You are on page 1of 61

Page1

0ptimal Powei Flow Papei S



Staff papei by
Richaiu P. 0Neill
Anya Castillo
Naiy B. Cain
Becembei 2u12
Theviewspresentedarethepersonalviews oftheauthors
andnottheFederalEnergyRegulatoryCommissionorany
ofitsCommissioners.








Page2


The Computational Testing of AC 0ptimal Powei Flow
0sing the Cuiient voltage Foimulations
0ptimal Powei Flow Papei S
Richaiu P 0Neill, Anya Castillo anu Naiy Cain
iichaiu.oneillfeic.gov; anya.castillofeic.gov; maiy.cainfeic.gov
Becembei 2u12


Abstiact anu Executive Summaiy
Solving the iteiative lineai appioximation of the cuiient voltage (Iv) AC0PF is
compaieu to solving the ILIvAC0PF with seveial nonlineai solveis. In geneial, the
lineai appioximation appioach is moie iobust anu fastei than seveial of the
commeicial nonlineai solveis. Foi the nonlineai solveis, the uefault paiameteis aie
useu. The accuiacy of the nonlineai solveis is highei when the optimal solution is
founu. 0n seveial staiting points, the nonlineai solveis faileu to conveige oi
containeu positive ielaxation vaiiables above the thiesholu value. We finu that the
flat stait is supeiioi to the aveiage of the ianuom staits. The iteiative lineai
piogiam appioximation is moie iobust anu consistent acioss staiting points. The
iteiative lineai piogiam appioach finus a neai feasible neai optimal in almost all
pioblems anu all staiting points. 0sing the lineai appioximation appioach opens up
the possibility of solving the AC optimal topology anu AC optimal unit commitment
pioblems fastei as a mixeu integei lineai piogiam (NIP). The iankeu peifoimance
of the nonlineai solveis changes uepenuing on the test pioblem.


Bisclaimei: The views piesenteu aie the peisonal views of the authois anu not the
Feueial Eneigy Regulatoiy Commission oi any of its Commissioneis.

Page3


Table of Contents

1. Intiouuction 4
2. Test Besign S
S. Results foi Each Test Pioblem 9
4. Summaiy of Results anu Conclusions 12
Refeiences 1S
Appenuix 14


Page4
1. Intiouuction
Aftei hunuieus of papeis ovei the last half centuiy on foimulating anu
solving the AC optimal powei flow (AC0PF) anu many attempts to solve the AC0PF
commeicially, in piactice, appioximate BC mouels with suiiogate AC constiaints
aie still wiuely useu. Cain et al (2u12)piesenteu a liteiatuie ieview of the histoiy
anu vaiious foimulations of the AC0PF. The AC0PF is a continuousvaiiable
nonconvex piogiam (see Lesieutie anu Biskens, 2uuS; Biskens, 2uu1). Foi
nonconvex pioblems, most solveis teiminate at a local optimal solution oi fail to
conveige. Local nonglobal optimal solutions coulu cost the electiic powei inuustiy
billions of uollais annually because of uniealizeu gains fiom suboptimal solutions.
We often take it on faith that the local solution obtaineu is also global oi at least a
goou one. In this papei, we stuuy small to meuiumsize pioblems to bettei
unueistanu the natuie anu stiuctuie of the pioblem.
Seveial small pioblems aie wellstuuieu oi at least welltesteu. They have
helpeu uemonstiate that theie aie high anu low voltage solutions to the powei flow
equations on small examples. Bowevei, little is known about the high anu low
voltage solutions to the AC0PF with moie than seveial buses. The hypothesis is that
the low voltage solution is moie expensive oi is constiaineu to be infeasible.
Although acauemic papeis usually auuiess optimal solutions, in piactice, we settle
foi neaioptimal solutions that can be obtaineu in a specifieu time winuow. Bue to
nonconvexities, theie is no guaiantee that the softwaie will finu an optimal solution.
Robustness is the ability to finu an optimal solution oi at least a feasible solution.
Castillo et al. piesent a liteiatuie ieview of computational stuuies anu test pioblem
peifoimance (Castillo 2u12).
Touay, most papeis offeiing algoiithms, that solve the AC0PF fastei anu
moie iobustly, test these algoiithms on the stanuaiu IEEENATP0WER test set
anuoi some piopiietaiy pioblems not geneially available. Schectei et al (2u12)
piesent an analysis of the stanuaiu test pioblems. Even though they appeai often in
the liteiatuie, we know ielatively little about these pioblems oi how iepiesentative
they aie of actual electiicity netwoiks. Foi example, aie these test sets easiei oi
haiuei to solve ielative to the bioauei univeise of actual AC0PFs. Test iesults on
piopiietaiy pioblems iequiie gieatei tiust in the iesults since they aie seluom if
evei iepiouuceu by inuepenuent paities. Since we have no piopiietaiy pioblems, in
this papei, we test using geneially available test pioblems.
To compaie nonlineai solveis, one can consiuei seveial uimensions
incluuing the quality of the optimal solution, time to solution, anu solution
iobustness, that is, the numbei of instances a goou feasible solution is founu. In this
papei, we test the peifoimance of lineai appioximations of AC0PFIv foimulation
against the nonlineai AC0PFIv; both foimulations aie piesenteu in a companion
papei (0Neill et al, 2u12).

Page5
Finally, we note that all mouels heie appioximate a physical system. A
steauystate poweiflow mouel appioximates a laigei uynamic mouel. Foi example,
powei flow mouels geneially uo not mouel fiequency excuisions anu any uynamic
constiaints oi contingencies fiom offline stuuies, which aie achieveu by aujusting
static constiaints. Noieovei, a featuie of the AC0PF is that many of the pioblem
constiaints aie soft, that is, aie not absolute. Theimal constiaints oi constiaints that
often contain significant engineeiing safety maigins, such as the voltage stability
constiaints, aie a couple examples of soft inequality constiaints. As it is common
accepteu piactice, penalty slack vaiiables aie assigneu to these inequalities. Such
ielaxation slack vaiiables aie given high objective function costs since exact costs
anu bounus aie haiu to quantify.


2. Test Besign
Appioach. We begin by geneiating a staiting point. To test the lineai foimulation,
we solve the nonlineai Iv foimulation with quauiatic costs (0PF2) to benchmaik
the solvei against the geneially accepteu optimal solution. Next, we solve the
nonlineai Iv foimulation with a lineai tenstep objection function (0PF)
appioximating the quauiatic costs to benchmaik the lineai objective function anu
examine how the softwaie peifoims with a lineai objective function. We then solve
a seiies of lineaiizeu Iv foimulations (0PFlin). At each step, employing the pievious
optimal solution as the new staiting point foi lineaiization, we ielineaiize the fiist
oiuei appioximation anu auu new constiaints that fuithei impiove oi bounu the
appioximations. When any one of thiee ciiteiia (explaineu below) is satisfieu, the
algoiithm stops. Finally, we make an auuitional iun thiough the nonlineai Iv
foimulation with the lineai objective function staiting with the optimal solution to
the last iteiation of the lineai Iv foimulation (0PFp). The uANS coue to test the
solveis is unchangeu except foi the names of the solveis.
Staiting Points. We solve each test pioblem with ten uiffeient staiting points. The
fiist staiting point is a flat stait (in polai cooiuinates, voltage magnituue = 1 anu
voltage angle = u) foi all buses. The iemaining nine points aie unifoimly ianuom
points inteiioi to the voltage magnituue constiaints. The staiting value foi cuiient
is I = Yv wheie v is the geneiateu point. This I, v point satisfies the cuiient flow
equations but not necessaiily the geneiatoi anu loau constiaints. This appioach
helps test the iobustness of the solvei. In piactice, choosing a goou staiting point
baseu on expeiience is most likely a bettei stiategy, but cannot be easily testeu. But
in a contingency, choosing a staiting point using the last available solution may not
be a goou staiting point.
Test Pioblems. We uefine small test pioblems (foi example, 14bus, Subus, S7bus)
as pioblems that incluue between 1u to 99 buses. They aie helpful because they

Page6
solve fast anu allow foi a bettei unueistanuing of what is happening in the solution
piocess at the noual anu bianch level. We uefine meuium size pioblems (foi
example, 118bus, Suubus) as pioblems with 1uu to 999 buses. These pioblems aie
uifficult to analyze at the noual anu bianch level, but solve ielatively quickly. Nost
veitically integiateu utility tiansmission netwoiks fit in this categoiy. Laige
pioblems aie between 1,uuu anu 9,999 buses anu most laige system opeiatois fit in
this categoiy. We uefine veiy laige pioblems as 1u,uuu oi moie buses; laige IS0s
like P}N anu NIS0 fit in this categoiy.
We solve each of the five now stanuaiu IEEENATP0WER test pioblems (14
bus, Subus, S7bus, 118bus, anu Suubus). Each test pioblem contains the same
voltage magnituue constiaints at each bus anu contains no constiaints on the
tiansmission element flows oi voltage angles. We scale each objective function.
Summaiy paiameteis, the wiuely accepteu quauiatic optimal value, anu the step
function appioximation of the objective function foi each test pioblems aie in Table
1.

Table 1. Test pioblem paiameteis anu optimal solutions
Noues Bianches ueneiatois uemanu v
max
v
min
Best Known
0ptimal value Numbei Capacity
Step
Function
Quauiatic
14 2u S 7.724 2.S9u 1.u6 u.94 82.76 8u.81
Su 41 6 S26.8u 42.42 1.1u u.9S S.918 S.74S
S7 8u 7 S26.78 2SS.26 1.u6 u.94 42S.7 417.4
118 186 S4 99.66 42.42 1.u6 u.94 1S11 1297
Suu 411 69 19.76 12.S1 1.u6 u.94 7488 7197

Auuitional infoimation on the test pioblems anu solutions is available at the
NATP0WER site (http:www.pseic.coinell.euumatpowei). Foi the quauiatic
foimulation, the local optimal solutions aie pioveu to be global by solving a
semiuefinite piogiam (SBP) uual foimulation anu finuing a solution without a
uuality gap (see Lavaei anu Low). Foi the lineai solveis, the BTBETA foimulation is
solveu anu piesenteu foi contiast. Since we test moie nonlineai solveis than lineai
solveis, the lineai solveis aie useu moie than once in the testing pioceuuie.
Baiuwaie. The pioblems weie solveu on a BELL Latituue laptop with an Intel
CoieTN iSS4uN (2.SSuBz, SN cache) anu 4uB memoiy. Ninoi uiffeiences in
solution times weie iecoiueu when the pioblems weie iun at uiffeient times of uay
anu with anu without netwoik connection but the uiffeiences weie small enough to
be consiueieu negligible.

Page7
Solveis. The nonlineai piogiams useu solveis NIN0S S.S1, SN0PT 7.24, IP0PT
veision S.8, C0N0PT veision S.14v anu KNITR0 veision 7.u.u. Thiee of the solveis
have oveilapping authoisuevelopeis see table 2.
Table 2. Solveis anu uevelopeis
C0N0PT SN0PT NIN0S IP0PT KNITR0
Aine Biuu X X
Waltei Nuiiay X X
Philip uill X X
Biuce Nuitagh X
Nichael Saunueis X
Cail Laiiu X
Anuieas Wchtei X
R.B. Byiu X
}. Noceual X
R.A. Waltz X
Lineai piogiams useu u0R0BI veision 4.u.u anu CPLEX veision 12.2.u.1. Theii
authoisuevelopeis also oveilap since Robeit Bixby, Zonghao uu anu Euwaiu
Rothbeig weie piincipal uevelopeis on both.
We useu the uefault settings foi all solveis. In touays maiket, a geneiic
solvei is valueu, in pait, by its ability to inteinally choose the best stiategy foi
solving a pioblem with uefault settings while inteinally choosing solution stiategies
uynamically. Asking the usei to set numeious paiameteis is vieweu as a weakness,
because it iequiies a level of expeitise that shoulu not be a pieiequisite foi using
the solvei. Neveitheless, expeit tuning is impoitant in commeicial installations
wheie the pioblem is solveu iepetitively. In geneial, a nonlineai solvei spenus most
of its time to get close to the optimal solution. Foi example, the ieuuceu giauient
geneially goes fiom 1u
2
to 1u
8
(oi less) in less than 1u% of the iteiations.
Pioblem Foimulations. The pioblems weie foimulateu in uANS veision 2S.6
staiting with coue wiitten by Nichael Feiiis anu mouifieu by Richaiu 0Neill. Foi all
foimulations, ielaxation vaiiables with high penalties, 1u
7
, aie auueu to the uppei
anu lowei bounus foi P, Q anu v to piouuce feasible solutions, albeit at times with
penalties active in the objective function. In actual uispatch maikets, penalty slack
vaiiables aie inseiteu foi vaiious ieasons, incluuing soft constiaints. 0ften neai
feasible solutions aie usable. Even though most commeicial applications use penalty
paiameteis, the lineai Iv foimulation piouuceu bettei iesults than the nonlineai
solveis.
As a quauiatic appioximation to c(S), we use
C
2
(PP) + C
1
P + a|C
2
(Q
+
Q
+
) + C
2
(Q

) + C
1
(Q
+
+Q

)].
ueneially, u < a < .1. Foi a lineai appioximation, a step function appioximation can
be useu. Beie a is set to u.1.

Page8
Peifoimance Neasuies. Foi powei system uispatch, iobustness is impoitant to
opeiations. An impoitant metiic of powei systems opeiation softwaie is a goou
feasible solution within a specifieu time winuow. Foi the test pioblems, we can
bieak this into two uimensions: CP0 time to conveige, that is, the fiaction of the
time that a goou usable local optimal solution was founu, anu iobustness, that is,
the ability to finu a goou usable solution.
The lineai piogiam is a fuithei appioximation of the nonlineai piogiam.
Theiefoie, we uevelop a nonlineai feasibility test to measuie the 1noim uistance
of the lineai piogiam optimal solution vaiiables fiom a nonlineai feasible solution.
Let v
i
*, v
j
*, I
i
*, I
j
*, P*, Q* be an optimal solution to the lineai piogiam. By the
foimulation, v
i
*, v
j
*, I
i
*, I
j
* is a feasible solution to the flow equations. Let
v
m
= (v
i
*v
i
*

+ v
j
*v
j
*)
12

P = v
i
*I
i
* + v
j
*I
j
*
Q = v
j
*I
i
* v
i
*I
j
*
P anu Q aie the implieu injections fiom the flow equations anu v
m
is the voltage
magnituue. We calculate feasible noimalizeu ueviation metiics foi v, P, anu Q as
follows. Foi voltage magnituue feasibility excuisions, foi each bus, n, set v
f
n
= u.
if v
m
n
> v
max
n
, v
f
n
= v
m
n
v
max
n
.
if v
m
n
< v
min
n
, v
f
n
= v
m
n
v
min
n
.
Theiefoie if v
f
n
= u, the voltage is within its magnituue bounus. if v
f
n
> u, the voltage
exceeus its maximum level. If v
f
n
< u, the voltage is less than its minimum level.
We now sum the absolute value of the ueviations anu noimalize by the numbei of
buses, obtaining the total infeasibility voltage metiic:
v
u
=
n
abs(v
f
n
)n.
The maximum infeasibility voltage metiic is:
v
umax
= max
n
{abs(v
f
n
)].
Foi noual ieal powei imbalances, feasibility excuisions, foi each bus, n,
set p
f
n
= u.
If p
n
> p
max
n
, p
f
n
= abs(p
n
p
max
n
).
If p
n
= p
max
n
anu p
max
n
= p
min
n
, p
f
n
= abs(p
n
p
flow
n
)
wheie p
flow
n
=
k
abs(p
flow
kn
)2.
If p
n
= p
max
n
anu p
max
n
= p
min
n
, p
f
n
= u.
If p
n
< p
min
n
, p
f
n
= abs(p
n
p
min
n
).
If p
f
n
= u, the ieal powei is within its magnituue bounus. If p
f
n
> u, the ieal powei
exceeus its maximum level oi p
n
= p
max
n
anu p
max
n
= p
min
n
. If p
f
n
< u, the ieal powei
is less than its minimum level.
We now sum the absolute value of the ueviations anu uiviue by the numbei of buses,
obtaining the total infeasibility ieal powei metiic:
p
u
=
n
abs(p
f
n
)n.
The total maximum infeasibility ieal powei metiic is:

Page9
p
umax
= max
n
{abs(p
f
n
)].
We uefine q
f
n
, q
u
, anu q
umax
in the same way.
v
u
, P
u
anu Q
u
measuie the total infeasibility implieu by the nonlineai equations.
v
umax
, P
umax
anu Q
umax
measuie the maximum infeasibility implieu by the nonlineai
equations.
We test solveis with anu without ielaxation slack vaiiables. Relaxation slack
vaiiables aie iemoveu by setting the uppei bounu on the slack vaiiables to zeio. We
also auu an auuitional noim foi active ielaxation slack vaiiables:
Tepsinoim =
n
(p
minepsi
n
+ p
maxepsi
n
)p
usum
+
n
(q
minepsi
n
+sq
maxepsi
n
)q
usum
+

n
(v
minepsi
+ v
maxepsi
n
)
If Tepsinoim is less than epsilon, we consiuei the solution acceptable anu
iemove the penalty values fiom the objective function. If not, it is iecoiueu as a
feasibility violation, that is, TEPSIN = TEPSIN +1.
Paiametei Setting. The ielaxation penalty was set at 1u
7
. NaxLin is the maximum
numbei of calls to the lineai solvei. It is set at 1u. feasum is the sum of noimalizeu
infeasibilities. It is set to u.uS. maxfeas is the maximum noimalizeu infeasibilty. It is
set to u.u1. The LP teiminates when any of the thiee conuitions aie met. Tepsi is the
infeasibility test baseu on the magnituue of the penalty slacks.


S. Results foi Each Test Pioblem
We stait by piesenting summaiy iesults foi each test pioblem. Noie uetaileu
iesults aie in the appenuix
14bus pioblem. Without ielaxation slack vaiiables, all solveis with the exception of
KNITR0 founu a neai global optimal solution foi each staiting point. 0n aveiage,
SN0PT, C0N0PT anu NIN0S weie the fastest solveis. u0R0BI was competitive with
the nonlineai solveis on the 0PF. Compaiing 0PF anu 0PFp, using the LP
appioximation final solution as the staiting point uiu not lowei cpu time. The flat
stait uoes bettei than the aveiage ianuom staiting point.
With ielaxation slack vaiiables, all solveis ueclaie a local optimal solution foi
each staiting point, but only C0N0PT anu IP0PT finu the global optimal value. The
othei solveis finu local optimal solutions neai the global optimal value. 0n aveiage
NIN0S is the fastest solvei followeu by C0N0PT. SN0PT is eiiatic, at times, woiking
well anu at times, appeaiing to get hung up. 0veiall, since C0N0PT is the seconu
fastest anu always finus a global optimum, C0N0PT outpeifoims the othei
nonlineai solveis.
Foi the lineai solveis, u0R0BI outpeifoims CPLEX in cpu time. The optimal
solution values aie within 1 peicent of the best nonlineai solveis, conveiging in 4 to
7 lineai piogiam passes. u0R0BI anu CPLEX aie fastei with ielaxation slack
vaiiables.

Page
10
We compaie a C0N0PT 0PF solution to a u0R0BI 0PFlin lineai
appioximation solution. The 0PF solution anu 0PFlin solution have only minoi
uiffeiences in noual piices foi ieal anu ieactive powei. The objective function values
uiffei by less than one peicent. The voltage magnituues anu uual vaiiables fiom the
bus equations aie similai, but the actual geneiatoi uispatch (ieal anu ieactive)
uiffeis somewhat. This is not suipiising since the suiface neai the optimal solution
is ielatively flat (see Schectei anu 0Neill, 2u1S).
A stanuaiu test foi iobustness is incieasing uemanu. The maximum ieal
powei geneiation is 7.724 anu the fixeu ieal uemanu is 2.S9. The maximum
possible uemanu inciease factoi is 2.98. We solve the 14bus pioblem with uemanu
multiplieu by 1.S, 2 anu 2.S. The pioblems iequiie incieasing CP0 time. When
uemanu is 2.S times gieatei, the penalty slacks aie gieatei than the set thiesholu. In
2x anu 2.Sx uemanu, C0N0PT uiu not conveige on two staiting points, but the
u0R0BI solutions weie iobust.
Subus pioblem. Without ielaxation vaiiables, all solveis with the exception of
KNITR0 finu the same optimal solution foi all staiting points. 0n aveiage, SN0PT,
C0N0PT anu NIN0S weie the fastest solveis anu solve 0PFp fastei than 0FP2 anu
0PF. NIN0S, C0N0PT anu IP0PT solve fastei than u0R0BI solveu 0PFlin.
With ielaxation slack vaiiables, all solveis ueclaie a local optimal solution,
but only C0N0PT anu IP0PT finu the global optimal value. The othei solveis finu
optimal solutions neai the global optimal value. 0n aveiage, SN0PT is the fastest
solvei followeu by C0N0PT. C0N0PT outpeifoims the othei nonlineai solveis on
the 0PFp wheie SN0PT appeais to get hung up on some pioblems. KNITR0 is veiy
eiiatic at times woiking well anu at times appeaiing to get hung up. 0veiall,
C0N0PT outpeifoims the othei nonlineai solveis in finuing the global optimal
solution.
Foi the lineai solveis, u0R0BI outpeifoims CPLEX. The optimal solution
values aie within 1 peicent of the best nonlineai solveis. u0R0BI is not as fast as
SN0PT, but finus a bettei solution on 0PF. C0N0PT outpeifoims u0R0BI. The lineai
solveis uo not to have pioblems with ielaxation slack vaiiables.
We compaie the 0PF C0N0PT solution to the u0R0BI 0PFlin solution anu
finu uiffeiences in P, Q anu v anu noual piices foi ieal anu ieactive powei. The
actual geneiatoi uispatch is ieasonably close anu the objective function value uiffeis
by less than one peicent. We solve the Subus pioblem with uemanu multiplieu by
1.2, 1.4 anu 1.6. The iesults aie not maikeuly uiffeient.
S7bus Pioblem. Without ielaxation slack vaiiables, all solveis except KNITR0 finu
the global optimal value foi all staiting points. SN0PT, C0N0PT anu NIN0S aie the
fastest nonlineai solveis. u0R0BI is fastei on 0PFlin than any nonlineai solvei.
Without ielaxation slack vaiiables foi 0PF2, C0N0PT peifoims bettei on the flat
stait than any ianuom staiting point. Foi 0PF2 on thiee staiting points, C0N0PT

Page
11
got hung up. Foi the 0PF, C0N0PT peifoims bettei on the flat stait than on most
ianuom staiting points.
With ielaxation slack vaiiables, only C0N0PT anu IP0PT finu the global
optimal value. The othei solveis finu optimal solutions neai the global optimal
value. SN0PT is the slowest solvei anu gets hung up on some pioblems. C0N0PT
outpeifoims the othei nonlineai solveis on the 0PF2 anu 0PF, but C0N0PT fails to
conveige on one pioblem, anu has highei optimal values. Foi lineai solveis, CPLEX
times bounu the u0R0BI times. The optimal solution values aie within 1 peicent of
the best nonlineai solveis. With uemanu incieaseu by 1.S, 2 anu 2.S, minoi
uiffeiences aie obseiveu.
We compaie the C0N0PT 0PF solution to the u0R0BI 0PFlin solution anu
finu only minoi uiffeiences in P, Q anu v anu noual piices foi ieal anu ieactive
powei. The voltages anu uual vaiiables aie similai but the actual geneiatoi uispatch
is uiffeient, anu the objective function value uiffeis by less than one peicent.
118bus Pioblem. Without ielaxation slack vaiiables, all solveis except KNITR0 finu
the global optimal value foi all staiting points. SN0PT, C0N0PT anu NIN0S aie the
fastest solveis. CPLEX anu u0R0BI aie fastei on 0PFlin than any nonlineai solvei,
except NIN0S. 0PFp uiu not solve fastei than the 0PF.
With ielaxation slack vaiiables, all nonlineai solveis have tiouble. 0nly
C0N0PT anu KNITR0 finu the optimal solution foi the 0PF2. C0N0PT, IP0PT anu
KNITR0 founu the optimal solution to the 0PF. With ielaxation slack vaiiables,
C0N0PT founu optimal values neai the optimal value anu peifoimeu bettei on the
flat stait than the aveiage ianuom staiting point foi 0PF2 anu 0PF.
With ielaxation slack vaiiables, all nonlineai solveis have tiouble solving the
pioblems. 0nly C0N0PT anu KNITR0 finu the optimal solution foi the 0PF2.
C0N0PT, IP0PT anu KNITR0 founu the optimal solution to the 0PF. Foi lineai
solveis, u0R0BI anu CPLEX have similai peifoimance with ielaxation slack
vaiiables. The optimal solution values aie within 2 peicent of the best nonlineai
solvei peifoimance. With uemanu incieaseu by 1.2 anu 1.4, 0PFlin is moie iobust.
Suubus pioblem. Without ielaxation slack vaiiables, all solveis ueclaie an infeasible
pioblem. With ielaxation slack vaiiables, all solveis except KNITR0 ueclaie a local
optimal solution, but occasionally fail to iemove the penalty slacks. 0nly C0N0PT
anu IP0PT finu the global optimal value.
Foi lineai solveis, u0R0BI outpeifoims CPLEX. The optimal solution values
aie within S peicent of the best nonlineai solveis. All lineai appioximations took
the full 1u passes anu uiu not meet the othei conveigence ciiteiia. The lineai solveis
weie moie iobust than the nonlineai solveis. With uemanu incieaseu by 1.2 anu
1.4, the pioblems iequiie moie CP0 time, the penalty slacks aie gieatei than the set
thiesholu anu conveigence becomes moie uifficult.


Page
12
4. Summaiy of Results anu Conclusions.
In geneial, the Iv lineai appioximation solves fastei than the nonlineai
solveis useu in this stuuy. The iteiative lineai piogiam appioach appeais to finu a
neai feasible anu neai optimal solution in almost all pioblems anu staiting points.
The accuiacy of the nonlineai solveis is bettei when the optimal solution is founu.
0n seveial staiting points, the nonlineai solveis faileu to conveige oi containeu
positive ielaxation vaiiables above the thiesholu. The flat stait is supeiioi to
ianuom staits. 0PF solves fastei than 0PF2. The iankeu peifoimance of the
nonlineai solveis changes uepenuing on the test pioblem.
The lineai piogiam appioximation is moie iobust anu consistent acioss
staiting points. In some instances, the lineai solution vaiiables uo not match closely
to the nonlineai solution vaiiables, but aie close to feasible solutions. This is not
suipiising since the suiface neai the optimal is ielatively flat (see Schectei anu
0Neill). Even the ciuue implementation of the iteiative lineai piogiam appioach is
fastei than the nonlineai solveis. 0sing the lineai appioximation appioach opens up
the possibility of solving the AC optimal topology anu AC optimal unit commitment
pioblems fastei using NIP algoiithm.
.

Page
13
Refeiences
Naiy B. Cain, Richaiu P. 0Neill, Anya Castillo, Bistoiy of 0ptimal Powei Flow anu
Foimulations, FERC technical papei, Becembei, 2u12

}avau Lavaei anu S. B. Low, Zeio uuality gap in optimal powei flow pioblem, IEEE
Tiansactions on Powei Systems, 27(1):921u7, Febiuaiy 2u12

Richaiu P. 0Neill, Anya Castillo Naiy B. Cain, The Iv Foimulation anu
Lineaiizations of the AC 0ptimal Powei Flow Pioblem, FERC staff technical papei,
Becembei, 2u12

A. Schectei anu R. P. 0Neill, Exploiation of the AC0PF Feasible Region foi the
Stanuaiu IEEE Test Set, FERC Technical Confeience, Bocket No. AB1u12uuS,
available at http:www.feic.govEventCalenuaiFiles2u12u626u8462S
Tuesuay,%2uSession%2uTC1,%2uSchectei%2u.puf, }une 26, 2u12.

B.C. Lesieutie, I.A. Biskens. Convexity of the Set of Feasible Injections anu Revenue
Auequacy in FTR Naikets. IEEE Tiansactions on Powei Systems, 2u(4), 179u1798,
2uuS.

I. A. Biskens, R.}. Bavy. Exploiing the powei flow solution space bounuaiy. IEEE
Tiansactions on Powei Systems, 16(S), S89S9S, 2uu1.

Page
14
Appenuix
Solution Statistics. The summaiy statistics aie ovei the set of staiting points. They incluue
0PTvAvu is the aveiage optimal solution.
CP0Avu is the aveiage cpu time.
CP0NAX is the maximum cpu time.,
C0NvRu foi nonlineai solveis is the numbei that iepoiteu conveigence ovei the
ten staiting points. Foi lineai solveis is the numbei of lineai piogiams solveu ovei
the ten staiting points.
0PTvSBv is the stanuaiu ueviation fiom the aveiage optimal solution.
CP0SBv is the stanuaiu ueviation fiom the aveiage cpu time.
0PTvCv is the coefficient of vaiiation (stanuaiu ueviationmean) of optimal values
CP0Cv is the coefficient of vaiiation (stanuaiu ueviationmean) of CP0 times
ITER is the numbei of uiffeient staiting points.
NIN0PF is the minimum objective function value founu.
0PTvAL is the optimal value ietuineu by the solvei.
0PTN is the objective solution value less the penalty value.
Pflow is the ieal powei flow thiough the bus
Qflow is the ieactive powei flow thiough the bus

In tables with the uual vaiiables, they aie fiom the bus equations,
P = v
i
*I
i
* + v
j
*I
j
*
Q = v
j
*I
i
* v
i
*I
j
*


Page
15
14bus pioblem
All solveis with the exception of KNITR0 founu the global optimal solution foi each
staiting point, see Table 14.1. 0n aveiage, SN0PT , C0N0PT anu NIN0S weie the fastest
solveis. Compaiing 0PF anu 0PFp, the LP appioximation staiting point uiu seem to help.
u0R0BI was competitive with the nonlineai solveis on the 0PF.

Table 14.1. Solution Statistics foi the 14Bus Pioblem foi the Nonlineai Solveis
Without Relaxation Slack vaiiables.
uate 2u12 S29
type optvavg optvcv cpuavg cpucv cpumax C0NvRu Tepsin
0PF2
C0N0PT 8u.81Su u.uuuu u.uS2u u.1S6u u.1u9u 1u u
NIN0S 8u.81Su u.uuuu u.u69u u.u8uu u.u94u 1u u
SN0PT 8u.81Su u.uuuu u.u7Su u.1S4u u.12Su 1u u
KNITR0 81.878u u.uu2u u.2SSu u.2u4u u.7SSu 1u u
IP0PT 8u.81Su u.uuuu u.S7uu u.u66u u.Suuu 1u u
0PF
SN0PT 82.762u u.uuuu u.uS9u u.u8Su u.u62u 1u u
C0N0PT 82.762u u.uuuu u.uS1u u.11Su u.u94u 1u u
NIN0S 82.762u u.uuuu u.uS2u u.u96u u.u78u 1u u
IP0PT 82.762u u.uuuu u.6u8u u.146u 1.S1uu 1u u
KNITR0 84.u92u u.uuSu 2.S8Su u.7SSu 2u.6u8u 1u
0PFp
NIN0S 82.762u u.uuuu u.u44u u.1S8u u.u78u 1u u
C0N0PT 82.762u u.uuuu u.u47u u.22Su u.11uu 1u u
SN0PT 82.762u u.uuuu u.u76u u.2u7u u.2uSu 1u u
IP0PT 82.762u u.uuuu u.SS6u u.uS8u u.484u 1u u
KNITR0 8S.u78u u.uu2u u.S6Su u.u68u u.S1Su 1u u
Btheta
u0R0BI 76.S91u u.uuuu u.uuSu u.u4uu u.uu4u 1u u
u0R0BI 76.S91u u.uuuu u.uu4u u.uSSu u.uuSu 1u u
u0R0BI 76.S91u u.uuuu u.uu4u u.u82u u.uu6u 1u u
CPLEX 76.S91u u.uuuu u.u18u u.uSuu u.u21u 1u u
CPLEX 76.S91u u.uuuu u.u19u u.117u u.u4uu 1u u
0PFlin
u0R0BI 82.29uu u.uuSu u.u41u u.S26u u.uu9u 6u u
u0R0BI 82.29uu u.uuSu u.u41u u.S7uu u.uu6u 6u u
u0R0BI 82.29uu u.uuSu u.u41u u.S8uu u.uu6u 6u u
CPLEX 82.SSuu u.uu4u u.u97u u.S1Su u.u18u S1 u
CPLEX 82.SSuu u.uu4u u.1u7u u.S1uu u.uSSu S1 u

Table 14.1a. Lineai Piogiam Passes foi u0R0BI by Staiting Point
With anu Without Relaxation Slack vaiiables
Staiting point 1 2 S 4 S 6 7 8 9 1u Total
Without 4 4 S 9 S 8 7 8 6 4 6u
With S 9 7 9 8 8 8 8 9 9 8u

Page
16

With ielaxation slack vaiiables, all solveis ueclaie a local optimal solution foi each
staiting point, but only C0N0PT anu IP0PT finu the global optimal value see Table 14.1i.
The othei solveis finu local optimal solutions neai the global optimal value. 0n aveiage
NIN0S is the fastest solvei followeu by C0N0PT. SN0PT is eiiatic, at times woiking well
anu at times appeaiing to get hung up. 0veiall, since it is the seconu fastest anu always
finus a global optimum, C0N0PT outpeifoims the iemaining nonlineai solveis.

Table 14.1i. Solution Statistics foi the 14Bus Pioblem with the Nonlineai Solveis
With Relaxation Slack vaiiables.
type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN NIN0PT
0PF2
NIN0S 82.279 u.uu2 u.uS2 u.2S4 u.u94 1u u 81.849
C0N0PT 8u.81S u.uuu u.u8S u.u6S u.1u9 1u u 8u.81S
KNITR0 81.148 u.uu1 u.197 u.261 u.67u 1u u 81.uu9
IP0PT 8u.81S u.uuu u.4S2 u.u62 u.S62 1u u 8u.81S
SN0PT 8S.uS1 u.uuS 14.1S7 u.714 1u4.17 1u u 81.267
0PF
NIN0S 8S.u77 u.uu2 u.u42 u.129 u.u6S 1u u 84.S82
C0N0PT 82.762 u.uuu u.uS6 u.1S2 u.1u9 1u u 82.762
KNITR0 8S.S8S u.u1S u.17u u.uSS u.2S4 1u u 8S.u42
IP0PT 82.762 u.uuu u.446 u.uS8 u.S46 1u u 82.762
SN0PT 84.249 u.uu2 28.2S8 u.947 281.89 1u u 8S.S46
0PFp
NIN0S 8S.S4u u.uu1 u.uu8 u.S17 u.u16 1u u 82.894
C0N0PT 82.762 u.uuu u.u2u u.Su7 u.u62 1u u 82.762
KNITR0 84.767 u.u18 u.1S6 u.17S u.S28 1u u 82.9S4
IP0PT 82.762 u.uuu u.2S7 u.uS9 u.281 1u u 82.762
SN0PT 82.9S4 u.uu1 44.842 u.948 448.17 1u u 82.762

Foi the lineai solveis, u0R0BI outpeifoims CPLEX in cpu time. The optimal solution
values aie within 1 peicent of the best nonlineai solveis. ueneially, conveiging in 4 to 7
lineai piogiam passes (see table 14.2i). u0R0BI anu CPLEX aie fastei with ielaxation slack
vaiiables.


Page
17
Table 14.2i. Solution statistics foi the 14bus pioblem with the lineai solveis
with ielaxation slack vaiiables.
Bate 2u12 S1S
type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu
BTBETA
u0R0BI 76.S91 u.uuu u.uu2 u.uSu u.uu2 1u
u0R0BI 76.S91 u.uuu u.uu2 u.uSu u.uu2 1u
u0R0BI 76.S91 u.uuu u.uu2 u.1SS u.uu4 1u
CPLEX 76.S91 u.uuu u.uu9 u.u49 u.u11 1u
CPLEX 76.S91 u.uuu u.u1u u.111 u.u2u 1u
0PFlin
u0R0BI 82.4u2 u.uuS u.u22 u.Su6 u.uuS S6
u0R0BI 82.4u2 u.uuS u.u22 u.281 u.uuS S6
u0R0BI 82.4u2 u.uuS u.u24 u.SS2 u.uu4 S6
CPLEX 82.S68 u.uuS u.u6S u.12S u.uu8 62
CPLEX 82.S68 u.uuS u.u6S u.214 u.uu8 62

Table 14.S iepoits the test iesults iun on a uiffeient uay. The notable uiffeience is in the
iepoiteu times.
Table 14.S. Solution statistics at a uiffeient time with ielaxation slack vaiiables.
0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN
0PF2
NIN0S 82.279 u.uu2 u.uS4 u.2uS u.1u9 1u u
C0N0PT 8u.81S u u.u67 u.147 u.1u9 1u u
KNITR0 81.48 u.uu4 u.194 u.187 u.484 1u u
IP0PT 8u.81S u u.49S u.u49 u.624 1u u
SN0PT 8S.uS1 u.uuS 16.429 u.7S4 12S.612 1u u
0PF
C0N0PT 82.762 u u.uSS u.178 u.u94 1u u
NIN0S 8S.u77 u.uu2 u.uS9 u.127 u.u94 1u u
KNITR0 12S.766 u.26S u.222 u.1S8 u.48S 1u u
IP0PT 82.77S u u.S16 u.u61 u.6SS 1u u
SN0PT 84.249 u.uu2 S2.uS2 u.947 S19.848 1u u
0PFp
NIN0S 8S.S4 u.uu1 u.u16 u.444 u.u78 1u u
C0N0PT 82.762 u u.uSS u.227 u.u6S 1u u
KNITR0 87.747 u.uSS u.1SS u.u74 u.171 1u u
IP0PT 82.762 u u.2S1 u.u68 u.SS9 1u u
SN0PT 82.967 u.uu1 46.272 u.948 462.496 1u u

0PFlin
u0R0BI 82.4u7 u.uuS u.u22 u.247 u.uu4 S7 u
CPLEX 82.S68 u.uuS u.u6S u.17S u.u2 62 u
BTBETA
u0R0BI 76.S91 u u.uu2 u.u97 u.uu2 1u u
CPLEX 76.S91 u u.u11 u.1S1 u.u18 1u u


Page
18
C0N0PT. With ielaxation slack vaiiables, the 0PF2 foimulation hau 211 iows, 2S9 columns
884 nonzeioes, 14S nonconstant fiist ueiivatives with S6S evaluation opeiations. The 0PF
anu 0PFp foimulations hau 22S iows, S79 columns, 1,1S9 nonzeioes, 14u nonconstant
fiist ueiivatives nonzeioes with S22 evaluation opeiations. The uiffeience is uue to the
lineai objective function. With anu without ielaxation slack vaiiables foi all staiting points,
C0N0PT founu the optimal solutions of 8u.81 foi 0PF2 anu 82.76 foi 0PF anu 0PFp.
C0N0PT peifoimeu consiueiably fastei on 0PF anu 0PFp. The flat staiting point uiu not
impiove the piocess.

Table 14.4. C0N0PT peifoimance on the 14bus with anu without ielaxation slack vaiiables.
0PF2 0PF 0PFp
without with without With without with
stait iteiations iteiations iteiations iteiations iteiations iteiations
1 SS 122 29 SS 21 2u
2 47 9S S8 42 27 29
S Su 117 28 27 SS 24
4 Su 112 SS S8 42 19
S 67 87 29 S9 2S 28
6 S8 129 S6 S6 2u 24
7 S9 64 28 SS 21 2u
8 Su 86 SS Su S1 S6
9 48 1uS SS SS 21 2u
1u Su 1u2 S1 29 2S 18
aveiage 49 1u2 SS S6 27 24

KNITR0. With ielaxation slack vaiiables, the 0PF2 foimulation hau 2S8 vaiiables, 21u
constiaints (28 lineai equalities, 42 nonlineai equalities, 14u lineai inequalities), 794
nonzeios in the }acobian, anu 89 nonzeios in the Bessian 0PF anu 0PFp S78 vaiiables, 224
constiaints, (42 lineai equalities, 42 nonlineai equalities, 14u lineai inequalities) 948
nonzeios in the }acobian anu 84 nonzeios in the Bessian. The flat stait uoes bettei than the
aveiage ianuom staiting point.

Table 14.4. KNITR0 peifoimance on the 14bus without ielaxation slack vaiiables.
stait 0PF2 0PF 0PFp
point iteiations optvalu iteiations optvalu iteiations optvalu
1 144 8u.81S 1u2 82.762 84 82.762
2 169 8u.81S 1SS 82.762 9u 82.762
S 166 8u.81S 1SS 82.762 87 82.762
4 2u2 8u.81S 14S 82.762 86 82.762
S 1S9 8u.81S 126 82.762 87 82.762
6 222 8u.81S 1S1 82.762 91 82.762
7 184 8u.81S 92 82.762 89 82.762
8 189 8u.81S 1S8 82.762 91 82.762
9 172 8u.81S 99 82.762 89 82.762
1u 162 8u.81S 1S2 82.762 91 82.762
avg 177 8u.81S 129 82.762 89 82.762


Page
19
With ielaxation slack vaiiables foi one ianuom staiting point KNITR0 ueclaieu
conveigence with an objective value of 169.8u with ielaxation slack vaiiables active at veiy
low levels anu an objective value of 99.u14 when the penalty values aie iemoveu. KNITR0
also hau tiouble with the ieactive powei foimulation of Q = Q
+
Q

not uiiving one membei


of each paii to zeio in the optimal solution.

Table 14.4. KNITR0 peifoimance on the 14bus with ielaxation slack vaiiables.
Stait 0PF2 0PF 0PFp
point Iteiation value Iteiation value Iteiation value
1 2u 8.1u1 24 8.788 27 18.64u
2 18 8.142 21 8.Su4 21 8.Su4
S 21 8.1uu 26 9.S91 24 8.Su4
4 2u 8.1u1 2S 14.21u 19 8.Su4
S 24 8.147 2S 8.Su4 42 11.u2u
6 14 9.721 4S 169.8uu 22 8.Su4
7 Su 8.11u S2 8.S1u 21 8.Su4
8 21 8.1u1 2S 8.Su4 2S 8.Su4
9 S6 8.1S7 24 8.Su4 21 8.Su4
1u 24 8.1u1 24 8.SuS 24 8.47u
avg 2S 8.276 27 2S.2u2 24 9.626

SN0PT. With ielaxation vaiiables, the 0PF2 foimulation hau 211 iows, 2S9 columns, 884
nonzeioes, anu 14S nonconstant fiist ueiivatives with S6S evaluation opeiations. The 0PF
anu 0PFp foimulations hau 22S iows, S79 columns, 1,1S9 nonzeioes, anu 14u with S22
evaluation opeiations. The flat stait uoes bettei than the aveiage ianuom point. With
ielaxation slack vaiiables, SN0PT occasionally got hung up on the 0PF2 anu 0PF (see Table
14.6).

Table 14.6 SN0PT peifoimance on the 14bus with anu without ielaxation slack vaiiables.
0PF2 0PF 0PFp
without with without with without with
stait iteis optval Iteis optval iteis optval Iteis optval iteis optval Iteis optval
1 1S9 8u.82 7 84.uS 8S 82.76 6 76.S9 6S 82.76 2 82.89
2 18S 8u.82 >eS 84.uS 1u6 82.76 4 8S.S4 71 82.76 2 82.88
S 2SS 8u.82 8 84.16 149 82.76 >e6 8S.79 66 82.76 S 82.76
4 17S 8u.82 S 84.28 141 82.76 8 81.27 7u 82.76 S 8S.uS
S 214 8u.82 8 81.27 116 82.76 8 84.29 67 82.76 S 82.76
6 199 8u.82 8 8S.Su 1SS 82.76 S 8S.S6 66 82.76 S 82.76
7 19u 8u.82 8 84.u1 149 82.76 S 84.u1 66 82.76 6 82.76
8 2uu 8u.82 8 81.27 1S4 82.76 6 84.11 77 82.76 S 82.76
9 222 8u.82 9 81.47 1uu 82.76 4 84.19 67 82.76 S 87.4S
1u 21S 8u.82 8 84.u7 1uu 82.76 4 8S.7S 64 82.76 S 82.76
avg 199 8u.82 8 8S.S9 124 82.76 6 82.9S 68 82.76 S 8S.4S



Page
20
NIN0S. Without ielaxation slack vaiiables, the flat stait uoes bettei than the aveiage
ianuom point. With ielaxation slack vaiiable, NIN0S was fast, but nevei founu the optimal
value.

Table 14.7 NIN0S peifoimance on the 14bus with anu without ielaxation slack vaiiables.
0PF2 0PF 0PFp
without with without with without with
stait iteis optval Iteis optval iteis optval Iteis optval iteis optval Iteis optval
1 18 8u.82 6 82.SS 29 82.76 6 8S.61 16 82.76 6 8S.78
2 29 8u.82 7 82.S4 46 82.76 6 8S.uS 1S 82.76 6 8S.82
S 28 8u.82 6 81.8S 42 82.76 6 84.S8 24 82.76 S 8S.6S
4 2S 8u.82 6 81.87 SS 82.76 7 84.88 S7 82.76 S 8S.88
S 2S 8u.82 6 81.87 9S 82.76 6 8S.Su 1S 82.76 4 82.9u
6 19 8u.82 6 81.8S 19 82.76 6 86.11 27 82.76 S 8S.94
7 2u 8u.82 6 81.87 28 82.76 6 84.6S 22 82.76 S 8S.79
8 S2 8u.82 6 82.48 S4 82.76 7 84.89 16 82.76 S 8S.68
9 44 8u.82 S 82.1S SS 82.76 S 8S.22 16 82.76 4 82.89
1u 28 8u.82 6 82.2S SS 82.76 6 84.7S Su 82.76 S 8S.4u
avg 28 8u.82 6 82.u8 4S 82.76 6 8S.u8 22 82.76 S 8S.S7

u0R0BI. u0R0BI piouuces a BTBETA foimulation with 247 iows, SuS columns anu 1,117
nonzeioes. u0R0BI piesolve iesulteu in 2S iows, 6S columns, 8S nonzeios. u0R0BI founu
an optimal solution in 1S iteiations with an objective value of 7.6S9 in less than .uuSs.
u0R0BI piouuces an 0PFlin foimulation with S94 iows, S79 columns anu 1,S6S non
zeioes. Bepenuing on the staiting point, takes 4 to 9 passes of the lineai piogiam.


Page
21
Table 14.8. Lineai Appioximation Nonzeios, Iteiations anu 0ptimal value by LP Pass anu
Staiting Point foi the 14Bus 0sing u0R0BI with Relaxation Slack vaiiables
pass non
zeios
iteis optval non
zeios
iteis optval non
zeios
iteis optval
stait point 1 2 S
1 1281 S6 77.22 1SuS 64 79.41 1SuS 62 78.9S
2 1SS9 71 81.69 1S6S 67 78.91 1S6S 6S 78.47
S 1S91 71 82.46 1S87 62 77.SS 1S87 6S 77.S2
4 1419 76 82.S4 1417 69 78.99 1417 71 8u.9S
S 1447 7u 82.Su 144S 71 82.29 144S 7u 82.48
6 147S 74 82.SS 147S 71 82.S6 147S 86 82.S6
stait point 4 S 6
1 1SuS S6 78.28 1SuS 62 79.1S 1SuS 6S 78.48
2 1S6S 68 77.S7 1S6S 62 78.S2 1S6S 6u 78.17
S 1S91 7u 79.98 1S87 67 77.49 1S87 S8 77.S6
4 1419 7S 82.27 1417 7S 81.81 1417 66 81.82
S 1447 7S 82.S4 144S 7u 82.6u
6 147S 7u 82.S2 147S 69 82.S7
7 1SuS 68 82.S2 1Su1 71 82.6S
stait point 7 8 9
1 1SuS S9 77.98 1SuS 62 79.SS 1SuS S7 78.1S
2 1S6S 61 77.S6 1S6S 62 78.96 1S6S 67 77.6S
S 1S91 68 82.u7 1S87 68 78.42 1S91 71 82.S2
4 1419 76 82.u4 141S S8 77.89 1419 68 82.6u
S 14S9 67 78.81 1447 66 82.S7
6 1469 66 82.2S 147S 8S 82.62
7 1SuS 71 82.SS
8 1SS1 77 82.62
stait point 1u
1 1SuS S6 77.7S
2 1S6S 71 78.77
S 1S91 7u 82.72
4 1419 77 82.64
S 1447 81 82.61
6 147S 7S 82.SS
7 1SuS 84 82.62
8 1SS1 7u 82.SS



Page
22
CPLEX. CPLEX piouuces a BTBETA foimulation with 247 iows, SuS columns, anu 1,117
nonzeioes that was ieuuceu to 2S iows, 64 columns, anu 86 nonzeios. Almost iuentical to
u0R0BI, At CPLEXuefineu iteiation 1, CPLEX founu an optimal solution with a value of
76.S9. CPLEX is veiy similai to u0R0BI.

Table 14.9. Lineai appioximation nonzeios, iteiations anu optimal value by LP pass anu
staiting point foi the 14bus using CPLEX
pass non
zeios
iteiations optval non
zeios
iteiations optval non
zeios
iteiations optval
stait point 1 2 S
1 1281 87 77.22 1SuS 7S 79.41 1SuS 7S 78.9S
2 1SS9 81 81.69 1S6S 9S 78.91 1S6S 8S 78.47
S 1S91 9u 82.6S 1S87 1uS 77.SS 1S87 1u2 77.S2
4 1419 87 82.S4 1417 8S 78.99 141S 14S 78.71
S 1447 91 82.62 144S 8u 82.16 14S9 189 82.18
6 147S 82 82.S6 147S 8S 82.S7 146S 1u1 82.49
stait point 4 S 6
1 1SuS 7S 78.28 1SuS 8u 79.1S 1SuS 7u 78.48
2 1S6S 77 77.S7 1S6S 8u 78.S2 1S6S 7u 78.17
S 1S91 86 81.61 1S87 1u2 77.49 1S87 12u 77.S6
4 1419 1u2 82.66 1417 117 81.81 1417 18S 81.82
S 1447 8S 82.66 144S 1u7 82.6u
6 147S 79 82.S2 147S 91 82.S7
7 1SuS 84 82.28 1Su1 84 82.6S
8 1S29 8u 82.SS
stait point 7 8 9
1 1SuS 81 77.98 1SuS 77 79.SS 1SuS 71 78.1S
2 1S6S 8S 77.S6 1S6S 84 78.96 1S6S 86 77.6S
S 1S91 86 82.u7 1S87 11S 78.42 1S91 92 82.29
4 1419 79 82.u6 141S 76 77.89 1419 8S 82.4S
S 1447 1u6 82.S6 14S9 124 78.81 1447 86 82.42
6 1469 118 82.2S
stait point 1u
1 1SuS 76 77.7S
2 1S6S 89 78.77
S 1S91 8S 82.72

Compaiing the 0PF to the lineai appioximation. We compaie the 0PF solution to the 0PFlin
solution anu finu only minoi uiffeiences in P, Q anu v anu noual piices foi ieal anu ieactive
powei. We examine a C0N0PT solution anu lineai appioximation using u0R0BI. The
objective function values uiffei by less than one peicent. In tables 14.1u anu 14.11, the
voltage magnituues anu uual vaiiables fiom the bus equations aie similai but the actual
geneiatoi uispatch (ieal anu ieactive) uiffeis somewhat.


Page
23
Table 14.1u Solution Infoimation fiom The C0N0PT Solvei With 0bjective value 0f 82.76
bus voltage geneiation loau Lambua ($NvAhi)
no. Nag
(pu)
Ang
(ueg)
P
(NW)
Q
(NvAR)
P
(NW)
Q
(NvAR)
P Q
1 1.u6 u 1.826 u.1 u u SS.7SS 2.SS6
2 1.u4 S.7S u.42 u.S u.217 u.127 S7.672 2.924
S u.984 9.19 u.S u u.942 u.19 4u.6S7 S.9S1
4 u.999 8.1S4 u u u.478 u.uS9 4u.u92 S.68S
S 1.uu4 6.9S6 u u u.u76 u.u16 S9.S96 S.66S
6 1.uS2 12.uS8 u.uSS u.u47 u.112 u.u7S 4u.1 4.uu1
7 1.u2 1u.S96 u u u u 4u.412 4.u42
8 1.u2S 9.6S1 u.1 u.uS2 u u 4u.277 4.uu1
9 1.u17 12.Su4 u u u.29S u.166 4u.682 4.2S8
1u 1.u12 12.729 u u u.u9 u.uS8 4u.88S 4.S89
11 1.u18 12.S2S u u u.uSS u.u18 4u.6S8 4.287
12 1.u16 12.9S7 u u u.u61 u.u16 4u.94 4.28S
1S 1.u11 1S.uu2 u u u.1SS u.uS8 41.18 4.449
14 u.996 1S.8 u u u.149 u.uS 42.uu2 4.726
totals 2.679 u.679 2.S9 u.7SS losses u.u89

Table 14.11 Solution Infoimation fiom Lineai Appioximation 0sing u0R0BI
With 0bjective value 0f 82.6S
bus voltage geneiation loau Lambua ($NvAhi)
no. Nag
(pu)
Ang
(ueg)
P
(NW)
Q
(NvAR)
P
(NW)
Q
(NvAR)
P Q
1 1.u7S u 1.662 u.1 u u SS.6S1 2.SS8
2 1.uSS S.41 u.42 u.S u.217 u.127 S7.SS9 2.868
S u.996 9.14S u.184 u u.942 u.19 4u.S S.961
4 1.u11 7.2u2 u u u.478 u.uS9 4u.192 S.627
S 1.u16 6.u87 u u u.u76 u.u16 S9.487 S.618
6 1.uSS 1u.221 u.1 u u.112 u.u7S 4u.274 S.989
7 1.u2S 8.164 u u u u 4u.S9S S.98S
8 1.u2 S.S21 u.S u.uuS u u 4u.S2S 4.uu1
9 1.u22 1u.449 u u u.29S u.166 4u.872 4.144
1u 1.u16 1u.722 u u u.u9 u.uS8 41.111 4.29
11 1.u21 1u.614 u u u.uSS u.u18 4u.881 4.219
12 1.u17 11.122 u u u.u61 u.u16 41.2u8 4.2uS
1S 1.u1S 11.17S u u u.1SS u.uS8 41.477 4.S46
14 u.998 11.866 u u u.149 u.uS 42.SS6 4.S2S
totals 2.666 u.6uS 2.S9 u.7SS losses u.u76


Page
24
Incieasing Bemanu. A stanuaiu test foi iobustness is incieasing uemanu. The maximum ieal
powei geneiation is 7.724 anu the fixeu ieal uemanu is 2.S9. The maximum possible
uemanu inciease factoi is 2.98. We solve the 14bus pioblem with uemanu multiplieu by
1.S, 2 anu 2.S. The iesults aie in table 14.12. The pioblems iequiie incieasing CP0 time.
When uemanu is 2.S times gieatei, the penalty slacks aie gieatei than the set thiesholu. In
2x anu 2.Sx uemanu, C0N0PT uiu not conveige on two staiting points. The u0R0BI
solutions weie iobust.

Table 14.12. Solution Statistics foi 14Bus Pioblem With Incieaseu Bemanu 0sing C0N0PT
anu u0R0BI With Relaxation Slack vaiiables
type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN
0PF2 8u.81S u.uuu u.u67 u.14S u.12S 1u u
0PF 82.762 u.uuu u.u61 u.1SS u.u94 1u u
0PFlin 81.S84 u.u18 u.u1S u.S21 u.uuS SS u
0PFp 82.762 u.uuu u.u41 u.22u u.u94 1u u
BTBETA 76.S91 u.uuu u.uu2 u.1uu u.uuS 1u
1.S*uemanu
0PF2 1SS.668 u.uuu u.u8u u.1S6 u.12S 1u u
0PF 1S7.1S2 u.uuu u.u78 u.u6S u.u94 1u u
0PFlin 1SS.96S u.uuS u.u1S u.S68 u.uu4 S9 u
0PFp 1S7.1S2 u.uuu u.u48 u.221 u.1u9 1u u
BTBETA 128.6u7 u.uuu u.uu2 u.u97 u.uu2 1u
2*uemanu
0PF2 217.4S8 u.uuu u.u94 u.u94 u.14u 9 1
0PF 222.S1u u.uuu u.u69 u.141 u.12S 1u u
0PFlin 212.uSS u.u17 u.uS1 u.48u u.uuS 72 u
0PFp 222.S1u u.uuu u.u78 u.18S u.12S 1u u
BTBETA 181.92S u.uuu u.uu2 u.u8S u.uu2 1u
2.S*uemanu
0PF2 268.SS2 u.uuu u.u81 u.119 u.12S 8 1u
0PF 27S.S22 u.uuu u.u9u u.u68 u.12S 1u 1u
0PFlin 272.S21 u.uu1 u.u19 u.S6S u.uu4 4S 1u
0PFp 27S.S22 u.uuu u.uS8 u.184 u.11u 1u 1u
BTBETA 2S6.97S u.uuu u.uu2 u.u7u u.uu2 1u

Page
25
Subus pioblem
Without ielaxation vaiiables, all solveis with the exception of KNITR0 finu the same
optimal solution foi all staiting points see table Su.1. NIN0S, C0N0PT anu IP0PT aie fastei
than the u0R0BI solution. 0n aveiage, SN0PT, C0N0PT anu NIN0S weie the fastest solveis.
0PFp solves fastei than 0FP2 anu 0PF .

Table Su.1. Solution Statistics foi the SuBus Pioblem Without Relaxation Slack vaiiables.
type optvavg optvcv cpuavg cpucv cpumax C0NvRu Tepsin
0PF2
NIN0S S.74S u.uuu u.1uS u.u72 u.141 1u u
C0N0PT S.74S u.uuu u.117 u.u87 u.1S6 1u u
SN0PT S.74S u.uuu u.1S6 u.u67 u.172 1u u
IP0PT S.74S u.uuu u.SS2 u.u61 u.7S4 1u u
KNITR0 7.S6S u.1uS u.674 u.221 1.716 1u u
0PF
SN0PT S.918 u.uuu u.uS9 u.u62 u.u78 1u u
C0N0PT S.918 u.uuu u.u7S u.12S u.14u 1u u
NIN0S S.918 u.uuu u.u8S u.u97 u.11u 1u u
IP0PT S.918 u.uuu u.722 u.u76 1.uSu 1u u
KNITR0 1u.u18 u.u8S 1.247 u.S82 S.S8S 1u u
0PFp
SN0PT S.918 u.uuu u.u24 u.1u4 u.uS2 1u u
NIN0S S.918 u.uuu u.u27 u.119 u.u47 1u u
C0N0PT S.918 u.uuu u.uS4 u.214 u.u78 1u u
IP0PT S.918 u.uuu u.4S1 u.u87 u.6S6 1u u
KNITR0 8.18u u.1u6 u.S46 u.224 1.S76 1u u
0PFlin
u0R0BI S.914 u.uu2 u.126 u.S12 u.u1S 8S u
u0R0BI S.914 u.uu2 u.129 u.292 u.u1S 8S u
u0R0BI S.914 u.uu2 u.1SS u.S2u u.u1S 8S u
CPLEX S.914 u.uu2 u.2SS u.289 u.uS1 87 u
CPLEX S.914 u.uu2 u.2SS u.Suu u.u29 87 u

Table Su.1a. LP Passes By Staiting Point foi the SuBus 0sing u0R0BI
With Anu Without Relaxation Slack vaiiables
Stait point 1 2 S 4 S 6 7 8 9 1u total
Without 2 1u 1u 1u 1u 7 9 7 9 9 8S
with S 9 7 9 8 8 8 8 9 9 8u

With ielaxation slack vaiiables, all solveis ueclaie a local optimal solution, but only
C0N0PT anu IP0PT finu the global optimal value see Table Su.1i. The othei solveis finu
optimal solutions neai the global optimal value. 0n aveiage SN0PT is the fastest solvei
followeu by C0N0PT. C0N0PT outpeifoims the othei nonlineai solveis on the 0PFp wheie
SN0PT appeais to get hung up on some pioblems. KNITR0 is veiy eiiatic at times woiking

Page
26
well anu at times appeaiing to get hung up. 0veiall, C0N0PT outpeifoims the othei
nonlineai solveis in finuing the global optimal solution.

Table Su.1i. Solution Statistics foi the SuBus Pioblem with The Nonlineai Solveis.
type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN NIN0PT
0PF2
SN0PT 6.722 u.u1u u.u67 u.u99 u.1u9 1u u 6.S19
C0N0PT S.74S u.uuu u.17S u.u98 u.296 1u u S.74S
NIN0S 6.8uS u.uu8 u.2u4 u.S76 u.8S8 1u u 6.4S9
KNITR0 7.472 u.u86 u.626 u.216 1.S7S 1u u S.8u7
IP0PT S.74S u.uuu 1.u19 u.u84 1.7uu 1u u S.74S
0PF
SN0PT 6.216 u.uu7 u.uS6 u.149 u.u94 1u u 6.u42
C0N0PT S.918 u.uuu u.u97 u.1S8 u.188 1u u S.918
NIN0S 6.68S u.u1u u.17S u.2SS u.468 1u u 6.298
KNITR0 62u.6 u.918 u.726 u.162 1.4u4 1u u 6.492
IP0PT S.919 u.uuu 2u1.272 u.628 1uuu.2 8 u S.918
0PFp
C0N0PT S.918 u.uuu u.u4u u.189 u.u78 1u u S.918
KNITR0 1S.492 u.166 u.S77 u.117 u.624 1u u 6.49S
NIN0S 6.727 u.u17 u.S12 u.S4S 1.84u 1u u 6.141
IP0PT S.919 u.uuu u.772 u.u7u 1.1S4 1u u S.918
SN0PT S.928 u.uu1 87.uS6 u.74S 67S.1 1u u S.918

Foi lineai solveis, u0R0BI outpeifoims CPLEX see table Su.2. The optimal solution
values aie within 1 peicent of the best nonlineai solveis. u0R0BI is not as fast as SN0PT,
but finus a bettei solution on 0PF. C0N0PT outpeifoims u0R0BI. The lineai solveis appeai
not to have pioblems with ielaxation vaiiables.
Table Su.2. Solution Statistics foi the SuBus Pioblem with the Lineai Solveis.
type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN
0PFlin
u0R0BI S.9u9 u.uu4 u.167 u.18S u.u2u 8u u
u0R0BI S.9u9 u.uu4 u.18S u.191 u.u28 8u u
u0R0BI S.9u9 u.uu4 u.189 u.216 u.u19 8u u
CPLEX S.91u u.uu4 u.Suu u.2u2 u.uSu 88 u
CPLEX S.91u u.uu4 u.Su8 u.181 u.u4S 88 u
BTBETA
u0R0BI S.6S9 u.uuu u.uu8 u.u26 u.uu9 1u u
u0R0BI S.6S9 u.uuu u.uu9 u.u7S u.u1S 1u u
u0R0BI S.6S9 u.uuu u.uu9 u.u48 u.u11 1u u
CPLEX S.6S9 u.uuu u.u24 u.u17 u.u28 1u u
CPLEX S.6S9 u.uuu u.u28 u.11u u.uS6 1u


Page
27
Table Su.S iepoits the test iesults iun at a uiffeient uay. The only notable uiffeience
is in the iepoiteu times.

Table Su.S Solution Statistics foi the SuBus Pioblem with Relaxation Slack vaiiables Fiom
Anothei Bay.
0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN
0PF2
SN0PT 6.71S u.u1 u.uSS u.221 u.1u9 1u u
C0N0PT S.74S u u.1 u.1S6 u.172 1u u
NIN0S 6.86S u.uu9 u.1S1 u.S8S u.6SS 1u u
KNITR0 8.492 u.118 u.462 u.SS7 1.716 1u u
IP0PT S.746 u u.616 u.u81 u.9S6 1u u
0PF
SN0PT 6.211 u.uu7 u.uS u.1S9 u.u94 1u u
C0N0PT S.918 u u.u69 u.1u7 u.1u9 1u u
NIN0S 6.718 u.u17 u.16S u.S48 u.S46 1u u
KNITR0 8.974 u.24S u.441 u.186 u.9S6 1u u
IP0PT S.918 u 2uu.61 u.6S 1uuu.1SS 8 u
0PFp
C0N0PT S.918 u u.u7S u.111 u.11 1u u
KNITR0 12.1u4 u.111 u.188 u.u89 u.28 1u u
NIN0S 6.89S u.u14 u.S29 u.22 u.796 1u u
IP0PT S.92 u u.S7 u.u71 u.S14 1u u
SN0PT 6.u91 u.u11 1u.1uS u.6S6 S6.89S 1u u
BTBETA
u0R0BI S.6S9 u u.uu4 u.u4S u.uu4 1u
CPLEX S.6S9 u u.u11 u.uS7 u.u14 1u
0PFlin
u0R0BI S.8u1 u.uuS u.u22 u.1u9 u.u1 29 u
CPLEX S.811 u.uu6 u.uS6 u.441 u.u1S S8 u


Page
28
C0N0PT. The 0PF2 foimulation hau 4S1 iows, S11 columns, 1,891 nonzeioes, anu Su6
nonconstant fiist ueiivatives with 7S9 evaluation opeiations. The 0PF anu 0PFp
foimulations hau 481 iows, 811 columns, 2,S9S nonzeioes, anu Suu nonconstant fiist
ueiivatives with 69u evaluation opeiations. With ielaxation slack vaiiables foi the flat stait
(staiting point 1), C0N0PT founu an optimal solution equal to S.74S with a ieuuceu
giauient of S.7Eu9 in 69 iteiations. Foi the fiist ianuom stait (staiting point 2), C0N0PT
founu an optimal solution equal to S.74S with a ieuuceu giauient of 6.6E1u in 1u4
iteiations. Foi the seconu ianuom stait (staiting point S), C0N0PT founu an optimal
solution equal to S.74S with a ieuuceu giauient of 2.SE8 in 11u iteiations.
C0N0PT founu an optimal solution of S.74S foi 0PF2 anu S.918 foi 0PF anu 0PFp.
The flat stait uoes bettei than the aveiage ianuom point foi 0PF2 anu 0PF.

Table Su.4. C0N0PT Peifoimance on the SuBus With anu Without Relaxation Slack
vaiiables.
0PF2 0PF 0PFp
stait without with without with without with
point iteiations iteiations iteiations iteiations iteiations iteiations
1 4S 69 2u 2u 28 S9
2 1u8 9u 89 12u 17 SS
S 48 92 8S 1uu 9 S7
4 121 1uS 9S S2 1u 4S
S S2 4u 9S 6u 12 46
6 178 126 84 42 11 S2
7 4u 148 4u S8 12 1S4
8 118 9S 64 42 26 26
9 146 212 24 SS 1u 6u
1u 1SS 118 74 S2 29 46
avg 99 1u9 67 S6 16 S4


Page
29
KNITR0. KNITR0 uoes not ietuin iteiations. Foi seveial staiting points, KNITR0 ueclaieu
conveigence with an objective value of with ielaxation slack vaiiables active anu solutions
fai off the optimal value when the penalty values aie iemoveu (see Table Su.S).

Table Su.S. KNITR0 Peifoimance on the SuBus With anu Without Relaxation Slack
vaiiables.
0PF2 0PF 0PFp
stait without with without with without with
point optval optval optm optval optval optm optval optval optm
1 11.21 8u.SS8 S.8u7 1u.u42 4S.228 29.64 12.S92 2S.SS 14.691
2 11.124 6.18 6.18 12.uS4 6.49S 6.49S 11.6S4 6.49S 6.49S
S S.7SS 6.176 6.176 11.786 18.SS7 18.SS7 6.u9 7S.662 26.12S
4 6S.74 411.272 S.889 6.1S9 2126u 6u2S.S 12.S9 6.49S 6.49S
S S.7S1 17.7u2 7.6SS 11.772 6.49S 6.49S 6.u89 6.496 6.496
6 S.7S8 11.69S 11.u7 6.17S 6.492 6.492 6.u89 7u.2S 2S.91S
7 S.817 26.614 11.S8S 11.u79 6.S 6.S 6.148 96.968 24.6S1
8 S.766 6.18 6.18 12.u68 9S.Su 9S.SuS 6.u89 6.49S 6.49S
9 S.7S2 S8.626 7.S29 1S.2S6 6.Su8 6.Su8 6.u2 21.S1 2u.674
1u 11.726 Su.8uS 6.6S4 6.21 6.49S 6.49S 8.6S9 16.9uS 16.9uS
avg 1S.24u 6S.SS9 7.472 1u.uS8 214S.8 62u.62 8.18u S2.861 1S.492

SN0PT. SN0PT without ielaxation slack vaiiables peifoim bettei on the flat stait than any
ianuom staiting point. With uiffeient staiting points, SN0PT ueclaieu many uiffeient
solution values optimal on 0PF2 anu 0PF (see Table Su.6). SN0PT peifoim bettei on the
flat stait than on any staiting point foi 0PF2 anu 0PF. With uiffeient staiting points, SN0PT
ueclaieu many uiffeient solution values optimal on any staiting point foi 0PF2 anu 0PF see
table Su.6. SN0PT founu the optimal solutions of S.74S foi 0PF2 anu S.918 foi 0PF anu
0PFp.

Table Su.6 SN0PT Peifoimance on the SuBus With anu Without Relaxation Slack vaiiables
0PF2 0PF2 0PF 0PF 0PFp 0PFp
stait without with without with without with
point iteis iteis optval iteis iteis optval iteis iteis optval
1 19S 92 6.7u7 11S 84 6.u42 11S 12S S.918
2 4Su 2u7 6.889 229 168 6.1u1 141 1uS S.989
S S47 2S7 6.7S7 277 2u4 6.u94 92 6S S.918
4 S98 22S 6.84u 2S7 1S9 6.SS7 9S 112 S.918
S SSS 174 6.S27 196 1S8 6.S72 97 12u S.918
6 48S 212 6.S19 289 292 6.S4 1u4 86 S.9S2
7 S18 247 6.962 222 19u 6.1u2 12S 71 S.9SS
8 44S 169 6.862 2S8 1S4 6.2S4 129 9S S.918
9 S68 171 6.471 22S 177 6.117 89 66 S.918
1u S64 2u6 6.88S 2S2 172 6.4u6 9u 98 S.918
avg S7u 194 6.722 226 174 6.217 1u7 94 S.928


Page
30
NIN0S. Without ielaxation slack vaiiables, NIN0S always founu the optimal value anu
peifoims bettei on the flat stait than any ianuom staiting point on 0PF2 anu 0PF. With
ielaxation slack vaiiables, NIN0S nevei founu the optimal value foi any pioblem foi any
staiting point. NIN0S founu the optimal solutions of S.74S foi 0PF2 anu S.918 foi 0PF anu
0PFp.

Table Su.7 NIN0S Peifoimance on the SuBus Without Relaxation Slack vaiiables
0PF2 0PF 0PFp
stait without with without with without with
point iteis iteis optval iteis iteis optval iteis iteis optval
1 2S8 SS4 6.966 142 187 6.496 1SS S846 6.SS8
2 Su8 18u 6.864 2S1 2S1 6.66S 1S9 217 6.141
S SS6 1299 6.6uS 2S1 76S 6.8S8 146 189S 6.677
4 S46 2462 6.9S4 18S 1S46 7.u14 1S6 1799 7.1Su
S S81 196 6.898 22S 21S 6.729 1S7 6uS2 6.87S
6 SS8 S92 6.727 262 2SS 6.298 162 S88 7.u67
7 28S 164 6.4S9 191 1S1S 6.929 1S9 684 6.7S4
8 2S8 17u 6.861 21u 189 6.41S 1S9 7SS 7.uS8
9 S41 217 6.7u6 229 2u7 6.6S7 161 212 6.21S
1u 268 2S6 6.989 21S 826 6.788 1S9 SS6 7.u14
avg S16 6u7 6.8uS 214 SSS 6.68S 149 16S8 6.727

u0R0BI. The BTBETA foimulation hau S17 iows, 1,SSS columns anu S,28S nonzeioes.
u0R0BI piesolve iesulteu in 41 iows, 12u columns, 16u nonzeios anu founu an optimal
solution in 2S iteiations with objective value of S.6S9. The fiist pass of 0PFlin hau 842
iows 811 columns anu 2,881 nonzeioes. u0R0BI finus the lineai Iv solution value is within
1 peicent of the 0PF solution value.

Table Su.8i. Lineai Appioximation Nonzeios, Iteiations anu 0ptimal value by LP Pass anu
Staiting Point foi the SuBus 0sing u0R0BI with Relaxation Slack vaiiables
pass nonzeios iteis optval nonzeios iteis optval nonzeios iteis optval
stait point 1 2 S
1 27u1 2S2 S.774 2749 14S S.8u2 2749 168 S.798
2 28S9 1S7 S.88u 287S 142 S.8uS 287S 16S S.79S
S 292S 1S9 S.842 29S1 1S6 S.8uS 29S1 1S4 S.78S
4 298S 1S1 S.8SS 2989 16S S.782 2989 167 S.786
S Su4S 1S2 S.8SS Su47 18S S.914 SuS1 164 S.9uS
6 S1uS 168 S.918 S111 167 S.917
7 S167 176 S.917 S171 1S6 S.917
8 S227 18S S.918
9 S287 17S S.918
stait point 4 S 6
1 2749 16u S.79S 2749 144 S.816 2749 16S S.794
2 287S 147 S.792 287S 148 S.811 287S 1S6 S.79u
S 29S1 147 S.787 29S1 1SS S.792 29S1 1Su S.787
4 2989 16u S.8S1 2989 166 S.787 2989 16S S.8uu

Page
31
S SuS1 177 S.87u Su47 168 S.9uS SuS1 171 S.899
6 S111 177 S.917 S1uS 168 S.917 S111 178 S.917
7 S171 162 S.917 S16S 18u S.918 S171 186 S.917
8 S2S1 169 S.918 S221 17S S.918 S2S1 172 S.918
9 S28S 17S S.918
stait point 7 8 9
1 2749 14S S.81u 2749 1SS S.8u6 2749 186 S.816
2 287S 1S2 S.8u7 287S 149 S.8u1 287S 141 S.814
S 29S1 1SS S.79S 29S1 14S S.789 29S1 17S S.8u4
4 2989 166 S.789 2989 172 S.79S 2989 1S4 S.788
S SuS1 166 S.9u6 SuS1 169 S.9uS Su47 16S S.9u7
6 S111 172 S.917 S111 171 S.917 S1uS 171 S.887
7 S171 18S S.917 S171 16S S.917 S167 17S S.918
8 S2S1 164 S.918 S2S1 172 S.918 S227 171 S.918
stait point 1u
1 2749 188 S.796
2 287S 1S4 S.79S
S 29S1 1S1 S.787
4 2989 17u S.84u
S SuS1 17u S.8S1
6 S111 18S S.91S
7 S171 1SS S.9u7
8 S2S1 176 S.918
9 S291 1S9 S.917


Page
32
CPLEX. CPLEX piouuceu essentially the same as u0R0BI.
Compaiing the 0PF to the lineai appioximation. We compaie the C0N0PT 0PF solution to
the u0R0BI 0PFlin solution anu finu uiffeiences in P, Q anu v anu noual piices foi ieal anu
ieactive powei. The actual geneiatoi uispatch is uiffeient, but the objective function value
uiffeis by less than one peicent.

Table Su.1u Solution Infoimation fiom C0N0PT With 0bjective value 0f S.918
bus voltage geneiation loau Lambua($NvAhi)
no. Nag
(pu)
Ang
(ueg)
P
(NW)
Q
(NvAR)
P
(NW)
Q
(NvAR)
P Q
1 1.uS u u.444 u u u S.76 u.17S
2 1.u48 u.78S u.S6 u.S89 u.217 u.127 S.79S u.177
S 1.u27 1.9u7 u u u.u24 u.u12 S.866 u.22
4 1.u2S 2.269 u u u.u76 u.u16 S.887 u.228
S 1.u29 2.218 u u u u S.862 u.2u8
6 1.u17 2.68 u u u u S.9uS u.2SS
7 1.u1S 2.99S u u u.228 u.1u9 S.926 u.242
8 1.uuS S.u76 u u u.S u.S S.928 u.261
9 1.uS S.82S u u u u S.929 u.2u2
1u 1.uS7 4.414 u u u.uS8 u.u2 S.94 u.18S
11 1.uS S.82S u u u u S.929 u.2u2
12 1.u11 S.S48 u u u.112 u.u7S S.911 u.271
1S 1.u11 2.292 u.16 u u u S.899 u.271
14 1.uuS 4.u97 u u u.u62 u.u16 S.97 u.286
1S 1.uu8 S.9S7 u u u.u82 u.u2S S.948 u.27S
16 1.u1S 4.2S4 u u u.uSS u.u18 S.9S u.247
17 1.u2S 4.S78 u u u.u9 u.uS8 S.96 u.21S
18 1.uu6 4.8SS u u u.uS2 u.uu9 4.uu9 u.26S
19 1.uu8 S.188 u u u.u9S u.uS4 4.u26 u.2S1
2u 1.u14 S.u46 u u u.u22 u.uu7 4.uu8 u.2S6
21 1.uS 4.6u2 u u u.17S u.112 S.9S4 u.1S
22 1.uS8 4.S46 u.2S u.6u8 u u S.919 u.1u6
2S 1.u2S 2.741 u.18 u u.uS2 u.u16 S.866 u.266
24 1.u28 S.444 u u u.u87 u.u67 S.92S u.2S4
2S 1.u17 1.667 u u u u S.869 u.287
26 u.999 2.u92 u u u.uSS u.u2S S.94S u.SS8
27 1.u19 u.298 u.SS u u u S.8uS u.291
28 1.u17 2.S2S u u u u S.891 u.247
29 u.999 1.SS u u u.u24 u.uu9 S.92S u.S2S
Su u.987 2.428 u u u.1u6 u.u19 4.uu8 u.SS9
totals 1.924 u.997 1.892 1.u72 ieal powei loss u.uS2


Page
33
Table Su.11 Solution Infoimation fiom Lineai Appioximation 0sing u0R0BI 0bjective
value of S.88.
bus voltage geneiation loau Lambua($NvAhi)
no. mag(pu) ang(ueg) P(NW) Q(NvAR) P(NW) Q(NvAR) P Q
1 1.u8 u u.4 u.SS1 u u S.7S8 u.2u2
2 1.u68 u.726 u.S6 u.6 u.217 u.127 S.77S u.199
S 1.u29 1.947 u u u.u24 u.u12 S.8SS u.2u6
4 1.u18 2.S29 u u u.u76 u.u16 S.87S u.2u6
S 1.uS4 2.S4S u u u u S.84S u.2u2
6 1.uu7 2.77 u u u u S.89S u.2uS
7 1.uu9 S.191 u u u.228 u.1u9 S.91S u.2u6
8 u.99S S.28S u u u.S u.S S.921 u.22
9 u.98S 4.SS7 u u u u S.9u4 u.17S
1u u.97 S.189 u u u.uS8 u.u2 S.9u9 u.1S6
11 u.98S 4.SS7 u u u u S.9u4 u.17S
12 u.971 4.426 u u u.112 u.u7S S.9u9 u.24
1S u.968 2.774 u.18S u u u S.9 u.244
14 u.9S9 S.177 u u u.u62 u.u16 S.966 u.2S2
1S u.96 4.86S u u u.u82 u.u2S S.948 u.227
16 u.964 S.17 u u u.uSS u.u18 S.9S6 u.2uS
17 u.96S S.48S u u u.u9 u.uS8 S.9S4 u.17S
18 u.9S1 S.984 u u u.uS2 u.uu9 S.994 u.19S
19 u.949 6.S74 u u u.u9S u.uS4 4.uuS u.179
2u u.9S4 6.149 u u u.u22 u.uu7 S.98S u.17S
21 u.967 S.u79 u u u.17S u.112 S.89 u.1u8
22 u.97 4.8S1 u.2S u u u S.869 u.u9
2S u.96S S.29 u.18 u u.uS2 u.u16 S.887 u.241
24 u.9S7 4.u12 u u u.u87 u.u67 S.916 u.198
2S u.96S 2.1u9 u u u u S.87S u.26S
26 u.946 2.749 u u u.uSS u.u2S S.9S7 u.284
27 u.979 u.S62 u.SS u u u S.814 u.291
28 1.uuS 2.612 u u u u S.884 u.219
29 u.9S9 2.16S u u u.u24 u.uu9 S.927 u.281
Su u.948 S.276 u u u.1u6 u.u19 4.uu1 u.26S
totals 1.9uS u.9S1 1.892 1.u72 ieal powei loss u.u11


Page
34
Incieasing Bemanu. A stanuaiu test foi iobustness is incieasing uemanu. We solve the Su
bus pioblem with uemanu multiplieu by 1.2, 1.4 anu 1.6. The iesults aie in Table Su.12. The
pioblems iequiie incieasing CP0 time. Foi 0PFlin, C0NvRu is the numbei of lineai
piogiams solveu ovei the ten staiting points.

Table Su.12. Solution Statistics foi SuBus Pioblem with Incieaseu Bemanu
0sing C0N0PT anu u0R0BI with Relaxation Slack vaiiables.
type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN
0PF2 S.74S u.uuu u.1uS u.u87 u.141 1u u
0PF S.918 u.uuu u.u76 u.1S4 u.14u 1u u
0PFlin S.8u1 u.uuS u.u21 u.1u4 u.uu9 29 u
0PFp S.918 u.uuu u.uS8 u.197 u.1u9 1u u
BTBETA S.6S9 u.uuu u.uu4 u.u81 u.uu6 1u
1.2uu*uemanu
0PF2 7.274 u.uuu u.1uS u.1uS u.171 1u u
0PF 7.Su4 u.uuu u.u46 u.111 u.u78 1u u
0PFlin 7.S66 u.u1u u.u27 u.289 u.u1u S4 u
0PFp 7.Su4 u.uuu u.u41 u.224 u.u9S 1u u
BTBETA 7.1SS u.uuu u.uuS u.u46 u.uu4 1u
RatioNlLp 1.7u1
1.4uu*uemanu
0PF2 8.912 u.uuu u.112 u.u76 u.1S6 1u u
0PF 9.2u8 u.uuu u.uS8 u.1S6 u.1u9 1u u
0PFlin 9.uSu u.u1S u.uSu u.286 u.u1u S8 u
0PFp 9.2u8 u.uuu u.u64 u.126 u.u94 9 1
BTBETA 8.7u6 u.uuu u.uu4 u.u67 u.uu6 1u
1.6uu*uemanu
0PF2 1u.719 u.uuu u.u97 u.11S u.14u 9 1
0PF 11.u8S u.uuu u.uS8 u.19S u.14u 1u u
0PF lin 1u.8S1 u.u1S u.uSu u.21S u.u1u S6 u
0PFp 11.u8S u.uuu u.u7S u.12S u.1u9 1u u
BTBETA 1u.41S u.uuu u.uu4 u.u71 u.uu6 1u

Page
35
S7bus Pioblem
The iesults with uiffeient nonlineai solveis aie in Table S7.1. Without ielaxation
slack vaiiables, all solveis except KNITR0 finu the global optimal value foi all staiting
points. SN0PT, C0N0PT anu NIN0S aie the fastest nonlineai solveis. u0R0BI is fastei on
0PF than any nonlineai solvei.
Table S7.1. Solution Statistics foi the S7Bus Pioblem without Relaxation Slack vaiiables.
type optvavg optvcv cpuavg cpucv cpumax C0NvRu Tepsin
0PF2
SN0PT 417.S78 u.uuu u.S74 u.u64 u.Suu 1u u
C0N0PT 417.S78 u.uuu u.4S1 u.u6u u.S92 1u u
NIN0S 417.S78 u.uuu u.484 u.242 1.498 1u u
IP0PT 417.S78 u.uuu u.9S1 u.161 1.887 1u u
KNITR0 42u.S74 u.uuS 1.789 u.4u4 8.222 1u u
0PF
C0N0PT 42S.7Su u.uuu u.242 u.1u8 u.S9u 1u u
SN0PT 42S.7Su u.uuu u.S1S u.u72 u.4u6 1u u
NIN0S 42S.7Su u.uuu u.4S4 u.212 1.u92 1u u
IP0PT 42S.7Su u.uuu 1.878 u.24S 6.u69 1u u
KNITR0 4Su.9S8 u.uuS 2.44u u.12S 4.82u 1u u
0PFp
SN0PT 42S.7Su u.uuu u.1u6 u.u28 u.12S 1u u
NIN0S 42S.7Su u.uuu u.118 u.uS8 u.14u 1u u
C0N0PT 42S.7Su u.uuu u.242 u.1u8 u.S9u 1u u
IP0PT 42S.7Su u.uuu u.777 u.uS6 u.9uS 1u u
KNITR0 426.248 u.uu4 1.718 u.18S S.744 1u u
Btheta
u0R0BI 41u.S82 u.uuu u.u19 u.u42 u.u24 1u u
u0R0BI 41u.S82 u.uuu u.u2u u.uS4 u.u27 1u u
CPLEX 41u.S82 u.uuu u.u4S u.uSu u.uSS 1u u
CPLEX 41u.S82 u.uuu u.u44 u.u42 u.uS7 1u u
CPLEX 41u.S82 u.uuu u.u44 u.u2S u.u48 1u u
0PFlin
u0R0BI 42u.u21 u.uuS u.19S u.S2S u.uS4 S8 u
u0R0BI 42u.u21 u.uuS u.196 u.S17 u.uS6 S8 u
CPLEX 42u.S77 u.uu2 u.2uS u.262 u.u4S S2 u
CPLEX 42u.S77 u.uu2 u.2u8 u.277 u.u4u S2 u
CPLEX 42u.S77 u.uu2 u.21S u.292 u.u61 S2 u

With ielaxation slack vaiiables, only C0N0PT anu IP0PT finu the global optimal
value. The othei solveis finu optimal solutions neai the global optimal value. SN0PT is the
slowest solvei anu hau uifficulty on some staiting points. C0N0PT outpeifoims the othei
nonlineai solveis on the 0PF2 anu 0PF but C0N0PT fails to conveige on one pioblem, but
has highei optimal values. KNITR0 is veiy eiiatic at times woiking well anu at times
appeaiing to get hung up.


Page
36

Table S7.1i. Solution Statistics foi the S7Bus Pioblem with Relaxation Slack vaiiables.
type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN NIN0PT
0PF2
C0N0PT 4S6.718 u.uS4 u.817 u.1S 1.279 9 2 417.S78
KNITR0 419.1S u.uu1 1.S21 u.149 2.886 1u u 417.849
IP0PT 417.S78 u 2.861 u.u8S 4.18 1u u 417.S78
NIN0S 42u.S22 u.uu1 2.961 u.61 17.176 1u u 418.442
SN0PT 42u.S22 u.uu1 S.749 u.6u8 SS.u1 1u u 418.442
0PF
C0N0PT 42S.7S u u.44S u.122 u.671 1u u 42S.7S
KNITR0 61u.4S9 u.242 2.u12 u.2u8 4.462 1u u 424.827
IP0PT 42S.7S8 u 2.718 u.u7S 4.1S 1u u 42S.7S
NIN0S 428.uSS u.uu1 1uu.128 u.947 999.998 9 u 426.227
SN0PT 428.12 u.uu1 1uu.268 u.946 999.998 9 u 426.2uS
0PFp
NIN0S 42S.uu4 u u.u64 u.171 u.141 1u u 424.249
C0N0PT 42S.7S u u.226 u.u87 u.S4S 1u u 42S.7S
KNITR0 42S.211 u u.8S u.uS 1.u6 1u u 424.824
IP0PT 42S.7S8 u 2.S48 u.178 6.u22 1u u 42S.7S
SN0PT 42S.222 u.uu1 44.SS7 u.947 444.712 1u u 42S.96S

Foi lineai solveis, CPLEX times bounu the u0R0BI times see Table S7.2. The
optimal solution values aie within 1 peicent of the best nonlineai solveis. Both lineai
solveis aie fastei anu moie iobust than the nonlineai solveis.

Table S7.2. Solution Statistics foi the S7Bus Pioblem with the Lineai Solveis.
type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvERu TEPSIN
0PFlin
CPLEX 42u.S88 u.uuS u.1SS u.22S u.u4S S6 u
u0R0BI 42u.878 u.uuS u.278 u.171 u.u46 Su u
u0R0BI 42u.878 u.uuS u.28u u.18S u.u44 Su u
u0R0BI 42u.878 u.uuS u.288 u.171 u.u4S Su u
CPLEX 42u.S88 u.uuS u.S84 u.2S7 u.1u2 S6 u
BTBETA
CPLEX 41u.S82 u.uuu u.u19 u.uS1 u.u24 1u u
u0R0BI 41u.S82 u.uuu u.u2u u.uS2 u.u2S 1u u
u0R0BI 41u.S82 u.uuu u.u21 u.u4S u.u27 1u u
u0R0BI 41u.S82 u.uuu u.u24 u.u6u u.uSu 1u u
CPLEX 41u.S82 u.uuu u.u47 u.u49 u.u6S 1u u

Table S7.1a. Lineai Appioximation Nonzeios, Iteiations anu 0ptimal value By LP Pass anu
Staiting Point foi the S7Bus 0sing u0R0BI With Anu Without Relaxation Slack vaiiables.
Stait 1 2 S 4 S 6 7 8 9 1u Total
Without S 6 S S 7 6 9 6 S 6 S8
With S 6 S S S S 6 S S S Su

Page
37
Table S7.S iepoits the test iesults iun at a uiffeient uay. The only notable uiffeience
is in the iepoiteu times.
Table S7.S. Solution Statistics foi the S7Bus Pioblem with Biffeient Solveis at Anothei
Time Without Relaxation Slack vaiiables.
type optvavg optvcv cpuavg cpucv cpumax C0NvRu Tepsin
0PF2
SN0PT 417.S78 u.uuu u.S74 u.u64 u.Suu 1u u
C0N0PT 417.S78 u.uuu u.4S1 u.u6u u.S92 1u u
NIN0S 417.S78 u.uuu u.484 u.242 1.498 1u u
IP0PT 417.S78 u.uuu u.9S1 u.161 1.887 1u u
KNITR0 42u.S74 u.uuS 1.789 u.4u4 8.222 1u u
0PF
C0N0PT 42S.7Su u.uuu u.242 u.1u8 u.S9u 1u u
SN0PT 42S.7Su u.uuu u.S1S u.u72 u.4u6 1u u
NIN0S 42S.7Su u.uuu u.4S4 u.212 1.u92 1u u
IP0PT 42S.7Su u.uuu 1.878 u.24S 6.u69 1u u
KNITR0 4Su.9S8 u.uuS 2.44u u.12S 4.82u 1u u
0PFp
SN0PT 42S.7Su u.uuu u.1u6 u.u28 u.12S 1u u
NIN0S 42S.7Su u.uuu u.118 u.uS8 u.14u 1u u
C0N0PT 42S.7Su u.uuu u.242 u.1u8 u.S9u 1u u
IP0PT 42S.7Su u.uuu u.777 u.uS6 u.9uS 1u u
KNITR0 426.248 u.uu4 1.718 u.18S S.744 1u u
Btheta
u0R0BI 41u.S82 u.uuu u.u19 u.u42 u.u24 1u u
u0R0BI 41u.S82 u.uuu u.u2u u.uS4 u.u27 1u u
CPLEX 41u.S82 u.uuu u.u4S u.uSu u.uSS 1u u
CPLEX 41u.S82 u.uuu u.u44 u.u42 u.uS7 1u u
CPLEX 41u.S82 u.uuu u.u44 u.u2S u.u48 1u u
0PFlin
u0R0BI 42u.u21 u.uuS u.19S u.S2S u.uS4 S8 u
u0R0BI 42u.u21 u.uuS u.196 u.S17 u.uS6 S8 u
CPLEX 42u.S77 u.uu2 u.2uS u.262 u.u4S S2 u
CPLEX 42u.S77 u.uu2 u.2u8 u.277 u.u4u S2 u
CPLEX 42u.S77 u.uu2 u.21S u.292 u.u61 S2 u

C0N0PT. The 0PF2 foimulation hau 8S6 iows, 97u columns, S,S87 nonzeioes, anu S77
nonconstant fiist ueiivatives with 1,S7u evaluation opeiations occupying 4 Nb. C0N0PT
has u pietiiangulai equations anu 7S posttiiangulai equations. Foi the flat stait with
ielaxation slack vaiiables, at iteiation 42 C0N0PT founu a feasible solution anu at iteiation
166 founu an optimal solution with an optimal value of 4.174E2 with a ieuuceu giauient of
1.4Eu8. With ielaxation slack vaiiables foi staiting point 2, at iteiation S4S C0N0PT founu
an optimal solution value of 417.4 .
The 0PF anu 0PFp foimulations hau 91S iows, 1,S4u columns, 4,SuS nonzeioes,
anu S7u nonconstant fiist ueiivatives with 1,S11 function evaluations. C0N0PT founu Su
pietiiangulai equations anu 1S8 posttiiangulai equations. At iteiation 1S, C0N0PT founu
a feasible solution anu at 44 an optimal solution of 4.2S7E2. At staiting point 2, C0N0PT

Page
38
founu a feasible solution at iteiation 1uu anu at 1S1 founu an optimal solution of 4.2S7E2
foi 0PFp.
Without ielaxation slack vaiiables foi 0PF2, C0N0PT peifoims bettei on the flat
stait than any ianuom staiting point. Foi the 0PF, C0N0PT peifoims bettei on the flat stait
than most ianuom staiting points. Without ielaxation vaiiables, C0N0PT finus the optimal
values of 417.S78 foi 0PF2 anu 42S.7Su foi 0PF anu 0PFp. 0n thiee points, C0N0PT got
hung up on 0PF2.

Table S7.4. C0N0PT Peifoimance on the S7Bus without Relaxation Slack vaiiables.
0PF2 0PF 0PFp
stait without with without with without with
point iteis iteis optval optml iteis iteis optval iteis iteis optval
1 81 166 4.17E2 417.4 48 44 42S.8 77 Su 42S.8
2 2S9 S4S 4.17E2 417.4 1SS 1S1 42S.8 S4 S8 42S.8
S 186 11u 1.S6E8 S21.S 267 7S 42S.8 S6 SS 42S.8
4 222 262 4.17E2 417.4 64 S9 42S.8 99 84 42S.8
S 212 S62 4.17E2 417.4 9S 2S8 42S.8 SS 128 42S.8
6 2u6 116 4.17E2 417.4 46 S4S 42S.8 29 46 42S.8
7 14S SS9 4.17E2 417.4 1S1 144 42S.8 42 1u8 42S.8
8 182 4Su 1.S7E8 S94.4 189 171 42S.8 S4 1S6 42S.8
9 1S1 SuS 1.S7E8 S94.4 S6 1SS 42S.8 S7 1u4 42S.8
1u 1S2 48S 4.17E2 417.4 7u 9S 42S.8 S4 64 42S.8
avg 177 SS4 4.1uE7 46S.2 114 149 42S.8 S2 8S 42S.8

KNITR0. Without ielaxation slack vaiiables, KNITR0 finus local optima neai the global
optimal with the exception of the 0PF pioblem. With ielaxation slack vaiiables, KNITR0
finus local optima neai the global optimal.
Table S7.4. KNITR0 Peifoimance on the S7Bus With anu Without Relaxation Slack
vaiiables.
0PF2 0PF2 0PF 0PF 0PFp 0PFp
stait without with without with without with
point optval optval optm optval optval optm optval optval optm
1 417.S 418.2 418.2 429.1 424.8 424.8 424.u 426.1 426.1
2 421.u 421.4 421.4 4S7.9 474.2 474.2 42S.9 42S.9 42S.9
S 42S.9 419.8 419.8 4SS.6 4S8.6 4S8.6 42S.9 426.S 426.S
4 417.4 418.7 418.7 4Su.1 4S4.S 4S4.S 42S.9 424.8 424.8
S 424.4 42u.8 42u.8 426.S 424.8 424.8 424.u 424.8 424.8
6 498.u 418.2 418.2 4S4.6 2uu9.2 2uu1.1 4Su.u 424.8 424.8
7 4S9.6 418.2 418.2 424.u 424.8 424.8 42S.9 424.8 424.8
8 417.4 42u.2 42u.2 4SS.6 424.9 424.9 42S.9 424.8 424.8
9 421.6 481.7 417.8 424.7 424.8 424.8 44S.2 424.8 424.8
1u 417.4 418.2 418.2 4S2.8 8S7.S 611.7 424.u 424.8 424.8
avg 429.8 42S.S 419.2 4S1.1 6SS.8 61u.4 426.7 42S.2 42S.2

SN0PT. Without ielaxation slack vaiiables, SN0PT peifoims bettei on the flat stait than any
ianuom staiting point foi 0PF2 anu 0PF. With ielaxation slack vaiiables, SN0PT peifoims

Page
39
bettei on the flat stait than any ianuom staiting point foi 0PF2, 0PF anu 0PFp. Without
ielaxation vaiiables, SN0PT founu the optimal solutions of 417.4 foi 0PF2 anu 42S.8 foi
0PF anu 0PFp.

Table S7.6 SN0PT Peifoimance on the S7Bus With anu Without Relaxation Slack vaiiables.
0PF2 0PF 0PFp
stait without with without with without with
point iteis iteis optval iteis iteis optval iteis iteis optval
1 6S8 729 421.6 298 4Su 426.2 268 1S6 424.S
2 1u29 89u 419.S 8S9 728 427.8 28S 28S 42S.u
S 14uS 1u72 418.4 971 9S7 427.9 2SS SSS 426.2
4 11S2 649u 419.7 617 777 426.6 SuS 222 42S.6
S 1287 11u7 42S.2 82S 12SS 4Su.9 2S8 1S79 42S.4
6 1uSS 114u 419.8 866 1uu4 427.S 24u 226 424.7
7 11S7 1u81 419.9 97u 896 428.6 27S 4u7 424.u
8 1u4u S182 419.7 961 824 4Su.1 2S7 2u8 42S.2
9 11u2 1419 42u.1 8S4 4S67 426.7 2SS 281 427.2
1u 11u7 1299 421.6 896 877 427.7 2S4 249 424.6
avg 1u9S 2u41 42u.S 812 1212 428.u 262 S7S 42S.2

NIN0S. Without ielaxation slack vaiiables, NIN0S always founu the optimal value anu
peifoim bettei on the flat stait than the aveiage ianuom staiting point foi 0PF2 anu 0PF.
With ielaxation slack vaiiables, NIN0S founu optimal values neai the the optimal value anu
peifoim bettei on the flat stait than the aveiage ianuom staiting point foi 0PF2 anu 0PF.
Without ielaxation vaiiables, NIN0S founu the optimal solutions of 417.4 foi 0PF2 anu
42S.8 foi 0PF anu 0PFp.

Table S7.7 NIN0S Peifoimance on the S7Bus With anu Without Relaxation Slack vaiiables.
0PF2 0PF 0PFp
stait without with without with without with
point iteis iteis optval iteis iteis optval iteis iteis optval
1 1S86 4217 422.9 41S SS7 426.9 SS6 26u9 428.2
2 1287 S19S 418.4 1829 92S 42S.9 S44 26u9 4Su.6
S 1uS2 969 422.6 1128 S972 428.4 S1S 49u 426.2
4 S7S8 1241 419.9 2178 2SS2 427.u S46 9u7 4S1.6
S 128S 2S99 42u.1 11uS 1286 4Su.S SS8 S49 424.9
6 1u8u S429 42u.8 8SS 8SSu 426.1 S72 S2u 427.7
7 9S2 772 418.8 1u8u S1S9 428.2 S28 2SSu 426.9
8 1492 11u6 418.S 2627 924 428.1 S88 1S12 429.S
9 2uSS 7u9 42u.u 1uuu 8S1 427.S S27 496 42S.9
1u 1u62 1499 419.8 14S7 2S78 428.S S48 S147 4Su.9
avg 1SSS 197S 42u.2 1S6S 2691 427.7 S48 1SS9 428.2
IP0PT. Without ielaxation slack vaiiables, IP0PT always founu the optimal value anu uoes
not peifoim bettei on the flat stait than the aveiage ianuom staiting point foi 0PF2 anu
0PF. With ielaxation slack vaiiables, IP0PT founu the optimal value anu peifoim bettei on
the flat stait than the aveiage ianuom staiting point foi 0PF2 anu 0PF. With anu without

Page
40
ielaxation vaiiables, IP0PT founu the optimal solutions of 417.4 foi 0PF2 anu 42S.8 foi 0PF
anu 0PFp.
Table S7.7 IP0PT Peifoimance on the S7Bus With anu Without Relaxation Slack vaiiables.
0PF2 0PF 0PFp
stait without with without with without with
point iteis iteis iteis iteis iteis optval iteis
1 SS 48 44 6S 21 S9
2 19 S7 4u 66 2S SS
S S8 7S 128 49 19 S7
4 S1 8u 41 S6 19 4S
S 19 97 28 6S 2u 46
6 26 41 S8 SS 22 S2
7 29 7S SS 4S 2S 1S4
8 21 S1 46 SS 2u 26
9 21 8S 2S 7u 2S 6u
1u 21 64 2S 71 2u 46
avg Su 67 4S 6u 21 S4


Page
41
u0R0BI. The BTBETA foimulation hau 82 iows, 4,489 columns, anu 9,272 nonzeioes
occupying S Nb. Piesolve time took u.u2s anu iesulteu in 8S iows, 176 columns, 26u
nonzeios. u0R0BI solveu in 46 iteiations anu u.u2 seconus objective of 4.1u6e 2. 0PFlin
hau 1,S98 iows, 1,S4u columns anu S,4S1 nonzeioes. Piesolve iesulteu in 612 iows, 7S1
columns, anu 27Su nonzeios. Foi 0PFlin, the piogiession of the lineai optimal values is not
monotonic because the pioblem has nonnonconvex constiaints, but the teiminal values aie
within a 1 peicent iange.

Table S7.8. Lineai Appioximation Nonzeios, Iteiations anu 0ptimal value By LP Pass anu
Staiting Point foi the S7Bus 0sing u0R0BI With Relaxation Slack vaiiables.
pass nonzeios iteis optval nonzeios itei optval nonzeios iteis optval
staitpoint 1 2 S
1 S1u9 41u 41S.92 S187 Su7 416.87 S187 287 417.u2
2 S41S S84 421.68 S417 S6u 416.7S S417 S14 417.2S
S SS27 S1S 421.9S SSS1 S78 416.21 SS27 S94 416.28
4 S64S SS9 41S.uS S6S9 4u8 416.61
S S7S9 S19 422.17 S7S1 S4u 417.74
6 S87S S42 421.S8
staitpoint 4 staitpoint S staitpoint 6
1 S187 292 416.67 S187 282 416.8S S187 286 416.89
2 S417 SS2 414.8S S417 SSS 417.S4 S417 S2u 417.14
S SSS1 Su7 414.S6 SS27 S22 417.S9 SS27 S78 416.42
4 S64S Su4 42u.74 S6S9 S97 414.98 S6S9 S61 41S.44
S S7S9 SuS 421.1u S7SS 424 422.42 S7S1 S4S 421.SS
staitpoint 7 staitpoint 8 staitpoint 9
1 S187 279 417.S9 S187 29S 416.84 S187 SuS 417.S7
2 S417 S2u 417.68 S417 S22 417.24 S417 Su4 416.S9
S SS27 S96 417.S4 SS27 SSu 416.S6 SS27 SS4 41S.1S
4 S6S9 S47 41S.u9 S6S9 S9S 41S.1S S6S9 S6S 418.27
S S7S1 S44 422.2S S7S1 4u2 421.S2 S7S1 S8S 42u.9S
6 S86S S46 42u.6S
staitpoint 1u
1 S187 S1S 417.2S
2 S417 S1S 416.76
S SS27 S76 41S.6u
4 S6S9 SS7 41S.14
S S7S1 S64 419.78


Page
42
Incieasing Bemanu. A stanuaiu test foi iobustness is incieasing uemanu. The maximum
possible uemanu inciease is 2.98. We solve the S7bus pioblem with uemanu incieaseu by
1.S, 2 anu 2.S. The iesults aie in table S7.9.
Table S7.9. Solution Statistics foi S7Bus Pioblem with Incieaseu Bemanu 0sing C0N0PT
Anu u0R0BI .
type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN
0PF2 4S6.718 u.uS4 u.SSu u.1S6 1.u4S 9 2
0PF 42S.7Su u.uuu u.267 u.146 u.484 1u u
0PFlin 417.6S8 u.uu8 u.u86 u.249 u.uS1 S9 u
0PFp 42S.7Su u.uuu u.1S6 u.1S2 u.2S4 1u u
BTBETA 41u.S82 u.uuu u.uu9 u.u2S u.u1u 1u
1.2*uemanu
0PF2 S6S.941 u.uuu u.479 u.124 u.811 9 1
0PF S7S.S47 u.uuu u.Su4 u.14S u.6u8 1u u
0PFlin SS2.646 u.uu8 u.1uS u.SS4 u.u16 4S u
0PFp S7S.S47 u.uuu u.187 u.u86 u.266 1u u
BTBETA S19.4Su u.uuu u.uu9 u.uS7 u.u11 1u
1.4*uemanu
0PF2 741.424 u.uuu u.S82 u.16u u.749 8 2
0PF 7SS.168 u.uuu u.248 u.122 u.421 1u u
0PFlin 7uS.462 u.u14 u.u77 u.2S9 u.uS2 S6 u
0PFp 7SS.168 u.uuu u.1SS u.1S4 u.2S4 1u u
BTBETA 678.164 u.uuu u.uu9 u.uS4 u.u11 1u
1.6*uemanu
0PF2 88S.914 u.uuu u.627 u.1S2 1.2S2 9 1
0PF 899.774 u.uuu u.S6S u.1Su u.718 1u u
0PFlin 899.SS9 u.uuu u.u97 u.1S1 u.u24 S8 u
0PFp 899.774 u.uuu u.18S u.u68 u.2Su 1u u
BTBETA 88S.9u7 u.uuu u.uu8 u.uSS u.u11 1u

Compaiing the 0PF to the lineai appioximation. We compaie the C0N0PT 0PF solution to
the u0R0BI 0PFlin solution anu finu only minoi uiffeiences in P, Q anu v anu noual piices
foi ieal anu ieactive powei. The voltages anu uual vaiiables aie similai but the actual
geneiatoi uispatch is not as close, but the objective function value uiffeis by less than one
peicent.

Table S7.1u Solution Infoimation fiom C0N0PT with 0bjective value 0f 42S.7.
bus voltage geneiation loau Lambua Bus flow
no. Nag
(pu)
Ang
(ueg)
P
(NW)
Q
(NvAR)
P
(NW)
Q
(NvAR)
P Q Pflow Qflow
1 1.u12 u.uuu 1.4S9 u.884 u.SSu u.17u 42.SS8 2.uu8 1.SS7 u.794
2 u.999 1.u29 u.91S u.uuu u.uSu u.88u 41.9uu 2.SS8 u.88S u.827
S 1.uuS 1.286 u.42u u.41S u.41u u.21u 42.899 2.u2S u.479 u.291
4 1.uuS 1.182 u.uuu u.uuu u.uuu u.uuu 42.872 2.u8S u.S86 u.u81
S 1.u1S u.14S u.uuu u.uuu u.1Su u.u4u 42.S2u 2.16S u.268 u.uSu
6 1.u22 u.777 u.8u6 u.uuu u.7Su u.u2u 41.7uu 2.129 u.S22 u.u72
7 1.u2S 1.S11 u.uuu u.uuu u.uuu u.uuu 41.Su7 2.26S u.77S u.27S

Page
43
8 1.u4S 4.17S 4.4uu u.964 1.Suu u.22u 4u.SS2 2.uu2 2.9uu u.777
9 1.uu2 u.SS7 1.uuu u.uuu 1.21u u.26u 42.Su6 2.S4S 1.711 u.684
1u u.984 S.666 u.uuu u.uuu u.uSu u.u2u 4S.741 2.Su7 u.S88 u.SS1
11 u.98S 2.4u6 u.uuu u.uuu u.uuu u.uuu 4S.47S 2.4S1 u.44S u.18u
12 u.992 S.44S S.69u u.427 S.77u u.24u 4S.769 2.uuS u.2SS u.262
1S u.977 S.242 u.uuu u.uuu u.18u u.u2S 4S.889 2.2u9 u.682 u.42S
14 u.97u S.6uS u.uuu u.uuu u.1uS u.uSS 4S.971 1.966 u.SS4 u.8S8
1S u.988 2.628 u.uuu u.uuu u.22u u.uSu 4S.48S 2.u66 1.uu6 u.SS2
16 u.992 S.914 u.uuu u.uuu u.4Su u.uSu 4S.9SS 2.uS6 u.4Su u.u29
17 u.994 2.868 u.uuu u.uuu u.42u u.u8u 4S.SS4 2.122 u.47S u.u61
18 1.u2S S.421 u.uuu u.uuu u.272 u.u98 4S.22S 2.166 u.S2S u.1u9
19 u.988 6.794 u.uuu u.uuu u.uSS u.uu6 4S.S6S 2.81S u.u49 u.u21
2u u.976 6.968 u.uuu u.uuu u.u2S u.u1u 4S.799 S.u2u u.u2S u.uSS
21 1.u1S 6.646 u.uuu u.uuu u.uuu u.uuu 4S.897 2.72S u.uu7 u.u26
22 1.u1S 6.S84 u.uuu u.uuu u.uuu u.uuu 4S.8S8 2.699 u.uS9 u.u28
2S 1.u14 6.S96 u.uuu u.uuu u.u6S u.u21 4S.917 2.7S9 u.u6S u.uSu
24 1.u17 6.uS1 u.uuu u.uuu u.uuu u.uuu 46.219 S.141 u.176 u.44S
2S 1.uuu 1u.967 u.uuu u.uuu u.u6S u.uS2 46.871 S.S77 u.14S u.uS1
26 u.976 S.SSu u.uuu u.uuu u.uuu u.uuu 46.2Su S.21u u.176 u.41S
27 1.u12 S.1S4 u.uuu u.uuu u.u9S u.uuS 4S.961 2.861 u.274 u.uS4
28 1.uS2 1.8S2 u.uuu u.uuu u.u46 u.u2S 42.628 2.644 u.S2S u.u64
29 1.u49 u.96S u.uuu u.uuu u.17u u.u26 41.6uu 2.S81 u.68u u.29u
Su u.979 11.S49 u.uuu u.uuu u.uS6 u.u18 48.u86 4.2SS u.u8u u.u46
S1 u.9Su 12.SS4 u.uuu u.uuu u.uS8 u.u29 49.8S6 S.2S7 u.uS8 u.u29
S2 u.96u 11.7u4 u.uuu u.uuu u.u16 u.uu8 48.91u 4.888 u.u69 u.u17
SS u.9S7 11.74S u.uuu u.uuu u.uS8 u.u19 49.u84 4.97S u.uS8 u.u19
S4 u.966 7.748 u.uuu u.uuu u.uuu u.uuu 48.S9S 4.S71 u.u69 u.u21
SS u.97S 7.S28 u.uuu u.uuu u.u6u u.uSu 47.97S 4.1S7 u.129 u.u6S
S6 u.982 7.271 u.uuu u.uuu u.uuu u.uuu 47.SSu S.816 u.1S4 u.1u4
S7 u.99u 7.116 u.uuu u.uuu u.uuu u.uuu 46.89S S.49u u.186 u.1SS
S8 1.u16 6.S4S u.uuu u.uuu u.14u u.u7u 4S.67u 2.61S u.S68 u.2S7
S9 u.988 7.142 u.uuu u.uuu u.uuu u.uuu 46.98S S.S47 u.uS1 u.u22
4u u.979 7.26S u.uuu u.uuu u.uuu u.uuu 47.S9S S.928 u.u24 u.u2S
41 1.uu6 6.S7u u.uuu u.uuu u.u6S u.uSu 4S.82S S.uSS u.227 u.u67
42 u.974 8.192 u.uuu u.uuu u.u71 u.u44 46.SuS S.798 u.u96 u.u44
4S 1.u19 S.64u u.uuu u.uuu u.u2u u.u1u 4S.S71 2.62S u.147 u.149
44 1.u19 6.u19 u.uuu u.uuu u.12u u.u18 4S.218 2.S64 u.271 u.u26
4S 1.uSS 4.16u u.uuu u.uuu u.uuu u.uuu 4S.SSS 2.Su6 u.27S u.244
46 1.u6u S.186 u.uuu u.uuu u.uuu u.uuu 4S.982 u.918 u.429 u.8Su
47 1.uS4 6.4u1 u.uuu u.uuu u.297 u.116 44.87S 1.72S u.424 u.24S
48 1.u29 6.4SS u.uuu u.uuu u.uuu u.uuu 4S.127 2.u4u u.1S6 u.176
49 1.uS8 6.4S9 u.uuu u.uuu u.18u u.u8S 44.8Su 2.u86 u.SS1 u.SSS
Su 1.u24 6.496 u.uuu u.uuu u.21u u.1uS 4S.444 2.678 u.21u u.1uS
S1 1.uS2 4.9uS u.uuu u.uuu u.18u u.uSS 4S.779 2.29S u.SS8 u.S8S
S2 1.u17 2.48S u.uuu u.uuu u.u49 u.u22 44.u88 S.u77 u.176 u.uS2
SS 1.uu7 S.142 u.uuu u.uuu u.2uu u.1uu 4S.118 S.2S4 u.2uu u.u68
S4 1.u27 2.478 u.uuu u.uuu u.u41 u.u14 4S.988 2.89S u.116 u.u4S

Page
44
SS 1.uS7 1.482 u.uuu u.uuu u.u68 u.uS4 42.S8u 2.SSu u.187 u.296
S6 u.974 8.91S u.uuu u.uuu u.u76 u.u22 47.26S S.4S7 u.11S u.u29
S7 u.969 9.S69 u.uuu u.uuu u.u67 u.u2u 47.9uS S.476 u.u67 u.u2u
totals 12.67u 2.688 12.Su8 S.S64 iealpoweiloss u.162

Table S7.11. Solution Infoimation fiom Lineai Appioximation 0sing u0R0BI
0bjective value of 421.9 anu Thiiu Pass
bus voltage geneiation loau Lambua Bus flow
no. Nag
(pu)
Ang
(ueg)
P
(NW)
Q
(NvAR)
P
(NW)
Q
(NvAR)
P Q Pflow Qflow
1 1.uSS u.uuu 1.1S2 1.192 u.SSu u.17u 41.974 2.uu8 1.11S 1.17S
2 1.u12 1.12S u.8uu u.uuu u.uSu u.88u 41.S29 2.466 u.786 u.94S
S u.992 u.27u u.42u u.uuu u.41u u.21u 42.74S 1.486 u.497 u.S17
4 u.988 u.2S7 u.uuu u.uuu u.uuu u.uuu 42.86S 1.S92 u.S1S u.2u8
S u.987 2.uu1 u.uuu u.uuu u.1Su u.u4u 42.S82 1.246 u.S14 u.1u2
6 u.992 S.Su8 1.uuu u.uuu u.7Su u.u2u 42.u91 1.122 u.67S u.146
7 u.98S S.879 u.uuu u.uuu u.uuu u.uuu 41.782 1.2S6 u.762 u.266
8 u.997 7.uS7 4.S77 u.uuu 1.Suu u.22u 4u.778 1.u16 S.u47 u.17S
9 u.986 1.474 1.uuu u.uuu 1.21u u.26u 42.466 1.6u4 1.841 u.72S
1u u.997 2.SS2 u.uuu u.uuu u.uSu u.u2u 4S.816 1.846 u.414 u.7SS
11 u.98u 1.u6S u.uuu u.uuu u.uuu u.uuu 4S.S7S 1.86S u.477 u.2S9
12 1.uS1 2.6S4 S.69u 1.S99 S.77u u.24u 4S.8u9 2.uuS u.289 1.241
1S u.987 2.2S2 u.uuu u.uuu u.18u u.u2S 4S.9Su 1.8u8 u.7S6 u.78u
14 u.977 2.698 u.uuu u.uuu u.1uS u.uSS 4S.98S 1.SS7 u.S48 u.8SS
1S u.99S 1.8SS u.uuu u.uuu u.22u u.uSu 4S.S77 1.699 u.898 u.617
16 1.u26 S.287 u.uuu u.uuu u.4Su u.uSu 4S.919 2.u19 u.4S4 u.uSu
17 1.u2S 2.S27 u.uuu u.uuu u.42u u.u8u 4S.S6u 2.11S u.426 u.u67
18 1.uu9 4.2u2 u.uuu u.uuu u.272 u.u98 4S.119 1.421 u.S24 u.1u6
19 u.977 S.78S u.uuu u.uuu u.uSS u.uu6 4S.S82 2.u6S u.u48 u.u1S
2u u.969 6.uS2 u.uuu u.uuu u.u2S u.u1u 4S.867 2.S17 u.u2S u.uSS
21 1.u14 S.6S9 u.uuu u.uuu u.uuu u.uuu 4S.971 2.u26 u.uu9 u.uSu
22 1.u1S S.S9S u.uuu u.uuu u.uuu u.uuu 4S.916 2.u28 u.u49 u.u79
2S 1.u14 S.S87 u.uuu u.uuu u.u6S u.u21 46.uu2 2.uS9 u.u64 u.u77
24 1.uuS 4.684 u.uuu u.uuu u.uuu u.uuu 46.S82 2.u14 u.17S u.484
2S u.988 9.822 u.uuu u.uuu u.u6S u.uS2 46.87u 2.S1S u.147 u.u47
26 u.96u 4.17S u.uuu u.uuu u.uuu u.uuu 46.4S9 2.1u8 u.174 u.4S4
27 u.98S 1.2uS u.uuu u.uuu u.u9S u.uuS 44.166 1.74u u.27S u.u21
28 u.999 u.S76 u.uuu u.uuu u.u46 u.u2S 42.8S7 1.S42 u.S24 u.u2S
29 1.u14 1.42u u.uuu u.uuu u.17u u.u26 41.8uS 1.29S u.68S u.27S
Su u.968 1u.471 u.uuu u.uuu u.uS6 u.u18 48.144 2.9u8 u.u81 u.u42
S1 u.941 11.S84 u.uuu u.uuu u.uS8 u.u29 49.982 S.817 u.uS9 u.u29
S2 u.9S4 1u.8S2 u.uuu u.uuu u.u16 u.uu8 49.uS4 S.791 u.u71 u.u22
SS u.9S2 1u.89S u.uuu u.uuu u.uS8 u.u19 49.219 S.86u u.uS8 u.u19
S4 u.966 6.7S2 u.uuu u.uuu u.uuu u.uuu 48.667 S.46S u.u71 u.u27
SS u.97S 6.SS9 u.uuu u.uuu u.u6u u.uSu 48.218 S.29S u.1S2 u.u69
S6 u.982 6.287 u.uuu u.uuu u.uuu u.uuu 47.SSS S.u4u u.1S7 u.11S
S7 u.991 6.141 u.uuu u.uuu u.uuu u.uuu 47.uS9 2.76u u.188 u.146

Page
45
S8 1.u18 S.6uu u.uuu u.uuu u.14u u.u7u 4S.7SS 2.uu6 u.S82 u.S1S
S9 u.989 6.164 u.uuu u.uuu u.uuu u.uuu 47.1S6 2.8u9 u.uSu u.u2S
4u u.98u 6.27S u.uuu u.uuu u.uuu u.uuu 47.6u9 S.149 u.u2S u.u2S
41 1.uu4 S.S79 u.uuu u.uuu u.u6S u.uSu 4S.862 2.4SS u.2S4 u.u68
42 u.972 7.u8S u.uuu u.uuu u.u71 u.u44 46.444 S.u1S u.u99 u.u44
4S 1.u17 2.S47 u.uuu u.uuu u.u2u u.u1u 4S.647 2.u18 u.1S2 u.149
44 1.u22 S.12u u.uuu u.uuu u.12u u.u18 4S.2SS 2.u24 u.264 u.u17
4S 1.uS9 S.SSS u.uuu u.uuu u.uuu u.uuu 4S.4S4 1.9SS u.27u u.247
46 1.u66 4.S14 u.uuu u.uuu u.uuu u.uuu 4S.9S7 u.491 u.44u u.867
47 1.uS9 S.Su9 u.uuu u.uuu u.297 u.116 44.891 1.21S u.428 u.268
48 1.uSS S.S22 u.uuu u.uuu u.uuu u.uuu 4S.1S8 1.Su1 u.169 u.21S
49 1.u44 S.S1S u.uuu u.uuu u.18u u.u8S 44.86u 1.S76 u.S42 u.SS9
Su 1.uSS S.S1u u.uuu u.uuu u.21u u.1uS 4S.S41 2.12u u.212 u.1u7
S1 1.u66 S.8S7 u.uuu u.uuu u.18u u.uSS 4S.81S 1.761 u.S61 u.6uS
S2 u.984 u.S6u u.uuu u.uuu u.u49 u.u22 44.S2S 1.964 u.177 u.u22
SS u.97S 1.142 u.uuu u.uuu u.2uu u.1uu 4S.S6S 2.1S8 u.2u1 u.u78
S4 1.uu2 u.6S4 u.uuu u.uuu u.u41 u.u14 44.181 1.968 u.119 u.u64
SS 1.uS7 u.246 u.uuu u.uuu u.u68 u.uS4 42.496 1.S78 u.192 u.286
S6 u.972 7.872 u.uuu u.uuu u.u76 u.u22 47.S8S 2.S99 u.11S u.uS2
S7 u.967 8.S7S u.uuu u.uuu u.u67 u.u2u 48.u27 2.S7S u.u68 u.u22
totals 12.6S9 2.S91 12.Su8 S.S64 iealpoweiloss u.1S1


Page
46
118bus Pioblem
Without ielaxation slack vaiiables, all solveis except KNITR0 finu the global optimal
value foi all staiting points see Table 118.1. SN0PT, C0N0PT anu NIN0S aie the fastest
solveis. CPLEX anu u0R0BI aie fastei on 0PF than any nonlineai solvei but NIN0S. 0PFp
uiu not solve fastei than the 0PF.
Table 118.1. Solution statistics foi the 118bus pioblem with uiffeient nonlineai solveis
without ielaxation slack vaiiables.
optvavg optvcv cpuavg cpucv cpumax C0NvRu Tepsin
0PF2
SN0PT 1297 u 2.9u u.u6 S.9 1u u
NIN0S 1297 u S.u6 u.1u S.u 1u u
IP0PT 1297 u S.19 u.14 6.S 1u u
C0N0PT 1297 u S.19 u.u9 8.9 1u u
KNITR0 1298 u.uu1 17S.97 u.S2 7SS.1 8 u
0PF
NIN0S 1S11 u u.87 u.1S 1.8 1u u
SN0PT 1S11 u 1.S2 u.uS 1.8 1u u
C0N0PT 1S11 u 1.7S u.uS 2.2 1u u
IP0PT 1S11 u 4.6S u.12 7.7 1u u
KNITR0 1S1S u.uu1 2u2.69 u.62 1uuu.1 8 2
0PFp
NIN0S 1S11 u 1.u9 u.SS S.S 9 u
C0N0PT 1S11 u 1.21 u.11 2.1 9 u
IP0PT 1S11 u S.22 u.S1 12.S 9 u
SN0PT 1S11 u S.47 u.S8 12.7 7 u
KNITR0 1S19 u.uu2 277.41 u.48 1uuu.1 7 S
0PFlin
CPLEX 129S u.uu6 u.91 u.S2 u.1 82 u
CPLEX 129S u.uu6 u.92 u.SS u.1 82 u
CPLEX 129S u.uu6 u.9S u.SS u.1 82 u
u0R0BI 1292 u.uu7 1.1S u.28 u.2 89 u
u0R0BI 1292 u.uu7 1.16 u.Su u.1 89 u
Btheta
u0R0BI 1261 u u.u8 u.u4 u.1 1u u
u0R0BI 1261 u u.u8 u.uS u.1 1u u
CPLEX 1261 u u.14 u.u6 u.2 1u u
CPLEX 1261 u u.14 u.u4 u.2 1u u
CPLEX 1261 u u.14 u.u2 u.2 1u u

Without ielaxation slack vaiiables, the iesults with uiffeient nonlineai solveis aie in
table 118.1i. With ielaxation slack vaiiables, all nonlineai solveis have tiouble solving the
pioblems. 0nly C0N0PT anu KNITR0 finu the optimal solution foi the 0PF2. C0N0PT,
IP0PT anu KNITR0 founu the optimal solution to the 0PF.


Page
47
Table 118.1i. Solution Statistics foi the 118Bus Pioblem with Biffeient Nonlineai Solveis
with Relaxation Slack vaiiables.
type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN NIN0PF
0PF2
SN0PT 1SS9.8S2 u.uu1 1.6S2 u.uS6 2.1S7 1u u 1SSu.S87
NIN0S 1694.146 u.uS2 S.424 u.19S 9.u79 S 9 1S42.u9S
C0N0PT 18u7.122 u.u49 8.441 u.1uS 14.227 9 9 1296.6u7
IP0PT 19SS.824 u.uS8 1u.9u6 u.11S 2u.S27 1u 1u 1S1u.u22
KNITR0 1S7u.uu9 u.u49 129.S87 u.722 1uuu.u44 9 2 1297.6S9
0PF
NIN0S 1S94.88S u.uu4 2.788 u.176 6.178 1u u 1S6S.S9S
C0N0PT 1467.S77 u.uS6 6.2u4 u.SSS 2S.uSS 1u S 1S11.498
IP0PT 1S11.S17 u 9.u76 u.uS7 1u.764 1u u 1S11.499
SN0PT 1S9S.91 u.uuS 1u.14S u.8S7 92.S71 1u u 1S61.S67
KNITR0 S682.S u.S44 14u.969 u.691 1uuu.u44 8 1 1S11.498
0PFp
SN0PT 1S8u.287 u.uS8 1.2S6 u.487 6.9S7 1u 1 1S2u.67S
C0N0PT 1S27.6SS u.u12 1.911 u.2S2 S.99 1u 1 1S1S.96S
NIN0S 1S92.89 u.u21 2.442 u.121 S.729 1u 1 1SS4.S94
KNITR0 1419.S99 u.u4 4.271 u.SuS 1S.694 1u 1 1S1S.96S
IP0PT 1S66.8S9 u.uS8 S.u22 u.2u4 1S.977 1u 1 1S11.499
Foi lineai solveis, u0R0BI anu CPLEX have similai peifoimance with ielaxation
slack vaiiables see Table 118.2. The optimal solution values aie within 2 peicent of the best
nonlineai solvei peifoimance.
Table 118.2. Solution Statistics foi the 118Bus Pioblem with the Lineai Solveis
with Relaxation Slack vaiiables.
type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN
0PFlin
u0R0BI 129S u.uu6 1.89 u.41 u.1S 86 u
u0R0BI 129S u.uu6 1.89 u.41 u.1S 86 u
CPLEX 129S u.uu6 1.89 u.4S u.17 82 u
u0R0BI 129S u.uu6 1.9S u.41 u.16 86 u
CPLEX 129S u.uu6 1.97 u.4S u.18 82 u
BTBETA
u0R0BI 1261 u u.u9 u.uS u.1u 1u u
u0R0BI 1261 u u.u9 u.uS u.11 1u u
u0R0BI 1261 u u.u9 u.uS u.11 1u u
CPLEX 1261 u u.1S u.u1 u.14 1u u
CPLEX 1261 u u.1S u.u7 u.21 1u u

When iun at uiffeient times, the iesults weie similai with the exception of cpu
times.
Table 118.S. Solution Statistics foi the 118Bus Pioblem with Relaxation Slack vaiiables at
A Biffeient Time.
0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN
0PF2
SN0PT 1,S42 u.uu u.69 u.uS u.8S 1u u
C0N0PT 1,797 u.u4 S.S9 u.u6 4.S2 1u 9
IP0PT 1,9S1 u.u4 S.91 u.u6 4.96 1u 1u
NIN0S 1,768 u.u4 S.97 u.S8 2S.S2 9 9
KNITR0 1,S48 u.uS Su.11 u.66 212.uS 9 2
0PF

Page
48
NIN0S 1,429 u.uS 1.S1 u.17 2.47 1u 1
C0N0PT 1,S7S u.u6 2.99 u.19 6.SS 1u S
IP0PT 1,S12 u.uu S.S7 u.u6 4.7u 1u u
SN0PT 1,S92 u.u1 4.17 u.86 S8.22 1u u
KNITR0 7,S49 u.76 Su.71 u.9S 497.64 9 1
0PFp
SN0PT 1,S81 u.u4 u.4S u.SS 2.6S 1u 1
C0N0PT 1,SS2 u.uS u.86 u.22 2.61 1u 1
NIN0S 1,S9S u.u2 1.u6 u.12 1.9S 1u 1
KNITR0 41u,6uu u.9S 1.46 u.Su S.1S 1u 1
IP0PT 1,S67 u.u4 1.86 u.17 4.6S 1u 1
BTBETA
u0R0BI 1,261 u.uu u.u4 u.u2 u.u4 1u u
CPLEX 1,261 u.uu u.u6 u.u2 u.u7 1u u
0PFlin
u0R0BI 1,289 u.u1 u.S1 u.6S u.u6 62 u
CPLEX 1,288 u.u1 u.62 u.62 u.u8 61 u


Page
49
C0N0PT. The 0PF2 foimulation hau 1,771 iows, 2,uu7 columns, 7,7uS nonzeioes, anu
1,2S4 nonconstant fiist ueiivatives with S,149 evaluation opeiations. C0N0PT founu 1S8
posttiiangulai equations. The iesults aie piesenteu in Table 118.4. At iteiation 411,
C0N0PT finus a feasible solution anu at iteiation 71u founu the optimal solution of 8.427E7.
At iteiation 8uS, C0N0PT founu a feasible solution anu at iteiation 1262, founu an optimal
solution with value of 1.889E8 with the following message: The eiioi in the optimal
objective function value. Bessian is less than the minimal toleiance on the objective. With
anothei staiting point, at iteiation 9S1, C0N0PT founu a feasible solution anu at iteiation
1SS7, an optimal solution of 1.297ES Reuuceu giauient 4.8Eu8. The 0PF anu 0PFp
foimulations hau 1,889 iows, S,187 columns, 9,961 nonzeioes, anu 1,18u nonconstant
fiist ueiivatives with 2,714 evaluation opeiations. C0N0PT founu 64 pietiiangulai
equations anu 212 posttiiangulai equations.
With ielaxation slack vaiiables, at iteiation 216, C0N0PT founu a feasible solution
anu at iteiation 291 founu a local optimal solution with a value of 1.S11ES. Foi staiting
point 2, at iteiation 126, C0N0PT founu a feasible solution anu at iteiation 22S founu an
optimal solution with a value of 1.S11ES. Foi staiting point of S, at iteiation 18S, C0N0PT
founu a feasible solution anu at iteiation 78u founu an optimal solution with a value of
7.u44E7 with a ieuuceu giauient of .SEu9. Foi 0PFp, at iteiation 91, C0N0PT founu a
feasible solution anu at iteiation 1S1, C0N0PT founu an optimal solution with a value of
1.S12ES with a ieuuceu giauient less than toleiance.
With ielaxation slack vaiiables, C0N0PT founu optimal values neai the optimal
value anu peifoimeu bettei on the flat stait than the aveiage ianuom staiting point foi
0PF2 anu 0PF. Without ielaxation vaiiables, C0N0PT founu the optimal solutions of 1297
foi 0PF2 anu 1S11 foi 0PF. The peifoimance with anu without ielaxation slack vaiiables
aie in Table 118.S.
Table 118.S. C0N0PT Peifoimance on the 118Bus With anu Without Relaxation Slack
vaiiables.
0PF2 0PF 0PFp
stai without with without with without with
pt itei itei opt opt itei itei opt opt itei opt itei opt opt
1 69 71u 8E 198 2u 198 1E 1S1 11 1S1 11 1E 1S1
2 S8 121 SE 167 Su 4u2 1E 1SS 92 1S1 21 1E 1S1
S S6 1S1 1E 129 Su S1S 1E 1S1 19 1S1 2S 1E 1S1
4 S1 98u 1E 2u2 S8 664 7E 197 24 16S S2 2E 147
S SS 1u7 1E 1S1 S2 S42 1E 1S1 1S 1S1 24 1E 1S1
6 6u 17u 6E 171 22 S11 1E 1S1 SS 1S1 S8 1E 1S1
7 4S 122 1E 218 27 S9S 1E 1S1 1S 1S1 16 1E 1S1
8 S6 1SS 1E 2u4 4S SS9 1E 1S1 2u 1S1 S7 1E 1S1
9 8u 611 2E 192 19 S6u 1E 1S1 88 1S1 1S 1E 1S1
1u 7S 1SS 9E 19S 21 1uS 7E 196 1S 1S1 16 1E 1S1
avg 6u 11S 8E 18S 28 4S8 2E 146 16 1S4 2S 2E 1S2


Page
50
KNITR0. The peifoimance with anu without ielaxation slack vaiiables aie in Table 118.4.
Table 118.4. KNITR0 Peifoimance on the 118Bus without Relaxation Slack vaiiables.
0PF2 0PF2 0PF 0PF 0PFp 0PFp
stait without with without with without with
point optval optval optm optval optval optm optval optval optm
1 1297 1.SES 1298 1SS8 1.SES 1S14 1S12 1.SES 1S14
2 1297 1.SES 1Suu 1S17 1.SES 1S14 1S12 1.SES 1S14
S 1297 1.SES 1299 1S14 1.SES 1S14 1S12 1.SES 1S14
4 1S24 1.SES 1298 1S12 2.uE4 186uu 1S12 1.SE7 18S9
S 1297 8.6E7 18SS 1S12 1.SE8 642Su 1S12 1.9ES 1S71
6 1297 1.2E8 1941 1S12 S.SE+1u S.SE+1u 1S12 1.SES 1S14
7 1297 1.SES 1298 1S12 1.6ES 1S7S 1S12 2.8ES 1S66
8 1297 1.SES 1299 1S12 2.9ES 29u9 1S1S 1.SES 1S14
9 1297 1.SES 1298 1S2u 1.SES 1S14 1S12 1.SES 1S14
1u 1298 1.SES 1298 1S12 1.SES 1S2u 1S12 1.SES 1S14
avg 1Suu 2.uE7 1418 1S16 S.SE9 S.SE9 1S12 1.SE6 1419

SN0PT. Without ielaxation slack vaiiables, SN0PT always founu the optimal value 0PF2 of
1297, anu 0PF of 1S11. SN0PT peifoimeu bettei on the flat stait than each ianuom staiting
point foi 0PF2, 0PF anu 0PFp. With ielaxation slack vaiiables, SN0PT hau uifficulty (see
Table 118.S).
Table 118.S SN0PT Peifoimance on the 118Bus With anu Without Relaxation Slack
vaiiables.
0PF2 0PF 0PFp
stait without with without with without with
point itei itei opt itei itei opt itei opt itei opt optm
1 1768 14SS 1S42 1S96 927 1S62 691 1S11 489 1S21 1S22
2 S8S8 2462 1SS6 24S4 1476 1S99 8u8 1S11 717 1S29 1S29
S SS27 298u 1S47 2Su9 17uS 14S8 6S1 1S11 S7u 1S26 1S26
4 S6S2 Su29 1SS1 2612 1818 14Su 178S9 2u48 88u2 SE7 1878
S S189 28SS 1SSS 2S6S 1992 1S8S 88S 1S11 629 1S26 1S26
6 S911 27u1 1S48 2SS7 192S 1S7S 11621 1S11 76u 1S22 1S22
7 SSSS 2949 1SS9 26S9 1862 1S89 SS6S 17u4 64S 1S21 1S21
8 S46u SSS9 1S41 2S8u 1886 14uS 9S1 1S11 Su6 1S26 1S26
9 SSu6 2S69 1SS8 2626 1714 1S76 799 1S11 7uu 1S26 1S26
1u 48u4 2S89 1S42 2SS7 17SS 14uS 9647 1828 77S 1S28 1S28
avg S479 2692 1S4u 24u6 17u4 1S96 4944 1476 14S9 SE6 1S8u


Page
51
NIN0S. Without ielaxation slack vaiiables, NIN0S always founu the optimal value of 1297
foi 0PF2 anu 1S11 foi 0PF anu 0PFp. NIN0S peifoims bettei on the flat stait than the
aveiage ianuom staiting point. With ielaxation slack vaiiables, NIN0S hau uificulty (see
Table 118.6).
Table 118.6 NIN0S Peifoimance on the 118Bus With anu Without Relaxation Slack
vaiiables.
0PF2 0PF 0PFp
stait withou with withou with without With
poin itei itei opt opt itei itei opt itei opt itei opt omp
1 176S 2S4 SE 1677 792 6uS 14u 66S 1S1 4u4 1S7 1S7
2 2u82 242 4E 1846 882 294 14u 67S 1S1 28S 1S6 1S6
S 18u8 S8u 1E 1S42 9S8 281 1S7 69u 1S1 4uu 1S8 1S8
4 S6u9 Su1 4E 17SS 847 181 141 8S9 199 268 SE7 167
S 287S 89S 1E 16u9 921 498 1S9 71u 1S1 1S9 1S4 1S4
6 41S7 291 4E 2u1S 21S7 2S4 1S6 7uS 1S1 S4u 1SS 1SS
7 28SS 2S6 SE 18u9 997 1S9 1S6 68S 1S1 149 1SS 1SS
8 1874 SSu SE 1864 877 462 141 711 1S1 1S4 1S4 1S4
9 2792 S47 SE 1696 19u7 19S 14u 712 1S1 S18 1S6 1S6
1u 2244 249 SE 174S 917 22S 141 692 1S1 S28 1S9 1S9
avg 26u4 S7S SE 17S4 1122 S1S 1S9 2u9 1S8 278 SE6 1S9

IP0PT. Without ielaxation slack vaiiables, IP0PT always founu the optimal value anu
peifoim bettei on the flat stait than the aveiage ianuom staiting point foi 0PF2 anu 0PF
(see Table 118.7).
Table 118.7 IP0PT Peifoimance on the 118Bus without Relaxation Slack vaiiables.
stait 0PF2 0PF 0PFp
point iteiations optval iteiations optval iteiations optval
1 S8 1296.6u7 6u 1S11.498 26 1S11.498
2 84 1296.6u7 8S 1S11.498 22 1S11.498
S S7 1296.6u7 Su 1S11.498 2S 1S11.498
4 28 1296.6u7 28 1S11.498 167 46S1.8u9
S 42 1296.6u7 4S 1S11.498 2S 1S11.498
6 SS 1296.6u7 77 1S11.498 47 1S11.498
7 S9 1296.6u7 8u 1S11.498 2S 1S11.498
8 S8 1296.6u7 72 1S11.498 28 1S11.498
9 S1 1296.6u7 46 1S11.498 28 1S11.498
1u 29 1296.6u7 S1 1S11.498 26 1S11.498
avg 44 1296.6u7 S9 1S11.498 41 164S.S29


Page
52
With ielaxation slack vaiiables, IP0PT hau uifficulty (see Table 118.8).
Table 118.8 IP0PT Peifoimance on the 118Bus with Relaxation Slack vaiiables.
stait 0PF2 0PF 0PFp
point iteis optval optm iteis optval optm iteis optval optm
1 76 1.u8E8 2u86.1 97 1.S1ES 1S11.S S9 1.S1ES 1S11.6
2 82 4.1uE7 1847.1 96 1.S1ES 1S11.S 47 1.S1ES 1S11.S
S 1uu 4.1uE7 1847.1 81 1.S1ES 1S11.S 41 1.S1ES 1S11.S
4 18S 6.u1E7 2u26.7 94 1.S1ES 1S11.S 12u S.11E7 1864.8
S 11u 1.48E8 22Su.u 96 1.S1ES 1S11.S S8 1.S1ES 1S11.S
6 81 6.9SE7 1917.S 112 1.S1ES 1S11.S 71 1.S1ES 1S11.S
7 1u2 S.81E7 1686.8 1u7 1.S1ES 1S11.S S9 1.S1ES 1S11.S
8 16u 4.S4E8 2Su6.S 9S 1.S1ES 1S11.S S9 1.S1ES 1S11.S
9 82 1.4SE7 1S1u.u 1uu 1.S1ES 1S11.S 44 1.S1ES 1S11.S
1u 91 2.74E8 2u8u.7 98 1.S1ES 1S11.S 46 1.S1ES 1S11.S

u0R0BI. The BTBETA foimulation hau 2,1S7 iows, 16,S2S columns anu S4,19S nonzeioes.
u0R0BI Piesolve ieuuceu the pioblem size to 199 iows, 497 columns, 69S nonzeios in
u.uSs. u0R0BI solveu in 1uS iteiations anu u.uS seconus with an optimal objective of
1.261eS. 0PFlin staiteu with Su7u iows, S187 columns anu 1127S nonzeioes. piesolve
iesulteu in 124u iows, 1961 columns anu 6u46 nonzeioes. 0PFlin pass S hau S,Su6 iows,
S,187 columns anu 12,u8S nonzeioes anu piesolve 124u iows, 1961 columns anu 6u46
nonzeioes.
Table 118.9. Lineai Appioximation LP Passes By Staiting Point foi The 118Bus
0sing u0R0BI without Relaxation Slack vaiiables.
stait 1 2 S 4 S 6 7 8 9 1u
Without S 1u 6 1u 1u 1u 1u 1u 1u 1u


Page
53
Table 118.1u. Lineai Appioximation Nonzeios, Iteiations anu 0ptimal value by LP Pass anu
Staiting Point foi the 118Bus 0sing u0R0BI.
pass Nonzeios iteis optval Nonzeios iteis optval Nonzeios iteis optval
stait point 1 2 S
1 1127S 711 1266 11S77 691 128u 11S77 692 1282
2 11849 S78 1297 1184S S97 1269 1184S 7u2 1271
S 12u8S SS6 1Su4 12u8S S78 127S 12u79 61u 127S
4 12S19 S67 1284 12S17 626 1281
S 12SSS S9S 1276 12SSS 62u 128S
6 12791 666 1296 12789 619 1SuS
7 1Su27 647 1289
8 1S26S 671 1287
9 1S499 61S 1279
1u 1S7SS 8uu 1291
stait point 4 S 6
1 11S77 7u8 128S 11S77 679 128S 11S77 7uu 1278
2 11849 8uu 127u 11849 S99 1274 1184S 67S 1268
S 12u69 6SS 1278 12u8S 62S 1276 12u8S 69S 1277
4 12SuS 1u21 1299 12S21 62S 1288 12S1S 6S6 1299
S 12SS7 978 1291 12SS7 624 128S 12SSS 628 1276
6 12769 127S 1282 1279S 6uS 128u 1278S 7uu 1294
7 1Suu1 987 1296 1Su29 6S6 1291 1Su19 686 1286
8 1S2SS 1S17 1287 1S26S 6uu 1281 1S2SS 711 128S
9 1S461 7SS 1291 1SSu1 72S 1294 1S491 66S 1296
1u 1S7uS 689 1296 1S7S7 666 1287 1S727 692 1286
stait point 7 8 9
1 11S77 67S 128u 11S77 66u 1281 11S77 694 1286
2 1184S 6S9 127u 11849 61u 127S 11841 761 1269
S 12u79 696 1276 12u8S 1u14 1269 12u7S 62S 1271
4 12S17 61S 129u 12S21 6S4 127S 12S1S 71S 128u
S 12SSS 664 1272 12SS7 6uS 1Su2 12S4S 988 1282
6 12789 691 1287 1279S 644 129u 1278S 6uu 1286
7 1Su2S 6u1 1278 1Su29 6S2 128S 1Su19 6S7 127S
8 1S261 6S9 1298 1S26S 64S 1279 1S2SS 6S8 1297
9 1SSu1 1418 129u 1S491 S72 1298
1u 1S7S7 7uS 128S
stait point 1u
1 11S77 684 1281
2 1184S 712 1277
S 12u8S 612 1288
4 12S19 629 1278
S 12SSS 71S 129S
6 12791 619 1284
7 1Su27 764 129S
8 1S26S 64u 1282
9 1S499 72S 129u
1u 1S7SS 721 128S


Page
54
Incieaseu Bemanu. With uemanu incieaseu by 1.2 anu 1.4, the iesults aie in table 118.11.
0PFlin is moie iobust.
Table 118.11. Solution Statistics foi 118Bus Pioblem with Relaxation Slack vaiiables with
Incieaseu Bemanu 0sing C0N0PT anu u0R0BI.
type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN
0PF2 1797 u.u44 S.6S u.u6 4.6S 1u 9
0PF 1S7S u.uS9 S.2S u.19 6.69 1u S
0PFlin 1289 u.uu7 u.SS u.6u u.u7 62 u
0PFp 1SS2 u.u28 u.9S u.21 2.61 1u 1
BTBETA 1261 u.uuu u.u4 u.u2 u.u4 1u u
1.2*uemanu
0PF2 1888 u.uSu 4.74 u.18 9.86 8 9
0PF 1814 u.u4S 2.18 u.24 6.u7 1u S
0PFlin 16Su u.uu7 u.64 u.S7 u.u7 69 u
0PFp 17u7 u.u27 u.99 u.SS 4.2S 1u 1
BTBETA 16u1 u.uuu u.u4 u.u2 u.u4 1u u
1.4*uemanu
0PF2 2141 u.u24 S.8S u.12 7.1S 1u 6
0PF 2142 u.uSu 2.S8 u.17 S.S4 1u S
0PFlin 1978 u.uu6 u.82 u.4u u.u7 84 u
0PFp 2uS9 u.u2u 1.S7 u.22 S.84 1u 1
BTBETA 1946 u.uuu u.u4 u.uS u.u4 1u u

Page
55
Suubus pioblem
Without ielaxation slack vaiiables, all solveis ueclaie an infeasible pioblem. The
iesults with uiffeient nonlineai solveis with ielaxation slack vaiiables aie in table Suu.1i.
All solveis (except KNITR0) ueclaie a local optimal solution, but occasionally fail to iemove
the penalty slacks. 0nly C0N0PT anu IP0PT finu the global optimal value.
Table Suu.1i. Solution Statistics foi the SuuBus Pioblem with the Nonlineai Solveis
with Relaxation vaiiables.
type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN NIN0PT
0PF2
SN0PT 7Su7 u.uu7 14.1 u.147 28.S 1u 2 72uS
C0N0PT 7229 u.uu4 24.S u.u61 Su 1u 1 7197
NIN0S 7SuS u.uu4 S2.S u.S1S 129 1u 1 7271
IP0PT 7S4S u.uu8 S4.7 u.1S 7S 1u 4 7197
KNITR0 7SSu u.u18 S29.S u.2S 1uuu.S S 6 72uS
0PF
NIN0S 7S74 u.uuS 6.4 u.121 12.2 1u 1 7SS1
C0N0PT 7488 u 6.6 u.u86 11.1 1u u 7488
SN0PT 7671 u.uu6 14.7 u.2u1 28.S 1u 4 7SS1
IP0PT 7488 u 16 u.2u9 S4.S 1u u 7488
KNITR0 14Suu u.271 71S.6 u.184 1uuu.1 S 8 7SS7
0PFp
NIN0S 77S2 u.uu9 4.9 u.192 9.2 1u S 7SuS
SN0PT 7S88 u.uu7 8.1 u.2S 17.1 1u 1 7499
IP0PT 7776 u.u27 8.S u.u91 1S.1 1u 2 7488
C0N0PT 77S1 u.u1 8.7 u.227 18.S 1u S 7488
KNITR0 9u89u u.842 694u u.916 6722u 7 S 7494

Foi lineai solveis, u0R0BI outpeifoims CPLEX see table Suu.2. The optimal solution
values aie within S peicent of the best nonlineai solveis. All lineai appioximation took the
full 1u passes anu uiu not meet the othei conveigence ciiteiia. The lineai solveis weie moie
iobust than the nonlineai solveis.


Page
56
Table Suu.2. Solution Statistics foi the SuuBus Pioblem with the Lineai Solveis.
type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN
0PFlin
CPLEX 7S28 u.u12 9.9 u.SS8 u.4 1uu u
CPLEX 7S28 u.u12 1u u.S28 u.S 1uu u
u0R0BI 761S u.u42 16.2 u.161 u.S 1uu u
u0R0BI 761S u.u42 16.4 u.146 u.S 1uu u
u0R0BI 761S u.u42 16.6 u.16S u.4 1uu u
u0R0BI 761S u.u42 16.6 u.148 u.4 1uu u
BTBETA
u0R0BI 719S u u.S u.u11 u.S 1u u
u0R0BI 719S u u.S u.u12 u.S 1u u
CPLEX 719S u u.S u.u22 u.4 1u u
u0R0BI 719S u u.S u.uu9 u.S 1u u
u0R0BI 719S u u.S u.uS4 u.4 1u u
CPLEX 719S u u.S u.u27 u.4 1u u

Table Suu.S. Solution Statistics foi the SuuBus Pioblem fiom Anothei Bay.
optvavg optvcv cpuavg cpucv cpumax C0NvRu Tepsin
0PF2
C0N0PT 7291 u.uu9 22.97 u.u81 S1.9 1u 2
NIN0S 7277 u.uuu S4.uu u.296 128.8 1u u
SN0PT 7266 u.uuS 1S.14 u.129 28.9 1u 1
IP0PT 742u u.uu8 7u.1S u.1S7 12S.2 1u 6
KNITR0 7S6u u.u1u S92.SS u.247 1uuu.2 S 6
KNITR0 7S6u u.u1u 61S.S8 u.226 1uuu.1 S 6
0PF
C0N0PT 7488 u.uuu 6.S8 u.u81 8.7 1u u
NIN0S 7S8u u.uuS 6.4S u.1u8 11.4 9 1
SN0PT 7696 u.uu6 17.1S u.186 S1.2 1u S
IP0PT 7488 u.uuu S2.86 u.274 1u7.8 1u u
KNITR0 S2u8u u.6S9 S72.12 u.2Su 1uuu.1 S 7
KNITR0 8Su7 u.u47 726u.1u u.8SS 661Su.u S 7
0PFp
C0N0PT 7682 u.u1u 9.86 u.282 28.8 1u 4
NIN0S 77SS u.uu9 S.S4 u.2uS 1u.2 1u S
SN0PT 7S89 u.uu7 8.21 u.2S4 17.1 1u 1
IP0PT 7776 u.u27 18.8S u.217 41.4 1u 2
KNITR0 S181u u.698 27S.2u u.48S 1uuu.u 8 S
0PFlin
CPLEX 7S2u u.u1S 7.76 u.681 u.4 77 u
u0R0BI 76u8 u.u4S 12.42 u.SSS u.S 77 u
BTBETA
u0R0BI 719S u.uuu u.SS u.u14 u.4 1u
CPLEX 719S u.uuu u.S4 u.u1S u.4 1u


Page
57
C0N0PT. The 0PF2 foimulation hau 4,Su1 iows, S,1u1 columns, 18,9S8 nonzeioes, anu
S,u69 nonconstant fiist ueiivatives with 7,4SS evaluation opeiations anu occupieu 29 Nb.
C0N0PT hau 41u posttiiangulai equations. With ielaxation slack vaiiables, C0N0PT founu
a feasible solution at 1u2u iteiations anu an optimal solution at iteiation 24u4 with a
ieuuceu giauient of 2.9Eu8. Foi staiting point 2, C0N0PT founu an optimal solution of
S.219E7 at iteiation 211u.
The 0PF anu 0PFp foimulations with ielaxation slack vaiiables hau 4,8u1 iows,
8,1u1 columns, 24,uS9 nonzeioes, anu S,uuu nonconstant fiist ueiivatives with 6,9uu
evaluation opeiations occupying 29 Nb with 2S1 pietiiangulai equations anu 716 post
tiiangulai equations. Foi 0PF, at iteiation 6u1, C0N0PT founu optimal solution of S.219E7
with a ieuuceu giauient 4.uEu9 anu with active penalty slacks. Foi 0PFp, at 126, C0N0PT
founu a feasible solution anu at iteiation 4S9, an optimal solution of S.219E7 with ieuuceu
giauient of 1.1Eu8 with active penalty slacks.
Table Suu.4. C0N0PT Peifoimance on the SuuBus Pioblem.
stait 0PF2 0PF 0PFp
point iteis optval optm iteis optval optm iteis optval optm
1 24u4 S.2E7 7197 644 S.2E7 7488 S9S S.2E7 7488
2 1992 S.2E7 7197 S18 S.2E7 7488 1Su7 4.8E7 7866
S Su84 S.SE7 7S11 1u88 S.2E7 7488 186 S.2E7 7488
4 294S S.2E7 7197 4S4 S.2E7 7488 14S2 8.7E7 79Su
S S248 S.2E7 7197 7S2 S.2E7 7488 17uS 1.2E8 8u26
6 S14S S.2E7 7197 8Su S.2E7 7488 211S 4.1E7 792S
7 Su2S S.2E7 7197 887 S.2E7 7488 S2S S.2E7 7488
8 2977 S.2E7 7197 742 S.2E7 7488 4S4 S.2E7 7488
9 S6S4 S.2E7 7197 862 S.2E7 7488 S74 S.2E7 7488
1u 171S S.2E7 7197 78u S.2E7 7488 1699 9.2E7 81uu
avg 2818 S.4E7 7228 7S6 S.2E7 7488 1uS7 S.SE7 77S1


Page
58
KNITR0 With ielaxation slack vaiiables, KNITR0 peifoimeu pooily.
Table Suu.4. KNITR0 Peifoimance on the SuuBus.
stait 0PF2 0PF 0PFp
point optval optm optval optm optval optm
1 S.22E7 72uS S.22E7 7SS7 S.22E7 7494
2 S.22E7 72S6 9.14E7 1216u 4.88E7 S867uu
S 4.4uE14 1.1uE9 1.4uE1S 9.48E7 S.22E7 114uu
4 S.S1E8 776S 8.SSE7 49uSuu 8.69E7 827u
S 6.6SE8 78SS S.6uE7 2SS4uu 1.12E8 84u4
6 1.S7E9 7867 S.u1E7 492uuu 1.u4E8 2uu9uu
7 4.SuE1S S927uuu 1.uuE1S 2SSSuu S.22E7 7494
8 4.14E7 7S22 S.22E7 2S44u S.22E7 7494
9 S.22E7 72Su 6.62E7 1192u S.22E7 7SSS
1u 9.6SE8 7919 S.S8E8 1897uu 7.u6E7 81S2
avg 4.8SE1S 1.11E8 2.4uE12 9.6SE6 S.8SE7 8.S4E4

SN0PT
Table Suu.6 SN0PT Peifoimance on the Suubus.
stait 0PF2 0PF 0PFp
point iteis optval optm iteis optval optm iteis optval optm
1 1S478 S.2E7 72uS 6Su7 S.2E7 7SS1 2Su2 S.2E7 7499
2 S8S29 S.6E7 7647 9966 S.2E7 7S66 2S64S S.2E7 76u2
S 1681S S.2E7 724u 19742 S.2E7 7S79 2988 S.2E7 7Su6
4 1969S S.2E7 72S1 18787 S.2E7 7S8S 884S S.2E7 7SS1
S 18uS1 S.2E7 72u9 29142 S.6E7 78Su 21Suu S.2E7 7S64
6 4Su6u S.6E7 76uS S9uS1 7.SE7 789S 18142 S.2E7 7SS8
7 14882 S.2E7 721S S8S42 S.1E7 782S Su66 S.2E7 7Suu
8 1S2S9 S.2E7 72Su 7478 S.2E7 7SS4 289S S.2E7 7SuS
9 14162 S.2E7 7226 1u2S8 S.2E7 7SS4 Su61 S.2E7 7SuS
1u 1419S S.2E7 724S S98S4 4.1E7 77S8 1794u 9.2E7 8114
avg 2u79S S.7E7 7Su7 21887 4.1E7 7671 1u4S8 S.8E7 7S88


Page
59
NIN0S
Table Suu.7 NIN0S Peifoimance on the Suubus.
stait 0PF2 0PF 0PFp
point iteis optval optm iteis optval optm iteis optval optm
1 246Su S.2E7 7277 S172 S.2E7 7SS7 2224 S.2E7 7SuS
2 2414u S.2E7 7271 6S64 S.2E7 7S4u 8228 4.8E7 78S8
S 1287S9 S.2E7 7278 8u76 S.2E7 761S 4S62 S.2E7 7S26
4 1872S S.2E7 7287 61S1 S.2E7 7S6u S9S1 8.7E7 797S
S 2S9u2 S.2E7 727S S412 S.2E7 7S4S 91u7 1.2E8 8u74
6 26SS8 S.2E7 7278 12799 4.1E7 7794 911S 1.uE8 8uS9
7 2S924 S.2E7 7278 7768 S.2E7 7SS1 SSS2 S.2E7 7S28
8 27414 S.2E7 7272 7279 S.2E7 7S4S 2414 S.2E7 7S1S
9 19u71 S.2E7 7274 82u9 S.2E7 7SS7 2S61 S.2E7 7S14
1u 1997S 4.1E7 7S67 SS76 S.2E7 7S41 98uu 4.SE7 7767
avg SS69u S.SE7 7Su6 7281 S.SE7 7S74 S7u7 S.7E7 77S2

IP0PT
Table Suu.8 IP0PT Peifoimance on the Suubus.
stait 0PF2 0PF 0PFp
point iteis optval optm iteis optval optm iteis optval optm
1 161 S.2E7 7197 127 S.2E7 7488 7S S.2E7 7488
2 S81 S.2E7 7197 148 S.2E7 7488 124 S.2E7 7488
S 47S 1.8E8 7627 SS1 S.2E7 7488 1u2 S.2E7 7488
4 S2S S.8E7 749S 1uu S.2E7 7488 76 8.7E7 9671
S SSu 6.SE7 7S2u 9S S.2E7 7488 81 S.2E7 7488
6 4SS S.2E7 7197 48S S.2E7 7488 126 1.uE8 8189
7 772 S.2E7 7197 71 S.2E7 7488 86 S.2E7 7488
8 Suu S.2E7 7197 78 S.2E7 7488 74 S.2E7 7488
9 SS9 1.9E8 76Su 1u6 S.2E7 7488 71 S.2E7 7488
1u 4u9 S.2E7 7197 S24 S.2E7 7488 74 S.2E7 7488
avg 417 6.8E7 7S4S 186 S.2E7 7488 89 4.SE7 7776

u0R0BI. The BTBETA foimulation hau S,1SS iows, 96,S19 columns, 194,899 nonzeioes
anu occupieu S2 Nb. in u.22s u0R0BI Piesolve iesulteu in S8S iows, 1u89 columns, 1471
nonzeios. u0R0BI solveu in 196 iteiations anu u.28 seconus with an optimal objective of
S.219e7 with active penalty slacks.
Foi u0R0BI, 0PFlin hau 8,4u2 iows 8,1u1 columns anu 29,1uS nonzeioes
occupying 27 Nb. Piesolve iesulteu in 29u6 iows, 4126 columns anu 1SS8S nonzeioes.
Buiing the solution piocess, the following waining was issueu Naikowitz toleiance
tighteneu to u.uS12S. The sequence of optimal values is not monotonic uue to lineai
appioximations to the nonconvex constiaints.

Page
60
Table Suu.9. Lineai Appioximation Nonzeios, Iteiations anu 0ptimal value by LP Pass anu
Staiting Point foi the SuuBus 0sing u0R0BI.
pass nonzeios iteis optval optm nonzeios iteis optval optm
Stait 1 Stait 2
1 27SuS 2S7S S.22E7 7247 27611 S189 S.22E7 7297
2 2881S 268u S.22E7 7444 288SS SSS2 S.22E7 7SSS
S 294S9 7uu4 S.22E7 7471 29449 9296 S.S9E7 79S1
4 SuuS9 S6u1 S.22E7 7474 SuuS1 71S2 S.SSE7 7718
S Su6S9 S988 S.22E7 747u Su627 69u2 S.26E7 79SS
6 S12S9 4742 S.22E7 7474 S121S 4284 S.26E7 8171
7 S18S9 4SS9 S.22E7 746S S181S 44S4 S.22E7 767u
8 S24SS Suu1 S.22E7 747S S241S Su61 S.4SE7 7918
9 SSu2S 4668 S.22E7 747S SSu17 S8S4 S.22E7 7S84
1u SSSSS 2674 S.22E7 7474 SS61S SuS2 S.22E7 7S72
Stait S Stait 4
1 27611 217S S.22E7 7297 27611 2S19 S.22E7 72SS
2 288S9 S221 S.22E7 7S21 288SS 761S S.22E7 744S
S 294S9 462S S.22E7 7S2u 2944S S499 S.22E7 7S98
4 SuuS9 28S8 S.22E7 7S88 SuuS9 4u8u S.22E7 7SSS
S Su6S9 S62S S.22E7 74S2 Su641 2644 S.22E7 744S
6 S12S9 4212 S.22E7 7462 S12S9 48S9 S.22E7 7S6u
7 S18S9 S86S S.22E7 747u S18S7 46SS S.22E7 749S
8 S24S9 SSuS S.22E7 7466 S24SS 8SS8 S.22E7 749S
9 SSuS9 2819 S.22E7 7471 SSu21 4817 S.22E7 7471
1u SS6S9 4149 S.22E7 746S SS61S S771 S.22E7 749S
Stait S Stait 6
1 27611 S721 S.22E7 7288 27611 2u7u S.22E7 7284
2 288S1 461S S.22E7 74S2 288SS SS46 S.2SE7 8u27
S 294S1 46u6 S.42E7 79S7 294S7 7S4S S.4uE7 796S
4 29941 49u2 S.2SE7 7821 2992S 6S4S S.S7E7 8u42
S SuSu9 S2SS S.22E7 7S69 Su48S 982u S.46E7 8u7S
6 S11u1 4919 S.2SE7 7S2S S1u67 6179 S.24E7 81S4
7 S1687 7197 S.28E7 8u49 S1649 769u S.6uE7 8u12
8 S2267 S424 S.24E7 8129 S22S9 4S9S S.2SE7 8u8S
9 S284S 7S76 S.2SE7 8u1S S28S9 4Su9 S.2SE7 84u6
1u SSS89 6S46 S.2SE7 774u SS447 4149 S.SSE7 8496
Stait 7 Stait 8
1 27611 44SS S.22E7 7284 27611 S21S S.22E7 729S
2 288S9 S698 S.22E7 7S47 288S9 62u9 S.22E7 7S48
S 294S9 4uS7 S.22E7 7S7S 294S9 4Su6 S.22E7 7SSS
4 SuuS9 S4S4 S.22E7 7S24 SuuS7 2792 S.22E7 7S99
S Su6S9 262S S.22E7 74u6 Su6S9 4u18 S.22E7 7449
6 S12S9 2716 S.22E7 7464 S12S9 2948 S.22E7 747S
7 S18S9 2S12 S.22E7 746S S18S9 712S S.22E7 7467
8 S24S9 S4S4 S.22E7 747S S24S9 794u S.22E7 7474
9 SSuS9 4714 S.22E7 7466 SSuS9 Su49 S.22E7 7467
1u SS6S7 4u16 S.22E7 7474 SS6S9 SS19 S.22E7 747S
Stait 9 Stait 1u
1 27611 28uS S.22E7 7271 27611 2S4u S.22E7 7266

Page
61
2 288S9 4SS8 S.22E7 7S66 288SS SS1S S.22E7 74uS
S 294S9 S4S8 S.22E7 7S77 294SS S261 S.2SE7 7762
4 SuuS9 6661 S.22E7 7SS6 Suu47 Su17 S.22E7 7484
S Su6S9 SSu2 S.22E7 7SS6 Su647 SS91 S.S6E7 8198
6 S12SS SuS2 S.22E7 742S S1221 696S S.S4E7 81S4
7 S18S7 4S91 S.22E7 7449 S1799 S9Su S.22E7 7421
8 S24S7 22S6 S.22E7 746u S2S9S S4S1 S.27E7 76S2
9 SSuS7 22Su S.22E7 747u S294S 7uSS S.22E7 7SS9
1u SS6S7 Su81 S.22E7 7469 SSS6S S9u4 S.22E7 7497

Incieasing Bemanu. With uemanu incieaseu by 1.2 anu 1.4, the iesults aie in table Suu.12.
When uemanu is 1.2 oi 1.4 times gieatei, the penalty slacks aie gieatei than the set
thiesholu anu conveigence becomes moie uifficult.
Table Suu.12. Solution Statistics foi SuuBus Pioblem with Incieaseu Bemanu
0sing C0N0PT anu u0R0BI.
type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN
0PF2 7291.214 u.uu9 2S.447 u.u71 S1.6S7 1u 2
0PF 7488.u69 u.uuu 6.8Su u.u77 8.92S 1u u
0PFlin 76u7.6SS u.u4S 11.67u u.S4S u.46S 77 u
0PFp 7681.821 u.u1u 9.6u7 u.2S2 2S.962 1u 4
BTBETA 719S.2S4 u.uuu u.S2u u.u1S u.S48 1u u
1.2*uemanu
0PF2 94uS.8S1 u.uuu 2u.247 u.u69 29.1S7 1u 1u
0PF 9817.u16 u.uuu 9.8uS u.u9S 14.SSS 9 1u
0PFlin 978S.84u u.u1u 18.u72 u.67S u.68u 76 9
0PFp 9826.498 u.uuu 7.8S9 u.164 1S.S8S 8 1u
BTBETA 9189.S62 u.uuu u.SS4 u.u1S u.S67 1u u
1.4*uemanu
Type 0PTvAvu 0PTvCv CP0Avu CP0Cv CP0NAX C0NvRu TEPSIN
0PF2 1uS6u.uuu u.uuu 27.11u u.u72 S6.1Su 7 1u
0PF 1u64u.uuu u.uuu 11.9u6 u.uS7 14.196 1u 1u
0PFlin 1u7Su.uuu u.u22 2S.S2S u.228 u.96S 1uu 1u
0PFp 1u64u.uuu u.uuu 8.7S1 u.u78 1S.9S1 1u 1u
BTBETA 1172u.uuu u.uuu u.S28 u.u22 u.S77 1u u

You might also like