Voltage Stability Based Placement of Distributed

Generation Against Extreme Events

Ahmet Öner, Ali Abur
Department of Electrical and Computer Engineering
Northeastern University
Boston, MA
aoner@ece.neu.edu, abur@ece.neu.edu

Abstract—This paper is concerned about improving the re- evidence that points to the need to urgently invest in tools
silience of power grids against extreme events which may and strategies to improve the resilience of power grids.
lead to the line and generator outages and subsequent voltage This work considers distributed generators (DG) which are
stability problems and blackouts. The reported study investigates
ways of eliminating or substantially reducing the chances of commonly used in both distribution and transmission systems
having such voltage stability problems during expected extreme and provide both reactive power support as well as alternative
events, by strategically placing a few distributed generators in real power dispatch options in order to address both voltage
the system. The problem is addressed in two stages, where a and line flow limit violations, minimize system losses, improve
reasonably inclusive list of credible contingencies are individually system stability and power quality. There are numerous studies
considered first. A minimum number of distributed generators
are selected and placed in order to maintain voltage stability reported in the literature on DG placement strategies [4]–[9].
under each considered contingency. In the second stage, the The authors of [6], [7] solve the DG placement problem by
number of generators is minimized by the strategic selection considering the voltage stability of the power grid. If more
of locations to reach a solution that ensures voltage stability than one DG is required to be installed, they are selected
under all considered contingencies in the system. Effectiveness and installed one DG at a time, thus not guaranteeing the
and computational performance of the developed strategy are
illustrated by simulating several outage scenarios using the IEEE optimality of the result. The authors of [8] allocate PV
118-bus system. generation and battery storage to restore the system following
an extreme event. Rather than using DGs in restoring the
Index Terms—Distributed generation (DG) placement, extreme
events, resilience, voltage stability system, DGs can be placed strategically ahead of the event
using outage forecasts based on historical outage data. Results
of a preliminary study using this approach are reported in [9].
I. I NTRODUCTION In this paper, this study is further extended by considering a
possible set of outage scenarios and determining a minimum
As the world’s overall energy demand is continually increas- number of required DG placements to maintain system feasi-
ing [1], so do the frequency and impact of extreme events on bility for all considered scenarios.
power grids. There are many power systems which are forced The developed approach aims to eliminate the potential
to operate closer to their limits due to the changing power adverse effects of possible outages by strategic placement of
flow patterns resulting from increased number of renewable new generators. These new DGs are introduced into the system
sources at remote locations. Operators of such systems find it model by switching the bus types from PQ bus to PV bus in the
very challenging to maintain voltage stability during line or problem formulation. In the process, their real power outputs
generator outages caused by such extreme events and look for are arbitrarily set equal to zero in this study. While this yields a
ways of easing this challenge by investing in appropriate tools solution where generators essentially act as synchronous con-
to eliminate or significantly reduce the probability of having densers, the same approach can be applied equally effectively
voltage problems during or after such outages. It should also be to solve the problem with any predetermined real power output
noted that according to the Department of Energy, the average assigned to the placed generators.
annual cost of power outages caused by severe weather is The proposed approach starts out by selecting a set of
estimated to be between $18 billion and $33 billion per year outage scenarios that are credible and significant based on
[2]. Moreover, the frequency of extreme events is expected the historical records of the given power grid. Also, a subset
to increase in the future [3]. Therefore, there is sufficient of the load (or PQ) buses is chosen as candidate locations
for placing new DGs. In this study, all PQ buses are assumed
The authors are grateful for the partial support provided by the NSF /CRISP to belong to this set, but in an actual system it is likely that
Type 2 Grant with Award Number:1638234. This work also made use of some of the load buses may be removed from this list due to
Engineering Research Center shared facilities supported by the Engineering other considerations or physical limitations. Next, a voltage
Research Center Program of the National Science Foundation and the Depart-
ment of Energy under NSF Award Number EEC-1041877 and the CURENT stability constrained optimal power flow (VSCOPF) problem
Industry Partnership Program. is repeatedly solved for each outage scenario by placing a

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
single DG at a time at the candidate PQ bus locations. If a made feasible by introducing a DG into the system, then this
feasible solution is obtained for a DG location, then it will be will also have to be verified. Both of these tasks require a
assigned as a candidate DG for the corresponding outage. If reliable optimization tool which can solve the well studied
among the single DGs considered for a given outage, none of nonlinear optimal power flow problem while ensuring voltage
them manages to yield a feasible solution then the VSCOPF stability under the operating conditions corresponding to the
solution for this specific outage scenario will be attempted obtained solution. In this work, such an optimization problem
to be obtained by using a pair of DGs. This will typically is formulated and solved repeatedly for all considered outage
require a large number of solutions each corresponding to scenarios and assumed DG locations. The details of the
one of the possible combinations of a pair of DG locations problem formulation including the objective function and the
among the chosen set of candidate PQ buses. At the end of constraint equations are provided below.
this process, each outage will be assigned a set of single and 1) Objective Function: The objective function to be mini-
possibly pairs of candidate DG locations. These candidate DG mized is chosen as the total production cost of generators:
locations are collectively saved in a binary matrix relating
outages to corresponding DG candidate buses. This matrix is
then used to formulate an integer programming (IP) problem min C1 PG2 + C2 PG (1)
whose solution will yield the desired numbers and locations
of DGs to be placed in order to maintain feasibility under all where PG is the vector of generator active power outputs, C1
considered outage scenarios. and C2 are the vector cost coefficients.
2) Constraints: There are 5 sets of constraints associated
II. P ROBLEM S TATEMENT with this problem. These constraints not only account for
Given a large power grid and a set of outage scenarios which network operation limits but also ensure voltage stability at
may involve multiple line and/or generator losses, this study the resulting solution. These sets of constraints involve the
is concerned about placing distributed generators at a selected following equations:
subset of candidate load (PQ type) buses in order to maintain a) Power Balance Equations: The first set reflects the
feasibility under each of the considered outage scenarios. power balance equations at each bus expressed for both active
Here, feasibility primarily refers to voltage stability since and reactive power flows. For a given bus i, these equations
such outage scenarios are becoming more commonly observed will be given by:
at the sub-transmission or distribution systems which are
more often subjected to extreme weather events causing such
outages. Hence, the study uses voltage stability constrained nB 
X
optimal power flow as the main tool for restoring feasibility, PG (i) − PD (i) = Gij | Vi || Vj |cos(θij )
but it is worth mentioning that this could be replaced by other j=1
feasibility restoration tools of choice without changing the

+ Bij | Vi || Vj |sin(θij ) (2)
steps of the proposed approach.

III. P ROBLEM F ORMULATION

nB 
The above described problem is formulated and solved in X
QG (i) − QD (i) = Gij | Vi || Vj |sin(θij )
multiple stages involving both nonlinear programming and
j=1
integer programming (IP) tools. The voltage stability con- 
strained optimal power flow part (nonlinear programming part) − Bij | Vi || Vj |cos(θij ) (3)
is utilized to find the binary feasibility decision matrix and
the cost vector considering all the selected outage scenarios. where:
IP solver is used to find an optimal solution for the final set i and j are bus numbers,
of DG placements. The details of voltage stability constrained nB represents the total number of buses considered for DG
optimal power flow problem will be described in section III-A, placement. While it designates all the system buses in this
building the required incidence matrices for the DG placement formulation, in practice buses with existing generation as well
problem will be described in section III-B and finally the use as some other buses which do not lend themselves to DG
of IP method to strategically assign DGs in a given power grid connection will be excluded from this list, making this number
will be explained in section III-C. much less than the total number of buses in the system,
Gij and Bij represent i, jth elements of the real and imaginary
A. Voltage Stability Constrained Optimal Power Flow parts of the bus admittance matrix, respectively,
(VSCOPF) | Vi | and | Vj | represent the voltage magnitudes at bus-i and
The core of the proposed DG placement approach relies bus-j respectively,
heavily on efficiently and reliably determining whether or θij is the phase angle difference between bus-i and bus-j,
not a given outage scenario will lead to voltage instability PG (i) PD (i), QG (i) and QD (i) are the active/reactive power
in the power grid. If the identified infeasible cases can be generation/demand, at bus-i, respectively.

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
 b) Line Flow Limits: Line flow equations for both active requiring voltage stability constraints. In this work, a similar
and reactive power flows can be written as: approach will be taken and a voltage stability index will be
used for this purpose. While this index is well documented in
2
the literature, in order to make the paper self contained, its
Pij = gij (| Vi | − | Vi || Vj |cos(θij )) derivation will be briefly reviewed below.
− bij | Vi || Vj |sin(θij ) (4) The problem of detecting and quantifying the proximity
to voltage instability for all load buses in a given system
2 has been investigated, and an effective and practical method
Qij = −bij (| Vi | − | Vi || Vj |cos(θij ))
was presented earlier in [10]. The method is based on the
− gij | Vi || Vj |sin(θij ) (5) calculation of a stability index which can then be used to
where: check voltage stability of a given operating point. The index is
Pij , Qij : active/reactive power flow on branch (i, j), computed in three steps using only the bus admittance matrix
gij , bij : conductance/susceptance of branch (i, j). and the bus voltages. System buses are initially separated into
The apparent power flow for branch (i, j), | Sij | which is two sets representing the generator and load buses. The load
given by: bus voltages and generator bus currents are then expressed
q in terms of load bus currents and generator bus voltages.
| Sij | = Pij2 + Q2ij (6) Derivation of these equations require partial inversion of the
bus admittance matrix and results in the following partitioned
will be limited by the power transfer capacity, hence: matrix equation:
      
| Sij | ≤ Sij (7) VL   IL ZLL FLG IL
= H = (11)
IG VG KGL YGG VG
where, Sij represents the upper limit for power transfer on
branch (i, j). where, ZLL , FLG , KGL and YGG are the submatrices of H
c) Generator Limits: Each generator is assigned upper matrix, VL , IL , VG and IG are load bus voltages, load bus
and lower power output limits. The upper limit is related to currents, generator bus voltages and generator bus currents,
the unit’s power generation capacity. The lower limit is used to respectively.
enforce mandatory power generation while operating a boiler Using the top part of the above equation, an equivalent
(if exists). The associated constraints for power generation voltage V0j for each load bus j is calculated using the elements
limits are given below: of the submatrix FLG and the generator bus voltages VG [10]:
X
P ≤ PG ≤ P (8) V0j = − Fji Vi (12)
i∈G

Q ≤ QG ≤ Q (9) Finally, a voltage instability proximity index Lj for all load

buses is calculated using V0j and load bus voltages Vj :
where, P , P , Q and Q are the vectors that represent lower  
and upper limits for the active and reactive power generation.  V0j 
Lj = 1 +
 (13)
d) Voltage Limits: Upper and lower voltage limits are Vj 
introduced for all buses in order to avoid damage to voltage
where:
sensitive loads and equipment.
G and L refer to generator and load buses,
i and j are generator and load bus indices,
V ≤|V |≤V (10)
Vi and Vj are complex voltage phasors.
where, V and V are the vectors that represent lower and upper Large number of simulated cases confirmed the reliability
limits for voltage magnitudes. of the index in predicting the voltage stability of a given
e) Voltage Stability Limit: Depending on the considered system and associated operating point. It is proven in [10]
scenario, line and generator outages may cause voltage sta- that the voltage stability will be guaranteed if Lj < 1. It turns
bility problems which may then trigger cascading failures out that this condition is sufficient, but not necessary, i.e.
and partial/complete blackouts in the system. The commonly systems may still remain stable even when the calculated index
encountered reason for voltage instability is lack of reac- violates above inequality. Hence, it is a conservative index
tive power support. While line outages may limit transfer but a reliable one if the main concern is avoiding voltage
of reactive power, generator outages may lead to reactive instability. It is therefore adopted in this work as a proxy for a
power shortage due to the missing reactive power source. The constraint that will maintain voltage stability when considering
problem of voltage stability is quite challenging and has been an outage scenario. The constraint can then be simply included
the topic of numerous studies in the past. Several of these as:
studies focus on the development of relevant indices which can
then be used as proxies in formulating optimization problems Lj ≤ Lobjective (14)

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
Choosing Lobjective as equal to 1.0 will yield voltage stable and/or generator outages. A flowchart of the proposed ap-
solutions as proven in [10]. On the other hand, choice of proach is given in Fig. 1. Note that the index i refers to the
values less than 1.0 for Lobjective will yield more conservative candidate buses to place DGs and the index k refers to the
solutions with respect to proximity to voltage instability. considered outage scenarios.
For certain considered outage scenarios, VSCOPF may fail
to find a feasible solution no matter where DG is placed. Such
Given the above description of various constraints needed to cases will be indicated by a null row in the binary assignment
be incorporated into the optimization problem, the VSCOPF matrix A. Such cases will then be studied by considering
problem can now be written in compact form as follows: placement of pairs of DGs instead of a single DG. Thus, all
possible combinations of DG pairs that can be placed for a
T given set of nB buses will have to be calculated as:
min CG PG (15)
PG ,QG ,|V |,θ

subject to: 
nB

nB !
1. Power Balance Equations: (2), (3) = (16)
2 (nB − 2)!2!
2. Line Flow Limits: (7)
3. Generator Limits: (8), (9) and placement of these possible pairs will have to be tested
4. Voltage Limits: (10) one pair at a time for feasibility. Note that, if none of the
5. Voltage Stability Limit: (14) pair combinations yields a feasible solution, then simultaneous
placement of larger numbers of DGs will have to be investi-
Active power generation (PG ), reactive power generation gated. In this work, only single DG and pairs of DGs are
(QG ), bus voltage magnitudes ( | V | ) and phase angles (θ) considered. Fig. 2 shows the way candidate DG pairs are
at each bus will be the control variables in the above problem. selected and placed when certain rows of initial A matrix
are null indicating the infeasibility of solving the optimization
B. Forming Matrices for DG Placement Method problem by placing only single DGs. Details of extending the
The impact of each considered outage scenario on the power rows of A matrix to account for simultaneous placement of
grid will be different and may require DGs to be placed at DG pairs will be described in section III-C below.
different buses. In order to determine all possible options of
placing a single DG for a given outage scenario, the VSCOPF
problem will be solved as many times as the number of Start
candidate locations for DGs (typically only PQ buses or a
subset of them) in the system. These solutions will then be i = 1, k =1, NonzeroRow = 0
repeated in a similar fashion for each outage scenario. At
the end of this process, one can build an assignment matrix Run VSCOPF

where for each outage scenario a set of candidate locations

for placing a single DG will be marked. A similar matrix L NonzeroRow=0
can be created where for each of these DG placements, the Yes Feasible? i=i+1 i=1
k=k+1
maximum element of the corresponding stability index vector
Lj for bus j will be recorded. The elements of L matrix will No
be used to gauge the effectiveness of a given DG placement A(k,i)=1
A(k,i)=0
at bus j for the outage scenario i, i.e. the smaller L(i, j), the NonzeroRow = 1
L(k,i)=0
more effective the placement. L(k,i)=max(Lj)
No
The DG assignment table is stored as a binary incidence No
matrix A where the rows and columns correspond to outage
i=nB?
scenarios and placed DGs respectively. Note that this is very
similar to the incidence matrix formed for optimal meter
placement in [11]. If the VSCOPF problem cannot be solved, Yes
i.e. it cannot find a feasible solution by placing a single DG
at any of the possible locations, then the corresponding row NonzeroRow=0? No k=M?
of A will be null indicating that for that outage scenario
voltage instability cannot be avoided by placing a single DG
anywhere in the system. In this work, such cases are marked Yes Yes

and subsequently processed by considering placement of pairs

Save k End
of DGs instead of a single DG.
Consider a power system having nB buses and M outage
scenarios, where each outage case may involve multiple line Fig. 1. Formation of A and L matrices for only single DG

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
 Start 
 1 if a DG or DG pair is assigned
xi = to bus i or bus pair i
i = nB+1, NonzeroRow = 0
0 otherwise


Run VSCOPF Note that in the case of DG pair placements, paired buses will
be associated with a separate fictitious bus whose number will
be larger than nB and the corresponding entry in x will be
assigned a 1.
Yes Feasible? i=i+1
b is a binary vector with the first nB elements equal to 1,
and the remaining ones are 0.
No A is a matrix which contains the candidacy information for
A(k,i)=1 all single and paired DGs. Details of forming this coefficient
A(k,i)=0
NonzeroRow = 1
L(k,i)=0
matrix will be explained next.
L(k,i)=max(Lj)

No Forming the [A] Matrix:

If VSCOPF finds at least one feasible solution for each
outage scenario by placing a single DG, then the algorithm
i=nB+C(nB,2)?
in Fig. 1 will be sufficient for building A matrix. Else, some
rows of A may remain null requiring simultaneous placement
Yes of more than one DG to maintain voltage stability. In this
work it was sufficient to consider only up to two (or a
pair of) DGs to find feasible solutions. The flowchart in
NonzeroRow=0? Yes Save k Fig. 2 shows how to form A using pairs of DGs for those
infeasible outage scenarios. Moreover, the constraints will be
augmented by additional equations and variables increasing the
No
row and column sizes of A as illustrated below using a tutorial
End example. Augmented matrix A can be written in partitioned
form as:
Fig. 2. Formation of A and L matrices for pairs of DGs
 
A11 A12
A= (19)
A21 A22
C. DG Placement Method Note that A11 refers to the initial A matrix formed consid-
ering single DG placements only. The augmented rows and
Let us assume that the repeated VSCOPF solutions are columns will be provided by the sub-matrices A12 , A21 and
carried out and assignment matrix A and stability index matrix A22 whose entries will be defined below:
L are formed for a considered set of outage scenarios and i) A11 : Apply the algorithm in Fig. 1 to form the initial A
candidate load buses for placing DGs as described above. matrix, let it be A11 as shown below.
Then, the following integer programming (IP) problem can
be formulated [11] to determine the minimum set of DGs that
Bus-1 Bus-2 Bus-3 ... Bus-i ... Bus-nB
are required to be placed in order to ensure feasible operating
0 1 0 1 0
 
Outage-1
conditions under all considered outage scenarios:
 1 0 0 0 0  Outage-2
0 0 0 0 0
 
  Outage-3
 .. .. .. .. .. ..  ..
min cT x . . . . . .
 
A11 =  .
(17)  .. .. .. .. 
s.t. Ax ≥ b
 
 . . . A(k, i) .  Outage-k
.. .. .. .. .. ..
 
..
.
 
. . . . . .
where: 1 0 0 0 1 Outage-M
c is the cost vector which is formed based on the entries of
the matrix L considering all M outage scenarios as follows:
Note that the third (3rd ) row is null, indicating that no fea-
sible solution can be found for outage scenario-3 considering
v
uM
single DG placements. Therefore, for this scenario, in addition
uX
ci = k(Li )k = t {L(k, i)}2 (18)
k=1
to single DGs, DG pair placements should also be investigated
in order to maintain voltage stability. This requires finding all
x is a binary vector whose dimension depends on matrix A possible DG pair combinations, sequentially assigning them
and defined as: fictitious bus numbers starting from nB + 1 and solving the

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
VSCOPF problem by placing these pairs one at a time. This (both active and reactive) are increased by 100% to create
will extend the columns of A matrix by the number of newly highly stressed operating conditions and pushing the system
assigned fictitious bus numbers and creates a submatrix A12 limits. Moreover, Lobjective is intentionally reduced in order
representing the newly added columns. to create cases requiring paired DG placements for some of
ii) A12 : Assume that applying the flowchart in Fig. 2 yields: the simulated scenarios.


Simulation results are shown in Table II. Out of the simu-
nB
nB + 1 nB + 2 nB + 3 ... nB + i ... nB + 2 lated 40 outage scenarios, 32 of them required only a single
 0 0 0 0 0  Outage-1
DG placement at bus 44 in order to maintain voltage stability.
0 0 0 0 0  Outage-2
The remaining 8 outage scenarios required three more DGs

0 1 0 1 0
 
A12 =  Outage-3
 .. .. .. .. ..  to be placed in order to maintain a voltage stable system.
. . . . .
 
0 0 0 0 0 Outage-M
To obtain those results, initially, the candidate locations are
found by using the single DG placement algorithm in Fig. 1.
In this example, several feasible solutions are assumed to However, voltage instability cannot be avoided for 8 out of
be found by placing DG pairs as indicated by 1’s in row 3. 40 scenarios where corresponding rows of A are found to be
iii) A21 and A22 : In assigning DG pairs, in order to null. Then, candidate locations are determined to place paired
establish consistency between the created fictitious buses (e.g. DGs for those remaining 8 scenarios. Finally, DGs are placed
representing the pair of buses (nB +2) and (nB +i)) and actual at buses 21, 44, 52, and 95 by using the integer programming
buses, additional equations are required to indicate which bus (IP) problem formulation and solution.
pair belongs to which actual buses. Thus, a truth table can In order to illustrate the overall procedure, one of the outage
be formed as shown in Table I for the example where the scenarios is chosen as an example. In this scenario three of
fictitious buses (nB + 2) and (nB + i) refer to bus pairs 2-3 the transmission lines are taken out of service, which makes
and 2-i respectively. the system voltage unstable. Applying the DG placement
procedure, it can be shown that by placing a DG pair at either
TABLE I one of the twelve possible combinations formed among the
T RUTH TABLE sets of buses {43, 44, 45}, and {51, 52, 53, 58}, the system
Solution 1 Solution 2 can be made voltage stable. Among these, since the pair 44-
Input Outputs Input Outputs 52 also helps other considered outage scenarios, it is chosen
nB + 2 Bus-2 Bus-3 nB + i Bus-2 Bus-i
1 1 1 1 1 1 as the solution. Fig. 3 shows the voltage magnitude profile of
0 unknown unknown 0 unknown unknown the system buses before and after the placement of DGs at
buses 44 and 52. Note that voltage magnitude limits used for
These tables can now be converted into the following this study are between 0.9 and 1.06 per unit in all simulations.
equations: VSCOPF solution is repeated tens of thousands of times for
the IEEE 118-bus system. While it is manageable for such a
x2 + x3 ≥ 2xnB +2 test system, the computational burden can rapidly increase for
x2 + xi ≥ 2xnB +i much larger size power grids. In anticipation of applying this
method to very large scale problems, utilization of the high-
and accordingly A21 and A22 can be formed as follows: performance computing facility of the institution is considered
and actually implemented for the 118-bus system. Note the
Bus-1 Bus-2 Bus-3 ... Bus-i ... Bus-nB  total number of required repeat solutions for the IEEE 118-bus
A21 = 0 1 1 0 0 Additional system given in Table III. Implementing the outage scenarios
0 1 0 1 0 in the Discovery Cluster by accessing 50 of its cpu cores,
yields each scenario solution on average in about 5 seconds.
nB

It should also be noted that it is possible to implement and
nB + 1 nB + 2 nB + 3 ... nB + i ... nB + 2  execute the proposed algorithm also for very large scale power
A22 = 0 −2 0 0 0 Additional
0 0 0 −2 0

IV. S IMULATION R ESULTS TABLE II

DG P LACEMENT R ESULTS
The proposed methodology is simulated in Matlab [12]
using the IEEE 118-bus system. The computations are carried No. of
DG(s) Placed
Algorithm Stabilized
out using the Northeastern University Discovery cluster which at Bus
Scenarios
is housed at the Massachusetts Green High Performance
Computing Center (MGHPCC), which provides access to over Single DG Placement
32 44
Algorithm
20,000 CPU cores and over 200 GPUs.
In this study, 40 extreme event outage scenarios are created. Multi DG Placement
40 21, 44, 52, 95
Algorithm
Each scenario involves outages of three random lines. Loads

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
 Fig. 3. Voltage Profiles for the Outage Scenario

TABLE III R EFERENCES

T OTAL NUMBER OF REPEATED SOLUTIONS
[1] World Bank. (2019) Electric power consumption. [Online]. Available:
Total Number https://data.worldbank.org/indicator/EG.USE.ELEC.KH.PC
Number of Number of [2] Department of Energy. (2013) Economic benefits of in-
of Candidate
Scenarios Solutions creasing electric grid resilience to weather outages. [On-
Combinations
line]. Available: https://www.energy.gov/sites/prod/files/2013/08/f2/
40 64 2560 Grid%20Resiliency%20Report FINAL.pdf
8 2016 16128 [3] NASA. (2013) More extreme weather events forecast. [Online]. Avail-
18688 able: https://www.nasa.gov/centers/langley/science/climate assessment
2012.html
[4] M. Vatani, D. S. Alkaran, M. J. Sanjari, and G. B. Gharehpetian, “Multi-
ple distributed generation units allocation in distribution network for loss
grids efficiently by using a lot more of the 20,000 cpu cores reduction based on a combination of analytical and genetic algorithm
of the cluster. methods,” IET Generation, Transmission & Distribution, vol. 10, no. 1,
pp. 66–72, 2016.
V. C ONCLUSION [5] K. Mahmoud, N. Yorino, and A. Ahmed, “Optimal distributed gener-
ation allocation in distribution systems for loss minimization,” IEEE
This paper is concerned about resiliency of power grids Transactions on Power Systems, vol. 31, no. 2, pp. 960–969, 2015.
during extreme events which may cause multiple outages in [6] M. Ettehadi, H. Ghasemi, and S. Vaez-Zadeh, “Voltage stability-based
the system. The primary focus of the study is on maintaining dg placement in distribution networks,” IEEE Transactions on Power
Delivery, vol. 28, no. 1, pp. 171–178, 2012.
voltage stability during such events and accomplishing this [7] H. Hedayati, S. A. Nabaviniaki, and A. Akbarimajd, “A method for
by installing a small number of DGs in the system. This placement of dg units in distribution networks,” IEEE Transactions on
is proposed to be done in two stages, where in the first Power Delivery, vol. 23, no. 3, pp. 1620–1628, 2008.
[8] B. Zhang, P. Dehghanian, and M. Kezunovic, “Optimal allocation
stage possible candidate locations for each considered outage of pv generation and battery storage for enhanced resilience,” IEEE
scenario are identified via repeated optimization solutions Transactions on Smart Grid, vol. 10, no. 1, pp. 535–545, Jan 2019.
and this candidate list is pruned via an integer programming [9] A. Öner and A. Abur, “Strategic placement of distributed generators
against extreme events,” in 2019 North American Power Symposium
formulation to reach a final set of recommended DG locations. (NAPS). IEEE, 2019, pp. 1–6.
While being an off-line procedure, its heavy cpu requirements [10] P. Kessel and H. Glavitsch, “Estimating the voltage stability of a power
are shown to be readily addressed by using a multi-core high system,” IEEE Transactions on Power Delivery, vol. 1, no. 3, pp. 346–
354, 1986.
performance computer cluster and one illustrative implementa- [11] A. Abur and F. H. Magnago, “Optimal meter placement for maintaining
tion of it is presented using the 118-bus system as an example. observability during single branch outages,” IEEE Transactions on
Power Systems, vol. 14, no. 4, pp. 1273–1278, Nov 1999.
[12] The MathWorks, Inc., Optimization Toolbox, Natick, Massachusetts,
United State, 2020. [Online]. Available: https://www.mathworks.com/
help/optim/index.html

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020