You are on page 1of 8

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

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2016.2591503, IEEE
Transactions on Power Systems
1

Computationally Efficient Adjustment of


FACTS Set Points in DC Optimal Power Flow
with Shift Factor Structure
M. Sahraei-Ardakani, Member, IEEE, and K. W. Hedman, Member, IEEE

𝑘 Index for transmission elements.


Abstract—Enhanced utilization of the existing transmission 𝒑 Upper limit vector of power generation.
network is a cheaper and paramount alternative to building new 𝒑 Lower limit vector of power generation.
transmission lines. Flexible AC transmission system (FACTS)
devices are advanced technologies that offer transfer capability
𝒇 Vector of transmission elements’ thermal limit.
improvements via power flow control. Although many FACTS 𝒇𝑬 Vector of transmission elements’ emergency thermal
devices exist in power systems, their set points are not frequently limit.
changed for power flow control purposes, which is mainly due to 𝒄 Vector of generators’ marginal costs.
the computational complexity of incorporating FACTS flexibility 𝒅 Vector of power consumptions.
within the market problem. This paper proposes a computation- 𝝌 Upper limit vector of power injections representing
ally efficient method for adjustment of variable impedance based FACTS devices.
FACTS set points, which is also compatible with existing market
solvers. Thus, the method can be employed by the existing solvers
𝝌 Lower limit vector of power injections representing
with minimal modification efforts. This paper models FACTS FACTS devices.
reactance control as injections to keep the initial shift factors 𝜹 Upper limit vector of FACTS relative susceptance
unchanged. Next, the paper formulates a DC optimal power flow adjustment.
that co-optimizes FACTS set points alongside generation dis- 𝜹 Lower limit vector of FACTS relative susceptance
patch. The resulting problem, which is in a nonlinear program, is
then reformulated to a mixed-integer linear program. Finally, an
adjustment.
engineering insight is leveraged to further reduce the computa- 𝛿! Upper limit for FACTS relative susceptance adjust-
tional complexity to a linear program. Simulation studies on ment on transmission element k.
IEEE 118-bus and Polish 2383-bus test cases show that the meth- 𝛿! Lower limit for FACTS relative susceptance adjust-
od is extremely effective in finding quality solutions and being ment on transmission element k.
very fast.
𝜞 Generation locator matrix.
Index Terms—FACTS devices, linear programming, optimal
𝑨 Adjacency matrix.
power flow, power system economics, power system operation, 𝑨 Reduced Adjacency matrix.
power transfer distribution factors. 𝑩!" Branch B matrix.
𝑩 Nodal B matrix.
I. NOMENCLATURE 𝑏! Base susceptance of transmission element k.
𝜱 PTDF matrix.
Matrices are represented by upper case bold, vectors by
𝜉!! LODF, representing the change on line l’s flow with
lower case bold, and scalars by lower case characters. Super-
respect to the outage of line k.
script D is used for the diagonal operator, while superscript T
𝟏 Vector of ones.
represents transpose of a matrix. is used to describe reduced 𝟎 Vector of zeros.
matrices and vectors where the elements related to the refer- 𝑰 Identity matrix.
ence bus are removed. Finally, is used to show the flow on 𝑀 A very large positive number.
the lines equipped with FACTS without the FACTS impact.
B. Variables
A. Parameters 𝒑 Vector of power generations.
𝐹 Set of transmission elements equipped with FACTS 𝒇 Vector of power flows on transmission elements.
devices. 𝝍 Vector of power injections.
𝑁! Number of transmission elements. 𝝍 Reduced vector of power injections.
𝝌 Vector of power injections representing FACTS de-
Mostafa Sahraei-Ardakani is with the Department of Electrical and Com- vices.
puter Engineering at the University of Utah, Salt Lake City, UT 84112 USA
(e-mail: mostafa.ardakani@utah.edu). 𝝌 Reduced vector of power injections representing
Kory W. Hedman is with the School of Electrical, Computer, and Energy FACTS devices.
Engineering, Arizona State University, Tempe, AZ 85287 USA (e-mail: 𝜹 Vector of FACTS relative susceptance adjustments.
Kory.Hedman@asu.edu).

0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2016.2591503, IEEE
Transactions on Power Systems
2

𝒛 Vector of binary variables identifying the direction of the initial LP formulation of the problem. A recent work by
power flows. the authors has investigated reformulation of the resulting
𝑓! Power flow on transmission element k. NLP to a mixed-integer linear program (MILP) [30]. Howev-
𝜃!,!" Voltage angle at the “to” bus of transmission element er, the formulation presented in [30] is developed using the 𝐵𝜃
k. DCOPF structure, which is not compatible with the industry
𝜃!,!"#$ Voltage angle at the “from” bus of transmission ele- practices. With a 𝐵𝜃 formulation, all the bus voltage angles
ment k. need to be calculated in order to compute the line flows. The
Δ𝑏! Changes in the susceptance of transmission element k formulation, thus, does not scale well with the size of the
by FACTS. power system [31]. Note that real power systems have tens of
thousands of buses and transmission elements. Therefore, a
reduced formulation with power transfer distribution factors
II. INTRODUCTION (PTDF) is often used in SCUC and SCED solvers. With the

I N many parts of the United States, the transmission network


is under stress and needs to be upgraded [1]-[2]. High con-
gestion costs, e.g. $670 million in 2013 for PJM [3], is one
use of PTDFs, the need for calculation of bus voltage angles
vanishes. Moreover, only transmission elements, suspected to
become overloaded, are monitored for thermal limit violations.
indicator of the scarcity of transfer capability over the existing DCOPF formulation with PTDF structure is explained in more
transmission network. A trivial but unattractive solution to this details in the next section.
problem would be to build new transmission lines. New Industrial implementations of SCED and SCUC employ a
transmission projects are extremely costly, take many years to variety of methods to improve the speed and accuracy of their
complete, and are not preferred by the people living near the linearized PTDF-based optimal power flow model [32]-[34].
new lines. NIMBY (Not In My Back Yard) and BANANA SCUC and SCED solvers become significantly faster as the
(Build Absolutely Nothing Anywhere Near Anything) are two number of transmission elements, whose flows are monitored
commonly used terms to refer to this problem. A cheaper, for thermal limit violations, decreases. For instance, California
faster, and environmentally more appealing alternative to Independent System Operator employs an adjustable threshold
building new transmission lines is more efficient utilization of for flows to select the lines that are needed to be monitored.
the existing system. This objective can be achieved with pow- Any transmission element that is loaded above this threshold
er flow control technologies, including transmission switching is included in SCED as a constraint [34]. Since the operator
(TS) [4]-[6] and flexible AC transmission system (FACTS) has information on the current flows, this threshold is often
devices [7]-[10]. With power flow control, the transfer capa- picked at levels very close to the thermal limit of the element
bility over the existing network can be significantly improved for the next round of SCED. Therefore, the subset of pre-
[11]. Transfer capability improvement can reduce operational contingency and post-contingency line flows that are moni-
cost [12]-[15], improve system reliability [16]-[21], and en- tored is significantly smaller than the full set (e.g., hundreds to
hance integration of renewable energy resources [22]-[23], thousands in comparison to millions).
thereby delaying the need for new transmission lines. PTDFs are calculated offline based on the topology and
Specific types of FACTS devices can provide significant susceptance information. FACTS devices change the suscep-
power flow control [24]-[25] through adjustment of transmis- tance of the transmission elements and make the original
sion elements’ reactances. This paper loosely employs the PTDFs invalid. Therefore, the PTDF matrix needs to be recal-
term “FACTS” to refer to those specific types of FACTS de- culated every time the transmission network is changed either
vices, which are known as variable impedance based FACTS: through transmission switching or adjustment of FACTS set
such as thyristor protected series compensator (TPSC) / thyris- point. Moreover, inclusion of FACTS adjustments makes
tor controlled series compensator (TCSC) and unified power DCOPF, an NLP. This paper contributes to the literature by
flow controller (UPFC). addressing these two problems. First, the impact of FACTS
Despite the ability of the existing FACTS devices to pro- devices is isolated form the initial PTDF matrix. Second, the
vide power flow control, day-ahead security-constrained unit resulting NLP is reformulated to an MILP and then to an LP.
commitment (SCUC) or real-time security-constrained eco- In order to avoid variable PTDFs, the impact of FACTS ad-
nomic dispatch (SCED) rarely capture their full flexibility. justments are modeled as additional injections at the two ends
This is mainly due to the lack of economic incentive [26]-[29] of the elements equipped with FACTS [22], [35]-[36]. Using
and computational complexity of modeling the flexibility of this technique, the PTDF matrix remains constant with any
FACTS devices in SCUC and SCED [20], [30]. possible adjustment of FACTS devices.
Solving the optimal power flow problem in its original non- The paper further shows that even with the constant PTDF
linear form is beyond today’s computational capabilities. The matrix, the resulting problem remains an NLP. Building upon
industry, thus, employs a linearized form of optimal power authors’ prior work [30], this NLP is reformulated to an
flow problem within the SCUC and SCED solvers. It is re- MILP. An engineering insight is then introduced to further
ferred to as DC optimal power flow (DCOPF), which in its reduce the computational complexity of the problem to an LP.
original form is a linear program (LP). Inclusion of FACTS The method developed in this paper employs a similar
flexibility would make the DCOPF a non-linear program structure to what is used in industry implementations of SCUC
(NLP), which is much more computationally burdensome than and SCED: PTDF-based power flow. Thus, the existing mar-

0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2016.2591503, IEEE
Transactions on Power Systems
3

ket solvers can be modified with minimal effort to incorporate 𝒇 = 𝜱𝝍 (6)


this method and exploit the flexibility of FACTS devices. 𝝍 = 𝜞𝒑 − 𝒅 (7)
Through the co-optimization of FACTS set points alongside
PTDFs are calculated offline through network topology in-
generation dispatch, cost savings [30] as well as reliability
formation as well as susceptance data:
gains can be achieved [20]. This framework will also enable
the system operators to effectively reduce spillage of intermit- 𝜱 = 𝑩𝑩𝒓 𝑨𝑩!𝟏 (8)
tent generation from renewable energy resources [22]. 𝑩 = 𝑨𝑩𝑩𝒓 𝑨! (9)
To show the effectiveness of the method, simulation studies
on IEEE 118-bus system, as well as the Polish system are Series variable impedance FACTS devices offer power
conducted. The results are compared with those obtained by flow control through adjustment of the transmission element’s
impedance. Such adjustment translates into a range for the
the 𝐵𝜃 formulation [30]. The results suggest that the algorithm
transmission element’s susceptance in the DCOPF formula-
developed in this paper is extremely effective in reducing the
tion. Therefore, the branch susceptance elements in (3) would
computational complexity of the problem and achieving quali-
no longer be constants for the transmission assets equipped
ty solutions. The algorithm is significantly faster than [30]
with FACTS devices. The susceptance for those assets can be
because it employs a PTDF structure and has better scaling
anywhere in the acceptable range of the FACTS device:
behavior with respect to the size of the power system. Alt-
hough optimality is not guaranteed, the results show that the 𝑏!!"# ≤ 𝑏! ≤ 𝑏!!"# ∀𝑘 ∈ 𝐹 (10)
method almost always find the optimal solution. Even when
the optimal solution is not achieved, the solution obtained by The changes to the susceptance would propagate through
the LP method is very close to the optimal solution found by PTDF calculations in (7)-(8). Since the PTDF calculations are
the MILP algorithm. performed offline, it is not suitable to model the flexibility of
The rest of this paper is organized as follows: Section 3 pre- FACTS devices through their direct impact on the PTDF ma-
sents the methodology and derives the formulation. Simulation trix. This problem has been discussed in the literature, for dif-
studies on IEEE 118-bus system and the large-scale Polish ferent problems, with the straightforward solution of modeling
system are presented in Section 4. Conclusions are drawn in FACTS devices as injections [33]-[34]. The power flow on a
Section 5. transmission element equipped with FACTS can be calculated
using DC power flow equation, with the impact of the FACTS
III. FORMULATION adjustment separated from the line’s initial susceptance:
A DCOPF in 𝐵𝜃 form is shown in (1)-(5). Note that all the 𝑓! = 𝑏! 𝜃!,!" −   𝜃!,!"#$ + Δ𝑏! 𝜃!,!" −   𝜃!,!"#$
comparison operators for the vectors, throughout this paper,
!!!
are meant to operate on an element to element basis. In this = 𝑓! + 𝑓! (11)
!!
formulation, the line flows are calculated individually through Thus, impact of FACTS adjustment can be modeled with
their susceptance and voltage angle different. The formulation injections at the two ends of the transmission element. This
also creates inter-dependence between the flows of the trans- procedure is shown in Fig. 1. Since this modeling technique
mission elements through bus voltage angles. Therefore, all keeps the line’s initial susceptances unchanged, the PTDF
the bus voltage angles need to be calculated before the flows matrix will remain unchanged as well. Each FACTS device
can be compared with the transmission elements’ thermal limit will be modeled by two additional injections at the two ends of
(4). Such calculations become increasingly challenging with the transmission element as defined in (12).
the size of meshed power systems.
from to
min 𝒄! 𝒑 (1) 𝑏𝑘 + 𝛥𝑏𝑘
𝒑≤𝒑≤𝒑 (2)
𝒇 = 𝑩𝑩𝒓 𝑨! 𝜽 (3)
−𝒇 ≤ 𝒇 ≤ 𝒇 (4)
𝜞𝒑 − 𝒅 + 𝑨𝒇 = 𝟎 (5) from to
Given the size of real power systems, energy management 𝑏𝑘
systems (EMS) and market management system (MMS) soft- 𝛥𝑏! 𝛥𝑏!
ware prevents such burdensome calculations by employing an 𝑓! ! !   𝑓! ! !  
𝑏! 𝑏!
alternative formulation based on shift factors. Using shift fac-
tors known as PTDFs, the need for calculation of bus voltage Fig. 1. Conversion of FACTS adjustments to additional injections, while
angles disappears and transmission flow calculations can be keeping the line’s susceptance unchanged.
limited to only a subset of transmission elements. System op- In order to include the FACTS injections in the PTDF-
erators are only concerned about a rather limited subset of based DCOPF formulation, they can be modeled as a vector:
transmission assets for thermal limit violations. Therefore, the
market solver can only include those transmission elements to
improve its computational capabilities. A PTDF-based OPF
can be obtained by replacing (3) with:

0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2016.2591503, IEEE
Transactions on Power Systems
4

!!! min 𝒄! 𝒑 (26)


!!
𝑝≤𝑝≤𝑝 (27)
𝜹= ⋮ (12)
!!!! 𝒇 = 𝜱(𝝍 + 𝑨𝝌) (28)
! !! 𝝍 = 𝜞𝒑 − 𝒅 (29)
!
𝝌 = 𝜹 𝒇     (13) 𝜹! 𝒇 − 𝑀𝒛 ≤ 𝝌 ≤ 𝜹! 𝒇 + 𝑀𝒛 (30)
𝜹≤𝜹≤𝜹 (14) 𝜹! 𝒇 − 𝑀 𝟏 − 𝒛 ≤ 𝝌 ≤ 𝜹! 𝒇 + 𝑀 𝟏 − 𝒛 (31)
𝒇 ≥ −𝑀𝒛 (32)
where superscript D is the diagonal operator, creating a diago- 𝑀 𝟏−𝒛 ≥𝒇 (33)
nal matrix from the elements of a vector. Note that the ele- −𝒇 ≤ 𝒇 + 𝝌 ≤ 𝒇 (34)
ments of 𝜹 vector, which correspond to transmission elements 𝜞𝒑 − 𝒅 + 𝑨(𝒇 + 𝝌) = 𝟎 (35)
with no FACTS device, will be zero since the numerator is 𝒛 ∈ 0,1 !! (36)
zero for those terms. Using (13), (6) can be rewritten as:
MILPs, though less challenging than MINLPs, are still
𝒇 = 𝜱 𝝍 + 𝑨𝝌 = 𝜱 𝝍 + 𝑨𝜹! 𝒇     , (15) computationally expensive problems and are, thus, not pre-
which can, then, be rearranged: ferred. As mentioned before, the binary variables in (26)-(36)
only identify the direction of the power flow on the lines
𝑰 − 𝜱𝑨𝜹! 𝒇 = 𝜱𝝍. (16) equipped with FACTS devices. Assuming that those directions
are known, the binary variables in (26)-(27) can be fixed and
It is clear from (16) that the resulting DCOPF is a nonlinear
program (NLP), since it includes production of FACTS ad- the problem will become an LP. FACTS devices are often
installed on major lines, where the power flow direction is
justment vector, 𝜹, and power flow, 𝒇. NLPs are computation-
relatively predictable. For instance, it is trivial to predict the
ally expensive and are, thus, avoided in SCED and SCUC
flow direction on key corridors like the California-Oregon
formulations. The nonlinearity that FACTS devices create in
intertie (COI). Even for the cases when power flow directions
the DCOPF problem is a major barrier for enhanced operation
are not known, the operator can use information from previous
of these devices. To avoid the product term in (16), FACTS
rounds of SCUC or SCED to estimate, with a high accuracy,
injection can be represented as a range with inequality con-
this direction. The operator can also solve an initial DCOPF
straints instead of the equality constraint presented in (13). To
without considering FACTS devices and assign the same
do so, (14) can be reformulated to:
power flow direction to (26)-(36). It is not likely that the pow-
𝝌≤𝝌≤𝝌 (17) er flow directions change after FACTS adjustments on the
lines equipped with FACTS. On the contrary, it is expected
The upper and lower limits of FACTS injection can be iden- that the flows increase, in the same direction, on the paths
tified based on the direction of the flow and the limits on parallel to the congested lines. Using this engineering insight,
FACTS adjustment: the MILP presented in (26)-(36) can be solved as a LP. Note
for  𝑓! ≥ 0:                  𝛿! 𝑓! ≤ 𝜒! ≤ 𝛿! 𝑓! (18) that, although it is highly unlikely that the power flow direc-
tions on the lines equipped with FACTS change in the optimal
for  𝑓! < 0:                𝛿! 𝑓! ≤ 𝜒! ≤ 𝛿! 𝑓! (19)
solution, it is not impossible. Therefore, there is no guarantee
These conditional inequalities can be modeled with binary that the solution to the LP version of the problem will be op-
variables: timal. However, the significance of the proposed method does
not vanish in absence of optimality; the algorithm will still
1 − 𝑧! 𝛿! 𝑓! + 𝑧! 𝛿! 𝑓! ≤ 𝜒! ≤ 1 − 𝑧! 𝛿! 𝑓! + 𝑧! 𝛿! 𝑓! (20) very quickly find a solution with significantly lower cost. The
1 − 𝑧! 𝑓! ≥ 𝑧! 𝑓! (21) results in the next section show that the algorithm almost al-
ways finds the optimal solution, and does so significantly fast-
where 𝑧! takes a value of 0 for positive power flows and a er than the original MILP formulation.
value of 1 for negative power flows. Equations (20)-(21) in- As mentioned before, not all power flows are monitored for
clude products of the binary variable 𝑧! and the power flow thermal limit violation in PTDF-based DCOPF problems.
𝑓! , making the resulting DCOPF a mixed integer nonlinear Therefore, (34) is only considered for a subset of transmission
program (MINLP). However, using the big M reformulation assets. However, all the transmission elements equipped with
[37] technique, (20)-(21) can be reformulated as a set of linear FACTS should be monitored in (34) to ensure the FACTS
constraints: injections remain within the acceptable range.
𝛿! 𝑓! − 𝑧! 𝑀 ≤ 𝜒! ≤ 𝛿! 𝑓! + 𝑧! 𝑀 (22) A. Contingency Constraints
1 − 𝑧! −𝑀 + 𝛿! 𝑓! ≤ 𝜒! ≤ 1 − 𝑧! 𝑀 + 𝛿! 𝑓! (23) Modern SCUC and SCED solvers include explicit repre-
𝑓! ≥ −𝑧! 𝑀 (24) sentation of a limited subset of contingency constraints. The
1 − 𝑧! 𝑀 ≥ 𝑓! (25) impact of each transmission outage is represented through
linear sensitivities known as line outage distribution factors
Thus, the DCOPF can be reformulated as a mixed integer (LODF). LODFs can be calculated through PTDFs and, thus,
linear program (MILP) with no loss of precision: are readily available [31]. A contingency constraint represent-
ing line l’s flow after the outage of line k is shown in (37).

0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2016.2591503, IEEE
Transactions on Power Systems
5

−𝑓!! ≤ 𝑓! + 𝜉!! 𝑓! ≤ 𝑓!! (37) range, for 𝐵𝜃 and PTDF based methods presented here and in
[30]. The figure shows that the PTDF-based DCOPF is faster
Note that, both 𝑓! and 𝑓! are represented through linear ex- than the 𝐵𝜃-based model both in the MILP and LP forms.
pansions using the PTDF structure (28). Thus, inclusion of Note that, monitoring fewer lines can further reduce the com-
contingency constraints like (37) would not add to the com- putational time. This is tested with simulations on a large-
plexity of the problem, and the same logic explained before scale Polish 2383-bus system.
can be used to turn the problem into an LP. In the case that
either, or both, of the lines k and l are equipped with FACTS
devices, the linear technique presented in (26)-(36) still holds
and can be used in contingency constraint modeling similar to
the normal operations.

IV. SIMULATION STUDIES


To evaluate the effectiveness of the proposed method, sim-
ulation studies on IEEE 118-bus system as well as the Polish
2383-bus system are conducted. The results obtained here are
compared to those presented in [30].
A. IEEE 118-Bus System
The data is taken from [38], modified, and supplemented
with cost data from [39]. The full dataset matches [4], [13]-
Fig. 3. Average computational time (over reactance control range) for 𝐵𝜃 and
[14] and in particular [30] PTDF-based models with different number of lines equipped with FACTS
Fig. 2 shows the savings for IEEE 118 bus system, for the devices.
case that FACTS devices are installed on the more heavily
B. Polish 2383-Bus System
utilized lines. The total system cost is $2074/h when there is
no FACTS device in the system. The results presented in Fig. The computational gains from PTDF-based formulation of
2 show the savings for the case where up to 20 FACTS devic- the DCOPF problem become more significant and important
es with different range of reactance controls are used. These when the size of the system increases. To test the effectiveness
results are consistent with those obtained in [30]. of the proposed method on large-scale systems, a number of
simulation studies were conducted on a Polish system. The
data is taken from [40]-[41], which represents the Polish elec-
tric grid in a peak winter hour in year 2000. The system in-
cludes 2383 buses, 2896 transmission lines, and 327 genera-
tors. The difference between the cost obtained by DCOPF and
transportation model is $30,886, which is the upper limit for
savings that can be achieved with power flow control technol-
ogies such as FACTS devices.
Fig. 4 shows the savings achieved with up to 20 FACTS
devices with different reactance control ranges, when FACTS
devices are installed on the lines that carry larger flows. This
is similar to simulation studies presented in [30] for the Polish
system.
Fig. 2. Savings for IEEE 118-bus system with different number of FACTS
devices and different reactance control ranges. The FACTS devices are in-
stalled on lines that are more heavily utilized and the reactance control range
is shown on the figure for each set of the results.

As previously mentioned, the major advantage of PTDF-


based formulation of SCED and SCUC is the ability to only
monitor the flow of the lines that are of concern. For the re-
sults shown in Fig. 2, only line flows that are more than 70%
of their capacity in the base case (no FACTS) are monitored.
For all the simulations conducted in Fig. 2, the computational
time was stored. Analysis of the results showed that the com-
putational time heavily depends on the number of FACTS
devices. The correlation between FACTS reactance range and
the computational time was very weak. Fig. 3 shows the com- Fig. 4. Savings for the Polish system with different number of FACTS devices
putational time, averaged over FACTS reactance control and different reactance control range. The FACTS devices are installed on
lines that carry larger power and the reactance control range is shown on the

0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2016.2591503, IEEE
Transactions on Power Systems
6

figure for each set of the results. before, the PTDF-based methods outperform the 𝐵𝜃-based
The computational time for the results shown in Fig. 4 is DCOPF models. The results shown in Fig. 7 were obtained by
presented in Fig. 5. For the PTDF-based models, only lines only monitoring the lines with flows more than 70% of their
whose flow is larger than 70% of the line’s capacity in the thermal capacity in the base case. Since the maximum number
base case (with no FACTS) are monitored. The computational of lines with FACTS devices is 50 in Fig. 7, as opposed to 20
time is averaged over the reactance control range of the for Fig. 3 and 5, Fig. 7 better shows how the computational
FACTS devices, as the computational time heavily depends on time grows in MILP-based implementation of FACTS adjust-
the number of lines equipped with FACTS, and not so much ments. The linear methods, however, maintain the same level
on the reactance control range of the devices. The results show of complexity for different number of lines with FACTS. It
that the PTDF-based formulation is consistently faster than the should be noted that the difference between the computational
𝐵𝜃 model. The results also show that unlike the MILP models time needed for the 𝐵𝜃-based and the PTDF-based models
that get computationally more expensive with additional num- becomes significantly larger with the size of the power sys-
ber of FACTS devices, the LP-based formulations show a ro- tem. The difference will also become larger in computational-
bust behavior with respect to the computational time. ly more difficult problems such as SCUC, which is an MILP
with many binary variables.

Fig. 5. Average computational time (over reactance control range) for 𝐵𝜃 and
PTDF-based models with different number of lines equipped with FACTS Fig. 7. Average computational time (over reactance control range) for 𝐵𝜃 and
devices. The FACTS devices are installed on lines that carry larger power. PTDF-based models with different number of lines equipped with FACTS
devices. The FACTS devices are installed on lines with larger flows relative
Fig. 6 shows the savings when up to 50 FACTS devices are to their thermal capacity.
installed on the lines that have larger flows relative to their
As discussed before, with PTDF-based methods the num-
thermal capacity. The reactance control range of the devices is
ber of monitored lines can be reduced. This subset of critical
shown on the figure. Similar to the results shown previously,
lines can be chosen either with the operator’s knowledge or
significant savings can be achieved with a limited number of
the information present at the time of operation. For instance,
FACTS devices.
real-time SCED can use the state estimation information in-
cluding the current line flows. Thus, SCED can use such in-
formation to identify the lines that need to be monitored for
thermal flow violation. As an example, only the lines with
flows larger than 95% of their thermal capacity can be fed into
SCED [34]. The computational time decreases as this subset
of lines becomes more exclusive. To examine the impact of
having fewer lines monitored in DCOPF, different thresholds
on the line flow are studied. In particular, lines at 70% and
80% of capacity or above in the base case are picked as fixed
thresholds. Three levels of adaptive thresholds are also tested.
The difference between adaptive and fixed thresholds is that
the information fed into the adaptive threshold includes the
most recent run, i.e. the power flow in the previous simulation
with one less number of FACTS devices. Therefore, the power
Fig. 6. Savings for the Polish system with different number of FACTS devices flow information used for the adaptive thresholds is more ac-
and different reactance control ranges. The FACTS devices are installed on
lines with larger flows relative to their thermal capacity and the reactance
curate than the fixed thresholds. With the adaptive threshold,
control range is shown on the figure for each set of the results. less number of lines can be monitored without violating the
thermal limits. Note that, the lines equipped with FACTS de-
The computational time for the results shown in Fig. 6 is vices should always be monitored to ensure the FACTS injec-
presented in Fig. 7. Similar to the two sets of results presented tions remain within the feasible range.

0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2016.2591503, IEEE
Transactions on Power Systems
7

Fig. 8 and Fig. 9 show the computational time with fixed and compatible with the industry implementations. Therefore,
and adaptive thresholds for the results shown in Fig. 4 and Fig. it is straightforward to add the proposed FACTS set point ad-
6 respectively. Both figures show that employment of a meth- justment routine to the existing SCED and SCUC solvers. The
od to reduce the number of lines, whose flows are monitored results presented in this paper show that:
in DCOPF, significantly reduces the computational complexi- 1. A few number of FACTS devices with mediocre reac-
ty of the FACTS adjustment problem. In particular, the meth- tance control range can bring significant savings.
od presented in this paper gained a speedup factor of around 2 2. The PTFD-based models outperform the 𝐵𝜃-based meth-
by using an adaptive threshold instead of the fixed 70% ods in terms of computational time.
threshold. Note that lines equipped with FACTS should al- 3. The engineering insight that was used to reformulate the
ways be monitored. Thus, monitoring fewer lines would lead MILP-based DCOPF to a LP model is extremely effec-
to a significant improvement of the computational efficiency tive. It is significantly faster than the MILP model and
with fewer FACTS devices. Increasing the number of FACTS almost always finds the optimal solution. In fact, only 3
devices, and consequently increasing the number of lines out of the 520 cases simulated in the paper, converged to
needed to be monitored, would increase the computational a suboptimal solution using the LP-based method. For this
time. The candidate lines for FACTS are likely already a part 0.6% of the cases, which occurred for the Polish 2383-bus
of the lines identified by the fixed 70% thresholds. However, system, the difference between the suboptimal solution
many of those lines are not included in the set specified by achieved by the LP-based model, and the optimal solution
tighter thresholds. Therefore, the computational time saving reached by the MILP-based method, was below $3. Note
reduces as the number of lines equipped with FACTS devices that the potential savings that can be achieved with
increases. FACTS devices is more than $30,000 for the Polish 2383-
bus system. Therefore, even when the LP-based method
does not find the optimal solution, it finds a solution that
is very close to the optimal solution and does that very
quickly relative to the MILP-based method.

V. CONCLUSIONS
Variable impedance FACTS devices can provide signifi-
cant power flow control and, thus, reduce the cost, improve
the system reliability, and enhance the integration of renewa-
ble resources. Despite the benefits of FACTS devices, their
utilization has been limited due to a variety of factors includ-
ing the computational complexity of FACTS set point adjust-
ment within the DCOPF problem. Inclusion of FACTS flexi-
bility in the DCOPF, originally an LP, would introduce non-
Fig. 8. Average computational time (over reactance control range) for the
PTDF-based models with different thresholds for the results shown in Fig. 4. linearities with significant computational burden. FACTS set
point adjustment would also change the initial PTDF matrix.
This paper first models the FACTS reactance control as addi-
tional injections and develops an NLP with a constant PTDF
matrix. The paper, then, reformulates this nonlinear problem
to an MILP. Moreover, the MILP is reformulated to an LP
using an engineering insight on the power flow directions.
Since the industry employs PTDF-based linear market solvers,
existing SCUC and SCED solvers can be straightforwardly
modified using the method developed in this paper. To include
the flexibility of FACTS devices, additional injections should
be added to the model, which does not impose significant ad-
ditional computational efforts. Additionally, all the lines
equipped with FACTS should be monitored for thermal flow
violation to ensure FACTS injections remain within the de-
Fig. 9. Average computational time (over reactance control range) for the vice’s reactance control range. The simulation studies on
PTDF-based models with different thresholds for the results shown in Fig. 6. IEEE-118 bus as well as the Polish 2383-bus system show that
a limited number of FACTS devices can provide significant
C. Discussion
cost savings. The method presented in this paper was able to
The method proposed in this paper models the flexibility of effectively identify quality solutions with minimal computa-
variable impedance FACTS devices within PTDF-based tional effort. In fact, the LP-based method almost always
DCOPF. Since SCUC and SCED solvers in the industry em- (more than 99% of the time) found the optimal solution. For
ploy the same structure, the model presented here is consistent the very few cases, where the optimal solution was not found,

0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2016.2591503, IEEE
Transactions on Power Systems
8

the method was still able to quickly find a solution that was [23] F. Qiu and J. Wang, “Chance-constrained transmission switching with
guaranteed wind power utilization,” IEEE Trans. Power Syst., vol. 30,
very close to the optimal solution (less than 0.01% different). no. 3, pp. 1270-1278, May 2015.
[24] N. G. Hingorani and L. Gyugyi, Understanding FACTS: concepts and
REFERENCES technology of flexible AC transmission systems, Wiley-IEEE Press, Dec.
1999.
[1] A. Spencer, “National transmission grid study,” US Department of En- [25] P. Fairley, “Flexible AC transmission: the FACTS machine,” IEEE
ergy, 2002, [Online]. Available: Spectrum Magazine, Jan. 2011.
http://www.ferc.gov/industries/electric/gen-info/transmission-grid.pdf. [26] M. Sahraei-Ardakani and S. Blumsack, “Marginal value of FACTS
[2] S. W. Snarr, “The commerce clause and transmission infrastructure devices in transmission-constrained electricity markets,” in Proc. IEEE
development: an answer to jurisdictional issues clouded by protections,” PES General Meeting, Jul. 2013.
Electr. J., vol. 22, no. 5, pp. 8–18, June 2009. [27] J. Cardell, “A real time price signal for FACTS devices to reduce trans-
[3] PJM, “PJM state of the market,” 2013, [Online]. Available: mission congestion,” in Proc. of 40th HICSS, 2007.
http://www.monitoringanalytics.com/reports/PJM_State_of_the_Market/ [28] M. Sahraei-Ardakani and S. Blumsack, “Transfer capability improve-
2013.shtml. ment through market-based operation of series FACTS devices,” IEEE
[4] E. B. Fisher, R. P. O'Neill, and M. C. Ferris, “Optimal transmission Trans. Power Syst., accepted for publication.
switching,” IEEE Trans. Power Syst., vol. 23, no. 3, pp. 1346-1355, [29] M. Sahraei-Ardakani, “Policy analysis in transmission-constrained elec-
Aug. 2008. tricity markets,” Ph.D. Dissertation, Department of Energy and Mineral
[5] C. Barrows and S. Blumsack, “Transmission switching in the RTS-96 Engineering, The Pennsylvania State University, University Park, 2013.
test system,” IEEE Trans. Power Syst., vol. 276, no. 2, pp. 1134-1135, [30] M. Sahraei-Ardakani and K. W. Hedman, “A fast LP approach for en-
May 2012. hanced utilization of variable impedance based FACTS devices,” IEEE
[6] W. Shao and V. Vittal, “Corrective switching algorithm for relieving Trans. Power Syst., vol. 31, no. 3, pp. 2204-2213.
overloads and voltage violations,” IEEE Trans. Power Syst., vol. 20, no. [31] P. A. Ruiz, A. Rudkevich, M. C. Caramanis, E. Goldis, E. Ntakou, and
4, pp. 1877-1885, Nov. 2005. C. R. Philbrick, “Reduced MIP formulation for transmission topology
[7] G. Hug, “Coordinated power flow control to enhance steady state secu- control,” in Proc. 50th Allerton Conference, pp. 1073-1079. Oct. 2012.
rity in power systems,” Ph.D. Dissertation, Swiss Federal Institute of [32] Electric Reliability Council of Texas, “Nodal protocols – Section 7:
Technology, Zurich, 2008. congestion revenue rights,” 2015, [Online]. Available:
[8] Y. Xiao, Y. H. Song, C. C. Liu, and Y. Z. Sun, “Available transfer capa- http://www.ercot.com/mktrules/nprotocols/current.
bility enhancement using FACTS devices,” IEEE Trans. Power Syst., [33] PJM, “PJM manual 11 – Energy and ancillary services market opera-
vol. 18, no. 1, pp. 305-312, Feb. 2003. tions,” 2015, [Online]. Available:
[9] R. S. Wibowo, N. Yorino, M. Eghbal, and Y. Zoka, “FACTS devices http://www.pjm.com/~/media/documents/manuals/m11.ashx
allocation with control coordination considering congestion relief and [34] California ISO, “Market optimization details – technical bulletin,” 2009,
voltage stability,” IEEE Trans. Power Syst., vol. 26, no. 4, pp. 2302- [Online]. Available: http://www.caiso.com/23cf/23cfe2c91d880.pdf.
2310, Nov. 2011. [35] S. N. Singh and A. K. David, “Optimal location of FACTS devices for
[10] T. Orfanogianni and R. Bacher, “Steady-state optimization in power congestion management,” Elect. Power Syst. Res., vol. 58, no. 2, pp. 71–
systems with series FACTS devices,” IEEE Trans. Power Syst., vol. 18, 79, Jun. 2001.
no. 1, pp. 19-26, Feb. 2003. [36] M. Noroozian and G. Anderson, “Power flow control by use of control-
[11] S. M. Amin, "Securing the electricity grid,” NAE The Bridge, vol. 40, lable series components,” IEEE Trans. Power Del., vol. 8, no. 3, pp.
no. 1, pp. 13-20, Spring 2010. 1420–1429, Jul. 1993.
[12] C. Lehmkoster, “Security constrained optimal power flow for an eco- [37] I. Griva, S. G. Nash, and A. Sofer, Linear and Nonlinear Optimization,
nomical operation of FACTS devices in liberalized energy markets,” 2nd edition. Society for Industrial Mathematics, 2008.
IEEE Trans. Power Del., vol. 17, no. 2, pp. 603–608, Apr. 2002. [38] Dept. Electr. Eng., Univ. Washington, “Power system test case archive,”
[13] K. W. Hedman, R. P. O'Neill, E. B. Fisher, and S. S. Oren, “Optimal 2007 [Online]. Available: http://www.ee.washington.edu/research/.
transmission switching-sensitivity analysis and extensions,” IEEE Trans. [39] S. A. Blumsack, “Network topologies and transmission investment,
Power Syst., vol. 23, no. 3, pp. 1469-1479, Aug. 2008. under electric-industry restructuring,” Ph.D. dissertation, Eng. Public
[14] K. W. Hedman, M. C. Ferris, R. P. O’Neill, E. B Fisher, and S. S. Oren, Pol., Carnegie Mellon Univ., Pittsburgh, PA, 2006.
“Co-optimization of generation unit commitment and transmission [40] R. D. Zimmerman, C. E. Murillo-Sánchez, and R. J. Thomas,
switching with N-1 reliability,” IEEE Trans. Power Syst., vol. 25, no. 2, “MATPOWER: steady-state operations, planning and analysis tools for
pp. 1052-1063, May 2010. power systems research and education,” IEEE Trans. Power Syst., vol.
[15] M. Sahraei-Ardakani and S. Blumsack, “Market equilibrium for dis- 26, no. 1, pp. 12-19, Feb. 2011.
patchable transmission using FACT devices,” in Proc. IEEE PES Gen- [41] R. D. Zimmerman, C. E. Murillo-Sánchez, and D. Gan, MATPOWER,
eral Meeting, Jul. 2012. [Online]. Available: http://www.pserc.cornell.edu/matpower/
[16] A. S. Korad and K. W. Hedman, “Robust corrective topology control for
system reliability,” IEEE Trans. Power Syst., vol. 28, no. 4, pp. 4042– Mostafa Sahraei-Ardakani (M’06) received the PhD degree in energy engi-
4051, Nov. 2013. neering from The Pennsylvania State University, University Park, PA in 2013.
[17] M. Sahraei-Ardakani, X. Li, P. Balasubramanian, K. W. Hedman, and He also holds the B.S. and M.S. degrees in electrical engineering from Univer-
M. Abdi-Khorsand, “Real-time contingency analysis with transmission sity of Tehran, Iran. He was a post-doctoral scholar in the School Electrical,
switching on real power system data,” IEEE Trans. Power Syst., vol. 31, Computer and Energy Engineering at Arizona State University from 2013 to
no. 3, pp. 2501-2502. 2016. Currently, he is an assistant professor at the Department of Electrical
[18] P. Balasubramanian, M. Sahraei-Ardakani, X. Li, and K. W. Hedman, and Computer Engineering at the University of Utah. His research interests
“Towards smart corrective switching: analysis and advancement of include energy economics and policy, electricity markets, power system opti-
PJM's switching solutions,” IET Gen. Trans. & Dist., vol. 10, no. 8, pp. mization, transmission network, and the smart grid.
1984-1992.
[19] J. D. Lyon, S. Maslennikov, M. Sahraei-Ardakani, T. Zheng, E. Litvi- Kory W. Hedman (S’ 05, M’ 10) received the B.S. degree in electrical engi-
nov, X. Li, P. Balasubramanian, and K. W. Hedman, “Harnessing Flexi- neering and the B.S. degree in economics from the University of Washington,
ble Transmission: Corrective Transmission Switching for ISO-NE,” Seattle, in 2004 and the M.S. degree in economics and the M.S. degree in
IEEE Power & Ener. Tech. Sys. J., accepted for publication. electrical engineering from Iowa State University, Ames, in 2006 and 2007,
[20] M. Sahraei-Ardakani and K. W. Hedman, “Day-ahead corrective ad- respectively. He received the M.S. and Ph.D. degrees in industrial engineering
justment of FACTS reactance: a linear programming approach,” IEEE and operations research from the University of California, Berkeley in 2008
Trans. Power Syst., vol. 31, no. 4, pp. 2867-2875. and 2010 respectively. Currently, he is an assistant professor in the School of
[21] W. Shao and V. Vittal, “LP-based OPF for corrective facts control to Electrical, Computer, and Energy Engineering at Arizona State University. He
relieve overloads and voltage violations,” IEEE Trans. Power Syst., vol. previously worked for the California ISO (CAISO), Folsom, CA, on transmis-
21, no. 4, pp. 1832–1839, Nov. 2006. sion planning and he has worked with the Federal Energy Regulatory Com-
[22] A. Nasri, A.J. Conejo, S. J. Kazempour, and M. Ghandhari, “Minimizing mission (FERC), Washington, DC, on transmission switching. His research
wind power spillage using an OPF with FACTS devices,” IEEE Trans. interests include power systems operations and planning, electricity markets,
Power Syst., vol. 29, no. 5, Sep. 2014. power systems economics, renewable energy, and operations research.

0885-8950 (c) 2016 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.

You might also like