EFFECTIVENESS OF THE DISTRIBUTION FACTOR

APPROXIMATIONS USED IN CONGESTION MODELING

Minghai Liu George Gross
University of Illinois at Urbana Champaign University of Illinois at Urbana Champaign
Illinois, U.S.A. Illinois, U.S.A.
[email protected] [email protected]

Abstract – Congestion has widespread impacts on the power systems computed for a specified network topol-
availability and utilization of the existing transmission ogy and parameter values. However, in many applica-
systems and consequently on the operation of competitive tions such as ATC evaluations and NERC’s TLR proce-
markets in electricity. The distribution factors play a key dures, distribution factors are considered to be constant
role in the modeling of congestion in various market appli-
in each time period even when the network parameters
cations. These factors are linear approximations of sensi-
tivities of variables with respect to various inputs and are and topology are slightly different from those for which
computed for a specified network topology and parameter the distribution factors are computed [1,2]. This usage
values. In practice, the factors are used over a wide range imposes questions on the robustness of the distribution
of system conditions. This paper investigates the analytical factor applications. In fact, systematic studies of the
characteristics, the robustness and the quality of the ap- behaviors of these distribution factors and the effective-
proximations provided by key distribution factors such as ness of their applications in the congestion modeling
injection shift factors (ISFs) and power transfer distribu- have received scant attention so far.
tion factors (PTDFs). We examine the range of conditions
This paper investigates the analytical characteristics,
over which these factors can provide a reliable approxima-
tion for large power system networks. The numerical simu-
the effectiveness and robustness of approximations
lation results indicate that the errors of the approximations provided by key distribution factors such as ISFs and
stay in an acceptable range under a broad spectrum of PTDFs. Starting with the actual derivation of these fac-
conditions including contingencies used to establish n-1 tors, we analyze their characteristics and examine the
security. range of conditions over which these factors can provide
a reliable approximation for large power system net-
Keywords: Distribution Factors, Injection Shift works. In particular, we examine the effect of contin-
Factor (ISF), Power Transfer Distribution Factor gencies represented by changes in the network parame-
(PTDF), Available Transfer Capability (ATC), Trans- ters and topology and investigate the robustness of the
mission Loading Relief (TLR), Congestion Modeling PTDF and ISF applications in congestion modeling.
Numerical studies on various systems are provided to
1 INTRODUCTION examine the robustness of the ISF and PTDF approxi-
mations for ATC information and TLR curtailments
Open access of the transmission network has posed under a variety of loadings, system conditions and pa-
serious problems in the management of the transmission rameter values.
system. The congestion on the network becomes the This paper consists of four additional sections. Sec-
main obstacle that, in great extend, impacts the opera- tion 2 reviews the definition and characteristics of the
tion and management of the power system. Critical in- basic distribution factors. In section 3, the role of the
formation such as available transfer capability (ATC) distribution factors in the congestion modeling is illus-
[1], congestion relief schemes such as the transmission trated by focusing on the evaluation of ATC and the
loading relief (TLR) procedure [2] and congestion man- deployment of NERC’s TLR procedure. We devote
agement approaches such as incremental/decremental section 4 to analyze the effects of the changes in the
auctions and financial transmission rights (FTR/FGR) network parameters and topology and their impacts on
[3,4] are all impacted by the congestion situations on the the quality of distribution factor approximations. We
grid. Solution of these problems requires explicit model- summarize representative numerical results in section 4
ing of congestion. to illustrate the robustness of the ISF and PTDF ap-
Distribution factors play an essential role in conges- proximations in ATC evaluation and TLR curtailments
tion modeling. These distribution factors, including the determination.
injection shift factors (ISFs) and the power transfer
distribution factors (PTDFs) have been widely used in
2 BASIC DISTRIBUTION FACTORS
the congestion modeling in many electricity market
applications by providing fast approximations of the We consider a system with N+1 buses and L lines.
active power flow changes due to various system opera- We denote by N ! {0, 1, 2," , N } the set of buses, with
tions. These factors are basically linear approximations the slack bus at bus 0, and by L ! {#1 , # 2 ," , # L } the set
of the first order sensitivities of certain relationships in
of transmission lines and transformers that connect the

Proceedings of the 14th Power Systems Computation

Conference. Sevilla, Spain. June 24-29, 2002.
buses in the set N . We denote each element # ∈ L by −1
∂h ∂h  ∂ g 
the ordered pair # = (i, j ) with the convention that the ∆f ≈ ∆s =   ∆p (7)
∂s ∂s  ∂s 
direction of the flow on line # is from node i to node j.
(0 )
The serial admittance of line # is g # − jb# , the active The sensitivity matrix in (7) depends on s and this
power flow is f # and f ! [ f1 , f 2 ," , f L ]T . The net dependence on the system operating point makes it less
than practical for power system applications.
active power injection at node n ∈ N is denoted by p n To simplify the computation of the sensitivity matrix,
and we define p ! [ p1 , p 2 ," , p N ]T . Transactions are we next introduce the assumptions used in the derivation
of DC power flow models and make use of the reduced
represented by the set of power injection-withdrawal (I-
nodal susceptance matrix [5], B ! A% T B ′A% , where
W) node pairs, W ! {w1 , w2 ," , wW } , with each element
B ′ ! diag [ b1 , b2 ," , bL ] is the diagonal branch suscep-
in this set denoted by the ordered triplet w = {m, n, t}
tance matrix and A% ! [ a% 1, a% 2 ," , a%L ] is the branch-to-
T
representing an I-W node pair with from node m, to node
n, in the amount t. node incidence matrix with the row # of the matrix:
We study the response of the active line flow to i j

changes in nodal injections. Consider the nodal injection a%# ! [0" 0 1 0" 0 -1 0" 0]T . We assume B to be non-
vector p and the corresponding active line flow vector singular. Under these assumptions, s reduces to θ and
T the expressions for the partial derivatives become
f . Denote the system state by s ! θ Τ , V T  , where ∂ g ∂θ ≈ B , ∂h# ∂θ ≈ b# a%# . We furthermore define
θ ! [θ 1 ,θ 2 ," ,θ N ]T ( V ! [V 1 ,V 2 ," ,V N ]T ) is the volt-
A ! B ′A% to be the “admittance weighted” branch-node
age phase angle (magnitude) vector. Denote the refer- incidence matrix, then
(0 ) (0 ) (0 )
ence conditions by p , s and f that satisfy: ∆ f ≈ AB −1 ∆ p ! Ψ ∆ p (8)
 g ( s (0 ) ) − p (0 ) = 0 (1) We henceforth replace the approximation by the equal-
 (0 ) ity:
 h( s ) − f = 0
(0 )
(2)
∆ f =Ψ ∆ p . (9)
where (1) represents a statement of the active power
The L × N matrix Ψ ! AB -1 is an approximation of the
flow equations and the component # of h(i) ,
sensitivity matrix and is called the injection shift factor
h# ( s ) = g# V i −V iV j cos(θ i − θ j )  +b#V iV j sin(θ i − θ j ) (3)
2
(ISF) matrix. Since A and B are solely determined by
 
is the expression for the active flow on line the network topology and the line parameters, Ψ is
(0 )
# = (i, j ), # ∈ L . For a small change ∆ p that changes the independent of s . The ISF of a line # ∈ L with re-
(0 ) spect to a change in injection at node n ∈ N, n ≠ 0 is the
value from p(0 ) to p + ∆ p , we denote by ∆ s ( ∆ f )
element ψ #n in row # , column n of Ψ . Note that ψ #n is
the corresponding change in the state s (active line
defined implicitly under the assumption that there is a
flows f ). We assume the system stays in balance for
corresponding change ∆ p0 in the injection at the slack
the change ∆ p and neglect the changes in losses so that,
node 0 with ∆ p0 = −∆ p n . Therefore, the ISF is depend-
for every MW increase in the injection at node n ≠ 0 , ent on the slack bus. As the location of the slack bus
there is a corresponding MW increase in the withdrawal changes, the values of the ISFs may change. The notion
at the slack node 0. In other words, ∆ p0 = − ∑ ∆ p n . of the ISF may be extended to include the slack bus 0.
n∈N, n ≠0
Since the injection and withdrawal buses are identical in
We apply the first order Taylor’s series expansion near this case, ψ #0 ≡ 0 for any # ∈ L .
(0 )
the reference point s : In many applications, the impacts of changes in the
 (0 ) (0 )
∂g quantity of an I-W node pair on the active line flows are
 g ( s + ∆ s) = g ( s ) + ∆ s + h.o.t. (4) of interest. We may evaluate the change in the active
 ∂ s s( 0 )
 flow on a line # due to a change ∆t in the transfer
 h( s (0 ) + ∆ s ) = h( s (0 ) ) + ∂ h ∆ s + h.o.t. (5) quantity of an I-W node pair w ={m, n, t}∈W with ISFs.
 ∂ s s(0 )
 This change is represented by setting ∆ p m =∆t =−∆ p n .
For “small” ∆ p , ∆s is “small” and so we neglect the The corresponding active flow change on line # is
higher order terms (h.o.t.). We furthermore assume ∆ f # = ψ #m ∆ p m + ψ #n ∆ p n = (ψ #m −ψ #n )∆t . (10)
∂ g ∂ s ( 0 ) to be nonsingular and henceforth drop the The ISF difference term is called the power transfer
s
distribution factor (PTDF) of line # with respect to the
bar in the notation so that:
−1
I-W node pair w ∈ W [1] and is defined by
∆ s ≈  ∂ g ∂ s  ∆ p (6) ϕ #( w ) ! ∆ f # ∆t = ψ #m −ψ #n . (11)
In this case, the compensation at the slack bus cancels on each line # ∈ L due to w . We wish to determine the
out since ∆ pm − ∆ pn = (∆ pm − ∆ p0 ) − (∆ pn − ∆ p0 ) . As maximum amount ∆t that can be transferred without
such, the PTDF is independent of the slack bus. causing any congestion. This problem is formulated as:
A line # = (i, j ) is radial if either H i = {#} or max ∆ t
H j = {#} , where H i ( H j ) is the set of lines that con- s.t. ∆ f # = ϕ #( w) ∆ t ≤ f #max − f #(0 ) ∀# ∈ L
nect to node i (j). For the radial line # with H i = {#} , where we explicitly use the PTDF representation of the
active line flow change. The optimal value of ∆ t is then
i ≠0 ,

{
the UTC quantity from node m to node n and
ψ #n = 10 if n = i (12)
otherwise  f #max − f #(0 )  f #max − f #(0 )
UTC m, n = min  ! (15)
 ϕ# ϕ (#w)
( w)
#∈L , ϕ #( w ) > 0
since the only impact on line # comes from the injec- 
tion at node i. For any other line #′ ≠ # , the injection where # is a line whose active flow limit determines
change at the terminal nodes i and j has the same impact, UTC m , n and is referred to as a binding constraint line.
ψ #i ′ = ψ #j′ ∀#′ ≠ # . (13) Clearly, the PTDFs play a key role in the determina-
tion of UTC. In fact, the PTDF representation results in
3 APPLICATIONS TO CONGESTION MODELING the straightforward computation of UTC since only one
Congestion has widespread impacts on the availabil- traversal of all the lines is required to compute UTC m , n
ity and utilization of the existing transmission systems. for each pair of nodes m and n.
Key problems in the electricity market including the We next discuss the role of the PTDFs in the trans-
determination of available transfer capability (ATC) [1], mission loading relief (TLR) procedure. TLR is the pro-
the implementation of the congestion management ap- tocol used by NERC to prevent insecure operation of the
proaches and the definition of transmission rights [3,4], interconnected grid in the Eastern Interconnection. The
are all impacted by the congestion situations on the grid. TLR procedure is invoked whenever some present or
The solution of these problems requires explicit model- future insecure situations are identified, such as those
ing of congestion. The models used have in common the arising when some proposed future transaction(s) load
application of distribution factors. In this section, we the network beyond specified operating security limits.
discuss the role of the distribution factors in congestion Of the five different levels associated with TLR, three
modeling by focusing on two important areas: the involve the rearrangement of the transactions and two
evaluation of ATC and the deployment of NERC’s TLR require the curtailment, in part or whole, of the transac-
procedure. tions. Our focus is on the levels 3 or 5 that involve
The available transfer capability (ATC) provides a transaction curtailments [2].
measure of the transfer capability remaining in the We represent the limit violation by the active over-
physical transmission network for further commercial load on a line # denoted by
activity over and above already committed uses [1]. δ f # = f # − f #max . (16)
ATC represents the maximum additional MW that can be We partition the I-W node pairs into two groups:
transferred between two specific areas while meeting all
the defined pre- and post-contingency system condi- W + ! {w ∈ W : ϕ #( w) > 0} and W − ! {w ∈ W : ϕ #( w ) ≤ 0 } .
−
tions. A key component for the evaluation of ATC is the Without loss of generality, we assume the set W does
so-called uncommitted transfer capability (UTC) which not impact the overflow and the overflow δ f # may be
is the total transfer capability (TTC) minus the existing
transmission commitments. The computation of the ATC allocated to each w ∈W + so that
from UTC is straightforward since ( w) ϕ #( w) t
δ&f # = δ f# (17)
ATC = UTC – CBM – TRM
where CBM is the capacity benefit margin and TRM is
(14) ∑ ϕ #( w′) t ′
w′∈W +
the transmission reliability margin [1]. All the transfer is considered to be the portion of overflow attributable
capability quantities are defined with respect to a send- to w ∈W + [6]. The corresponding transfer amount is:
ing node/area and a receiving node/area. ( w)
The modeling of congestion is a key issue in the de- ' ( w) δ&f # t
δ t = ( w) = δ f# (18)
termination of UTC. We denote by f max the maximum ϕ# ∑ ϕ #( w′)t ′
w′∈W +
(0 )
active line flow limits and by f the active line flow To relieve the congestion, we may curtail each I-W node
( w)
corresponding to the existing transmission commitments according to δ't . Since ϕ ( w) are very small for many
#
constituting the reference case. Without loss of general-
w ∈W + , their contributions to the congestion relief,
ity, we assume that the UTC quantities are determined ( w)
by the dominant flows [6]. We consider UTC m , n , where ϕ ( w ) δ't , is small. Consequently, NERC defined the set
#

m is the from node and n is the to node. We introduce an Wˆ = {w ∈ W +: ϕ (#w) ≥ 0.05} which excludes the I-W
additional I-W node pair, w = {m, n, ∆t} , to the system
node pairs with PTDF less than 0.05. NERC used the
and evaluate the change ∆ f # in the active power flow allocation rule given by:
 (ϕ ) t δ f
ˆ ˆ
( w) 2 ψ#i −ψ#j
δf ( w)
=
#
(19) φ#,#ˆ = − ˆ ˆ
is proportional to the quantity
b ˆ# ∆b #ˆ + (ψ#ˆi −ψ#ˆj )
∑ (ϕ ) t ′
# #
( w) 2
#
ψ #i − ψ #j . Note that if both B and B + A% LTˆ ∆B ′Lˆ A% Lˆ are
ˆ ˆ
w′∈Wˆ

to determine the overflow attributed to w ∈ Wˆ . Then nonsingular, B ′Lˆ ∆B ′Lˆ +Ψ Lˆ A% LTˆ is invertible [8].
−1

the curtailment associated with each w ∈ Wˆ is:

Network topology changes such as line outages and
δ f ( w) ϕ #( w) t line additions may be considered as special cases of
δ t ( w) = (#w ) = δ f# . (20)
ϕ# (
∑ # ϕ )
( w′ ) 2
t ′ parameter changes. For example, for the outage of a line
w′∈Wˆ #ˆ = (iˆ, ˆj ) , Lˆ = {#ˆ } , A% Lˆ = a% T#ˆ and ∆b #ˆ = −b #ˆ , so that:

4 IMPACT OF CHANGES IN NETWORK

−ψ Τˆ if # = #ˆ
Τ
 #
TOPOLOGY AND PARAMETER VALUES ∆ψ # =  ψ i#ˆ − ψ #ˆj (22)
The ISFs and PTDFs play a key role in congestion  iˆ ˆj
ψ Τ#ˆ otherwise
 1 − (ψ #ˆ − ψ ˆ# )
modeling used in the new competitive environment.
ˆ ˆ
Clearly, these factors are evaluated for a given topology ψ i# −ψ #j
and parameter values and an operating point that satis- where the factor φ # , #ˆ = ˆ ˆ is called line outage
1 − (ψ i#ˆ −ψ #ˆj )
fies, to a greater or lesser extent, the assumptions cited
in the previous section. However, in many cases of in- distribution factor [1] which establishes the relationship
terest, there are changes in the network topology, between the pre-outage active flow f #ˆ on line #̂ and
parameter values and the operating point, while the ISFs
the change ∆ f# on the active flows on line # ≠ #ˆ due to
and PTDFs are held constant in the applications in
which they are used. Such usage, in effect, neglects the the outage of line #̂ with ∆ f # = φ # ,Lˆ f #ˆ . Note that
impacts of these changes. In this section, we evaluate the
ψ i#ˆ − ψ #ˆj = 1 only when {#} is a cutset of the network
ˆ ˆ
effect of these changes and their impacts on the quality
of distribution factor approximations. [7]. In that case, the outage of line # breaks the system
We first consider the impacts of changes in network into two separate subnetworks and the ISFs needs to be
parameters. Let us denote by Lˆ ! #ˆ , #ˆ ," , #ˆ ˆ ⊆ L the { 1 2 L } redefined for each subnetwork.
Another example is the addition of a line #ˆ = (iˆ, ˆj ) .
subset of lines whose parameters are changed. For each
Two possible situations of interest are:
line #ˆ ∈ Lˆ , its line susceptance is changed from b #ˆ to
(i) #̂ is a radial line with iˆ ∉ N whose addition re-
b #ˆ + ∆b #ˆ . Denote the analogues of the matrices
sults in L% = L ∪ #ˆ and N% = N ∪ iˆ . We may apply
B ′ ( L × L ), A% and Ψ ( L × N ) corresponding to the (12), (13) to construct the augmented ISF matrix
lines in L̂ by B ′Lˆ ! diag[b #ˆ , b #ˆ ,", b #ˆ ] ( Lˆ × Lˆ ), Ψ ψ iˆ 
1 2 Lˆ Ψ% =  T  (23)
A% Lˆ ! [a% #ˆ , a% #ˆ ," , a% #ˆ ˆ ] T
and Ψ Lˆ ! [ψ #ˆ ,ψ #ˆ ," ,ψ #ˆ ˆ ]T  0 1 
1 2 L 1 2 L ˆ ˆ
where ψ i = ψ j , the column ĵ of Ψ ;
( L̂ × N ) where ψ T
ˆ# is row #̂ of Ψ , the ISF matrix. Let
(ii) #̂ is a new line with iˆ, ˆj ∈ N whose addition re-
∆B ′Lˆ ! diag[∆b #ˆ , ∆b #ˆ ," , ∆b #ˆ ˆ ] , ∆b#ˆ ≠ 0, ∀#ˆ ∈ Lˆ . The
1 2 L
sults in L% = L ∪ #ˆ . We define a new ISF row vec-
changes in L̂ result in changing the B matrix into tor ψ% Τ#ˆ ! a T#ˆ B = b #ˆ a% T#ˆ B and construct the aug-
B + A% LTˆ ∆B′ Lˆ A% Lˆ . This, in turn, changes each row of the mented ( L + 1) × N ISF matrix
ISF matrix by:
 Ψ + ∆Ψ 
∆b# Τ b# +∆b# Τ % T Ψ% =  Τ T (24)
ψ# ALˆ (B′Lˆ ∆B′Lˆ +ΨLˆ A%LTˆ )−1ΨLˆ # ∈ Lˆ
−1
Τ b
ψ# −
b ψ% #ˆ + ∆ψ #ˆ 
∆ψ # =  # # (21)
−ψΤ A % Tˆ (B′ˆ ∆B′−ˆ1 +Ψˆ A % Tˆ )−1Ψˆ ˆ where ∆ψ T#ˆ and each row of ∆Ψ is determined
 # L L L L L L
# ∉L
ˆ ˆ
Τ ψ i# −ψ #j
The derivation of (21) is straightforward using the by ∆ψ # = − ˆ
%Τ .
ˆ ψ #ˆ
Sherman-Morrison-Woodbury formula [8]. 1 + (ψ% i#ˆ − ψ% #ˆj )
For # ∉ Lˆ , the L̂ -dimensional row vector A change in the network topology or parameter values
φ # ,Lˆ ! −ψ Τ# A% LTˆ ( B ′Lˆ ∆B ′Lˆ +ΨLˆ A% LTˆ ) −1 establishes the rela-
T −1 will change the PTDF values from those in the base
case. However, industry practice typically does not
tionship between the pre-change active flows update the values; instead, the base case values are kept.
f Lˆ ! [ f ˆ# , f #ˆ ," , f #ˆ ˆ ]T and the change ∆ f # in the active This practice, in turn, impacts the quality of the PTDF
1 2 L
applications in congestion modeling. We investigate
flows on line # ∉ Lˆ due to the parameter changes with these impacts by evaluating the relative errors in the
∆ f # = φ T# ,Lˆ f Lˆ . Particularly, if Lˆ = {#ˆ = (iˆ, ˆj )} , then
calculation of ATC and TLR actions due to the neglect of activities. The error bounds indicate that these impacts
the PTDF changes. are minor.
We first consider the ATC application. Denote by Ψ
and ϕ #( w) the ISF matrix and the PTDF value of each 5 SIMULATION RESULTS
line # ∈ L with respect to each I-W node pair w ∈W To investigate the quality and robustness of the dis-
for the modified (reference) network, respectively, and tribution factor approximations for congestion model-
by Ψ + ∆Ψ and ϕ #( w ) + ∆ϕ #( w ) those that are used to ing, we have simulated various cases on a number of test
systems including the IEEE 118-bus system and portions
evaluate UTC m, n . From (15)
of the Eastern Interconnection of the United States. Our
 f #max − f #(0 )  f max −f (0 )
simulations indicate that the errors of the approxima-
UTC m , n (Ψ ) = min  !
# #

 ϕ# ϕ (#w) tions stay in an acceptable range under a broad spectrum

(w) ( w)
ϕ # >0

of conditions. In this section, we summarize representa-
 f max − f #(0 )  f # − f #
max (0 )

UTC m, n (Ψ +∆Ψ ) = ( w )min( w )  (#w) ! tive results of our studies.

( w) 
ϕ # +∆ϕ # > 0 ϕ
 # + ∆ϕ #  ϕ # + ∆ϕ #
( w) ( w)
We focus our discussion on three sets of representa-
tive network conditions: (a) the base case condition; (b)
f max
−f (0 )
f max
−f (0 )
the 50% reactance cases where, for a set of specified
It follows that # #
≤ # #
and
ϕ # + ∆ϕ #
( w) ( w)
ϕ ( w)
#
+ ∆ϕ ( w)
#
lines, we decrease by 50% the reactance of each line
individually; and (c) the line outage cases where each
UTC m, n (Ψ +∆Ψ ) − UTC m, n (Ψ ) ϕ (#w) −∆ϕ (#w) line in the same set is outaged individually.
≤ − 1 ≈
UTC m, n (Ψ ) ϕ (#w) + ∆ϕ (#w) ϕ (#w) We use the AC power flow results for benchmark
The relative errors in the UTC are bounded by the rela- purposes and use the absolute value of the relative error
tive errors in the PTDF of the binding constraint lines. as the metric in our assessment. Let ytrue be the quantity
Next we investigate the impact of the errors in Ψ on computed using the AC power flow and y be the result
the curtailment quantity computation in the TLR proce- obtained using the distribution factors, then the absolute
dure. We adopt the same notation and assume line # is value of the relative error,
congested with the active overflow δ f # . The curtail- ε y ! ( y − ytrue ) ytrue (25)
ment quantities associated with each w ∈ Wˆ in the two is a unit free quantity. We henceforth refer to ε y as
cases are determined by:
simply the relative error. The relative errors of interest
ϕ #( w) t
δ t ( w) (Ψ ) = δ f# are those associated with the active line flows to evalu-
∑ (ϕ ) ( w′ ) 2
# t′ ate the overall quality of the distribution factor ap-
w′∈Wˆ (Ψ ) proximations. In addition, we evaluate the relative errors
(ϕ ( w)
+ ∆ϕ #( w ) ) t in the UTC quantities and the TLR curtailments in the
(Ψ +∆Ψ ) =
#
δt ( w)
δ f# ATC and TLR studies, respectively.
∑ (ϕ + ∆ϕ #( w′) ) t ′
( w′ ) 2
# We discuss first the quality of the ISF approximations
w′∈Wˆ (Ψ + ∆Ψ ) in evaluating the active line flow changes due to changes
Without loss of generality, we assume ∆ϕ #( w) > 0 and in nodal injections under various conditions. We use the
consider two situations of interest: ISF values of the base case for all the cases studied. For
the base case condition (a), we study the approximation
(i) ϕ #( w ) < 0.05 ≤ ϕ #( w) + ∆ϕ #( w) : δ t ( w ) (Ψ ) = 0 , the error
with different values of injections at a set of specified
δ t ( w ) (Ψ +∆Ψ ) − δ t ( w) (Ψ ) = δ t ( w ) (Ψ +∆Ψ ) is typi- buses. We vary the injection in step of 10% from 0 to
cally small since both ϕ #( w) and ∆ϕ #( w) are small 200% of the base case value for each bus in the set and
evaluate the active power flow changes on all the lines
(ii) ϕ #( w ) ≥ 0.05, ϕ #( w ) + ∆ϕ #( w ) ≥ 0.05 : the error using the ISF expression. This sensitivity study is re-
∆ϕ #( w ) t peated for every bus in the set. We collect the relative
δ t ( w) (Ψ +∆Ψ ) − δ t ( w ) (Ψ ) ≈ δ f# errors in these results and evaluate their density distribu-
∑ (ϕ ) ( w′ ) 2
# t′ tion. The resulting plot is shown in Figure 1(a). We
w′∈Wˆ (Ψ )
perform the same study for every case in the set for (b)
is also small since ∆ϕ ( w)
# t << ∑ (ϕ ) ( w′ ) 2
# t′ and (c). We collect the relative errors of all the cases for
w′∈Wˆ (Ψ ) the respective set and plot their distribution density in
Therefore, the impact of the errors in the PTDFs on the Figure 1(b) and (c), respectively. We also display in
TLR curtailment quantity is bounded by a small quantity. Figure 2 the cumulative distributions of the relative
Notice that the 0.05 threshold helps reduce the impact of errors for these cases.
the errors in PTDFs by excluding I-W node pairs with Each of the density plots shows that the frequency for
small-valued PTDFs which are typically associated with the relative errors is high for small values and very low
more pronounced relative errors. for large values in each of the cases studied. The cumu-
In this section, we analytically determined the bounds lative plot makes clear that the base case ISF values
on the relative errors in the evaluation of UTC and TLR hold over a wide range of conditions including contin-
gencies. In fact, the scatter plot in Figure 3, showing the
size of relative error as a function of the magnitude of determined using the base case PTDFs with respect to
the corresponding ISF, further enforces the notion that those using the AC power flow for the cases (a), (b) and
large errors in ISF approximations are associated pri- (c). We collect the errors for each of these cases and
marily with the small magnitude ISFs. The plots in Fig- compute the distribution density of these errors. Figure 4
ure 1-3 are representative of the results we obtained with shows the cumulative distribution of the relative errors
the different systems studied. for these three cases. The plots indicate that, ignoring
0.03
the impacts of the changes in the network parameters
(a) base case and topology on the PTDFs results in the errors of the
0.025 UTC values that stay in an acceptable range over a wide
spectrum of conditions.
0.02
1.2
density

0.015
1
0.01

relative error
0.8

0.005 0.6

0 0.4
0 0.5 1 1.5 2 2.5 3
relative error
0.2
0.045
(b) 50% line re actance cases 0
0.04
-0.4 -0.2 0 ISF 0.2 0.4 0.6
0.035

0.03 Figure 3: Relationship of relative errors and ISFs

density

0.025

0.02 1

0.015

0.01 0.8
cumulative distribution

0.005

0 0.6
0 0.5 1 1.5 2 2.5 3
relative error
0.4
0.035
(c) line outage cases
base case
0.03 0.2 50% reactance cases
line outage cases
0.025
0
0 0.1 0.2 0.3 0.4 0.5
density

0.02 relative error

Figure 4: Cumulative error distribution of UTC approximation
0.015

0.01 0.05

0.005
0.04
0
0 0.5 1 1.5 2 2.5 3
density

relative error 0.03

Figure 1: Error distribution of ISF line flow approximations.
0.02
1
0.01
cumulative distribution

0.8
0
0 0.2 0.4 0.6 0.8 1
0.6 PTDF of binding constraint line
Figure 5: Distribution of PTDFs of binding constraint lines
0.4

base case
We display in Figure 5 the distribution density of the
0.2 50% re actance case s PTDF values of the binding constraint lines in the UTC
line outage case s calculations. The plot indicates that the binding con-
0
straint lines that determine the UTC quantities are typi-
0 0.5 1 1.5 2 2.5 3
re lative error cally associated with large-valued PTDFs. This result is
Figure 2: Cumulative distribution of the relative errors. particularly important considering the analytical bounds
derived in section 4. The analytical bounds derived for
We next assess the errors in the UTC calculations. A the UTC evaluations depend on the relative errors in the
similar study is performed in computing the UTC values PTDFs associated with the binding constraint lines and
the relative errors associated with large-valued PTDFs indicate that the errors of the approximations stay in an
are typically small. Consequently, the analytical bounds acceptable range under a broad spectrum of conditions
determined in section 4 are small. including contingencies used to establish n-1 security.
We also study the impacts of the PTDF errors on the An attractive characteristic is that larger errors are typi-
TLR curtailment quantities. We consider a set of conges- cally associated with the small-valued PTDFs and these
tions for each case in (b) and (c). We compute the TLR errors fail to affect the overall results in either ATC or
curtailment quantities using the base case PTDFs and TLR calculations. We will report the role of these factors
compare them with the corresponding curtailment quan- in the definition and implementation of the financial
tities computed using the PTDFs associated with the transmission rights (FTRs) and flowgate rights (FGRs)
modified network. We collect the errors and plot the in a future paper.
distribution density of the absolute errors for the cases
in (b) and (c) in Figure 6. ACKNOWLEDGMENT
0.45 The research reported here was performed under the
50% reactance cases
0.4 sponsorship of PSERC, CERTS and NSF.
0.35

0.3 REFERENCES
density

0.25
[1] Transmission Transfer Capability Task Force,
0.2 “Available Transfer Capability Definitions and De-
0.15 termination”, North American Electric Reliability
0.1 Council, Princeton, New Jersey, June 1996
0.05 [2] North American Electric Reliability Council, “Ap-
0 pendix 9C – Transmission Loading Relief Proce-
0 10 20 30 40 50
absolute error dure”, www.nerc.com
0.25 [3] W. Hogan, “Contract Networks for Electric Power
line outage cases
Transmission”, Journal of Regulatory Economics,
0.2 Vol. 4, 1992, pp. 211-242
[4] H. Chao and S. Peck, “A Market Mechanism for
0.15 Electric Power Transmission”, Journal of Regulatory
density

Economics, Vol. 10, 1996, pp. 25-59

0.1
[5] A. Wood and B. Wollenberg, “Power Generation
Operation and Control”, New York, John Wiley &
0.05
Sons, ISBN 0-471-58699-4, pp. 105-108
0
[6] G. Gross and S. Tao, “A physical-flow-based ap-
0 10 20 30
absolute error
40 50 proach to allocating transmission losses in a transac-
Figure 6: Error distribution of TLR curtailment
tion framework”, IEEE Transactions on Power Sys-
tems. Vol. 15 n. 2 May 2000, pp. 631-637
Figure 6 indicates that the errors in the PTDFs have [7] L. Chua and P. Lin, “Computer-Aided Analysis of
very limited impact on the TLR curtailment quantities in Electronic Circuits: Algorithms & Computational
both cases. Techniques”, Englewood Cliffs, Prentice-Hall, ISBN
0-13-165415-2, pp.134-140
6 CONCLUSION [8] G. Golub and C. Loan, “Matrix Computations”,
We reported on our investigation in the robustness Baltimore, The John Hopkins University Press,
and quality of approximations provided by key distribu- ISBN 0-8018-3772-3, pp. 49-53
tion factors such as ISF and PTDF. We examined the [9] Transmission Transfer Capability Task Force,
range of conditions over which these factors can provide “Transmission Transfer Capability”, North American
a reliable approximation for large power system net- Electric Reliability Council, Princeton, New Jersey,
works. This constitutes the first effort to systematically May 1995
assess the impacts of errors in the distribution factors in
the area of congestion modeling. Numerical results