Documentos de Académico
Documentos de Profesional
Documentos de Cultura
PROYECTO DE GRADO
PRESENTADO ANTE LA ILUSTRE UNIVERSIDAD SIMN BOLVAR
COMO REQUISITO PARCIAL PARA OPTAR AL TTULO DE
INGENIERO ELECTRICISTA
PROYECTO DE GRADO
PRESENTADO ANTE LA ILUSTRE UNIVERSIDAD SIMN BOLVAR
COMO REQUISITO PARCIAL PARA OPTAR AL TTULO DE
INGENIERO ELECTRICISTA
UNIVERSIDAD
SIMN BOLVAR
Decanato de Estudios Profesionales
Coordinacin de Ingeniera Elctrica
EN ~
ESTUDIANTES:1:l
JOSf
ESO~A
\.ce..c\o
c;-...
--:::::>
'~~<2-S
~-'<
JURADo:profs.~A1/1JO
370 (
'P1
'ft ~b2
TTULODELTRABAJO:+~\\Ce--.C\;
TUTOR: Prof.\co,-..>\.o
EP
e.
CARNET:
CARNET:
D<{
3(; 04
03 3(,313
a\ '-.)Q.\,c......
CO-TUTOR:
t\GUl?1AS
APROBADO:
~
OBSERV ACIONES:
Prof.
y ftP~O
'E_A_\V~A __
REPROBADO:
~: ,1
Co-Tutor
Nota: Colocar los sellos de los respectiyos Departamentos.
Coordinacin.
Jurado
Para jurados externos, usar el sello de la
En este trabajo se implementa un anlisis de flujo de carga, (FDC ), para ser aplicado
en redes de distribucin de gran tamao. El FDC implementado es un mtodo de barrido
unidireccional directo definido por el uso de una matriz nica que caracteriza la resistencia y
admitancia de las lneas y la topologa de la red. El algoritmo fue codificado en MATLAB. El
mismo fue desarrollado en forma modular utilizando un patrn de diseo de software ModeloVista-Controlador, (MVC ). El FDC implementado en el presente trabajo est basado en Teng
(2003) y se incluye la modificacin propuesta por De Oliveira (2010) para utilizar nmeros
reales en lugar de nmeros complejos. Como contribucin principal se establece un esquema de
datos en RAM (Random Access Memory), lo cual optimiza el proceso de clculo del FDC. El
algoritmo fue validado con redes de 4, 7, 12 y 69 barras. Adems, se verificaron los resultados
del circuito de 7 barras con un Newton-Raphson (NR), (MATPOWER). Finalmente, se aplic
exitosamente en una muestra significativa de la red de la Gran Caracas; conformada por 530
circuitos distribuidos en 78 S/E y 64.251 nodos, que corresponden al 70 % de la demanda de
las Gran Caracas. El resultado del FDC para el 70 % de la Gran Caracas se obtuvo en 0, 76
segundos. El FDC directo estudiado es un mtodo eficiente con un gran potencial para ser
aplicado en el anlisis y planificacin del Sistema Elctrico de Distribucin (SDEE ).
iv
AGRADECIMIENTOS
Quisiera agradecer a Dios y a las tres personas que influyeron en quien soy hoy en da: mi
madre, mi padre y mi abuela Pancha. Gracias a su esfuerzo es que he llegado hasta aqu.
A mis tas Carmen y Zulay que me vieron crecer y siempre han estado apoyndome.
A mi abuelo Santiago y mis tos con los que siempre he contado y podr contar en cualquier
momento.
A mi hermano, mis primos y mi sobrina con los que he crecido y he compartido valiosos
momentos de mi vida.
A mis amigos y compaeros del equipo de rugby subacutico CONGRIOS USB con los
que comparto el vicio por el agua y la adrenalina.
A mis amigos: Martha, Gabriela, Debora, Leopold, Ariaam, Fedora y Jos Rafael. Por ser
unos excelentes amigos y personas.
A todos; no existen palabras suficientes para agradecer lo que han hecho por m.
Quisiera agradecer a mi madre por ser la persona que siempre ha estado conmigo.
Gracias!!!
Gracias al Prof. Paulo De Oliveira por ayudarnos y apoyarnos en todo momento durante la
tesis. Gracias a Benicia, Mara Teresa y al Prof. Miguel Martnez por estar a la orden cada
vez que necesitbamos ayuda. Gracias a Angel y Luis Gerardo por darnos una mano justo
cuando era necesario.
Gracias Totales.
vi
LISTA DE SMBOLOS
xi
LISTA DE ABREVIATURAS
xiii
ndice general
INTRODUCCIN
1. ANTECEDENTES
10
2. METODOLOGA
14
14
15
18
22
22
23
24
xiv
25
25
26
28
3. IMPLEMENTACIN
29
29
34
3.3. Datos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
34
37
37
37
39
39
40
43
47
48
51
51
54
xv
CONCLUSIONES
63
REFERENCIAS BIBLIOGRFICAS
65
71
71
72
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
74
74
83
87
98
100
F. A COMPENSATION-BASED POWER FLOW METHOD FOR WEAKLY MESHED DISTRIBUTION AND TRANSMISSION NETWORKS,
SHIRMOHAMMADI ET AL. 1988, [11]
103
114
121
139
J. SIMPLE AND EFFICIENT COMPUTER ALGORITHM TO SOLVE RADIAL DISTRIBUTION NETWORKS, RANJAN, R. ET AL. 2003, [27] 150
xvii
ndice de tablas
. . . . . . . . . . . . . . . . . . .
25
40
41
44
45
45
46
46
47
50
4.10. Tiempo de cmputo del ASP para red de 530 circuitos de la EDC . . . . . .
51
4.11. Tiempo de cmputo del algoritmo implementado para red de 530 circuitos de
la EDC
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
52
4.12. Espacio Ocupado en Disco por redes de 10, 50, 100, 200 y 530 circuitos . . .
54
62
98
99
xviii
ndice de figuras
10
19
22
27
30
30
32
33
35
36
49
52
53
55
xix
56
56
57
58
. . . . . . . .
59
60
61
62
71
xx
INTRODUCCIN
En la actualidad hay diversas aplicaciones computacionales que son utilizadas como herramientas para el anlisis de flujo de carga. La mayora de stas estn basadas en mtodos
numricos, como el Newton-Raphson (NR), [1]. Dicha metodologa fue desarrollada y usada
para la operacin, control y planificacin de sistemas de transmisin; cuya configuracin es
tpicamente mallada [2, 3]. Pero al ser empleado en SDEE presenta problemas de convergencia
1
2
y eficiencia ya que se produce singularidad en la matriz Jacobiana. Debido, principalmente,
a su estructura radial, baja relacin x/r, lneas no traspuestas y cargas desbalanceadas entre
fases [2, 4, 5].
Tomando en cuenta lo anterior y considerando la topologa radial del sistema de distribucin, se han desarrollado nuevas metodologas de flujo de carga que explotan la radialidad
del circuito; es decir, no requieren inversin de Jacobiano al utilizar tcnicas conocidas como
el Gauss-Seidel GS o de barrido, [18, 22, 31, 33, 44]. Finalmente, dichos algoritmos se extendieron para poder ser aplicados en sistemas dbilmente mallados [11, 12, 33].
Los mtodos desarrollados para solucionar el FDC aplicado a sistemas radiales de distribucin, llamados tpicamente flujos de carga de barrido (FDCB ), pueden ser catalogados
en dos grupos [2]. En el primer grupo, aquellos que hacen ciertas modificaciones a tcnicas
ya existentes como el NR [6]-[10]. Y en el segundo, estn aquellos que hacen un proceso
de barrido hacia delante y hacia atrs, (forward and backward sweep), usando las leyes de
Kirchhoff [11]-[21] o utilizando la ecuacin bicuadrtica [22]-[36].
Tpicamente, los FDCB presentan una tasa lenta de convergencia. Sin embargo, son altamente eficientes desde el punto de vista computacional porque no es necesario invertir
matrices [20]; hecho que resulta importante al resolver redes de gran tamao. Con las herramientas computacionales disponibles en la actualidad se ha reducido considerablemente
el tiempo empleado en obtener una solucin del FDC en redes de distribucin de gran tamao.
3
aplicacin web multiplataforma, control de versiones mediante un esquema colaborativo, [50].
Objetivos
Para lograr el alcance propuesto, se plantearon un objetivo general y varios especficos,
stos se muestran a continuacin:
Objetivo General: Implementar un flujo de carga en base a la tcnica de barrido unidireccional, basada en la construccin de una matriz nica que incluye impedancias de lnea y
la topologa del sistema, en redes de distribucin de gran tamao.
Objetivos Especficos:
Desarrollar un flujo de carga prototipo.
Realizar un estudio comparativo de los tiempos de cmputo entre el FDC implementado
y otros FDC eficientes como el NR y el Teng.
4
Implementar eficientemente el programa en plataforma computacional.
Aplicacin en redes ejemplo reportadas en literatura y redes de distribucin reales de
gran tamao.
Justificacin de la Tesis
El proyecto surge de la necesidad de hacer estudios de red a gran escala en tiempo til
para cumplir los fines de la operacin y planificacin de SDEE.
Organizacin de la Tesis
El presente trabajo se divide en 4 captulos. En el primer captulo se introducen los
antecedentes de los FDC aplicados a redes de distribucin. En el captulo 2, se presenta el
marco metodolgico desarrollado a lo largo del proyecto. En el tercer captulo, se describe el
proceso de implementacin del algoritmo: empezando por la arquitectura de diseo utilizada
y la descripcin de los mdulos desarrollados. En el captulo 4 se habla de los casos de estudio:
el primero, es un estudio comparativo de los tiempos de cmputo del FDCT y TRX al ser
aplicados en redes de 4, 7, 12 y 69 nodos. Se mostrar la eficiencia del segundo respecto
al primero; en segundo lugar, se realiza el caso de 7 barras paso por paso para ilustrar el
proceso de clculo del TRX ; en tercer lugar, se validan los resultados obtenidos en el caso
de 7 barras utilizando un NR eficiente, (MATPOWER); en cuarto lugar, se implement a
gran escala en SDEE y se compararn los tiempos de cmputo del TRX con el programa
utilizado por la EDC para este tipo de estudios. Finalmente, se presentan las conclusiones y
recomendaciones pertinentes. Los anexos se encuentran al final del documento.
CAPTULO 1
ANTECEDENTES
El Sistema Elctrico de Potencia, SEP, se encarga de la generacin, transmisin y distribucin de la energa elctrica. Como consecuencia del gran tamao y la alta complejidad
en funciones de dicho sistema; ste se encuentra dividido en dos grandes sub-sistemas:
1.1.
El SDEE presenta una serie de caractersticas especficas que lo diferencian considerablemente del sistema de transmisin. Estas caractersticas se presentan a continuacin:
Relacin x/r. Los conductores de la red de distribucin presentan una baja relacin
x/r, ya que x r. Mientras que, en los SEDT se tiene que normalmente x >> r; razn
6
sobre la cual se basa el Principio de Desacople:
los cambios en la potencia activa deben manifestarse sobre los ngulos de fases del
sistema (...) los cambios de potencia reactiva deben reflejarse en las magnitudes de
tensin [38, pg. 159].
Debido a la baja relacin x/r presente en las redes de distribucin, no se cumple dicho
principio. Razn por la cual, el Newton-Raphson, NR, tiene problemas de convergencia
y el Newton-Raphson Desacoplado, NRD, no es funcional.
Diversidad de Cargas. Debido a la diversidad de consumidores en la red es posible
encontrar diversos tipos de demanda, desde zonas rurales con densidades del orden de
kV A/km2 , hasta zonas urbanas con densidades en el orden de M V A/km2 [3]. Adems,
el mismo circuito puede ser usado para suplir cargas residenciales, comerciales y/o
industriales.
Cargas Desbalanceadas. Usualmente se emplean acometidas de dos, tres y cuatro
hilos para alimentar cargas tanto trifsicas como bifsicas y/o monofsicas. Esta caracterstica tiende a desbalancear las redes de distribucin.
1.2.
El FDC es un algoritmo que permite calcular las tensiones nodales, en mdulo y ngulo,
en rgimen permanente de un SEP. El rgimen permanente es aquel estado cuasiestacionario
en el cual existe un equilibrio de las potencias y las variables de tensin y frecuencia no presentan variaciones significativas [38]. Es bien conocido que la carga vara constantemente en
el tiempo, razn por la cual se utiliza el termino cuasiestacionario. Sin embargo, se entiende
que el FDC analiza el sistema en estados puntuales [3]. Tambin es utilizado en el rea de
planificacin por la capacidad de introducir suposiciones de estados futuros de la red. Razn
por la cual la herramienta posee una gran versatilidad y extraordinario potencial en el anlisis
de SEP.
7
Existen dos tipos de datos de entrada para el algoritmo, stos son:
Datos de Red. Hacen referencia a las especificaciones de las conexiones y a los
parmetros de las lneas, transformadores y compensadores presentes en el circuito.
Datos de Nodos. Hacen referencia a las especificaciones de los datos operacionales
de cada nodo, como tipo de nodo, su potencia consumida o generada, su magnitud de
tensin, etc.
El problema es no-lineal y no puede ser resuelto analticamente. Razn por la cual, se
recurre a tcnicas numricas iterativas para hallar la solucin [3]. Las ecuaciones que describen
al problema se pueden plantear como se presenta en las ecuaciones 1.1 y 1.2, [37].
Pi = |Vi |2 Gii +
n
X
(1.1)
j=1
Qi = |Vi | Bii
n
X
(1.2)
j=1
Donde, i, j son los nodos del sistema. n representa el nmero de nodos del sistema. Pi y
Qi es la potencia activa y reactiva en el nodo i. Vi y Vj son las tensiones en los nodos i y j.
Gii y Bii son la conductancia y la susceptancia propias del nodo i. Yij es la admitancia entre
los nodos i-j. ij es el ngulo de la admitancia de una lnea entre los nodos i-j. i y j son los
ngulos de las tensiones en los nodos i y j, respectivamente. En [37] se encuentra con mayor
detalle las ecuaciones planteadas en los FDC.
Una vez calculadas las tensiones, es posible calcular las corrientes por el circuito, flujos
de potencia, capacidades amperimtricas de conductores y prdidas en la red.
8
Tipos de Nodos Dependiendo de las condiciones de contorno que se especifiquen, se
pueden clasificar los nodos del sistema en tres grupos; [37]:
1.2.1.
9
Su simplicidad matemtica es notoria, no requirindose la inversin de matrices en lo absoluto por lo cual su programacin digital es muy rpida [38, pg. 109]. En [3, 38] se encuentra la
descripcin de las ecuaciones utilizadas por el FDC y la descripcin del proceso algortmico
del mismo.
En la literatura [4, 11, 14, 18, 19, 33, 39, 40, 41, 42, 43, 44] es posible encontrar una
amplia discusin sobre los problemas de convergencia o tiempos ineficientes de convergencia
de estos mtodos al ser aplicados en redes del tipo enfermas. Las redes de distribucin, por
su estructura radial y baja relacin x/r entran dentro de esta categora.
1.2.2.
Debido a los problemas presentados por los FDC NR y GS al ser aplicados en redes de
distribucin, se ha hecho ms frecuente el uso del FDC de barrido [16]. Estos algoritmos
sacan provecho de la topologa, tpicamente radial, de la red. Este mtodo consiste en un
proceso iterativo en el cual se computan tensiones y corrientes mediante evaluaciones secuenciales, llamados barridos. Est compuesto por dos etapas: barrido hacia atrs (backward
sweep) desde las cargas hacia la fuente; y el barrido hacia adelante (forward sweep) desde la
fuente hacia las cargas [2].
En el backward sweep se calculan las corrientes de rama y/o potencias del sistema, partiendo desde el nodo ms lejano a la fuente hasta llegar al ms cercano a sta. En el forward
sweep se calculan los voltajes en las barras del sistema, partiendo desde la barra ms cercana
a la fuente hasta la ms lejana a sta [2]. De esta forma, se utiliza el backward sweep para
actualizar las corrientes y/o potencias del sistema, y con el forward sweep se actualizan los
voltajes en las barras.
10
1.2.3.
Vs Vr
V2
cos(Z s + r ) r cos(Z )
Z
Z
Vs Vr
Vr2
Qr =
sin(Z s + r )
sin(Z )
Z
Z
Pr =
(1.3)
(1.4)
cos2 (Z s + r ) + sin2 (Z s + r ) = 1
(1.5)
(1.6)
11
v
u
u
(P 2 + Q2s Z 2 )
Vr = tVs2 2(Ps R + Qs X) + s
2
Vs
(1.7)
Tambin, se puede aplicar la ley de Kirchhoff de voltajes (KVL) para obtener los voltajes
en las barras del sistema. Aplicando KVL en el circuito de la figura 1.1. En la literatura
es posible encontrar varios algoritmos desarrollados en base a la ecuacin cuadrtica. En el
presente trabajo se enfoca la investigacin en los FDCB basados en las leyes de Kirchhoff.
Vs = Vr + Is .Z
(1.8)
Vr = Vs Is .Z
(1.9)
Muchos de los algoritmos usados en sistemas de distribucin ([11]-[21]) usan las leyes
de Kirchhoff (KVL y KCL, por sus siglas en ingls) para calcular las corrientes de rama
y los voltajes nodales en el backward y forward sweep, respectivamente. En [11], los autores presentan un mtodo de compensacin para redes de distribucin balanceadas radiales
y/o dbilmente malladas. Para esto, utilizan una tcnica de compensacin multipuertos y
la formulacin bsica de las leyes de Kirchhoff. Los sistemas radiales son resueltos usando
un procedimiento de dos pasos: las corrientes de rama son calculadas (backward sweep) y
luego se actualizan los voltajes en los nodos usando la ecuacin 1.8 para cada rama (forward
sweep). Se utiliza la diferencia de la potencia activa y reactiva en las cargas entre iteraciones
como criterio de convergencia. En [12] se aplica el algoritmo de [11] en redes desbalanceadas
(radiales o dbilmente malladas).
En los algoritmos [13] y [14] se hacen pequeas modificaciones a los algoritmos mencionados anteriormente. En stos, los circuitos radiales son resueltos tal como lo explica [11] y se
usa la diferencia de las tensiones entre iteraciones como criterio de convergencia. Sin embargo,
en [13] se incorpora el modelo de trasformadores trifsicos, propuesto por [45], en el anlisis.
Mientras que, en [14] se describe un algoritmo computacional que permite encontrar el valor
12
exacto de las corrientes en todas las ramas del circuito.
En [15], se propone una versin modificada del mtodo de barrido para redes radiales
balanceadas o desbalanceadas. En el backward sweep cada corriente de rama es calcula usando KCL. Luego, conociendo las corrientes, se calculan los voltajes en las barras, usando la
ecuacin 1.9, durante el forward sweep. La magnitud del voltaje en cada barra es comparada
con el valor de la iteracin anterior. Si el error est dentro de determinada tolerancia el proceso se detiene (criterio de convergencia). En caso contrario, el proceso de barrido contina
hasta que todas las tensiones cumplan dicho criterio. En el algoritmo propuesto en [16] se
calculan las corrientes de cada rama y, usando 1.8, se calculan los voltajes de barra. El voltaje
en la barra slack es calculado y se compara con un valor definido anteriormente. Si el error
est dentro de cierta tolerancia el proceso iterativo se detiene.
Liu et al, en [17], desarrollan un algoritmo aplicable a redes radiales y/o redes dbilmente
malladas. La parte radial es resuelta como se describe en [16]. Sin embargo, [17] se diferencia
de [16] porque primero se ajustan los voltajes en las barras a travs de un rango obtenido
por medio de la relacin entre el voltaje computado en la fuente y el especificado. Se utiliza
la diferencia de los voltajes entre iteraciones como criterio de convergencia.
.V K+1 = DLF.I K
(1.10)
13
V K+1 = V0 .V K+1
(1.11)
Este algoritmo es mejorado en [19] para redes radiales y redes dbilmente malladas. Para
esto se realizan algunas modificaciones en las matrices BIBC y BCBV. El criterio de convergencia utilizado es la diferencia de voltajes entre iteraciones. Una modificacin al algoritmo
anterior es realizada por [20]. En sta se sigue la misma lgica utilizada por [19] pero se introduce un cambio para poder realizar todos los clculos con nmeros reales, dicho algoritmo
es denominado TRX. Con esta modificacin se mejora la eficiencia del algoritmo propuesto
por [19].
CAPTULO 2
METODOLOGA
2.1.
Algoritmos Implementados
Sistema balanceado.
Datos del sistema transformado en por unidad (p.u) en una base comn.
15
Sistema exento de nodos PV.
Numeracin del nodo slack como el nmero 1.
Redes de distribucin radiales.
Para todos los algoritmos implementados, el proceso de numeracin de los nodos se realiza
como se describe en [11].
2.1.1.
Este algoritmo se caracteriza por realizar los clculos en nmeros complejos, como la
mayora de los algoritmos de FDC. Adems, utiliza una matriz nica (DLF) para determinar
el estado del sistema y el proceso se realiza de forma unidireccional. La matriz [DLF ] refleja
las impedancias del sistema asociadas a la topologa del mismo.
Debido a las tcnicas utilizadas en este algoritmo; se tiene que procesos como el de inversin de la matriz Jacobiana o la construccin de la matriz de admitancias nodales ([Y ])
requerido en los FDC tradicionales ya no son necesarios. Lo que hace del FDCT un algoritmo
robusto y eficiente en tiempo, [19]. De acuerdo a [19], el FDCT presenta un gran potencial
para ser utilizado en aplicaciones de automatizacin de redes de distribucin. El artculo se
encuentra disponible en el apndice G.
Descripcin General
El proceso inicia numerando los nodos del circuito con el mtodo propuesto por [11]; segn
el cual se divide el circuito en capas y la primera va de la fuente hacia el(los) nodo(s) ms
16
cercano(s) a la misma, la segunda capa va desde stos ltimos hacia los nodos ms cercanos
a los mismos, se repite el proceso hasta haber dividido todo el circuito en capas. Luego se
enumeran los nodos empezando desde la capa nmero 1 hasta llegar a la ltima pero no se
puede realizar cambio de capas hasta que todos los nodos de la capa previa sean enumerados,
ver apndice F.
Luego, se calculan las corrientes inyectadas en los nodos para la iteracin K-sima, IiK ,
(ecuacin 2.1). Donde, Pi y Qi en la potencia activa y reactiva demandada en el nodo i, la
cual no cambia enter iteraciones. ViK es la tensin en el nodo i durante la iteracin K.
IiK = (
Pi + jQi
)
ViK
(2.1)
(2.2)
En el siguiente paso, se calcula la diferencia de voltaje entre iteraciones, [.V K+1 ], utilizando la ecuacin 2.3. Dicho clculo se realiza en forma matricial. [I K ] es el vector columna
de inyecciones de corriente.
(2.3)
Finalmente, se calculan las tensiones nodales 2.4. Donde, V K+1 es el vector de tensiones
nodales en la iteracin K + 1 y V0 es el vector de tensiones nodales iniciales, tpicamente la
tensin inicial en cada nodo se asume 1 p.u.
17
(2.4)
||ViK | |ViK1 ||
(2.5)
Procesos Algortmicos
18
Solucin del FDC
Paso 1. Enumerar los nodos del circuito utilizando el mtodo propuesto por Shirmohammadi, [11].
Paso 2. Leer datos referentes a las conexiones del sistema, potencia aparente en los
nodos e impedancias de las lneas.
Paso 3. Construir matrices BIBC y BCBV.
Paso 4. Pre-especificar V0 = Vslack e inicializar todas las tensiones al valor del nodo
slack.
Paso 5. Calcular IiK (2.1).
Paso 6. Calcular la matriz DLF (2.2).
Paso 7. Calcular la variacin de voltaje en la iteracin K + 1 (2.3).
Paso 8. Calcular la tensin en la iteracin K + 1 (2.4).
Paso 9. Verificar si se cumple (2.5).
Paso 10. En caso de no cumplirse (2.5), volver al paso 5.
2.1.2.
Este algoritmo se caracteriza por realizar los clculos en nmeros reales, lo cual optimiza
el proceso de cmputo. Adems, utiliza una matriz nica (TRX ) para determinar el estado del sistema y el proceso se realiza de forma unidireccional. La matriz [T RX] refleja las
impedancias del sistema asociadas a la topologa del mismo.
Este algoritmo es til al ser aplicado con propsitos de planificacin y evaluacin del sistema de distribucin en tiempo real. El estado del sistema se obtiene usando el histrico de
19
Figura 2.1: Diagrama de Flujo del Algoritmo Propuesto por Teng, [19].
las medidas tomadas al mismo; teniendo en cuenta su topologa en el presente y en el futuro.
El TRX resulta ms eficiente que el FDCT porque el mismo trabaja con nmeros reales a
diferencia del segundo lo hace en nmeros complejos. Desde un punto de vista computacional,
consume implica ms tiempo realizar operaciones en nmeros complejos que en nmeros
reales. El artculo se encuentra disponible en el apndice H.
Descripcin General
20
0
0
]
...Vxi0 ...Vxn
Vx0 = [Vx1
(2.6)
0
0
...Vyi0 ...Vyn
]
Vy0 = [Vy1
DR
DR1
0
=
...
DR2 . . .
..
.
0
0
..
.
DRn
0
...
DX1
DX2 . . .
0
DX =
..
0
.
0
0
0
..
.
DXn
(2.7)
(2.8)
T DR T
TRX =
T T DX T
T DX T
T T DR T
(2.9)
K
Ixi
I=
K
Iyi
(2.10)
K
K
K
K
Donde, Ixi
y Iyi
se obtienen de las ecuaciones, (2.11) y (2.12), respectivamente. Ixi
y Iyi
21
Qi son las potencias activa y reactiva, respectivamente. VxiK y VyiK son las componentes real
e imaginaria de la tensin en la barra i durante la iteracin K.
K
Ixi
Re{IiK }
Pi VxiK Qi VyiK
=
(VxiK )2 + (VyiK )2
(2.11)
Qi VxiK Pi VyiK
(VxiK )2 + (VyiK )2
(2.12)
K
Iyi
= Im{IiK } =
V = V0 T RX.I
(2.13)
El proceso se detiene cuando la diferencia de voltaje entre iteraciones es menor a determinada tolerancia, ecuacin (2.14).
||ViK | |ViK1 ||
(2.14)
Procesos Algortmicos
1. Paso 1. Leer datos referentes a las conexiones del sistema, potencia aparente en los
nodos e impedancias de las lneas.
2. Paso 2. Construir matrices T , DR y DX .
3. Paso 3. Pre-especificar V0 = Vslack e inicializar todas las tensiones al valor del nodo
slack, (2.6).
4. Paso 4. Calcular matriz de corrientes inyectadas, I.
5. Paso 5. Calcular la matriz T RX.
6. Paso 6. Calcular la tensin en la iteracin K + 1 (2.13).
22
7. Paso 7. Verificar si se cumple (2.14).
8. Paso 8. En caso de no cumplirse (2.14), volver al paso 4.
En la figura 2.2 se muestra el diagrama de flujo del algoritmo propuesto por De Oliveira
[20].
Figura 2.2: Diagrama de Flujo del Algoritmo Propuesto por De Oliviera, [20].
2.2.
Herramientas Computacionales
2.2.1.
MATPOWER, [52]
23
niendo un cdigo simple de entender y modificar.
MATPOWER presenta tres algoritmos para resolver FDC : un FDC estndar y dos FDCO). El FDC estndar est basado en el mtodo Newton-Raphson (NR), en el cual la matriz
Jacobiana es actualizada en cada iteracin. Este mtodo se describe con detalle en [37, 38].
Los otros algoritmos son variaciones del NR desacoplado rpido, NRDR, [46]. El algoritmo
basado en el NR tradicional presente un excelente rendimiento al ser aplicado en sistemas
de potencia de gran escala. Esto se debe a que el algoritmo trabaja con la esparsidad de
la matriz Jacobiana. Es decir, con esta tcnica se evita construir la matriz Jacobiana y su
posterior inversin.
El primer FDCO est basado en la funcin constr de Matlab, la cual usa una tcnica
de programacin cuadrtica sucesiva para trabajar con la matriz Hessiana del sistema. El
segundo algoritmo est basado en programacin lineal (LP, por sus siglas en ingls), [53].
Sin embargo, el desempeo de los FDCO del MATPOWER depende de muchos factores.
En primer lugar, la funcin constr utiliza un algoritmo que no preserva la esparsidad de la
matriz. Por lo tanto, el primer FDCO queda limitado a sistemas de potencia de poco tamao.
Por otro lado, el algoritmo basado en LP preserva la esparsidad de la matriz pero no le saca
provecho, [53]. El programa se describe con mayor detalle en [52].
2.2.2.
ASP, [54]
24
Compensacin capacitiva para mnima perdida y correccin de bajo voltaje.
Anlisis de sensibilidad de parmetros.
Simulacin de interrupciones y recuperacin con otros circuitos interconectados.
Configuracin para mnima perdida.
Muestra el nodo con mayor cada de tensin y los nodos en los que la tensin es menor a
0,95 p.u. El programa seala la ubicacin grficamente de los nodos con la caracterstica
arriba mencionada e indica su tensin.
Muestra el tramo con mayor Capacidad de Emergencia, CE, y seala con colores diferentes aquellos tramos en los que la demanda es superior a un 67 % de su CE, los que
superan su Capacidad Nominal, CN y aquellos que superan el 100 % de la CE.
Muestra un reporte en el que se encuentran las prdidas totales en kVAR y kW, tanto
en valores reales como en porcentaje ( %).
2.3.
Al conocer el estado del sistema, obtenido al aplicar el FDC, es posible calcular las
caractersticas de operacin del sistema estudiado; entre otros anlisis se puede estudiar la
operacin del sistema ante condiciones de fallas, disear y/o planificar posibles expansiones
o mejoras en la red, ente otros estudios. Slo se har nfasis en el clculo de las condiciones
de operacin de la red. En la literatura ([16], [37], [38]) es posible encontrar informacin
referente a otros estudios que se pueden realizar al conocer el estado de la red.
25
2.3.1.
Este criterio indica la mxima cada de tensin que puede ocurrir en circuitos primarios,
tanto areos como subterrneos. Los lmites permitidos se encuentran en la tabla 2.1.
Tabla 2.1: Cada de Tensin Mxima Permitida, [56]
Condiciones de Operacin
VMAX( %)
Normal
0, 95 < V < 1, 05
Emergencia
0, 92 < V < 1, 08
.Vi =
2.3.2.
VSlack Vi
100
VSlack
(2.15)
Con las ecuaciones 2.16 y 2.17 se pueden obtener las prdidas activas (Lmp ) y reactivas
(Lmq ) totales en el circuito de estudio, [48].
Lmp =
n X
n
1X
Gij [Vi2 Vi Vj cos ij ]
2 i=1 j=1
(2.16)
Lmq =
n X
n
1X
Bij [Vi2 Vi Vj sin ij ]
2 i=1 j=1
(2.17)
i, j = 1, 2, ..., n
i 6= j
26
Donde, i, j y n representan el nodo de salida, de llegada y el nmero de nodos del circuito,
respectivamente. Adems, Gij , Bij y ij se obtienen de 2.18, 2.19 y 2.20, respectivamente,
[48]. Gij , Bij , Rij y Xij son la conductancia, susceptancia, resistencia y reactancia entre los
nodos i-j. ij es el ngulo de la admitancia de una lnea entre los nodos i-j.
Rij
(Rij + (Xij )2
Xij
Bij =
2
(Rij ) + (Xij )2
ij = i j
Gij =
)2
(2.18)
(2.19)
(2.20)
Las prdidas en una subestacin (S/E ), LS/E , corresponde a la suma de las prdidas de
cada circuito de dicha S/E, ecuacin 2.21. En esta ecuacin, C corresponde al nmero total
de circuitos que la componen.
LS/E =
C
X
Lm
(2.21)
k=1
Las prdidas totales en la red, LT , corresponden a la suma de las perdidas por subestacin,
ecuacin 2.22. En esta ecuacin, S corresponde al nmero total de S/E que componen toda
la red.
LT =
S
X
Lm
(2.22)
k=1
2.3.3.
27
(2.23)
(2.24)
Sij =
(Pij )2 + (Qij )2
(2.25)
Gij , Bij y ij se obtienen de 2.18, 2.19 y 2.20, respectivamente. Sij corresponde al flujo
de potencia aparente del conductor en condiciones de operacin.
Para verificar que un conductor est operando en condiciones normales se debe cumplir
la relacin de la ecuacin 2.26, [56].
.Sijmax Sij
(2.26)
28
corresponde a un factor de seguridad; tpicamente corresponde a 67 %. Sijmax corresponde
al flujo mximo de potencia de un conductor ubicado entre los nodos i y j.
2.3.4.
CF = C.(
N
X
kV Am kV Amax )
(2.27)
m=1
CAPTULO 3
IMPLEMENTACIN
3.1.
Incompatibilidad de datos.
Duplicidad de Informacin.
Mayor trabajo del necesario.
Retardos administrativos.
Ineficiencia operativa.
Efectos acumulativos con tendencia al caos.
Debido a esto, se busca unificar y sincronizar la base de datos con la que cuenta dicha
empresa. Al unificar la base de datos se busca tener en la misma base de datos la informacin
suministrada por los sistemas SAP (Software de Gestin y Estrategia), GIS (Software de
Estimacin de Demanda) y SCADA (Software de registro de carga y data del sistema), ver
figura 3.1.
30
31
1. Acceso inmediato y fiable a la informacin requerida para el proceso.
2. Desarrollo de herramientas tcnicas especficas en forma modular, garantizando sostenibilidad y escalabilidad de la solucin tecnolgica.
3. Manejo de informacin tcnica a gran escala integrando los distintos sistemas existentes.
4. Eliminar dependencias en cuanto a plataformas de cdigo cerrado o propietario.
El presente trabajo forma parte de la primera etapa del proyecto; siendo el flujo de carga
(FDC ) desarrollado uno de los mdulos a implementar en el prototipo. En la segunda y
tercera etapa se definirn otras funciones de la herramienta computacional.
Como se explica en [50], este patrn se compone de tres capas que permiten separar los
mbitos de trabajo de la aplicacin. El nivel superior es la Vista, la misma corresponde a la
interfaz de usuario en la cual se realizan las interacciones con el operador del programa. En la
parte ms baja se encuentra la capa correspondiente al Modelo. En ste se gestionan todas
las interacciones y validaciones con las fuentes de datos. Usualmente este proceso se delega en
paquetes conocidos como Mapeadores Objeto-Relacionales (o ORM por sus siglas en ingls)
que proveen un nivel de acceso de alto nivel convirtiendo las interacciones con las bases de
32
33
34
3.2.
3.3.
Datos
Los archivos utilizados para la implementacin del programa en redes de gran tamao
fueron obtenidos de una data suministrada por la Electricidad de Caracas (EDC ) en el ao
2006; dichos archivos se encuentran en formato .DAT. Los mismos son archivos de datos
generados automticamente por los sistemas de adquisicin de datos y mediciones, como
35
En la EDC, los archivos .DAT son utilizados por el programa ASP, desarrollado por
el profesor Alberto Naranjo, el cual tiene como funciones principales realizar un anlisis
consecuente con el (FDC ), y clculo de corto circuito. Una vez que se ejecuta el (FDC ) se
muestran las magnitudes de variables elctricas como tensin y corrientes pertenecientes al
sistema elctrico ordenadas en columnas, [50].
36
37
3.4.
3.4.1.
Inicialmente, se desarroll un algoritmo para filtrar de los archivos .DAT los datos correspondientes a la potencia de las cargas, impedancias de lneas, tipos de nodo, distancia de
las lneas, factor de potencia, tensin nominal de operacin y topologa de la red.
En la seccin B.1 del apndice B se puede encontrar el cdigo del algoritmo desarrollado.
3.4.2.
Proceso Iterativo
Con las matrices TRX y S de todos los circuitos disponibles en memoria RAM se realiza
el proceso iterativo, descrito en la seccin 2.1.2 del captulo 2. Como es de esperar, por la
simplicidad de los clculos, los resultados son obtenidos en pocas iteraciones y en un perodo
de tiempo considerablemente reducido en comparacin con otros algoritmos de FDC. Como
resultado de este modulo se obtienen las tensiones en mdulo y ngulo de todos los nodos de
la(s) red(es) estudiada(s).
38
Conocidas las tensiones de la red se realiza el anlisis correspondiente al estudio de cadas de tensin (seccin 2.3.1), prdidas tcnicas de potencia (ver seccin 2.3.2), capacidad
amperimtrica de conductores (ver seccin 2.3.3) y capacidad firme (seccin 2.3.4) en condiciones normales de operacin. Adems, se muestra un sumario de los resultados obtenidos,
incluyendo tensiones nodales, prdidas tcnicas, cadas de tensin, cantidad de conductores
que operan por debajo del 67 % de su capacidad, cantidad de conductores que superan el
67 % y 100 % de su capacidad. Los resultados son almacenados en un archivo con la extensin .MAT. En el apndice C se encuentra el reporte que muestra el programa para un
circuito ejemplo. En este caso el circuito es el GRA_A01 correspondiente al circuito A1 de
la subestacin (S/E) Granada.
En la seccin B.2 del apndice B se puede encontrar el cdigo del algoritmo desarrollado.
CAPTULO 4
CASOS DE ESTUDIOS Y ANLISIS DE RESULTADOS
4.1.
Datos de Lnea El circuito est conformado por 6 lneas, descritas en la tabla 4.1. Los datos
se presentan en el siguiente orden: nodo de salida, nodo de llegada, resistencia y reactancia
40
de cada lnea en p.u. Las bases utilizadas son las siguientes: VBASE = 12, 47kV y SBASE =
100MVA
Tabla 4.1: Datos de Lnea (Ldat)
salida
llegada
r(p.u)
x(p.u)
0,0265
0,0462
0,1005
0,0693
0,0670
0,0462
0,0265
0,0462
0,1005
0,0693
0,0670
0,0462
Datos de Nodo Los datos de los 7 nodos se encuentran en la tabla 4.2. Los mismos se
presentan en el siguiente orden: 1. n es el nmero de la barra. 2. Tipo corresponde al tipo
de barra: 1 barra slack, 2 barra PV y 3 barra PQ. 3. P gen0 es el valor inicial de la potencia
activa generada (p.u). 4. Qgen0 es el el valor inicial de la potencia reactiva generada (p.u).
5. Pload es la potencia activa demandada (p.u). 6. Qload es la potencia reactiva demandada
(p.u). 7. V0 es la tensin inicial de la barra (p.u). 8. fcap es el factor de capacidad de la barra.
4.1.1.
Como se describe en la seccin 2.1.2 del captulo 2, para calcular la matriz es necesario
hallar la matriz que relaciona las corrientes inyectadas con las corrientes de rama, [T], y las
matrices diagonales de resistencias de lnea, [Dr], y reactancias de lneas, [Dx].
Matriz [T]
En el apndice A se explica paso a paso cmo construir la matriz [T]. En esta matriz se
representa la topologa del circuito estudiado. El nmero 1 implica la existencia de una lnea
41
Tabla 4.2: Datos de Nodos, (Bdat)
Tipo
P gen0 (p.u)
Qgen0 (p.u)
Pload (p.u)
Qload (p.u)
V0
fcap
0,1017
0,0635
0,0547
0,0342
0,0809
0,0596
0,1017
0,0635
0,0547
0,0342
0,0809
0,0596
1 0 0 0 0 0
0 1 1 1 1 1
1 0 0 0
0 1 1 1
0 0 1 0
0 0
T =
0 0
0 0
0 0 0 0 0 1
0,0265
0
0
0,1005
Dr =
0,0670
0,0265
0,1005
0,0670
42
Matriz diagonal de reactancias de lneas, Dx. En la ecuacin 2.8 se observa la forma de
la matriz.
0,0462
0
0
0,0693
Dx =
0,0462
0,0462
0,0693
0,0462
Matriz TRX
T DR T =
0,1005 0,1005
0,127 0,127
0,2275 0,127
0.0265
0
0
0
0
0
0
0,1005 0,1005 0,1005 0,1005 0,1005
T
T DX T =
0,1005 0,1005
0,127
0,1005 0,1005
0,127
0,1005 0,1005
0,127
-0.0462
0,127
0,194
43
T T DX T =
0,0693 0,0693
0,1155 0,1155
0,1848 0,1156
0.0462
0
0
0
0
0
0
0,0693 0,0693 0,0693 0,0693 0,0693
T
T DR T =
0,1005 0,1005
0,127 0,127
0,2275 0,127
0.0265
0
0
0
0
0
0
0,1005 0,1005 0,1005 0,1005 0,1005
4.1.2.
0,1005 0,1005
0,127
0,1005 0,1005
0,127
0,1005 0,1005
0,127
0,127
0,194
Proceso Iterativo
Valores iniciales
Los valores de potencia y tensiones iniciales utilizados para empezar proceso algortmico
se muestran en la tabla 4.3. Los datos se presentan en el siguiente orden: 1. nmero del nodo
2. Potencia activa demandada en p.u. (Pload ) 3. Potencia activa consumida en p.u. (Qload ) 4.
Componente real de la tensin inicial en p.u. (VR0 ) 5. Componente imaginaria de la tensin
inicial en p.u. (VI0 ).
44
Tabla 4.3: Valores de potencia y tensin inicial (V0 , P0 )
Nodo
Pload (p.u)
Qload (p.u)
VR0 (p.u)
VI0 (p.u)
0,1017
0,0635
0,0547
0,0342
0,0809
0,0596
0,1017
0,0635
0,0544
0,0342
0,0809
0,0596
fue 0, 001. Se debe tomar en cuenta que se utiliz un criterio de convergencia bastante
conservador por motivos acadmicos. Los resultados de las iteraciones se muestran en las
tablas 4.4, 4.5 y 4.6, respectivamente.
Primera Iteracin
En la tabla 4.4 se muestran los resultados de la primera iteracin. Al terminar la primera
iteracin se observa que no se cumple el criterio de convergencia. En la mayora de los casos
la variacin es mayor a la tolerancia predeterminada.
45
Tabla 4.4: Resultados de la Primera Iteracin
Nodo
Ir (p.u)
Ij (p.u)
Vr (p.u)
Vj (p.u)
Vr
Vj
-0,1017
0,0635
0,9944
-0,003
0,0056
0,003
-0,0547
0,0342
0,9452
-0,0005
0,054
0,0006
-0,0809
0,0596
0,9370
-0,0053
0,0631
0,0052
-0,1017
0,0635
0,9316
-0,0074
0,0684
0,0074
-0,0544
0,0342
0,9237
-0,0077
0,0762
0,0077
-0,0809
0,0596
0,9234
-0,0071
0,0766
0,0071
Segunda Iteracin
Tabla 4.5: Resultados de la Segunda Iteracin
Nodo
Ir (p.u)
Ij (p.u)
Vr (p.u)
Vj (p.u)
.Vr
.Vj
-0,1025
0,0642
0,9943
-0,003
0,0001
0,0001
-0,0579
0,0362
0,9411
-0,0005
0,0040
0,0003
-0,0867
0,0641
0,9324
-0,0059
0,0631
0,0008
-0,1097
0,0690
0,9265
-0,0074
0,0051
0,0001
-0,0592
0,0375
0,9180
-0,0082
0,0058
0,0001
-0,0881
0,0652
0,9172
-0,0075
0,0058
0,0001
46
Tabla 4.6: Resultados de la Tercera Iteracin
Nodo
Ir (p.u)
Ij (p.u)
Vr (p.u)
Vj (p.u)
.Vr
.Vj
-0,1042
0,0636
0,9943
-0,003
0,0001
0,0001
-0,0582
0,0363
0,9411
-0,0005
0,0001
0,0001
-0,0868
0,0642
0,9323
-0,0059
0,0001
0,0001
-0,1092
0,0676
0,9265
-0,0078
0,0001
0,0001
-0,0590
0,0367
0,9175
-0,0082
0,0005
0,0001
-0,0877
0,0643
0,9171
-0,0075
0,0001
0,0001
La tabla 4.7, muestra las tensiones del circuito en mdulo y ngulo. Adems se muestra
el porcentaje de cada de tensin respecto a la barra slack.
Tabla 4.7: Tensiones nodales en mdulo y ngulo
Nodo
|V |(p.u)
(V )(grados)
V( %)
0,9943
-0,1728
0,57
0,9411
-0,0204
5,89
0,9323
-0,0036
6,77
0,9265
-0,4823
7,35
0,9175
-0,5121
8,25
0,9171
-0,4685
8,29
Se observa que 5 nodos estn por debajo del criterio del 5 % de cada de tensin permitido
en condiciones de operacin normal, ver tabla 2.1. La mxima cada de tensin en el circuito
es de 8, 29 %.
47
4.2.
Validacin de Resultados
Para realizar la validacin se utiliz como caso de ejemplo el circuito explicado en la seccin 4.1 del presente captulo. Las tensiones nodales se muestran en la tabla 4.8. Los datos de
presentan de la siguiente forma: 1. Nombre del nodo. 2. Mdulo de tensin en p.u. (|V |). 3.
ngulo de tensin en grados (6 (V )). 4. Error porcentual del mdulo de las tensiones obtenidas
con el algoritmo implementado respecto a las del MATPOWER. 5. Error porcentual del ngulo de las tensiones obtenidas con el algoritmo implementado respecto a las del MATPOWER.
Nodo
|V |(p.u)
(V )(grados)
Error |V |( %)
Error 6 (V )( %)
NA
0,9943
-0,1737
0,5181
0,9407
-0,0206
0,0425
0,9708
0,9319
-0,0039
0,0429
7,6923
0,9260
-0,4699
0,0540
2,6389
0,9174
-0,4937
0,0109
3,7269
0,9171
-0,4526
3,5130
Se observa que el mayor error en el mdulo de las tensiones es slo del 0, 0540 %; mientras
48
que en el ngulo de las tensiones se tuve un error mximo correspondiente al 7, 6923 %. En el
apndice E, se encuentra el reporte generado por el MATPOWER como resultado del flujo
de carga del circuito de 7 barras explicado en la seccin 4.1.
4.3.
Estudio Comparativo
El caso de estudio fue resuelto utilizando una red de n nodos variando la variable n de
1.000 hasta 3.000 nodos. La figura 4.1, obtenida de dicho artculo, muestra el tiempo de
convergencia utilizado en el proceso iterativo de cada algoritmo. El grfico muestra que el
TRX presenta un mejor tiempo de cmputo que el FDCT y el NR.
El TRX es desde un punto de vista computacional ms eficiente que los otros dos algoritmos porque el proceso est basado en la suma y multiplicacin de nmeros reales que se
encuentran previamente alojados en memoria RAM. La matriz TRX no requiere ser actualizada en cada iteracin tal como se hace con la matriz Jacobiana del NR. Motivo por el cual,
49
Por otro lado, se decidi corroborar dicho estudio de forma independiente. Slo se comprob la eficiencia del (TRX ) respecto al FDCT. Dicho estudio se realiz tomando en cuenta
el nmero de iteraciones y el tiempo de computo de ambos algoritmos.
50
algoritmos ya que se basan en el mismo principio.
Para poder estudiar el tiempo de respuesta de ambos algoritmos, se forz el ciclo de iteraciones para que realizaran un total de 5.000 iteraciones. De esta forma, fue posible hacer la
medicin del tiempo en el que los dos algoritmos realizan las operaciones matemticas. Esto
se fundamenta en el hecho de que los procesadores actuales realizan estos procesos en tiempos
muy pequeos, lo cual dificulta su medicin. Sin embargo, al aumentar apreciablemente el
nmero de iteraciones se puede medir el tiempo del proceso.
En la tabla 4.9 se muestran los resultados obtenidos. Se observa que los tiempos se reducen
aproximadamente a la mitad al aplicar la modificacin de [20]. Para este estudio se utiliz
un computador de 2GB de memoria RAM y un procesador Intel Core 2 Duo.
Tabla 4.9: Tiempo de cmputo de [19] y [20] para circuitos de 4, 7, 12 y 69 barras
[20] (mseg)
[19] (mseg)
69
243,61
432,03
12
45,48
84,84
35,61
79,54
30,33
74,22
Los resultados demuestran que el cmputo en nmeros reales utilizado por el TRX es
ms eficiente en comparacin con el clculo en nmeros complejos utilizado por FDCT, a
pesar de que las matrices son de mayor dimensin en el TRX.
51
4.4.
4.4.1.
Tiempo de cmputo
Para probar la eficiencia del algoritmo se compar el tiempo de cmputo del mismo con
el utilizado por el programa ASP ; programa desarrollado por el profesor Alberto Naranjo.
Dicho programa es utilizado actualmente por la EDC para realizar estudios de planificacin
a corto y mediano plazo. La utilizacin de esta herramienta permite con ciertas limitaciones
obtener resultados en tiempo til, [50]. En la tabla 4.2 se muestra el tiempo de cmputo del
ASP. Para este estudio se utiliz un computador de 2GB de memoria RAM y un procesador
Intel Core 2 Duo.
Tabla 4.10: Tiempo de cmputo del ASP para red de 530 circuitos de la EDC
Tiempo Total(seg)
ASP
438,82
52
Figura 4.2: Tiempo de Cmputo del ASP para la Red de la Gran Caracas.
anlisis de condiciones de la red.
Tabla 4.11: Tiempo de cmputo del algoritmo implementado para red de 530 circuitos de la
EDC
Mdulo
Tiempo Total(seg)
Adquisicin de Datos
1.700
Proceso Iterativo
0,76
Resultados
3,22
A primera vista se observa que, en trminos generales, el ASP requiri de menor tiempo
para obtener un resultado del FDC. Sin embargo, es importante resaltar que varios hechos
relevantes que demostrarn que la afirmacin inicial no es del todo valida.
En primer lugar, hacer una comparacin en igualdad de condiciones ente ambos algorit-
53
Figura 4.3: Tiempo de Cmputo del Algoritmo Implementado para la Red de la Gran Caracas.
mos no es del todo vlido ya que el programa del profesor Naranjo es un programa compilado
en un lenguaje de bajo nivel como lo es DELPHI a diferencia del algoritmo implementado
que no ha sido compilado y utiliza MATLAB como interpretador.
Por otro lado, el mdulo que requiri de mayor tiempo de proceso fue el modulo de
adquisicin de datos. El tiempo utilizado fue de 1.700 segundos; es decir, el mdulo tard en
promedio 3, 21 segundos en procesar la data de cada circuito. No obstante, como se explic
en la seccin 3.4.1 del captulo 3, en este mdulo se realiza el filtrado de los datos y se colocan
las matrices TRX y S en memoria RAM para que estn disponibles en cualquier momento
que se requiera, lo cual optimiza el proceso iterativo.
Adems, es vlido acotar que el espacio ocupado por las matrices TRX y S de los 530
circuitos es de 40, 6MB. Es decir, el espacio requerido para almacenar las matrices de los
54
circuitos de una red de gran tamao es insignificante en comparacin con la capacidad que
poseen los procesadores actualmente. En la tabla 4.12, se muestra el espacio ocupado en disco
por 10, 50, 100, 200 y 530 circuitos elegidos aleatoriamente. Los datos son presentados de la
siguiente forma: 1. Nmero de circuitos. 2. Espacio ocupado en disco en MB. 3. Nmero de
nodos totales.
Tabla 4.12: Espacio Ocupado en Disco por redes de 10, 50, 100, 200 y 530 circuitos
Nmero de Circuitos
Espacio en Disco(MB)
Nmero de Nodos
10
2,029
1.168
50
3,78
5.934
100
6,6
11.346
200
20
23.115
530
40,6
64.251
Por otra parte, se observa que el tiempo requerido para obtener el resultado del FDC
de toda la red estudiada fue 0, 76 segundos utilizando un computador de uso comn. Es
decir, en promedio se obtuvo el resultado del FDC de cada circuito en 1, 43 milisegundos.
Al compararlo con el ASP, que en promedio tena la solucin en 0, 75 segundos por circuito,
se puede concluir que el TRX es ms eficiente. Dicha eficiencia se debe a la sencillez de la
lgica de clculo, al hecho de que los clculos se realizan en nmeros reales y a la disposicin
de la informacin de los circuitos (en forma matricial) en memoria RAM.
4.4.2.
55
tmano. Para este circuito se estudi la cada de tensin, la capacidad amperimtrica y las
prdidas tcnicas.
La figura 4.4 muestra el perfil de la cada de tensin en el circuito Antmano A04, conformado por un total de 318 nodos. Es lgico obtener que a medida que los nodos se alejan de la
barra Slack, stos tengan un mayor porcentaje de cada de tensin. Sin embargo, la mxima
cada de tensin de este circuito es de 15 %, valor que supera el criterio de cada de tensin
del 5 % de la EDC, ver seccin 2.3.1 del captulo 2.
La figura 4.6 muestra la potencia reactiva demandada y las prdidas tcnicas en relacin
con la potencia activa total entregada. La potencia reactiva entregada al circuito es de
11, 72MVAr, de los cuales un 20 % representa las prdidas tcnicas en el circuito, valor que
est por encima del 4 % permitido, [56].
56
Figura 4.5: Potencia Activa Demandada y Prdidas Tcnicas del Circuito AN T _A01.
Figura 4.6: Potencia Reactiva Demandada y Prdidas Tcnicas del Circuito AN T _A01.
57
La figura 4.7 muestra la cantidad de conductores del circuito que estn operando por debajo del 67 % de la capacidad de emergencia (CE ), aquellos que operan por encima del 67 %
de la CE y aquellos que operan por encima del 100 % de la CE. La capacidad amperimtrica
de los conductores se calcul como se explica en la seccin 2.3.3 del captulo 2. Para un total
de 318 ramas, se observa que 7 (2 %) de los conductores superan el 100 % de la CE, 141
(44 %) de los conductores operan por encima del 67 % de la CE. Slo 171 (54 %) operan por
debajo del 67 % de la CE. Esto indica que se debe tratar de equilibrar la distribucin de las
cargas en el circuito para reducir el nmero de conductores que operan por encima 67 % o
del 100 % de la CE.
58
Anlisis Ejemplo para 1 S/E
Conocidas las tensiones nodales en mdulo y ngulo se realiz un anlisis de las condiciones operacionales de la S/E Antmano (AN T ). Para esta S/E se estudi la capacidad
amperimtrica de los conductores de los circuitos que conforman dicha S/E, se compar la
potencia entregada por circuito respecto a las prdidas tcnicas de los mismos en ( %) y
estudi la capacidad firme de dos S/E: Antmano AN T y Santa Rosa SRO.
En la figura 4.8 se muestra la potencia entregada junto con el porcentaje de prdidas por
circuito de la S/E Antmano. Se observa que el circuito con mayor potencia entregada es el
AN T _B04 con 22MVA; mientras que, el que menos potencia demanda es el AN T _B02 con
2, 3MVA. Por otro lado, el circuito que presenta ms prdidas es el AN T _A012 con 3, 1MVA
en prdidas tcnicas. Para este anlisis no se incluyen prdidas en los transformadores.
59
Para finalizar, en la figura 4.10 se muestra la potencia entregada por 2 S/E: Antmano y
Santa Rosa. La potencia total de cada circuito se calcul sumando la potencia suministrada
por cada circuito perteneciente a dicha S/E. Como caso ejemplo, se asume que cada S/E
cuenta con 4 transformadores de 33, 3MVA. Aplicando el criterio de capacidad firme CF,
explicado en la seccin 2.3.4 del captulo 2, se obtiene que la CF de las S/E debe ser aproximadamente 100MVA. Las potencias suministradas por los circuitos se compararon con el
valor terico de CF por S/E.
La potencia suministrada por las S/E AN T es 114MVA y SRO es 49, 5MVA, respectivamente. Se observa que la S/E AN T supera el valor terico por 14MVA mientras que la
S/E SRO est aproximadamente 50MVA por debajo de dicha capacidad. Se podra pensar
60
La figura 4.11 muestra la potencia activa demandada y las prdidas tcnicas en relacin
con la potencia activa total entregada. Los clculos se realizaron como se explica en la seccin
2.3.2 del captulo 2. La potencia activa entregada al circuito es de 3, 28GW, de los cuales un
558MW representa las prdidas tcnicas en el circuito, valor que est por encima del porcentaje de prdidas permitido. En promedio el valor de las prdidas para toda la red oscila
alrededor del 2 %, [54]. Sin embargo, se debe recordar que la data con la que se trabaj estaba
corrupta. Por lo tanto, los resultados ac mostrados se vieron afectados por el estado de la
61
data con la que se trabaj.
62
Parmetro Analizado
Resultado
557, 27MW
808, 25MVAr
3, 28GW
2, 03GVAr
24.227
22.130
64.251
CONCLUSIONES
64
Comentarios y Recomendaciones
Debido a todos los problemas ocasionados por la data de los circuitos de la EDC se
recomienda depurar la data existente, haciendo especial nfasis en los datos de las
impedancias.
Se recomienda ir a un modelo comn de datos en XML.
A fin de tener un registro de las cargas y poder tener disponibles sus respectivas curvas
se recomienda tipificar el consumo de los clientes.
REFERENCIAS BIBLIOGRFICAS
[1] Tinney, W. F., Hart, C. E., Power Flow Solution by Newtons Method, IEEE
Transactions on Power Apparatus and Systems, Vol. 86, No. 11, pp. 1449-1460, Nov.
1967.
[2] Eminoglu, U., Hocaoglu, M. H., Distribution Systems Forward/Backward Sweep-based
Flow Algorithms: A Review and Comparison Study, Electric Power Components and
Systems, Vol. 37, No. 1, 2009, pp. 91-110.
[3] Rypin Cazal, F., Barrenechea Gruber, R., Flujo de Carga en los Sistemas Integrales
de Distribucin, Proyecto de Grado presentado a la Universidad Simn Bolvar, Mar.
2006.
[4] Iwamoto, S., Tamura, Y., A Load Flow Calculation Method for Ill-Conditioned Power
Systems, IEEE Transactions on Power Apparatus and Systems, Vol. PAS-100, No. 4,
pp. 1736-1743, Abr. 1981.
[5] Tripathy, S. C., Durga, G., Malik, O. P., Hope, G. S., Load-Flow Solutions for IllConditioned Power Systems by Newton-Like Method, IEEE Transactions on Power
Apparatus and Systems, Vol. PAS-101, No. 10, pp. 3648-3657, Oct. 1982.
[6] Zhang, F., Cheng, C. S., A Modified Newton Method for Radial Distribution System
Power Flow Analysis, IEEE Transactions on Power Systems, Vol. 12, No. 1, pp. 389-397,
Feb. 1997.
[7] Da Costa, V. M., Martins, N., Pereira, J. L., Development in the Newton Raphson
Power Flow Formulation Based on Current Injections, IEEE Transactions on Power
Systems, Vol. 14, No. 4, pp. 1320-1326, 1999.
65
66
[8] Garcia, P. A. N., Pereira, J. L. R., Carneiro, S., Da Kosta, V. M., Martins, N., ThreePhase Power Flow Calculations Using the Current Injection Method, IEEE Transactions
on Power Systems, Vol. 15, No. 2, pp. 508-514, May. 2000.
[9] Lin, W. M., Teng, J. H., Three-Phase Distribution Network Fast-Decoupled Power Flow
Solutions, Electrical Power and Energy Systems, Vol. 22, No. 5, pp. 375-380, 2000.
[10] Teng, J. H., A Modified Gauss-Seidel Algorithm of Three-Phase Power Flow Analysis
in Distribution Networks, Electrical Power and Energy Systems, Vol. 24, No. 2, pp.
97-102, 2002.
[11] Shirmohammadi, D., Hong, H. W., Semlyen, A., Luo, G. X., A Compensation-Based
Power Flow Method for Weakly Meshed Distribution and Transmission Networks, IEEE
Transactions on Power Systems, Vol. 3, No. 2, pp. 753-762, May. 1988.
[12] Cheng, C. S., Shirmohammadi, D., A Three-Phase Power Flow Method for Real-Time
Distribution System Analysis, IEEE Transactions on Power Systems, Vol. 10, No. 2,
pp. 671-679, May. 1995.
[13] Zhu, Y., Tomsovic, K., Adaptive Power Flow Method for Distribution Systems with
Dispersed Generation, IEEE Transactions on Power Delivery, Vol. 17, No. 3, pp. 822827, Jul. 2002.
[14] Ghosh, S., Das, D., Method for Load-Flow Solution of Radial Distribution Networks,
IEE Proc. Generation Transmission and Distribution, Vol. 146, No. 6, pp. 641-648, 1996.
[15] Thukaram, D., Banda, H. M. W., Jerome, J., A robust Three Phase Power Algorithm
for Radial Distribution Systems, Electrical Power Systems Research, Vol. 50, No. 3, pp.
227-236, 1999.
[16] Kersting, W. H., Distribution System Modeling and Analysis, CRC Press, Nuevo Mxico, pp. 269-302 (2002).
[17] Liu, J., Salama, M. M. A., Mansour, R. R., An Efficient Power Flow Algorithm
for Distribution Systems with Polynomial Load, International Journal of Electrical
Engineering Education", Vol. 39, No. 4, pp. 371-386, 2002.
67
[18] Teng, J. H., A Network-Topology Based Three-Phase Load Flow for Distribution
Systems, Proc. Natl. Sci. Counc. ROC (A), Vol. 24, No. 4, pp. 259-264, 2000.
[19] Teng, J. H., A Direct Approach for Distribution System Load Flow solution, IEEE
Transactions on Power Delivery, Vol. 18, No. 3, pp. 882-887, Jul. 2003.
[20] De Oliveira, P. M., The Distribution TRX-Power Flow Method, Universidad Simn
Bolvar, artculo aprobado para publicacin, 2010.
[21] Augugliaro, A., Dusonchet, L., Favuzza, S., Ippolito, M. G., and Sanseverino, E. R.,
A New Backward/Forward Method for Solving Radial Distribution Network with PV
Nodes, Electrical Power Systems Research, Vol. 78, No. 3, pp. 330-336, Mar. 2008.
[22] Cspedes, G. R., New Method for the Analysis of Distribution Networks, IEEE
Transactions on Power Delivery, Vol. 5, No. 1, pp. 391-396, Ene. 1990.
[23] Haque, M. H., Load Flow Solution of Distribution Systems with Voltage Dependent
Load Models, Electrical Power Systems Research, Vol. 36, No. 3, pp. 151-156, 1996.
[24] Haque, M. H., Efficient Load Flow Method for Distribution Systems with Radial or
Mesh Configuration, IEE Proc. Generation Transmission and Distribution, Vol. 143,
No. 1, pp. 33-38, 1996.
[25] Haque, M. H., A General Load Flow Method for Distribution Systems, Electrical Power
Systems Research, Vol. 54, No. 1, pp. 106-114, 2000.
[26] Asfari, M., Singh, S. P., Raju, G. S., Rao, G. K., A Fast Power Flow Solution of Radial
Distribution Networks, Electric Power Components and Systems, Vol. 31, No. 1, pp.
1065-1074, 2002.
[27] Ranjan, R., Das, D., Simple and Efficient Computer Algorithm to Solve Radial
Distribution Networks, Electric Power Components and Systems, Vol. 31, No. 1, pp.
95-107, 2003.
[28] Das, D., Nagi, H. S., Kothari, D. P., Novel Method for Solving Radial Distribution
Networks, IEE. Proc. Generation Transmission and Distribution, Vol. 141, No. 4, pp.
291-298, 1994.
68
[29] Das, D., Kothari, D. P., Kalam, A., Simple and Efficient Method for Load Flow Solution
of Radial Distribution Networks, Electrical Power and Energy Systems, Vol. 17, No. 5,
pp. 335-346, 1995.
[30] Eminoglu, U., Hocaoglu, M. H., A New Power Flow Method for Radial Distribution
Systems Including Voltage Dependent Load Models, Electrical Power Systems Research,
Vol. 76, No. 1-3, pp. 106-114, 2005.
[31] Baran, M. E., Wu, F. F., Optimal Sizing of Capacitors Placed on a Radial Distribution
System, IEEE Transactions on Power Delivery, Vol. 4, No. 1, pp. 735-742, 1989.
[32] Baran, M. E., Wu, F. F., Network Reconfiguration in Distribution Systems for Loss
Reduction and Load Balancing, IEEE Transactions on Power Delivery, Vol. 4, No. 2,
pp. 1401-1407, Abr. 1989.
[33] Luo, G. X., Semlyen, A., Efficient Load Flow for Large Weakly Meshed Networks,
IEEE Transactions on Power Systems, Vol. 5, No. 4, pp. 1309-1313,1990.
[34] Jasmon, G. B., Lee, L. H. C., Distribution Network reduction for Voltage Stability
Analysis and Load Flow Calculations, Electrical Power and Energy Systems, Vol. 13,
No. 1, pp. 9-13, 1991.
[35] Rajicic, D., Ackovski, R., Taleski, R., Voltage Correction power Flow, IEEE
Transactions on Power Delivery, Vol. 9, No. 2, pp. 1056-1062, 1994.
[36] Satyanarayana, S., Ramana, T., Sivanagaraju, S., and Rao, G. K., An Efficient Load
Flow Solution for Radial Distribution Network Including Voltage Dependent Load
Models, Electric Power Components and Systems, Vol. 35, No. 5, pp. 539-551, 2007.
[37] Graigner, J. J., Stevenson, W. D., Anlisis de Sistemas de Potencia, MC Graw Hill,
Mxico, pp. 309-357 (2004).
[38] Bermdez, J. F., Anlisis de Flujo de Carga en sistemas Elctricos de Potencia, Versin
B, Mar. 1994.
[39] Rajicic, D., Bose, A., A Modification to the Fast decoupled Power Flow for Methods
with High R/X Ratios, IEEE Transactions on Power Systems, Vol. 3, No. 2, pp. 743746,1988.
69
[40] Baldwin, T. L., Lewis, S. A., Distribution Load Flow Methods for Shipboard Power
Systems, Vol. 40, No. 5, pp. 812-819, Sept. 2004.
[41] Lin, W. M., Su, Y. S., Chin, H. C., Teng, J. H., Three-phase unbalanced distribution
power flow solutions with minimum data preparation, IEEE Transactions on Power
Systems, Vol. 14, No. 3, pp. 1178-1183, Aug. 1999.
[42] Expsito. G. A., Augmented Rectangular Flow Model, IEEE Transactions on Power
Systems, Vol. 17, No. 2, May. 2002.
[43] Expsito. G. A., Ramos, E. R., Reliable Load Flow Technique for Radial Distribution
Networks, IEEE Transactions on Power Systems, Vol. 14, No. 3, pp. 1063-1069, Aug.
1999.
[44] Jovanovic, S., Milicevic, F., Triangular Distribution Load flow, IEEE Power
Engineering Review, pp. 60-62, May. 2000.
[45] Baran, M. E., Staton, E. A., Distribution Transformer Models for Branch Current Based
Feeder Analysis, IEEE Transactions on Power Systems, Vol. 12, pp. 698-703, 1997.
[46] Stott, B., Alsac, O., Fast Decoupled Load Flow, IEEE Transactions on Power
Apparatus and Systems, pp. 859-869, May/Jun. 1974.
[47] Younes, M., Rahli, M., Abdelhakem K, L., Economic Power Dispatch Using Evolutionary Algorithm, Journal of ELECTRICAL ENGINEERING, Vol. 57, No. 4, pp.
211-217, 2006.
[48] Expsito. G. A., Conejo. J. A., Caizares. C.,Electric Energy Systems, Analysis and
Operation, CRC Press, Florida, pp. 95-126 (2009).
[49] Alvarez, M., Anteproyecto para el Desarrollo de Herramienta Computacional de Sistemas Elctricos de Distribucin, INDENE-Parque Tecnolgico Sartenejas, 2010.
[50] Arteaga, A., Rojas, A., Ruiz C., Informe Final: CONTRATO N NCO/ 0909-589:
ANTEPROYECTO PARA EL DESARROLLO DE HERRAMIENTA COMPUTACIONAL PARA ESTUDIOS DE PLANIFICACIN CORTO-MEDIANO PLAZO DE
70
SISTEMAS ELCTRICOS DE DISTRIBUCIN"DOC. No. EDC2010 10 001,
INDENE-Parque Tecnolgico Sartenejas, 2010.
[51] PSS/ADEPT 4, User Manual, Power Technologies INC, Nueva York,pp. B1-B22
(Febrero 2002).
[52] Zimmerman, R., Murillo S., C. E., MATPOWER: A MATLAB Power System
Simulation Package, Users Manual, Power Systems Engineering Research Center
(PSerc), Nueva York, Versin 3.2 (Septiembre 2007).
[53] Younes, M., Rahli, M., Abdelhakem K, L., Economic Power Dispatch Using Evolutionary Algorithm, Journal of ELECTRICAL ENGINEERING, Vol. 57, No. 4, pp.
211-217, 2006.
[54] Rodrgues Goncalves, C., Estudio de Planificacin a Mediano Plazo de la S/E Esmeralda de 12,47KV de la Electricidad de Caracas, Proyecto de Grado presentado a la
Universidad Simn Bolvar, Ene. 2008.
[55] Naranjo, A., Gua de Comandos del Porgrama ASPV7A, Manual, Caracas, 2003.
[56] Pia, Y., Estudio de Planificacin a Mediano Plazo de dos Subestaciones de 100MVA
de la Regin Este de La Electricidad de Caracas, Proyecto de Grado presentado a la
Universidad Central de Venezuela, Nov. 2005.
Apndice A
CASO EXPLICATIVO: MATRICES BIBC Y BCBV
A.1.
Matriz BIBC
B1 = I2 + I3 + I4 + I5 + I6
71
(A.1)
72
B2 = I3 + I4 + I5 + I6
B3 = I4 + I5
B4 = I5
B5 = I6
De esta forma, la relacin entre las corrientes de rama y las corrientes de inyectadas se
puede escribir de forma matricial como se muestra en la ecuacin A.2.
B1
1 1 1 1 1 I2
0 1
B2
B = B3 =
0 0
0 0
B4
B5
I
1 1 1
3
1 1 0
I4
(A.2)
0 1 0 I5
0 0 0 0 1
I6
La ecuacin A.2 puede ser reescrita en forma general como se muestra en la ecuacin A.3
[B] = [BIBC][I]
(A.3)
BIBC es una matriz triangular superior que slo contiene 1 0. Adems, la matriz T del
TRX se calcula de igual forma que la matriz BIBC del FDCT.
A.2.
Matriz BCBV
Utilizando la Ley de Voltajes de Kirchhoff (KLV ) se pueden obtener los voltajes nodales
en funcin de las corrientes de rama. Para este caso,
V2 = V1 B1 Z12
V3 = V1 B1 Z12 B2 Z23
(A.4)
73
V4 = V1 B1 Z12 B2 Z23 B3 Z34
V5 = V1 B1 Z12 B2 Z23 B3 Z34 B4 Z45
V6 = V1 B1 Z12 B2 Z23 B5 Z36
De esta forma, la relacin entre los voltajes nodales y las corrientes de rama se puede
escribir de forma matricial como se muestra en la ecuacin A.5.
V1
V2
Z12
Z12
V1 V3
.V = V1 V4 =
Z12
Z12
V1 V5
V1
V6
0 B1
Z23
B
0
2
0
B3
Z23 Z34
Z12 Z23
0
Z36
(A.5)
B4
B5
La ecuacin A.5 puede ser reescrita en forma general como se muestra en la ecuacin A.6.
(A.6)
Apndice B
ALGORITMO IMPLEMENTADO TRX
Los algoritmos mostrados a continuacin fueron desarrollados e implementados utilizando un programa comercial; los mismos estn en formato .MAT. En el captulo 3 se puede
encontrar el diagrama de flujo del programa implementado.
B.1.
Algoritmo desarrollado para el filtrado y adquisicin de datos de los archivos .DAT previamente transformados a archivos .MAT. Adems, arma las matrices T RX y S; las guarda
en un archivo .MAT y las almacena en disco.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%% DECLARACIN DE VARIABLES
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
75
directorio(1)=[]; directorio(1)=[]; [T0 T1]=size(directorio); [Nc Tc
Rc]=xlsread(C:\Documents and
Settings\JR\Escritorio\integracion\conductor.xlsx,); zk=0; t=0;
w1=0; Vbase=0;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% CICLO QUE ABRE LOS ARCHIVOS DE LA CARPETA %%%%%%%%%%%%%%%%
for w=1:T0
tic;
clear N T R nodos0 lineas0
if strcmp(finfo(directorio(w).name),xlsx)
w1=w1+1;
archivo=directorio(w).name;
[N T R]=xlsread(archivo,);
pro=0;
i=0; fc=1; fp=0.85;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%% INICIO DE CILCO QUE BUSCA LOS DATOS %%%%%%%%%%%%
for j=1:length(T)
%%%%% Busco datos de Tension, potencia y factor de potencia del sistema. %%
if strcmp(T(j,1),END/TITLE)
Sbase=cell2mat(R(j+2,1));
Vbase=cell2mat(R(j+2,2));
fp=cell2mat(R(j+2,3));
if fp==0 || fp==1
fp=0.85;
end
end
if Vbase > 100
Vbase=Vbase/1000;
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
76
%%%%%%%%%%%%%%%% SE BUSCAN DATOS DE NODOS DEL SISTEMA %%%%%%%%%%%%%%%%%%%%%
if strcmp(T(j,1),END/PARAMS)
j=j+1;
while strcmp(T(j,1),END/NODES)==0
if cell2mat(R(j,2))==Vbase
i=i+1;
nodos0{i,2}=cell2mat(R(j,1));
end
j=j+1;
end
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% ASIGNA VALORES Y TIPO A LOS NODOS %%%%%%%%%%%%%%%%%%%%%%%%
if strcmp(T(j,1),END/NODES)
%Encuentra y crea la variable con la barra Slak
slak=T(j+1,1);
Q=size(nodos0,1);
for i0=1:Q
%potencia activa Generada
nodos0{i0,4}=0;
%potencia reactiva Generada
nodos0{i0,5}=0;
nodos0{i0,6}=0;
nodos0{i0,7}=0;
%Le pone numero a los nodos
nodos0{i0,1}=i0;
%Definir Barra Slak
if strcmp(nodos0{i0,2},slak)
nodos0{i0,3}=1;
else
nodos0{i0,3}=3;
77
end
end
end
%Armar la matriz de lineas
i=0;
i2=0;
if strcmp(T(j,1),END/SOURCE)
j=j+1;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% ENCUENTRA LOS DATOS DE LINEAS Y CARGAS %%%%%%%%%%%%%%%%%%%
while strcmp(T(j,1),END/BRANCH)==0
if strcmp(T(j,3),T)
i2=i2+1;
pro=1;
trans{i2,1}=R(j,1);
trans{i2,2}=R(j,2);
trans{i2,3}=(cell2mat(R(j,7)))*3;
trans{i2,4}=trans{i2,3}*fp*fc;
trans{i2,5}=sqrt((trans{i2,3})^2-(trans{i2,4})^2);
else
i=i+1;
lineas0{i,1}=cell2mat(R(j,1));
lineas0{i,2}=cell2mat(R(j,2));
if strcmp(T(j,3),L)
lineas0{i,3}=cell2mat(R(j,7));
lineas0{i,4}=R(j,6);
lineas0{i,5}=0;
lineas0{i,6}=0;
else
lineas0{i,3}=0;
lineas0{i,4}=0;
78
lineas0{i,5}=0;
end
end
j=j+1;
end
end
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% ENCUENTRA Y ASIGNA LA CARGA AL NODO ADECUADO %%%%%%%%%%%%%
clear i i0 j slak i2 K=0; if pro==1 [q y]=size(trans);
for i=1:size(nodos0,1)
for j=1:q
if strcmp(nodos0{i,2},trans{j,1})
nodos0{i,6}=trans{j,4}+nodos0{i,6};
nodos0{i,7}=trans{j,5}+nodos0{i,7};
K=K+1;
end
end
end
clear i j
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%% ENCUENTRA CONDUCTORES Y ADJUNTA VALORES DE IMPEDANCIA
%
%%%%%%%%%%%%%%%%%%
%%%%%
%
[wl rl]=size(lineas0);
for i=1:wl
for j=1:length(Tc)
if lineas0{i,3}~=0
if strcmp(lineas0{i,4},Tc{j,2})
lineas0{i,5}=Nc(j,3);
lineas0{i,6}=Nc(j,4);
79
lineas0{i,7}=(Nc(j,7))/((Sbase)/(Vbase*sqrt(3)));
end
end
end
if lineas0{i,5}==0
lineas0{i,5}=0.18026;
lineas0{i,6}=0.12779;
lineas0{i,7}=324.09/((Sbase)/(Vbase*sqrt(3)));
end
end
clear fc fp i j pro q rl trans u y wl K N Q R T T1
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% INICIO DEL FORMATEO DE LAS MATRICES BDAT Y LDAT %%%%%%%%%%
nl=size(lineas0,1); %nmero de lneas
nn=size(nodos0,1); %nmero de nodos
for k=1:nn
if cell2mat(nodos0(k,3))==1
bdat0{1,1}=1;
bdat0{1,2}=nodos0(k,2);
bdat0{1,3}=nodos0{k,3};
bdat0{1,4}=(cell2mat(nodos0(k,4)))/Sbase;
bdat0{1,5}=(cell2mat(nodos0(k,5)))/Sbase;
bdat0{1,6}=(cell2mat(nodos0(k,6)))/Sbase;
bdat0{1,7}=(cell2mat(nodos0(k,7)))/Sbase;
bdat0{1,8}=Vbase/Vbase;
end
end
%ciclo que ordena los datos de lnea y construye la matriz bdat0
%con base a la metodologa propuesta por Shirmohammadi[10].
%Esta metodologa es la base para la matriz BIBC del flujo de
%carga implementado.
80
l=2; k=0; for u=1:nn
for g=1:nl
if strcmp(bdat0{u,2},lineas0{g,2})
nlle=lineas0{g,2};
nsal=lineas0{g,1};
lineas0{g,1}=nlle;
lineas0{g,2}=nsal;
end
end
for k=1:nl
if strcmp(bdat0{u,2},lineas0{k,1})
for y=1:nn
if
strcmp(lineas0{k,2},nodos0{y,2})
bdat0{l,1}=l;
bdat0{l,2}=nodos0(y,2);
bdat0{l,3}=nodos0{y,3};
bdat0{l,4}=nodos0{y,4}/Sbase;
bdat0{l,5}=(nodos0{y,5})/Sbase;
bdat0{l,6}=(nodos0{y,6})/Sbase;
bdat0{l,7}=(nodos0{y,7})/Sbase;
bdat0{l,8}=Vbase/Vbase;
ldat0{l-1,1}=bdat0{u,1};
ldat0{l-1,2}=bdat0{l,1};
ldat0{l-1,3}=lineas0{k,7};
ldat0{l-1,4}=(lineas0{k,5}*lineas0{k,3})/
(Vbase^2/(Sbase/1000));
ldat0{l-1,5}=(lineas0{k,6}*lineas0{k,3})/
(Vbase^2/(Sbase/1000));
if ldat0{l-1,4}==0
ldat0{l-1,3}=0;
end
81
lineas0{k,1}=0;
lineas0{k,2}=0;
l=l+1;
end
end
end
end
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% CONVIERTE LAS CELDAS EN MATRICES %%%%%%%%%%%%%%%%%%%%%%%%%
clear g k l nl nlle nn nsal u y nll ns bdat ldat
bdat(:,1)=cell2mat(bdat0(:,1)); bdat(:,3)=cell2mat(bdat0(:,3));
bdat(:,4)=cell2mat(bdat0(:,4)); bdat(:,5)=cell2mat(bdat0(:,5));
bdat(:,6)=cell2mat(bdat0(:,6)); bdat(:,7)=cell2mat(bdat0(:,7));
bdat(:,8)=cell2mat(bdat0(:,8)); ldat(:,1)=cell2mat(ldat0(:,1));
ldat(:,2)=cell2mat(ldat0(:,2)); ldat(:,3)=cell2mat(ldat0(:,3));
ldat(:,4)=cell2mat(ldat0(:,4)); ldat(:,5)=cell2mat(ldat0(:,5));
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% SE INICIA EL ARMADO DE LA MATRIZ "T" Y LA "DLF" %%%%%%%%%%
m=size(ldat,1); %nmero de lneas
n=size(bdat,1); %nmero de nodos
T1(m,n)=0;
% matriz previa a T
ns=ldat(k,1);
% nodo de salida
82
end
for k=2:n %ciclo que crea la matriz T
T(:,k-1)=T1(:,k);
end
TR=T*diag(Vectr)*T; %Sub-matriz de Rs de la matriz TRX
TX=T*diag(Vectx)*T; %Sub-matriz de xs de la matriz TRX
TRX=vertcat(horzcat(TR,-TX),horzcat(TX,TR)); %Matriz TRX Matriz real.
DLF=TRX; S(:,1)=bdat(:,6); S(:,2)=bdat(:,7); for ll=1:size(bdat0,1)
nno(ll,1)=bdat0{ll,2}; end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%% SE CREA LA VARIABLE "RED" CON LOS DATOS DEL SISTEMA %%%%%%%%%%
RED{w1,1}=(directorio(w).name);
RED{w1,2}=DLF;
RED{w1,3}=S;
RED{w1,4}=nno;
RED{w1,5}=T;
RED{w1,6}=ldat;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
else
zk=zk+1;
malos{zk,1}=directorio(w).name;
end
bdat k
m n nll ns z
83
B.2.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%% DECLARACIN DE VARIABLES
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
84
(V0(k+(n-1),1))^2));
I(k+(n-1),1)=(-S(k+1,2)*V0(k,1)+S(k+1,1)*V0(k+(n-1),1))/(((V0(k,1))^2+
(V0(k+(n-1),1))^2));
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%% OPERACIONES PARA INICIALIZAR VECTORES DE ITERACION %%%%%%%%%%
DV=TRX*I; %Multiplicacin inicial para hallar el delta de tensin inicial
%para corregir Vo.
V1=V0-DV; %V1 coregida, V0-DV
conve=1; % inicializacion de variabel para convergencia.
Vp(1,1)=1; V(1,1)=1; ang(1,1)=0; DV2=DV; k=0;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%% CICLO ITERATIVO DEL FLUJO DE CARGA %%%%%%%%%%%%%%%%%%
tic; % INICIA EL CONTEO DE TIEMPO
while
85
ni=(size(I,1))/2; for L=1:ni
Ii(L,1)=I(L,1);
Ii(L,1)=Ii(L,1)+I(L+ni,1)*i;
end Ip=abs(T*Ii);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%% CICLO QUE ORDENA LOS VOLTAJES Y ANGULOS %%%%%%%%%%%%%%%
for M=2:n
V(M,1)=abs((V1(M-1+(n-1),1))+i*(V1(M-1,1)));
ang(M,1)=rad2deg(angle((V1(M-1+(n-1 ),1))*i+(V1(M-1,1))));
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%% GUARDA RESULTADOS EN LA VARIABLE "RED" %%%%%%%%%%%%%%%%%
RED{qp,6}(:,6)=Ip; RED{qp,7}(:,1)=V; RED{qp,7}(:,2)=ang(:,1);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%% CALCULO DE PERDIDAS POR CIRCUITOS, CAIDAS EN LAS RAMAS%%%%%%%%%%
K=size(RED{qp,6},1); for J=1:K
RED{qp,6}(J,7)=(RED{qp,6}(J,6))^2*(RED{qp,6}(J,4));
RED{qp,6}(J,8)=(RED{qp,6}(J,6))^2*(RED{qp,6}(J,5));
end J=0;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%% CALCULO DE VARIACION DE TENSION EN CADA UNO DE LOS NODOS%%%%%%%%
for J=2:n
RED{qp,7}(J,3)=((RED{qp,7}(1,1)-RED{qp,7}(J,1))/(RED{qp,7}(1,1)))*100;
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%% CALCULO DE NODOS POR CIRCUITOS Y TOTALES DE LA RED%%%%%%%%%%%%
K2=size(RED{qp,7},1);
%Calculamos el nmero de nodos totales del sistema.
npc=size(RED{qp,4},1); RED{qp,8}=npc; ntot=ntot+npc;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
86
clear DV DV2 I M S TRX V V0 V1 Vk Vp ang conve k m n K J npc ni Ii L
Ip T toc; t1=t1+toc; end ntot t t1 save CCS2 RED clear all
Apndice C
EJEMPLO DE REPORTE DEL ALGORITMO DESARROLLADO
# Nodo:
Nodo:
Tensin:
ngulo:
5201_1
1.000000
0.000000
0.000000
5201_2
0.996748
-0.011527
0.325188
5201_70
1.000000
0.000000
0.000000
5201_43
0.994400
-0.019907
0.559986
5201_3
0.993386
-0.023515
0.661434
5201_89
0.996708
-0.010766
0.329241
5201_71
1.000000
0.000000
0.000000
5201_51
0.993381
-0.023554
0.661860
5201_44
0.994213
-0.020578
0.578715
10
5201_5
0.993386
-0.023515
0.661434
11
5201_90
0.996708
-0.010766
0.329241
12
5201_55
0.993381
-0.023554
0.661860
13
5201_52
0.993381
-0.023554
0.661860
14
5201_45
0.994213
-0.020578
0.578715
15
5201_68
0.994213
-0.020578
0.578715
87
Cada:
88
16
5201_46
0.994213
-0.020578
0.578715
17
5201_7
0.990692
-0.033176
0.930784
18
5201_54
0.993381
-0.023554
0.661860
19
5201_53
0.993381
-0.023554
0.661860
20
5201_52
0.993381
-0.023554
0.661860
21
5201_53
0.993381
-0.023554
0.661860
22
5201_47
0.994213
-0.020578
0.578715
23
5201_49
0.994213
-0.020578
0.578715
24
5202_133
0.994213
-0.020578
0.578715
25
5201_8
0.990648
-0.032334
0.935247
26
5201_10
0.989429
-0.037725
1.057105
27
5201_53
0.993381
-0.023554
0.661860
28
5201_56
0.992600
-0.026356
0.740000
29
5201_48
0.994190
-0.020377
0.580982
30
5201_50
0.994207
-0.020526
0.579297
31
5201_9
0.990648
-0.032334
0.935247
32
5201_13
0.989429
-0.037725
1.057105
33
5201_57
0.992581
-0.025996
0.741908
34
5201_59
0.991759
-0.029377
0.824109
35
5201_11
0.989429
-0.037725
1.057105
36
5201_12
0.989429
-0.037725
1.057105
37
5201_58
0.992581
-0.025996
0.741908
38
5201_60
0.991530
-0.025054
0.847034
39
5201_62
0.991054
-0.031910
0.894558
40
5201_80
0.989352
-0.037042
1.064767
41
5201_19
0.988295
-0.041816
1.170520
42
5201_61
0.991530
-0.025054
0.847034
43
5201_63
0.991039
-0.031965
0.896062
44
5201_64
0.991054
-0.031910
0.894558
45
5201_14
0.989250
-0.036132
1.074988
46
5201_81
0.989144
-0.035187
1.085595
89
47
5201_20
0.988290
-0.041732
1.170967
48
5201_67
0.988237
-0.040723
1.176300
49
5201_23
0.987049
-0.046320
1.295086
50
5201_6
0.990475
-0.033996
0.952472
51
5202_49
0.991039
-0.031965
0.896062
52
5202_47
0.991039
-0.031965
0.896062
53
5201_65
0.991054
-0.031910
0.894558
54
5201_15
0.989249
-0.036102
1.075148
55
5201_17
0.989152
-0.035257
1.084806
56
5201_82
0.989144
-0.035187
1.085595
57
5201_21
0.988290
-0.041732
1.170967
58
5201_22
0.988237
-0.040723
1.176300
59
5201_24
0.987049
-0.046320
1.295086
60
5201_86
0.990454
-0.033595
0.954597
61
5201_25
0.990475
-0.033996
0.952472
62
5201_16
0.989249
-0.036102
1.075148
63
5201_18
0.989152
-0.035257
1.084806
64
5201_84
0.989144
-0.035187
1.085595
65
5201_83
0.989144
-0.035187
1.085595
66
5201_27
0.986254
-0.049202
1.374616
67
5201_87
0.990454
-0.033595
0.954597
68
5201_85
0.990475
-0.033996
0.952472
69
5201_28
0.986203
-0.048231
1.379741
70
5201_30
0.985663
-0.051343
1.433652
71
5201_88
0.990454
-0.033595
0.954597
72
5201_29
0.986203
-0.048231
1.379741
73
5201_32
0.985527
-0.050125
1.447291
74
5201_72
0.985575
-0.051665
1.442519
75
5201_34
0.985365
-0.052426
1.463477
76
5201_33
0.985527
-0.050125
1.447291
77
5201_69
0.985575
-0.051665
1.442519
90
78
5201_73
0.985569
-0.051562
1.443064
79
5201_35
0.985365
-0.052426
1.463477
80
5201_31
0.985575
-0.051665
1.442519
81
5201_74
0.985569
-0.051562
1.443064
82
5201_36
0.985365
-0.052426
1.463477
83
5201_37
0.985349
-0.052485
1.465098
84
5201_38
0.985342
-0.052511
1.465818
85
5201_75
0.985019
-0.053683
1.498064
86
5201_39
0.985342
-0.052511
1.465818
87
5201_76
0.984992
-0.053441
1.500764
88
5201_40
0.984910
-0.054080
1.508981
89
5201_77
0.984992
-0.053441
1.500764
90
5201_41
0.984856
-0.053045
1.514433
91
5201_79
0.984992
-0.053441
1.500764
92
5201_78
0.984992
-0.053441
1.500764
93
5201_42
0.984856
-0.053045
1.514433
# Nodo sal
# Nodo lle:
Perdida P:
Perdida Q:
2.770122e-004
1.963796e-004
0.000000e+000
0.000000e+000
7.354838e-005
5.213995e-005
1.728360e-004
1.225270e-004
1.247080e-007
3.039511e-008
0.000000e+000
0.000000e+000
2.143848e-005
1.519818e-005
1.925370e-006
1.364934e-006
10
0.000000e+000
0.000000e+000
91
6
11
0.000000e+000
0.000000e+000
12
0.000000e+000
0.000000e+000
13
0.000000e+000
0.000000e+000
14
0.000000e+000
0.000000e+000
15
0.000000e+000
0.000000e+000
16
0.000000e+000
0.000000e+000
10
17
1.384522e-004
9.815158e-005
12
18
0.000000e+000
0.000000e+000
12
19
0.000000e+000
0.000000e+000
12
20
0.000000e+000
0.000000e+000
13
21
0.000000e+000
0.000000e+000
16
22
0.000000e+000
0.000000e+000
16
23
0.000000e+000
0.000000e+000
16
24
0.000000e+000
0.000000e+000
17
25
2.302853e-007
5.612751e-008
17
26
5.971812e-005
4.233540e-005
18
27
0.000000e+000
0.000000e+000
19
28
1.644385e-005
1.165738e-005
22
29
1.593068e-007
6.769693e-008
23
30
2.723222e-008
1.157225e-008
25
31
0.000000e+000
0.000000e+000
26
32
0.000000e+000
0.000000e+000
28
33
5.897191e-008
1.437324e-008
28
34
1.562133e-005
1.107428e-005
32
35
0.000000e+000
0.000000e+000
32
36
0.000000e+000
0.000000e+000
33
37
0.000000e+000
0.000000e+000
34
38
1.181952e-006
2.880776e-007
34
39
1.017943e-005
7.216409e-006
35
40
1.244838e-006
5.289901e-007
36
41
3.744148e-005
2.654303e-005
92
38
42
0.000000e+000
0.000000e+000
39
43
2.173545e-007
1.540870e-007
39
44
0.000000e+000
0.000000e+000
40
45
7.942518e-007
3.375149e-007
40
46
1.765661e-006
7.503123e-007
41
47
6.943301e-009
1.692293e-009
41
48
3.005762e-007
7.325951e-008
41
49
3.439619e-005
2.438416e-005
43
50
8.150792e-006
5.778263e-006
43
51
0.000000e+000
0.000000e+000
43
52
0.000000e+000
0.000000e+000
44
53
0.000000e+000
0.000000e+000
45
54
1.237258e-009
3.015572e-010
45
55
6.935722e-007
2.947314e-007
46
56
0.000000e+000
0.000000e+000
47
57
0.000000e+000
0.000000e+000
48
58
0.000000e+000
0.000000e+000
49
59
0.000000e+000
0.000000e+000
50
60
2.194010e-007
5.347467e-008
50
61
0.000000e+000
0.000000e+000
54
62
0.000000e+000
0.000000e+000
55
63
0.000000e+000
0.000000e+000
56
64
0.000000e+000
0.000000e+000
56
65
0.000000e+000
0.000000e+000
59
66
2.196064e-005
1.556835e-005
60
67
0.000000e+000
0.000000e+000
61
68
0.000000e+000
0.000000e+000
66
69
2.656918e-007
6.475715e-008
66
70
1.385303e-005
9.820693e-006
67
71
0.000000e+000
0.000000e+000
69
72
0.000000e+000
0.000000e+000
93
70
73
6.461058e-007
2.745607e-007
70
74
5.520172e-008
3.913363e-008
70
75
5.572751e-006
3.950637e-006
73
76
0.000000e+000
0.000000e+000
74
77
0.000000e+000
0.000000e+000
74
78
4.240176e-009
1.033459e-009
75
79
0.000000e+000
0.000000e+000
77
80
0.000000e+000
0.000000e+000
78
81
0.000000e+000
0.000000e+000
79
82
0.000000e+000
0.000000e+000
82
83
3.028669e-007
2.147085e-007
83
84
5.979455e-008
4.238958e-008
83
85
3.422693e-006
2.426417e-006
84
86
0.000000e+000
0.000000e+000
85
87
1.277531e-007
5.428831e-008
85
88
6.801111e-007
4.821447e-007
87
89
0.000000e+000
0.000000e+000
88
90
4.246517e-007
1.035005e-007
89
91
0.000000e+000
0.000000e+000
89
92
0.000000e+000
0.000000e+000
90
93
0.000000e+000
0.000000e+000
CAPACIDADES AMPERIMETRICAS
***************************************************************
#Nodo sal
#Nodo lle:
I(max):
CAP(67)
CAP (max)
1.042209e-001
4.689942e-002
6.999914e-002
0.000000e+000
4.689942e-002
6.999914e-002
3.832317e-002
4.689942e-002
6.999914e-002
6.288734e-002
4.689942e-002
6.999914e-002
94
2
3.010414e-003
1.852592e-002
2.765062e-002
0.000000e+000
0.000000e+000
0.000000e+000
2.574611e-002
4.689942e-002
6.999914e-002
1.257706e-002
4.689942e-002
6.999914e-002
10
6.288734e-002
0.000000e+000
0.000000e+000
11
3.010414e-003
0.000000e+000
0.000000e+000
12
2.574611e-002
0.000000e+000
0.000000e+000
13
0.000000e+000
0.000000e+000
0.000000e+000
14
0.000000e+000
2.692495e-002
4.018649e-002
15
0.000000e+000
4.689942e-002
6.999914e-002
16
1.257706e-002
0.000000e+000
0.000000e+000
10
17
6.288734e-002
4.689942e-002
6.999914e-002
12
18
0.000000e+000
0.000000e+000
0.000000e+000
12
19
2.574611e-002
0.000000e+000
0.000000e+000
12
20
0.000000e+000
0.000000e+000
0.000000e+000
13
21
0.000000e+000
0.000000e+000
0.000000e+000
16
22
7.546220e-003
0.000000e+000
0.000000e+000
16
23
5.030841e-003
0.000000e+000
0.000000e+000
16
24
0.000000e+000
0.000000e+000
0.000000e+000
17
25
5.049839e-003
1.852592e-002
2.765062e-002
17
26
5.783750e-002
4.689942e-002
6.999914e-002
18
27
0.000000e+000
0.000000e+000
0.000000e+000
19
28
2.574611e-002
4.689942e-002
6.999914e-002
22
29
7.546220e-003
2.692495e-002
4.018649e-002
23
30
5.030841e-003
2.692495e-002
4.018649e-002
25
31
5.049839e-003
0.000000e+000
0.000000e+000
26
32
5.783750e-002
0.000000e+000
0.000000e+000
28
33
3.023646e-003
1.852592e-002
2.765062e-002
28
34
2.272247e-002
4.689942e-002
6.999914e-002
32
35
1.744872e-002
0.000000e+000
0.000000e+000
32
36
4.038877e-002
0.000000e+000
0.000000e+000
95
33
37
3.023646e-003
0.000000e+000
0.000000e+000
34
38
5.044779e-003
1.852592e-002
2.765062e-002
34
39
1.767769e-002
4.689942e-002
6.999914e-002
35
40
1.744872e-002
2.692495e-002
4.018649e-002
36
41
4.038877e-002
4.689942e-002
6.999914e-002
38
42
5.044779e-003
0.000000e+000
0.000000e+000
39
43
1.767769e-002
4.689942e-002
6.999914e-002
39
44
0.000000e+000
2.692495e-002
4.018649e-002
40
45
8.344978e-003
2.692495e-002
4.018649e-002
40
46
9.103745e-003
2.692495e-002
4.018649e-002
41
47
1.518756e-003
1.852592e-002
2.765062e-002
41
48
5.088027e-003
1.852592e-002
2.765062e-002
41
49
3.378199e-002
4.689942e-002
6.999914e-002
43
50
1.767769e-002
4.689942e-002
6.999914e-002
43
51
0.000000e+000
0.000000e+000
0.000000e+000
43
52
0.000000e+000
0.000000e+000
0.000000e+000
44
53
0.000000e+000
0.000000e+000
0.000000e+000
45
54
7.585761e-004
1.852592e-002
2.765062e-002
45
55
7.586402e-003
2.692495e-002
4.018649e-002
46
56
9.103745e-003
0.000000e+000
0.000000e+000
47
57
0.000000e+000
0.000000e+000
0.000000e+000
48
58
0.000000e+000
0.000000e+000
0.000000e+000
49
59
3.378199e-002
0.000000e+000
0.000000e+000
50
60
1.010156e-002
1.852592e-002
2.765062e-002
50
61
7.576129e-003
0.000000e+000
0.000000e+000
54
62
0.000000e+000
0.000000e+000
0.000000e+000
55
63
7.586402e-003
0.000000e+000
0.000000e+000
56
64
0.000000e+000
0.000000e+000
0.000000e+000
56
65
0.000000e+000
0.000000e+000
0.000000e+000
59
66
3.378199e-002
4.689942e-002
6.999914e-002
60
67
1.010156e-002
0.000000e+000
0.000000e+000
96
61
68
0.000000e+000
0.000000e+000
0.000000e+000
66
69
5.073849e-003
1.852592e-002
2.765062e-002
66
70
2.870814e-002
4.689942e-002
6.999914e-002
67
71
0.000000e+000
0.000000e+000
0.000000e+000
69
72
5.073849e-003
0.000000e+000
0.000000e+000
70
73
5.087528e-003
2.692495e-002
4.018649e-002
70
74
7.615905e-004
4.689942e-002
6.999914e-002
70
75
2.285902e-002
4.689942e-002
6.999914e-002
73
76
0.000000e+000
0.000000e+000
0.000000e+000
74
77
0.000000e+000
4.689942e-002
6.999914e-002
74
78
7.615905e-004
1.852592e-002
2.765062e-002
75
79
2.285902e-002
0.000000e+000
0.000000e+000
77
80
0.000000e+000
2.692495e-002
4.018649e-002
78
81
0.000000e+000
0.000000e+000
0.000000e+000
79
82
2.285902e-002
0.000000e+000
0.000000e+000
82
83
2.285902e-002
4.689942e-002
6.999914e-002
83
84
1.015693e-002
4.689942e-002
6.999914e-002
83
85
1.270209e-002
4.689942e-002
6.999914e-002
84
86
1.015693e-002
0.000000e+000
0.000000e+000
85
87
5.080493e-003
2.692495e-002
4.018649e-002
85
88
7.621598e-003
4.689942e-002
6.999914e-002
87
89
5.080493e-003
0.000000e+000
0.000000e+000
88
90
7.621598e-003
1.852592e-002
2.765062e-002
89
91
0.000000e+000
0.000000e+000
0.000000e+000
89
92
0.000000e+000
0.000000e+000
0.000000e+000
90
93
7.621598e-003
0.000000e+000
0.000000e+000
97
Apndice D
CASO DE ESTUDIO: CIRCUITO DE 4 BARRAS
El sistema incluye 1 generador y 3 cargas equivalentes, conectadas a una red de distribucin radial. Los datos de ramas y nodos del circuito se muestran en las tablas D.1 y D.2,
respectivamente.
Datos de lnea
El circuito est conformado por 3 lneas, descritas en la tabla D.1.
salida
llegada
r(p.u)
x(p.u)
0,0265
0,0462
0,1005
0,0693
0,0670
0,0462
Datos de nodo
Los datos de los 4 nodos se encuentran en la tabla D.2.
98
99
Tabla D.2: Datos de nodos (Bdat)
Tipo
P gen0 (p.u)
Qgen0 (p.u)
Pload (p.u)
Qload (p.u)
V0
fcap
0,1017
0,0635
0,0547
0,0342
0,0809
0,0596
Donde,
N. Es el nmero de la barra.
Tipo. Corresponde al tipo de barra: 1 barra slack, 2 barra PV y 3 barra PQ.
P gen0 . Generacin de potencia activa (valor inicial)(p.u).
Qgen0 .Generacin de potencia reactiva (valor inicial)(p.u).
Pload . Potencia activa consumida (p.u).
Qload . Potencia activa consumida (p.u).
V0 . Tensin inicial de la barra (p.u).
fcap. Factor de capacidad de la barra.
Apndice E
REPORTE DEL MATPOWER PARA EL CASO DE 7 BARRAS
runpf(MATPOWERexample)
System Summary
================================================================================
How many?
How much?
-------------
P (MW)
-------------------
-----------
Q(MVAr)
------------
Buses
250.0
-300.0 to 300.0
Generators
On-line Capacity
250.0
-300.0 to 300.0
Generation (actual)
0.5
0.3
Load
0.5
0.3
Fixed
0.5
0.3
Committed Gens
Loads
Fixed
Dispatchable
Dispatchable
-0.0 of -0.0
-0.0
Shunts
Shunt (inj)
Branches
Losses (I^2*Z)
Transformers
Inter-ties
0
0
-0.0
0.0
0.03
0.02
Branch Charging(inj)
0.0
0.0
0.0
100
101
Areas
Minimum
Maximum
-------------------------
--------------------------------
Voltage Magnitude
Voltage Angle
-0.49 deg
0.00 deg
P Losses (I^2*R)
Q Losses (I^2*X)
@ bus 6
-
@ bus 1
0.02 MW
@ line 1-3
0.02 MVAr
@ line 1-3
================================================================================
|
Bus Data
================================================================================
Bus
#
Voltage
Mag(pu) Ang(deg)
Generation
P (MW)
Load
Q (MVAr)
--------
--------
1 1.000000 0.000000
0.50
0.34
2 0.994335 -0.173777
3 0.940742 -0.020595
P (MW)
--------
Q (MVAr)
--------
0.10
0.06
0.05
0.03
4 0.931971 -0.003891
0.08
0.06
5 0.925992 -0.469903
0.10
0.06
6 0.917416 -0.493752
0.05
0.03
7 0.917079 -0.452656
0.08
0.06
-------Total:
0.50
-------0.34
-------0.47
-------0.31
102
================================================================================
|
Branch Data
================================================================================
Brnch
From
To
Bus
Bus
P (MW)
Q (MVAr)
P (MW)
----
----
----
-------
------
-------
--------
------
-------
0.10
0.06
-0.10
-0.06
0.000
0.00
0.40
0.27
-0.38
-0.26
0.024
0.02
0.08
0.06
-0.08
-0.06
0.001
0.00
0.24
0.16
-0.24
-0.16
0.003
0.00
0.06
0.03
-0.05
-0.03
0.000
0.00
0.08
0.06
-0.08
-0.06
0.001
0.00
--------
--------
0.029
0.02
Q (MVAr) P (MW)
Total:
Q (MVAr)
Apndice F
A COMPENSATION-BASED POWER FLOW METHOD FOR WEAKLY
MESHED DISTRIBUTION AND TRANSMISSION NETWORKS,
SHIRMOHAMMADI ET AL. 1988, [11]
103
753
D. Shirmoharmnadi
Member
H. W. Hong
Senior Member
ABSTRACT
This paper describes a new power flow method for
solving weakly meshed distribution and transmission
networks, using a multi-port compensation technique and
basic formulations of Kirchhoff's laws. ThiS method
has excellent convergence characteristics and is very
robust. A computer program implementing this power
flow solution scheme was developed and successfully
applied to several practical distribution netwdrks with
radial and weakly meshed structure. This program was
also successfully used for solving radial and weakly
meshed transmissioh networks. The method
can be
applied tb the solution of both the three-phase
(unbalanced) and single-phase (balanced) representation
of the netbork.
In this paper, however, only the
single phase representation is treated in detail.
I. INTRODUCTION
Recently.at the Pacific,Gasand Electric Company
we developed a distkibution network optimization
software package. This development work called for a
power flew solution algorithm with the following
general characteristics:
2.
A. Semlyen
Senior Member
G. X. Luo
754
JL(k) =
currents in
-ILt(k)+C(branches emanating)
from node L2
L=b,b-1,
(2)
...,1
I,@)
...,n
= (si/vi(k-l) )*-Y~v~(~-I)i=1,2,
(1)
LAYER 10
2. Backward
Fig.2
755
Convergence Criterion
We used the maximum real and reactive power
mismatches at the network nodes as our convergence
criterion. As described in the solution algorithm, the
nodal current injections, at
iteration k, are
calculated using the scheduled nodal power injections
and node voltages from the previous iteration (equation
(1)).
The node voltages at the same iteration are then
calculated using these nodal current
injections
(equations (2) and (3)).
He c , the power injection
for node i at kth iteration,
is calculated as:
AND INITIALCONDITION
SET ITERATONCWNT
(4 )
= Re[Si(k)
AQ~(~)= ImIsi(k)
Si]
i=1,2,.
- si]
...n
i are
(4)
DISTRIBUTION
~~~~
CALCULATEMAXIMUMREAL AND
REACTIVEPOWER MISMATCHES
EQUATION (4)
MAXlMUMlTERAT
Fig.4
lEEzRArn1
1
2
3
4
5
6.134
0.301
0.008
0.000
0.000
13.092
0.567
0.024
0.001
0.000
1
2
3
4
5
6
7
8
5.994
1.691
0.402
0.126
0.031
0.010
0.003
O.OO0
5.597
1.132
0.377
0.085
0.029
0.007
0.002
0.000
4.573
1,
4.691
0.719
0.088
0.011
0.001
0.000
WXFIANGE
0.245-5.065
544 NODES
WXFIANGE
0.409-5.063
1411 NODES
0.000-5.480
Table 1
INT
Fig.3
244 NODES
YES
0.037
0.003
156
j = 1, 2,
..., p
(5)
Tj(6
$Tky
BREAKPOINT
Pig.7
Fig.5
The breakpoint
equivalent impedance)
following method:
impedance matrix
(Thevenin
can be determined using the
ji=O,
..
1)
757
1. Examine
all
connected to
layer
branches
and
select those
the root node for the first branch
2. Store
3. Examine
5. Repeat
This fixed
tangent
solution
method
is
schematically depicted in Figure 8 for a network having
a single breakpoint. Computationally there is no need
f r the inversion
the breakpoint impedance matrix
[&. Complex matrix
1 is factorized once in the
beginning
of the iterations and the forward fiand
backward substitution is then used to calculate
Our test cases on practical
in equation (E').
distribution networks showed that the number of
iterations required
for the calculation of the
breakpoint currents was less than 5 in most cases.
steps
processed.
2-4
'
until
$ ( l ) $12)
Fig.8
Selection of Breakpoints
Breakpoints are selected in order to convert
meshed network into a radial configuration.
In
addition to this function breakpoints should be
selected in such a way as to ensure the convergence of
the overall solution algorithm. The latter requirement
for the selection of breakpoints usually is satisfied
by concentrating on the parts of the meshed network
where the power flows are low. On the other hand, the
power flows are the end product of the solution method
and are not known at the time of breakpoint selection.
In weakly meshed distribution networks breakpoint
selection does not affect the convergence performance
of the solution method in any noticeable manner.
Hence, we select them for the main purpose of opening
the network loops.
Under these circumstances the
algorithm for identifying the breakpoints is very
simple and becomes part of the branch numbering scheme
described below:
Fig.9
Compensation-based power
weakly meshed networks
flow method
for the
758
WSCCPOWERW
IETwow(
ZNFWRATION
ml"S
-METHOD
MAX MISMATCH
&!$
m!
l"
=
S
REAL
RUCTNE
E,
NODES
WLOCPS
0.97
021
O.CQ8
0.024
SLOOPS
037
025
0.003
0.008
5.31
2)
1411 NOMS
42 BUS, WOW
N3WNoDEs
N3Loops
36 LOOPS
527
1.47
0011
NO LOOPS
10
1 LOOP
3 LOOPS
SLOOPS
42 BUS, 5W KV
4
w
m
NO LOOPS
1 LOOP
3 LOOPS
5 LOOPS
2.
3.
159
[lo] N. Vempati, R. R.
Shoults, M. S. Chen,
L. Schwobel, "Simplified Feeder Modeling for Load
Flow Calculations", 1EF.E Paper No. 86WM102-8, PES
Winter Power Meeting, New York, 1986.
APPENDIX
At the kth iteration, the reactive power injection
required to maintain the voltage at the generator bus
i, can be calculated using the secant method:
where IVi(k-l)l
and
IVi(k-2)l
are the voltage
magnitudes at the node i calculated in the previous
two iterations (equation (3) in step 3).
The actual
reactive
power
injection
is determined as:
ACKNOWLEDGEMENT
The authors thank Messrs R.L. Smith and J.
Monasterio of PGhE's Electric Distribution Engineering
Department for their useful coments on the practical
applications of the proposed method.
The Canadian
authors gratefully acknowledge the financial support
from the National Sciences and Engineering Research
Council of Canada.
and
111Like
pp.
Qic <Qimh
760
currents in the reverse trace diverged, whereas the method of summing the
powers converged.
A brief and heuristic explanation of this phenomenon is as follows.
Initially, when the currents are summed in the reverse trace, each current
will contain an error proportional to the initially guessed voltage. If the
initially guessed voltages are maintainedconstant and a succession of power
flow problems are solved in which the loads &e continually increased, the
errors that are proportional to the initially guessed voltages wiU grow. For a
sufficiently heavily loaded system, the initially guessed voltages fall outside
the region of convergence, and the algorithm will diverge.
When the powers are summed in the reverse trace, the errors that exist
when the source bus is reached involve only the power losses, and not the
load powers. The power losses are always a small fraction of the load
powers. Hence, using the power sum leads to good convergence for even
heavily loaded systems.
2) Even though it is mentioned that the multiphase unbalanced systems
are easily handled, the convergence characteristics claimed for the singlephase system may get severely impaired, since the error involved in current
summation may become excessive.
3) The details of the distribution networks given in Table 1 do not include
the loading levels of the systems. It would be instructive to know whether
the systems are nominally loaded or lightly loaded. Further, the number of
nodes may not be a direct indication of the size of the system since
distributed loads can be modeled using any number of node points.
General
6) The CPU time given in the table is said to include the initial processing
time also. This may show the WSCC Power Flow Program in a bad light.
Exact CPU time required for the iterations alone should be a better index.
7)In Fig. 2, the numbering of the distribution network is laid out on a
grid in a very orderly fashion. With this scheme, it appears that choosing
buses at large load centers may lead to conflicts. For instance, in Layer 2
developed for a load flow problem, the discaqers believe that it is effective
to develop an algorithm which utiliis special features of the problem; the
radial power system is one of such special feahres. In load flow problem of
a radial power system, the number of variables or computational burden can
be reduced extremely by taking voltages of the ends (or tips) of the branches
as independent variables, in comparison with taking all node voltages as
variables. The method developed by the authors also utilizes such features.
The discussers have a question that it might be fast to make a rcduced
variable problem as above and solve it directly by the Newton-Fbphson
method. Could the authors comment on whether they have ever tried such a
reduced variable Newton-Raphson method, and if tried, how was the
comparison of computation time between the two methods?
suppose that a large load exists at the center of branch 4, but there is little or
no load in the centers of branches 5 and 6. With the authors proposed
scheme, it appears that additional buses would be inserted in branches 5 and
6 that would not be necessary for accurate load modeling.
8) It appears that if a component is inserted or added somewhere in the
network, a complete renumbering of the network is required. Is this true?
9) Even though the WSCC Power Flow Program is shown to converge
more slowly for the examples chosen in comparison to the proposed
method, it is a moot point whether a few extra seconds of CPU time alone
would be sufficient to choose any alternative method. The Newtons method
has been so finely honed during the last two decades that negative
experiences must be very few to be deemed almost nonexistent. Hence, the
virtues of the proposed method must be highlighted from a different
viewpoint.
Dr. Dromey: The authors have investigated an aspect of load flow analysis
that has received little attention in the literature to date. The wide range of
R/Xratios in a distribution system can lead to difficulty with convergence
and this problem is aggravated by the presence of adjacent long and short
branches. In particular, the Newton-Raphson and decoupled algorithm
can, in some circumstances, fail to converge for larger or illconditioned
systems. The method presented in the paper produces an optimal ordering
for solution and is very efficient, particularly where the number of loops is
limited and the system is essentially radial in nature.
-*-
76I
(a)
762
It can be seen that the incidence matrices are square, lower triangular,
and consist of elements f l only. Their inverses could also have been
formed directly by inspection of the network graph. This reflects the fact
that equations (a) represent directly the two Kirchhoffs laws. They
correspond to the forward and backward sweepsused in the paper.
Let Yb (diagonal) and Y& be the admittance matrices of the network. We have
Y,=AI YbwAV
Clearly, the two incidence matrices are the transpose of each other.
Let us now solve the nodal problem
ybw v&=Ibur
Factorization of Y& yields
Y&=LDL~
we obtain
Comparing (d) with 0)
L-'=AI, LTT=Av. and D=Ybr
vbur =L-T D-' L-' IbyI =AV (& (AI[&))=AV (ZbrIt,,) =AV V,
(r)
Equation (f) shows that the branch-oriented solution reproduces the
matrix operations of the nodal approach without the need of preliminary
formulation and factorization of a bus admittance matrix. To achieve
this, it was essential to replace all shunt branches connected to buses by
corresponding injections.
We note that reference [6]claims that the efficiency of its algorithm is comparable to that of the Newton-Raphson Power Flow while
we have shown a substantial improvement over the Newton-Raphson
Power Flow method using our algorithm.
We appreciate the information provided by the discussers on problems related to transformer modeling. We have not investigated this
topic.
(g)
The resultant change in voltage is, taking (g) into account with
V=l,
AVNW=-ZN=&
AV*
resistances.
Fig. 8 of the paper is used only to depict the basic mechanism of
the fixed tangent solution algorithm used for calculating breakpoint
current injections according to equation (8). Nevefieless. our
experience with all distribution networks studied indicates a monotone and rapid convergence in the calculation of bRakpoint
currents.
After a considerable amount of experimentation we found that a
better convergence in the calculation of breakpoiit c m n t s can be
achieved when the breakpoint impedance matrix is calculated
neglecting the shunt components (mainly capacitors). Unfortunately this important conclusion is not reflected in the paper and
we would like to thank the discussers for providing us with this
opportunity.
Dr. Dromey:
(1)
(h)
The method presented in the paper can handle any load characteristics, since the load current is calculated at each step as a function of the
prevailing voltage. We did not use however any convergence
(accelerating) factors. We feel that such factors (not necessarily uniform
and real) could improve the convergence of the method and appreciate
the suggestion implied in the question.
We have developed our methodology and the accompanying program mainly for primary distribution networks. Primary distribution networks are, in almost all cases, either radial or weakly meshed. The program is, however, capable of representing up to 5000 nodes and 300
loops which is adequate for almost all practical cases including secondary distribution networks in downtown metropolitan areas. Even with
such large dimensidns. the memory requirement for the program is less
than 500 Kbytes which makes it ideal for microcomputer applications.
The largest network studied using this program consisted of 241 1 nodes
and 183 loops. It took the program a total CPU time of 12 seconds to
process the input data, solve the power flow and print the results for this
network on an IBM 3090-200 computer, which is very reasonable considering the size of the network.
Partitioning of distribution networks, or even of transmission networks, involves network equivalencing. Reference [A] is pertinent to this
topic.
[AI F.F. Wu, A. Monticelli, "A Critical Review on External Network
Modelling for On-Line Security Analysis", Electrical Power and
Energy Systems, Vol. 5, October 1983, pp. 222-235.
Apndice G
A DIRECT APPROACH FOR DISTRIBUTION SYSTEM LOAD FLOW
SOLUTION, TENG. 2003, [19]
114
882
I. INTRODUCTION
883
A simple distribution system shown in Fig. 2 is used as an example. The power injections can be converted to the equivalent
current injections by (5), and the relationship between the bus
current injections and branch currents can be obtained by applying Kirchhoffs Current Law (KCL) to the distribution network. The branch currents can then be formulated as functions
of equivalent current injections. For example, the branch currents
and
, can be expressed by equivalent current
injections as
(7a)
884
From (9), it can be seen that the bus voltage can be expressed
as a function of branch currents, line parameters, and the substation voltage. Similar procedures can be performed on other
buses; therefore, the relationship between branch currents and
bus voltages can be expressed as
(10b)
(11)
(10a)
Equation (10a) can be rewritten in general form as
(13)
885
(17a)
Fig. 3. Simple distribution system with one loop.
(14b)
C. Modification for Solution Techniques
And the modified BIBC matrix can be obtained as
(15a)
(19)
Applying Krons Reduction to (19), the modified algorithm for
weakly meshed networks can be expressed as
(20)
The building algorithm of Step 2) for BIBC matrix can be modified as follows:
Step 2a)If a new branch
makes the system become
meshed (the new branch is between bus and ), copy the elements of the -th bus column to the -th column and minus the
elements of the -th bus column. Finally, fill a
value to the
position of the -th row and the -th column.
V. TEST RESULTS
The proposed three-phase load flow algorithm was implemented using Borland C++ language and tested on a
Windows-98-based Pentium-II PC. Two methods are used for
tests and the convergence tolerance is set at 0.001 p.u.
Method 1: The Gauss implicit -matrix method [7].
Method 2: The proposed algorithm.
886
TABLE II
TEST FEEDER
TABLE III
NUMBER OF ITERATION AND NORMALIZED EXECUTION TIME
Fig. 4. Eight-bus distribution system.
TABLE I
FINAL CONVERGED VOLTAGE SOLUTIONS
TABLE IV
TEST RESULTS FOR THE WEAKLY MESHED FEEDERS
A. Accuracy Comparison
For any new method, it is important to make sure that the final
solution of the new method is the same as the existent method.
An eight-bus system (equivalent 13-node system), including the
three-phase, double-phase, and single-phase line sections and
buses as shown in Fig. 4 is used for comparisons. The final
voltage solutions of method 1 and method 2 are shown in Table I.
From Table I, the final converged voltage solutions of method
2 are very close to the solution of method 1. It means that the
accuracy of the proposed method is almost the same as the commonly used Gauss implicit -matrix method.
B. Performance Test
A main feeder trunk with 3 90-phase buses, which was acquired from the Taiwan Power Company (TPC), is used for this
test. The single and double-phase laterals have been lumped to
form the unbalanced loads for testing purposes. This trunk is
then chopped into various sizes for tests as shown in Table II.
The substation is modeled as the slack bus.
1) Radial Network Test: Table III lists the number of iterations and the normalized execution time for both methods. It can
be seen that method 2 is more efficient, especially when the network size increases, since the time-consuming processes such as
LU factorization and forward/backward substitution of -admittance matrix are not necessary for method 2. For a 270-node
system, method 2 is almost 24 times faster than method 1.
2) Weakly-Meshed Network Test: Some branches are connected to the test feeder to make the system meshed. Table IV
shows the number of iterations and normalized execution time
for the weakly meshed network. Table IV shows that the number
of iterations of method 2 is stable. The normalized execution
time increases, since the meshed network increase the nonzero
terms of the BIBC and BCBV matrices, and extra procedure
needs to be done.
C. Robustness Test
One of the major reasons, which make the load flow program
diverge, is the ill-condition problem of the Jacobian matrix or
admittance matrix. It usually occurs when the system contains some very short lines or very long lines. In order to prove
that the proposed method can be utilized in severe conditions,
IEEE 37-bus test feeder is used [14]. The test feeder is adjusted
by changing the length of eight line sections. Four of them are
multiplied by ten, and the other four are divided by ten. The test
result shows the number of iterations for this case is 4 and the
execution time is 0.0181 s. It means that the proposed method
is robust and very suitable for online use.
VI. DISCUSSION AND CONCLUSION
In this paper, a direct approach for distribution load flow solution was proposed. Two matrices, which are developed from the
topological characteristics of distribution systems, are used to
solve load flow problem. The BIBC matrix represents the relationship between bus current injections and branch currents, and
the BCBV matrix represents the relationship between branch
currents and bus voltages. These two matrices are combined
to form a direct approach for solving load flow problems. The
time-consuming procedures, such as the LU factorization and
forward/backward substitution of the Jacobian matrix or admittance matrix, are not necessary in the proposed method. The
ill-conditioned problem that usually occurs during the LU factorization of the Jacobian matrix or admittance matrix will
not occur in the proposed solution techniques. Therefore, the
proposed method is both robust and efficient. Test results show
that the proposed method is suitable for large-scale distribution
systems. Other issues involved in the distribution system operation, such as multiphase operation with unbalanced and distributed loads, voltage regulators, and capacitors with automatic
tap controls, will be discussed in future work.
887
[7] T.-H. Chen, M.-S. Chen, K.-J. Hwang, P. Kotas, and E. A. Chebli, Distribution system power flow analysisA rigid approach, IEEE Trans.
Power Delivery, vol. 6, pp. 11461152, July 1991.
[8] T. H. Chen and J. D. Chang, Open wye-open delta and open delta-open
delta transformer models for rigorous distribution system analysis, in
Proc. Inst. Elect. Eng., vol. 139, 1992, pp. 227234.
[9] K. A. Birt, J. J. Graffy, J. D. McDonald, and A. H. El-Abiad, Three
phase load flow program, IEEE Trans. Power Apparat. Syst., vol.
PAS-95, pp. 5965, Jan./Feb. 1976.
[10] D. Shirmohammadi, H. W. Hong, A. Semlyen, and G. X. Luo, A compensation-based power flow method for weakly meshed distribution and
transmission networks, IEEE Trans. Power Syst., vol. 3, pp. 753762,
May 1988.
[11] G. X. Luo and A. Semlyen, Efficient load flow for large weakly meshed
networks, IEEE Trans. Power Syst., vol. 5, pp. 13091316, Nov. 1990.
[12] C. S. Cheng and D. Shirmohammadi, A three-phase power flow method
for real-time distribution system analysis, IEEE Trans. Power Syst., vol.
10, pp. 671679, May 1995.
[13] R. D. Zimmerman and H. D. Chiang, Fast decoupled power flow for
unbalanced radial distribution systems, IEEE Trans. Power Syst., vol.
10, pp. 20452052, Nov. 1995.
[14] W. M. Kersting and L. Willis, Radial Distribution Test Systems, IEEE
Trans. Power Syst.,, vol. 6, IEEE Distribution Planning Working Group
Rep., Aug. 1991.
REFERENCES
[1] IEEE Tutorial Course on Distribution Automation.
[2] IEEE Tutorial Course on Power Distribution Planning.
[3] W. M. Lin and M. S. Chen, An overall distribution automation structure, Elect. Power Syst. Res., vol. 10, pp. 719, 1986.
[4] B. Stott and O. Alsac, Fast decoupled load flow, IEEE Trans. Power
Apparat. Syst., vol. 93, pp. 859869, May/June 1974.
[5] J. H. Teng and W. M. Lin, Current-based power flow solutions for distribution systems, in Proc. IEEE Int. Conf. Power Syst. Technol., Beijing,
China, 1994, pp. 414418.
[6] T. S. Chen, M. S. Chen, T. Inoue, and E. A. Chebli, Three-phase cogenerator and transformer models for distribution system analysis, IEEE
Trans. Power Delivery, vol. 6, pp. 16711681.2, Oct. 1991.
Jen-Hao Teng (M99) was born in 1969 in Tainan, Taiwan, R.O.C. He received
the B.S., M.S., and Ph.D. degrees in electrical engineering from the National
Sun Yat-Sen University, Taiwan, R.O.C., in 1991, 1993, and 1996, respectively.
Currently, he is with I-Shou University, Taiwan, R.O.C., where he has been
since 1998. His current research interests include energy management system,
distribution automation system, and power system quality.
Apndice H
THE DISTRIBUTION TRX-POWER FLOW METHOD, DE OLIVEIRA.
2010, [20]
121
1217
I. Introduction
ISTRIBUTION system operation and planning require robust and reliable power flow analysis
techniques. Modern distribution management systems (DMS) need this support in order to
perform applications as distributed generation dispatch, service restoration, feeder reconfiguration,
phase balancing, volt/var control, optimal location of capacitors, etc. For a long time, traditional
and efficient methods as Newton Raphson (NR) [1] and fast decoupled power flow [2] have been
successfully applied in large power systems. Concerning distribution systems, early attempts to
develop power flow studies can be found in [3] [12].
Special features of distribution systems as radial or weakly meshed structure, low ratio between
reactance and resistance values, unbalancing and distributed generation have undesired results in
both performance and robustness of traditional power flow techniques. Weakly meshed distribution
networks may have thousands of busbars being considered ill-conditioned by causing numerical
problems for the conventional power flow algorithms [4], [5]. The degree of ill-conditioning is
evaluated by the number of iterations or the value maximum eigenvalues of the inverse Jacobian
matrix. These deficiencies lead to the development of alternative power flow techniques specially
designed for distribution systems.
Several methods have been reported in Literature to solve distribution systems. They can be
divided into three categories: Jacobian-based methods, direct methods, and backward/forward
(BW/FW) sweep methods. The first type of methods is based on modification of existing methods
such as Newton-Raphson [6], [7], [8], [9] where no node ordering is required.
Direct methods [10], [11] require the construction of an impedance matrix. These methods
usually present a heavy computational burden requiring a specific numbering scheme for nodes
and branches.
Finally, BW/FW sweep algorithms have been developed based on the assumption that nodes and
network branches are properly ordered. In 1967, Berg presented a paper which can be considered
P. M. De Oliveira is with the Department of Electrical Engineering, Simon Bolivar University, Caracas, Venezuela, Ap.89000.
Phone: +58 212 9063913, e-mail: pdeoliveira@usb.ve
1218
as the source for the all variants of BW/FW sweep methods [12]. Later, a similar approach was
presented in [13] based on ladder network theory. In general, these algorithms can use the Kirchhoff laws [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [26], [25], [27] or the
bi-quadratic equation [28], [29], [30], [31] in the iterative process. BW/FW sweep methods typically present a slow convergence rate but computationally efficient at each iteration. Using these
methods, power flow solution for a distribution network can be obtained without solving any set of
simultaneous equations.
The effectiveness of the BW/FW sweep algorithm has already been proven by comparing it to
the traditional NR methods [32] [34]. Recently, some comparative and convergence studies have
been presented by [33] [35].
All contributions mentioned above have been valuable tools to perform power distribution analysis. However, under modern DMS environment new considerations must be raised up about the
appropriateness of a given distribution power flow methodologies from system implementation point
of view:
1. The system data model: topology structure and nodal measurements can be stored in memory of DMS environment reflecting the most credible state of the network at given time.
Standardization processes based on Extended Markup Language (XML) as Common Information Model [36] and Open Data Model (ODM) [37], [46] have been carried out. In this
sense, power flow analysis in real time or off-line studies can be applied using the most
realistic data (present and historical) stored in memory or media with fast transfer rates.
Then, performance and robustness assessment of power flow methods implemented under
DMS must be made considering the impact of I/O data access.
2. Under DMS environment, power flow applications are object oriented [39] suitable to be
applied with distributed processing [38].
In this context, this paper proposes a new methodology to assess the power flow solution of
distribution networks by merging the standard BW/FW sweep steps into one unique state-ofthe-system calculation formula based upon a TRX matrix suitable to be stored in memory in the
context of a DMS environment. The TRX matrix is formed by real numbers with three fundamental
elements: the triangular matrix T that relates nodal currents with branch currents, resistance R
and reactance X vectors that characterize the branches, lines or transformers, of the distribution
network. The T matrix and R and X vectors reflect the present condition of the network being
constructed directly from the data exchange scheme adopted.
The proposal is simple and suitable to be applied in distribution systems, radially or weakly
meshed operated with distributed generation. The power flow is iteratively solved using only one
state-of-the system calculation step. This contribution is meaningful for actual assessment of the
distribution system and expansion system planning process, since the state of the system is reached
through multiplication and summation operations where no matrix inverse is required
Proposed methodology has been applied in a simple 4-node network for illustration purposes and
compared with a robust Newton Raphson solver [44] and a standard Backward/Forward Sweep
algorithm [24] in three test systems: a 12-bus [29], 33-bus [40] and 69-bus [41] networks. Finally,
a convergence study is performed using a uniformly loaded distribution network from 1000 to 3000
nodes.
This paper is organized as follows: Section II describes the standard BW/FW sweep method under
an appropriate notation for DMS implementation. Section III presents the proposed methodology
for balanced and unbalanced networks. Case studies are discussed in Section IV. Conclusions are
drawn in Section V. Nomenclature, list of symbols are provided in Appendix A. Proposed algorithm
1219
Z 01 ... Z ij ... Z mn
Z ij = Rij + jXij
Bus data is given by
S1
.
..
S=
Si
..
.
i, j = 1, ..., n i 6= j
P1 + jQ1
..
(1)
(2)
= Pi + jQi
..
(3)
Pn + jQn
Sn
where net nodal active and reactive powers are given by generated and demanded powers:
Pi = PGi PDi
Qi = QGi QDi
(4)
(5)
The numbering of branches in one layer begins only after all the branches in the previous layer
k=0
and
have been numbered. Considering that initial voltages are known: voltage at substation V
an initial voltage vector given by:
V0 =
V 1 ... V i ... V n
(6)
1220
(7)
The relationship between nodal currents Ik and branch currents Jk is set through an upper
k
triangular matrix T accomplishing the Kirchhoff Current Laws (KCL) [18]. Each element I i of Ik
k
associated to node i is calculated as function of injected powers S i and its voltage profile V i as
shown below:
k
Ii
Si
i = 1, ..., n
Vi
(8)
(9)
where V0 is a n-elements vector with all entries set at voltage at origin (swing node) V 0 and
branch impedances DZ is the diagonal matrix of vector Z:
C. Convergence
Updated voltages are compared with previous voltages in order to perform convergence check in.
k+1
|V i
V i | i = 1, ..., n
(10)
h
h
0
0
Vx1
... Vxi0 ... Vxn
0
0
Vy1
... Vyi0 ... Vyn
(11)
(12)
Reference voltages Vx0 and Vy0 can be acquired from measurements of DMS support. Alternatively,
it can be used Vxi0 = 1 and Vyi0 = 0 for i = 1, ..., n.
The state of the system is updated at each iteration k by the following expression:
Vxk+1 + jVyk+1 = Vx0 + jVy0 TT (DR + jDX ) (Jkx + jJky )
(13)
1221
where,
DR
DX
Jx
Jy
= e{DZ }
= Im{DZ }
= T Ix
= T Iy
(14)
(15)
(16)
(17)
This expression correspond to Kirchhoff Voltage Laws (KVL) through all nodes and, where Jx +
jJy term corresponds to all nodal Kirchhoff Current Laws (KCL). After some algebra, real and
imaginary parts of 13 can be rewritten as:
Vx0 = Vx0 TT DR T Ix + TT DX T Iy
Vy0
Vy0
(18)
T DX T Ix T DR T Iy
(19)
Vxk+1
Vyk+1
Vx0
Vy0
T T DR T
T T DX T
TT DX T
T T DR T
Ikx
Iky
(20)
"
TT DR T TT DX T
TT D X T T T D R T
(21)
The dimension of the TRX matrix for balanced systems is 2nx2n. This matrix is not sparse but
suitable to be allocated using storage devices with high data transfer rate. However, it must be
pointed out that only TT DR T and TT DX T matrices should be allocated due to symmetry in the
TRX matrix. For typical, 1000 node distribution circuits, the TRX matrix will require al least
6MB using double-precision numbers
A general expression that relates each nodal voltage with the origin can be written as:
1222
V = V0 TRX I
(22)
Note that, at each iteration, voltage drops V = V0 V are directly obtained multiplying the
TRX matrix and the nodal current vector I.
k
k
The nodal currents Ixi
and Iyi
for i = 1, ..., n are given as function of present nodal power injections
(demands, capacitors or distributed generators) and operational voltages:
k
Ixi
= e{Iki } =
k
Pi Vxi
Qi Vyik
(23)
k
Qi Vxi
Pi Vyik
(24)
k
Iyi
= Im{Iki } =
k )2 + (V k )2
(Vxi
yi
k )2 + (V k )2
(Vxi
yi
For simplicity, this contribution only refers constant power P Q load models. Formulae for
constant current and constant impedance models will be provided in the future. In addition,
this model is suitable to be extended to weakly meshed networks using the compensation method
described in [14].
Convergence check: recent updated voltages are compared with previous voltages in all nodes in
order to perform convergence check in.
k+1
|V i
V i | i = 1, ..., n
(25)
..
S = S pi
..
S pn
Pp1 + jQp1
..
.
Ppi + jQpi
..
.
Ppn + jQpn
p = a, b, c
(26)
Branch impedances are given as a rectangular 3nx3 phase impedance matrix Zabc
Zabc =
(27)
where Zij is the 3-phase matrix impedance corresponding to ij line section [42]:
(28)
0
Given an initial condition (k = 0), nodal voltage 3xn-elements vector Vabc
are decomposed into
real and imaginary parts:
0
Vabcx
=
0
Vabcy
0
Vax1
0
... Vaxn
0
Vbxn
0
Vcxn
0
Vay1
0
Vayn
0
Vbyn
0
Vcyn
...
(29)
(30)
1223
Initial values of three phase nodal apparent power injections vector S and three-phase voltage
profiles are online scanned or estimated using DMS.
The state of the system is an n-elements voltage vector that is updated at each iteration k by
the following formula:
k+1
Vabcx
k+1
Vabcy
Vabcx0
Vabcy0
A B
B A
Ikabcx
Ikabcy
(31)
where;
A = Tabc T Dabc
R Tabc
T
B = Tabc Dabc
X Tabc
(32)
(33)
Tabc is the upper triangular matrix, with dimension 3nx3n, according current Kirchhoff laws
considering three phase connection. A general expression can be written as:
Vabc = Vabc0 TRXabc Iabc
(34)
Note that, at each iteration, voltage drops per phase Vabc = Vabc0 Vabc are directly obtained
multiplying the TRXabc matrix and the nodal current vector Iabc .
The dimension of the TRXabc matrix for unbalanced systems is 6nx6n. For instance, the corresponding matrix for a 1000 node-distribution circuits will require al least 56MB using doubleprecision numbers.
k
k
Three phase currents Ipxi
and Ipyi
for i = 1, ..., n are given as function of present three phase
power injections (demands, capacitors or distributed generators) and three phase operational voltages:
k
Ipxi
= e{Ikpi } =
k
Ipyi
= Im{Ikpi } =
k
k
Ppi Vpxi
Qpi Vpyi
(35)
k
k
Qpi Vpxi
Ppi Vpyi
(36)
k
k
(Vpxi
)2 + (Vpyi
)2
k
k
(Vpxi
)2 + (Vpyi
)2
Convergence check: recent updated voltages are compared with previous voltages in all nodes in
order to perform convergence check in.
k+1
|V pi V pi | i = 1, ..., n p = a, b, c
IV. Testing
The proposed methodology was applied to a list of test systems:
A 4-node illustrative example
Comparison Analysis: a 12, 33 and 69 node network
Comparison Analysis: a uniformly distributed test system from 1000 to 3000 nodes
(37)
1224
"
0.4576 + j1.0780
0.1560 + j0.5017
0.1535 + j0.3849
0.1560 + j0.5017
0.4666 + j1.0482
0.1580 + j0.4236
0.1535 + j0.3849
0.1580 + j0.4236
0.4615 + j1.0651
.0296
.0683
Z = R + jX = .0296 + j .0683
.0296
.0683
(38)
1 1 1
T=
0 1 0
0 0 1
(39)
(40)
(41)
T
T DX T = .0683 .1367 .0683
.0683 .0683 .1367
Using 22, solution reached at iteration 3 for = 104 and displayed in Table I. Results are
presented in per unit and degrees.
TABLE I
4 Node State of the System - balanced Approach
V0
0
V1
1
V2
2
V3
3
1.000 0.00 0.987 -1.59 0.981 -2.40 0.981 -2.40
1225
Using SB =3.33MW and VB = 12.47/ 3kV. Three phase data is given in p.u.:
Sa = Sb = Sc =
0 0.2 0.2
Three phase impedance matrix is the same for all line sections:
(42)
(43)
(44)
(45)
As this system is modeled with three phase circuits in all section lines, three phase upper triangular matrix is given by the following expression:
Tabc
U U
= 0 U
0 0
U
0
U
where
1 0
U= 0 1
0 0
0
0
1
(46)
T
abc
Using 31 ,three phase TRXabc matrix is formed by Tabc T Dabc
R Tabc and Tabc DX Tabc shown
in Tables II and IV, respectively.
Reference three phase voltages at slack node 0 and initial values in nodes 1,2 and 3 are given by:
1226
Tabc
.0294
.0100
.0098
.0294
.0100
.0098
.0294
.0100
.0098
.0100
.0299
.0101
.0100
.0299
.0101
.0100
.0299
.0101
.0098
.0101
.0296
.0098
.0101
.0296
.0098
.0101
.0296
TABLE II
Dabc
R Tabc Matrix
.0294
.0100
.0098
.0587
.0200
.0197
.0294
.0100
.0098
.0100
.0299
.0101
.0200
.0599
.0203
.0100
.0299
.0101
.0098
.0101
.0296
.0197
.0203
.0592
.0098
.0101
.0296
.0294
.0100
.0098
.0294
.0100
.0098
.0587
.0200
.0197
.0100
.0299
.0101
.0100
.0299
.0101
.0200
.0599
.0203
.0098
.0101
.0296
.0098
.0101
.0296
.0197
.0203
.0592
.0692
.0322
.0247
.0692
.0322
.0247
.1383
.0644
.0494
.0322
.0672
.0272
.0322
.0672
.0272
.0644
.1345
.0544
.0247
.0272
.0683
.0247
.0272
.0683
.0494
.0544
.1367
TABLE III
Tabc T Dabc
X Tabc Matrix
.0692
.0322
.0247
.0692
.0322
.0247
.0692
.0322
.0247
.0322
.0672
.0272
.0322
.0672
.0272
.0322
.0672
.0272
.0247
.0272
.0683
.0247
.0272
.0683
.0247
.0272
.0683
Vabc0 =
.0692
.0322
.0247
.1383
.0644
.0494
.0692
.0322
.0247
0
Vabc1
.0322
.0672
.0272
.0644
.1345
.0544
.0322
.0672
.0272
0
Vabc2
.0247
.0272
.0683
.0494
.0544
.1367
.0247
.0272
.0683
0
Vabc3
"
16
16 0
120o
16 120o
(47)
Using 34, solution reached at iteration 3 for = 104 and displayed in Table IV. Results are
presented in per unit and degrees.
TABLE IV
4 Node State of the System - balanced Approach
p
a
b
c
V0
1
1
1
0
V1
1
V2
2
V3
3
0
0.989 -0.95 0.984 -1.43 0.984 -1.43
-120 0.994 -120.86 0.991 -121.30 0.991 -121.30
120 0.993 119.02 0.989 118.52 0.989 118.52
These illustrative examples were solved using Microsoft Excel, and spreadsheets can be requested
to the author.
1227
Number of Iterations
12-bus 33-bus 69-bus
Complex Back/Forward Sweep
5
6
7
NR (Matpower)
4
4
4
TRX Method
5
6
7
CPU Time (103 Seconds)
12-bus 33-bus 69-bus
Complex Back/Forward Sweep
0.78
0.94
4.38
NR (Matpower)
8.17
10.44
12.15
TRX Method
0.64
0.82
1.25
Results show that NR method has better convergence behavior than TRX and Backward/Forward
Sweep. However, despite TRX method requires the same number of iterations than BW/FW sweep
method, it has the better CPU time. Regarding NR and BW/FW sweep methods, these results
confirm the conclusions reported by Eminoglu in broad comparative convergence analysis among
several BW/FW sweep based methods [33]. In this paper, it was used the same test systems in
order to compare with results reported in this study. Proposed TRX method presents the slowest
convergence rate but with the highest computationally efficiency. Note that in these small examples
CPU time corresponding to NR results is extremely different than other methods. This is due to
required I/O data exchange time in sparse matrix handling.
1228
%V0n =
V0 Vn
S3
=
(R cos + X sin )
V0
5V02
(49)
where S3 = 0.4, the load factor is equal to 1. For V0 = 1, 49 solution is Vn is 0.9940pu. In order
to verify this solution, an exact power flow solution is obtained using NR for n=1000: at the end
of the feeder, voltage is Vn =0.99398 and angle n =-0.016 radians.
The n-node network has been solved using the proposed TRX method and two alternative methods (NR, Back/Forward Sweep) varying n parameter from 1000 to 3000 nodes. All simulations lead
to the same solution.
CPU process time results are depicted in Fig. 6. Convergence time spent in the iterative process
does not consider I/O data exchange time.
When I/O data exchange is not considered, and CPU time is only related to the while loop (see
Appendix B), results show that TRX method has better CPU time behavior respect to NR and
1229
BW/FW sweep method. It must be pointed out that BW/FW sweep method developed in [24] is
not direct, as indicated in paper title but iterative like all sweep-based methods. Not considering
I/O data exchange is justified by the fact that under DSM environment, a sentinel program should
maintain all TRX matrices in memory with the most realistic topology.
As a result, the proposed TRX method presents the slower convergence rate but with the higher
computationally efficiency. The overall process is based on summation and multiplication of fixed
real numbers numbers previously allocated in memory being highly competitive respect to NR with
LU factorization and sparse handling. The TRX matrix does not require update at each iteration
like Jacobian in Full NR method. In this case, CPU time corresponding to NR results is the second
best. The worst behavior correspond to the standard BW/FW sweep in complex form which is
extremely affected by operations required by summation and multiplication of complex numbers.
The proposed method is a valuable tool to analyze aerial distribution systems, since they can
be modeled as a single series impedance. Research efforts are focused in the generalization of
the method including shunt impedances associated to cables. Future research is also oriented in
compare the proposed methodology with other distribution-oriented power flow methodologies, in
particular decoupled versions of NR and BW/FW sweep methods based on biquadratic formula.
Ill-conditioned systems will be also considered.
V. Conclusion
This paper proposes a new methodology to assess the power flow solution of distribution networks
by merging the standard Backward/Forward Sweep process in a unique state-of-the-system calculation formula. The proposal is suitable to be applied in balanced and unbalanced aerial distribution
systems, radially or weakly meshed operated with distributed generation. Proposed methodology
has been applied in test-case systems showing better performances than other large-scale methods.
Appendices
A. Nomencalture
TRXabc
TRX
DZ
DR
DZ
I
Ii
Ix
Iy
Iabc
Iabcx
Iabcy
J
J ij
Jx
Jy
n
P
1230
Q
Pi
Qi
PDi
QDi
PGi
QGi
R
Rabc
Rij
S
Sabc
Si
SDi
SGi
T
Tabc
V
Vi
Vx
Vy
Vabc
Vabcx
Vabcy
X
Xabc
Xij
Z
Zabc
Z ij
Zij
Z aaij
Z abij
1231
References
[1] W. F. Tinney and C. E. Hart, Power Flow Solution by Newtons Method, IEEE Transactions
on Power Apparatus and Systems, Vol. PAS-86, No. 11, pp. 1449-1460, Nov. 1967
[2] B. Stott and O. Alsac, Fast decoupled load flow, IEEE Transactions on Power Apparatus and
Systems Vol. PAS-93, No. 3, pp. 859-869, 1974
[3] R.D. Hoestenbach, Load Flow Analysis of Electrical Distribution Systems Using the Digital
Computer, IEEE Transactions on Industry Applications, Volume IA-12, No. 2, pp. 152 - 160,
March 1976
[4] S.C. Tripathy, G.D. Prasad, O.P. Malik, O. P. and G.S. Hope, Load flow solutions for illconditioned power systems by a Newton-like method, IEEE Transactions on Power Apparatus and
Systems, Vol. PAS-101, No. 10, pp. 3648-3657, 1982.
[5] M.M.M. El-Arini, Decoupled power flow solution method for well-conditioned and illconditioned power systems, IEE Proceedings Generation, Transmission and Distribution, Vol.
140, No.1,pp.7-10, Jan. 1993
[6] H. Chiang, and Hsiao, A decoupled load flow method for distribution power networks: algorithms, analysis and convergence study, Electrical Power & Energy Systems, Vol.13, No.3, pp.
130-138, 1991
[7] R. D. Zimmerman and H. D. Chiang, Fast decoupled power flow for unbalanced radial distribution systems, IEEE Transactions on Power Systems, vol. 10, pp. 2045-2052, Nov. 1995.
[8] F. Zhang and C. Cheng, A modified newton method for radial distribution system power flow
analysis, IEEE Transactions on Power Systems vol. 12, pp. 389-397, 1997
[9] P. Aravindhababu, R. Ashokkumar A Fast Decoupled Power Flow for Distribution Systems,
Electric Power Components and Systems, Vol. 36, No.9, pp. 932-940, September 2008
[10] S.K. Goswami and S.K. Basu, Direct solution of distribution systems IEE Proceedings C
Generation, Transmission and Distribution Vol. 138, No. 1 pp.78-88, Jan 1991
[11] T.Chen K. Hwang, P. Kotas and E. Chebli, Distribution system power flow analysis - a rigid
approach, IEEE Transactions on Power Delivery vol. 6, pp 1146-1152, 1991
[12] R. Berg, E.S. Hawkins, and W.W. Pleines, Mechanized Calculation of Unbalanced Load Flow
on Radial Distribution Circuits, IEEE Transactions on Power Apparatus and Systems, Volume
PAS-86, No. 4, pp.415-421, Apr 1967
[13] W. H. Kersting and D. L. Mendive, An application of ladder network theory to the solution
of three phase radial load flow problem, in Proc. IEEE PES Winter Meeting, 1976, New York,
paper A76044-8 (IEEE, New York, 1976).
[14] D. Shirmohammadi, H.W. Hong, A. Semlyen, and G.X. Luo, A compensation-based power
flow method for weakly meshed distribution and transmission networks, IEEE Transactions on
Power Delivery, Vol. 3, No. 2, pp.753-762, 1988.
[15] S. Ghosh and D. Das, Method for load-flow solution of radial distribution networks, IEE
Proc. Generat. Transm. Distrib., Vol. 146, No. 6, pp. 641-648, 1996.
[16] Y Fukuyama, Y Nakanishi, H-D Chiang Fast distribution power flow using multi-processors,
Electrical Power & Energy Systems, Vol. 18, No. 5, pp. 331-337, 1996
[17] D. Thukaram, H.M.W.Banda, and J. Jerome, A robust three-phase power flow algorithm for
radial distribution systems, Electric Power Systems Research, Vol. 50, No. 3, pp. 227-236, 1999.
1232
[18] S. Jovanovic and F. Milicevic, Triangular distribution load flow, IEEE Power Engineering
Review, pp. 60-62, 2000.
[19] D. Rajicic, R. Taleski, Two novel methods for radial and weakly meshed network analysis,
Electric Power Systems Research, Vol. 48, No.2, pp. 79-87, Dec. 1998
[20] Aravindhabuba, P., Ganapathy, S., and Nayar, K. R., A novel technique for the analysis of
radial distribution systems, Electric Power Systems Research, Vol. 23, No. 3, pp. 167-171, 2001.
[21] Y. Zhu and K. Tomsovic, Adaptive power flow method for distribution systems with dispersed
generation, IEEE Transactions on Power Delivery, Vol. 17, No. 3, pp. 822-827, 2002.
[22] J. Liu, M.M.A. Salama and R.R. Mansour, An efficient power flow algorithm for distribution
systems with polynomial load, Int. J. Elect. Eng. Educat., Vol. 39, No. 4, pp. 371-386, 2002.
[23] M. Afsari, S. P. Singh G. S. Raju G. K. Rao A Fast Power Flow Solution of Radial Distribution
Networks, Electric Power Components and Systems, 30:1065-1074, 2002
[24] J.H. Teng,A direct approach for distribution system load flow solutions, IEEE Transactions
on Power Delivery, Vol. 18, No. 3, pp. 882-887, 2003.
[25] K, Prasad, N.C. Sahoo, A. Chaturvedi and R. Ranjan. A simple approach for branch current
computation in load flow analysis of radial distribution systems, Int. J. Elect. Eng. Educat. Vol.
44 No. 1 pp. 49-63, Jan. 2007
[26] G.W. Chang, S.Y. Chu and H.L. Wang An Improved Backward/Forward Sweep Load Flow
Algorithm for Radial Distribution Systems, IEEE Transactions on Power Systems, Vol. 22, No.
2, pp.882-884, May 2007
[27] W.C. Wu and B.M. Zhang, A three-phase power flow algorithm for distribution system power
flow based on loop-analysis method, Electrical Power & Energy Systems, pp. 8-15, No. 30 (2008)
[28] R. Cespedes, New method for the analysis of distribution networks, IEEE Transactions on
Power Delivery, Vol. 5, No. 1, pp. 391-396, 1990
[29] D. Das,H.S. Nagi and D.P. Kothari, Novel method for solving radial distribution networks,
IEE Proc. Generat. Transm. Distrib., Vol. 141, No. 4, pp. 291-298, 1994.
[30] A. Gomez Exposito and E. Romero Reliable Load Flow Technique for Radial Distribution
Network, IEEE Transactions on Power Systems, Vol. 14. 1999, pp 1063-1069.
[31] U. Eminoglu and M.H. Hocaoglu, A new power flow method for radial distribution systems
including voltage dependent load models, Electric Power Systems Research, Vol. 76, No. 1-3,
pp. 106-114, 2005
[32] L.R.Araujo, D.R.R. Penido, S.Carneiro, J.L.R.Pereira and P.A.N. Garcia, A Comparative
Study on the Performance of TCIM Full Newton versus Backward-Forward Power Flow Methods
for Large Distribution Systems, IEEE Power Systems Conference and Exposition, 2006 pp.522526, Nov. 2006
[33] U. Eminoglu amd M.H. Hocaoglu Distribution Systems Forward/Backward Sweep-based
Power Flow Algorithms: A Review and Comparison Study, Electric Power Components and
Systems Vol. 37 No.1 pp.91-110, 2009
[34] C.E.Lin, Y.W. Huang and C.L. Huang. Distribution system load-flow calculation with microcomputer implementation,. Electric Power Systems Research,13,pp.139-45,1987
[35] E. Bompard, E. Carpaneto, G. Chicco, R. Napoli Convergence of the backward/forward sweep
method for the load-flow analysis of radial distribution systems, Electrical Power & Energy
Systems No. 22 pp. 521-530, 2000
[36] X. Wang, N. N. Schulz, and S. Neumann, CIM extensions to electrical distribution and CIM
XML for the IEEE radial test feeders, IEEE Transactions on Power Systems, vol. 18, pp.
1021-1028, Aug. 2003.
1233
[37] F. Milano, M. Zhou, and H. Guan Ji, Open model for exchanging power system data, IEEE
Power & Energy Society General Meeting, 2009. PES 09. pp.1-7 26-30 July 2009
[38] M. Kleinberg,K. Miu, K.and C. Nwankpa, A Study of Distribution Power Flow Analysis Using Physically Distributed Processors, IEEE Power Systems Conference and Exposition, 2006,
pp.518-521, Nov.2006
[39] A. Losi and M. Russo Object-oriented load flow for radial and weakly meshed distribution
networks, IEEE Transactions on Power Systems Vol. 18, No. 4, pp.1265-1274, Nov. 2003
[40] M.E. Baran and F.F. Wu, Optimal sizing of capacitors placed on a radial distribution system,
IEEE Transactions on Power Delivery, Vol. 4, No. 1, pp. 735-742, 1989
[41] M.E. Baran and F.F. Wu, Network reconfiguration in distribution systems for loss reduction
and load balancing, IEEE Transactions on Power Delivery, Vol. 4, No. 2, pp. 1401-1407, 1989.
[42] W.H. Kersting, Distribution System Modeling and Analysis, Boca Raton, FL: CRC Press, 2002.
[43] T. Gonen, Electric Power Distribution System Engineering, McGraw-Hill Book Company, 1986
[44] MATPOWER 3.2, Cornell University, [Online]. Available: http://www.pserc.cornell.edu/matpower
[45] Mathworks, Matlab 7.2 Users manual 2006
[46] IEEE Task Force on Open Source Sofware fo Power Systems [Online]. Available:
http://ewh.ieee.org/cmte/psace/CAMS taskforce/format.htm
Apndice I
OPTIMAL SIZING OF CAPACITORS PLACED ON RADIAL
DISTRIBUTION SYSTEM, BARAN, M. E. ET AL. 1989, [31]
139
725
Mesut E . Baran
Felix F.
Wu
88 WM 064-8
A paper recommended and approved
by the IEEE Transmission and Distribution Committee
of the IEEE Power Engineering Society for presentation at the IEEE/PES 1988 Winter Meeting, New York,
New York, January 31
February 5, 1988. Manuscript
submitted September 1, 1987; made available for
printing November 13, 1987.
:F+>;
QL@) =
Q~W)
(1)
S"t
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
726
i =0,1,. . . .nt
(7)
... =uM
(8.i)
OIUiSU,o
Where,
The objective terms. namely the real power and energy loss and the
capacitor cost. can be formulated ,asfollows. For each load level i ,let the
power loss in the system be pi(x'). Then the total cost of energy loss can
be wrimn as
Pk ,Qk : real and reactive power flows into the receiving end of
= fk+l(xk
nl
ke
(3.i)
&k+d
= [U1 . . . U,]
= [Qcn,.
. . Q-]
where, Ti is the duration of the load for load level i and the constant k, is
the energy cost per unit. The cost for the real power loss at peak load level
can be added to this sum by modifying To accordingly.
The capacitor cost h c t i o n is usually step like as shown in Fig.3
since in practice capacitors are grouped in banks of standard discrete capacities (usually 300 kvar sizes at 23 kV level).
cvb
4)
= v& = v l = XOk,
xh,=Ph=O
xhr=Qh=O
(3.iii)
Hence, for the general feeder considered, there are 3(n+l+1) DistFlow equations corresponding to Eq.(2) and Eq.(3). They will be
represented by the following equations.
G(x,u)= 0
"0
(3.ii)
(ii) at the end of the lateral, there is no power sent to the other branches,
i.e.,
(4)
O S u o Su".e
(10)
Where, uo and r, represent the size and the marginal cost of the capacitor
respectively. Note that e = 0 corresponds to the decision that the capacitor
not to be placed.
To summarize, let the types and places of nc "candidate" capacitors,
initially considered for installment, be given and let the sets C 1 ,C2 contain the switched and the fixed capacitors respectively. Then we can write
the general capacitor placement problem as a standard optimization problem as follows.
s.t.
Gi($,ui)=O
i =0,1...nr
(5)
Where, xi , ui represent the state and the contml variables corresponding
to the load profiie i respctively.
The voltage constraints can be taken into account by specifying
upper and lower bounds on the magnitude of the node voltages as follows,
vm2< v;'=Y;'(x') 1v-2
j = l . . . n i -- 0 , 1 . . . n t
(6)
(9)
x=TiPi(X')
i 4
H'(x')SO
0 I uo I u-.e
0 I U; I U:
U:
=:U
i=O,1, . . . , nt
k E C 1 = {W. cap.)
k E C z = lfixed c a p . }
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
127
This is a non-linear, mixed integer programming problem. Decision
variables, e = [e, . . .e,IT are to be used to choose the capacitors among
initially designated ones and continuous variables ui , i = 0.1.. . . .nt will
be used to determine the optimal settings of the capacitors.
Voltage regulators are not explicitly represented in the capacitor
placement problem presented here. However, the formulation and the
solution algorithm introduced in this paper can be generalized to include
the voltage regulators in the following fashion:
Voltage regulators. such as regulating transformers, can be
represented by their equivalent circuits in the DistFlow equations.
The solution algorithm for sizing problem, proposed in [14] for
determining the optimal sizes of capacitors placed on the system, can
be generalized to obtain the optimal settings of the voltage regulators
placed on the system,
Voltage regulation for a given set of capacitors and voltage regulators can be obtained again by using the solution algorithm developed
for the sizing problem. This is demonstrated in I141 by using only
the capacitors.
It seems appropriate to put the voltage regulator placement problem
(to find the locations for the voltage regulators) at the top level of
hieravhical decomposition scheme proposed in this paper in solving
the overall problem.
Further investigation is required to complete the generalization.
I e E ;U E U ]
(1 1)
where, e and U correspond to the decision vector, and the control vector,
respectively, and E and U represent the constraint sets imposed on these
vectors. We adapt a general solution approach which decomposes the
problem by making use of the following property of the optimization
f, (e& = min
{ i f f ,(e&]
WE.UEU
W E UE
min
(12)
Pfx
mill f , = ke CTiPi(Xi) +
i=O
s.t.
Gi(xi,u)=O
rckuk
kEC2
i=O,l,. .. ,nt
H'(x') 5 0
OIUIU-
assuming that for each decision, e the problem in braces, called the slave
problem,
has a solution and it is "easy" to find it. Then the main problem becomes,
min g(e)
(EE
(14)
U' -U'SO
i = I , . . . ,nr
(17)
Note that the two constraint sets (15) and (16) are coupled through (17), to
indicate that the capacitor sizes, 'U are the upper bund constraints for
capacitor settings at the other load levels, U' . This weak coupling
between 'U and ui's can be exploited to decompose the problem into
smaller subproblems. In appendix, it is shown that Psw can be decomposed into the fdlowing nt +1 base problems.
The main problem, SW,
728
43. Mixed Type Capacitor Problem
In this section, we go back to consider the general slave problem in
which it is not known a priori which capacitor is of fixed and which one is
of switched type and one has to determine the types of capacitors in addition to their settings. Note that, in general, the switching capacitors are
more expensive in both fixed cost, c and the marginal cost, r, than the
Subproblems, SWi i = 1 . . . nt
SWi
C'(x',u') = 0
H'(xi) S 0
OIUU' s u o
The subproblems, SWi involve only one load level and therefore,
they can be solved by the algorithm developed for the sizing problem for
given capacitor sizes,.'U The solutions will correspond to the oprimal
capacitor senings ,8' for the off-peak load 'eye's considered.
The main problem, SW,,is also a sizing problem; but computationally it is not of the easy type due to existence of fi
- the extra terms
coupled to the subproblems - in the objective function. Updating these
terms at each iteration during the solution of SWo requires solution of the
subproblems. However, we can use a simple, heuristic scheme to update
fi(uO)and Vfi(u? in the main problem SWo. To begin with, let the solutions of the subproblems SWi for a given 'U be denoted (ui)* . As we
move 'U from iteration to iteration in SW, ,f i (u4 and Vfi (U') should be
calculated by solving SWi with the new i? for a new (U'): , say (iii)*
We prppose, however, instead of solving SWi for a new 6")'
, to use the
old (U')' unless the constraint (17) is violated, i.e., we set the k'th component of 0'
(114's
if ii;S(u;)*
ii;
a; =
(U;)*
otherwise
(18)
f;(if)
= k, Tipi (2' ,tii)
(19)
,
Note that if 'U and i? do not differ very much, the approximation will be
good. To assure this, we start the procedure by first solving the subproblems with the capacitor sizes set to their maximum, U'"=.
The ovcrall iterative algorithm is shown in Fig.4. In the algorithm,
an iteration comprises the solution of the subproblems, SWi first and then
the main problem, WO.
Convergence check at end of an iteration involves
checking if there is a status change in the constraint set of Eq.(17). (i.e., a
non-binding constraint becomes binding or vice versa ). If there is such a
status change, then we go back and update the slave problems; otherwise
we stop iterating since the solution is converged.
s=s+l
fori = 1,...,nt
Solve SWo
Use Eq.(19) and Eq.(20) to calculate f,Oand Vfi&)
i = 1. . . . ,nl
1
1
converged
- given a node which is identified by its decision vector, e and the solution
of the corresponding Slave Problem, ( i.e., the control vector
ii = [ z i 1 . . .
and the objective f, (ii) ).
- for all existing capacitors, k = 1, . . . ,nc s.t. ek z 0
- construct U by removing the capacitor k and keeping the rest, i.e.,
0 ifj=k
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
729
- calculate the
The A f t s can used to choose the child to branch out (i.e., capacitor to
take out) since the child with the smallest Af ," is more liiely the one who
contributes to the objective the least.
The first procedure, depth-first search, visits the children of a parent
node (i.e., solves the associated slave problems) according to the order
determined by the sorting procedure and branches out on the, first one
which gives a lower objective than that of the parent. This search is of
order nc and therefore computationally attractive. However the branching
criteria is appropriate only for unconstrainted case. When the voltage constraints are active, a better criteria would be to check how well the
deficiency created by a capacitor removal be compensated by the rest of
the capacitors by Nnning the corresponding slave problems. This
corresponds to the breath-& type search which is conducted by visiting
all the children first and then branching out on the one with the smallest
objective. This search is of order nc2 and requires at most nc(nc+l)/2
slave problem solutions. However, the search can be made much faster by
noting that the objectives, f," calculated by sorting procedure will be
bigger than the ones calculated by the search. Therefore, in practice, the
search can be conducted only on the children with negative Af ,"s.
AE
VI. TESTSTUDIES
The proposed solution methodology has been implemented in Fortran 77 on both VAX llh'50 and IBM AT. The program uses the algorithm developed for the sizing problem in [14] as a subroutine to solve the
base problems. We present the test results of two systems in this section
to illustrate the performance of the proposed solution scheme.
The first test system, TS1 is a 9-branch main feeder test system
developed by Grainger et. al., [9].The second test system, TS2 is a 69branch, %lateral test system derived from a portion of the PG&E distribution system. The network data of this system is given in Table 1.
We adopted the following cost figures : energy cost,
k, =0.06 $/kwh ; capacitor fixed cost, c = lOOO$ ; capacitor marginal
cost, r, = 3 $/bur. The load duration data assumed for the systems is
given in Table 2. It is also assumed that substations have regulating
transformers which are tapped to +5% during peak load and set to nominal
otherwise. The substation voltage without tap is taken as the base kV and
the lower and upper voltage limits are assumed to be 0.9 and 1.1 p.u.
respectively.
No
Nd. N d . r(ohm)
1 0
24 ;
2e
3
4
5
6
7
8
11 9
12 10
13 11
14 12
1 5 13
16 14
17 15
18 16
19 17
20 1 8
21 19
2 2 20
23 21
24 2 2
25 23
2 6 24
27 25
5
6
7
8
9
10
2e
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
0.0005
0.0005
0.
0.0015
0.0251
0.3660
0.3811
0.0922
0.0493
0.8190
0.1872
0.7114
1.0300
1.0440
1.0580
0.1966
0.3744
0.0047
0.3276
0.2106
0.3416
0.0140
0.1591
0.3463
0.7488
0.3089
0.1732
Par.
x(ohm)
Rv. N d . Load
P(KW) Q(KVRR)
0.0012
0.0012
0.
0.0036
0.0294
0,1864
0.1941
0.0470
0.0251
0.2707
0.0619
0.2351
0.3400
0.3450
0.3496
0.0650
0.1238
0.0016
0.1083
0.0696
0.1129
0.0046
0.0526
0.1145
0.2475
0.1021
0.0572
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
2.60
2.20
40.40 30.00
75.00 54.00
30.00 22.00
28.00 19.00
145.00 104.00
145.00 104.00
8.00
5.50
8.00
5.50
0.
0.
45.50 3 0 . 0 0
60.00 35.00
60.00 3 5 . 0 0
0.
0.
1.00
0.60
114.00 81.00
5.30
3.50
0.
0.
28.00 20.00
0.
0.
14.00 10.00
14.00 10.00
9?0
270
6
5
2
1
3700
191
.901
5
4550 960
226 660
.90 360
700
1160
210
200
System
TS1.
TS2
Br. S d . Rv.
4700
225
.90
910
630
160
175
.90
9
V,,,i,,
itr.
Br. S d . Rv. - B r .
3 800
225
rev
x(ohm)
So
S,
S2
To
TI
T2
1.1
0.6
1.
0.3
0.5
1OOO.
1OOO.
6760.
6760.
1OOO.
1OOO.
1.8
N d . Load
P(Kw)
Q(KVAR)
No
Nd.
N d . r(ohm)
28
29
30
31
32
33
34
35
2
27
28
29
30
31
32
33
27
28
29
30
31
32
33
34
0.0044
0.0640
0.3978
0.0702
0.3510
0.8390
1.7080
1.4740
0.0108
0.1565
0.1315
0.0232
0.1160
0.2816
0.5646
0.4873
26.00
26.00
0.
0.
0.
14.00
19.50
6.00
18.60
18.60
0.
0.
0.
10.00
14.00
4.00
36
37
38
39
40
41
42
43
44
45
46
2e
27e
28e
65
66
67
68
69
70
88
89
27e
28e
65
66
67
68
69
70
88
89
90
0.0044
0.0640
0.1053
0.0304
0.0018
0.7283
0.3100
0.0410
0.0092
0.1089
0.0009
0.0108
0.1565
0.1230
0.0355
0.0021
0 .E509
0.3623
0.0478
0.0116
0.1373
0.0012
26.00
26.00
0.
24.00
24.00
1.20
0.
6.00
0.
39.22
39.22
18.55
18.55
0.
17.00
17.00
1.00
0.
4.30
0.
26.30
26.30
47
48
49
50
3
35
36
37
35
36
37
38
0.0034
0 .0851
0.2898
0.0822
0.0084
0.2083
0.7091
0.2011
0.
0.
56.40
79.00
384.70 274.50
384.70 274.50
B r . S d . Rv. - Br. P a r .
No N d . N d . r(ohm) X ( 0 h m
Load
R V . Nd.
P(KW)
Q(")
40.50
3.60
28.30
2.70
51
52
7
40
40
41
0.0928
0.3319
0.0473
0.1114
53
54
55
56
57
58
59
60
61
62
63
64
65
8
42
43
44
45
46
47
48
49
50
51
52
53
42
43
44
45
46
47
48
49
50
51
52
53
54
0.1740
0.2030
0.2842
0.2813
1.5900
0.7837
0.3042
0.3861
0.5075
0.0974
0.1450
0.7105
1.0410
4.35
3.50
0.0886
26.40 19.00
0 .lo34
24.00 17.20
0.1447
0.
0.
0.1433
0.
0
..
0.5337
0.
0.
0.2630
0.1006 100.00 72.00
0
.
0.
0.1172
0.2585 1244.00 888.00
3
2
.
0
0
2
3.00
0.0496
0.
0.
0.0738
2
2
7
.
0
0
1
6
2.00
0.3619
59.00 42.00
0.5302
66
67
10
55
55
56
0.2012
0.0047
0.0611
0.0014
18.00
18.00
13.00
13.00
68
69
11 57
57 5 8
0.7394
0.0047
0,2444
0.0016
28.00
28.00
20.00
20.00
S u b s t a t i o n V o l t a g e (kv)
B a s e KVA
B a s e V o l t a g e (kV)
t o t a l load: P(KW)
3 8 0 2 . 1 9 , Q(KVAR)
--
12.66
10.000
12.660
2694.60
730
Test run for TSl, summarized in Table 3, starts with 4 capacitors.
The solution of the slave problem for the mot (initial one) gives the
optimal settings for these 4 capacitors. Then the capacitor's contributions
to the objective, A j t s are calculated by using the sorting procedure. The
QC6 - have
results indicate that three of the capacitors - Q,,
Af: e 0; which implies that they are not economicallyfeasible ( i.e., their
economic contribution due to energy loss reduction is less than their cost).
Therefore, the search is conducted only on these three capacitors at the
second search level. As a result, QC2is found to be the least economical
and hence is taken out. This corresponds to the branching out on the first
child of the mot in the table. Then a new search process resumes from this
new node; first sorting out its children by using the sorting procedure and
then visiting the ones that are economically infeasible. The solution is
obtained at the third level of search when the evaluation of capacitors Qc6
and Q,, in the last node by the sorting procedure indicated that they are
economically feasible.
The total run time for this test on VAX is about 45 sec. of CPU and 8
sec. of vo.
. e,,,
Test System 2
Test run for TS2 is summarized in Table 4. The test starts with 5
capacitors. After visiting the root. the capacitor with the zero setting, Qcas
is taken out and the other two Q, and
are found to be economically
infeasible by the sorting procedure. The search therefore is conducted
only on these two capacitors and as a result the last node containing the
capacitors QCl8and Qc52 is identified as the solution node. The total run
time for this test on VAX is recorded as 165 sec. of CPU and 8 sec. of VO.
We now present the test runs for the switched capacitor problem on
the same test systems. The optimal places obtained from the general fixed
capacitor problem tests are used in these tests also to avoid the search.
Test System 1
The first step of the test runs for TS1 is the solution of the
corresponding slave problems SWi ,i = 1,2 to get the optimal settings for
the off-peak load levels assuming no limit on the size of the capacitors.
Each of such solution is obtained in about less than 2 nc iterations by calling the sizing problem Subroutine. In the second step of the test, the main
problem, SW, is solved in about 3 nc iterations by using the solution procedure described in section 5. In the third step, convergence checks indicate that the setting of QC5 for the first load level is binding, i.e.,
= Q,"S. This constitutes the end of the first iteration. Another iteration
is performed to see if the binding status of the capacitors will change. The
convergence is obtained at the second iteration when the capacitor settings, Q2 ,Q:, are updated and it is found that there is no binding status
change.
The solution is summarized in Table 5. The box in the table is similar to that of fixed capacitor case; except here, in addition to the capacitor
sizes,,:Q capacitor settings at the off-peak load levels, d: ,d:, are given
also. The total run takes about 12 sec. of CPU and 2 sec. of VO on VAX.
Table 5 :Test run results for TSl - Switched Capacitor case.
QZ
Q:
Q,"
bus
rev.
AE
Vmin
1 1
Test System 2
Test run for TS2 is similar to that of TSl. The solution is obtained in two
iterations and found out that only the setting of Qc18 for the first load level
The solution is summarized in Table 6. The
is binding, i.e., QJ18 =
total run takes about 82 sec. of CPU and 4 sec. of VO on VAX.
Table 6 : Test run results for TS2 - Switched Capacitor case
3-y
137.;;
1 200
rev.
AE
vmi,
39180
785
.go9
QZ
Q,'
Q,"
197
620
330
1240
330
1700
bus
18
52
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
73 I
Analysis of the Results
Consider the starting point of TS2 in Table 4, where the fixed cost of
capacitors m not included. The solution of capacitor sizing gives 1040
kvar and 210 kvar on the nodes 52 and 47 of a lateral respectively, and 170
kvar and 230 kvar on the nodes 11 and 18 of the main feeder respectively.
In the system, the loads in the laterals are more concentrated whereas the
loads in the main feeder are more evenly distributed. The result is that the
size of the capacitors are also more concentrated in the lateral and more
evenly distributed in the main feeder. This further reaffirms the fact that
the nature of reactive power compensation is rather local.
W.CONCLUSIONS
In this paper, a general formulation and an efficient solution metho-
REFERENCES
[l] Distribution Systems, East Pittsburgh, PA. : Westinghouse Electric
Corp., 1965
[2] R. F. Cook, "Optimizing the Application of Shunt Capacitors for
Reactive Volt-Ampere Control and Loss Reduction" , AIEE Trans.,
vol. 80, pp. 430-444, August 1961.
[3] Y.G. Bae, "Analytical Method of Capacitor Application on Distribution Primary Feeders", IEEE Trans. on Power Apparatus and Systems, vol. 97, pp. 1232-1237, July/Aug. 1978.
[4] H. Duran, "Optimum Number, Location, and Size of Shunt Capacitors in Radial Distribution Feeders: A Dynamic Programming
Approach, IEEE Trans. on Power Apparatus and Systems, vol. 87,
pp. 1769-1774, Sept. 1968.
U 1
U = { U I Eq.(16) is satisfied )
U o ={ uo I Eq.(15) is satisfied )
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
i=l
132
Then Psw can be re-written as
min
u9u
if,
=f,(u?+fr(u)
; U E U ; F(U',U)L;O l(a.1)
UOEU,
f s (U?
:$
i f r (u)-F(uo*u) 5 0 1
min f 0 ('4
=fm
UWO
9
'
( +fJ('4
6.3)
To assure that the above assumption holds, we must avoid the values of 'U
such that f,@) does not have a feasible solution. For this, we define a
new set. V as follows.
V = {U'
The set V can be thought of as the projection of the constraint set defined
by F(uo,u) and U onto the space defined by 'U alone as illustrated in
Fig.a.1.
U, c
m0
U'E
nv 1
(a.4)
U0
= ic[fri (U')
Mesut E. Baran received his B.S. :ind M.S. from Middle East Technical
University, Turkey. He is currently a Ph.D. student at the University of
California, Berkeley.
Felix F. Wu received his B.S. from National Taiwan University, M.S.
U' EUi
from the University of Pittsburgh, and Ph.D. from the University of California, Berkeley. He is a professor of Electrical Engineering and ComWhere, Ui = { ui I Eq.(I6) } and f,(u? =
(U?.
of the main problem Swo and the subproblems puter Sciences at the University of Califomia, Berkeley.
The explicit
SWi i = 1. . . . ,nt are given in "4.2.
S.t.
ui - U 0 1 0
us,
Benders Decomposition
Assuming that the duality conditions holds for the switched capacitor
problem of (a.2) [16], the main problem can be transcribed into the following form.
m o b
min
uo E U,
i = 1.. . . .nt
(a.5.ii)
Where, U" and I.' corresponds to the solution of the subproblem SW, at
iteration j.
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
133
Discussion
Roger C. Dugan (McGraw-Edison Power Systems, Cooper Industries
Inc., Canonsburg, PA): The reader will discover in a few moments that I
have several objections to the methods presented in this paper. However, 1
do not wish for my objections to reflect poorly upon the efforts of the
authors. It would appear that a great deal of good work has been done and I
suspect that this paper was intended to emphasize the application of more
sophisticated techniques to the problem. Therefore, it may be unfair to
expect the authors to respond to all of my objections because they are based
on more practical aspects and also apply to previous investigators in this
field who have made similar assumptions.
First, one would hope that a method employing nonlinear programming
techniques would be faster than simple exhaustive searches. It is not
apparent from the paper that this would be the case, and the times quoted
during the papers presentation and discussion lead me to believe otherwise.
The number of discrete solutions to this problem are not necessarily large
due to the practical constraints that I will mention below. Therefore, an
intelligent exhaustive search method that automatically discards many cases
due to its knowledge of the way the feeder operates can be reasonably fast. I
have investigated a number of search techniques that give very good, but
perhaps not optimal, solutions, and the execution times grow approximately
linearly with problem size rather than geometrically.
The method proposed in the paper is based on the assumption that the
feeder is discrete and the capacitor size is continuous. Perhaps, a more
realistic choice would have been the opposite. There are many nodes in a
circuit, but utilities generally wish to consider only two or three different
capacitor sizes, for example 600- and 1200-kvar banks. I have generally
approached the problem by assuming that both the feeder and capacitor
sizes are discrete. When considering such a s d number of sizes, a simple
search can often be done quickly.
Another assumption I would question is that the capacitor is a source of
assume that this implies a constant source because I
reactive power (Q). (I
am not able to ascertain otherwise from the paper.) Of course, this will
introduce inaccuracies because a capacitor is a constant impedance element.
Since capacitor placement on a feeder significantly affects the voltage, this
assumption weakens any claim that the method results in an optimal
location. It would seem to be a simple matter to correctly represent
capacitors and avoid this difficulty.
My examination of this problem has also indicated that line regulator tap
position and control characteristics affect the optimal solution. It is not
clear how the method presented in the paper properly accounts for these
effects.
The constant P-Q load model employed inthe proposed method also leads
to inaccuracies. The P-Q load model is a peculiar bias of transmission
analysts, and it should not be employed on distribution systems without
question. Frantz et al. [I] have clearly shown that distribution system loads
are sensitive to voltage. My experimentation with different load models has
shown that one will usually get a different optimal solution for each load
model. The P-Q load model is best employed to establish the base case
voltages from known load conditions. Then one should switch to a more
realistic model when studying capacitor additions. In the absence of better
information, I will typically use a load model in which the P varies linearly
with voltage and the Q varies by the square of the voltage. Lacking this
capability, I suspect that a simple constant impedance load model would be
better than a constant P-Q model.
Differing economic evaluation criteria among utilities require different
approaches to the optimization problem. Utilities using very high values for
released substation and generation capacity savings may achieve a more
economical solution by optimizing the peak load condition first, although I
must admit to being skeptical of this. Then the load is decremented in steps
to determine when switched banks should be turned off. For utilities where
the costs of losses is more important, a more economical solution can
generally be achieved by fist optimizing the location of fixed banks at
minimum load and then incrementing the load in steps to determine the
location, size, and switching levels of the switched banks.
I believe that the latter approach is more practical for most utilities. One
reason is that most feeders operate near minimum or average load levels
much more than they operate near peak loads. Another reason is that I
question whether values for substation and generation costs are based on
assumptions compatible with assumptions made for capacitor economic
evaluations. This approach can be easily programmed using an intuitive
algorithm that recognizes how a feeder typically operates. It yields a nearoptimal solution that is difficult to improve upon significantly. I will state
the algorithm in words, giving the reader the freedom of choice in selecting
techniques for solving the load flow and making decisions.
systems, ZEEE Trans. Power App. Syst., April 1984, pp. 819-831.
Manuscript received February 19, 1988
M. E. Baran and F. F. Wu :We would like to thank Mr. Dugan for his
interest in the paper and his insightful questions about the capacitor placement problem.
The algorithm outlined by Mr. Dugan is a special case of switched
capacitor problem introduced in this paper. In Sec. 4.2 it is shown that
assuming all the capacitors are of switched type, and for a given set of
capacitors placed on the system, the problem can be decomposed into
smaller subproblems each of which corresponds to minimization of losses
at each of the load levels considered. However, these subproblems are
coupled to each other due to the cost of capacitors. It is easy to show that
when the cost of capacitors are neglected the subproblems become
independent and hence the optimization for each load level can be carried
out independently. Mr. Dugans algorithm incorporates both placement
and sizing problems into this decomposition scheme. It is indeed a good
idea especially since he considers the capacitor sizes as discrete. Therefore, his algorithm will work, as he points out, when the objective function
is power loss minimization or voltage regulation.
However, for the general case, where the cost of capacitor is important and it effects the number of capacitors to be placed, the method may
not give good answers because of the coupling between the subproblems.
Hence, in this case, it is not easy to answer the question how capacitors
should be added as the load level increases. It seems that the best aid to
answer this question would be the use of switched capacitor problem.
Starting from a candidate set of capacitors and assuming them all switchable the problem can be solved by using the switched capacitor algorithm
introduced in. this paper. The solution will give the capacities of the capacitors, U and their settings at other load levels, ui ,i = 1, . . . ,nt assuming
the capacities are continuous variables. These results then can be used in
answering the question mentioned above and hence the capacitors can be
placed by the method proposed by h4r. Dugan. This way the search intro-
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
134
duced in this paper may not need to be c a n i d out any further than the
mot.
Another point raised by Mr. Dugan is the assumption made in
modeling the capacitors. It involves approximating the reactive power
injected by a capacitor as constant, independent of the voltage. This
assumption is justified based on the fact that V, = 1 pa. and sizes of capacitors determined by the solution need to be munded off to get the practical size of capacitors. However, the exact model can be incorporated in
the method if needed. This is explained in the closure of [14]. Note that
this approximation will most likely affect the sizes, not the location, of
capacitors.
The power Row model used in this paper (DistFlow equations) can
handle voltage dependent loads and the solution algorithm can be generalized to take into account such loads, as explained again in the closure of
~41.
Manuscript received May 2 , 1988.
Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.
Apndice J
SIMPLE AND EFFICIENT COMPUTER ALGORITHM TO SOLVE
RADIAL DISTRIBUTION NETWORKS, RANJAN, R. ET AL. 2003, [27]
150
To cite this Article Ranjan, Rakesh and Das, D.(2003)'Simple and Efficient Computer Algorithm to Solve Radial Distribution
D. DAS
Electrical Engineering Department
Indian Institute of Technology
Kharagpur, India
A simple and ecient algorithm is presented to solve radial distribution networks (RDN). It solves the simple algebraic recursive expression of voltage
magnitude and all the data are stored in vector form. The algorithm uses the
basic principle of circuit theory and can be easily understood. The proposed
algorithm has been tested with several distribution networks and results are
compared with two other existing methods. The eectiveness of the proposed
algorithm is demonstrated through two examples.
Keywords
Nomenclature
NB
LN 1
PL(i)
QL(i)
|V (i)|
R(jj)
X(jj)
Z(jj)
I(jj)
P (m2)
Q(m2)
(m2)
LP (jj)
LQ(jj)
95
96
IS (jj)
IR(jj)
PLOSS
QLOSS
1. Introduction
In the past few years, the developments of automated distribution systems solutions have increased considerably. With the development of the microcomputer,
the requirement of distribution substation-owned computer programs has become
a necessity. However, the choice of a solution method for the practical application is
dicult. It requires a careful analysis of comparative advantages and disadvantages
of those methods available in respect to storage, computation speed, and convergence criterion. Generally, radial distribution networks has a high R/X ratio. Due
to this, conventional Newton Raphson [1] and fast decoupled load ow [2] methods
fail to converge. Many other researchers [35] have suggested modied versions of
conventional load ow methods with a high R/X ratio.
Kersting and Mendive [6] and Kersting [7] have developed load-ow techniques
based on ladder theory and Stevens et al. [8] modied it and proved faster than
earlier methods. However, it fails to converge in ve out of twelve case studies.
Baran and Wu [9] have developed a load-ow method based on the Newton-Raphson
method but it requires a Jacobian matrix, a series of matrix multiplications, and
at least one matrix inversion. Hence, it is not computationally ecient. Chiang [10]
has developed decoupled and fast decoupled load-ow methods based on a method
suggested by Baran and Wu [9]. The very fast decoupled method is impressive
because it does not require any Jacobian matrix. Many other researchers [115] have
proposed generalized methods of modeling and analysis of distribution systems.
However, the diculty arises from the fact that no method posseses all the desirable
features.
In this article, a simple algorithm that is based on basic systems analysis
methods and circuit theory is developed. The purpose of this article is to develop a
new calculation model that requires less computer memory and is computationally
fast for radial distribution networks. The proposed method involves only recursive
algebraic equations to be solved to get the following information:
1. Status of the feeder line, overloading of the conductor and feeder line currents;
2. Whether the system can maintain adequate voltage level for the remote
loads;
3. The line losses in each segment;
4. Suggestion of the necessity of rerouting or network reconguration for the
existing distribution networks.
The proposed method is compared with those of Das et al. [12] and Baran and
Wu [9]. It is observed that the proposed algorithm is computationally very ecient. Several distribution networks have been tested with this algorithm with the
consideration that all loads are constant power. However, the algorithm can easily
accommodate composite load modeling, if the composition of the load is known.
The algorithm has a good convergence property for practical radial distribution
networks.
97
2. Circuit Model
In this section, a circuit model of a radial distribution network (RDN) is presented.
It is assumed that a three-phase RDN is balanced and can be represented by
equivalent single line diagram. Line shunt capacitance at distribution voltage level
is negligibly small. Figure 1 shows single line diagram of a sample radial distribution
network.
The electrical equivalent of Figure 1 is shown in Figure 2.
|V (m1)|(m1) |V (m2)|(m2)
Z(jj)
(1)
and
P (m2) jQ(m2) = V (m2) I(jj)
(2)
where Z(jj) = R(jj)+X(jj), m1, and m2 are the sending and receiving end nodes,
respectively, {m1 = ISS (jj) and m2 = IRR(jj)}.
P (m2) = sum of the real power loads of all the nodes beyond node m2 plus the
real power load of the node m2 itself plus the sum of the real power losses of
all the branches beyond node m2.
Q(m2) = sum of the reactive power loads of all the nodes beyond node m2 plus the
reactive power load of the node m2 itself plus the sum of the reactive power
losses of all the branches beyond node m2.
98
Table 1
Branch number, sending end node, and receiving end
nodes of Figure 1
Branch no.
(jj)
1
2
3
4
5
6
7
8
9
10
1
2
3
4
2
8
9
3
6
9
2
3
4
5
8
9
10
6
7
11
{B(jj) A(jj)}
(3)
(4)
(5)
(6)
LQ(jj) =
(7)
99
number, sending end, and receiving end nodes of the feeder shown in Figure 1 are
given.
First set IR(jj) = IRR(jj) and IS (jj) = ISS (jj) for jj = 1, 2, . . . , LN 1.
Now for jj = 1 (rst branch of Figure 1 and Table 1), IR(jj) = IR(1) = 2,
and set P (2) = PL(2), Q(2) = QL(2). Now computer logic will check whether
IR(1) = IS (i) or not for i = 1, 2, . . . , LN 1. It is seen that (Table 1) for i = 2
(Branch 2), IR(1) = IS (2) = 2 and for i = 5 (Branch 5), IR(1) = IS (5) = 2,
and corresponding receiving end nodes are IR(2) = 3 and IR(5) = 8. Now it will
compute
P (2) = P (2) + PL(3) + PL(8) + LP (2) + LP (5)
= PL(2) + PL(3) + PL(8) + LP (2) + LP (5)
and
Q(2) = Q(2) + QL(3) + QL(8) + LQ(2) + LQ(5)
= QL(2) + QL(3) + QL(8) + LQ(2) + LQ(5)
Proposed computer logic will again check whether nodes 3 and 8 are connected with
the other nodes. It is seen that node 3 is connected with node 4 (branch 3) and
node 6 (branch 8). Node 8 is connected with the node 9 (branch 6). Now computer
will compute
P (2) = P (2) + PL(4) + PL(6) + LP (3) + LP (8) + PL(9) + LP (6)
= PL(2) + PL(3) + PL(8) + LP (2) + LP (5) + PL(4)
+ PL(6) + LP (3) + LP (8) + PL(9) + LP (6)
and
Q(2) = QL(2) + QL(3) + QL(8) + LQ(2) + LQ(5) + QL(4)
+ QL(6) + LQ(3) + LQ(8) + QL(9) + LQ(6)
Similarly, computer logic will check whether nodes 4, 6, and 9 are connected
with the other nodes. It is seen that 4 is connected with node 5 (branch 4), node 6
is connected to node 7 (branch 9) and node 9 is connected with node 10 (branch 7)
and node 11 (branch 10). Therefore,
P (2) = P (2) + PL(5) + LP (4) + PL(7) + LP (9) + PL(10) + LP (7)
+ PL(11) + LP (10)
P (2) = PL(2) + PL(3) + PL(8) + LP (2) + LP (5) + PL(4) + PL(6)
+ LP (3) + LP (8) + PL(9) + LP (6) + PL(5) + LP (4) + PL(7)
+ LP (9) + PL(10) + LP (7) + PL(11) + LP (10)
For jj = 2 (branch 2), IR(jj) = IR(2) = 3 (from Table 1), set P (3) = PL(3)
and Q(3) = QL(3). Now computer logic will check whether IR(2) = IS (i) for
i = 1, 2, . . . , LN 1. It is seen that (Table 1) for i = 3 (branch 3), IR(2) = IS (3) = 3
100
and for i = 8 (branch 8), IR(2) = IS (8) = 3 and correspondingly receiving end
nodes are IR(3) = 4 and IR(8) = 6. Now it will compute
P (3) = P (3) + PL(4) + PL(6) + LP (3) + LP (8)
= PL(3) + PL(4) + PL(6) + LP (3) + LP (8)
Similarly,
Q(3) = Q(3) + QL(4) + QL(6) + LQ(3) + LQ(8)
(a)
Figure 3. (a) Flow chart of load ow technique (continues).
101
102
(b)
Figure 3. (b) Flow chart of load ow technique.
103
6. Example
To demonstrate the eectiveness of the proposed algorithm, two examples are
selected. The rst example is a 69-node radial distribution network [9] as shown in
Figure 4 in the appendix. The system data is available in [9]. Results of the load
ow study are tabulated in Table 2.
For the load ow study, we have considered the following:
substation voltage = 12.66 kV,
base kVA = 10 and base voltage = 12.66 kV.
The second example is a 33-node radial distribution network [14]. Data are
given in Table 5 in the appendix. Load ow results of the system are given in
Table 3.
Comparison of relative speed and memory requirements is given in Table 4
below.
Table 2
Load ow result of 69-node radial distribution network
Node
number
Voltage
magnitude
(p.u.)
Node
number
Voltage
magnitude
(p.u.)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
1.00000
0.99997
0.99993
0.99984
0.99902
0.99009
0.98079
0.97858
0.97745
0.97245
0.971135
0.96819
0.96526
0.96237
0.995950
0.95897
0.95809
0.95808
0.95761
0.95732
0.95683
0.95683
0.95676
0.95660
0.95643
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
0.99992
0.99975
0.99959
0.99954
0.99884
0.99884
0.99855
0.99850
0.99850
0.99841
0.99840
0.99979
0.99854
0.99470
0.99415
0.97854
0.97853
0.97466
0.97142
0.96694
0.96257
0.94010
0.92904
0.92476
0.91974
(continued )
104
Table 2
(Continued )
Node
number
Voltage
magnitude
(p.u.)
26
27
28
29
30
31
32
33
34
35
0.95636
0.95634
0.99993
0.99973
0.99985
0.99971
0.99961
0.99935
0.99901
0.99895
Node
number
Voltage
magnitude
(p.u.)
61
62
63
64
65
66
67
68
69
0.91234
0.91205
0.91166
0.90976
0.90919
0.97129
0.97129
0.96786
0.96786
Table 3
Load ow result of 33-node radial distribution network
Node
number
Voltage
magnitude
(p.u.)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
1.00000
0.99703
0.98289
0.97538
0.96796
0.94948
0.94595
0.93230
0.92597
0.92009
0.91922
0.91771
0.91153
0.90924
0.90782
0.90643
0.90439
Node
number
Voltage
magnitude
(p.u.)
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
0.90377
0.99650
0.99292
0.99221
0.99158
0.97931
0.97264
0.96931
0.94755
0.94499
0.93354
0.92532
0.92177
0.91760
0.91669
0.91640
105
Table 4
Comparison of relative speed and memory
Method
Relative memory
Proposed method
D. Das et al. [12]
Baran and Wu [9]
1
4
6
1.0
1.5
2.9
It is reported in the literature that authors have tried to solve above systems
by NR (Newton-Raphson) and GS (Gauss-Siedel) methods but for both the cases
NR and GS did not converge.
7. Conclusions
In this paper a simple and ecient computer algorithm has been presented to solve
radial distribution networks. The proposed method has a good convergence property
for any practical distribution networks with practical R/X ratio. Computationally,
this method is extremely ecient, as compared to Baran and Wu [9] and Das
et al. [12], as it solves a simple algebraic recursive equation for voltage magnitude.
Another advantage of the proposed method is that all the data are stored in
vector form, thus saving an enormous amount of computer memory. The method is
successfully implemented on PIII with several realistic distribution networks. The
proposed algorithm can be used eectively with SCADA (supervisory control and
data acquisition) and DAC (distribution automation and control) as the algorithm
quickly solves the system and even suggests rerouting or network reconguration
for ecient operation of the system.
References
[1] W. F. Tinny and C. E. Hart, 1967, Power Flow Solution of the Newton Method,
IEEE Trans. PAS, Vol. PAS-86, No. 11.
[2] B. Stott and O. Alsac, 1974, Fast Decoupled Load Flow, IEEE Trans., Vol. PAS-93,
pp. 859869.
[3] B. Stott, 1984, Review of Load Flow Calculation Methods, Proc. IEEE, Vol. 62,
No. 7.
[4] D. Rajicic and Y. Tamura, 1988, A Modication to Fast Decoupled Power Flow for
Network with High R/X Ratios, IEEE Trans., Vol. PWRS-3, pp. 743746.
[5] S. C. Tripathy, D. Prasad, O. P. Malik, and G. S. Hope, 1982, Load Flow Solution for
Ill Conditioned Power Systems by Newton Like Method, IEEE Trans., Vol. PAS-101,
pp. 36843657.
[6] W. H. Kersting and D. L. Mendive, 1976, An Application of Ladder Network Theory
to the Solution of Three Phase Radial Load Flow Problem, IEEE PES Winter
Meeting.
[7] W. H. Kersting, 1984, A Method to Design and Operation of Distribution System,
IEEE Trans., Vol. PAS-103, pp. 19451952.
[8] R. A. Stevens, D. T. Rizy, and S. L. Puruker, 1986, Performance of Conventional
Power Flow Routines for Real Time Distribution Automation Applications, Proc.
of 18th Southeastern Symposium on Systems Theory, (IEEE), pp. 196200.
106
[9] M. E. Baran and F. F. Wu, 1989, Optimal Sizing of Capacitor Placed on Radial
Distribution Systems, IEEE Trans., Vol. PWRD-2, pp. 735743.
[10] H. D. Chiang, 1991, A Decoupled Load Flow Method for the Distribution Power
Network Algorithm: Analysis and Convergence Study, Electrical Power and Energy
Systems, Vol. 13, No. 3, pp. 130138.
[11] D. Das, H. S. Nagi, and D. P. Kothari, 1994, Novel Method for Solving Radial
Distribution Networks, IEE Proc. C, Vol. 141, No. 4, pp. 291298.
[12] D. Das, D. P. Kothari, and A. Kalam, 1995, Simple and Ecient Method for Load
Flow Solution of Radial Distribution Networks, Electrical Power & Energy Systems,
Vol. 17, No. 5, pp. 335346.
[13] S. Bhowimik, S. K. Goswami, and P. K. Bhattacherjee, 2000, A New Power Distribution System Planning through Reliability Evaluation Technique, Electrical Power
Systems Research, Vol. 24, pp. 169179.
[14] M. A. Kashem, V. Ganapathy, G. B. Jasmon, and M. I. Buhari, 2000, A Novel
Method for Loss Minimization in Distribution Networks, Proc. of International Conference on Electric Utility Deregulation and Restructuring and Power Technologies,
pp. 251255.
[15] T. Gonen, 1986, Electric Power Distribution Systems Engineering, McGraw-Hill.
Appendix
A.1. 69-Node Radial Distribution Networks
107
Table 5
Br.
no.
Send.
node
Rec.
node
Resis.
(ohm)
Reac.
(ohm)
Real load
(kW)
Reac. load
(kVAr)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
2
19
20
21
3
23
24
6
26
27
28
29
30
31
32
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
0.0922
0.4930
0.3660
0.3811
0.8190
0.1872
1.7114
1.0300
1.0400
0.1966
0.3744
1.4680
0.5416
0.5910
0.7463
1.2890
0.7320
0.1640
1.5042
0.4095
0.7089
0.4512
0.8980
0.8960
0.2030
0.2842
1.0590
0.8042
0.5075
0.9744
0.3105
0.3410
0.0477
0.2511
0.1864
0.1941
0.7070
0.6188
1.2351
0.7400
0.7400
0.0650
0.1238
1.1550
0.7129
0.5260
0.5450
1.7210
0.5740
0.1565
1.3554
0.4784
0.9373
0.3083
0.7091
0.7011
0.1034
0.1447
0.9337
0.7006
0.2585
0.9630
0.3619
0.5302
100.0
90.0
120.0
60.0
60.0
200.0
200.0
60.0
60.0
45.0
60.0
60.0
120.0
60.0
60.0
60.0
90.0
90.0
90.0
90.0
90.0
90.0
420.0
420.0
60.0
60.0
60.0
120.0
200.0
150.0
210.0
60.0
60.0
40.0
80.0
30.0
20.0
100.0
100.0
20.0
20.0
30.0
35.0
35.0
80.0
10.0
20.0
20.0
40.0
40.0
40.0
40.0
40.0
50.0
200.0
200.0
25.0
25.0
20.0
70.0
600.0
70.0
100.0
40.0