IET Generation Trans Dist - 2019 - Wu - Microgrid Planning Considering The Resilience Against Contingencies

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

IET Generation, Transmission & Distribution

Research Article

Microgrid planning considering the resilience ISSN 1751-8687


Received on 13th November 2018

against contingencies
Revised 11th May 2019
Accepted on 31st May 2019
E-First on 19th July 2019
doi: 10.1049/iet-gtd.2018.6816
www.ietdl.org

Xiong Wu1 , Zhao Wang1, Tao Ding1, Xiuli Wang1, Zhiyi Li2, Furong Li3
1School of Electrical Engineering, Xi'an Jiaotong University, Shaanxi 710049, People's Republic of China
2Robert W. Galvin Centre for Electricity Innovation, Illinois Institute of Technology, 10 West 35th Street, Chicago, USA
3Department of Electronic and Electrical Engineering, University of Bath, BA2 7AY, Bath, UK

E-mail: wuxiong@mail.xjtu.edu.cn

Abstract: Increasing the redundancy of distribution lines and increasing the penetration of distributed energy resources (DERs)
both help microgrids ride through contingencies. However, it remains a challenging problem for how to coordinate these two
measures for minimising the deployment cost while guaranteeing a pre-specified degree of resilience in case of contingencies.
Accordingly, this study proposes a novel microgrid planning model to site and size candidate sets of DERs and distribution lines
in close coordination, which is mathematically equivalent to a two-stage robust optimisation problem. In particular, the resilience
level of microgrid operations is quantified and maintained such that the load loss is constrained within a given bound under any
realisation of N–k contingencies. The proposed model also incorporates a practical strategy to maintain the radial topology of
islanding network sections in any N–k contingency. Finally, numerical experiments based on two microgrid test systems are
performed. The results show that the system resilience is adaptively enhanced through optimally placing DERs and distribution
lines compared with the conventional economics based model. Moreover, the employed robust method is at least ten times
faster than the reliability-based method in identifying the worst contingency.

Nomenclature Q reactive power (kVar)


ρ cost per unit ($/kWh)
Sets r interest rate
G set of distributed generators (DGs) E energy capacity of BSs (kWh)
ℬ set of battery storage systems (BSs) η efficiency
W set of renewable energy resources R line resistance (Ω)
N set of buses X line reactance (Ω)
N0 set of root buses M sufficiently big number
ℒ set of existing distribution lines S resilience level
ℒc set of candidate distribution lines
Decision variables
Indices x binary variable indicating the deployment state of candidate
b script for BSs DERs
g script for DGs z binary variable indicating the deployment state of candidate
w script for renewable energy resources distribution lines
d script for days α binary variable indicating the active state of existing
h script for hours distribution lines in a contingency
c script for candidate distribution lines β binary variable indicating the active state of backup
dl script for distribution lines distribution lines in a contingency
ld script for loads p variable associated with active power (kW)
lc script for curtailed loads q variable associated with reactive power (kVar)
gd script for the utility grid e stored energy of BSs (kWh)
in script for injection v voltage magnitude of buses (pu)
+, − script for export/import l power flow in the fictional network (pu)
( ⋅ ), ( ⋅ ) upper/lower bounds s power injection at source nodes in the fictional network (pu)
u binary variable indicating the active state of source nodes in
π( ⋅ ), σ( ⋅ ) parents/children of buses the fictional network
(⋅) variables in the resilience model
⋅ number of elements
1 Introduction
Parameters and constants Microgrids, as localised intelligent distribution systems, utilise on-
site distributed energy resources (DERs) to meet the load demand.
K present-worth coefficient Owing to their remarkable advantages of controllability and
I investment cost ($) flexibility, microgrids are gaining increasing popularity in power
IE energy-related investment cost of BSs ($/kWh) engineering practises. In addition to the common concern over the
IP power-related investment cost of BSs ($/kW) large budget of implementing microgrid projects, microgrids may
T life span of DERs (year) even suffer from prolonged outages when confronted with
P active power (kW) unexpected disruptions due to man-made faults or natural disasters.

IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 16, pp. 3534-3548 3534
© The Institution of Engineering and Technology 2019
How to maintain the resilience of microgrids, as referred to as the hardening existing distribution lines, deploying distributed
load restoration ability after an outage, is, therefore, a practical generators (DGs), and automatic switches. Nevertheless, the model
issue to be addressed adequately at the planning stage. On the one mainly aims at distribution system, and the islanded cases caused
hand, the deployment of more DERs will contribute to the load by contingencies are not considered. Nazar and Heidari [21]
restoration ability of microgrids as they are able to restore loads develop an optimal robust microgrid expansion planning model to
directly in case of contingencies. On the other hand, the determine the deployment of distribution lines and DGs. The
configuration of redundant distribution lines provides an additional resilience is indirectly reflected in the objective function in terms
means to supply power to customers on the outage. Given that both of energy not supplied (ENS) cost, and the system resilience is not
strategies have a distinct budget and operational requirements, it is adequately characterised. Madathil et al. [22] conducts an N−1
of practical significance to consider them in close coordination in contingency analysis on off-grid microgrids. In fact, more
order to strike a trade-off between economics and resilience of contingencies are needed to be considered from the perspective of
planning microgrids. robust planning.
The optimal planning and expansion of microgrids have been In summary, conventional microgrid planning models [1–9] are
extensively studied in the existing work. Most of them construct mostly based on economic or reliability criterion. Additionally, the
the DERs sizing and placement models in interconnected or planning objects are usually DERs, and the role of distribution line
isolated microgrids following the economic or reliability criterion expansion in enhancing the microgrid resilience has not been
[1–3]. For example, Lotfi and Khodaei [2] propose a microgrid adequately investigated. Recently, an increasing number of studies
planning model, which aims at minimising the total investment and [23–28] take the resilience criterion into account in the planning
operation cost, to determine the optimal size and mix of distributed model by deploying distribution lines and DGs. Nevertheless,
generation. Mitra et al. [3] determine the optimal size and location many works are not aimed at microgrids. Additionally, some
of distributed generation in a microgrid according to the reliability details such as islanded cases in a contingency and the explicit
criterion. On the basis of the basic microgrid planning framework, description of resilience level are commonly ignored. Accordingly,
many factors are taken into account in the pertinent studies, e.g. the this paper proposes a microgrid planning model considering the
uncertainties of loads, renewable energy generation, and electricity resilience against contingencies. To sum up, the main contributions
prices [4, 5], the multiple energy integration [6, 7], and the of this paper are:
coordinated operation of multi-microgrids [8, 9].
Recently, the resilience of power grids has attracted the (i) A tri-level two-stage robust microgrid planning model from the
attention of some researchers. With regards to microgrids, three perspectives of economics and resilience is proposed.
categories of resiliency enhancing strategies are studied in the (ii) The resilience level of microgrids is quantified and maintained
literature. (i) Preventative strategies to hedge against possible such that the load loss is constrained within a given bound under
natural disasters or man-made faults, investigating the microgrids any realisation of N–k contingencies.
planning or formation to increase the system resiliency against (iii) A practical strategy is developed to maintain the radial
contingencies [10]. (ii) Resilient operation after blackouts caused topology of sub-systems of a microgrid in each N–k contingency.
by extreme weather conditions or human errors, focusing on the
resilient operation to restore the systems either through The remainder of this paper is laid out as follows. Detailed
rescheduling the DERs or reconfiguring the network [11]. (iii) optimisation models are provided in Section 3. Algorithms
Resilient control strategies for a system recovery of frequency and proposed for solving the models are presented in Section 4. Case
voltage to maintain the security of supply [12]. This paper, studies are illustrated in Section 5, and finally, conclusions are
however, focuses on the resilient planning of microgrids given in Section 6.
considering the contingencies, which belong to the scope of
preventative strategies against contingencies. Regarding
contingency analysis, N–k contingency is widely used in the power
2 Proposed microgrid planning model
system to analyse the system security and resilience. When In this section, the basic microgrid planning model is formulated
integrated into a system operation or expansion planning problem, first taking into account the costs of both DERs and distribution
heuristic method [13], exhaustive method [14], and robust method lines. Furthermore, we extend the model by exploring the
[15, 16] are usually employed to solve it. Comparatively speaking, formulations for characterising the resilience of microgrids against
the robust method gains more attention in recent years for its contingencies. The radial topology formation of microgrids in an
advantage in quickly and precisely identifying the worst N–k contingency is also analysed and integrated into the model.
contingencies. For example, Wang et al.[15] develop an N–k Finally, the model is formulated as a robust microgrid planning
contingency-constrained unit commitment model and use a robust model and written in a compact form.
method to analyse contingencies. Moreira et al. [16] adopt the
same method to address the contingency-constrained transmission 2.1 Basic planning model
expansion planning problem. In this paper, we conduct the N–k
contingency analysis using the robust method. With respect to microgrid planning, the most common concern is
In fact, most existing studies focus on the resilient operation of related to economics. The total cost of implementing a microgrid
microgrids but few of them model the microgrid resilience against includes the investment cost and operation cost of DERs and the
contingencies explicitly at the planning stage. Zhang et al. [17] investment cost of distribution lines. Accordingly, the objective of
propose an optimal sizing and siting model for the battery storage the microgrid planning problem, as formulated in (1), is to
system (BS) and photovoltaic (PV) generation to improve the minimise the overall investment and operation cost (including the
system resilience. This paper mainly concerns the deployment of DER investment cost, the distribution line investment cost, and the
DERs and ignores the role of distribution line in enhancing the microgrid operation cost) throughout the life cycle. Since the life
system resilience. Ma et al. [18] propose an optimal hardening cycle of a microgrid lasts for decades, the total cost is calculated in
strategy with distribution lines to enhance the resilience of the terms of the present-worth value.
power distribution system against extreme weather events. Zare- The first term in the objective is the investment cost of DGs
Bahramabadi et al. [19] present a resilience-based framework for such as microturbines and fuel cells. The investment cost of DGs is
determining optimal switch placement in distribution systems. assumed to be proportional to their capacity with the consideration
However, Ma et al. [18] and Zare-Bahramabadi et al. [19] of discounted costs. The second term represents the investment
contribute to the system resilience merely by installing new cost of renewable energy resources such as wind turbines (WTs)
distribution lines or switches without realising the value of DERs and PVs. The third term is the investment cost of BSs. Note that
in resilience enhancement. In contrast, some works [20, 21] use the capital cost of BSs includes two parts. One is relevant to the
both distribution line and DER to enhance the system resiliency. capacity, while the other one depends on the maximum discharging
Specifically, Ma et al. [20] develop a mixed integer model to power. The fourth term denotes the investment cost of distribution
protect distribution grid against extreme weather events by lines. The remaining terms are the operation cost of the microgrid

IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 16, pp. 3534-3548 3535
© The Institution of Engineering and Technology 2019
over the year. The generation costs of WTs and PVs are assumed to In addition, the power output of DERs is constrained by their
be zero as they consume no fuel. The fifth term is the fuel cost of availability, which means that only deployed DERs can generate
DGs. The sixth and seventh terms are related to the energy traded power as requested. The corresponding constraints are shown in
with the utility grid. The microgrid is assumed to either buy or sell the equations below:
power from or to the utility grid
g
0 ≤ pigjdh ≤ P¯ i xigj, ∀i ∈ G, ∀ j ∈ N (12)
min ∑ KigIigP¯ gi xig + ∑ KiwIiwP¯ wi xiw
i∈G i∈W g
0 ≤ qigjdh ≤ Q¯ i xigj, ∀i ∈ G, ∀ j ∈ N (13)
b b
+ ∑ Kib(IPibP¯ i + IEibE¯ i )xib + ∑ K dlIidlj zi j
i∈ℬ (i, j) ∈ ℒc (1) b
0 ≤ pibjdh
+
≤ P¯ i xibj, ∀i ∈ ℬ, ∀ j ∈ N (14)
365 24
g gd + gd −
+∑ ∑(∑ ρig pidh + ρgd + pdh − ρgd − pdh ) b
d =1h=1 i∈G 0 ≤ pibjdh

≤ P¯ i xibj, ∀i ∈ ℬ, ∀ j ∈ N (15)

In this objective, the investment cost is converted to the annual 0 ≤ piwjdh ≤ Piw xiwj , ∀i ∈ W, ∀ j ∈ N (16)
value at present by multiplying the present-worth coefficient as
follows: Equation (17) characterises the energy storage of BSs, where the
increased energy storage level is the difference between the
r(1 + r)T y, i charging and discharging powers after considering the pertinent
K y, i = , ∀y ∈ G, W, ℬ, ℒc (2)
(1 + r)T y, i − 1 efficiency. The energy storage level, the charging, and discharging
powers are restricted by their physical limits, as stated in (18)–(20),
where Ty,i is the life span of the ith device y. The operation cost is respectively. Equation (21) enforces the stored energy at the end of
daily based and then accumulated throughout the year. Considering the scheduling period to be equal to that at the beginning so as to
the temporal correlation, several typical days instead of all days are facilitate the scheduling in the next period
selected to simulate the daily operation of the microgrid to lessen
the computational complexity of the model.
b
eidh b
= eid b+ b+
, h − 1 − pidh /η
b− b−
+ pidh η , ∀i ∈ ℬ, ∀d, ∀h (17)
The numbers of DGs, BSs, and renewable energy resources are
subject to resource limits such as financial budgets, and they are b+
0 ≤ pidh ≤ P¯ i xib,
b
∀i ∈ ℬ, ∀d, ∀h (18)
bounded by their maximum values, as stated in the equations
below: b
b−
0 ≤ pidh ≤ P¯ i xib, ∀i ∈ ℬ, ∀d, ∀h (19)
0≤ ∑ xig ≤ G (3) b
i∈G E bi ≤ eidh
b
≤ E¯ i , ∀i ∈ ℬ, ∀d, ∀h (20)

0≤ ∑ xib ≤ ℬ (4)
b
eid b
, 24 = Eid , 0, ∀i ∈ ℬ, ∀d (21)
i∈ℬ

A typical daily generation profile is used to simulate the generation


0≤ ∑ xiw ≤ W (5) behaviours of renewable energy resources, and their generation
i∈W outputs are constrained by the corresponding profile as stated in the
equation below:
As each DER is connected to only one bus in the microgrid, the
aggregated sum of the location indicators should be equal to the w
0 ≤ pidh w w
≤ Pidh xi , ∀i ∈ W, ∀d, ∀h (22)
deployment state of the DER, as indicated in (6)–(8). If the DER is
installed, only one location indicator is active and equal to 1;
otherwise, all the location indicators are inactive and equal to 0 The model above involves various operation constraints. However,
the power flow equations of the network should also be taken into
account. The active (or reactive) power injection amount at each
xig = ∑ xigj, ∀i ∈ G (6) bus is equivalent to the difference between the cumulated power
j∈N
output of local DERs and the load, as shown in (23)–(26). Note
that the power injection of the utility grid (i.e. the imported/
xib = ∑ xibj, ∀i ∈ ℬ (7) exported power amount) is also considered at the root bus. Only
j∈N
DGs are assumed to generate reactive power, as shown in (25) and
(26)
xiw = ∑ xiwj , ∀i ∈ W (8)
j∈N
pin gd + gd −
jdh = pdh − pdh + ∑ pigjdh + ∑ (pibjdh − pibjdh)
+ −

i∈G i∈ℬ
To avoid placing two and more DERs at a single bus, constraint (9) (23)
should be met + ∑ piwjdh − Pld
jdh, ∀ j ∈ N0
i∈W

∑ xigj + ∑ xibj + ∑ xiwj ≤ 1, ∀j ∈ N (9)


i∈G i∈ℬ i∈W pin
jdh = ∑ pigjdh + ∑ (pibjdh − pibjdh)
+ −

i∈G i∈ℬ
The power exchange with the utility grid is constrained by the (24)
maximum limits as (10), while the power output of each DG is
+ ∑ piwjdh − Pld
jdh, ∀ j ∈ N∖N0
i∈W
constrained by its lower bound (LB) and upper bound (UBs) as
(11)
qin gd
jdh = qdh + ∑ qigjdh − Qldjdh, ∀ j ∈ N0 (25)
i∈G
gd
0≤ gd +
pdh , gd −
pdh ≤ P¯ , ∀d, ∀h (10)

g
qin
jdh = ∑ qigjdh − Qldjdh, ∀ j ∈ N∖N0 (26)
0≤ g
pidh ≤ P¯ xg,
i i ∀i ∈ G, ∀d, ∀h (11) i∈G

3536 IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 16, pp. 3534-3548
© The Institution of Engineering and Technology 2019
The LinDistFlow model is employed to formulate power flow
constraints, which is extensively used in radial distribution systems Φ= α ∑ αi j = ℒ − k (37)
and microgrids [23–25], as expressed in (27)–(33): (27) and (28) (i, j) ∈ ℒ

address the balance of active and reactive powers at each bus.


Equation (29) relates the active, reactive power flows of a branch Similar to those in the basic model, (38) and (39) specify the
to the voltage magnitude of its two terminal buses. Equations (30)– bounds of the active (reactive) power exchanged with the utility
(33) present the limits on the active power, reactive power and grid. Equations (40)–(44) limit the power outputs of DERs.
voltage, where V0 is the reference voltage magnitude. If the bus is Equation (45) denotes the curtailed load should be less than the
load at present. Similarly, (46)–(61) address the power flow
the root bus, then the voltage retains the reference voltage as
constraints of microgrids in an N–k contingency. Note that (52)–
shown in (32); otherwise, the bus voltage is allowed to vary in the
(59) are restricted by the states of both the existing distribution
determined interval as shown in (33)
lines and the backup ones. If a distribution line is damaged, the
restriction on the voltage is relaxed and does not take effect
pin
jdh = ∑ p jkdh − ∑ pi jdh, ∀j ∈ N (27) according to (52)–(55). Furthermore, the power flow in that
k ∈ σ( j) i ∈ π( j)
distribution line is also enforced to be zero, as indicated in (56)–
(59); otherwise, (52)–(59) will take effect and be incorporated into
qin
jdh = ∑ q jkdh − ∑ qi jdh, ∀j ∈ N (28) the power flow constraints. Equations (60) and (61) represent the
k ∈ σ( j) i ∈ π( j)
limits on the voltage magnitude
(pi jdhRi j + qi jdh Xi j)
vidh − v jdh = , ∀(i, j) ∈ ℒ (29) −P̄
gd
≤ p~gd ≤ P¯
gd
(38)
V0

−P̄i j ≤ pi jdh ≤ P¯ i j, ∀(i, j) ∈ ℒ (30) −Q̄


gd
≤ q~gd ≤ Q¯
gd
(39)

g
−Q̄i j ≤ qi jdh ≤ Q¯ i j, ∀(i, j) ∈ ℒ (31)
g
0 ≤ p~i j ≤ P¯ i xigj, ∀i ∈ G, ∀ j ∈ N (40)

vidh = V 0, ∀i ∈ N0 (32) g
0 ≤ q~i j ≤ Q¯ i xigj,
g
∀i ∈ G, ∀ j ∈ N (41)

V i ≤ vidh ≤ V¯ i, ∀i ∈ N∖N0 (33) b


(42)
0 ≤ p~bi j+ ≤ P¯ i xibj, ∀i ∈ ℬ, ∀ j ∈ N
It is worth noting that the basic model is conducted regardless of b
contingencies. To enhance the resilience of microgrids against 0 ≤ p~bi j− ≤ P¯ i xibj, ∀i ∈ ℬ, ∀ j ∈ N (43)
contingencies, specific constraints for characterising the resilience
of microgrids are to be developed. 0 ≤ p~w w w
i j ≤ Pi xi j , ∀i ∈ W, ∀ j ∈ N (44)

2.2 Resilience of microgrids against contingencies 0 ≤ p~lcj ≤ Pld


j , ∀j ∈ ℬ (45)
For a planning problem, the detailed operation of the resilience
measures and the transient operation of the system are commonly p~inj = p~gd + ∑ p~gi j + ∑ (p~bi j +
− p~bi j− )
ignored. What we care is whether the system has the ability to i∈G i∈ℬ
(46)
restore the loads in case of contingencies. In this regard, we first ~ld
+ ∑ p~w ~lc
ij − Pj + pj , ∀ j ∈ N0
characterise the resilience and then model the system recovery i∈W
ability in the worst scenario in the planning model.

Definition 3.1: The resilience level of microgrids S is defined as p~inj = ∑ p~gi j + ∑ (p~bi j +
− p~bi j− )
i∈G i∈ℬ
the inverse load loss rate in N–k contingencies, as indicated in (34), (47)
~ld
lc
where P¯ is the maximum allowable load loss in N–k contingencies + ∑ ~w
pi j − P j + p~lcj , ∀ j ∈ N∖N0
ld i∈W
and P is the maximum loads
¯

~
1 P¯
ld q~inj = q~gd + ∑ q~gi j − Qldj + q~lcj , ∀ j ∈ N0 (48)
S=
¯ lc ld
=
¯ lc
(34) i∈G
P /P̄ P
~
q~inj = ∑ q~gi j − Qldj + q~lcj , ∀ j ∈ N∖N0 (49)
Obviously, the resilience level increases as the maximum allowable i∈G
load loss decrease. Specifically, the resilience level becomes
positively infinite when the maximum allowable load loss is zero. p~inj = ∑ p~ jk − ∑ p~i j, ∀j ∈ N
This is also consistent with practises as the system is the most (50)
k ∈ σ( j) i ∈ π( j)
resilient when no load is lost in any N–k contingencies. According
to the definition, the domain of resilience level is [1, + ∞). q~inj = ∑ q~ jk − ∑ q~i j, ∀j ∈ N
The goal at this stage is to make the system resilience larger (51)
k ∈ σ( j) i ∈ π( j)
than a given resilience level, as indicated in (35). The model can be
reformulated as (36), where the curtailed load is less than a given v~i − v~ j − (p~i jRi j + q~i j Xi j) ≥ − M(1 − αi j), ∀(i, j) ∈ ℒ (52)
bound under the worst realisation of the N–k contingency. Φ is the
uncertainty set of N–k contingencies as indicated in (37) v~i − v~ j − (p~i jRi j + q~i j Xi j) ≤ M(1 − αi j), ∀(i, j) ∈ ℒ (53)
ld
P¯ v~i − v~ j − (p~i jRi j + q~i j Xi j) ≥ − M(1 − βi j), ∀(i, j) ∈ ℒc (54)
S≤ (35)
max min ∑ p~lcj
α∈Φ j∈ℬ v~i − v~ j − (p~i jRi j + q~i j Xi j) ≤ M(1 − βi j), ∀(i, j) ∈ ℒc (55)
lc ld
max min ∑ p~lcj ≤ P¯ = P¯ /S (36) −P̄i jαi j ≤ p~i j ≤ P¯ i jαi j, ∀i, j ∈ ℒ (56)
α∈Φ j∈ℬ

−Q̄i jαi j ≤ q~i j ≤ Q¯ i jαi j, ∀i, j ∈ ℒ (57)

IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 16, pp. 3534-3548 3537
© The Institution of Engineering and Technology 2019
−P̄i j βi j ≤ p~i j ≤ P¯ i j βi j, ∀i, j ∈ ℒc (58) lines. Path line (i, j) is active when αi,j is equal to one and is
inactive when αi,j is equal to zero. According to the graph theory
−Q̄i j βi j ≤ q~i j ≤ Q¯ i j βi j, ∀i, j ∈ ℒc (59) [28], the number of lines is equal to the number of nodes minus
that of sub-graphs when a radial graph is partitioned into several
v~i = V 0, ∀i ∈ N0 (60) radial sub-graphs, as revealed in (68). Equations (64)–(67) ensure
the connectivity of the network, as there must be routeing from the
V i ≤ v~i ≤ V¯ i, ∀i ∈ N (61) source node to the load node according to energy balance. The
network is further enforced to be radial when (68) is added
It should be noted that the ‘Big-M’ method generally introduces a
computational burden. To reduce the computational burden, we ∑ l jk − ∑ li j = s j − 1, ∀ j ∈ N0 (64)
would estimate the LB of the ‘M’ to narrow the feasibility region k ∈ σ( j) i ∈ π( j)

in practises. For example, the ‘M’ in (52)–(55) is set to 0.2 as the


maximum voltage gap between two buses is 0.2. ∑ l jk − ∑ li j = − 1, ∀ j ∈ N∖N0 (65)
k ∈ σ( j) i ∈ π( j)

2.3 Radial formation of microgrids s j ≥ 1, ∀ j ∈ N0 (66)


Sub-systems may be formed in the microgrid in case a contingency
occurs. On the one hand, to keep a regular operation, all the sub- −Mαi j ≤ li j ≤ Mαi j, ∀(i, j) ∈ ℒ (67)
systems are required to maintain a radial structure. On the other
hand, to recover the power supply of loads as much as possible, ∑ αi j = N − N0
one can either use local DERs to generate power if they are at hand (68)
(i, j) ∈ ℒ
or reconnect the isolated sub-system to the active one through
backup distribution lines if they are available. We can use at most k A simple case (shown in Fig. 1) is used to illustrate this idea.
backup distribution lines to link the sub-systems to form a few Assume node 1 is a source node, and the other nodes are all load
larger systems or leave the sub-systems alone in an N–k nodes [as shown in (1)). To ensure the energy balances at nodes 2–
contingency, as shown in (62). β represents the active state of 4, paths from the source node 1 to all load nodes 2–4 have to be
backup distribution lines in a contingency, whereas z represents the formed [as shown in (2)], and an interconnected radial network is
deployment state of candidate distribution lines. Backup formed.
distribution line is available only if it is deployed, as stated in (63). The model above is incapable of dealing with islanded cases as
It should be noted that the active state of existing distribution lines the source nodes have to be predetermined. To overcome this
in a contingency α does not rely on z but is associated with the drawback, a modified single commodity flow method is proposed.
specific contingency In a similar manner, the proposed method assumes that each sub-
system is dominated by one source node. Each node of the network
∑ βi j ≤ k (62) is referred to as a virtual source node, and a power injection is
(i, j) ∈ ℒc assigned. Whether the virtual source node is a real source node
depends on the source node indicator. If the source node indicator
βi j ≤ zi j, ∀(i, j) ∈ ℒc (63) is active, the virtual source node turns out to be a real source node
and the power injection amount is non-zero; otherwise, the virtual
However, constraint (62) is not a sufficient condition to ensure a source node is a load node and the power injection amount is zero.
radial structure of sub-systems. To guarantee the radiality of a Specifically, the constraints are shown in (69)–(73). Equation (69)
distribution network, some effective models have been developed represents the energy balance of the virtual source nodes, where s′j
in existing works [25–27]. Those models partition the distribution is the actual power injection related to the virtual power injection
network into several sub-systems, each of which is dominated by a and source node indicator. The bilinear terms sjuj can be easily
source node. However, islanded areas without source nodes may be handled by the ‘Big-M’ method mentioned later. Equations (70)–
formed in an N–k contingency. Accordingly, some improvements (72) share the same meanings with those in the model with
should be made to handle this new situation. In the following predetermined source nodes. As each sub-system is supplied by
section, we would first introduce the single commodity flow only one real source node, the sum of source node indicators is
method [25], which suffices to ensure the radiality of a distribution equal to the number of sub-systems, as indicated in (73)
network with predetermined source nodes. After that, we modify
the model to make it capable of handling islanded cases. ∑ l jk − ∑ li j = s ju j − 1 = s′j − 1, ∀j ∈ N
The main idea of the method is as follows: a fictitious graph (69)
k ∈ σ( j) i ∈ π( j)
that has the same topology with the distribution system is
constructed. Assume each sub-system is dominated by one source s j ≥ 1, ∀j ∈ N (70)
node (e.g. nodes corresponding to the utility grid or the location of
DGs), and each source node is assigned a power injection. The −Mαi j ≤ li j ≤ Mαi j, ∀(i, j) ∈ ℒ (71)
other nodes are set to be sink (load) nodes, and each is assigned a
one unit load. Equations (64)–(68) constitute the complete
constraints for maintaining the radiality of a distribution network. ∑ αi, j = N − ∑ uj (72)
(i, j) ∈ ℒ j∈N
Equations (64) and (65) address the energy balance at the source
nodes and the load nodes, respectively. The power injection
amount at the source nodes should be larger than one unit, as ∑ uj = k (73)
indicated in (66). Equation (67) specifies the availability of path j∈N

To clarify the method, a simple case (shown in Fig. 2) same as


Fig. 1 is analysed. As shown in (1), all the nodes are set to be
virtual source nodes. If k = 1, one radial system is formed and one
virtual source node (e.g. node 1) will adaptively become a real
source node limited by (73), while the other nodes all become load
nodes naturally. Similarly, if k = 2, two radial sub-systems are
formed and two virtual source nodes (e.g. node 1 and node 3)
adaptively become real source nodes.
Fig. 1 Radial formation of a network with predetermined source nodes In this way, the sub-systems formed in an N–k contingency can
always be radial. In particular, the constraints are shown in (74)–

3538 IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 16, pp. 3534-3548
© The Institution of Engineering and Technology 2019
(80) when the developed method is applied to the planning
problem

∑ l jk − ∑ li j = s′j − 1, ∀j ∈ N (74)
k ∈ σ( j) i ∈ π( j)

s j ≥ 1, ∀j ∈ N (75)

−Mαi j ≤ li j ≤ Mαi j, ∀(i, j) ∈ ℒ (76)

−Mβi j ≤ li j ≤ Mβi j, ∀(i, j) ∈ ℒc (77)

∑ αi, j + ∑ βi, j = N − ∑ uj (78)


(i, j) ∈ ℒ (i, j) ∈ ℒc j∈N
Fig. 2 Radial formation of a network with virtual source nodes
∑ uj ≤ k + 1 (79)
j∈N

0 ≤ s′j ≤ Mu j, s j − M(1 − u j) ≤ s′j ≤ s j, ∀j ∈ N (80)

It should be noted that the microgrid is interconnected with the


main grid. The root bus is always considered as a source node.
Therefore, at most k + 1 sub-systems (including the main grid) are
formed in an N–k contingency, as indicated in (79). The actual
number of sub-systems is to be optimised in the procedure.
Equation (80) accounts for the linearisation of bilinear terms using
the ‘Big-M’ method. Similarly, the UBs of lij and s’j are both |N|.
Consequently, M can be |N| in (76), (77), and (80).

Remark: The model above is a mixed integer linear
programming (MILP) problem, thus the constraints are easily
constructed for their relatively low computational cost. The
advantage of the model is that it overcomes the drawback of
conventional approaches [25–27] incapable of dealing with
islanded cases. In addition, it introduces fewer binary variables
than the conventional spanning tree approach [27], where each
branch introduces two binary variables.

2.4 Compact model of the planning problem


To better grasp the structure of the model and further explore the Fig. 3 Overall algorithm
solution to this problem, the model is rewritten in a compact form.
An equation can be written in the form of two inequalities, and all
equations are expressed as inequalities. Therefore, the proposed 3 Proposed algorithms
model can be written in a compact form as shown in the equations 3.1 Overall algorithm
below:
The proposed model is a tri-level two-stage robust optimisation
problem; the column-and-constraint generation (C&CG) procedure
min aT x + bT y0
[29] is suitable to solve it. The key idea of this procedure is to find
x, y0 (81) the worst-case event and then add the event-related constraints into
s . t . A x + B y0 ≤ d the original problem to tighten the feasible region until the
algorithm converges. Specifically, the proposed model is
ld
max min cT y ≤ P¯ /S decomposed into a master problem and a sub-problem. The master
α∈Φ β, y (82) problem is a relaxation of the original problem, whereas the sub-
s . t . Dx + Eα + Fβ + Gy ≤ e problem is used to identify extreme event scenarios. The obtained
extreme event is taken as the worst-case event and corresponding
Equation (81) corresponds to the basic planning model as shown in constraints are added to the master problem. The master problem
(1)–(33), whereas (82) corresponds to the resilience model as and the sub-problem will be solved iteratively until the objective of
displayed in (36)–(63) and (74)–(80). x is the investment-related the sub-problem is less than a given load loss level. The explicit
binary decision vector; y0 is the operation-related continuous algorithm using this procedure is as follows (Fig. 3).
decision vector in the basic planning model; y is the operation- The master problem is an MILP problem, which could be
related continuous decision vector in the resilience model; α and solved directly by off-the-shelf solvers such as CPLEX. The sub-
are binary decision vectors representing the states of existing problem is, however, a bi-level programme. An existing algorithm
distribution lines and backup distribution lines, respectively; and a, [30] is adopted to solve it.
b, c, d, and e are corresponding constant vectors, whereas A, B, D,
E, F, and G are corresponding constant matrices. 3.2 Algorithm to solve the sub-problem

Remark: The advantage of the model above is that it is able to The innermost minimisation problem of the sub-problem contains
characterise the resilience of the microgrid by the resilience level. both binary and continuous variables, which cannot be directly
Thus, the decision makers can adjust the system resilience by converted to a linear program by using duality theory [31, 32].
choosing a proper value, which is more straightforward than some Observe that the sub-problem is equivalent to the tri-level problem
methods [20] which consider the load loss in the objective by separating binary variables from continuous variables expressed
function. as the equation below:

IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 16, pp. 3534-3548 3539
© The Institution of Engineering and Technology 2019
f (x^ ) = max min min cT y
α∈Φ β y
(86)
s . t . Gy ≤ e − Dx^ − Eα − Fβ: λ

Specifically, by dualising the innermost minimisation problem, the


sub-problem can be further written as the tri-level problem as
shown in the equation below:

f (x^ ) = max min max (e − Dx^ − Eα − Fβ)T λ


α∈Φ β λ
T (87)
s.t. G λ = c
λ≤0

By enumerating all possible assignments of , the monolithic


equivalent form of the sub-problem can be acquired, as shown in
the equation below:

f (x^ ) = max θ
^k
s . t . θ ≤ (e − Dx^ − Eα − Fβ )T λk 1≤k≤m
(88)
T k
G λ = c, 1 ≤ k ≤ m;
α ∈ Φ, λk ≤ 0, 1≤k≤m

Obviously, a C&CG procedure can be used again to solve the tri-


level formulation. The explicit algorithm to solve the sub-problem
is given in Fig. 4. Fig. 4 Explicit algorithm to solve the sub-problem
The first constraints in (89) contain bilinear terms αTλ, which
can be reformulated as an MILP problem using the ‘Big-M’
method by introducing auxiliary variable vector

−Mα ≤ γ ≤ 0, λ ≤ γ ≤ λ + M(1 − α) (92)

Finally, the sup-problem can also be efficiently solved by off-the-


shelf solvers such as CPLEX.

4 Case study
To verify the effectiveness of the proposed model, two microgrid
cases are studied based on the IEEE 33-bus distribution system
[23] and IEEE 123-bus distribution system [33]. Numerical
experiments are implemented in MATLAB/YALMIP on a desktop
with an Intel Pentium 3.6 GHz processor and 8 GB memory. All
the optimisation problems are solved by invoking CPLEX. Fig. 5 Structure of the IEEE 33-bus distribution system-based microgrid

4.1 Case study I: IEEE 33-bus distribution system


The configuration of the tested microgrid is shown in Fig. 5. The
peak load of the microgrid is about 3 MW. The candidate DERs
include four DGs, two WTs, two PVs, and two BSs, while five
candidate distribution lines are displayed in Fig. 5. The parameters
of DERs, the costs of distribution lines, the daily electricity prices,
and the candidate buses for DER integration are given in Figs. 6
and 7 and Tables 1–5, whose data are mainly adapted from [34–
36]. Both the charging and discharging efficiencies of BSs are
assumed to be 0.9. The initial stored energy in BSs is assumed to
be half of their capacity. The selling price is assumed to be 0.8
times the buying price. Four typical days are selected to represent
four seasons of a year. Typical generation profiles of WTs and PVs
and the typical load profile are also given in Figs. 6 and 7 and
Tables 1–5.
Since several line outages seldom happen simultaneously in
distribution systems, the maximum number of contingencies is set Fig. 6 Typical generation profiles of WTs and PVs in four seasons
to three. Eight cases are analysed and compared in Table 6. Cases
I-1–I-4 plan both DERs and distribution lines considering different (i) Economic analysis of Case I-1: The resilience level of Case
resilience levels in N–1 contingencies. Note that the resilience I-1 is set to be relatively low such that the backup distribution lines
levels are calculated according to (34), e.g. S = 6 derives from S =  are unnecessary. The smart deployment of DERs is adequate to
3000/500, which also means that the maximum allowable load loss satisfy an appropriate power supply in N−1 contingencies. The
is 500 kW. Cases I-5–I-6 study the cases at two resilience levels in optimisation result finds the optimal number and placement of
N−2 contingencies, whereas Cases I-7–I-8 conduct the cases in N– DERs in the microgrid. The annual cost and power generation of
3 contingencies. DERs are shown in Table 7.
According to the optimisation, DERs including three DGs, two
BSs, four WTs, and two PVs are installed in the microgrid. These
3540 IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 16, pp. 3534-3548
© The Institution of Engineering and Technology 2019
Fig. 7 Typical load profile in four seasons

Table 1 Technical Parameters of DERs


Type Rated power, kW Investment, $/kW Lifespan, year Fuel cost, $/kWh Number limit
DG 400 300 10 0.16 6
WT 200 1200 20 0 5
PV 250 1600 20 0 6

Table 2 Technical parameters of BSs


Type Pmax (kW) Emax (kWh) Power cost, $/kW Energy cost, $/kWh Lifespan, year Number limit
BS 200 500 270 150 15 4

Table 3 Investment costs of distribution lines


Type Investment, $/m Lifespan, year
Distribution line 20 30

Table 4 Electricity prices


Time of use Off-peak 1 (23–07) Off-peak 2 (07–10, 15–18, 21–23) Peak 1 (10–11, 13–15, 18–21) Peak 2 (11–13)
Electricity price ($/kWh) 0.057 0.126 0.198 0.216

Table 5 Candidate buses for DER integration


DER type Candidate buses
DG 5 6 7 15 19 23 25 29 31
BS 2 3 4 11 13 17 30
WT 9 8 10 18 20 22 24 27
PV 12 14 16 21 26 28 32 33

Table 6 Case illustration for a 33-bus microgrid


Case Resilience level k Case Resilience level k
I-1 6 1 I-5 6 2
I-2 10 1 I-6 10 2
I-3 15 1 I-7 6 3
I-4 +∞ 1 I-8 10 3

Table 7 Annual cost and generation of DERs in the microgrid


DER Investment cost, 103$ Operation cost, 103$ Generation amount, MWh Average cost, $/kWh
DG1 23.4 200.3 1241 0.181
DG2 23.4 217.6 1349 0.178
DG3 23.4 198.6 1231 0.180
BS1 32.0 - 1174 -
BS2 32.0 - 1174 -
WT1–WT4 39.4 0 668 0.059
PV1–PV2 43.8 0 441 0.099

IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 16, pp. 3534-3548 3541
© The Institution of Engineering and Technology 2019
Fig. 8 Generation schedule of DERs in the microgrid in the typical days

DERs coordinate to satisfy the load in the microgrid. Surplus deployed in Cases I-2–I-4 as the resilience level increases. Take
power generated by DERs is sold to the utility grid. Case I-2, for example. Compared to Case I-1, the size and site of
Table 7 records the detailed cost and power generation of DERs are nearly the same; however, one more distribution line is
DERs. It can be observed that the average cost of DGs is higher deployed. In case an outage occurs in the first feeder, the backup
than that of other DERs, while the average cost of renewable distribution line (8, 21) can reconnect isolated loads to active
energy resources is relatively low. Three out of four candidate DGs networks, thereby reducing load loss in a contingency and
are installed, even though the generation of DGs is not cheap. increasing the system resilience. It is as expected to find that the
However, there are some periods (e.g. Peak 1 and Peak 2 h) during number of backup distribution lines increases as the resilience level
which the local generation of DGs is more economic. With regards rises. Specifically, all the candidate DERs and three candidate
to renewable energy resources, all the candidates WTs and PVs are distribution lines are deployed to defend against possible line
placed in the microgrid as their generation is cheap. Note that the outages since load loss is entirely intolerable in any N−1
average cost of WTs is less than that of PVs. On the one hand, the contingencies in Case I-4. Obviously, the results are compatible
unit investment of WTs is less than that of PVs. On the other hand, with practise, which verify the effectiveness of the proposed
the annual energy generation amount (668 MWh) of WTs is more model.
than that (441 MWh) of PVs. All the candidate BSs are deployed, From Table 8, the investment and operation costs of DERs in
which is because volatile electricity prices put BSs in such a Cases I-1–I-3 are identical. As we know, the total loads in the
favourable position that BSs can take full advantage of the microgrid are the same in the three cases. From the perspective of
electricity price gap to diminish the cost. energy balance, the capacities of the installed DERs in Case I-2 and
Fig. 8 gives the generation schedule of DERs in the microgrid Case I-3 should be at the same level as that in Case I-1. However,
in the typical days of four seasons. Since the electricity price is this does not mean that they have no influence on resilience. Their
cheap most of the time, the majority of the power supply is derived placement, which may influence the system resilience, makes
from the utility grid. During certain high-price periods, the power difference, as explained in the paragraph above. In contrast to Case
generated from local DERs is even sold to the utility grid to I-1, additional investment cost of distribution lines is imposed on
increase the microgrid revenue. We find that DGs which are off Case I-2 and Case I-3 to defend against N–1 contingencies, even
most of the time is started up to satisfy the load only during peak though the investment and operation costs in three cases are the
hours when the electricity price is high. The generation of same. Since the resilience requirement to Case I-3 is severer, more
renewable energy resources is relatively cheap. Therefore, WTs investment cost of distribution lines is demanded.
and PVs are adequately utilised to generate power. The BSs take The total cost in Case I-4 is the most among the four cases as
advantage of the electricity price gap to make a profit during the load loss is completely intolerable in this case. Accordingly, more
day. As can be seen in Fig. 8, they charge during off-peak 1/off- backup DERs and distribution lines are desired for Case I-4. As
peak 2 h while discharge during peak 1/peak 2 h. Obviously, the can be seen in Table 8, the DER investment cost and distribution
results verify the economics of the operation. line investment cost in Case I-4 are higher than those in other
(ii) Influence of the resilience level: Cases I-1–I-4 at different cases. Nevertheless, the operation cost in Case I-4 is less than that
resilience levels are compared to investigate the influence of the in other cases. This can be explained by the fact that one more DG
resilience level to the planning results. The placement of DERs and can improve the flexibility of the microgrid and better decrease the
distribution lines in the four cases are displayed in Fig. 9, while the operation cost. We can find that the system would prefer to
detailed annual investment and operation costs, as well as numbers establish new distribution lines rather than install more DERs to
of DERs and distribution lines in the four cases, are presented in defend against contingencies. This is because the investment cost
Tables 8 and 9, respectively. of distribution lines is much lower than that of DERs in terms of
Fig. 9 displays the placement of DERs and distribution lines in maintaining a resilience level. From the analysis above, we can
the four cases. As can be seen in Fig. 9, only DERs are deployed in conclude that the planning problem is essentially to make a proper
Case I-1. The DERs seem to be distributed randomly in the trade-off between the investment and resilience of the microgrid.
network. A careful look at Case I-1 of Fig. 9 reveals that DGs are Table 9 shows the detailed numbers of DERs and distribution lines
distributed in three feeders. It should be noted that DGs have a in the four cases. Obviously, the result is in accordance with that of
large capacity and they are assumed to generate reactive power Table 8, which is not analysed here for simplicity.
only among the DERs, and therefore their placement greatly (iii) Influence of the number of contingencies: To investigate the
influences the system resilience. Once an upstream line outage impact of the number of contingencies on the planning results,
occurs in a feeder, the local DGs can at least support a part of loads Cases I-5 and I-7, Cases I-6 and I-8, which are at the same
in that feeder. Therefore, the deployment of DERs actually resilience level but with a different number of contingencies are
enhances the resilience of the microgrid. Conversely, assume all compared. Cases I-6 and I-8, however, are infeasible as their load
the DGs are located in one feeder. Majority of the loads are lost loss fails to meet the required resilience level. Hence, only the
once an upstream line outage occurs in that feeder. Therefore, the results of Cases I-5 and I-7 are displayed in Fig. 10. The detailed
placement of DERs also plays an important role in influencing the annual investment and operation costs, as well as numbers of DERs
system resilience. DERs together with backup distribution lines are

3542 IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 16, pp. 3534-3548
© The Institution of Engineering and Technology 2019
Fig. 9 Placement of DERs and backup distribution lines in the four cases

Table 8 Detailed annual investment and operation costs in the four cases
Case DERs Investment, 103$ DERs operation, 103$ AC lines, 103$ Total cost, 103$
I-1 379.5 1331.6 0 1711.2
I-2 379.5 1331.6 8.8 1719.9
I-3 379.5 1331.6 10.9 1722.1
I-4 402.9 1320.4 13.1 1736.5

Table 9 Numbers of DERs and backup distribution lines in the four cases
Case DGs BSs WTs PVs Distribution lines
I-1 3 2 4 2 0
I-2 3 2 4 2 1
I-3 3 2 4 2 2
I-4 4 2 4 2 3

and distribution lines in the four cases, are presented in Tables 10 contingencies grow. However, backup distribution lines are given
and 11, respectively. The results of Cases I-6 and I-8 are empty. priority with respect to defending against contingencies as their
As can be seen in Fig. 10, more and more backup distribution cost is relatively low.
lines are established as the number of contingencies grows. (iv) Size and computational time of the cases: The size of the
Specifically, in Case I-5, two more distribution lines are reserved studied cases in terms of the number of variables and constraints
compared with that in Case I-1 at the same resilience level but in are given in Table 12. It should be noted that the size of the cases
N–1 contingencies. This is easy to understand as it may incur more corresponds to that of the master problem at the last iteration here.
serious outages (e.g. isolated areas) as the number of contingencies The computational time of the studied cases is also displayed in
grows. To restore the lost load in serious contingencies, backup this table.
distribution lines are indispensable to be established. If the number of contingencies is fixed, e.g. Cases I-1–I-4, the
In some extreme cases such as Case I-7, where three size of the optimisation problem increases with the rise of the
contingencies happen simultaneously, not only all the candidate resilience level. This is because more iterations are needed to
distribution lines are deployed but also all the candidate DERs are satisfy a higher resilience level, which also requires more
installed to defend against contingencies, as shown in Fig. 10 (2). computational time. As can be seen in Table 12, Case I-4, which
Tables 10 and 11 record the detailed costs and numbers of has the highest resilience level, costs the most among the first four
deployed DERs and distribution lines in the four cases. Since Cases cases. On the other hand, if the resilience level is fixed, e.g. Cases
I-6 and I-8 are infeasible, they have no records. Similar to the I-1, I-5, and I-7, the size and computational time of the
analysis above, more and more costs are invested in increasing the optimisation problem grow with the increase of the number of
number of DERs and backup distribution lines as the number of contingencies. This is easy to understand as more contingencies

IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 16, pp. 3534-3548 3543
© The Institution of Engineering and Technology 2019
Fig. 10 Placement of DERs and backup distribution lines in Cases I-5 and I-7

Table 10 Detailed annual investment and operation costs in the four cases
Case DERs Investment, 103$ DERs operation, 103$ AC lines, 103$ Total cost, 103$
I-5 379.5 1331.6 4.4 1715.5
I-6 — — — —
I-7 402.9 1320.4 30.6 1754.0
I-8 — — — —

Table 11 Numbers of DERs and backup distribution lines in the four cases
Case DGs BSs WTs PVs Distribution lines
I-5 3 2 4 2 2
I-6 — — — — —
I-7 4 2 4 2 5
I-8 — — — — —

Table 12 Size and computational time of the cases in case study I


Case Number of variables Number of constraints Number of iterations Computational time, s
I-1 83,831 178,439 7 54
I-2 85,717 182,691 9 133
I-3 86,660 184,817 10 191
I-4 88,546 189,069 12 208
I-5 86,660 184,817 10 205
I-6 — — — —
I-7 117,779 254,975 43 7830
I-8 - — — —

particular, it takes Case I-7 43 iterations and 7830 s to find the


solution.
(v) Comparison with the conventional model: To further
validate the proposed model, a conventional model in [4] assuming
the uncertainty is ignored is used for comparison. The planning
result is displayed in Fig. 11.
It is easily found that only DERs are placed in the microgrid.
The number and size of DERs are completely the same as that in
Case I-1. However, the placement of DERs makes a difference.
Three DGs are distributed in two feeders. As analysed before, the
resilience of this design is lower than that in Case I-1, where three
DGs are distributed in three feeders. Therefore, it proves that the
proposed model has fully considered the resilience and validates
the proposed model.
(vi) Comparison with the reliability-based method: A recent
publication such as Nazar and Heidari [21] identify the worst
contingencies by using the reliability-based method. The cases
whose ENS is more than a predefined threshold are considered as
Fig. 11 Planning result of the conventional model in case study I
the worst-case planning scenarios. This paper, however, uses a
robust optimisation method to identify the worst scenarios. The
correspond to more possible combinations of faults, which also advantage of the robust optimisation method is that it can fast and
requires more iterations and more time to search for the solution. In precisely identify the worst contingency. The reliability-based

3544 IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 16, pp. 3534-3548
© The Institution of Engineering and Technology 2019
method, which is commonly based on the Monte Carlo simulation 4.2 Case study II: IEEE 123-bus distribution system
(MCS), needs a lot of simulations and may cost more time. The
proposed method is compared with the MCS method in identifying To explore the generality of the proposed model, the IEEE 123-bus
the worst contingencies. The final deployment of Case I-1 as distribution system (whose configuration is shown in Fig. 11) is
shown in Fig. 9 (1) is used as the test system. Assume the failure also studied. The candidate DERs include ten DGs, six WTs, four
probability of the distribution lines is 0.001. About 10,000 samples PVs, and four BSs, while eight candidate distribution lines are
are simulated in the MCS. Subsequently, the worst-case shown in Fig. 12. The parameters of candidate DERs and the
contingencies identified by the robust method with k = 1, 2, and 3, candidate buses for DER integration are presented in Tables 14–16.
and three contingencies with the largest ENSs from the MCS The other parameters such as the electricity price and loading level
method are shown in Table 13. The numbers in braces denote the are identical to those in case study I.
damaged lines in the contingency. As can be seen, the worst three Cases that have the same settings as those in case study I are
cases from MCS are all N–1 contingencies. The N–2 or N–3 studied. It is found that the procedure fails to find an optimal
contingencies are hardly sampled for their low ENSs. In solution within a given time when the number of contingencies
comparison, the proposed method is more robust as it can consider exceeds three, as too many binary variables are introduced in the
worst cases by adjusting k. In addition, with regards to N–1 iterations. Therefore, only the cases in N−1 and N−2 contingencies
contingencies, the proposed method identifies the worst are studied. Specifically, the placement of DERs and distribution
contingency the same as that from the MCS completely. However, lines in Cases II-1, II-2, II-3, and II-5 are shown in Fig. 13. Take
the computational time only takes 1.8 s, whereas the MCS method Case II-3, for example. According to the optimisation, four DGs,
takes 24.8 s. four BSs, six WTs, and four PVs are installed in the microgrid. The
placement of DERs is as expected. All the candidate renewable
energy resources are installed for their cheap generation. Since the
variation of the time of use electricity price is relatively large, all
the BSs are deployed to make profits by taking advantage of the
Table 13 Worst contingencies of robust and MCS methods price gap. Several candidate DGs are placed to cooperate with
in case study I other DERs for satisfying loads. Moreover, three backup
Methods Robust method MCS distribution lines are established to defend against possible line
k=1 k=2 k=3 outages in contingencies. By observing the planning results in the
worst contingencies 1 {22} {1,14} {1,14,30} {22} four cases, we can draw the same conclusions as those in case
2 — — — {18} study I: more backup distribution lines tend to be deployed as the
resilience level increases and more contingencies are considered.
3 — — — {2}
The annual costs and the specific numbers of deployed DERs
time, s 1.8 1.8 1.9 24.8 and distribution lines are given in Tables 17 and 18, respectively.
Note that Case II-4 is infeasible as load loss is inevitable in this
case, and hence its results are empty here.
As can be seen in Table 17, the investment and operation costs
of DERs in N–1 contingency cases are all the same. The
investment cost of distribution lines also increases as the resilience
level rises. Furthermore, more and more backup distribution lines
are established as the number of contingencies grows. This is also
in accordance with the results in Table 18. Similar to case study I,
establishing new distribution lines instead of installing more DERs
is the main measure to deal with contingencies for its relatively low
cost. The investment cost of distribution lines is much lower than
that in case study I, and hence the investors would prefer to
establish new distribution lines to maintain the resilience level
against contingencies.
The size and computational time of the cases are demonstrated
in Table 19. Generally, the size and computational time increase
with the rise of resilience level and the inclusion of more
Fig. 12 Structure of the IEEE 123-bus distribution system-based contingencies. Note that the sizes of Case II-1 and Case II-2 are the
microgrid same. This is because the resilience levels of Cases II-1 and II-2

Table 14 Technical parameters of DERs


Type Rated power, kW Investment, $/kW Lifespan, year Fuel cost, $/kWh Number limit
DG 200 300 10 0.16 10
WT 150 1200 20 0 8
PV 250 1600 20 0 6

Table 15 Technical parameters of BSs


Type Pmax, kW Emax, kWh Power cost, $/kW Energy cost, $/kWh Lifespan, year Number limit
BS 200 500 270 150 15 6

Table 16 Candidate buses for DER integration


DER type Candidate buses
DG 123 13 18 35 47 51 115 101 57 60 67 76 86 89
BS 25 26 121 38 108 109 112 61 100 70 74 82 90 91 53
WT 3 7 120 58 65 38 46 50 108 107 97 71 75 79 82 95 56
PV 122 33 37 39 52 97 117 118 73 85 90 94 66

IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 16, pp. 3534-3548 3545
© The Institution of Engineering and Technology 2019
are relatively low, and they are easily satisfied after several without considering the resilience. Therefore, this case further
iterations. However, to further increase the resilience level, much validates the proposed model.
more iterations are needed, as can be seen in Case II-3. The comparison between the robust method and MCS method
The planning result of the conventional model is also given in in identifying the worst contingencies is studied in case study II as
Fig. 14. It is found that one less DG is installed in the microgrid well. Case II-1 is used as the test system. The results are given in
compared with Case II-1. Since the conventional model is Table 20. Similarly, the worst cases from MCS are still N−1
completely based on an economic criterion, one less DG is a better contingencies. In addition, the computational cost is far more than
option for saving investment. However, as mentioned before, that with a robust method.
increasing the number of DGs enhances system resilience. Overall, we can conclude that the proposed model can always
Therefore, Case II-1 which considers the resilience enhancement effectively plan the DERs and distribution lines while maintaining
actually is more resilient than that of the conventional model a certain degree of resilience at least cost. Both two case studies
verify the effectiveness of the proposed model.

Fig. 13 Placement of DERs and backup distribution lines in case study II

Table 17 Detailed annual investment and operation costs in case study II


Case DERs investment, 103$ DERs operation, 103$ AC lines, 103$ Total cost, 103$
II-1 432.4 1273.1 0 1705.5
II-2 432.4 1273.1 0.2 1705.7
II-3 432.4 1273.1 0.7 1706.2
II-4 — — — —
II-5 432.4 1273.1 0.4 1705.9
II-6 — — — —

Table 18 Numbers of DERs and backup distribution lines in case study II


Case DGs BSs WTs PVs Distribution Lines
II-1 4 4 6 4 0
II-2 4 4 6 4 1
II-3 4 4 6 4 3
II-4 — — — — —
II-5 4 4 6 4 2
II-6 — — — — —

3546 IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 16, pp. 3534-3548
© The Institution of Engineering and Technology 2019
Table 19 Size and computational time of the cases in case study II
Case Number of variables Number of constraints Number of iterations Computational time, s
II-1 549,540 174,187 5 39
II-2 549,540 174,187 5 45
II-3 591,085 189,069 12 289
II-4 — — — —
II-5 686,045 223,085 28 11,701
II-6 — — — —

7 References
[1] Arriaga, M., Caizares, C. A., Kazerani, M.: ‘Long-term renewable energy
planning model for remote communities’, IEEE Trans. Sustain. Energy, 2016,
7, pp. 221–231
[2] Lotfi, H., Khodaei, A.: ‘AC versus DC microgrid planning’, IEEE Trans.
Smart Grid, 2017, 8, pp. 296–304
[3] Mitra, J., Vallem, M.R., Singh, C.: ‘Optimal deployment of distributed
generation using a reliability criterion’, IEEE Trans. Ind. Appl., 2016, 52, pp.
1989–1997
[4] Khodaei, A., Bahramirad, S., Shahidehpour, M.: ‘Microgrid planning under
uncertainty’, IEEE Trans. Power Syst., 2015, 30, pp. 2417–2425
[5] Guo, L., Liu, W., Jiao, B., et al.: ‘Multi-objective stochastic optimal planning
method for stand-alone microgrid system’, IET Gener. Transm. Distrib., 2014,
8, pp. 1263–1273
[6] Basu, A.K., Chowdhury, S., Chowdhury, S.P.: ‘Impact of strategic deployment
of CHP-based DERs on microgrid reliability’, IEEE Trans. Power Deliv.,
2010, 25, pp. 1697–1705
[7] Hakimi, S.M., Moghaddas-Tafreshi, S.M.: ‘Optimal planning of a smart
microgrid including demand response and intermittent renewable energy
resources’, IEEE Trans. Smart Grid, 2014, 5, pp. 2889–2900
Fig. 14 Planning result of the conventional model in case study II [8] Wang, H., Huang, J.: ‘Cooperative planning of renewable generations for
interconnected microgrids’, IEEE Trans. Smart Grid, 2016, 7, pp. 2486–2496
Table 20 Worst contingencies of robust and MCS methods [9] Che, L., Zhang, X., Shahidehpour, M., et al.: ‘Optimal interconnection
planning of community microgrids with renewable energy sources’, IEEE
in case study II Trans. Smart Grid, 2017, 8, pp. 1054–1063
Methods Robust Method MCS [10] Gazijahani, F.S., Salehi, J.: ‘Optimal bilevel model for stochastic risk-based
planning of microgrids under uncertainty’, IEEE Trans. Ind. Inf., 2018, 14,
k=1 k=2 k=3 pp. 3054–3064
worst contingencies 1 {13} {73,115} {73,115,122} {13} [11] Manshadi, S.D., Khodayar, M.E.: ‘Resilient operation of multiple energy
2 — — — {61} carrier microgrids’, IEEE Trans. Smart Grid, 2015, 6, pp. 2283–2292
[12] Xiong, L., Zhuo, F., Wang, F., et al.: ‘Static synchronous generator model: a
3 — — — {62} new perspective to investigate dynamic characteristics and stability issues of
time, s 6 7 13 561 grid-tied PWM inverter’, IEEE Trans. Power Electron., 2016, 31, pp. 6264–
6280
[13] Majidi-Qadikolai, M., Baldick, R.: ‘Integration of N−1 contingency analysis
with systematic transmission capacity expansion planning: ERCOT case
5 Conclusions study’, IEEE Trans. Power Syst., 2016, 31, pp. 2234–2245
[14] Correa, C.A., Bolanos, R., Garces, A.: ‘Enhanced multiobjective algorithm
Planning the microgrid considering the resilience enhancement is for transmission expansion planning considering N − 1 security criterion’, Int.
Trans. Electr. Energy Syst., 2015, 25, pp. 2225–2246
challenging in characterising the resilience and determining the [15] Wang, Q., Watson, J.-P., Guan, Y.: ‘Two-stage robust optimization for N−k
optimal site and size of DERs and distribution lines. In this paper, a contingency-constrained unit commitment’, IEEE Trans. Power Syst., 2013,
two-stage robust microgrid planning model is proposed for 28, pp. 2366–2375
addressing this issue, which takes full consideration of the [16] Moreira, A., Street, A., Arroyo, J.M.: ‘An adjustable robust optimization
approach for contingency-constrained transmission expansion planning’,
resilience modelling and radial formation of islanding network in IEEE Trans. Power Syst., 2015, 30, pp. 2013–2022
case of contingencies. The proposed model is solved by a double- [17] Zhang, B., Dehghanian, P., Kezunovic, M.: ‘Optimal allocation of PV
level C&CG algorithm. Specifically, the resilience of the microgrid generation and battery storage for enhanced resilience’, IEEE Trans. Smart
is characterised and integrated into the constraints. This design Grid, 2019, 10, pp. 535–545
[18] Ma, S., Chen, B., Wang, Z.: ‘Resilience enhancement strategy for distribution
suffices the decision makers to adjust the system resilience by systems under extreme weather events’, IEEE Trans. Smart Grid, 2018, 9, pp.
choosing a proper value. In addition, the radial formation model of 1442–1451
the network in contingencies is developed, which overcomes the [19] Zare-Bahramabadi, M., Abbaspour, A., Fotuhi-Firuzabad, M., et al.:
drawback of conventional approaches incapable of dealing with ‘Resilience-based framework for switch placement problem in power
distribution systems’, IET Gener. Transm. Distrib., 2018, 12, pp. 1223–1230
islanded cases. As illustrated by case studies, the proposed model [20] Ma, S., Su, L., Wang, Z., et al.: ‘Resilience enhancement of distribution grids
can not only prioritise and deploy candidate DERs and distribution against extreme weather events’, IEEE Trans. Power Syst., 2018, 33, pp.
lines in an optimal manner but also demonstrate a robust 4842–4853
performance against possible contingencies. [21] Nazar, M.S., Heidari, A.: ‘Optimal robust microgrid expansion planning
considering intermittent power generation and contingency uncertainties’, in
This paper assumes the generation of renewable energy sources Shahnia, F., Arefi, A., Ledwich, G. (Eds.): ‘Robust optimal planning and
such as WTs and PVs is deterministic. However, the generation of operation of electrical energy systems’ (Springer, Singapore, 2019), pp. 177–
renewable energy sources demonstrates great volatility and 198
uncertainty, which may influence the planning results. Therefore, [22] Madathil, S.C., Yamangil, E., Nagarajan, H., et al.: ‘Resilient off-grid
microgrids: capacity planning and N−1 security’, IEEE Trans. Smart Grid,
the uncertainty of renewable energy sources will be considered in 2018, 9, pp. 6511–6521
our future work. [23] Baran, M. E., Wu, F.F.: ‘Network reconfiguration in distribution systems for
loss reduction and load balancing’, IEEE Trans. Power Deliv., 1989, 4, pp.
1401–1407
6 Acknowledgments [24] Wang, Z., Chen, B., Wang, J., et al.: ‘Robust optimization based optimal DG
placement in microgrids’, IEEE Trans. Smart Grid, 2014, 5, pp. 2173–2182
This work was supported by the National Natural Science [25] Ding, T., Lin, Y., Li, G., et al.: ‘A new model for resilient distribution systems
Foundation of China (51807149) and the China Postdoctoral by microgrids formation’, IEEE Trans. Power Syst., 2017, 32, pp. 4145–4147
Science Foundation (2016M600791). [26] Lavorato, M., Franco, J.F., Rider, M.J., et al.: ‘Imposing radiality constraints
in distribution system optimization problems’, IEEE Trans. Power Syst.,
2013, 28, p. 568

IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 16, pp. 3534-3548 3547
© The Institution of Engineering and Technology 2019
[27] Jabr, R.A., Singh, R., Pal, B.C.: ‘Minimum loss network reconfiguration [32] Balinski, M., Tucker, A.: ‘Duality theory of linear programs: a constructive
using mixed-integer convex programming’, IEEE Trans. Power Syst., 2012, approach with applications’, SIAM Rev., 1969, 11, pp. 347–377
27, pp. 1106–1115 [33] Kersting, W.H.: ‘Radial distribution test feeders’, IEEE Transactions on
[28] Balakrishnan, R., Ranganathan, K.: ‘A textbook of graph theory’, in Axler, S., Power Systems, 2001, 6, pp. 975–985
Capasso, V., Casacuberta, C., et al. (Eds.): ‘Mathematical gazette’, vol 84 [34] Zou, K., Agalgaonkar, A.P., Muttaqi, K.M., et al.: ‘Distribution system
(Springer, New York, NY, USA, 2000) planning with incorporating DG reactive capability and system uncertainties’,
[29] Zeng, B., Zhao, L.: ‘Solving two-stage robust optimization problems using a IEEE Trans. Sustain. Energy, 2012, 3, pp. 112–123
column-and- constraint generation method’, Oper. Res. Lett., 2013, 41, pp. [35] Zhang, J., Cheng, H., Wang, C.: ‘Technical and economic impacts of active
457–461 management on distribution network’, Int. J. Electr. Power Energy Syst.,
[30] Zhao, L., Zeng, B.: ‘An exact algorithm for two-stage robust optimization 2009, 31, pp. 130–138
with mixed integer recourse problems’, University of South Florida, 2012 [36] Cao, X., Wang, J., Zeng, B.: ‘A chance constrained information-gap decision
[31] Boyd, S., Vandenberghe, L.: ‘Convex optimization’ (Cambridge University model for multi-period microgrid planning’, IEEE Trans. Power Syst., 2018,
Press, Cambridge, UK, 2004) 33, pp. 2684–2695

3548 IET Gener. Transm. Distrib., 2019, Vol. 13 Iss. 16, pp. 3534-3548
© The Institution of Engineering and Technology 2019

You might also like