1524 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 6, NO.

4, OCTOBER 2015

Contingency Assessment and Network

Reconfiguration in Distribution Grids Including
Wind Power and Energy Storage
Pilar Meneses de Quevedo, Javier Contreras, Fellow, IEEE, Marcos J. Rider, Member, IEEE, and Javad Allahdadian

Abstract—In case of abnormal conditions, distribution systems C ss Storage scheduled cost ($/MWh).
should be reconfigured to overcome the impacts of outages such C rp Production scheduled reserve cost ($/MWh).
as overloads of network components and increased power losses. C rs Storage scheduled reserve cost ($/MWh).
For this purpose, energy storage systems (ESS) and renewable
energy sources (RES) can be applied to improve operating con- Ĉ rs Real-time energy storage cost ($/MWh).
ditions. An optimal contingency assessment model using two-stage Ĉ rp Real-time energy production cost ($/MWh).
stochastic linear programming including wind power generation Cij State of a contingency in branch ij.
and a generic ESS is presented. The optimization model is applied ditω Real power demand at node i (MW).
to find the best radial topology by determining the best switching qitω Reactive power demand at node i (MVAr).
sequence to solve contingencies. The proposed model is applied to
a 69-node distribution system and the results of all possible con- iniij Initial state of a switch in branch ij.
tingencies in the network are examined considering three different I ij Maximum current flow in branch ij (A).
case studies with several scenarios. In addition, a reconfiguration mijrtω Slope of the rth block for branch ij.
analysis including all the contingencies is presented for the case Nloop Number of switches in a loop.
studies. PFi Power factor (PF) of the load at node i.
fore
Index Terms—Contingency analysis (CA), distributed genera- Pitω Wind real power forecast at node i (MW).
tion (DG), islanding, optimal power flow, stochastic programming, tot
R Total number of blocks in the piecewise
storage devices. linearization.
ssi , ssi Minimum/maximum power storage at node I
N OTATION (MW).
Indexes: spi , spi Minimum/maximum production at node i (MW).
i, j, k Node index. Rij Resistance of branch ij (Ω).
r Block index of linearization. Xij Reactance of branch ij (Ω).
t Real-time period index on a 10-min basis. Zij Impedance of branch ij (Ω).
ω Scenario index. Vnom Nominal voltage of the distribution network (kV).
V i, V i Minimum/maximum voltage at node i (kV).
W Upper bound of variable Wijtω (kV).
Parameters:
xsi0 Initial energy level of storage unit at node i (MWh).
C sw Switching cost ($).
xi , xi Min/max storage capacity at node i (MWh).
C loss Real power losses cost ($/MWh).
ηis , ηip Efficiency storage/production rates.
C w_curt Wind curtailment cost ($/MWh).
ΔSijrtω Upper bound for the rth block of the power flow
C d_curt Real power demand curtailment cost ($/MWh).
through branch ij.
C sub Cost of real power from substation ($/MWh).
Δ Real-time period (10 min) (h).
C sp Production scheduled cost ($/MWh).
Manuscript received February 23, 2015; revised May 19, 2015; accepted Nonnegative variables:
July 05, 2015. Date of publication August 27, 2015; date of current version 2
Iijtω Square of the current flow of branch ij (A2 ).
September 16, 2015. This work was supported in part by the European Union wind
7th Framework Programme under Grant 309048, project SiNGULAR, and in
Pitω Real power wind generation at node i (MW).
w_curt
part by the Ministry of Science of Spain under CICYT Project ENE2012- Pitω Real power wind curtailment at node i (MW).
dcurt
30679. Paper no. TSTE-00134-2015. Pitω Real power demand curtailment at node i (MW).
P. Meneses de Quevedo and J. Contreras are with E.T.S. de Ingenieros 2
Pijtω Square of the real power in branch ij (MW2 ).
Industriales, University of Castilla–La Mancha, 13071 Ciudad Real, Spain
(e-mail: [email protected]; [email protected]). Q2ijtω Square of the reactive power in branch ij (MVAr2 ).
M. J. Rider is with the Department of Systems and Energy, University +
Pijtω Real power flow in branch ij, downstream (MW).
of Campinas, 13083-852 Campinas, Brazil (e-mail: [email protected]. −
unicamp.br). Pijtω Real power flow in branch ij, upstream (MW).
+
J. Allahdadian is with the Department of Energy (DENERG), Politecnico di Qijtω Reactive power flow (downstream) (MVAr).
Q−
Torino, 10129 Turin, Italy (e-mail: [email protected]).
Color versions of one or more of the figures in this paper are available online ijtω Reactive power flow (upstream) (MVAr).
at http://ieeexplore.ieee.org. rip , ris Scheduled power production/storage reserve
Digital Object Identifier 10.1109/TSTE.2015.2453368 (MW).
1949-3029 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on January 14,2025 at 15:19:08 UTC from IEEE Xplore. Restrictions apply.
 MENESES DE QUEVEDO et al.: CONTINGENCY ASSESSMENT AND NETWORK RECONFIGURATION IN DISTRIBUTION GRIDS 1525

p s
r̂itω , r̂itω Real-time production/storage reserve (MW). resources near loads and compensate energy demand, prob-
p s
si , si Scheduled power production/storage (MW). lems may arise due to the intermittency and uncertainty of RES
2
Vitω Square of the voltage magnitude at node i (kV2 ). [4], [25] and the integration of ESS into electrical distribution
2
Wijtω Square of the voltage drop for branch ij (kV2 ). systems (EDS) [26], [27].
x̂itω Storage level at node i (MWh). A mixed integer linear programming (MILP) reconfigura-
ΔPijrtω Value of the rth block of real power (MW). tion model under an N − 1 reliability criterion using stochastic
ΔQijrtω Value of the rth block of reactive power (MVAr). programming considering wind energy and ESS in EDS is
introduced. The steady-state operation of a radial EDS is com-
Free variables: plicated to model linearly; hence, an alternative current (ac)
sub
Pitω Real power of a substation at node i (MW). flow is approximated through linear expressions. The proposed
Qsub Reactive power of a substation at node i (MVAr). model also includes switching operation, intermittent RES, and
itω
Qwind Reactive power of wind generation at node i generic ESS. The analysis of N − 1 contingencies is performed
itω
(MVAr). using a contingency parameter per branch to locate the fault in
Qditω
curt
Demand reactive power curtailment (MVAr). the branch.
Due to several technical reasons such as low cost opera-
Binary variables: tion, simplicity of analysis and coordination, and reduction of
P+
vijtω Variable related to real power (upstream). short circuit current, EDS must operate with a radial topology
P− (even with a meshed structure). In addition, MILP is applied,
vijtω Variable related to real power (downstream). in this paper, due to the following advantages: 1) the math-
Q+
vijtω Variable related to reactive power (upstream). ematical model is robust; 2) the computational behavior of
Q−
vijtω Variable related to reactive power (downstream). a linear solver is more efficient than nonlinear solvers; and
s p
vit , vit Variables related to storage or production. 3) using classical optimization techniques, convergence can be
yij State of the switch in branch ij. guaranteed.
The main contributions of this paper are as follows.
1) Regarding the methodology, a two-stage stochastic math-
ematical MILP for contingency response is introduced
I. I NTRODUCTION
to consider the inconsistency and intermittency of the

C ONTINGENCY analysis (CA) provides tools to study

contingencies in electrical networks and analyze their
behavior in case of outages in one of the electrical com-
renewable power sources.
2) From a modeling perspective, a joint programming model
of the optimal reconfiguration and contingency evalua-
ponents such as lines, transformers, and generators. Several tion is presented here for a distribution system. However,
studies have been carried out to assess contingencies in elec- in the literature, optimal switching under contingency
trical networks. The comparison of some techniques such as conditions is reported only for transmission networks
ranking, screening, bounding, and fast power flow techniques [11]–[13].
has been done in [1] and [2]. Methods such as genetic algo- 3) The benefits of combining RES with ESS in distribution
rithms [3], [4], bi-level programming [5], and other new power networks, especially under contingencies and abnormal
flow calculation approaches [6] have been developed to solve conditions, are analyzed.
CA problems. Likewise, in terms of behavior assessment of Another relevant aspect of the presented optimization model
power networks, some works have been presented to evaluate is the minimization of the overall operation costs, including
voltage security [7], [8], operation of transmission networks switching and demand and wind curtailment costs. Note that
[9], [10], and optimal switching under N − 1 contingencies the problem formulation assumes that the overall investment
[11]–[13]. costs have already been minimized in a previous planning
Moreover, reconfiguration of distribution systems (RDS) phase where the best locations for wind and storage have been
aims at finding the best topology of the system concerning selected in advance.
power losses, energy demand, and operational performance. In This paper is organized as follows. Section II describes the
general, two major groups of optimization methods have been mathematical formulation of generation and storage models.
represented to solve the RDS problem: 1) exact techniques Section III defines the mathematical model as a stochastic
and 2) heuristic and metaheuristic techniques. In literature, MILP. The main results of the case studies considering contin-
exact techniques such as branch and bound are only used to gencies and reconfigurations are shown in Section IV. Finally,
solve relaxed models [14]–[18]. On the contrary, heuristic and conclusion is presented in Section V.
metaheuristic methods such as ant colony [19], [20], genetic
algorithms [21]–[23], and particle swarm optimization [24]
have been applied for complete models.
In particular, RDS can be used to improve the operation II. S TORAGE M ODELING
of distribution systems in case of contingencies, considering In modern power systems, storage devices have grown
renewable energy sources (RES) and energy storage systems rapidly. ESSs are integrated into EDS to consider several pur-
(ESS) in electrical networks. Although the availability of RES poses such as meeting real-time power demand, smoothing
and ESS can be seen as an opportunity to exploit available output power of RES, improving power system reliability,
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on January 14,2025 at 15:19:08 UTC from IEEE Xplore. Restrictions apply.
 1526 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 6, NO. 4, OCTOBER 2015

and being economically efficient [28]. Thus, a generic storage

system is modeled as follows:

ssi ≤ ssi + ris ≤ ssi vit

s
(1)

spi ≤ spi + rip ≤ spi vit

p
(2)
p
0 ≤ r̂itω ≤ rip (3)

0 ≤ r̂itω
s
≤ ris (4)
Fig. 1. Illustrative radial distribution system.
xi ≤ x̂itω ≤ xi (5)
A. Objective Function
x̂itω = x̂it−1ω + Δ [ηis ssi − (1/ηip ) spi ] (6)
The objective function is formulated through two-stage
+ Δ [ηis r̂itω
s
− (1/ηip ) r̂itω
p
] stochastic programming as shown in (9)–(13). At the first
stage, regardless of scenarios, i.e., only considering contin-
x̂it=0ω = xs0 (7) gency conditions, the cost of opening and closing switches
p and the scheduling of storage units are assigned. At the sec-
vit + vit
s
≤ 1. (8)
ond stage, the expected values of the total cost of power losses
in the network, the cost of real-time production/storage of the
The minimum and maximum storage limits (charge) and pro- storage units, and the cost of curtailment of both generation
ductions (injection into the network) are defined in (1) and (2). and demand, with respect to different scenarios and contin-
It is worth noting that ssi and spi are implemented in these equa- gency conditions, are assigned. At the first stage, the variables
tions to smooth the output power of the storage unit and ris /rip are only related to contingency conditions and variables are
to guarantee that real-time power balance is within the storage relevant to both contingencies and scenarios at the second
capacity. In (3) and (4), the actual real-time reserves are con- stage.
fined to the maximum reserve bounds ris /rip . The upper and Objective function:
lower bounds of energy capacity in storage units are defined
by (5). The ESS transition function is represented by (6). min φ + γ + E [ψ (ω)] + E [K (ω)] (9)
Equation (7) limits the remaining energy in the storage unit.
p s
Finally, in (8), binary variables vit and vit are defined and where the first stage is given by φ + γ:
added to (1) and (2) to avoid producing and storing energy 1) Switching cost
simultaneously.  
φ= yij C sw + (1 − yij ) C sw . (10)
ij ij
iniij = 0 iniij = 1
III. S TOCHASTIC F ORMULATION OF C ONTINGENCY
A SSESSMENT AND N ETWORK R ECONFIGURATION 2) Scheduled production/storage cost function
The following assumptions are defined to represent the sim- 
plified operation of EDS including switches, generation, and γ= [ssi C ss + spi C sp + ris C rs + rip C rp ] . (11)
i
storage devices.
1) The load pattern in EDS is altered every 10 min. Next, the second stage is given by ψ (ω) + K (ω):
2) EDS are balanced three-phase systems and can be repre- 1) Network costs (losses and substation costs)
sented by an equivalent single-phase circuit.     
2
3) Storage energy losses are ignored. ψ (ω) = Δ Rij Iijtω C loss + sub sub
Pitω C .
t ij t i
4) Storage and production of energy occur at constant power
(12)
in storage units during real-time periods (10 min).
5) The location of storage units has already been defined in 2) Corrective action costs (curtailment and real-time produc-
the planning phase. tion/storage costs)
6) Wind turbines are located at nodes with high demand.   
7) The maximum contingency duration is 1 h, considering K (ω) = Δ w_curt w_curt
Pitω C + Pitω
d_curt d_curt
C
t i
wind scenarios, which is divided into 10-min periods (six
periods). 
p
+ r̂itω Ĉ rp + r̂itω
s
Ĉ rs . (13)
8) Faults are considered in feeders and circuit breakers.
9) Islanded radial operation of the system is possible with
distributed generation (DG) integration.
B. Constraints
To distinguish the direction (sense) of the current and power
flow (forward or backward), two positive separate variables are The above objective function is subject to a set of constraints
shown in Fig. 1. to assure optimal operational conditions.
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on January 14,2025 at 15:19:08 UTC from IEEE Xplore. Restrictions apply.
 MENESES DE QUEVEDO et al.: CONTINGENCY ASSESSMENT AND NETWORK RECONFIGURATION IN DISTRIBUTION GRIDS 1527

1) Real-Time Power Balance Equations: Real power of binary variables to avoid considering forward and backward
wind including curtailment is represented in (14) power flows simultaneously. The constraints (23)–(27) define
the limits through switching devices in branches if they are
wind
Pitω = Pitω
fore
− Pitω
wcurt
. (14) closed; otherwise, all magnitudes are equal to zero. Note that
(24)–(27) are auxiliary constraints to improve the convergence
Equations (15) and (16) represent the real and reactive power
of the proposed model
balances at node i, respectively (see Fig. 1). For both active
and reactive power, wind power generation, scheduled produc- 2 2
0 ≤ Iijtω ≤ I ij Cij yij (23)
tion/storage, real-time production/storage power of ESS, the
power flow from/to the substation, and power demand curtail- + P+
Pijtω ≤ Vnom I ij vijtω (24)
ment are considered. In addition, wind generation curtailment
has already been formulated in (14) − P−
Pijtω ≤ Vnom I ij vijtω (25)
     + 
+ − −
wind
Pitω + Pjitω − Pjitω − Piktω − Piktω
j k Q+ Q+
ijtω ≤ Vnom I ij vijtω (26)
2 p
+Rij Iijtω + sub
Pitω + (r̂itω − s
r̂itω ) + (spi − ssi )
Q− Q−
ijtω ≤ Vnom I ij vijtω (27)
= ditω − dcurt
Pitω (15)
   P+ P−
Q+ − vijtω + vijtω ≤ yij (28)
itω +
Qwind jitω − Qjitω
j
   Q+ Q−
− Q+ − 2
iktω − Qiktω + Xij Iijtω + Qitω
sub vijtω + vijtω ≤ yij . (29)
k
= qitω − Qditω
curt
. (16)
C. Radial Configuration
2) Load PF: The following constraint is considered to keep
the initial load PF: The configuration of the network is radial by introducing
constraint (30), i.e., the number of closed switches in any
Qditω
curt
= Pitω
dcurt
∗ (tan (arc cos (PFi ))) . (17) loop has to be less than the total number of switches in that
loop. Therefore, by having at least one open branch in a loop,
3) Voltage Drop Equations: The square value of the volt- only a radial configuration of the network is feasible with the
age drop between nodes is represented in (18) as an auxiliary possibility of having one or more radial island(s) [29], [30]
2
variable Wijtω , which is related to switching operations and
contingencies 
Cij yij ≤ Nloop − 1. (30)
  + −
   ij
2
Vitω − 2 Rij Pijtω − Pijtω + Xij Q+ −
ijtω − Qijtω
2 2 2 2 Obviously, a branch between nodes i and j is disconnected in
− Zij Iijtω − Vjtω + Wijtω = 0. (18)
case of a contingency in that branch. Therefore, binary param-
The upper and lower bounds of the square of the voltage eter Cij is added to (30) to show that the branch is open in the
deviation for node i are defined by (19) mentioned case.
2
V 2 ≤ Vitω
2
≤V . (19)
D. Linearization Procedure
2
In constraints (20) and (21), Wijtω = 0 in case of operation
The applied equations in the optimization problem are lin-
of branch ij (Cij = 1) or switch closure (yij = 1). To satisfy
2 ear, excluding (22), which makes the model nonlinear. To cope
constraints (15) and (16), a proper value for parameter W with this problem, a linear approximation is proposed [31].
2
should be set to give a sufficient degree of freedom for Wijtω Both the left and right sides of (22) are nonlinear and both
2 2
2 should be linearized separately. Note that Vjtω and Iijtω are
Wijtω ≥ −W̄ 2 (1 − yij ) (1 − Cij ) (20)
variables that represent the square magnitude values of voltages
2 2
Wijtω ≤ W̄ (1 − yij ) (1 − Cij ) (21) and currents, respectively, and they are used in (15)–(19), (22),
and (23). In the following, the linearization process of (22) is
4) Nonlinear Apparent Power Equation: The current flow described.
magnitude calculation is represented by nonlinear constraint 2
1) Vjtω 2
Iijtω : The product of two variables is linearized by
(22), where both sides of the constraint are linearized as discretizing Vitω2
in small intervals. However, this leads to
explained in Section III-D an increase in the number of binary variables and compu-
2
Vjtω 2
Iijtω 2
= Pijtω + Q2ijtω . (22) tation time. Since the voltage magnitude is within small
2
range in EDS, a constant value Vnom is selected and sub-
2
5) Current and Power Magnitude Limits: Due to thermal stituted for Vjtω in (22) for the first iteration. Then, the
2
limits in EDS, a set of constraints is introduced for current, model is run again and Vjtω takes the value resulting from
real power, and reactive power in (23), (24)–(25), and (26)– 2
the first iteration. Note that Vjtω hardly changes after the
P+ P− Q+ Q−
(27), respectively. In addition, vijtω , vijtω , vijtω , and vijtω are second iteration.
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on January 14,2025 at 15:19:08 UTC from IEEE Xplore. Restrictions apply.
 1528 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 6, NO. 4, OCTOBER 2015

2
Fig. 2. Modeling the piecewise linear function Pijtω . Fig. 3. 69-node network topology under a contingency in branch 3-29.

TABLE I
2
2) Pijtω + Q2ijtω : The linearization of both terms on the S WITCH DATA FOR THE 69-N ODE S YSTEM
right side of (22) is carried out by a piecewise linear
approximation, as follows:

2
Pijtω + Q2ijtω = (mijrtω ΔPijrtω )
r
 TABLE II
+ (mijrtω ΔQijrtω ) (31) C OST DATA FOR 69-N ODE S YSTEM ($/MW H )
r

+ −
Pjitω + Pjitω = ΔPijrtω (32)
r

Q+ −
jitω + Qjitω = ΔQijrtω (33)
r
0 ≤ ΔPijrtω ≤ ΔSijrtω (34) depicts the 69-node network where the green nodes are not con-
0 ≤ ΔQijrtω ≤ ΔSijrtω (35) nected to loads and the locations of wind turbines and storage
units are represented. The network is connected to a substation
where at node 1. The network contains 74 branches, 5 sectionalizing
mijrtω = (2r − 1) ΔSijrtω (36) switches, and 5 tie-switches as shown in Table I, considering
  negligible opening and closing operation costs.
ΔSijrtω = Vnom I ij /Rtot (37) Other network costs are shown in Table II. The maximum
Note that mijrtω and ΔSijrtω are constant parameters and current flow in the branches is 150 A, excluding branches L1−2
(32) and (33) are a set of linear expressions. Likewise, (31) is (line from nodes 1 to 2), L2−3 , and L3−4 with a maximum
a linear approximation of (Pijtω2
+ Q2ijtω ) and, (32) and (33) current limit of 100 A. Since the voltage values vary through
iterations, the upper and lower voltage limits are defined as 1.1
represent that (Pjitω + Pjitω ) and (Q+
+ − −
jitω + Qjitω ) are equal to and 0.9 p.u., respectively. In the linearization, 20 discrete blocks
the sum of the values in each section of the discretization. As
are considered. The total real power demand is 1.103 MW and
shown in (31)–(33) and Fig. 2, the right side of (22) can be
the total time horizon is 1 h divided into 10 min periods. The
replaced with the right side of (31) to form a linear equation
demand data, recorded every 10 min, is collected from the
 
2
Vnom 2
Iijtω = (mijrtω ΔPijrtω ) + (mijrtω ΔPijrtω ). Iberian Electricity Market [33], on March 20, 2014, and the
r r contingency is assumed to occur at 4 A . M . Moreover, wind
(38) turbines are located at nodes 12, 34, and 62 with a produc-
The linear form of (22) is shown in (38). In this equa- tion share of 16%, 34%, and 50%, respectively, and the total
2 installed wind generation capacity is 0.45 MW. Two equal stor-
tion, Vnom is constant and r (mijrtω ΔPijrtω ) and r
(mijrtω ΔQijrtω ) are linear. The linearization of the active age units are located at nodes 30 and 51, with a capacity of
power flow is shown in Fig. 2. 0.15 MW allocated to each, and their initial energy levels are
set to 50% of their capacities. The upper and lower limits of
power storage/production in ESS are 0.075 MW.
IV. C ASE S TUDIES
A. Network Overview B. Wind Production Scenarios
To evaluate the behavior of the proposed model, a 69-node The exploitation of RES, especially wind, has had an
network is considered and network data come from [32]. Fig. 3 upward trend leading to intermittency and uncertainty of power
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on January 14,2025 at 15:19:08 UTC from IEEE Xplore. Restrictions apply.
 MENESES DE QUEVEDO et al.: CONTINGENCY ASSESSMENT AND NETWORK RECONFIGURATION IN DISTRIBUTION GRIDS 1529

TABLE III
N − 1 C ONTINGENCY C URTAILMENT C OST A NALYSIS ($)

Fig. 4. Three different wind power point forecast and stochastic scenarios.

generation. In order to analyze the network under different

conditions, three different stochastic levels of wind genera-
tion are introduced (low, mean, and high) as 1.5%, 16%, and
34% of the total demand. In this regard, to represent intermit-
tency of wind power, 200 wind scenarios have been generated
randomly by implementing Monte Carlo simulation under a
cone of uncertainty, with the maximum error of 15% in 1 h
for each forecast, as applied in [34] and [35]. The number of
wind scenarios has been limited to 10 to minimize the dimen-
sionality in the problem formulation maintaining accuracy in
the results, by implementing a clustering mechanism, the K-
means method, taking into account sequences of wind values
every 10 min (six values) per scenario. This method can be TABLE IV
used to partition a given set of scenarios into a given number N − 1 C ONTINGENCY L OSS A NALYSIS
of clusters. As a result of this partition, scenarios with similar
features are assigned to the same cluster. The centroid of each
cluster represents a somewhat average pattern of all the sce-
narios included in a cluster. Since this centroid is artificial, the
original scenario with the lowest probability distance from the
centroid is used to represent the cluster [36]. Three levels of
wind power generation and their corresponding scenarios are
illustrated in Fig. 4. Due to the different PF of loads, commonly
between 0.7 and 0.9, reactive power compensation is consid-
ered in wind turbines to keep grid security especially in case of
islanding.

C. Contingency Analysis
In order to analyze the benefits of introducing renewable
generation and storage units in EDS, three case studies under
an N − 1 criterion are represented: 1) CA without RES and
ESS; 2) CA including RES; and 3) CA including RES and losses are more than the other case studies due to the location
ESS. This is performed by implementing a contingency param- of loads at remote nodes. Also, in some situations, a contin-
eter Cij per branch to locate the fault in the network. The gency leads to the isolation of the network from the substation,
results are shown in terms of demand and wind curtailment cost e.g., C1−2 (a contingency in the branch between nodes 1 and
(Table III), active power losses (Table IV), and production and 2) or C2−3 resulting in a blackout. Regarding power losses,
storage cost (Table V). some selected results are shown in Table IV. For case study
1) Case Study 1—CA Without RES and ESS: In this case, 1, the highest amount is related to C60−61 and C61−62 . In this
the only power resource is the power injection from the sub- case, due to the significant amount of load at node 62, which is
station to the disturbed system. Consequently, the total power far from the power source, power losses increase dramatically.
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on January 14,2025 at 15:19:08 UTC from IEEE Xplore. Restrictions apply.
 1530 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 6, NO. 4, OCTOBER 2015

TABLE V
N − 1 C ONTINGENCY S TORAGE C OST A NALYSIS ($)

Fig. 5. Energy production/storage under a contingency in branch 3-29.

(a) Energy stored (p.u.). (b) Power produced/stored (p.u.).

the level of wind production (from a low level to higher lev-

els), the cost of wind curtailment rises. On the other hand, with
more wind power production, more loads can be fed, thus, the
In addition, demand curtailment (see Table III), especially for cost of demand curtailment drops.
C60−61 and C61−62 , is remarkable due to the high amount 3) Case Study 3—CA Including RES and ESS: Using mean
of demand at node 62 and the power flow limitation of the wind production level and compared with case study 2, ESS
connection branches. reduces power losses, wind and demand curtailment slightly.
2) Case Study 2—CA Including RES: An obvious differ- As illustrated in Table IV, excluding C1−2 , C2−3 , and C3−4 , in
ence with respect to case study 1 is the reduction of power the other cases, power losses are reduced moderately. In spite
losses. As shown in Table IV, real power losses drop roughly of a fair growth of power losses for C1−2 , C2−3 , and C3−4 (see
50% for most contingencies considering the mean level of Table IV), the reduction of demand curtailment is noticeable, as
wind generation. The reduction of power losses is expected shown in Table III. Storage units improve the operation of the
due to the availability of wind power for local loads. Although network, particularly in case of islanding for the whole or a part
the maximum power losses are relevant to contingency C4−5 , of the network.
power losses for C60−61 and C61−62 are still remarkable. In As shown in Table III, in cases with islanding, i.e., C1−2 ,
case of contingency C4−5 , the main connection branch to C2−3 , and C3−4 , the total load connected to the network is
transfer power from the substation to the remote loads is discon- increased compare with case study 2. Likewise, for C3−29 , the
nected and power losses are unavoidable, however, as shown in total wind generation capacity is exploited in the presence of a
Table III, demand curtailment is zero, in this case. Moreover, storage unit in the isolated area.
total demand curtailment is reduced in some cases, especially In extreme situations with a high level of wind in islanding
for contingencies C60−61 and C61−62 , due to the availability of areas, as in C3−29 and C29−30 , the storage units store the extra
wind power for loads at node 62. power production to avoid demand curtailment and minimize
In case of low wind generation, the amount of demand cur- their costs.
tailment cost and power losses are similar to Case 1, as wind Since demand is higher than generation in some isolated
power generation feeds very few loads. (C1−2 , C2−3 , and C3−4 ) areas, with a higher level of wind
In addition, C1−2 , C2−3 , and C3−4 create an island for generation, wind and demand curtailments decrease.
the whole network and, despite demand curtailment reduction As presented in Table IV, with a higher level of wind produc-
under these conditions (unlike case study 1), the main part of tion, excluding islanding conditions, the power losses decline
the demand is not served yet, due to a lack of real power gener- due to the availability of local power for loads.
ation. Likewise, an island appears in case of contingency C3−29 Fig. 5 shows the behavior of the storage units with mean
for a part of the network, as shown in Fig. 3. Here, the mean wind production in case of contingency C3−29 . In Fig. 5(a), the
level of wind generation is able to meet the demand in this energy level of the storage units is illustrated. As C3−29 creates
area, however, wind curtailment is inevitable due to the imbal- an isolated area with more generation than load, the storage unit
ance of load and generation. Lastly, since nodes 31, 32, and in that section (node 30) is scheduled to store energy to avoid
33 are not connected to any load, wind curtailment is equal for wind curtailment. On the other hand, the storage unit at node 51
contingencies C30−31 , C31−32 , C32−33 , and C33−34 , as shown is scheduled to produce and compensate the lack of real power.
in Table III. Since the demand is constant in islanding situ- Fig. 5(b) shows the real-time power storage of unit 30 (green
ations (C1−2 , C2−3 , C3−4 , C3−29 , and C29−30 ) by increasing line). Real-time production/storage is represented when the
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on January 14,2025 at 15:19:08 UTC from IEEE Xplore. Restrictions apply.
 MENESES DE QUEVEDO et al.: CONTINGENCY ASSESSMENT AND NETWORK RECONFIGURATION IN DISTRIBUTION GRIDS 1531

TABLE VI TABLE VII

R ECONFIGURATION A NALYSIS : O PEN S WITCHES C OMPUTATION T IME ( S )

and the other having wind and storage. Therefore, in case of

disconnection of the EDS from the external grid, the optimiza-
tion problem can manage the EDS to be energized avoiding a
blackout in the isolated part. Noteworthy, demand and genera-
tion curtailment is reduced by having separated radial islands
in areas isolated from the substation. In order to feed loads
next to the contingency and to avoid demand curtailment, the
optimization problem decides to close the switches near loads.
On the other hand, contingencies C57−58 , C58−59 , C59−60 , and
C56−57 occur near to nodes without any connected loads and
the switches in these areas remain open under any condition.
In addition, the location of both, generation and contin-
gency, influences the RDS. For instance, S31−32 and S51−60
are always closed, as they are near to generation and storage
units. Finally, reconfiguration of the EDS is not changed for
contingencies in branches close to each other, as in cases L5−51
(feeder from nodes 5 to 51), L7−10 , L12−68 , L13−70 , L24−28 ,
L29−36 , and L37−44 .
All case studies have been solved using CPLEX 11 solver in
GAMS [37]. An Intel Xeon E7-4820 computer with four pro-
cessors at 2 GHz and 128 GB of RAM has been used. MATLAB
[38] has been used to implement scenario reduction using K-
means. Table VII illustrates the computation time of every case
study assuming a mean wind power and for all the contingen-
scheduled production/storage of energy is not sufficient to com- cies. Note that, the CPU times in Table VII correspond to the
pensate the power imbalance with mean wind production of the whole set of contingencies for each case, which are solved one
network, as it occurs for storage unit 30. by one. It could be possible to reduce the CPU time by analyz-
Table V contains the costs of the storage units per contin- ing the branches, which are more likely to be out of operation.
gency. The total costs of the storage units are lower than the The CPU time could be further reduced by considering a lower
other costs in the network and, in that regard, they are effi- number of linearization blocks.
cient. With high wind production, in C3−29 and C29−30 , the
storage units in isolated areas store more in real time to avoid
wind curtailment. On the other hand, with low wind power, V. C ONCLUSION
the ESS units produce in real time to compensate the power In this paper, a two-stage stochastic MILP reconfiguration
imbalance. model under N − 1 contingency conditions, considering wind
energy and ESS with the possibility of having isolated radial
grids in distribution systems, has been investigated.
D. Reconfiguration Analysis
An exhaustive analysis has been done to evaluate the model
Network reconfiguration is important to analyze to meet and, as a result, the model has been successfully applied to a
some goals such as reduction of power losses and isolation distribution network to minimize total costs. Following that, the
of faults in the EDS. In this paper, reconfiguration is carried work shows the advantages of having energy storage units and
out under contingency conditions to reach the optimal point renewable energy in terms of real-time operation under contin-
of the optimization problem. As mentioned before, Table I gencies in distribution networks. Consequently, the penetration
shows the status of the switches in the EDS. Initially, under of storage devices and wind generation improves the opera-
normal conditions, S12−44 (switch between nodes 12 and 44), tion of the network under contingency conditions and reduces
S16−47 , S19−20 , S28−66 , and S56−57 are normally open and power losses and demand and generation curtailment. In addi-
others are normally closed. The status of the open switches is tion, this method improves reliability, allows for an optimal
represented for case studies 2 and 3 (for the mean value of wind) radial reconfiguration, radial islanded operation, and the reduc-
in Table VI, which is the same for both cases. tion of real power losses. Storage units improve the operation
In case of contingencies C1−2 , C2−3 , and C3−4 , the same of the network, especially in case of islanding for the whole
reconfiguration is defined by the optimization problem split- network or a part of it. The location of storage units is impor-
ting the EDS into two radial islands, one having wind only, tant to compensate local power, i.e., power injection to local
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on January 14,2025 at 15:19:08 UTC from IEEE Xplore. Restrictions apply.
 1532 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 6, NO. 4, OCTOBER 2015

demand and absorption of additional power from local gen- [21] K. Prasad, R. Ranjan, N. C. Sahoo, and A. Chaturvedi, “Optimal recon-
eration. In conclusion, the proposed model can be a valuable figuration of radial distribution systems using a fuzzy mutated genetic
algorithm,” IEEE Trans. Power Del., vol. 20, no. 2, pp. 1211–1213, Apr.
tool for an electrical distribution company to optimally recon- 2005.
figure the system by evaluating all possible contingencies of the [22] J. Mendoza, R. López, D. Morales, E. López, P. Dessante, and R. Moraga,
network using wind power and energy storage. “Minimal loss reconfiguration using genetic algorithms with restricted
population and addressed operators: Real application,” IEEE Trans.
Power Syst., vol. 21, no. 2, pp. 948–954, May 2006.
[23] E. Carreño, R. Romero, and A. Padilha-Feltrin, “An efficient codification
R EFERENCES to solve distribution network reconfiguration for loss reduction problem,”
IEEE Trans. Power Syst., vol. 23, no. 4, pp. 1542–1551, Nov. 2008.
[1] F. M. Echavarren, E. Lobato, and L. Rouco, “Contingency analysis: [24] P. Rezaei and M. Vakilian, “Distribution system efficiency improvement
Feasibility identification and calculation algorithm,” IEE Proc. Gener. by reconfiguration and capacitor placement using a modified particle
Transm. Distrib., vol. 152, no. 5, pp. 645–652, Sep. 2005. swarm optimization algorithm,” in Proc. Int. Symp. Mod. Electr. Power
[2] G. K. Stefopoulos, Y. Fang, G. J. Cokkinides, and A. P. S. Meliopoulos, Syst. (MEPS), Sep. 20–22, 2010, pp. 1–6.
“Advanced contingency selection methodology,” in Proc. 37th Annu. [25] R. Billinton and D. Huang, “Incorporating wind power in generat-
North Amer. Power Symp., Oct. 23–25, 2005, pp. 67–73. ing capacity reliability evaluation using different models,” IEEE Trans.
[3] M. H. Sulaiman, O. Aliman, and S. R. A. Rahim, “Contingency analysis Power Syst., vol. 26, no. 4, pp. 2509–2517, Nov. 2011.
of embedded generation deployment in distribution network,” in Proc. 4th [26] Y. M. Atwa and E. F. El-Saadany, “Optimal allocation of ESS in distribu-
Int. Power Eng. Optim. Conf. (PEOCO), Jun. 23–24, 2010, pp. 412–417. tion systems with a high penetration of wind energy,” IEEE Trans. Power
[4] A. Mendes, N. Boland, P. Guiney, and C. Riveros, “(N-1) contingency Syst., vol. 25, no. 4, pp. 1815–1822, Nov. 2010.
planning in radial distribution networks using genetic algorithms,” in [27] J. M. Gantz, S. M. Amin, and A. M. Giacomoni, “Optimal mix and
Proc. IEEE/PES Transm. Distrib. Conf. Expo. Latin Amer. (T&D-LA), placement of energy storage systems in power distribution networks for
Nov. 8–10, 2010, pp. 290–297. reduced outage costs,” in Proc. IEEE Energy Convers. Congr. Expo.
[5] S. Fliscounakis, P. Panciatici, F. Capitanescu, and L. Wehenkel, (ECCE), Sep. 20, 2012, vol. 15, pp. 2447–2453.
“Contingency ranking with respect to overloads in very large power sys- [28] D. Pozo and J. Contreras, “Unit commitment with ideal and generic
tems taking into account uncertainty, preventive, and corrective actions,” energy storage units,” IEEE Trans. Power Syst., vol. 29, no. 6, pp. 2974–
IEEE Trans. Power Syst., vol. 28, no. 4, pp. 4909–4917, Nov. 2013. 2984, Nov. 2014.
[6] B. Liu and Y. Zhang, “Power flow algorithm and practical contingency [29] B. Moradzadeh and K. Tomsovic, “Reconfiguration of a distribution sys-
analysis for distribution systems with distributed generation,” Eur. Trans. tem with load uncertainty,” in Proc. Int. Conf. Intell. Syst. Appl. Power
Elect. Power, vol. 19, pp. 880–889, 2009. Syst., Tokyo, Japan, Jul. 1–4. 2003, pp. 1–6.
[7] P. A. Ruiz and P. W. Sauer, “Voltage and reactive power estimation [30] B. Moradzadeh and K. Tomsovic, “Mixed integer programming-based
for contingency analysis using sensitivities,” IEEE Trans. Power Syst., reconfiguration of a distribution system with battery storage,” in Proc.
vol. 22, no. 2, pp. 639–647, May 2007. North Amer. Power Symp. (NAPS), Urbana-Champaign, IL, USA, Sep.
[8] F. Milano, C. Cañizares, and M. Invernizzi, “Voltage stability constrained 9–11, 2012, pp. 1–6.
OPF market models considering N-1 contingency criteria,” Electr. Power [31] J. F. Franco, M. J. Rider, M. Lavorato, and R. Romero, “A mixed integer
Syst. Res., vol. 74, pp. 27–36, 2005. LP model for the reconfiguration of radial electric distribution systems
[9] M. F. Akorede, H. Hizam, I. Aris, and M. Z. A. Kadir, “Contingency considering distributed generation,” Electr. Power Syst. Res., vol. 97,
evaluation for voltage security assessment of power systems,” in Proc. pp. 51–60, Apr. 2013.
IEEE Student Conf. Res. Develop. (SCOReD), Nov. 16–18, 2009, [32] H. Chiang and R. Jean-Jumeau, “Optimal network reconfigurations in dis-
pp. 345–348. tribution systems. II. Solution algorithms and numerical results,” IEEE
[10] P. J. M. Lacanina, A. M. Marcolini, and J. L. M. Ramos, “DCOPF con- Trans. Power Del., vol. 5, no. 3, pp. 1568–1574, Jul. 1990.
tingency analysis including phase shifting transformers,” in Proc. 18th [33] Red Eléctrica de España. (2014). Electricity Demand Tracking in Real-
Power Syst. Comput. Conf. (PSCC’14), Wroclaw, Poland, Aug. 18–22, time, Associated Generation Mix and CO2 Emissions [Online]. Available:
2014, pp. 1–4. https://demanda.ree.es/
[11] K. W. Hedman, R. P. O’Neill, E. B. Fisher, and S. S. Oren, “Optimal [34] P. Pinson, G. Papaefthymiou, B. Klockl, and H. A. Nielsen, “Generation
transmission switching with contingency analysis,” IEEE Trans. Power of statistical scenarios of short-term wind power production,” in Proc.
Syst., vol. 24, no. 3, pp. 1577–1586, Aug. 2009. IEEE Power Tech, Lausanne, Switzerland, Jul. 1–5, 2007, pp. 491–496.
[12] K. W. Hedman, M. C. Ferris, R. P. O’Neill, E. B. Fisher, and [35] K. Orwig et al., “Recent trends in variable generation forecasting and its
S. S. Oren, “Co-optimization of generation unit commitment and trans- value to the power system,” IEEE Trans. Sustain. Energy, vol. 6, no. 3,
mission switching with N-1 reliability,” IEEE Trans. Power Syst., vol. 25, pp. 924–933, Jul. 2015.
no. 2, pp. 1052–1063, May 2010. [36] D. Arthur and S. Vassilvitskii, “K-means++: The advantages of careful
[13] G. Ayala and A. Street, “Energy and reserve scheduling with post- seeding,” in Proc. 18th Annu. ACM-SIAM Symp. Discr. Algorithms, New
contingency transmission switching,” Electr. Power Syst. Res., vol. 111, Orleans, LA, USA, 2007, pp. 1027–1035.
pp. 133–140, 2014. [37] A. Brooke, D. Kendrick, A. Meeraus, and R. Raman, GAMS/CPLEX 9.0.
[14] C. S. Chen and M. Y. Cho, “Energy loss reduction by critical switches,” User Notes. Washington, DC, USA: GAMS Development Corp., 2003.
IEEE Trans. Power Del., vol. 8, no. 3, pp. 1246–1253, Jul. 1993. [38] The Mathworks Inc. (2015). Matlab [Online]. Available:
[15] A. Merlin and G. Back, “Search for minimum-loss operational spanning http://www.mathworks.com
tree configuration for an urban power distribution system,” in Proc. 5th
Power Syst. Conf., Cambridge, U.K., 1975, pp. 1–18.
[16] E. R. Ramos, A. G. Expósito, J. R. Santos, and F. L. Iborra, “Path-based
distribution network modeling: Application to reconfiguration for loss Pilar Meneses de Quevedo received the B.S. degree in industrial engi-
reduction,” IEEE Trans. Power Syst., vol. 20, no. 2, pp. 556–564, May neering from the University of Cantabria, Spain, in 2005, and the M.Sc.
2005. degree in the electric power industry from the Universidad Pontificia Comillas,
[17] H. Andrei, I. Caciula, and G. Chicco, “A comprehensive assessment of Madrid, Spain, in 2008. She is currently pursuing the PhD. degree in electrical
the solutions of the distribution system minimum loss reconfiguration engineering at the University of Castilla-La Mancha, Ciudad Real, Spain.
problem,” in Proc. 7th World Energy Syst. Conf., Romania, Europe, Jun. Her research interests include power systems operations, planning, and
30/Jul. 2, 2008, pp. 1–6. electricity markets.
[18] E. Romero-Ramos, J. Riquelme-Santos, and J. Reyes, “A simpler and
exact mathematical model for the computation of the minimal power
losses tree,” Int. J. Electr. Power Syst. Res., vol. 80, pp. 562–571, May Javier Contreras (SM’05–F’15) received the B.S. degree in industrial engi-
2010. neering from the University of Zaragoza, Zaragoza, Spain, the M.Sc. degree in
[19] S. Ching-Tzong, C. Chung-Fu, and C. Ji-Pyng, “Distribution network electrical engineering from the University of Southern California, Los Angeles,
reconfiguration for loss reduction by ant colony search algorithm,” Int. J. CA, USA, and the Ph.D. degree in electrical engineering from the University of
Electr. Power Syst. Res., vol. 75, pp. 190–199, Aug. 2005. California, Berkeley, CA, USA, in 1989, 1992, and 1997, respectively.
[20] C. F. Chang, “Reconfiguration and capacitor placement for loss reduc- He is currently a Full Professor with the University of Castilla–La Mancha,
tion of distribution systems by ant colony search algorithm,” IEEE Trans. Ciudad Real, Spain. His research interests include power systems planning,
Power Syst., vol. 23, no. 4, pp. 1747–1755, Nov. 2008. operations and economics, and electricity markets.
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on January 14,2025 at 15:19:08 UTC from IEEE Xplore. Restrictions apply.
 MENESES DE QUEVEDO et al.: CONTINGENCY ASSESSMENT AND NETWORK RECONFIGURATION IN DISTRIBUTION GRIDS 1533

Marcos J. Rider (S’97–M’06) received the B.Sc. (Hons.) and P.E. degrees from Javad Allahdadian received the M.Sc. degree in electrical engi-
the National University of Engineering, Lima, Perú, in 1999 and 2000, respec- neering from the Universiti Teknologi Malaysia (UTM), Johor
tively, the M.Sc. degree from the Federal University of Maranhão, Maranhão, Bahru, Malaysia, in 2009, and the Ph.D. degree in electri-
Brazil, in 2002, and the Ph.D. degree from the University of Campinas cal engineering from the Politecnico di Milano, Milano, Italy,
(UNICAMP), Campinas, Brazil, in 2006, all in electrical engineering. in 2013.
Currently, he is a Professor with the Department of Systems and Energy, Currently, he is a Researcher with the Department of Energy (DENERG),
UNICAMP. His research interests include development of methodologies Politecnico di Torino, Turin, Italy. His research interests include power system
for the optimization, planning, and control of electrical power systems, and analysis and optimization, power system dynamics and stability, and renewable
applications of artificial intelligence in power systems. energy and smart grids.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on January 14,2025 at 15:19:08 UTC from IEEE Xplore. Restrictions apply.