You are on page 1of 15

EUROPEAN TRANSACTIONS ON ELECTRICAL POWER

Euro. Trans. Electr. Power (2011)


Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/etep.618

Including of branch resistances in linear power transmission


distribution factors for fast contingency analysis

Vladan D. Krsman1*,†, Andrija T. Sarić2 and Neven V. Kovački3


1
DMS Group LLC, 8 Micurinova Street, 21000 Novi Sad, Republic of Serbia
2
Faculty of Technical Sciences, University of Novi Sad, 6 Dositeja Obradovića Square, 21000 Novi Sad,
Republic of Serbia
3
DMS Group LLC, 4 Sremska Street , 21000 Novi Sad, Republic of Serbia

SUMMARY
This paper proposes an extension of the traditional DC-based power flow model for obtaining the linear
power transmission distribution factors (PTDFs) that includes the influence of the branch resistances. The
proposed model is adopted for the fast contingency analysis, including cases of injection changes, branch
outages and their combination. Proposed formulation saves the efficiency of the DC-based model, such as
linearity, fast matrix inversion by matrix inversion lemma for branch outages as well as fast block matrix
inversion. This model is appropriate for subtransmission and transmission networks, where neglecting the
branch resistances and other approximations introduced in DC-based power model is not acceptable. The
proposed generalized linear PTDF-based model is successfully verified on the basis of the results for two
examples: small (for educational purposes) test system and large-scale (for industry applicability purposes)
test system of Continental Europe Synchronous Area (ex-UCTE) for characteristic contingencies (including
the generation/load changes and branch outages). Copyright © 2011 John Wiley & Sons, Ltd.

key words: power systems; power flow; linear approximation; power transmission distribution factor
(PTDF); contingency analysis

1. INTRODUCTION

With ongoing industry restructuring, planning for and responding to network security issues have
increasingly become a challenge. Of particular concern is the ability of portions of the transmission
grid to allow transmission of power from generating stations to (possibly distant) loads.
Standard industrial practice for contingency analysis proceeds in following steps (so-called
“simultaneous feasibility test”): 1) calculate base AC power flow (or state estimation in real time);
2) check all limiting elements for violations; 3) screen all the contingencies—this is a process of
simulating each contingency from the given set one-by-one by DC-based power flow analysis; 4) check
each for potential violations and 5) run all suspicious contingencies through the AC power flow analysis.
Additional to the contingency analysis, where basic DC-based power flow model [1], or their
extensions [2,3] are standard tools, these models are widely used in different power system
applications and optimizations, where a large number of cases (and consequently the time-consuming
calculations) are requested for analysis, including:
• (fast) power flow analysis [4];
• power system equivalencing [5];
• optimal power flow [6];
• security assessment [7,8];

*Correspondence to: Vladan D. Krsman is with DMS Group LLC, 8 Micurinova Street, 21000 Novi Sad, Republic of
Serbia.

E-mail: vladan.krsman@dmsgroup.rs

Copyright © 2011 John Wiley & Sons, Ltd.


V. D. KRSMAN, A. T. SARIĆ AND N. V. KOVAČKI

• contingency analysis and screening [9–13];


• real-time security-constrained economic dispatch [14];
• calculation of available transfer capability [15];
• day-ahead security-constrained unit commitment [16];
• calculation of locational marginal prices, where prices are constrained by network congestion and
involve active independent system operator control [6,14];
• auctions and allocations associated with transmission rights for transmission load relief and
transfer analysis [17–19];
• medium to long-term production and transmission network planning [20], etc.
The problem of accuracy of DC-based power flow results is especially important in so-called
subtransmission networks, where these networks represent the part of power system that practically
connects transmission subsystem (“bulk transmission”) and distribution subsystems. Typical example of a
subtransmission network is a HV network around major cities and networks that are used for large power
consumers (such as large industrial plants, or large office buildings), where the continuity of supply is very
important. Meshed configuration for subtransmission networks is typical in modern power industry
practice. Secondly, typical ratio R/X for overhead 400kV lines is 1/11, while ratio R/X for overhead 110kV
lines is usually higher than 1/2, which implies that neglecting the branch resistances in calculations of
subtransmission networks (mainly based on 110kV, 66kV or similar voltages) is generally not justified.
Different DC-based power flow models are inherently approximate. It is well known that their accuracies
are system and case dependent, following that sometimes these models are not appropriate for industrial
practice (for details, for example, see [3,4,14,21,22]). In [3] is reported that R/X=1/3, 1 and 3, respectively,
and the errors in calculated power flows by the DC-based model can be 11 %, 100 % and 900 %.
There are two contributions of this paper. Firstly, it proposes the generalized linear PTDFs, without
standard simplifications used in DC-based power flow model and their extensions (one of them is that
the branch resistances are included in the formulation). Secondly, the proposed method saves
numerical efficiency of PTDFs, as for ones derived from the standard DC-based power flow model,
using the sparse matrix technique for the characteristic contingency cases (injection changes, branch
outages and their combinations).
The paper is organized as follows: Section 2 represents extended standard DC-based power flow with
the generalized linearized power flow formulation, following their application to the contingency analysis,
for calculation PTDF matrices for injection changes, branch outages and their combination, in Section 3.
Results obtained for small- and large-scale test examples are presented in Section 4, while Section 5
provides conclusions. A derivation of particular sensitivities from the power flow equations, block matrix
inversion, PTDFs for branch outages, cold start model for the base case solution and matrix inversion after
branch outage is given in Appendices A, B, C, D and E, respectively.

2. MODEL FORMULATION

Generalized linearized (GL) node injection model in incremental form can be written as (details of
linearization are given in Appendix A):
2 ΔP 3
 Pθ  
6 U 7 S S PV Δθ
4 5¼  ; (1)
ΔQ S Qθ S QV ΔU
U
where:
n o n o
ΔQ ΔQi
ΔP
U ¼ ΔP Ui
i
, U ¼ Ui  (NPQ +NPV)- and NPQ-dimensional vectors of node active
and reactive power injection increments, divided by voltage magnitude at the same
node, respectively;
SPθ, SPV, SQθ, SQV  [(NPQ +NPV)(NPQ +NPV)]-, [(NPQ +NPV)NPQ]-, [NPQ (NPQ +
NPV)]- and (NPQ NPQ)-dimensional sensitivity sub-matrices with elements determined
from (A1a, b) and (A3a, b);

Copyright © 2011 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power (2011)
DOI: 10.1002/ete
INCLUDING OF BRANCH RESISTANCE IN PTDFS

Δθ, ΔU  (NPQ +NPV)- and NPQ-dimensional vectors of voltage magnitude and angle
increments in PQ+PV and PQ nodes, respectively.
Similarly, the GL-based model for power flows through branches in incremental form can be written as:
2 3
ΔPb  b;Pθ  
6 7
6 U 7 ¼ S b;Qθ S b;QV  Δθ ;
b;PV
4 ΔQb 5 (2)
S S ΔU
U

where:
nΔPb o n b o nΔQb o n b o
ΔP ΔQ
¼ Uiij ¼ Uil , ΔQ
b
ΔPb
U U ¼ Ui
ij
¼ Ui l  L-dimensional vectors of active and
reactive power flow increments through branches, divided by voltage magnitude at the
sending node (for branch ij, the sending node is i), respectively;
Sb,Pθ, Sb,PV, Sb, Qθ, Sb,QV  [L(NPQ +NPV)]-, (LNPQ)-, [L(NPQ +NPV)]- and
(LNPQ)-dimensional sensitivity submatrices with elements determined from (A4a,b)
and (A5a, b).
Note that the sensitivity submatrix SPθ in (1) corresponds to the bus susceptance matrix in DC-based
power flow model (for Cold Start, these submatrices are completely equal—see Section 3.4).

3. CONTINGENCY ANALYSIS

3.1. PTDF matrix for node injection change

We now introduce the PTDF matrix for node injection change, in order to calculate changes in
active and reactive power flows in branches when the node injection changes. Afterwards, replacing
(1) with (2), the perturbed linear matrix equation for active/reactive power flows through branches is:
2 3 2 3
ΔPb  b;Pθ    ΔP
6 U 7 1 6 U 7
6 7 S S b;PV S Pθ S PV
4 ΔQb 5 ¼ S b;Qθ S b;QV  S Qθ S PV 6 7
4 ΔQ 5; (3)

U U
where the PTDF matrix for node injection changes is defined as:
 b;Pθ   Pθ 
S S b;PV Z Z PV
PTDF ¼  ; (4a)
S b;Qθ S b;QV Z Qθ Z QV
where:
   1
Z Pθ Z PV S Pθ S PV
¼ : (5)
Z Qθ Z QV S Qθ S QV
Matrix inversion in (5) is performed using the block matrix approach, described by (B2) and (B3) in
Appendix B.
The PTDF matrix from (4a) can be separated on four PTDF submatrices:
2 ( b;i ) ( b;i ) 3
ΔPl ΔPl
  6 7
a P=P
aP=Q 6 ΔP ΔQi 7
PTDF ¼ Q=P ¼66 (
i
) ( ) 7: (4b)
a aQ=Q b;i
ΔQb;i 7
4 ΔQl l 5
ΔPi ΔQi
Nevertheless, calculation of the complete PTDF matrix is not required in practice. It is sufficient to
calculate some of its members only. It can be performed efficiently using sparse matrix techniques,
illustrated on example of submatrix aP/P in (4b), for sensitivity of active power flow in branch l
(between nodes m and n) to change in active power injection at node i (similar approach can be
applied for other sensitivities):

Copyright © 2011 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power (2011)
DOI: 10.1002/ete
V. D. KRSMAN, A. T. SARIĆ AND N. V. KOVAČKI

 i ΔPb;i b;Pθ Pθ b;Pθ Pθ


aP=P ¼ l
¼ S lm Zmi þ S ln Zni þ S b;PV b;PV PV
lm Zmi þ S ln Zni :
PV
(6)
l ΔPi
In the end, based on (3) and (A5a,b), it is possible to calculate the new active and reactive power
flows in branch l after change of injection at node i, respectively as:
 
b;i;ð1Þ Um  P=P i  i
Pl ¼ Pl þ
b
a ΔPi þ a P=Q
ΔQi ; (7a)
Ui l l
 i  i 
b;ið1Þ Um
Ql ¼ Qbl þ aQ=P ΔPi þ aQ=Q ΔQi ; (7b)
Ui l l

where Um and Ui are voltage magnitudes in nodes m and i, respectively (node m is sending node of
branch l) before injection change.

3.2. PTDF matrix for branch outage


The second characteristic case of PTDFs concerns sensitivity matrix for branch outage, to calculate
how the active and reactive branch flows will change for a single branch outage. This contingency
should be reflected on change of susceptance and conductance matrices in the GL-based power flow
method. If we consider an outage of branch k (between nodes s and t), the appropriate sensitivity
matrix from (1) can be written as:
2   T
3
 Pθ;ð1Þ  ′ Pθ
S  S st e1k e1k
Pθ T
S  f1k S
PV ′ PV
S S PV;ð1Þ ¼ 6
6 
7
7
Qθ;ð1Þ QV;ð1Þ 4  T 5; (8)
S S QV
S QV  f2k S ′

S Qθ  S ′ st e2k eT1k

where from (A4a, b) and (A5a, b) respectively:


Pθ  Qθ 
S′ st ¼ Ut Gbst sinθst þ Bbst cosθst ; S′ st ¼ Ut Gbst cosθst þ Bbst sinθst ; (9)
S′ PV, S′ QV - NPQ-dimensional vectors related with branch k, respectively with elements:

b

′ PV Gst ; i¼s ′ QV Bbst ; i¼s


Si ¼ ; S ¼ ; (10)
Gst cosθst þ Bst sinθst ; i ¼ t
b b i
Gst sinθst  Bst cosθst ; i ¼ t
b b

while vectors in (8) are defined as:


e1k, e2k  (NPQ +NPV)- and NPQ-dimensional vectors related to branch k, respectively with elements:
8 8
< 1; i ¼ s 2 PQ þ PV < 1; i ¼ s 2 PQ
e1i ¼ 1; i ¼ t 2 PQ þ PV ; e2i ¼ 1; i ¼ t 2 PQ ; (11)
: :
0; otherwise 0; otherwise
f1k, f2k  (NPQ +NPV)- and NPQ-dimensional vectors, related to branch k respectively with elements:

1; i ¼ s 2 ðPQ þ PVÞ 1; i ¼ s 2 PQ
f1i ¼ ; f2i ¼ : (12)
0; otherwise 0; otherwise

The active and reactive power flows through branch l (between nodes m and n), after outage of
branch k (between nodes s and t) respectively can be calculated as (particular sensitivities are derived
in Appendix C):
 
b;k;ð1Þ Um  P=P k b  P=Q k b
Pl ¼ Pbl þ d Pk þ d Qk ; (13a)
Us l l

 
b;k;ð1ÞUm  Q=P k b  Q=Q k b
¼ Qbl þ
Ql d Pk þ d Qk ; (13b)
Us l l
 k  k  k  k
where dP=P l , d P=Q l , d Q=P l and d Q=Q l are corresponding PTDFs of active and reactive power
flows through branch l, after change of active and reactive power flows through branch k, defined from
(C7a, b) as:

Copyright © 2011 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power (2011)
DOI: 10.1002/ete
INCLUDING OF BRANCH RESISTANCE IN PTDFS

 k  s;t  P=P k  s;t  Q=P k


dP=P ¼ aP=P a′ þ aP=Q a′ ; (14a)
l l s;t l s;t
 k  s;t  P=Q k  s;t  Q=Q k
d P=Q ¼ aP=P a′ þ aP=Q a′ ; (14b)
l l s;t l s;t
 k  s;t  P=P k  s;t  Q=P k
d Q=P ¼ aQ=P a′ þ aQ=Q a′ ; (14c)
l l s;t l s;t
 k  s;t  P=Q k  s;t  Q=Q k
d Q=Q ¼ aQ=P a′ þ aQ=Q a′ ; (14d)
l l s;t l s;t

where particular sensitivities


 of active/reactive
s;t  s;t flows through
s;t  power s;t branch l to simultaneous injection
changes in nodes s and t aP=P l , aP=Q l , aQ=P l and aQ=Q l are derived in (C3a, b, c, d), while k
′ P=P
particular
   sensitivities
 of active/reactive
  injections in nodes s and t to branch k outage a ,
P=Q k Q=P k Q=Q k s;t
a′ , a′ and a′ are derived in (C5a, b, c, d).
s;t s;t s;t

Matrix inversion of (8) is derived in Appendix E.


In practical implementation of proposed algorithm, several specific outages exist:
1. If the radial branch connecting the part of passive network, or “pure” consumption area (without
generators) is outaged, such branch outage is replaced with the injection change in the sending
node (s) of the outaged branch (k). This injection change is equal to the power flow on branch k
in the Base Case condition (ΔPs ¼ Pbk and ΔQs ¼ Qbk, where branch power flows Pbk and Qbk are
given on the node s branch side). In such case, the methodology for branch outage works as in
the case for node injection change from Section 3.1.
2. If the outaged branch splits the system into two islands, from the original PTDF matrix, based on
nodes and branches in the particular islands, the appropriate rows and columns should be
selected. In this way, the original PTDF matrix is divided into two submatrices which
correspond to the respective islands. Such branch k outage is replaced with the injection changes
in the sending and receiving nodes (s and t) which is equal to the branch k power flow for Base
Case (as described in point 1).

3.3. PTDF matrix for simultaneous node injection change and branch outage
The third characteristic case of PTDFs is matrices for simultaneous node injection change and
branch outage, to calculate how active and reactive branch power flows will change with a single
branch outage, combined with simultaneous injection change.
By the superposition principle, the active and reactive power flows in branch l after simultaneous
injection change at node i and branch k outage are:
 
b;i;k;ð1Þ Um  P=P k b Um  P=P i  P=P k  P=P i
Pl ¼ Pbl þ d Pl þ a þ d a ΔPi
Um l U
 i l l
 k
Um  P=Q k b Um  P=Q i  P=Q k  P=Q i
(15a)
þ d Ql þ a þ d a ΔQi ;
Um l Ui l l k
       i 
b;i;k;ð1Þ Um Q=P k b Um i k
Ql ¼ Qbl þ d Pl þ aQ=P þ d Q=P aQ=P ΔPi
Um l Ui
 l l
 k
Um  Q=Q k b Um  Q=Q i  Q=Q k  Q=Q i
(15b)
þ d Ql þ a þ d a ΔQi :
Um l Ui l l k

3.4. Initial voltages


Initial values of the voltages for the defined linearization in the GL-based PTDFs model can be
obtained as follows [3]:
a) Cold Start Model: In this case, the reliable initial condition is unavailable in advance, so the
model derived in Appendix D can be applied. Typically, Cold Start model with Single or

Copyright © 2011 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power (2011)
DOI: 10.1002/ete
V. D. KRSMAN, A. T. SARIĆ AND N. V. KOVAČKI

Distributed Slack Node can be applied for the Base Case solution (without injection changes or
branch outages).
b) Hot Start Model: In this case, the value of Ui,θi, i=1, 2,⋯, N is known and taken from AC-based
power flow solution. This approach can be applied for contingency analysis, where the AC-based
power flow solution for Base Case (single AC run) is used as initial condition.
Alternative approach can be application of the fast linearized power flow models with contingencies,
such as [24,25].

4. APPLICATION

A procedure for solving the GL-based PTDFs for fast contingency analysis (or screening, if additional
AC power flow analysis is performed for suspicious contingencies) was implemented in C++. The
effectiveness of the proposed approach was evaluated on the example of small- (for educational
purposes) and large-scale (for industry applicability purposes) test systems.

4.1. Small test system


Small test system from [23] consists of 6 nodes, 3 generators, 3 loads, and 11 lines (Figure 1). Values
of line parameters (resistances and reactances), specified active power/voltage values of generators
and active/reactive values of loads are shown in the same figure. All values are given in [p.u.],
normalized to SB =100 MVA and UB =230 kV. Rated power of all branches is Sbn ¼ 100 MVA. To
underline the analyzed effects of PTDFs-based power flow solutions, the line resistances are increased
from the original ones, given in [23].

Base Case power flow. “Base Case” condition presents a system without perturbation. The solution
was calculated using Cold Start (Appendix D) for both GL-based and DC-based PTDFs models, and
the effectiveness of these two power flow models is compared. Compared with the DC-based model,
the GL-based model also provides the possibility for calculation of reactive powers, since in the
traditional DC-based power flow model the reactive powers are neglected (the active powers are equal
to the apparent powers).
For comparison purposes of the proposed PTDFs-based power flow model (GL-based vs. DC-based),
the following global performance indices were used (AC-based solution is used as a reference) [26]:

P = 0.60 3
r = 0.12; x = 0.20
U = 1.07 8 6
~ r = 0.05; x = 0.10
9
r = 0.10 P = 0.70
x = 0.25 4
r = 0.10; x = 0.20 Q = 0.70
P = 0.50 7 11
U = 1.05
~
2 5 r = 0.15
r = 0.15; x = 0.30
r = 0.05 x = 0.30
6
x = 0.10 5
r = 0.20; x = 0.40
10
P = 0.70 P = 0.70
1 Q = 0.70 3 Q = 0.70
r = 0.10; x = 0.20 r = 0.10
4 x = 0.30
2
r = 0.10; x = 0.20 1 Slack node
~ U = 1.05
=0

Figure 1. Small test system with 6 nodes, 3 generators, 3 loads and 11 lines [23].

Copyright © 2011 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power (2011)
DOI: 10.1002/ete
INCLUDING OF BRANCH RESISTANCE IN PTDFS

1. Mean Arithmetic Error for active power flows through branches (MAEP):

L b b;AC
1X P l  P l
MAEP ¼ 100½%: (16a)
L l¼1 Sbnl

2. Mean Square Error for active power flows through branches (MSEP):
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

u L b;AC !2
u1 X Pb  P
MSEP ¼ t l l
100½%; (16b)
L l¼1 Sbnl
where Pbl denotes approximately calculated power flow solution through branch (Pb;DC l or Pb;GL
l ), and
Sbnl denotes rated apparent power of branch l. In Equations (16a, b), power flows through branch (Pb;DC
l
and Pb;GL
l ) are in absolute values because they can differ in sign.
Also, MAE and MSE for reactive and apparent power flows through branches, as well as node
voltage magnitudes and angles, are defined in similar way as in (16a, b).
Calculated node voltage magnitudes and angles are given in Table I, while in Table II are given
active and apparent power flows through branches for Base Case condition.
Only in branches 5 and 8the DC-based PTDFs model provided better results for active power flows
through branches than the GL-based model. To summarize the accuracy of the results, defined
performance indices (16a,b) are presented in Table III.
We can conclude that the GL-based PTDFs model has lower values of all analyzed performance
indices in comparison with the DC-based PTDFs model, due to fewer approximations introduced in
the mathematical model.
Due to limited space, the performance indices of voltage magnitudes and angles for injection
changes and branch outages analyzed in sequel are omitted from the paper.

Injection change. Proposed methodology for node injection change is verified for the outage of
generator in node 3. The PTDFs-based model directly calculates apparent power flows through
branches after perturbation, as a consequence of changed active and reactive power injections. The
summary of results for Cold and Hot Starts are given in Table IV. Hot Start initial condition includes
AC-based power flow results from Base Case.
Considering the presented results, it can be concluded that the GL-based PTDFs model has lower all
used performance indices, comparing with the DC-based model.

Branch outage. Proposed methodology for branch outage is verified for the outage of branch 11. The
summary of calculated results is given in Table V.
On the basis of Tables IV and V, it can be concluded that both DC- and GL-based PTDFs models
provide the lower performance indices for Hot Start initial condition, but the GL-based power flow
model for Hot Start provides better performance indices than the DC-based power flow model,
especially for the calculation of apparent power flows through branches. The reason is that the

Table I. Node voltage magnitudes and angles for small test system (Base Case).
Bus i Ui [kV] θi [degrees]
AC GL AC DC GL
1 (SL) 241.50 241.50 0.00 0.00 0.00
2 (PV) 241.50 241.50 3.95 3.45 3.48
3 (PV) 246.10 246.10 4.95 3.72 3.76
4 (PQ) 225.13 224.94 4.16 5.80 5.20
5 (PQ) 223.76 226.43 5.22 6.75 6.37
6 (PQ) 227.81 227.66 5.81 6.94 6.68

Copyright © 2011 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power (2011)
DOI: 10.1002/ete
V. D. KRSMAN, A. T. SARIĆ AND N. V. KOVAČKI

Table II. Active and apparent power flows through branches for small test system (Base Case).
Branch l Pbl [MW] Sbl [MVA]
AC GL DC AC GL
1 30.902 25.793 24.112 34.457 28.421
2 45.318 42.494 40.516 47.724 43.453
3 36.396 37.368 35.372 39.223 38.302
4 3.874 2.330 1.626 12.168 9.192
5 32.926 33.687 32.808 66.458 33.888
6 16.863 16.235 15.367 24.334 17.263
7 26.191 25.907 24.311 31.412 29.785
8 16.771 17.496 16.777 35.326 17.604
9 47.001 47.890 44.849 77.139 48.595
10 4.114 3.588 3.323 7.802 6.512
11 0.423 0.817 0.839 7.326 4.065

Table III. Summary of results for small test system (Base Case).

Performance index, in [%] DC GL


MAEP 1.937 1.295
MAES 12.180 9.405
MAEV  3.191
MAEθ 20.919 16.309
MSEP 2.801 1.908
MSES 16.340 13.490
MSEV  4.568
MSEθ 24.398 18.502

Table IV. Summary of results for injection change (generation outage in node 3) in small test system.

Performance DC GL
index, in [%]
Cold Start Hot Start Cold Start Hot Start
MAEP 3.065 1.145 2.361 0.802
MAES 14.563 12.665 10.179 8.314
MSEP 4.449 1.772 3.841 1.154
MSES 23.504 21.940 15.328 12.114

standard DC-based model introduces an error in calculation of apparent power flows through branches
(by neglecting reactive power flows).

4.2. Large-scale test system


The second (large-scale) power system presents a merged network model of Continental Europe
Synchronous Area (ex-UCTE) which consists of 7231 nodes, 10 058 lines and 1439 transformers. The
model represents regime of morning peak load for a typical working day in September. Total active/
reactive loads for Base Case are 211981.6MW/69777.9MVAr, while total active/reactive losses are
3655.7MW/15540MVAr. Among all branches in the system, 871 (7.57%) of them have ratio X/R<3.
The summary of the Base Case results for all branches and branches with X/R<3 in large-scale test
system is given in Table VI and Table VII, respectively.
Based on results presented in Tables VI and VII, it can be concluded that the GL-based PTDFs model
has better performance indices in the calculation of active power flows through branches than by the DC-
based model, especially for branches with small X/R ratio (XR). Also, the GL-based PTDFs model

Copyright © 2011 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power (2011)
DOI: 10.1002/ete
INCLUDING OF BRANCH RESISTANCE IN PTDFS

Table V. Summary of results for analysis of branch 11 outage in small test system.

Performance DC GL
index, in [%]
Cold Start Hot Start Cold Start Hot Start
MAEP 2.081 0.161 1.557 0.120
MAES 12.421 10.446 9.636 0.645
MSEP 2.991 0.211 1.987 0.141
MSES 16.462 15.344 14.489 0.922

Table VI. Summary of Base Case results in large-scale power system.

Performance index, in [%] DC GL


MSEP 16.37 8.64
MSES 44.49 22.49
MAEP 18.92 10.80
MAES 48.44 30.45
Calculation time [ms] 94 172

Table VII. Summary of Base Case results for part of large-scale system where X/R<3.

Performance index, in [%] DC GL


MSEP 18.54 8.29
MSES 52.46 21.38
MAEP 20.51 7.02
MAES 54.83 16.93

enables to calculate approximation of reactive/apparent power flows through branches. The important
result presented in Table VI is that the ratio of calculation time for DC- and GL-based models is
approximately 1 : 2. The reason is that the GL-based model requires the solving of the system of linear
equations that has larger dimension than those for the DC-based model (2NPQ +NPV to NPQ +NPV).
The proposed methodology is also verified for the outage of generator with real/reactive generation 295
MW/107.5MVAr. The initial system condition for this outage is calculated using DC-based and Cold and
Hot Starts in GL-based PTDFs models, and compared by respecting AC-based power flow solution as a
reference. Results of MAE and MSE performance indices for all branches in the system are given in
Table VIII.
Based on results presented in Table VIII, it can be concluded that both DC- and GL-based PTDFs
and corresponding power flow models for Hot Start have lower performance indices for active power
flows through branches. The reason is the system dimensionality (system has more than 10000
branches). In large-scale systems, the consequences of outage decline with the increase of distance
from the outage location. Both DC- and GL-based power flow models for calculation of the effects of
the outage based on Hot Start initial condition only superpose the increments on AC-based branch
active power flows in Base Case, which makes results of both of them very close to AC-based
solution. However, the GL-based PTDFs model has additionally better performance indices in
calculation of apparent power flows through branches than the DC-based power flow model. The
reason for this is neglecting the reactive power flows in branches in the DC-based power flow model.
PTDFs calculated from GL-based power flows for Hot Start have decreased all analyzed performance
indices which makes it quite efficient for calculations of the injection change. Also, for branches with
X/R<3, lower performance indices for the GL-based PTDFs power flow solution are obtained
(similarly with results for the Base Case condition in Table VII).

Copyright © 2011 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power (2011)
DOI: 10.1002/ete
V. D. KRSMAN, A. T. SARIĆ AND N. V. KOVAČKI

Table VIII. Summary of results for analysis of injection (generation) change in large-scale power system.

Performance DC GL
index, in [%]
Cold Start Hot Start Cold Start Hot Start
MAEP 15.38 3.131 9.94 1.236
MAES 16.95 11.381 10.71 3.834
MSEP 47.21 3.342 28.91 1.754
MSES 56.12 21.56 30.53 5.983

Similar conclusions were derived for branch outages (due to limited space, these results are omitted
from the paper).

5. CONCLUSION

Standard approach for contingency analysis that is based on the DC-based power flow model is not
appropriate in subtransmission and transmission networks with branches (dominantly lines) that have
low X/R ratio (usually X/R<3), because this model contains the unacceptable approximation of
neglecting the line resistances. The proposed model introduces less assumption in the linearization,
leading to the increased accuracy in calculated power flows through branches, compensated with
reasonable increase of the calculation time.
Generalized linear power flow model and PTDFs derived from them are more appropriate for fast
contingency analysis and security assessment of large-scale (subtransmission and/or transmission)
power systems that include thousands of nodes/branches, critical outages as well as monitored
elements. This model can be applied for different purposes in power system analysis, planning and
operation [4–20]. The proposed model is successfully verified on example of real large-scale power
system that recommends it for industry application.

APPENDIX A: BASIC POWER FLOW EQUATIONS


Standard equations for node active and reactive power injections respectively are:
X 
Pi ¼ Pi ðU; θÞ ¼ Gii Ui2 þ Ui Uj Gij cosθij þ Bij sinθij ; i 2 aPV þ aPQ ; (A1a)
j2ai
X 
Qi ¼ Qi ðU; θÞ ¼ Bii Ui2 þ Vi Uj Gij sinθij  Bij cosθij ; i 2 aPQ ; (A1b)
j2ai

where elements of bus admittance matrix are:


8
< Y i 6¼ j
b

Gij þ jBij ¼
P  ij  
: Y 0
þ Y
 ik ik
b
i ¼ j: (A2)
k2ai

By dividing the active and reactive node injections with voltage at the same bus, and by applying
the linearization for (A1a, b) we obtain:

ΔPi @Pi ðU; θÞ @Pi ðU; θÞ


¼ Δθi þ ΔUi ; (A3a)
Ui @ ðθÞ @ ðU Þ
ΔQi @Qi ðU; θÞ @Qi ðU; θÞ
¼ Δθi þ ΔUi ; (A3b)
Ui @ ðθÞ @ ðU Þ
where particular derivatives are obtained from node injections (A1a, b). Note that standard
simplifications used in the DC-based power flow model used in [27] ð j Gijsinθijj <<j Bijcosθijj ,
jQi j << jBii Ui2 j, sinθij θij, cosθij 1 and Ui =Uj =Const.Þ are not introduced in these derivatives.

Copyright © 2011 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power (2011)
DOI: 10.1002/ete
INCLUDING OF BRANCH RESISTANCE IN PTDFS

The active and reactive power flows through branch l (between nodes i and j), respectively, are:

Pbij ¼ Gbij Ui2  Gbij Ui Uj cosθij  Bbij Ui Uj sinθij ; (A4a)

Qbij ¼ Bbij Ui2 þ Bbij Ui Uj cosθij  Gbij Ui Uj sinθij : (A4b)

By dividing the active and reactive power flows through branches with voltage magnitude at the
branch’s sending node (for branch ij, the sending node is i) and by applying the linearization in
(A4a,b) we obtain:

ΔPbij @Pbij ðU; θÞ @Pbij ðU; θÞ


¼ Δθi þ ΔUi ; (A5a)
Ui @ ðθÞ @ ðU Þ
ΔQbij @Qbij ðU; θÞ @Qbij ðU; θÞ
¼ Δθi þ ΔUi ; (A5b)
Ui @ ðθÞ @ ðU Þ
where particular derivatives are derived from active and reactive power flows through branches
(A4a, b).

APPENDIX B: BLOCK MATRIX INVERSION


Block matrix inversion can be written as [28]:
 1 "  1  1 #
A B A  BD1 C A1 B D  CA1 B
¼  1  1 : (B1)
C D D1 C A  BD1 C D  CA1 B

To avoid inverting the particular submatrices, after reformulation introduced in [27] we can apply
the following matrix equation:
 1    
A B I A1 B A1 0 I 0
¼ ; (B2)
C D 0 I 0 S1
A CA1 I

where SA =DCA1B.
For inversion S1
A , the well-known Sherman–Morrison–Woodbury equation can be used:
 1  1
A þ CDT ¼ A1  A1 C I þ DT A1 C DT A1 : (B3)

APPENDIX C: PTDFs FOR BRANCH OUTAGE


The idea for derivation of PTDFs for branch outage is to replace branch outage with the change of
injections in branch incident nodes (similarly as in [23]). Changes of active and reactive power
injections correspond to active and reactive branch power flows after outage. This approach is
illustrated in Figure C1.

ΔPs = Pkb, s,t ,(1) −ΔPt = Pkb, s,t ,(1)


ΔQs = Qkb, s,t ,(1) branch k −ΔQt = Qkb, s,t ,(1)
b, s,t ,(1)
node s Pk ,Qkb, s,t ,(1) node t

Figure C1. Treatment of branch k outage with injection changes in branch’s incident nodes (s and t).

Copyright © 2011 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power (2011)
DOI: 10.1002/ete
V. D. KRSMAN, A. T. SARIĆ AND N. V. KOVAČKI

Changes of voltage angles and magnitudes in incident nodes (s and t) of by outage affected branch
(k), after substituting ΔPt/Ut ΔPs/Us and ΔQt/Ut ΔQs/Us, can be written as:
 ΔPs  PV ΔQs
Δθs ¼ ZssPθ  ZstPθ þ Zss  ZstPV ; (C1a)
Us Us

 ΔPs  PV ΔQs
Δθt ¼ ZtsPθ  ZttPθ þ Zts  ZttPV ; (C1b)
Us Us

 ΔPs  QV ΔQs
ΔUs ¼ ZssQθ  ZstQθ þ Zss  ZstQV ; (C1c)
Us Us

 ΔPs  QV ΔQs
ΔUt ¼ ZtsQθ  ZttQθ þ Zts  ZttQV : (C1d)
Us Us
Introducing Δθs, Δθt, ΔUs and ΔUt from (C1a,b,c,d), for branch k (between nodes s and t) in
outage (7a,b) can be written as follows:

b;s;t;ð1Þ Us h P=P s;t  s;t i


Pk ¼ Pbk þ a ΔPs þ aP=Q ΔQs ; (C2a)
Us k k

b;s;t;ð1Þ Us h Q=P s;t  s;t i


Qk ¼ Qbk þ a ΔPs þ aQ=Q ΔQs ; (C2b)
Us k k

where particular sensitivities are:


 s;t
b;Pθ  Pθ   Qθ  Qθ
aP=P ¼ Sks Zss  ZstPθ þ Sktb;Pθ ZtsPθ  ZttPθ þ Sb;PV
ks Zss  ZstQθ þ Sb;PV
kt Zts  ZttQθ ; (C3a)
k
 s;t
b;Pθ  PV  PV  QV  QV
aP=Q
¼ Sks Zss  ZstPV þ Sb;Pθkt Zts  ZttPV þ Sb;PV
ks Zss  ZstQV þ Sb;PVkt Zts  ZttQV ; (C3b)
k
 s;t  Pθ  Pθ  Qθ  Qθ
aQ=P ¼ Sb;Qθ
ks Zss  ZstPθ þ Sb;Qθ
kt Zts  ZttPθ þ Sb;QV
ks Zss  ZstQθ þ Sb;QV
kt Zts  ZttQθ ; (C3c)
k
 s;t  PV  PV  QV  QV
aQ=Q ¼ Sb;Qθ
ks Zss  ZstPV þ Sb;Qθkt Zts  ZttPV þ Sb;QVks Zss  ZstQV þ Sb;QV
kt Zts  ZttQV :
k
(C3d)
b;s;t;ð1Þ b;s;t;ð1Þ
The next step is introducing of Pk ¼ ΔPs and Qk ¼ ΔQs from Figure C1 to (C2a, b). The
solution for the obtained system of linear equations is:
 P=P k  P=Q k
ΔPs ¼ a′ Pbk þ a′ Qbk ; (C4a)
s;t s;t
 Q=P k  k
ΔQs ¼ a′ Pbk þ aQ=Q Qbk ; (C4b)
s;t s;t

where:
 k  s;t
′ P=P
1  aQ=Q k
a ¼ s;t  s;t  s;t s;t ; (C5a)
s;t 1  ðaQ=Q Þk 1  ðaP=P Þk  ðaP=Q Þk ðaQ=P Þk
 P=Q k  P=Q s;t

a
a ¼ s;t  s;t 
k
s;t s;t ; (C5b)
s;t 1  ða Þk 1  ðaP=P Þk  ðaP=Q Þk ðaQ=P Þk
Q=Q

 Q=P k  Q=P s;t



a
a ¼ s;t  s;t 
k
s;t s;t ; (C5c)
s;t 1  ða Þk 1  ðaP=P Þk  ðaP=Q Þk ðaQ=P Þk
Q=Q

 Q=Q k  s;t

1  aP=P k
a ¼ s;t  s;t  s;t s;t : (C5d)
s;t 1  ðaQ=Q Þk 1  ðaP=P Þk  ðaP=Q Þk ðaQ=P Þk

Copyright © 2011 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power (2011)
DOI: 10.1002/ete
INCLUDING OF BRANCH RESISTANCE IN PTDFS

Now, the perturbation of active and reactive power flows over branch l (between nodes m and n),
after outage of branch k (between nodes s and t), or node injection changes that correspond to branch k
flow after outage are:

b;s;t;ð1Þ Um h P=P s;t  s;t i


Pl ¼ Pbl þ a ΔPs þ aP=Q ΔQs ; (C6a)
Us l l

b;s;t;ð1Þ Um h Q=P s;t  s;t i


Ql ¼ Qbl þ a ΔPs þ aQ=Q ΔQs ; (C6b)
Us l l

 s;t  s;t  s;t  s;t


where sensitivity factors aP=P l , aP=Q l , aQ=P l and aQ=Q l are defined in similar way as in
(C3a, b, c, d).
After introducing (C4a, b) into (C6a, b), we have the following relations for active and reactive
power flows through branch l as:
 s;t  Q=P k 
b;k;ð1Þ Um  P=P s;t  ′ P=P k 
Pl ¼ Pbl þ a a þ aP=Q a′ Pbk
Us l s;t l s;t
 s;t  Q=Q k 
Um  P=P s;t  ′ P=Q k  (C7a)
þ a a þ aP=Q a′ Qbk ;
Us l s;t l s;t

 s;t  Q=P k 
b;k;ð1Þ Um  Q=P s;t  ′ P=P k 
Ql ¼ þ
Qbl a a þ a Q=Q
a′ Pbk
Us l s;t l s;t
 s;t  Q=Q k  (C7b)
Um  Q=P s;t  ′ P=Q k 
þ a a þ a Q=Q
a′ Qbk :
Us l s;t l s;t

APPENDIX D: COLD START MODEL FOR BASE CASE SOLUTION


For the Cold Start model, we have the following approximations in node injections (A1a,b): cosθij 1,
sinθij 0, and Uj 1, following that the power flow model is completely linear [not only incrementally
linear as in (1)]:
2P 3
 Pθ  
6U 7 S SPV θ
4 5¼ QV  ; (D1)
Q S Qθ
S U
U
where:
P Q
U , U  (NPQ +NPV)- and NPQ-dimensional vectors of node active and reactive power injection
increments, with elements obtained from (1) and (A1a,b) respectively as:
8 sp
> P X
>
> i  Uj YijPV ; i 2 aPQ
>
P < Ui j2ai;PV
¼ X ; (D2a)
U >> Psp
>
>
i
 U Y PV
 U Y PV
; i 2 a
: Ui i ii j ij PV
j2ai;PV
(
Q Qsp X
¼ i
 Uj YijQV ; i 2 aPQ ; (D2b)
U Ui j2ai;PV

θ, U  (NPQ +NPV)- and NPQ-dimensional vectors of voltage angles in PV+PQ and voltage
magnitudes in PQ nodes (obtained as a solution of the GL-based power flow model).
Base Case branch power flows are calculated by introducing the system state vector (θ and U) from
(D1) into (A4a, b) and by dividing by voltage magnitude in branch’s sending node [as in (A5a, b)].

Copyright © 2011 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power (2011)
DOI: 10.1002/ete
V. D. KRSMAN, A. T. SARIĆ AND N. V. KOVAČKI

APPENDIX E: MATRIX INVERSION AFTER BRANCH OUTAGE


Matrix inversion in (8) is performed using the combination of block matrix inversion (B2) and
Sherman–Morrison–Woodbury equation (B3). Considering (B3), it can be concluded that for every
branch outage, two inversions and multuplication of matrices are required.
The first matrix that should be inverted in (B3) is:
 Pθ
1
A1 ¼ SPθ  S′ st e1k eT1k ; (E1)
while the second matrix which should be inverted in (B3) is:
"   #
 1 T T 1  
T 1 ′ QV ′ Pθ ′ Qθ
IþD A C ¼ I þ S  f 2k S
QV
S  S st e1k e1k
Pθ T
S  S st e2k e1k :
Qθ T
(E2)

These inversions are performed using sparse matrix techniques, in similar way as described in [23]
for DC-based power flow.

LIST OF SYMBOLS AND ABBREVIATIONS

Note that vectors and matrices are denoted in bold letters.


A. Indices:
AC, DC, GL Full nonlinear (AC), linear active (DC) and generalized linear (GL) based
power flow models (upper index)
b Branch (upper index)
i, j, s, t, m, n Current indices for nodes (upper index denotes element in outage)
l, k Current lower indices for branches (branch l between nodes m and n; branch
k between nodes s and t) (upper index denotes element in outage)
PV (PQ) PV (PQ) type of nodes (upper index)
sp Specified value
T Matrix transposes
0 Shunt elements (upper index)
(1) Condition after injection change and branch outage (or their combination)
(upper index)

B. Variables:
P, Q, S Active, reactive and apparent powers
S Sensitivity
U (θ) Magnitude (angle, θij =θi θj) of node voltage phasor

C. Parameters:
G, B Conductance and susceptance
N (L) Number of nodes (branches)
ai Set of nodes incident to node i
aPV (aPQ) Set of PV (PQ) nodes
ai,PV (ai,PQ) Set of PV (PQ) nodes incident to node i

D. Abbreviations:
DC Direct Current
AC Alternating Current
GL Generalized Linearized
PTDF Power Transmission Distribution Factors
MAE Mean Arithmetic Error
MSE Mean Square Error
HV High Voltage

Copyright © 2011 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power (2011)
DOI: 10.1002/ete
INCLUDING OF BRANCH RESISTANCE IN PTDFS

REFERENCES
1. Stott B. Review of load flow calculation methods. Proceedings of the IEEE 1974; 62: 916–929. DOI: 10.1109/
PROC. 1974.9544
2. Sauer P. On the formulation of power distribution factors for linear load flow methods. IEEE Transactions on
Power Apparatus and Systems 1981; PAS-100: 764–779. DOI: 10.1109/TPAS.1981.316928
3. Stott B, Jardim J, Alsaç O. DC power flow revisited. IEEE Transactions on Power Systems 2009; 24: 1290–1300.
DOI: 10.1109/TPWRS.2009.2021235
4. Purchala K, Meeus L, Van Dommelen D, Belmans R. Usefulness of DC load flow for active load flow analysis.
Proceedings of the IEEE PES General Meeting 2005.
5. Cheng X, Overbye TJ. PTDF-based power system equivalents. IEEE Transactions on Power Systems 2005; 20:
1868–1876. DOI: 10.1109/TPWRS.2005.857013
6. Li F, Bo R. DCOPF-based LMP simulation: Algorithm, comparison with ACOPF, and sensitivity. IEEE
Transactions on Power Systems 2007; 22: 1475–1485. DOI: 10.1109/TPWRS.2007.907924
7. Hur D, Park J-K, Kim BH. An efficient methodology for security assessment of power systems based on distributed
optimal power flow. European Transactions on Electrical Power 2003; 13: 161–165. DOI:10.1002/etep.4450130304
8. Lim J, Jiang JN. Bibliography review on applications of correlation analysis in power system operation and
planning. European Transactions on Electrical Power 2010; 20: 114–122. DOI: 10.1002/etep.294
9. Donde V, Lopez V, Lesieutre B, Pinar A, Chao Y, Meza J. Severe multiple contingency screening in electric power
systems. IEEE Transactions on Power Systems 2008; 23: 406–417. DOI: 10.1109/TPWRS.2008.919243
10. Brandwajn V. Efficient bounding method for linear contingency analysis. IEEE Transactions on Power Systems
1988; 3: 38–43. DOI: 10.1109/59.43179
11. Taylor DG, Maahs LJ. A reactive contingency analysis algorithm using MW and MVAr distribution factors. IEEE
Transactions on Power Systems 1991; 6: 349–355. DOI: 10.1109/59.131082
12. Santos JR, Gomez Exposito A, Martinez Ramos JL. Distributed contingency analysis: practical issues. IEEE
Transactions on Power Systems 1999; 14: 1349–1354. DOI: 10.1109/59.801895
13. Carpentier JL, Di Bono PJ, Tournebise PJ. Improved efficient bounding method for DC contingency analysis using
reciprocity properties. IEEE Transactions on Power Systems 1994; 9: 76–84. DOI: 10.1109/59.317555
14. Sarić AT, Stanković AM. Model uncertainty in security assessment of power systems. IEEE Transactions on Power
Systems 2005; 20: 1398–1405. DOI: 10.1109/TPWRS.2005.851922
15. Grijalva S, Sauer PW, Weber JD. Enhancement of linear ATC calculations by the incorporation of reactive load
flows. IEEE Transactions on Power Systems 2003; 18: 619–624. DOI: 10.1109/TPWRS.2003.810902
16. Pinto H, Magnago F, Brignone S, Alsaç O, Stott B. Security constrained unit commitment: Network modeling and
solution issues. Proceedings of IEEE Power Systems Conference & Exposition. Atlanta: GA, USA, 2006; 1759–1766.
DOI: 10.1109/PSCE.2006.296179
17. Liu M, Gross G. Effectiveness of the distribution factor approximations used in congestion modeling. Proceedings
of 14th Power Systems Computation Conference, Seville, Spain, 2002.
18. Sarkar V, Kharparde SA. A comprehensive assessment of the evolution of financial transmission rights. IEEE
Trans. on Power Systems 2008; 23: 1783–1795. DOI: 10.1109/TPWRS.2008.2002182
19. Hao S, Shirmohammadi D. Transmission reservation and transactions modeling in electricity markets. European
Transactions on Electrical Power 2009; 19: 612–622. DOI: 10.1002/etep.312
20. Sadegheih A. New formulation and analysis of the system planning expansion model. European Transactions on
Electrical Power 2009; 19: 240–257. DOI: 10.1002/etep.210
21. Baldick R. Variation of distribution factors with loading. IEEE Transactions on Power Systems 2003; 18: 1316–1323.
DOI: 10.1109/TPWRS.2003.818723
22. Duthaler C, Emery M, Andersson G, Kurdizem M. Analysis of the use of PTDF in the UCTE transmission grid.
Proceedings of 16th Power Systems Computation Conference, Glasgow, U.K., 2008.
23. Wood AJ, Wollenberg BF. Power Generation, Operation and Control, 2nd ed. Wiley: New York, 1996.
24. Mamandur KRC, Berg GJ. Efficient simulation of line and transformer outages in power systems. IEEE
Transactions on Power Apparatus and Systems 1982; PAS-101: 3733–3741. DOI: 10.1109/TPAS.1982.317058
25. Ruiz PA, Sauer PW. Voltage and reactive power estimation for contingency analysis using sensitivities. IEEE
Transactions on Power Systems 2007; 22: 639–647. DOI: 10.1109/TPWRS.2007.894868
26. Van Hertem D, Verboomen J, Purchala K, Belmans R, Kling W. Usefulness of DC power flow for active power
flow analysis with flow controlling devices. The 8th IEE International Conference on AC and DC Power
Transmission, London, UK, 2006.
27. Stott B, Alsac O. Fast decoupled load flow. IEEE Transactions on Power Apparatus and Systems 1974; PAS-93:
859–867. DOI: 10.1109/TPAS.1974.293985
28. Watt SM. Pivot-free block matrix inversion. Eight International Symposium on Symbolic and Numeric Algorithms
for Science Computing  SYNASC 2006; 151–155. DOI: 10.1109/SYNASC.2006.61

Copyright © 2011 John Wiley & Sons, Ltd. Euro. Trans. Electr. Power (2011)
DOI: 10.1002/ete

You might also like