
outage in typical disasters.
Node Pmax, kW Qmax, kVar
9 78.6 18.6 4.1 Case study 1
17 29.32 14.34
27 66.97 36.94 In this case, it is assumed that the incident occurs in midday at
16:00, parking operates normally before the incident occurs, and
30 61.2 30.37
charging/discharging is controlled by the aggregator according to
37 18.21 31.6 the electricity tariff. In this case, the technical approach is
46 42.5 27.9 considered in selecting the objective function. The formed MG in
66 30.5 25.9 this case study is shown in Fig. 6. As predicted and shown in
94 30.12 21.3 Fig. 7a, the supplied load is increased substantially during the early
87 40.1 66.84 hours when using EVs for restoration. As we approach the end of
79 21.84 63.78 the restoration time, the participating percentage of EVs in PLs is
reduced due to EVs leaving the PL. Generally, when resources
101 50.56 19.87
(EVs’ batteries) are available for a limited time, the problem is
solved in a way that limited access resources charge other
resources which their presence is longer. For example, in Fig. 7b,
$10/kWh [30]. Also, it is assumed that the probability of the in the first two hours, the RP charging rate in MG-1 is much higher
presence of any type in the PLs is the same. In cases 1 and 2, the than the other hours, because, after the incident, which happened at
impact of the EVs penetration rate and different charging modes of 16:00, these EVs have more time than the EVs in PL. So, in the
EV on the amount of the supplied loads are investigated, PLs, EVs that want to leave in the next few hours and probably
respectively. In case 3, the impact of the PLs participation on the have a high amount of stored energy, discharged for supplying
number of supplied loads in the distribution system is investigated. loads and charging the EVs that have recently entered RP and need
As mentioned in the above section, the purpose of designing case to stay for a long time. Moreover, as shown in Figs. 7c and d, using
studies 1–3 is based on studying the technical aspects of EVs’ load control in the nodes, and also increasing EVs rate of
participation, and it is assumed that due to the implementation of penetration, the supplied load in the restoration stage is improved
appropriate incentive plans and the high importance of existing considerably. The resilience index when using and non-using of the
loads, there are no economic restrictions on the available energy of EVs is 0.82 and 0.56, respectively. As a result and as expected EVs
the EVs and DGs. In case 4, the economic benefits of EVs’ participation improve the system resilience after the outage.
participation are considered in selecting the objective function. The
IET Gener. Transm. Distrib., 2020, Vol. 14 Iss. 18, pp. 3750-3761 3757
© The Institution of Engineering and Technology 2020
restoration time is investigated. For this purpose, the simulation
was performed with three EVs charging modes: controlled
charging with V2G and with cost minimisation objective (mode I);
controlled charging without V2G and uniform charging during
charging period (mode II); uncontrolled charging without V2G and
with a maximum charging power (mode III). In this case, to show
the impact of the charging scheduling, the time of the incident is
assumed to be 21:00. It is of note that the energy available in PLs
at this time is considered 0 due to the absence of EVs. Moreover, in
this case, the technical approach is considered in selecting the
objective function. The amount of available energy of each RP
under three charging modes is given in Table 8. As can be seen, the
amount of energy available in the first mode is the lowest among
the three modes. This minimum level of energy is explained by the
fact that in the first mode, the energy price is high in the hours
leading up to the incident so that the EVs will not be charged
during these hours and will discharge their stored energy as much
as possible. In the two first modes, the formed MGs are the same as
case study 1, but in mode III, it is different. Figs. 8a and b show the
results of the MGs formation under operation in mode III and a
comparison of the supplied loads in the three different charging
modes, respectively. The resilience index in this case and for three
charging modes is 0.84, 0.87, and 0.95, respectively.
IET Gener. Transm. Distrib., 2020, Vol. 14 Iss. 18, pp. 3750-3761 3759
© The Institution of Engineering and Technology 2020
DGs’ power outputs and the power consumption by critical loads
through smart metering, and (iii) providing the means of enabling
the development and use of V2G technologies.
6 Conclusion
In this study, we attempted to restore critical loads through
dynamic MG formation using EV PLs and DGs after a natural
disaster. To do this optimally, several factors are required; e.g. the
use of a proper control method to coordinate resources within each
MG, the use of load control to assist the operator in enhancing
system performance in the restoration stage, and the use of load
restoration method in the mesh network. We satisfied all three
cases using the master–slave control method, applying load control
blocks and employing a novel proposed method for restoration in
the mesh grid, respectively. It is demonstrated that the use of EVs
increases the supplied load on restoration time significantly. In
particular, when there are PLs and RP in the formed MG, at any
given hour of the day that incident occurs, it can be expected that
the MG will be able to use significant energy from discharging
EVs to supply its loads. According to the resilience index, the use
of EVs in the MG formation problem has increased the energy
supplied to the critical loads by ∼45%. However, as has been
shown in case study 2 and previously stated, the use of proper
charge scheduling before the power outage, as a preventive
approach of enhancing the resilience, can increase the energy
supplied to the critical loads by up to 70%. Another advantage of
using PLs is that in the case that the damage of the network is very
severe, and access to DGs is not possible. It was shown that each
PL with frequency and voltage control capability could supply
local loads under such conditions. In this study, the uncertainty of
EVs and loads were modelled by the random sampling method.
Moreover, in case study 4, it was concluded that using the
economic approach in the objective function increases the
participation of EVs in the load restoration, and also reduces the
Fig. 10 Comparison of considering a different approach in case study 4 fuel consumption of the DGs.
(a) Charge/discharge power of RP3, (b) Generation power of DG-2
7 References
Table 9 Economic results of the simulation under
[1] Mousavizadeh, S., Haghifam, M.R., Shariatkhah, M.H.: ‘A linear two-stage
considering a technical and economical approach method for resiliency analysis in distribution systems considering renewable
Technical Economic energy and demand response resources’, Appl. Energy, 2018, 211, pp. 443–
approach approach 460
[2] Che, L., Khodayar, M., Shahidehpour, M.: ‘Only connect: MGs for
total interruption cost of CLs, $ 12,450 12,730 distribution system restoration’, IEEE Power Energy Mag., 2014, 12, (1), pp.
total generation cost of DGs, $ 1128 996 70–81
[3] Brown, R.: ‘Cost-benefit analysis of the deployment of utility infrastructure
EV owners’ profit, $ 164 207 upgrades and storm hardening programs prepared for: table of contents’,
Satell. Commun., 2009, 3021, pp. 919–961
[4] Bie, Z., Lin, Y., Li, G., et al.: ‘Battling the extreme: a study on the power
system resilience’, Proc. IEEE, 2017, 105, (7), pp. 1253–1266
improvement method for weather-based disasters, and general [5] Poudel, S., Dubey, A.: ‘Critical load restoration using distributed energy
behaviour of EVs owner is that they want to be at their home and resources for resilient power distribution system’, IEEE Trans. Power Syst.,
do not go out in harsh situation (weather-based disasters). So, the 2019, 34, (1), pp. 52–63
travelling cost is not considered in this study. It is to be noted that [6] Li, J., Ma, X.Y., Liu, C.C., et al.: ‘Distribution system restoration with MGs
using spanning tree search’, IEEE Trans. Power Syst., 2014, 29, (6), pp.
in some disasters, the severity of event is very high and in those 3021–3029
occasions, people usually leave their living place before disaster [7] Zhang, M., Chen, J.: ‘Islanding and scheduling of power distribution systems
start. However, as explained earlier, the number of these kinds of with distributed generation’, IEEE Trans. Power Syst., 2014, 30, (6), pp.
disasters is very low and we can assume that in those occasions 3120–3129
[8] Gao, H., Chen, Y., Xu, Y., et al.: ‘Resilience-oriented critical load restoration
people usually have tendency to be at their home. In practice, after using MGs in distribution systems’, IEEE Trans. Smart Grid, 2016, 7, (6), pp.
receiving weather forecast information and determining the 2837–2848
charging scheduling for the EVs, which participated in the V2G [9] Chen, C., Wang, J., Qiu, F., et al.: ‘Resilient distribution system by MGs
emergency programme, a large number of EVs stay in the PL in the formation after natural disasters’, IEEE Trans. Smart Grid, 2016, 7, (2), pp.
958–966
early hours of the incident until conditions return to normal. [10] Ding, T., Lin, Y., Bie, Z., et al.: ‘A resilient MG formation strategy for load
Therefore, EVs have the considerable available energy for injecting restoration considering master slave distributed generators and topology
into the grid due to the use of charging modes II and III. However, reconfiguration’, Appl. Energy, 2017, 199, pp. 205–216
we have assumed in this study that EVs behave normally. This [11] Zadsar, M., Haghifam, M.R., Larimi, S.M.M.: ‘Approach for self-healing
resilient operation of active distribution network with MG’, IET Gener.
assumption makes us consider the worst-case scenario, in terms of Transm. Distrib., 2017, 11, (18), pp. 4633–4643
available energy, at the restoration time. Therefore, the authors [12] Han, S., Soo, H.H., Sezaki, K.: ‘Design of an optimal aggregator for vehicle-
believe that the available energy is higher than the value stated in to-grid regulation service’. Innovative Smart Grid Technologies Conf. (ISGT
the presented case studies. 2010), Gothenburg, Sweden, 2010, vol. 1, pp. 65–72
[13] Mu, Y., Wu, J., Ekanayake, J., et al.: ‘Primary frequency response from
On the other hand, the restoration of the critical loads through electric vehicles in the Great Britain power system’, IEEE Trans. Smart Grid,
the MG formation is done in the context of the smart grid. 2013, 4, (2), pp. 1142–1150
Important aspects of the smart grid used in the MG formation [14] Liu, H., Hu, Z., Song, Y., et al.: ‘Decentralized vehicle-to-grid control for
problem are: (i) the two-way communication between remote primary frequency regulation considering charging demands’, IEEE Trans.
Power Syst., 2013, 28, (3), pp. 3480–3489
control switches and outage management system, (ii) monitoring of
3760 IET Gener. Transm. Distrib., 2020, Vol. 14 Iss. 18, pp. 3750-3761
© The Institution of Engineering and Technology 2020
[15] Wu, C., Mohsenian-Rad, H., Huang, J., et al.: ‘PEV-based combined
frequency and voltage regulation for smart grid’. 2012 IEEE PES Innovative
Smart Grid Technologies (ISGT 2012), Washington, DC, USA, 2012
[16] Xu, N.Z., Chung, C.Y.: ‘Reliability evaluation of distribution systems
including vehicle-to-home and vehicle-to-grid’, IEEE Trans. Power Syst.,
2016, 31, (1), pp. 759–768
[17] Farzin, H., Fotuhi-Firuzabad, M., Moeini-Aghtaie, M.: ‘Reliability studies of
modern distribution systems integrated with renewable generation and
parking lots’, IEEE Trans. Sustain. Energy, 2017, 8, (1), pp. 431–440
[18] Liu, Z., Wang, D., Jia, H., et al.: ‘Aggregation and bidirectional charging
power control of plug-in hybrid electric vehicles: generation system adequacy
analysis’, IEEE Trans. Sustain. Energy, 2015, 6, (2), pp. 325–335
[19] Guner, S., Ozdemir, A.: ‘Stochastic energy storage capacity model of EV
parking lots’, IET Gener. Transm. Distrib., 2017, 11, (7), pp. 1754–1761
[20] Vagropoulos, S.I., Bakirtzis, A.G.: ‘Optimal bidding strategy for electric
vehicle aggregators in electricity markets’, IEEE Trans. Power Syst., 2013,
28, (4), pp. 4031–4041
[21] Conejo, A.J., Carrión, M., Morales, J.M.: ‘Decision making under uncertainty
in electricity markets’ (Springer, USA, 2010)
[22] Wu, C., Mohsenian-Rad, H., Huang, J.: ‘Vehicle-to-aggregator interaction
game’, IEEE Trans. Smart Grid, 2012, 3, (1), pp. 434–442
[23] Yen, J.Y.: ‘Finding the K shortest loopless paths in a network’, Manage. Sci.,
1971, 17, (11), pp. 712–716
[24] Yeh, H.G., Gayme, D.F., Low, S.H.: ‘Adaptive VAR control for distribution
circuits with photovoltaic generators’, IEEE Trans. Power Syst., 2012, 27, (3),
pp. 1656–1663
[25] Winston, W.L., Venkataramanan, M., Goldberg, J.B.: ‘Introduction to
mathematical programming’ (Thomson/Brooks/Cole Duxbury, Pacific Grove,
CA, 2003)
[26] Abessi, A., Vahidinasab, V., Ghazizadeh, M.S.: ‘Centralized support Fig. 11 Information flow in master–save control technique
distributed voltage control by using end-users as reactive power support’,
IEEE Trans. Smart Grid, 2016, 7, (1), pp. 178–188
[27] Abessi, A., Zakariazadeh, A., Vahidinasab, V., et al.: ‘End-user participation
in a collaborative distributed voltage control and demand response
programme’, IET Gener. Transm. Distrib., 2018, 12, (12), pp. 3079–3085
[28] Moeini-Aghtaie, M., Abbaspour, A., Fotuhi-Firuzabad, M.: ‘Online
multicriteria framework for charging management of PHEVs’, IEEE Trans.
Veh. Technol., 2014, 63, (7), pp. 3028–3037
[29] ‘evspecifications’. Available at https://www.evspecifications.com, accessed 3
September 2019
[30] Yao, S., Wang, P., Zhao, T.: ‘Transportable energy storage for more resilient
distribution systems with multiple microgrids’, IEEE Trans. Smart Grid,
2019, 10, (3), pp. 3331–3341
[31] Wang, Y., Chen, C., Wang, J., et al.: ‘Research on resilience of power systems
under natural disasters – a review’, IEEE Trans. Power Syst., 2016, 31, (2),
pp. 1604–1613
[32] Tenti, P., Caldognetto, T.: ‘Master/slave power-based control of low-voltage
microgrids’ (Elsevier Inc., UK, 2017)
[33] Fusheng, L., Ruisheng, L., Fengquan, Z.: ‘Composition and classification of
the microgrid’ (Academic Press, 2016) Fig. 12 Power flow in the node on which the slave unit is located
8 Appendix frequency set point for master unit-1 to control the voltage and
frequency of the MG-1 (set point for U/f control), and (ii) current
set point as power leveling signal for slave unit-1 to adjust the
One of the most commonly control strategies, which is used for a power balance in the MG-1 (set point for P/Q control).
small MG is master–slave technique. In this technique, different The following are considered in the utilisation of master–slave
resources in the MG are controlled with different modes (in the control technique in the MG formation problem:
island operation, one generation unit act as a master unit and uses
U/f control, and other generation units act as slave units and use (i) Since the voltage of the master units is controlled, the voltage of
P/Q control). It should be noted that in the grid-connected mode all these units is always constant and equal to the base value. On the
generation units use P/Q control. The master unit controller keeps other hand, the voltage of the slave units can vary in the
the bus voltage constant, and has the ability to respond fast to the permissible range (±5%).
load fluctuations. Therefore, master unit can control the voltage (ii) Since the output power of the slave units is controlled, a new
and frequency of the MG. On the other hand, the slave units act as term added to the power flow equations in the nodes, in which
current driven source and follow the voltage and frequency of the slave units are located at each hour. This term is the hourly
MG. The slave units control their output power by following the constant active/reactive power injected to the mentioned nodes by
current pattern, which is sent from the master unit. Therefore, the slave units. When the slave units generate the active/reactive
master unit is responsible for coordination of slave units in the MG power, this term is considered with a negative sign (see Fig. 12),
[32, 33]. As shown in Fig. 11, after formation of the MGs, the and when the slave units consumed the active/reactive power, this
master controller receive measurement data such as voltage of the term is considered with a positive sign in the equation and slave
buses, current of the lines, available active and reactive power of units act as loads (see Fig. 12). For example, if the slave unit is a
the slave units, and load consumption data. Then the master PL, in charging mode of the PL, the related sign is positive and in
controller sends two types of control commands: (i) voltage and the discharging mode, the related sign is negative.
IET Gener. Transm. Distrib., 2020, Vol. 14 Iss. 18, pp. 3750-3761 3761
© The Institution of Engineering and Technology 2020