UNIVERSIDAD SIMN BOLVAR

DECANATO DE ESTUDIOS PROFESIONALES

COORDINACIN DE INGENIERA ELCTRICA

APLICACIN DE FLUJO DE CARGA DIRECTO A

REDES DE DISTRIBUCIN DE GRAN TAMAO
POR

HENRY ROMAN ESCOBAR MELGAREJO

JOS RAFAEL PREZ ZORRILLA

PROYECTO DE GRADO
PRESENTADO ANTE LA ILUSTRE UNIVERSIDAD SIMN BOLVAR
COMO REQUISITO PARCIAL PARA OPTAR AL TTULO DE
INGENIERO ELECTRICISTA

Sartenejas, Noviembre de 2010

UNIVERSIDAD SIMN BOLVAR

DECANATO DE ESTUDIOS PROFESIONALES
COORDINACIN DE INGENIERA ELCTRICA

APLICACIN DE FLUJO DE CARGA DIRECTO A

REDES DE DISTRIBUCIN DE GRAN TAMAO
POR
HENRY ROMAN ESCOBAR MELGAREJO
JOS RAFAEL PREZ ZORRILLA

TUTOR:PROF. PAULO DE OLIVEIRA

PROYECTO DE GRADO
PRESENTADO ANTE LA ILUSTRE UNIVERSIDAD SIMN BOLVAR
COMO REQUISITO PARCIAL PARA OPTAR AL TTULO DE
INGENIERO ELECTRICISTA

Sartenejas, Noviembre de 2010

UNIVERSIDAD
SIMN BOLVAR
Decanato de Estudios Profesionales
Coordinacin de Ingeniera Elctrica

ACT A DE EVALUACIN DEL PROYECTO DE GRADO

CDIGO DE LA ASIGNATURA:

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

JQ -C\u~o ~e. ce-...~Oc...

,sA\\~uc,;<')
~Q
'n,c....",
~.,-..f'\c..;\C>

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

APLICACIN DE FLUJO DE CARGA DIRECTO A REDES DE DISTRIBUCIN DE

GRAN TAMAO
Por:
HENRY ROMAN ESCOBAR MELGAREJO
JOS RAFAEL PREZ ZORRILA
RESUMEN

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

A mi madre; Gloria, por su infinito amor y sabidura.

a mi padre; Henry, por su esfuerzo.
Henry Escobar.
A mi madre; Olivia, por su apoyo y constancia.
Jos Prez.

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.

Todo es cuestin de actitud!!!

Henry Escobar.

Quisiera agradecer a mi madre por ser la persona que siempre ha estado conmigo.
Gracias!!!

Jos Rafael Prez.

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

i, j Nodos del sistema.

n Nmero de nodos del sistema.
Pi Potencia activa en el nodo i.
Qi Potencia reactiva en el nodo i.
Vi Tensin en el nodo i.
Vj Tensin en el nodo j.
Gii Conductancia propia en el nodo i.
Bii Susceptancia propia en el nodo i.
Yij Admitancia entre los nodos i-j.
ij ngulo de la admitancia de una lnea entre los nodos i-j.
i ngulo de la tensin en el nodo i.
j ngulo de la tensin en el nodo j.
s Nodo emisor.
r Nodo receptor.
Pr Potencia activa en el nodo r.
Qr Potencia reactiva en el nodo r.
Ps Potencia activa en el nodo s.
Qs Potencia reactiva en el nodo s.
Vs Tensin en el nodo s.
Vr Tensin en el nodo r.
Z Impedancia de la lnea entre los nodos s-r.
vii

R Resistencia de la lnea entre los nodos s-r.

X Reactancia de la lnea entre los nodos s-r.
Z ngulo de la impedancia de la lnea entre los nodos s-r.
s ngulo de la tensin en el nodo s.
r ngulo de la tensin en el nodo r.
Vs Tensin en el nodo s forma compleja.
Vr Tensin en el nodo r forma compleja.
Is Corriente en el nodo s forma compleja.
Z Impedancia de la lnea entre los nodos s-r en forma compleja.
BIBC Matriz de corrientes inyectadas a corrientes de rama (Bus Injection to Branch
Current), [19].
BCBV Matriz de corrientes de rama a voltajes nodales (Branch Current to Bus Voltage),
[19].
DLF Matriz de flujo de carga (Distribution Load Flow ), [19].
K Iteracin K-sima.
[I K ] Vector columna de corrientes inyectadas.
V K+1 Vector de variacin de tensiones nodales en la iteracin K + 1.
V K+1 Vector de tensiones nodales en la iteracin K + 1.
V0 Vector de tensiones nodales iniciales.
(p.u) Por Unidad.
[Y ] Matriz de Admitancias Nodales.
IiK Corriente inyectada en el nodo i durante la iteracin K
Pi Potencia activa demandada en el nodo i.
viii

Qi Potencia reactiva demandada en el nodo i.

ViK Tensin en el nodo i durante la iteracin K.
ViK+1 Tensin en el nodo i durante la iteracin K + 1.
 Tolerancia utilizada como criterio de convergencia.
S Vector de potencias inyectadas.
[Vx0 ] Vector de la componente real de las tensiones iniciales, [20].
[Vy0 ] Vector de la componente imaginaria de las tensiones iniciales, [20].
Vxi0 Componente real de la tensin inicial en la barra i, [20].
Vyi0 Componente imaginaria de la tensin inicial en la barra i, [20].
VxiK Componente real de la tensin en la barra i durante la iteracin K, [20].
VyiK Componente imaginaria de la tensin en la barra i durante la iteracin K, [20].
[T ] Matriz de corrientes inyectadas a corrientes de rama, [20].
[DR ] Matriz diagonal de resistencias de lneas, [20].
[DX ] Matriz diagonal de reactancias de lneas, [20].
[T RX] Matriz de flujo de carga, [20].
[I] Matriz de corrientes inyectadas, [20].
K
Ixi
Componente real de la corriente en la barra i durante la iteracin K, [20].
K
Iyi
Componente imaginaria de la tensin inicial en la barra i, [20].

VSlack Tensin en la barra de referencia o slack.

Vi Cada de tensin en el nodo i respecto a la barra slack ( %).
Lmp Prdidas activas totales.
Lmq Prdidas reactivas totales.
ix

Gij Conductancia entre los nodos i-j.

Bij Susceptancia entre los nodos i-j.
Rij Resistencia entre los nodos i-j.
Xij Reactancia entre los nodos i-j.
LS/E Prdidas de potencia en una subestacin
LT Prdidas totales en la red.
Pij Flujo de potencia activa entre los nodos i-j.
Qij Flujo de potencia reactiva entre los nodos i-j.
Sij Flujo de potencia aparente entre los nodos i-j.
Sijmax Potencia nominal mxima permitida por un conductor.
Factor de seguridad.
CF Capacidad Firme.
C Factor de escalamiento.
N Nmero de transformadores operativos.
kV Am Potencia nominal del transformador.
kV Amax Potencia del transformador de mayor capacidad.
VBASE Base de voltaje.
SBASE Base de potencia.
kV Kilovoltio.
M V A Mega voltio-Amper.
P gen0 Valor inicial de la potencia activa generada (p.u).
Qgen0 Valor inicial de la potencia reactiva generada (p.u).
x

Pload Potencia activa demandada (p.u).

Qload Potencia reactiva demandada (p.u).
f p Factor de potencia.
Ir Componente real de la corriente en el nodo i (p.u).
Ij Componente imaginaria de la corriente en el nodo i (p.u).
Vr Componente real de la tensin en el nodo i (p.u).
Vj Componente imaginaria de la tensin en el nodo i (p.u).
.Vr Diferencia de la componente real de tensin entre iteraciones.
.Vj Diferencia de la componente imaginaria de tensin entre iteraciones.

xi

LISTA DE ABREVIATURAS

FDC Flujo de Carga.

MVC Patrn de Diseo Modelo-Vista-Controlador.
MATLAB Herramienta Computacional de Clculo.
NR Newton-Raphson.
SDEE Sistema Elctrico de Distribucin.
GS Gauss-Seidel.
FDCB Flujo de Carga de Barrido.
INDENE Instituto de Energa de la Universidad Simn Bolvar.
SEDDGE Sistemas Elctricos de Distribucin a Gran Escala.
FDCT Flujo de Carga Directo desarrollado por Teng, [19].
TRX Flujo de Carga Directo desarrollado por De Oliveira, [20].
SEDDGT Sistema Elctrico de Distribucin de Gran Tamao.
EDC Electricidad de Caracas.
SEP Sistema Elctrico de Potencia.
SEDT Sistema Elctrico de Transmisin.
NRD Newton-Raphson Desacoplado.
slack Barra de Referencia del Sistema.
backward sweep Barrido Hacia Atrs en el FDCB.
forward sweep Barrido Hacia Adelante en el FDCB.
KVL Ley de Kirchhoff de Voltajes.
KCL Ley de Kirchhoff de Corrientes.
xii

BIBC Matriz de Corrientes Inyectadas a Corrientes de Rama o Bus Injection to Branch

Current.
BCBV Matriz de Corrientes de Rama a Voltajes Nodales o Branch Current to Bus
Voltage.
DLF Matriz de Flujo de Carga o Distribution Load Flow.
backward and forward sweep Proceso de Barrido Hacia Adelante y Hacia Atrs.
MATPOWER Paquete de Simulacin de Sistemas de Potencia de MATLAB.
ASP Sistema de Anlisis y Simulacin de Redes Primarias.
FDCO Flujo de Carga ptimo.
NRDR Newton-Raphson Desacoplado Rpido.
LP Programacin Lineal.
CE Capacidad de Emergencia .
CN Capacidad Nominal.
S/E Subestacin.
CF Capacidad Firme.
SAP Software de Gestin y Estrategia.
GIS Software de Estimacin de Demanda.
SCADA Software de registro de carga y data del sistema.
ORM Mapeadores Objeto-Relacionales.
.DAT Formato Original de la Base Datos.
.MAT Formato de los Archivos Generados con el Algoritmo Implementado.
RAM Memoria de Acceso Aleatorio (Random Access Memory).
GW Gigavatio.

xiii

ndice general

INTRODUCCIN

1. ANTECEDENTES

1.1. Caractersticas del Sistema de Distribucin . . . . . . . . . . . . . . . . . . .

1.2. El Flujo de Carga de Distribucin . . . . . . . . . . . . . . . . . . . . . . . .

1.2.1. FDC para Topologas Malladas . . . . . . . . . . . . . . . . . . . . .

1.2.2. FDC para Topologas Radiales . . . . . . . . . . . . . . . . . . . . .

1.2.3. Flujos de Carga de Barrido (FDCB ) o forward and backward sweep

10

2. METODOLOGA

14

2.1. Algoritmos Implementados . . . . . . . . . . . . . . . . . . . . . . . . . . . .

14

2.1.1. Flujo de Carga Directo Complejo, (FDCDC ), [19] . . . . . . . . . . .

15

2.1.2. Flujo de Carga Directo T RX . . . . . . . . . . . . . . . . . . . . . .

18

2.2. Herramientas Computacionales . . . . . . . . . . . . . . . . . . . . . . . . .

22

2.2.1. MATPOWER, [52] . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22

2.2.2. ASP, [54] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

23

2.3. Condiciones de Operacin de la Red . . . . . . . . . . . . . . . . . . . . . . .

24

xiv

2.3.1. Anlisis por Cada de Tensin . . . . . . . . . . . . . . . . . . . . . .

25

2.3.2. Anlisis por Prdidas de Potencia . . . . . . . . . . . . . . . . . . . .

25

2.3.3. Anlisis por Capacidad Amperimtrica de Conductores . . . . . . . .

26

2.3.4. Anlisis de Capacidad Firme de una S/E . . . . . . . . . . . . . . . .

28

3. IMPLEMENTACIN

29

3.1. Arquitectura del Sistema . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

29

3.2. Descripcin del Algoritmo Implementado . . . . . . . . . . . . . . . . . . . .

34

3.3. Datos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

34

3.4. Estructura del Programa . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

37

3.4.1. Adquisicin de Datos . . . . . . . . . . . . . . . . . . . . . . . . . . .

37

3.4.2. Proceso Iterativo . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

37

4. CASOS DE ESTUDIOS Y ANLISIS DE RESULTADOS

39

4.1. Aplicacin en Circuito de 7 Barras . . . . . . . . . . . . . . . . . . . . . . .

39

4.1.1. Clculo de la Matriz TRX . . . . . . . . . . . . . . . . . . . . . . . .

40

4.1.2. Proceso Iterativo . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

43

4.2. Validacin de Resultados . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

47

4.3. Estudio Comparativo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

48

4.4. Aplicacin en Redes de Gran Tamao . . . . . . . . . . . . . . . . . . . . . .

51

4.4.1. Tiempo de cmputo . . . . . . . . . . . . . . . . . . . . . . . . . . .

51

4.4.2. Anlisis de Condiciones Operacionales de una Red de Gran Tamao .

54

xv

CONCLUSIONES

63

REFERENCIAS BIBLIOGRFICAS

65

A. CASO EXPLICATIVO: MATRICES BIBC Y BCBV

71

A.1. Matriz BIBC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

71

A.2. Matriz BCBV

72

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

B. ALGORITMO IMPLEMENTADO TRX

74

B.1. Algoritmo de Adquisicin de Datos . . . . . . . . . . . . . . . . . . . . . . .

74

B.2. Flujo de Carga TRX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

83

C. EJEMPLO DE REPORTE DEL ALGORITMO DESARROLLADO

87

D. CASO DE ESTUDIO: CIRCUITO DE 4 BARRAS

98

E. REPORTE DEL MATPOWER PARA EL CASO DE 7 BARRAS

100

F. A COMPENSATION-BASED POWER FLOW METHOD FOR WEAKLY MESHED DISTRIBUTION AND TRANSMISSION NETWORKS,
SHIRMOHAMMADI ET AL. 1988, [11]

103

G. A DIRECT APPROACH FOR DISTRIBUTION SYSTEM LOAD FLOW

SOLUTION, TENG. 2003, [19]

114

H. THE DISTRIBUTION TRX-POWER FLOW METHOD, DE OLIVEIRA.

2010, [20]

121

I. OPTIMAL SIZING OF CAPACITORS PLACED ON RADIAL DISTRIBUTION

SYSTEM, BARAN, M. E. ET AL. 1989, [31]
xvi

139

J. SIMPLE AND EFFICIENT COMPUTER ALGORITHM TO SOLVE RADIAL DISTRIBUTION NETWORKS, RANJAN, R. ET AL. 2003, [27] 150

xvii

ndice de tablas

2.1. Cada de Tensin Mxima Permitida, [56]

. . . . . . . . . . . . . . . . . . .

25

4.1. Datos de Lnea (Ldat) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

40

4.2. Datos de Nodos, (Bdat) . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

41

4.3. Valores de potencia y tensin inicial (V0 , P0 ) . . . . . . . . . . . . . . . . . .

44

4.4. Resultados de la Primera Iteracin . . . . . . . . . . . . . . . . . . . . . . .

45

4.5. Resultados de la Segunda Iteracin . . . . . . . . . . . . . . . . . . . . . . .

45

4.6. Resultados de la Tercera Iteracin . . . . . . . . . . . . . . . . . . . . . . . .

46

4.7. Tensiones nodales en mdulo y ngulo . . . . . . . . . . . . . . . . . . . . .

46

4.8. Tensiones Nodales Validadas en Mdulo y ngulo . . . . . . . . . . . . . . .

47

4.9. Tiempo de cmputo de [19] y [20] para circuitos de 4, 7, 12 y 69 barras . . .

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

4.13. Condiciones de Operacin de la Red, 530 circuitos. . . . . . . . . . . . . . .

62

D.1. Datos de lnea (Ldat) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

98

D.2. Datos de nodos (Bdat) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

99

xviii

ndice de figuras

1.1. Red de Distribucin. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

10

2.1. Diagrama de Flujo del Algoritmo Propuesto por Teng, [19]. . . . . . . . . . .

19

2.2. Diagrama de Flujo del Algoritmo Propuesto por De Oliviera, [20]. . . . . . .

22

2.3. Porcentaje de Carga de un Cable Subterrneo, [56]. . . . . . . . . . . . . . .

27

3.1. Base de Datos Unificada, [49]. . . . . . . . . . . . . . . . . . . . . . . . . . .

30

3.2. Sincronizacin de Base de Datos, [49]. . . . . . . . . . . . . . . . . . . . . . .

30

3.3. Modelo MVC, [50]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

32

3.4. Esquema General de Herramienta de Planificacin Corto-Mediano Plazo (HPCMP)

- Visualizacin de Largo Plazo, [50]. . . . . . . . . . . . . . . . . . . . . . . .

33

3.5. Esquema MVC del programa implementado. . . . . . . . . . . . . . . . . . .

35

3.6. Diagrama de Flujo del Algoritmo Desarrollado. . . . . . . . . . . . . . . . .

36

4.1. Estudio Comparativo: Tiempo de Cmputo, [20]. . . . . . . . . . . . . . . .

49

4.2. Tiempo de Cmputo del ASP para la Red de la Gran Caracas. . . . . . . . .

52

4.3. Tiempo de Cmputo del Algoritmo Implementado para la Red de la Gran

Caracas. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

53

4.4. Perfil de Cada de Tensin en ( %) del circuito AN T _A01. . . . . . . . . . .

55

xix

4.5. Potencia Activa Demandada y Prdidas Tcnicas del Circuito AN T _A01. .

56

4.6. Potencia Reactiva Demandada y Prdidas Tcnicas del Circuito AN T _A01.

56

4.7. Capacidad Amperimtrica de los conductores del circuito AN T _A01. . . . .

57

4.8. Potencia Entregada y Perdida por Circuito de la S/E AN T . . . . . . . . . .

58

4.9. Capacidad Amperimtrica de los conductores de la S/E AN T

. . . . . . . .

59

4.10. Capacidad Firme de las S/E AN T y SRO. . . . . . . . . . . . . . . . . . . .

60

4.11. Potencia Activa Demandada y Prdidas Tcnicas de la red. . . . . . . . . . .

61

4.12. Capacidad Amperimtrica de los conductores de la red. . . . . . . . . . . . .

62

A.1. Sistema de Distribucin de Ejemplo. . . . . . . . . . . . . . . . . . . . . . .

71

xx

INTRODUCCIN

El Flujo de Carga, FDC, es un mtodo de anlisis numrico para la determinacin del

estado de los Sistemas Elctricos de Potencia, SEP. El FDC, como herramienta de anlisis,
cobra mayor importancia en el sistema de gestin elctrica actual. Esta herramienta no slo
permite conocer el estado de la red en un momento determinado sino que tambin permite
realizar aplicaciones como despacho de generacin distribuida, equilibrio de fases, control de
tensiones, colocacin ptima de capacitores y estudios de planificacin de corto y mediano
plazo, [20].

El FDC es un estudio que requiere la resolucin de sistemas de ecuaciones no lineales. Por

lo tanto, los clculos implicados se caracterizan por presentar un alto grado de complejidad y
resulta necesario el uso de procesos iterativos para encontrar la solucin con la mayor exactitud posible. An ms, el nmero de variables a manejar aumenta en funcin del tamao del
sistema elctrico en estudio y con esto aumenta la complejidad del clculo a realizar. Adems,
es bien conocido que los Sistemas Elctricos de Distribucin, SEDD, actuales se caracterizan
por ser de gran tamao; lo cual implica que el nmero de variables a manejar en el FDC
tambin lo es. Debido a esto, se hace imprescindible el uso de herramientas computacionales
para poder estudiar dichos sistemas de forma eficiente.

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.

El presente trabajo se realiz partiendo de una lnea de investigacin desarrollada por el

Instituto de Energa de la Universidad Simn Bolvar (INDENE ) para el desarrollo de una
herramienta computacional para el anlisis de Sistemas Elctricos de Distribucin a Gran Escala, SEDDGE. El proyecto general surge por la necesidad de desarrollar nuevos sistemas de
anlisis con una estructura abierta (cdigo abierto) a fines de integrar efectivamente los procesos atendiendo los siguientes criterios: cdigo modular, programacin orientada a objeto,

3
aplicacin web multiplataforma, control de versiones mediante un esquema colaborativo, [50].

Este trabajo aborda la resolucin de Sistema Elctrico de Distribucin de Gran Tamao,

SEDDGT, mediante la aplicacin de la metodologa de FDC unidireccional denominada
Flujo de Carga Directo. Se denomina directo por cuanto se obtiene el estado de la red
iterativamente a partir de una matriz nica que contiene informacin del sistema: dimensionamiento y topologa. El algoritmo se basa en Teng, [19], el cual ha sido modificado por
De oliveira [20] con el objetivo de realizar operaciones con nmeros reales y, de esta forma,
mejorar la eficiencia del algoritmo una vez implementado. El algoritmo se caracteriza por una
lgica de clculo sencilla (suma y multiplicacin de matrices), con lo cual se evitan operaciones que requieran invertir matrices como en el caso de la matriz Jacobiana del NR. An
ms, las matrices requeridas para realizar el FDC estarn disponibles en memoria RAM cada
vez que se requiera. Con esto, se optimiza el algoritmo y es posible obtener resultados para
SEDDGT en el orden de los segundos. Razones que hacen del FDC un algoritmo poderoso
y muy til en el rea de planificacin de SDEE.

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:

Sistema Elctrico de Transmisin, SEDT. Conformado por el rea de generacin

y la red de transmisin; los cuales operan a grandes niveles de tensin. Se encarga de
generar y transmitir grandes bloques de potencia.
Sistema Elctrico de Distribucin, SDEE. Conformado por las subestaciones y
la red de distribucin; los cuales operan a niveles intermedios y bajos de tensin. La
funcin de este sistema es distribuir los grandes bloques de potencia a los consumidores
finales.

1.1.

Caractersticas del Sistema de Distribucin

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 Flujo de Carga de Distribucin

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

|Vi ||Vj ||Yij | cos(ij + j i )

(1.1)

j=1

Qi = |Vi | Bii

n
X

|Vi ||Vj ||Yij | sin(ij + j i )

(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.

Para realizar ciertos anlisis no es estrictamente necesario considerar el desbalance de la

red. Cuando este es el caso, es posible asumir la red balanceada y realizar la modelacin de la
red mediante un equivalente unifilar [11]. Para efecto del anlisis realizado, las modelaciones
fueron realizadas considerando redes balanceadas.

8
Tipos de Nodos Dependiendo de las condiciones de contorno que se especifiquen, se
pueden clasificar los nodos del sistema en tres grupos; [37]:

Nodo de Referencia o Slack. Es una barra de generacin, en la que se asume

conocido el mdulo y ngulo de la tensin.
Nodos PQ. Son aquellos nodos en los que se especifican las potencias activa y reactiva
netas inyectadas. Los nodos PQ son los ms abundantes en los sistemas de distribucin.
Nodos PV. Son nodos en los que se especifican la potencia activa y el mdulo de la
tensin. Este tipo de nodos son poco comunes en sistemas de distribucin.

1.2.1.

FDC para Topologas Malladas

Los primeros algoritmos de FDC, como el Newton-Raphson (NR) o el Gauss-Seidel (GS ),

fueron desarrollados para la operacin, control y planificacin de SEDT, [2, 37, 1]. La principal diferencia entre estos ltimos respecto a los FDC para topologas radiales, es que los
mismos pueden ser aplicados independientemente de la configuracin de la red [3].

El NR es un mtodo basado en un slido fundamento matemtico como lo son las series

de Taylor [3]. Este mtodo utiliza un proceso iterativo mediante el cual se aproxima a la
solucin linealizando las ecuaciones de potencia (ver ecuaciones 1.1 y 1.2). En [3, 37, 38] es
posible encontrar la descripcin del proceso algortmico de este FDC. Adems, este mtodo
requiere de un gran esfuerzo computacional y un elevado tiempo de cmputo.

La aplicacin del mtodo a un sistema no lineal de ecuaciones de orden (n) va a implicar

en cada iteracin, la formacin del Jacobiano de orden (nxn) y su subsiguiente inversin, lo
cual toma un tiempo no despreciable [38, pg. 129].

El mtodo GS presenta la ventaja de no utilizar matrices Jacobianas; lo cual lo convierte

en un algoritmo mucho ms sencillo y requiere un menor esfuerzo computacional que el NR.

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.

Sin embargo, el GS requiere de un gran nmero de iteraciones para converger. Por lo

general, realiza un nmero de iteraciones aproximadamente igual al nmero de nodos del
sistema estudiado [38].

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.

FDC para Topologas Radiales

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.

Flujos de Carga de Barrido (FDCB) o forward and backward

sweep

Clculo de Voltaje en Redes de Distribucin

Como se describe en [2], se va a considerar el circuito de distribucin mostrado en la figura
1.1. La potencia activa y reactiva en la barra receptora, puede escribirse como se indica en
las ecuaciones 1.3 y 1.4

Figura 1.1: Red de Distribucin.

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)

Usando la identidad trigonomtrica, (ecuacin 1.5):

cos2 (Z s + r ) + sin2 (Z s + r ) = 1

(1.5)

Despejando los trminos cos(Z s + r ) y sin(Z s + r ) de 1.3 y 1.4, respectivamente

y sustituyndolos en 1.5 se obtiene la forma general de la ecuacin bi-cuadrtica, dada en
1.6. Con la mxima raz real de la ecuacin 1.6 se obtiene la magnitud del voltaje en el nodo
receptor. El voltaje en el nodo receptor se puede escribir en funcin de la impedancia de la
lnea, de la potencia y voltaje del nodo emisor, tal como se muestra en 1.7.

Vr4 + 2Vr (Pr R + Qr X) Vs2 Vr2 + (Pr2 + Q2r )Z 2 = 0

(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)

Descripcin de FDCB basados en las leyes de Kirchhoff, [2]

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.

Un algoritmo para sistemas de distribucin desbalanceadas es dado por [18]. En este

algoritmo, se utiliza la topologa de la red para resolver el problema. Propone la creacin de
dos matrices: una matriz que relaciona las corrientes inyectadas con las corrientes de rama,
BIBC (Bus Injection to Branch Current), y una matriz que relaciona las corrientes de rama
con los voltajes de nodo, BCBV (Branch Current to Bus Voltage). Luego, se obtiene la matriz
de flujo de carga de distribucin, DLF (Distribution Load Flow ) al multiplicar BCBV.BIBC.
Posteriormente, se calculan las cadas de tensin en las lneas multiplicando DLF y la matriz
de corrientes inyectadas, I K , 1.10. Las tensiones nodales se obtienen restando la tensin en
la fuente menos la cada de tensin en las lneas 1.11.

.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].

Un algoritmo basado en la KVL para sistemas de distribucin monofsicos es propuesto

por [21]. El problema es resuelto considerando las cargas como impedancias constantes durante el backward sweep. El voltaje en cada nodo es calculado usando la ecuacin 1.8 durante
el backward sweep. Luego, se usa la relacin entre la tensin impuesta en el nodo slack y su
nuevo valor calculado para obtener el valor actual del voltaje en las barras y de las corrientes
inyectadas y de rama.

CAPTULO 2
METODOLOGA

En este captulo se describen en detalle los algoritmos implementados: en primer lugar, el

flujo de carga, FDC, desarrollado por Teng, denominado (FDCT ), [19], y en segundo lugar,
el FDC desarrollado por De Oliveira, denominado (TRX ), [20]. Adems se describen dos
herramientas computacionales utilizadas para comparar y validar el algoritmo implementado;
es decir, el (TRX ). Luego, se muestran las ecuaciones requeridas para realizar el anlisis del
sistema posterior a la solucin del flujo de carga.

2.1.

Algoritmos Implementados

Partiendo de una lnea de investigacin desarrollada por el profesor Paulo de Oliveira se

trabaj con el TRX. El mismo utiliza la misma lgica de clculo que el FDCT pero incluye
una modificacin para trabajar en nmeros reales y no en complejos como lo hace FDCT.
Razn por la cual, es vlido afirmar que el TRX es ms eficiente que el FDCT. Dicha afirmacin se comprueba en la seccin 4.3 del captulo 4

En la implementacin de dichos algoritmos se asumieron las siguientes premisas:

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.

Flujo de Carga Directo Complejo, (FDCDC ), [19]

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.

En el mismo, se utilizan dos matrices -la matriz de inyecciones de corriente a corriente de

rama (Bus Injection to Branch Current), BIBC, y la matriz de corrientes de rama a voltajes
nodales (Branch Current to Bus Voltage, BCBV - y una simple multiplicacin de matrices
son usadas para obtener la solucin del FDC.

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)

Luego, se construyen las matrices BIBC y BCBV. En el apndice A se ilustra el proceso de

armado de ambas matrices. BIBC est relacionada con la topologa del circuito y BCBV est
relacionada con las impedancias de las lneas. Posteriormente, se multiplican ambas matrices
como lo muestra la ecuacin 2.2 para obtener la matriz de impedancias de lnea asociadas a
la topologa del sistema. La matriz [DLF ] es de orden (n 1)x(n 1), donde n es el nmero
de nodos del sistema.

[DLF ] = [BCBV ][BIBC]

(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.

[.V K+1 ] = [DLF ][I K ]

(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

[V K+1 ] = [V 0 ] + [.V K+1 ]

(2.4)

El proceso se detiene cuando la diferencia de voltaje entre iteraciones es menor a cierta

tolerancia, ecuacin 2.5. Tpicamente,  se asume igual a 0, 001 p.u.

||ViK | |ViK1 || 

(2.5)

Procesos Algortmicos

Matriz BIBC. En el apndice A se muestra un ejemplo ilustrativo para la construccin

de BIBC.

Paso 1. Para un sistema de m ramas y n nodos, la dimensin de la matriz BIBC es

m (n 1)
Paso 2. Si una lnea (Bl ) est entre los nodos i y j. Se copia la columna de la barra i,
se pega en la columna de la barra j y se coloca +1 en la fila l y la columna de la barra
j.
Paso 3. Repetir el paso 2 hasta que todas las lneas estn en la matriz BIBC.

Matriz BCBV. En el apndice A se muestra un ejemplo ilustrativo para la construccin

de BCBV.

Paso 1. Para un sistema de m ramas y n nodos, la dimensin de la matriz BCBV es

(n 1) m
Paso 2. Si una lnea (Bl ) est entre los nodos i y j. Se copia la fila de la barra i, se
pega en la fila de la barra j y se coloca la impedancia de la lnea Zij en la fila de la
barra j y la columna l.
Paso 3. Repetir el paso 2 hasta que todas las lneas estn en la matriz BCBV.

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.

En la figura 2.1 se encuentra el diagrama del FDCT, [19].

2.1.2.

Flujo de Carga Directo T RX

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

Para este algoritmo se requiere la numeracin nodal desarrollada por Shirmohammadi

(explicada en la seccin 2.1.1, ver apndice F). El FDC utiliza el vector de potencias inyectadas (S ), la topologa del circuito y la impedancia de las lneas (Z = R + jX). En este FDC
se trabaja con la parte real y la parte imaginaria de los datos por separado. De esta forma,
se parte de un vector de voltajes iniciales ([V0 ]) separado en parte real e imaginaria, ecuacin
(2.6).

20

0
0
]
...Vxi0 ...Vxn
Vx0 = [Vx1

(2.6)

0
0
...Vyi0 ...Vyn
]
Vy0 = [Vy1

Luego, se calcula la matriz [T ] siguiendo el mismo procedimiento utilizado para construir

la matriz BIBC en el FDCT, ver apndice A. Adems, utilizan dos matrices diagonales: en
la primera se definen las resistencias, [DR ], y en la segunda las reactancias de las lneas del
circuito, [DX ], ecuaciones 2.7 y 2.8, respectivamente.

DR

DR1

0
=

...

DR2 . . .
..
.
0

0
..
.

DRn

0
...
DX1

DX2 . . .
0
DX =

..
0
.
0

0
0
..
.

DXn

(2.7)

(2.8)

Posteriormente, se multiplican las matrices T , DR y Dx , como lo muestra la ecuacin 2.9,

para obtener la matriz de impedancias de lnea asociadas a la topologa del sistema, T RX.
La matriz [T RX] es de orden 2(n 1)x2(n 1), donde n es el nmero de nodos del sistema.

T DR T

TRX =

T T DX T

T DX T
T T DR T

(2.9)

Luego, se calcula la matriz de corrientes inyectadas [I], ecuacin 2.10.

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

es la componente real e imaginaria de la corriente en el nodo i durante la iteracin K. Pi y

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 } =

Finalmente, se calculan las tensiones en los nodos, ecuacin (2.13).

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)

Para los casos estudiados, se asumi Vxi0 = 1 y Vyi0 = 0 para i = 1, ..., n.

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

A continuacin se describirn las herramientas computacionales utilizadas para comparar

y validar el algoritmo implementado.

2.2.1.

MATPOWER, [52]

MATPOWER es un paquete de Matlab para resolver problemas de FDC y flujo de carga

ptimo, FDCO. MATPOWER est diseado para dar el mejor desempeo posible mante-

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]

El Sistema de Anlisis y Simulacin de Redes Primarias, ASP, est orientado al ingeniero

de distribucin especializado en planificacin y proyectos de redes primarias. Las capacidades
de esta aplicacin son, [55]:

Analizar y editar circuitos.

Simulacin de crecimiento de redes.

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.

Con el ASP se pueden obtener los siguientes resultados, [54]:

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.

Condiciones de Operacin de la Red

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.

Anlisis por Cada de Tensin

Criterio de Cada de Tensin Mxima, [56]

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( %)

Banda Permitida (p.u)

Normal

0, 95 < V < 1, 05

Emergencia

0, 92 < V < 1, 08

La ecuacin 2.15 muestra, en trminos porcentuales, la cada de tensin en cada nodo

respecto a la tensin de la barra Slack, [56].

.Vi =

2.3.2.

VSlack Vi
100
VSlack

(2.15)

Anlisis por Prdidas de Potencia

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.

Anlisis por Capacidad Amperimtrica de Conductores

Criterio de Capacidad de Carga, [56]

Todo circuito debe tener un porcentaje mximo de carga igual al 67 % de su capacidad

de emergencia, ya que debe cumplirse que cada circuito primario pueda ser asistido por
dos o ms circuitos. En la figura 2.3 se puede observar el porcentaje de carga de un cable

27

Figura 2.3: Porcentaje de Carga de un Cable Subterrneo, [56].

subterrneo en condiciones normales y de emergencia, as como tambin la reserva que puede
ser utilizada a mediano o a largo plazo.
Para una lnea conectada entre los nodos i y j, el flujo de potencia que va de la barra i a
la j, puede ser obtenida de la siguiente forma, [48].

Pij = Gij [Vi Vj cos ij Vi2 ] + Bij Vi Vj sin ij

(2.23)

Qij = Bij [Vi2 Vi Vj cos ij ] + Gij Vi Vj sin ij

(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.

Anlisis de Capacidad Firme de una S/E

Criterio de Capacidad de Firme, [54]

La capacidad firme (CF ) es la que se debe manejar cuando se disea una S/E, para que
ante una posible salida forzada de algn transformador se pueda seguir supliendo la carga
demandada de una forma segura y continua, sin necesidad de realizar interconexiones con
otros circuitos para suplir la demanda.

La demanda actual y la proyectada para el futuro no deben exceder la CF de la S/E.

Adems, se debe tratar que todos los transformadores de la misma S/E operen a la misma
capacidad nominal con el fin de tener la mayor CF, lo cual proporciona una mayor CF ante
posibles contingencias. La Electricidad de Caracas, EDC, opera con un mximo de 4 transformadores por S/E.

Utilizando la ecuacin 2.27 es posible calcular la CF de una S/E.

CF = C.(

N
X

kV Am kV Amax )

(2.27)

m=1

Donde, N es el nmero de transformadores operando en la S/E. kV Am es la potencia

nominal del transformador. kV Amax es la potencia del transformador de mayor capacidad.
C es un factor que puede ser:
120 por ciento para transformadores de distribucin cuyo tiempo til sea menor o igual
a 40 aos.
100 por ciento para transformadores de distribucin cuyo tiempo til sea mayor o igual
a 40 aos.

CAPTULO 3
IMPLEMENTACIN

3.1.

Arquitectura del Sistema

El estado actual del sistema de adquisicin y procesamiento de datos utilizado en la

Electricidad de Caracas, EDC, presenta las siguientes caractersticas, [49]:

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

Figura 3.1: Base de Datos Unificada, [49].

Al sincronizar la base de datos es posible accesar a la misma desde cualquier computador
que haya sido sincronizado a dicha red. Esto elimina la necesidad de un computador central
que contenga toda la informacin de la base de datos, ver figura (3.2).

Figura 3.2: Sincronizacin de Base de Datos, [49].

El Instituto de Energa de la Universidad Simn Bolvar, (INDENE ), en conjunto con
otras organizaciones y empresas, ha venido desarrollando un proyecto con el cual se busca
automatizar el proceso de estudios de la red de distribucin y desarrollar una plataforma
basada en cdigo abierto que permita, [49]:

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.

En, [50], se ha visualizado la elaboracin del proyecto en tres etapas:

Etapa 1. Implementacin del prototipo.

Etapa 2. Implementacin de Funciones Bsicas.
Etapa 3. Implementacin de Funciones Avanzadas.

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.

Para la aplicacin del proyecto se ha considerado que una arquitectura adecuada es el

patrn de diseo Modelo-Vista-Controlador (MVC ), mostrado en la figura 3.3. Este patrn
ha tomado especial relevancia a raz de las nuevas implementaciones que se han logrado hacer
del mismo y la explosin en cuanto a programacin orientada a objetos.

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

Figura 3.3: Modelo MVC, [50].

datos en algo ms acorde con la programacin orientada a objetos y liberando la definicin
de los datos de la implementacin (gestor de bases de datos seleccionado). Por ltimo, entre
las capas arriba mencionadas se encuentra la capa de los Controladores, stos se encargan
de las funciones inherentes a la lgica del programa, negociar las solicitudes de acciones de
las vistas y gestionar los datos en los modelos. En la figura 3.4 se puede apreciar el esquema
general del sistema que se est diseando, [49].

De acuerdo a [50], esta segmentacin de funciones permite mejorar el rendimiento a la

hora de hacer mantenimiento a la aplicacin o el control de cambios solicitados. Por ejemplo,
si se requieren cambios en la interfaz de usuario solamente es necesario realizar cambios en
el diseo de la capa correspondiente a la vista. De igual forma, si existe una reorganizacin
de la estructura de los datos o una implementacin en un nuevo gestor de bases de datos
se ver afectada solamente la capa del modelo. Del mismo modo, una nueva opcin dentro
del programa implicar la incorporacin de un controlador adecuado que gestione esta nueva
funcionalidad.

33

Figura 3.4: Esquema General de Herramienta de Planificacin Corto-Mediano Plazo

(HPCMP) - Visualizacin de Largo Plazo, [50].

34

3.2.

Descripcin del Algoritmo Implementado

El algoritmo implementado fue desarrollado bajo la arquitectura MVC, descrita en la

seccin 3.1 del presente captulo. El programa est desarrollado de forma modular para optimizar el tiempo de cmputo del mismo. ste est compuesto por dos mdulos: en el primero
se realiza el proceso de adquisicin de datos; en el segundo, se realiza el proceso iterativo y,
con las tensiones nodales, se realiza el anlisis de las condiciones de operacin de la red. Con
el primer mdulo se filtra la informacin de los archivos .DAT y se obtienen las matrices
TRX, S y los datos de nodos y lneas para clculos futuros; las cuales se guardan en archivos
.MAT y pueden ser almacenadas en disco o en memoria RAM. En el segundo mdulo se
realiza el proceso iterativo, se obtienen los voltajes nodales y se analizan los parmetros
correspondientes al estudio de prdidas tcnicas en conductores, cadas de tensin y capacidad amperimtrica de los conductores en condiciones normales de operacin. Es importante
destacar que MATLAB es utilizado como programa interpretador del algoritmo implementado.

En la figura 3.5 se muestra el esquema basado en la arquitectura MVC del programa

desarrollado en el presente proyecto. En el mdulo M se realiza el proceso de la adquisicin
de datos. En el mdulo C se lleva a cabo el proceso iterativo. Finalmente, en el mdulo V
se realiza el anlisis posterior de las condiciones de operacin de la red.

En la figura 3.6 se observa el diagrama de flujo del programa implementado.

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

Figura 3.5: Esquema MVC del programa implementado.

el SCADA y el GIS, y en los mismos se almacena informacin referente a dicho programa
para uso interno del mismo. Los datos estn separados por comas (,) y la informacin de la
red est agrupada en bloques; cada bloque termina con la palabra END [51]. La estructura de los archivos DAT, su explicacin exhaustiva y un ejemplo se pueden encontrar en [51].

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

Figura 3.6: Diagrama de Flujo del Algoritmo Desarrollado.

37

3.4.
3.4.1.

Estructura del Programa

Adquisicin de Datos

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.

Con este programa se obtiene la matriz de impedancias asociadas a la topologa de la red

(TRX, [20]) y la matriz de potencia aparente (S ). Las matrices pueden ser almacenadas en
el disco o en memoria RAM. Para optimizar el proceso de flujo de carga se busca que las
matrices TRX y S estn disponibles en la memoria RAM del sistema y se actualizan offline
con cierta frecuencia. Es importante el proceso de actualizacin de ambas matrices porque
en la primera, (TRX ), se registran los cambios en la red y en la segunda, (S ), se registran
los cambios en la potencia demandada. La segunda matriz requiere un proceso de actualizacin ms frecuente que la primera ya que la potencia demandada vara con ms frecuencia
que la conexin de la red. De esa forma se obtienen resultados ms objetivos y reales 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

El algoritmo implementado fue probado en diferentes casos de estudio. En primer lugar,

se hizo un estudio en un circuito de 7 barras. Luego, se realiz la validacin de los resultados
del caso anterior con un programa de uso comercial. Despus, se realiz un estudio comparativo entre los algoritmos propuestos por Teng (FDCT, [19]) y por De Oliveira (TRX, [20])
para comprobar la eficiencia del ltimo respecto al primero. Finalmente, se implement el
algoritmo en circuitos de gran tamao de la Electricidad de Caracas, EDC. Con los resultados
obtenidos del flujo de carga (FDC ) se realiz un anlisis de las condiciones de operacin de
la red estudiada.

4.1.

Aplicacin en Circuito de 7 Barras

Para ilustrar el proceso de clculo del algoritmo se utilizar un circuito de 7 barras. El

proceso de clculo se realizar paso por paso para demostrar la lgica del mismo, dicho proceso se describe en la seccin 2.1.2 que se encuentra en el captulo 2. El sistema incluye 1
generador y 6 cargas equivalentes, conectadas a una red de distribucin radial. Los datos de
lneas y de nodos del circuito se muestran en las tablas 4.1 y 4.2, respectivamente.

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.

Clculo de la Matriz TRX

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

entre el nodo correspondiente a la fila y la columna respectiva.

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

Matriz de Resistencias y Reactancias de lnea, Dr y Dx

Matriz diagonal de resistencias de lneas, Dr. En la ecuacin 2.7 se observa la forma de
la matriz.

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

La estructura de la matriz TRX se muestra en la ecuacin . Por motivos de espacio,

se presentarn las cuatro submatrices que conforman la matriz TRX por separado. De esta
forma,

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 0,1675 0,1005

0,1005 0,1005

0,127

0,1005 0,1005

0,127

0,1005 0,1005

0,127

-0.0462

0,127

0,194

-0,0693 -0,0693 -0,0693 -0,0693 -0,0693

-0,0693 -0,1155 -0,0693 -0,0693 -0,0693

-0,0693 -0,0693 -0,1155 -0,1155 -0,1155

-0,0693 -0,0693 -0,1155 -0,1848 -0,1156

-0,0693 -0,0693 -0,1155 -0,1156 -0,1617

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

0,0693 0,1155 0,0693

0,0693 0,0693 0,1155

0,0693 0,0693 0,1155

0,0693 0,0693 0,1155 0,1156 0,1617

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 0,1675 0,1005

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 ).

Resultados de las Iteraciones

Construida la matriz TRX y conociendo el vector de potencias y tensiones iniciales es

posible realizar el FDC tal como se describe en la seccin 2.1.2 en el captulo 2. Para el circuito de 7 barras se lleg a un resultado en 3 iteraciones. El criterio de convergencia utilizado

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.

Los resultados se presentan con la siguiente nomenclatura:

Ir (p.u) y Ij (p.u). Componente real e imaginaria de la corriente en el nodo i (p.u).

Vr (p.u) y Vj (p.u). Componente real e imaginaria de la tensin en el nodo i (p.u).
.Vr y .Vj . Diferencia de la componente real e imaginaria de tensin entre iteraciones.

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

Al terminar la segunda iteracin se observa que no se cumple el criterio de convergencia

en una de las barras. La diferencia de tensin entre iteraciones del nodo 4 es mayor a la
tolerancia predeterminada.
Tercera Iteracin
Al terminar la tercera iteracin se observa que se cumple el criterio de convergencia en
todas las barras.

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 validar los resultados se utiliz el programa MATPOWER. En la seccin E del

captulo 2 se describe dicho programa. Este programa est basado en el algoritmo NewtonRaphson, NR. Sin embargo, ste trabaja con la esparsidad de las matrices para evitar construir la matriz Jacobiana, tpica en el NR. Principio que hace del programa un algoritmo
bastante poderoso y eficiente. De esta forma, se evitan problemas de convergencia tpicos en
redes de distribucin, (ver seccin 1.1 del captulo 1). Para mayor informacin referente al
programa se recomienda revisar el manual de usuario, [52].

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.

Tabla 4.8: Tensiones Nodales Validadas en Mdulo y ngulo

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

En [20], se estudia el tiempo de proceso de CPU utilizando tres algoritmos:

1. El algoritmo propuesto en [20], (TRX ), (ver seccin 2.1.2 en el captulo 2).

2. El algoritmo propuesto en [19], (FDCT ), (ver seccin 2.1.1 en el captulo 2).
3. Newton-Raphson, (NR), utilizado por el programa MATPOWER (ver seccin E en el
captulo 2).

Para dicho estudio no se consider el tiempo de entrada/salida de datos. Esto ltimo se

justifica porque en la aplicacin del mismo se planea que un programa centinela mantenga
las matrices de los datos de entrada en memoria RAM y actualizados peridicamente; razn
por la cual, no se invertira tiempo en cargar los datos sino que se tendran disponibles en
cualquier momento que se requiera.

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

Figura 4.1: Estudio Comparativo: Tiempo de Cmputo, [20].

el TRX es un algoritmo altamente competitivo respecto al utilizado por el MATPOWER,
el cual utiliza la esparsidad de las matrices para no armar la matriz Jacobiana y, en consecuencia, reducir el tiempo de computo al evitar tener que invertir y actualizar dicha matriz
en cada iteracin.

En el caso propuesto, el NR tuvo el segundo mejor comportamiento por lo anteriormente

descrito. El peor tiempo se registr al utilizar el algoritmo FDCT en nmeros complejos ya
que el tiempo requerido para realizar las operaciones se incrementa al trabajar con nmeros
complejos, [20].

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.

Para esto se utilizaron cuatro (4) circuitos de diferentes tamaos: 4, 7, 12 y 69 barras.

Los datos del circuito de 4 barras se encuentra en el apndice D, el circuito de 7 se describi
en la seccin 4.1 y los circuitos de 12 y 69 barras se encuentran disponibles en los apndices
J e I, respectivamente. El nmero de iteraciones para la convergencia es el mismo en ambos

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

Estudio con 5000 iteraciones

N 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.

Aplicacin en Redes de Gran Tamao

4.4.1.

Tiempo de cmputo

Utilizando la data proporcionada por la EDC se implement el programa desarrollado en

un total de 530 circuitos distribuidos en 78 S/E, compuestos por un total de 64.251 nodos
que corresponden a un 80 % de la demanda mxima de la Gran Caracas (3, 28GW ).

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

En la figura 4.2 se muestra de forma esquemtica el proceso algortmico que utiliza el

ASP y el tiempo requerido para obtener la solucin del flujo de carga para el caso planteado
anteriormente.

En contraste, el tiempo utilizado por el algoritmo implementado se muestra en la tabla

4.11. Los resultados presentados son de cada uno de los mdulos que componen el algoritmo.
Para este estudio, se dividi el algoritmo mostrado en la seccin B.2 del apndice B en dos
submdulos: el primero correspondiente al proceso iterativo y el segundo correspondiente al

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

En la figura 4.3 se muestra de forma esquemtica el proceso algortmico que utiliza el

algoritmo desarrollado y el tiempo requerido para obtener la solucin del flujo de carga para
el caso planteado anteriormente.

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.

An ms, se debe recordar que el proceso de adquisicin de datos y actualizacin de las

matrices arriba mencionadas se realiza offline. Razn por la cual no resulta afectado el proceso iterativo del programa. A diferencia del ASP que tiene que recurrir al disco duro para
adquirir la informacin y luego realizar el flujo de carga.

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.

Anlisis de Condiciones Operacionales de una Red de Gran

Tamao

Anlisis Ejemplo para 1 Circuito

Conocidas las tensiones nodales en mdulo y ngulo se realiz un anlisis de las condiciones operacionales del circuito AN T _A04 correspondiente al circuito A04 de la S/E An-

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.

Figura 4.4: Perfil de Cada de Tensin en ( %) del circuito AN T _A01.

La figura 4.5 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 16, 59MW, de los cuales
un 8 % representa las prdidas tcnicas en el circuito, valor que est por encima del 4 %
permitido, [56]. En promedio el valor de las prdidas para toda la red oscila alrededor del
2 %, [54].

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.

Figura 4.7: Capacidad Amperimtrica de los conductores del circuito AN T _A01.

Tambin, se analiz la potencia entregada por circuito por la S/E comparndola con las
prdidas de los mismos. Adems, se verific la carga de los conductores de todos los circuitos
de la S/E.

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.

Figura 4.8: Potencia Entregada y Perdida por Circuito de la S/E AN T .

En la figura 4.9 se 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

59

Figura 4.9: Capacidad Amperimtrica de los conductores de la S/E AN T

2.3.3 del captulo 2. Para un total de 1,649 ramas, se observa que 20 (1 %) de los conductores
superan el 100 % de la CE, 678 (41 %) de los conductores operan por encima del 67 % de la
CE. Slo 952 (58 %) conductores operan por debajo del 67 % de la CE. En trminos generales
ms del 40 % de los conductores trabajan por encima del 67 % de su CE.

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

Figura 4.10: Capacidad Firme de las S/E AN T y SRO.

en una mejor distribucin de la carga.

Anlisis Ejemplo para Toda la Red

Conocidas las tensiones nodales en mdulo y ngulo se realiz un anlisis de las condiciones operacionales de toda la red de la Gran Caracas; conformada por 530 circuitos correspondientes a 78 S/E. La cantidad total de nodos de la red estudiada es 64,251. Se estudi la
capacidad amperimtrica de todos los conductores de la red y las prdidas activas en comparacin con la potencia activada entregada.

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.

Figura 4.11: Potencia Activa Demandada y Prdidas Tcnicas de la red.

En la figura 4.12 se muestra la cantidad de conductores de la red que estn operando por
debajo del 67 % de la capacidad de emergencia (CE ), aquellos que operan por encima del
67 % y aquellos que operan por encima del 100 % de 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 63.721 ramas, se observa que 22.130 (3 %) de los conductores superan el 100 % de la CE, 24.227 (35 %) de los conductores operan por encima del
67 % de la CE. Slo 39.494 (62 %) conductores operan por debajo del 67 % de la CE. En
trminos generales ms del 35 % de los conductores trabajan por encima del 67 % de su CE.

En la tabla 4.13 se muestra un reporte general de las condiciones de operacin de la red.

62

Figura 4.12: Capacidad Amperimtrica de los conductores de la red.

Tabla 4.13: Condiciones de Operacin de la Red, 530 circuitos.

Parmetro Analizado

Resultado

Prdidas Activas Totales

557, 27MW

Prdidas Reactivas Totales

808, 25MVAr

Potencia Activas Totales Demandada

3, 28GW

Potencia Reactivas Totales Demandada

2, 03GVAr

Conductores que violan 67 % de la CE

24.227

Conductores que violan 100 % de la CE

22.130

Total de Nodos Analizados

64.251

CONCLUSIONES

A continuacin se plantean las conclusiones de este trabajo de investigacin:

Se implement un flujo de carga de barrido unidireccional directo basado en el BIBC
(desarrollado por Teng, [19]). Incluyendo la modificacin propuesta por De Oliveira
(TRX ), [20], para hacer ms eficiente al algoritmo inicial.
El algoritmo fue codificado en MATLAB y se caracteriza por presentar una lgica de
clculo de gran sencillez: se basa en una suma y multiplicacin de matrices. Adems, el
mismo fue desarrollado de forma modular, compuesto por dos mdulos diferentes: en
el primero se adquieren los datos y en el segundo se realiza el proceso de clculo.
El proyecto se aplic con un patrn Modelo-Vista-Controlador (MVC ).
Se demostr que la modificacin del flujo de carga propuesta por De Oliveria, [20], convierte al algoritmo original en un algoritmo ms eficiente; reduciendo aproximadamente
a la mitad los tiempos de cmputo, ver seccin 4.3 del captulo 4.
El algoritmo desarrollado fue validado utilizando un flujo de carga directo complejo,
[19], y un flujo de carga basado en el Newton-Raphson, [52].
Para hacer an ms eficiente al algoritmo se estableci un esquema de datos en RAM,
en el cual se colocaron las matrices que caracterizan al circuito (TRX y S ). De esta
forma, estarn disponibles en cualquier momento que se requiera hacer el flujo de Carga.
An ms, las matrices en el caso de estudio de 530 circuitos slo ocuparon un espacio
en memoria de 40MB.
El algoritmo fue implementado eficientemente en una red de 530 circuitos correspondientes a 78 S/E de la EDC. Las tensiones nodales para los 64.251 nodos se obtuvo en
un tiempo de 0, 76 segundos.
63

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

Para ilustar el proceso de armado de la matriz BIBC se utiliza el circuito de 6 barras

disponible [19]. El sistema incluye 1 generador y 5 cargas equivalentes, conectadas a una red
de distribucin radial. El circuito se muestra en la figura C.

Figura A.1: Sistema de Distribucin de Ejemplo.

A.1.

Matriz BIBC

Utilizando la Ley de Corriente de Kirchhoff (KLC ) se pueden obtener las corrientes de

rama en funcin de las corrientes inyectadas. Para este caso,

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

Z23 Z34 Z45

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.

[.V ] = [BCBV ][B]

(A.6)

Apndice B
ALGORITMO IMPLEMENTADO TRX

A continuacin se mostrarn las lneas de cdigo de los algoritmos desarrollados a lo largo

del proyecto:

1. Algoritmo de adquisicin de datos.

2. Algoritmo iterativo [20] y algoritmo de anlisis de resultados. .

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 de Adquisicin de Datos

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

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

clear all clc delete CCS.mat path=C:\Documents and

Settings\JR\Escritorio\integracion\; directorio=dir(path);
74

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
%

%%%%%%%%%%%%%%%%%%

%%%%%
%

CAPACIDAD AMPERIMETRICA DEL CONDUCTOR. %%%%%%%%%%%%%%%%

[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

V0((2*n)-2,1)=0; %inicializar vector Voltaje con "0"

Vectx=ldat(:,5); %Definimos el vector de X de las ramas.
Vectr=ldat(:,4); %Definimos el vector de R de las ramas.
for k=1:m

%ciclo que crea la matriz T1

ns=ldat(k,1);

% nodo de salida

nll=ldat(k,2); % nodo de llegada

B(:,1)=T1(:,ns);
B(nll-1,1)=1;
T1(:,k+1)=B(:,1);

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

clear nno B BCBV BCBVI BIBC BIBCI DLF I S V0

bdat k

m n nll ns z

ll nodos0 lineas0 ldat0 bdat0 t=toc+t;

end
end save CCS RED clear all
Load_flow

% SE LLAMA AL MODULO QUE EJECUTA EL FLUJO DE CARGA%

83

B.2.

Flujo de Carga TRX

Cdigo desarrollado para el algoritmo de flujo de carga implementado. El mismo coloca

las matrices T RX y S en memoria RAM y realiza el proceso iterativo descrito en la seccin
2.1.2 del captulo 2. Luego, almacena las tensiones en archivos .MAT. Con los resultados de
tensin se calculan las prdidas de potencia en los circuitos, las capacidades amperimtricas
de los conductores y las cadas de tensin; todos los resultados son almacenados en el mismo
archivo .MAT.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%% DECLARACIN DE VARIABLES

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

clear all clc

load CCS % SE CARGA LA MATRIZ DE DATOS DEL SISTEMA A LA RAM
iter=10;

%definir nmero mximo de iteraciones

con=0.0001; %definir criterio de convergencia

t=0; t1=0; Y=size(RED,1); ntot=0;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% CICLO QUE INICIA Y CARGA CADA CIRCUITO %%%%%%%%%%%%%%%%%%%
for qp=1:Y
iter2=0;
TRX=RED{qp,2};
S=RED{qp,3};
n=size(S,1);
m=n-1;
V0((2*n)-2,1)=0;
T=RED{qp,5};
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%% CICLO QUE CREA LOS VECTORES DE "S" E "I" INICIALES %%%%%%%%%%
for k=1:n-1
V0(k,1)=1;
I(k,1)=((-S(k+1,1)*V0(k,1))-(S(k+1,2)*V0(k+(n-1),1)))/(((V0(k,1))^2+

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

conve>con & iter>iter2

iter2=iter2+1; %Contador de Iteraciones.

Vk=V1;%Definimos una variable Vk para actualizar los voltajes corregidos.
for k=1:n-1 %Proceso para corregir las corrientes del sistema.
I(k,1)=((S(k+1,1)*Vk(k,1))-(S(k+1,2)*Vk(k+(n-1),1)))/(((Vk(k,1))^2+
(Vk(k+(n-1),1))^2));
I(k+(n-1),1)=(-S(k+1,2)*Vk(k,1)+S(k+1,1)*Vk(k+(n-1),1))/(((Vk(k,1))^2+
(Vk(k+(n-1),1))^2));
end
DV=TRX*I; %corrige el delta de tensin para la prxima iteracin.
V1=V0-DV; %corrige el voltaje final antes de comprobar la convergencia.
conve=max(abs(DV2-DV)); %Define la convergencia midiendo la variacin
%de DV
DV2=DV;
end toc; t=t+toc; tic;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%% CICLO QUE CONVIERTE A "I" EN IMAGINARIO %%%%%%%%%%%%%%%%%%

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

NOMBRE DEL CIRCUITO: GRA_A01.xlsx

VOLTAJES EN LOS NODOS

***************************************************************

# 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

PERDIDAS EN LAS RAMAS

***************************************************************

# 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

TOTALES DEL CIRCUITO

***************************************************************

Perdidas activas totales (pu): 9.215657e-004

97

Perdidas reactivas totales (pu): 6.504554e-004

Cantidad de conductores violan el 67% capacidad: 28

Cantidad de conductores violan el 100% capacidad: 25

Total de nodos en este circuito: 93

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.

Tabla D.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

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)

Converged in 0.01 seconds

================================================================================
|

System Summary

================================================================================

How many?

How much?

-------------

P (MW)

-------------------

-----------

Q(MVAr)
------------

Buses

Total Gen Capacity

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

Total Inter-tie Flow

0.0

0.0

100

101

Areas

Minimum

Maximum

-------------------------

--------------------------------

Voltage Magnitude

0.917079 p.u. @ bus 7

1.000 p.u. @ bus 1

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

From Bus Injection To Bus Injection Loss (I^2*Z)

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

IEEE Transactionson Power Systems, Vol. 3, No. 2, May 1988

A COMPENSATION-BASED POWER FLOW METHdD

FOR WEAKLY MESHED DISTRIBUTION AND TRANSMISSION NETWORKS

D. Shirmoharmnadi
Member

H. W. Hong
Senior Member

Systems Engineering Group

Pacific Gas and Electric Company
San Francisco, California

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:

1. Capable of Solving radial and weakly meshed

distributation networks with up to several
thousand line sections (branches) and nodes
(buses)

2.

Robust and efficient.

The efficiency of such a power flow algorithm is

of utniost importance as each optimization study
requires numerous power flow runs. Furthermore, the
extension of the application of this power flow method
to three phase networks with distributed loads was also
envisaged.
The Newton Raphson and the fast decoupled power
flow solution techniques and a host of their
derivatives have efficiently solved "well-behaved"
power systems for more than two decades. Researchers,
however, have been aware of the shortcomings of these
solution algorithms when they are "generically"
implemented and applied to ill-conditioned andlor
Hence,
poorly
initialibed power systems [ 1 , 2 , 3 ] .
conrnerical power flow packages always modify these

This pap& was sponsored by the IEEE Power Engineering Society

for presentation at the IEEE Power Industry Computer Application Conference,Montreal,Canada,May 18-21,1987. Manuscript
was published in the 1987 PICA Conference Record.

A. Semlyen
Senior Member

G. X. Luo

Department of Electrical Engineering

University of Toronto
Toronto, Ontario, Canada

algorithms for enhanced robustness. The nature of

modifications and the degree of improvement obtained
varies for different packages. The Gauss-Seidel power
flow technique, another classicial power flow Ethod,
although very robust, has shown to be extremely
inefficient in solving large power systems.
Distribution networks, due to their wide ranging
resistence and reactance values and radial structure,
fall into the category of ill-conditioned power systems
for the generic Newton-Raphson and fast decoupled power
flow algorithms. Our experience with a basic NewtonRaphson power flow program for solving distribution
networks was mostly unsuccessful as it diverged for the
majority of
the networks
studied.
Later we
successfully used the Newton-Raphson based Western
System Coordinating Council (WSCC) power flow program.
This program, which is commonly used by the WSCC member
utilities,
includes
several
enhancements
for
increasing its convergence capabilities. Although the
robustness of
the program was acceptable, the
computation time was excessive.
In addition, the
extension of the Newton-Raphson algorithm to the
solution of the three phase networks, not even
considering distributed
load, would
result in
substantial deterioration of the numerical efficiency
of the solution algorithm [ 4 1 .
Efficient power flow algorithms for solving single
and three phase radial distribution networks [5,6,71
have been extensively used by distribution engineers.
These algorithms are not, however, designed to solve
meshed networks.
In this paper, we propose a new method for the
solution of weakly meshed networks. In this method, we
first break the interconnected grid at a number of
points (breakpoints) in order to convert it into one
radial network. Each breakpoint will open one simple
loop. The radial network is solved efficiently by the
direct application of Kirchhoff's voltage and current
laws (KVL and KCL). We then account for the flows at
the breakpoints by injecting currents at their two end
nodes. The breakpoint currents are calculated using
the multi-port compensation method [ 8 , 9 1 . In presence
of constant P,Q loads, the network is nonlinear causing
the compensation process to become iterative. The
solution of the radial network with the additional
breakpoint current injections completes the solution of
the weakly meshed network.
Our studies have shown that, typically, only a few
iterations were required for the solution of
distribution networks using the proposed power flow
solution technique. For the weakly meshed transmission
networks the number of iterations was higher, due to
the additional nonlinearities introduced by generator
buses (PV nodes).
In all the cases studied the
proposed power flow technique was significantly more
efficient than the Newpn-Raphson power flow algorithm
while converging to the same solution.

0885-8950/88/05oo-O753$01 .WO 1988 IEEE

754

The numerical efficiency of

the proposed
compensation-based
power flow method, however,
diminishes as the number of breakpoints required to
convert the meshed network to a radial configuration
increases. This restricts the practical application of
the method to weakly meshed networks.
In this paper we emphasize the application of the
compensation-based power flow method
to
the
distribution networks and provide several practical
examples. A comparison with the Newton-Raphson based
WSCC power flow program is also presented. We then
discuss the application of the algorithm to weakly
meshed transmission networks where again comparison
with the WSCC power flow program is provided. A brief
discussion of the extension of this method to three
phase networks as well as the treatment of distributed
loads and other practical considerations concludes the
paper.

11. SOLUTION OF A RADIAL DISTRIBUTION NETWORK

JL(k) =

currents in
-ILt(k)+C(branches emanating)
from node L2
L=b,b-1,

(2)

...,1

where IL2(k) is the current injection at node L2.

This is the direct application of the KCL.
3. Forward sweep:
Nodal voltages are updated in
a forward sweep starting from branches in the
first layer toward those in the last. For
each branch, L, the voltage at node L2 is
calculated using the updated voltage at node L1
and the branch current calculated
in the
preceding backward sweep:

where z~ is the series impedance of branch L.

This is the direct application of the KVL.
Steps 1, 2 and 3
achieved.

are repeated until convergence is

In our algorithm, regardless of its original

topology, the distribution network is first converted
to a radial network. Hence, an efficient algorithm for
the solution of radial networks is crucial to the
viability of the overall solution method.
The solution method used for radial distribution
networks is based on the direct application of the KVL
and KCL. Similar techniques are also described in
[5,6,7]. For our implementation, we developed a branch
oriented approach using an efficient branch numbering
scheme to enhance the numerical performance of the
solution method.
We first describe this branch
numbering scheme.
Branch Numbering
In contrast to all classical power flow techniques
which use nodal solution methods for the network, our
algorithm is branch-oriented. Figure 1 shows a typical
radial distribution network with n nodes, b(=n-1)
branches and a single voltage source at the root node.
In this tree structure, the node of a branch L closest
to the root note is denoted by L1 and the other node by
L2. We number the branches in layers away from the root
node as shown in Figure 2. The numbering of branches
in one laver starts onlv after all the branches in the
implemented in our power flow program.
Solution Method
Given the voltage at the root node and assuming a
flat profile for the initial voltages at all other
nodes, the iterative solution algorithm consists of
three steps:
1. Nodal current calculation: At
iteration k,
the nodal current injection,Ii(k),
at network
node i is calculated as,

I,@)

...,n

= (si/vi(k-l) )*-Y~v~(~-I)i=1,2,

(1)

where Vi(k-l) is the voltage at node i calculated

during the (k-l)th iteration and Si is the
specified power injection at node i. Yi is the
sum of all the shunt elements at the node i.

LAYER 10

2. Backward

sweep: At iteration k, starting

from the branches in the last layer and moving
towards the branches connected to the root node
the current in branch L, J,, is calculated as:

Fig.2

Branch numbering of the radial distribution

network

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 )

The real and reactive power mismatches at bus

then calculated as:
mi(k)

= Re[Si(k)

AQ~(~)= ImIsi(k)

Si]
i=1,2,.

- si]

...n

i are

(4)

Table 1 shows the values of the maximum real and

reactive power mismatches at the various iterations of
this radial network power flow solution algorithm for
three practical distribution networks. An excellent
rate of convergence can be observed for all three
networks studied. This convergence behavior can be
briefly explained as follows. The error that is
incurred in estimating initial node voltages
is
propagated first to nodal and then to branch currents
via equations (1) and ( 2 ) . In the process of updating
node voltages using equation ( 3 1 , the error in branch
currents is multiplied by the small line impedance. ZL
()ZLI<<l). and thereby rapidly attenuated.
~

DISTRIBUTION

~~~~

CALCULATEMAXIMUMREAL AND
REACTIVEPOWER MISMATCHES
EQUATION (4)

MAXlMUMlTERAT

Power flow solution algorithm

networks

Fig.4

A weakly meshed distribution network

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

lTERATlON MAXIMUM REAL

MAXIMUM REACTIVE
NUMBER
POWER MISMATCH POWER MISMATCH
(WAR)
(W

244 NODES

YES

for the radial

0.037
0.003

Convergence characteristics of the radial

network solution algorithm

Figure 3 shows the flow chart of the overall power

flow solution method for radial networks.
111. SOLUTION OF WEAKLY MESHED DISTRIBUTION NETWORKS
Figure 4 shows an example of a weakly meshed
distribution network containing four simple loops. The
radial network solution algorithm can not be directly
applied to this network. Nevertheless, by selecting
four breakpoints, this network can be converted to a
radial configuration. The branch currents interrupted
by the creation of every breakpoint can be replaced by

current injections at its two end nodes, without

affecting the network operating
condition.
This
p solved by the
resulting radial network can now t
radial network solution technique described earlier.
In applying the radial network solution algorithm,
the current at breakpoint j, Jj, must be injected with
opposite polarity at the two end nodes of the
breakpoint. At iteration k:

156

j = 1, 2,

..., p

(5)

where jl and j2 corr s ond to h two end nodes of the

breakpoint j, and Ijirkf and Ijjke become he nodal
current injections at these nodes,
is the
breakpoint current and p the total number of
breakpoints. In the presence of nodal currents at the
t a point no e
due to shunt elements andlor loads,
$.?kt and -3jh7'
must be added to these nodal currents.
rocess is c ematically shown in Figure 5.
Once
and I-$?
are updated, steps 2 and 3 of the
radial networi solution algorithm can be directly
applied.

Tj(6

In the presence of constant power loads the

distribution network is, however, nonlinear and equation
(6) cannot be directly used.
Instead, as we shall
explain, we calculate breakpoint currents iterively
using the Thevenin equivalent circuit.

$Tky

BREAKPOINT

Pig.7

Thevenin equivalent circuit of the network as

seen from the breakpoint ports

Calculation of Breakpoint Impedance Uatrix

Fig.5

Breakpoint representation using nodal current

injections

The breakpoint
equivalent impedance)
following method:

impedance matrix
(Thevenin
can be determined using the

Equation (6) can be written as,

Calculation of Breakpoint Currents Using Compensation
Method
Breakpoint
currents are calculated using the
multi-port compensation method
[81.
Figure 6
illustrates the concept used in this approach. In this
figure the radial network resulting from the opening of
the breakpoints is shown as a multi-port circuit with
breakpoint nodes forming the ports of the circuit. The
calculation of breakpoint currents requires that the
multi-port equivalent circuit for the radial network as
seen from the ports of the breakpoints be established.
For a linear network, this multi-port equivalent
circuit can be the Thevenin equivalent circuit of the
radial network seen from the open ports created b the
is
breakpoints. In this circuit the Thevenin voltage
the (pxl) vector of open circuit breakpoint voltages,
obtained from the power flow solution of the radial
network
,[dl the (pxp) non-sparse matrix of the
breakpoint impedances (coefficients relating breakpoint
currents and voltages) and 2 is the (pxl) vector of the
desired breakpoint currents (Figure 7):

.. zpj ..... C '

A.

According to equation (71, column j of the

breakpoint impedance ma rix will be e ual t o vector of
breakpoint voltages for $j=l p.u. and
i=1,2,. .p
and i # j. This corresponds to the application of
1 p.u. current at the breakpoint j with all loads and
the source at the root node removed, which is, in turn,
equivalent to the injection of 1 p.u. currents with
opposite polarity at the two end nodes of the
breakpoint j (equation (5)).
In the absence of loads,
the accurate solution of the power flow for the radial
network can be achieved in one iteration. Each of the
breakpoint voltages can be determined by subtracting
the voltages at the two end nodes of the breakpoint.
This process must be repeated for all breakpoints until
all the columns of the breakpoint impedance matrix are
calculated.

ji=O,

..

Iterative Compensation Process

The iterative compensation process for calculating
the breakpoint currents, using the Thevenin equivalent
circuit of Figure 7 , is described in the following:

1. Calculate the Thevenin equAvalent impedance

(breakpoint impedance matrix [Z] .of the radial
network) maintaining it constant throughout the
compensation process.

1)

Fig.6 Multi-port equivalent of the network as seen

from the breakpoint ports

2. Calculate the Thevenin e uivalent voltage

of the radial
(breakpoint voltage vector
network using the radial network
solution
algorithm (Figure 3 ) including the breakpoint
currents calculated from the previous iteration
of the compensation process. The initial values
of the breakpoint currents are zero.

757

3. Calculate the incremental change in the

breakpoint currents using the Thevenin equivalent
circuit.
At iteration m of the compensation
process:

1. Examine
all
connected to
layer

branches
and
select those
the root node for the first branch

2. Store

the node number of the far node of the

branches in the branch layer just formed. For
all these nodes raise a flag indicating that they
have already been used

4. Update the breakpoint currents. At iteration m:

(8")
steps 2, 3 and 4 until convergence is
breakpoint
voltage
reached
(the maximum
calculated
at
step 2 is within prescribed
limits)

3. Examine

all the remaining branches and select

those connected to any of far nodes of the
branches in the previous layer and place them in
a new branch layer

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.

4. If the end node of a branch numbered in step 3

has been used before (flag identification of step
2) a loop has been formed and a breakpoint must
be created at this node
5 , Repeat

steps
processed.

2-4

'

until

all branches are

Figure 9 shows the flow chart of the overall power

flow solution schome.

SOLVE RADIAL NRWORK LOAD R O W ( W R E 3)

UPDATE THEVEMN VOLTAGE MISMATCHAT BREAKPOINTS

CALCULATE BREMPOhlT CURRENTS ENATKIN (8)

ADD B E A P O M CURRENTSTO NODAL CURRENTS

$ ( l ) $12)

Fig.8

Graphic representation of the iterative

compensation process

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

IV. RESULTS FOR THE DISTRIBUTION NETWORKS

We developed the program WNETPF (Weakly meshed
NETwork Power Flow) based on the proposed power flow
solution algorithm.
This program successfully solved
several practical distribution networks with radial and
meshed structures.
Table 2 shows the performance
results of this program alongside those of
the
Newton-Raphson based WSCC power flow program. We also
used two other power flow programs using the generic
Newton-Raphson and the Gauss-Seidel solution algorithms
at this application stage. The generic Newton-Raphson
solution algorithm only converged for the smallest
network. The Gauss-Seidel power flow method converged
in all cases while requiring in excess of 20,000
iterations.

758

For the cases reported in Table 2, the proposed

algorithm converged in less than 14 iterations in 0.21
to 2.1 CPU seconds on a mainframe computer. Each
iteration of this program corresponds to one iteration
in the radial network solution algorithm prior to,
during, and after the calculation of breakpoint
currents (variable k in Figure 3 ) . The number of outer
iterations for calculating the breakpoint currents in
the compensation process (variable m in Figure 9) was
less than 6 in all three cases. A flat start was used
in all cases with the tolerance for the real and
reactive power mismatches set to 0.05 kW and 0.05 kvar.
Table 2 indicates that the proposed power flow
program is significantly more efficient than the
Newton-Raphson power flow method when studying radial
and weakly meshed
distribution networks.
This
conclusion is particularly crucia1,for: (a) on-line
applications; (b) multiple power flow studies; (c)
micro- and mini-computer applications.
The W T P F
program does not require double
precision variables and uses only one two dimensional
array (breakpoint impedance matrix), hence, avoids
taxing computer resources.
IISTRIWTION

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)

is then used for calculating the voltages at the

Vi
end nodes of the branches in the next layer. The
reactive power at the generator node i is then updated
using the secant method as described in the Appendix.
Inclusion of generator nodes using this approach
does not noticeably deteriorate the convergence
properties of the radial network power flow solution
algorithm. The efficiency of the solution method for
the weakly meshed networks will, however, be affected
by the introduction of the generator nodes.
The
nonlinearity in the transmission network caused by the
additiorial generator nodes is more pronounced than that
of the distribution networks having constant power
loads alone. This results in an increased number of
iterations for calculating the breakpoint currents
using equation ( 8 ) .
Table 3 shows the performance results of applying
the proposed algorithm and the Newton-Raphson based
WSCC power flow program to one radial and three weakly
meshed transmission networks. These networks were
synthesized from a practical 500kV transmission network
with a high degree of series compensation. The table
shows that for the networks with low number of
breakpoints, the proposed load flow techhiques is more
efficient than the Newton-Raphson method. However, as
the number of breakpoints increased, the proposed
method required significantly higher number of
iterations while the Newton-Raphson algorithm converged
with the same number of iterations.

1411 NOMS

42 BUS, WOW

N3WNoDEs
N3Loops

36 LOOPS

527

1.47

0011

NO LOOPS

10

1 LOOP

3 LOOPS

Table 2 Performance results for distribution networks

(IBM 3090-200 mainframe computer) - The total
CPU time in seconds includes the time for
initial processing of the network data and the
iterative solution algorithm.

SLOOPS
42 BUS, 5W KV
4
w
m

NO LOOPS
1 LOOP

V. APPLICATION TO WEAKLY MESHED TRANSMISSION NETWORKS

3 LOOPS

In a weakly meshed transmission network, the swing

bus is assigned as the root node. Then the branches
are numbered and breakpoints selected in exactly the
same manner used for the weakly meshed distribution
networks. As a result a radial transmission network is
created.
The solution algorithm for a radial transmission
network is identical to that of a radial distribution
network except for the processing of the generator (PV)
nodes. For the PV node i having a specified power
injection PT and voltage magnitude I VT I , we start
the iterations of the radial network solution algorithm
by assuming Vi(o) = lVrl /O
and
= 0. Steps 1
and 2 of the radial network solution method are then
performed in the same manner as before.
Step 3 ,
however, must be modified.
At iteration k, the voltage magnitude at the
generator node i, calculated using equation ( 3 ) must
modified as,

5 LOOPS

Table 3 Performance results for transmission networks

(IBM 3090-200 mainframe computer)

VI. PRACTICAL CONSTDERATIONS

Detailed representation of a distribution network
requires [lo]:
1.

The three phase representation of the network to

account for the actual load unbalances.

2.

The distributed load representation along the

distribution lines.

3.

The representation of load tap changers, voltage

regulators, boosters, etc.

159

Requirement 1 can be directly incorporated in the

proposed power flow method by replacing voltage and
current scalars in equations (1) thru (lo), by (3x1)
vectors of voltages and currents of the three phases.
(31, (4)
Under these circumstances, in equations (l),
and (71, the admittances and impedances should be
represented by (3x3) matrices.
Distributed loads along distribution lines can be
approximated by lumped loads at system nodes for power
flow calculations purposes. This, however, may require
the addition of pseudo-nodes along some of the lines
where part of the line load is lumped. The proposed
power flow method is also capable of directly including
distributed loads. This can be achieved by
the
modification of equations(1) and (3).
Distribution
network equipment
(regulators,
boosters,' etc.) can be modeled in the proposed
algorithm without any restriction.
This method is directly applicable to distribution
planning studies, where single phase representation of
the network with lumped nodal loads are considered
to be adequate.
In addition we have successfully
used
this power flow method
for our optimal
network reconfiguration studies.
In the case of weakly meshed
transmission
networks, a complete system representation requires the
inclusion of tap changers and phase shifters. The
proposed solution algorithm is capable of including
these components directly. This requires that the KVL
and KCL be written for the mathematical model of the
tap changers and phase shifters.
VII. CONCLUSION
This paper presents a new, compensation-based
power flow method, for the solution of weakly meshed
distribution and transmission networks. This technique
is simple, straightforward, computationally efficient
and numerically robust. Extensive study of the
performance of this compensation-based power flow
scheme shows that it is significantly more efficient
than the Newton-Raphson power flow technique when used
for solving radial and weakly meshed distribution and
transmission networks.

K. A. Birt, J. J. Graffy, J. D. McDonald,

A. H. El-Abiad. "Three Phase Load Flow Program",
IEEE Trans., PAS-95, JanuaryIFebruary 1976, pp.
59-65.
W. H. Kersting, D. L. Mendive, "An Application of
Ladder Network Theory to the Solution of Three
Phase Radial Load-Flow Problems", IEEE PAS Winter
Meeting, New York, 1976, IEEE Paper No. A76 044-8.
D. I. H. Sun, S. Abe, R. R. Shoultz, M. S. Chen,
P. Eichenberger, D. Farris, "Calculation of Energy
Losses in a Distribution System", IEEE Trans.,
PAS-99, July/August 1980, pp. 1347-1356.
R. Berg Jr.,
E. S. Hawkins, W. W. Pleines,
"Mechanised Calculation of Unbalanced Load Flow
for Radial Distribution Circuits", IEEE Trans.,
PAS-86, April 1967, pp. 415-421.
W. F. Tinney, "Compensation Methods for Network
Solutions by Triangular Factorization", Proc. of
PICA Conference, Boston, Mass., May 24-26, 1971.
G. Gross, H. W. Hong, "A Two-step Compensation
Method for Solving Short Circuit Problems", IEEE
Trans., PAS-101, June 1982, pp. 1322-1331.

[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.

111 S . C. Tripathy, G. Durga Prasad, 0. P. Malik

G. S. Hope, "Load Flow Solutions for
Conditioned Power Systems By a Newton
Method", IEEE Trans., PAS-101, October 1982,
3648-3657.

and
111Like
pp.

[2] S. Iwamoto, Y. Tamura, "A Load Flow Calculation

Method for Ill-Conditioned Power Systems", TEEE
Trans., PAS-100, April 1981, pp. 1736-1743.
[31 D. Rajicic, A. Bose, "A Modification to the Fast
Decoupled Power Flow for Networks with High R/X
Ratios", PICA '87 Conference, Montreal, Canada.

Qic <Qimh

where Qimin and Qiare the respective m i n i m and

maximum reactive power limits for the generator node i.

R. P. Broadwater and A. Cbandrasekaran (Tennessee Technologkd

University, Cookeville, TN): The authors are to be complimented for
addressing an area that has received little attention, distributionpower flow
analysis.
Radial Distribution System Analysis
1) A suggested improvement to the radial power flow is to sum load
powers and power losses in the reverse trace (i.e., moving from the endiog
buses to the source bus) instead of summing load currents. This suggestion
has been tested on a four-line section systemas illustratedin Fig. D. 1. For a
nominally loaded case, both methods converged in four iterations.
However, for a very heavily loaded case, the method of summing the

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.

Weakly Meshed Transmission System

4) The impedance matrix of (7)of the paper appears to be the loop
impedance matrix of the system with the breakpoint currents chosen as the
loop currents. At the exact solution, the breakpoii voltage vector must go
to zero. There appears to be a paradox here since for the constant
impedance matrix assumed the solution for the currents is either trivial or
illfinite.
5 ) The handling of PV buses explained in section V does not mention
whether the alternating current directions in the traces would affect
convergence if the P value is higher than downstream lo&.

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?

N. Vempnti, B. K. Chea, and R. R. Shoolts (University of Texas at

Arlington, Arlington, TX): The authors have presented an algorithm for the
load flow solution of weakly meshed distribution and transmission
networks. The simplicity of the algorithm makes it a very interesting paper
to read. However, the practical limitations of the algorithm prevent its
general usage for networks with multiple loops. It would be inadvisable to
have two programs, one to deal with strongly meshed networks and yet
another for weakly meshed ones.
The results were based on tests performed on radial and weakly meshed
networks using the positive sequence representation. The authors attempt to
extrapolatethe results to three-phasedistribution and transmission networks
without a report of such an analysis. Until such results are presented, the
efficacy of such an algorithm is still in doubt.
The authors make an erroneous observation that power flow algorithms
for meshed distribution systems have not been developed. One of the
algorithms referred to [6] has the ability to analyze three-phase networks,
irrespective of the complexity of the meshing. However, this algorithm was
designed for distribution systems and therefore has the limitation of only
v) bus. Subsequently this
one swing bus and no other voltagecontrolled(Palgorithm was modified to accept numerous P-V buses, thereby enabling its
usage in the analysis of three-phase transmission networks [A]. We feel that
the algorithm [6] based on an implicit Z-bus (i.e., bifactorcd Y-bus)
formulation is superior to the one proposed by the authors.
How do the authors propose to model the open-wyelopendelta transformers in the distribution system analysis? Our simulations have shown
that the injection currents due to the model are so large that the currents due
to the loads are negligible. This affects the convergence characteristics of
the algorithm directly. ~ n insight
y
into the modeling and simulatioh of
these transformers would be welcome.
References
[A] B. K. Chen, Transmission System Unbalance Analysis, Ph.D.
Diss., The University of Texas at Arlington, December 1986.

Fig. D.1. Four-lie section test system.

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.

K. Aolri, K. Nan, and T.Sat0 (Hiroshima University, Higashihiroshima,

Japan): The authors have written an interesting paper on a load flow
problem of radial power system. Although many algorithms have been

M.E. Baraa and F. F. Wu (University of California, Berkeley, CA): This

paper points out the reasons why a special power flow for distribution
systems is needed and provides a computationally attractive method. It is a
valuable contribution.
There are a few points on which we would appreciate the authors
claritlcation.
1) How does the convergence of the method for radial networks depend
on the system p e t e r s , in particular, the line resistances?
2) The amlication of the comuensation method for weaklv meshed
networks ismavery clever idea. ~e method uses an appro&te of the
linearized V-J function at the breakpoiit. We wonder if the convergence of
the method is always monotonic as implied in Fig. 8 or it exhibits
oscillatory behavior.
3) In calculating the breakpoiit impedance matrix, the authors have
observed that the corresponding power flow solutions can be achieved in
one iteration. Is this also true in cases where the shunts in the system are
significant and cannot be neglected?

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 number of questions arise in connection with the results. The paper

infers that constant power loads are assumed. What is the effect on
convergence for loads that have a mixture of p / u and q / u characteristics?
No mention is made of the use of convergence factors. Was this studied and
are there indications of optimal factors that can be considered generic for
the radial solution and for breakpoint injection currents?
The authors mention that the efficiency of the algorithm deteriorates with
the increase in the number of loops. In a large city with low voltage
downtown networks, the number of loops required to interconnect a large
number of radial sections can be in the hundreds. There is the additional
problem of a significant number of very short cable sections used to balance
the flows between sections.What degree of deterioration can be expected in
the efficiencyof solution for such a network where the coupling can be quite
strong?
\
There is a significant advantage in being able to solve the networks
described above on a desktop microcomputer which will be restricted in the
memory available for processing. Would the authors like to suggest a
sensible method of partitioning such a network to achieve solutions in an
acceptable time? How would the compensation method be modified to
account for the partitioning?
D. Shirmohammadi, H. W. Hong, A. Semlyen, and G. X. Luo: We
would like to thank the discussers for their interest in our paper and for
their questions anti comments. Many of these constitute contributions to
the topic of the paper. We will give our answers to each discusser
separately.
'

Messrs. ~roamVaterand Chandrmehrun:

We believe that the discussers' suggestion of adding up powers
rather than currents in the backward sweep is interesting and their
explanation quite plausible. We suspect however that the improved
convergence is obtained with more computations. Moreover, we
have not experienced any problems in dealing with heavily loaded
networks and low node voltages. In fact, in the 544 node network
example of the paper, due to heavy loading and lack of VAR s u p
port the voltages of some of the nodes a as low as 0.75 per unit.
This network was efficiently solved by the proposed algorithm.
The p m s S of the attenuation of the e m r s described in the section
of the paper on "Convergence Criterion" is completely general and
would apply to multi-phase unbalanced as well as single-phase networks.
AU thm examples included in the paper repment peak load conditions on the feeders. These were 4 MW, 3 M W , and 8 MW loads
on the 244 node, 544 node, and 1411 node networks, respectively.
We would like to clarify the discussers' question related to equations (6) and (7): the voltages there are internal Thevenin voltages
while the breakpoint voltages are zero since the breakpoint ports
are shortcircuited. We would also like to point out that the actual
relationship between the break point voltages and currents is established through equation (8).
We have not experienced any problem in the overall convergence
of the proposed method in solving a variety of networks that
included P,V buses. We agree, however, that there may be more
efficient ways of handling P,V buses in order to minimize the
impact on the convergence characteristics of the method.
In contrast to the discussers assumption, the inclusion of initial processing time puts in m3re favorable light the WSCC Power Flow
Program. For example, the time required for each iteration of the
proposed algorithm is around 0.005 seconds for the 244 node
example of the paper. At the same time, every iterator of the
Newton-Raphson based WSCC Power Flow Program took 0.08
seconds for the same network (16 times more). Similar results
were obtained for the 544 node and 1411 node networks.
In our distribution network model we have assumed that loads are
concentrated at network nodes. As a result, the introduction of a
load at the center of branch 4 means the addition of a new node at
this location. This will also add a new branch to the network. If
there are no loads at the end node of branch 5, this branch and
branch 10 can be combined and represented with a single branch.
Branch 6 must exist, because two separate branches emanate from
its end node.
There is no need for the renumbering of the entire network as a
result of the insertion or deletion of network components. Only

branches in the layers below the inserted or deleted component

must be renumbered.
(9) We do not agree with the discussers view on the merits of the proposed algorithm. These are well documented throughout the paper.
Furtliermore, we would like to point out that many major advances
in the development of power system analysis techniques have
resulted from methodologies that exploit the special structure of the
power system. An example is the Fast Decoupled Power Flow
which provides improved efficiency of "a few CPU-seconds" in the
solution of the transmission network by exploiting the low RIX
ratios prevalent in these networks. Yet, the impact of the Fast
Decoupled Power Flow in the field of the transmission network
analysis has been very significant.

Messrs. Aoki, Nara, and Sato:

As pointed out by the discussers, special features of particular load
flow problems can be exploited to procluce more efficient solution
methods. In the approach of the paper, the transversal elements have
been lumped witH the loads (see our answers to Messrs. Vempati, Chen.
and Shoults): this has made the factorization and the subsequent algebraic operations with the L and U matrices of a nodal approach
equivalent to using tree-branch voltages and currents. These matrix calculations are implicit in the method and did not have to be performed
explicitly. This is the clue of the computational efficiency of branchoriented calculations in a radial network. However, at each node we have
had to enforce the power (P,Q) constraints. Therefore, all bus voltages
are used as variables. Because of this only little improvement of
efficiency can be derived from the radial network structure in a Newtontype load flow.

Messrs. Vempati, chen, and Shoults:

It is true that the solution method of the paper becomes less
efficient as the number of loops increases. However, the ovenvhelming
majority of all distribution networks and also the transmission systems of
many countries are weakly meshed. Therefore, a special program, if it is
more efficient than a general one, is celtainly of interest Very often
advances in any field of knowledge are based on the special structure of a
palticular problem. One could cite innumerable examples where Special
programs are developed and used for particular situations.
We have not yet attempted the application of our method to multiphase networks. However, as we explained in our response to Messrs.
Broadwater and chandrasekaran we do not foresee any deterioration in
the convergence characteristics of this method when applied to multiphase networks.
With regard to our statement about the lack of power flow algorithms for meshed distribution network, we note that we made two
erroneous remarks. First, as the discussers rightly argue, the 2-bus solution algorithm of Ref. [6] is capable of solving meshed distribution networks. Second, we stated that Ref. [6] proposes a popular algorithm for
distribution network analysis. Based on further investigation, however,
we have found that this solution method is by no means popular among
distribution engineers.
The discussers raise the problem of the relative computational
efficiency of the method of Ref.[6] compared to the method of the paper.
It is easy to show that a nodal approach (for example the one of Ref.[61)
requires more computation than the branch-oriented approach of the
paper. We will show that the WOdiffer computationally by two facts:
(a) The computations in the branch-oriented method are equivalent to
the forward and backward substitutions of the nodal method but the
factorization of a matrix is not needed.
(b) The forward and backward substitutions in a nodal approach are
replaced in the branch-oriented method by additions and subtractions and no multiplications and divisions are needed.
Consider, for example, the simple radial network of Fig.D.l of the
discussion by Messrs. Broadwater and cbandrasekaran. Connect the mot
node to ground via a voltage source with V=O but do not connect
impedance branches to ground from the other nodes. Number the nodes
and branches moving outward from the mot. We will then have branch
and bus voltages and currents. We can relate these to each other by the
incidence matrices AV and A,,
VbW=AvVb and l b = A r l k

(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.

Messrs. Bamn and Wu:

The convergence of the method for 4 radial network is linear and
dependent essentially on the line impedances IZ I and the apparent
powers IS I of the loads. This can be seen from the following
simplified convergence analysis, for a single line of impedance
Z = R + j X and load S = P + j Q . For this, eqn.(l) becomes I=S*/V*
or, in incremental form,
N=-K
V"A Y '

(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:

so that the solution of (c) becomes

(1)

This equation shows that the convergence rate of I AV 1 (not A I V I

!) is given by Iz(i I = IZ I IS I . It depends only indirectly on line

(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

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 18, NO. 3, JULY 2003

A Direct Approach for Distribution System Load

Flow Solutions
Jen-Hao Teng, Member, IEEE

AbstractA direct approach for unbalanced three-phase

distribution load flow solutions is proposed in this paper. The
special topological characteristics of distribution networks have
been fully utilized to make the direct solution possible. Two
developed matricesthe bus-injection to branch-current matrix
and the branch-current to bus-voltage matrix and a simple
matrix multiplication are used to obtain load flow solutions. Due
to the distinctive solution techniques of the proposed method, the
time-consuming LU decomposition and forward/backward substitution of the Jacobian matrix or admittance matrix required
in the traditional load flow methods are no longer necessary.
Therefore, the proposed method is robust and time-efficient. Test
results demonstrate the validity of the proposed method. The
proposed method shows great potential to be used in distribution
automation applications.
Index TermsDistribution load flow, distribution automation
system, radial network, weakly meshed network.

I. INTRODUCTION

ANY programs of real-time applications in the area of

distribution automation (DA), such as network optimization, Var. planning, switching, state estimation, and so forth, require a robust and efficient load flow method [1][3]. Such a
load flow method must be able to model the special features of
distribution systems in sufficient detail. The well-known characteristics of an electric distribution system are
radial or weakly meshed structure;
multiphase and unbalanced operation;
unbalanced distributed load;
extremely large number of branches and nodes;
wide-ranging resistance and reactance values.
Those features cause the traditional load flow methods
used in transmission systems, such as the Gauss-Seidel and
Newton-Raphson techniques, to fail to meet the requirements
in both performance and robustness aspects in the distribution
system applications. In particular, the assumptions necessary
for the simplifications used in the standard fast-decoupled
Newton-Raphson method [4] often are not valid in distribution
systems. Therefore, a novel load flow algorithm for distribution
systems is desired. To qualify for a good distribution load flow
algorithm, all of the characteristics mentioned before need to
be considered.
Several load flow algorithms specially designed for distribution systems have been proposed in the literature [5][13].
Manuscript received October 17, 2002. This work was sponsored by National
Science Council under research Grant NSC 88-2213-E-214-041.
The author is with the Department of Electrical Engineering, I-Shou University, Kaohsiung 840, Taiwan, R.O.C.
Digital Object Identifier 10.1109/TPWRD.2003.813818

Some of these methods were developed based on the general

meshed topology like transmission systems [5][9]. From those
methods, the Gauss implicit -matrix method [7] is one of the
most commonly used methods; however, this method does not
explicitly exploit the radial and weakly meshed network structure of distribution systems and, therefore, requires the solution
of a set of equations whose size is proportional to the number
of buses. Recent research proposed some new ideas on how
to deal with the special topological characteristics of distribution systems [10][13], but these ideas require new data format
or some data manipulations. In [10], the authors proposed a
compensation-based technique to solve distribution load flow
problems. Branch power flows rather than branch currents were
later used in the improved version and presented in [11]. Since
the forward/backward sweep technique was adopted in the solution scheme of the compensation-based algorithm, new data
format and search procedure are necessary. Extension of the
method, which emphasized on modeling unbalanced loads and
dispersed generators, was proposed in [12]. In [13], the feederlateral based model was adopted, which required the layer-lateral based data format. One of the main disadvantages of the
compensation-based methods is that new databases have to be
built and maintained. In addition, no direct mathematical relationship between the system status and control variables can be
found, which makes the applications of the compensation-based
algorithm difficult.
The algorithm proposed in this paper is a novel but classic
technique. The only input data of this algorithm is the conventional bus-branch oriented data used by most utilities. The goal
of this paper is to develop a formulation, which takes advantages of the topological characteristics of distribution systems,
and solve the distribution load flow directly. It means that the
time-consuming LU decomposition and forward/backward substitution of the Jacobian matrix or the admittance matrix, required in the traditional Newton Raphson and Gauss implicit
matrix algorithms, are not necessary in the new development. Two developed matrices, the bus-injection to branch-current matrix and the branch-current to bus-voltage matrix, and a
simple matrix multiplication are utilized to obtain load flow solutions. The treatments for weakly meshed distribution systems
are also included in this paper. The proposed method is robust
and very efficient compared to the conventional methods. Test
results demonstrate the feasibility and validity of the proposed
method.
II. UNBALANCED THREE-PHASE LINE MODEL
Fig. 1 shows a three-phase line section between bus and .
The line parameters can be obtained by the method developed by

0885-8977/03$17.00 2003 IEEE

TENG: A DIRECT APPROACH FOR DISTRIBUTION SYSTEM LOAD FLOW SOLUTIONS

883

Fig. 1. Three-phase line section model.

Carson and Lewis [2]. A 4 4 matrix, which takes into account

the self and mutual coupling effects of the unbalanced threephase line section, can be expressed as

Fig. 2. Simple distribution system.

A. Relationship Matrix Developments

(1)

After Krons reduction is applied, the effects of the neutral or

ground wire are still included in this model as shown in (2)
(2)

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

The relationship between bus voltages and branch currents in

Fig. 1 can be expressed as
(6)
Therefore, the relationship between the bus current injections
and branch currents can be expressed as
(3)
For any phases failed to present, the corresponding row and
column in this matrix will contain null-entries.

(7a)

III. ALGORITHM DEVELOPMENT

The proposed method is developed based on two derived matrices, the bus-injection to branch-current matrix and the branchcurrent to bus-voltage matrix, and equivalent current injections.
In this section, the development procedure will be described in
detail.
For distribution networks, the equivalent-current-injectionbased model is more practical [5][13]. For bus , the complex
load
is expressed by
(4)
And the corresponding equivalent current injection at the -th
iteration of solution is
(5)
and
are the bus voltage and equivalent current
where
injection of bus at the -th iteration, respectively. and are
the real and imaginary parts of the equivalent current injection
of bus at the -th iteration, respectively.

Equation (7a) can be expressed in general form as

(7b)
where BIBC is the bus-injection to branch-current (BIBC) matrix.
The constant BIBC matrix is an upper triangular matrix and
contains values of 0 and
only.
The relationship between branch currents and bus voltages as
shown in Fig. 2 can be obtained by (3). For example, the voltages
of bus 2, 3, and 4 are
(8a)
(8b)
(8c)
where is the voltage of bus , and
is the line impedance
between bus and bus .
Substituting (8a) and (8b) into (8c), (8c) can be rewritten as
(9)

884

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 18, NO. 3, JULY 2003

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

can be reduced and the proposed method can be easily integrated

into the existent DA.
C. Solution Technique Developments

(10b)

The BIBC and BCBV matrices are developed based on the

topological structure of distribution systems. The BIBC matrix represents the relationship between bus current injections
and branch currents. The corresponding variations at branch
currents, generated by the variations at bus current injections,
can be calculated directly by the BIBC matrix. The BCBV matrix represents the relationship between branch currents and bus
voltages. The corresponding variations at bus voltages, generated by the variations at branch currents, can be calculated directly by the BCBV matrix. Combining (7b) and (10b), the relationship between bus current injections and bus voltages can
be expressed as

where BCBV is the branch-current to bus-voltage (BCBV) matrix.

(11)

(10a)
Equation (10a) can be rewritten in general form as

B. Building Formulation Development

Observing (7), a building algorithm for BIBC matrix can be
developed as follows:
Step 1) For a distribution system with -branch section and
-bus, the dimension of the BIBC matrix is
.
Step 2) If a line section
is located between bus and
bus , copy the column of the -th bus of the BIBC
matrix to the column of the -th bus and fill a
to
the position of the -th row and the -th bus column.
Step 3) Repeat procedure (2) until all line sections are included in the BIBC matrix. From (10), a building
algorithm for BCBV matrix can be developed as
follows.
Step 4) For a distribution system with -branch section and
-bus, the dimension of the BCBV matrix is
.
Step 5) If a line section
is located between bus and
bus , copy the row of the -th bus of the BCBV
matrix to the row of the -th bus and fill the line
impedance
to the position of the -th bus row
and the -th column.
Step 6) Repeat procedure (5) until all line sections are included in the BCBV matrix.
The algorithm can easily be expanded to a multiphase line
section or bus. For example, if the line section between bus
and bus is a three-phase line section, the corresponding branch
current
will be a 3 1 vector and the
in the BIBC matrix will be a 3 3 identity matrix. Similarly, if the line section
between bus and bus is a three-phase line section, the
in
the BCBV matrix is a 3
impedance matrix as shown in (2).
It can also be seen that the building algorithms of the BIBC
and BCBV matrices are similar. In fact, these two matrices were
built in the same subroutine of our test program. Therefore,
the computation resources needed can be saved. In addition,
the building algorithms are developed based on the traditional
bus-branch oriented database; thus, the data preparation time

And the solution for distribution load flow can be obtained by

solving (12) iteratively
(12a)
(12b)
(12c)
According to the research, the arithmetic operation number
of LU factorization is approximately proportional to
. For a
large value of , the LU factorization will occupy a large portion of the computational time. Therefore, if the LU factorization can be avoided, the load flow method can save tremendous
computational resource. From the solution techniques described
before, the LU decomposition and forward/backward substitution of the Jacobian matrix or the admittance matrix are no
longer necessary for the proposed method. Only the DLF matrix is necessary in solving load flow problem. Therefore, the
proposed method can save considerable computation resources
and this feature makes the proposed method suitable for online
operation.
IV. TREATMENTS FOR WEAKLY MESHED NETWORKS
Some distribution feeders serving high-density load areas
contain loops created by closing normally open tie-switches.
The proposed method introduced before can be extended for
weakly-meshed distribution feeders.
A. Modification for BIBC Matrix
Existence of loops in the system does not affect the bus current injections, but new branches will need to be added to the
system. Fig. 3 shows a simple case with one loop. Taking the
new branch current into account, the current injections of bus 5
and bus 6 will be

(13)

TENG: A DIRECT APPROACH FOR DISTRIBUTION SYSTEM LOAD FLOW SOLUTIONS

885

Combining (16) and (10a), the new BCBV matrix is

(17a)
Fig. 3. Simple distribution system with one loop.

The general form for the modified BCBV matrix is

The BIBC matrix will be
(17b)
(14a)

Equation (14a) can be rewritten as

The building algorithm of Step 5) for the BCBV matrix can be

modified as follows:
Step 5a)If a new branch
makes the system become
meshed, adds a new row to the original BCBV matrix by KVL.
The general form of KVL for a loop can be expressed as
(18)
where
is the number of branches in this loop, and
is the
line impedance corresponding to the branch current .

(14b)
C. Modification for Solution Techniques
And the modified BIBC matrix can be obtained as

Substituting (15) and (17) into (11), (11) can be rewritten as

(15a)

(19)
Applying Krons Reduction to (19), the modified algorithm for
weakly meshed networks can be expressed as

The general form for the modified BIBC matrix is

(15b)

(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.

Note that except for some modifications needed to be done for

the BIBC, BCBV, and DLF matrices, the proposed solution
techniques require no modification; therefore, the proposed
method can obtain the load flow solution for weakly meshed
distribution systems efficiently.

B. Modification for BCBV Matrix

Considering the loop shown in Fig. 3, KVL for this loop can
be written as
(16)

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

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 18, NO. 3, JULY 2003

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

TENG: A DIRECT APPROACH FOR DISTRIBUTION SYSTEM LOAD FLOW SOLUTIONS

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

The Distribution TRX-Power Flow Method

Paulo M. De Oliveira-De Jesus, Member, IEEE
Abstract
This paper proposes a new methodology to assess the power flow solution of distribution networks
by merging the classic Backward/Forward Sweep process in a unique state-of-the-system calculation
formula based on a global matrix denominated TRX. The TRX matrix is suitable to be stored in
memory in the context of Distribution Management Systems (DMS) environment and data exchange
framework supported in standard common information model. The TRX matrix reflects the most
credible state of the network topology. The proposal is suitable to be applied in balanced and unbalanced aerial distribution systems, radially or weakly meshed operated with distributed generation.
This contribution is meaningful under real-time distribution system assessment and planning purposes, since the state of the system is reached using present or historical system measurements as
well as present or future network topology arrangement. Proposed methodology has been applied in
a group of test systems showing better performances than other large-scale implementations like the
standard Newton Raphson with sparse matrix handling
Index Terms
Backward/forward sweep, load flow, power flow, distribution system analysis

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

THE DISTRIBUTION TRX-POWER FLOW METHOD

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

THE DISTRIBUTION TRX-POWER FLOW METHOD

1219

Fig. 1. Branch and node numbering of a radial distribution network

coded in Matlab is presented in Appendix B.

II. Standard Backward/Forward Sweep Power Flow Method in Complex Form
This method was widely described by several authors [14], [24]. The input data of this algorithm
is given by node-branch oriented data used by most utilities. Basic data required is: nodal powers
and sending and receiving nodes of a given line impedance.
In the following, the standard BW/FW sweep power flow method is written in matricial notation
using complex variables. Branch impedances are stated as a vector Z corresponding to a distribution
line model containing a series impedance or transformer. Shunt impedances are not considered in
this first approach. Fig. 1 shows a radial distribution network with n + 1 nodes, and n branches
and a single voltage source at the root node 0. Branches are organized according to an appropriate
numbering scheme (list), which details are provided in [14].
Z=
where,

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

The state of the system is reached solving two steps iteratively.

(6)

1220

THE DISTRIBUTION TRX-POWER FLOW METHOD

A. Step 1 - Backward Sweep

For each iteration k, branch currents are aggregated from loads to origin:
Jk = T I k

(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)

B. Step 2 - Forward Sweep

Nodal voltage vector V is updated from the origin to loads according the Kirchhoff Voltage Laws
(KVL), using previously calculated branch currents vector J 7 branch impedances vector Z:
Vk+1 = V0 TT DZ Jk

(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)

III. The Proposed Methodology

This section describes the proposed power flow approach. Two versions are developed in detail:
a balanced and a three phase unbalanced power flow analysis. In both cases, the general algorithm
flowchart is depicted in Fig. 2.
A. Balanced TRX Power Flow Algorithm
Similar to standard model presented in Section II, the proposed balanced power flow analysis
requires the apparent power injections vector S and distribution line or transformer model based on
series impedances provided from the data exchange model Z = R + jX. Given an initial condition
(k = 0), the nodal voltage n-elements vector V0 are decomposed into real and imaginary parts:
Vx0 =
Vy0 =

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)

THE DISTRIBUTION TRX-POWER FLOW METHOD

1221

Fig. 2. TRX Load Method Flowchart

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)

Eqs. 18 and 19 are settled in matricial form as:



Vxk+1
Vyk+1

Vx0
Vy0

T T DR T
T T DX T

TT DX T
T T DR T



Ikx
Iky

(20)

where the proposed TRX matrix is given by:

TRX =

"

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

THE DISTRIBUTION TRX-POWER FLOW METHOD

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)

B. Unbalanced TRX Power Flow Algorithm

Under the unbalanced approach, nodal power injections vector S are given per phase.
S p1

..

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 =

Z01 ... Zij ... Zmn

(27)

where Zij is the 3-phase matrix impedance corresponding to ij line section [42]:

Z aaij Z abij Z acij

Zij = Z baij Z bbij Z bcij

Z caij Z cbij Z ccij

(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)

THE DISTRIBUTION TRX-POWER FLOW METHOD

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

THE DISTRIBUTION TRX-POWER FLOW METHOD

A. Illustrative Example: Simply 4-node Network

To illustrate the proposed balanced and unbalanced power flow methodology, it is used a 4-node
example shown in Fig. 3. An overhead 12.47kV three-phase distribution line is constructed as shown
in Fig. 4. Length of all sections is 1 mile. The phase conductors are 336,400 26/7 ACSR (Linnet),
and the neutral conductor is 4/0 6/1 ACSR. 336,400 26/7 ACSR: GMR = 0.0244 ft Resistance 0.306
/mile 4/0 6/1 ACSR: GMR = 0.00814 ft. Resistance = 0.5920 /mile. Load demand at nodes 2
and 3 are 2MW with cos = 1.0. The phase impedance matrix of the line is computed according [42]:
Zabc =

"

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

A.1 Balanced TRX Power Flow

Using the following bases SB =10MW and VB =12.47kV, data and results are given in per unit.
Loads areh 0.2 in nodes
2 and 3.h Referencei voltage at node 0 is V 0 = 1 + j0 and initial voltages are
i
set Vx0 = 1 1 1 and Vy0 = 0 0 0 Branches are represented by:

.0296
.0683

Z = R + jX = .0296 + j .0683
.0296
.0683

(38)

Network topology is represented through a 3x3 upper triangular matrix T.

1 1 1

T=
0 1 0
0 0 1

(39)

Then, TRX matrix is built by aggregating TT DR T and TT DX T matrices as indicated in 21:

(40)

(41)

.0296 .0296 .0296

TT DR T = .0296 .0592 .0296

.0296 .0296 .0592
.0683 .0697 .0683

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

THE DISTRIBUTION TRX-POWER FLOW METHOD

1225

Fig. 3. 4-Node Network Topology

Fig. 4. 4-Node Network Three-phase distribution line spacing

A.2 Unbalanced TRX Power Flow

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)

Zabc = Rabc + jXabc

(43)

0.0294 0.0100 0.0098

Rabc = 0.0100 0.0299 0.0100
0.0098 0.0100 0.0296
Xabc

(44)

0.0692 0.0322 0.0247

= 0.0322 0.0672 0.0322
0.0247 0.0322 0.0683

(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

THE DISTRIBUTION TRX-POWER FLOW METHOD

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.

THE DISTRIBUTION TRX-POWER FLOW METHOD

1227

B. Comparative Analysis - Three Test Networks

The proposed methodology was applied in the three (3) distribution test networks widely used
in literature (12-bus [29], 33-bus [40] and 69-bus [41]).
The performance and robustness of the method was assessed under Matlab platform [45] and
compared with two alternative methods. First, the standard BW/FW sweep power flow approach
presented in [24] and, second the Matpowers Newton-Raphson solver [44]. The BW/FW sweep
method has been coded with complex variables using the theoretical basis presented in Section II.
Matpower-NR solver is based on a standard Newtons method [1] using a full Jacobian, updated
at each iteration. Matpower performance is excellent even on very large-scale test cases, since the
algorithms and implementation take advantage of Matlabs built-in sparse matrix handling. In
Appendix B, it is included the the codification of the proposed method. It is important to note
that overall iteration process only require ten programming lines demonstrating the simplicity of
the proposal.
Convergence and robustness of the algorithms are analyzed through the number of iterations
needed to reach a solution and the CPU time spent in the iterative process. Due to small size of
the problems, CPU time registered includes input-output (I/O) access time.
Stop criteria is 106 in all nodal voltages. A Macbook Intel Core 2 Duo T8300@2.4GHz with
2GB RAM under OSX Leopard 10.5.5 has been used for all simulations.
Results are presented in Table V
TABLE V
Comparison between TRX, Standard BW/FW sweep and NR Power Flow

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

THE DISTRIBUTION TRX-POWER FLOW METHOD

Fig. 5. The n-Node Network

Fig. 6. CPU time required to convergence

C. Comparative Analysis - Uniformly Loaded Distribution Feeder

The proposed method has also applied in a large-scale test network and its performance compared
with the BW/FW sweep and the Matpowers Newton-Raphson. Same computer conditions than
previous comparison analysis are used. The test case is a 12.47kV n-node radial distribution network
with a load of 4MW uniformly distributed through L=1mile. Each branch is 336,400 26/7 ACSR
(Linnet) conductors with impedance given in pu (SB =10MW):
0.0293 + j0.697
(48)
n
To illustrate the impact of the number of nodes in the CPU time required by each power flow
method, line section depicted in Fig. 48 has been divided in n branches connected in radial form
and total load has been fractionated in n nodal loads.
It is well known that analytical solution of this type of system is given by [43]:
Z 01 = Z 12 = .. = Z n1n =

%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

THE DISTRIBUTION TRX-POWER FLOW METHOD

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

Three Phase TRX matrix

TRX matrix
Diagonal matrix of branch impedance vector Z
Diagonal matrix of branch resistance vector R
Diagonal matrix of branch reactance vector Z
Convergence criteria
Current vector
Current at node i
Real part of Current vector I
Imaginary part of Current vector I
Three Phase Current vector
Real part of Three Phase Current vector Iabc
Imaginary part of Three Phase Current vector Iabc
Branch Current vector I
Branch current between node i and node j
Real part of Branch Current vector I
Imaginary part of Branch Current vector I
Number of nodes, excluding origin
Active Power Injected vector

1230

THE DISTRIBUTION TRX-POWER FLOW METHOD

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

Reactive Power Injected vector

Active Power Injected at node i
Reactive Power Injected at node i
Active Power Demanded at node i
Reactive Power Demanded at node i
Active Power Generated at node i
Active Power Generated at node i
Branch resistance vector
Three Phase Branch resistance vector
Resistance between node i and node j
Apparent Power Injected vector
Three Phase Apparent Power Injected vector
Apparent Power Injected at node i
Apparent Power Demanded at node i
Apparent Power Generated at node i
Triangular matrix
Three Phase Triangular matrix
Voltage vector
Voltage at node i
Real part of Voltage vector V
Imaginary part of Voltage vector V
Three Phase Voltage vector
Real part of Three Phase Voltage vector Vabc
Imaginary part of Three Phase Voltage vector Vabc
Branch reactance vector
Three Phase Branch reactance vector
Reactance between node i and node j
Branch Impedance vector Z
Three Phase Branch Impedance matrix Z
Branch Impedance between node i and node j
Impedance matrix between node i and node j
Self impedance phase a between node i and node j
Mutual impedance phase ab between node i and node j
B. TRX Algorithm Coded in MATLAB

%-----Calculations Under DSM environment------------T %Triangular matrix (topology) is known

r; x; %List of branch impedances are known
P; Q; % List of Active and reactive powers are known
Vo; %Slack voltage is known
A=T*diag(r)*T; B=T*diag(x)*T;
TRX=vertcat(horzcat(A,-B),horzcat(B,A));
%-----TRX Power Flow Analysis-----------------------Vx=[ones];Vy=[zeros];V=vercat(Vx,Vy)%Initial voltage profile
starttime=cputime;
while max(abs(delta))> 0.0001 %Stop Criteria
for j=1:n %Number of line sections and nodes
I(j)=(-P(j)*V(j)-Q(j)*V(j+n))/(V(j)^2+V(j+n)^2);
I(j+n)=(Q(j)*V(j)-P(j)*V(j+n))/(V(j)^2+V(j+n)^2); end
V2= Vo-TRX*I.; %Voltage Update
delta=V-V2; %Voltage MISMATCH
V=V2;

THE DISTRIBUTION TRX-POWER FLOW METHOD

1231

end %end while

processtime=cputime-starttime;

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

THE DISTRIBUTION TRX-POWER FLOW METHOD

[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.

THE DISTRIBUTION TRX-POWER FLOW METHOD

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

IEEE Transactions on Power Delivery, Vol. 4, No. 1, January 1989

725

OPTIMAL CAPACITOR PLACEMENT ON RADIAL DISTRIBUTION SYSTEMS

Mesut E . Baran

Felix F.

Wu

Department of Electrical Engineering and Computer Sciences

University of Califomia, Berkeley
Berkeley, CA 94720

Abstract - Capacitor placement problem on radial distribution systems is

formulated and a solution algorithm is proposed. The location, type, and
size of capacitors, voltage constraints, and load variations are considered
in the probIem. The objective of capacitor placement is peak power and
energy loss reduction by taking into account the cost of capacitors. The
problem is formulated as a mixed integer programming problem. The
power flows in the system are explicitly represented and the voltage constraints are incorporated. The proposed solution methodology decomposes the problem into a master problem and a slave problem. The master problem is used to determine the location of the capacitors. The
slave problem is used by the master problem to determine the type and
size of the capacitors placed on the system. In solving the slave problem, an efficient phase I - phase II algorithm is used. Proposed solution
methodology has been implemented and the test results are included in
this paper.
I. INTRODUCTION
The general capacitor placement problem consists of determining
the location, type, and the size of capacitors to be installed in the nodes
of a radial distribution system such that the economic benefits due to
peak power and energy loss reduction be weighted against the cost of
installment of such capacitors while keeping the voltage profile of the
system within defined limits.
The optimal capacitor placement problem as defined above has
many parameters, such as; the location, type, and cost of capacitors, voltage constraints, and load variations on the system. These parameters
determine the complexity of the problem.
Conventionally, the problem has been formulated by using a voltage independent reactive current model and solved by fixing some of
the parameters and using analytical methods [l-31. Recently there has
been some studies to solve the problem in its general form. There are
basically three approaches. The first one is the dynamic programming
type approach by treating the sizes of capacitors as discrete variables,
[4-51. The second approach is to combine the conventional analytical
methods with heuristics 16-71. Third approach, pioneered by Grainger
et. al., is to formulate the problem as a nonlinear progmming problem
by treating the capacitor sizes and the locations as continuous variables
[8-131. The application of this approach to general problem with the
voltage regulator problem is given in [ 131.
In this paper, a formulation for the general capacitor placement
problem as a mixed integer programming problem will be given first.
The formulation considers all the parameters of the problem stated above
and also the voltage constraints. Furthermore, the ac power flow formulation is used to represent the power flows and the voltage profile in the
radial distribution systems.
A solution methodology for the general problem is proposed in this

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.

paper. The solution is based on the decomposition of the problem into

hierarchical levels. The problem at the top level, called the muster problem, is an integer programming problem and is used to place the capacitors ( i.e. U) determine the number and the location of the capacitors).
A search algorithm has been developed for the master problem. The
problem at the bottom level is called the slave problem and is used by
the master problem to determine the types and the settings of the capacitors placed. Decomposition schemes are also used to further decompose
the slave problems into base problems. Base problems are solved by
using the efficient solution algorithm developed for a special capacitor
placement problem called the sizing problem. The sizing problem and
the associated solution algorithm are presented in another paper [141.
This paper consists of seven sections. In section 11, the general
capacitor placement problem is formulated and its complexity is discussed. h section 111, it is shown that the problem can be decomposed
into a muster problem and a slave problem. Section IV and V are
devoted to the development of solution methodologies for the slave and
master problems respectively. Section VI contains the test studies of the
solution method applied to two different systems. Conclusions are given
in section VII.

IL. FORMULATION OF THE PROBLEM

We consider the general capacitor placement problem as determining the places (number and location), types, and settings (capacities) of
the capacitors to be placed on a radial distribution system. The objectives are to reduce the power and energy losses on the system and to
maintain the voltage regulation while keeping the cost of capacitor addition at a minimum.
Since we are interested in energy loss in the system, it is necessary
to take into account the load variations for a given period of time, T.
We assume that the load variations can be approximated in discrete levels. Furthermore, the loads are assumed to vary in a conforming way
(i.e., all the loads enjoy the same pattem of variations). We let S(T) be
the common Loud Duration Curve as shown in Fig.1. Then a load, say
load QL , can be represented as

:F+>;
QL@) =

Q~W)

(1)

Where, Q2represents the peak value.

S"t

y 4;"qTIFigure 1 : Load Duration Curve

Under these assumptions, the time period, T can be divided into intervals
during which the load profile of the system is assumed to be constant.
Let there be nt such loud levels (loud projiles) .
Then for each load level, we have: (i) power flow equations, (ii)
voltage constraints as bounds on the magnitude of the system node voltages. (iii) capacity and control constraints on the control variables
(capacitors).
We will represent the constraints imposed by power flows on a
radial distribution system by a new set of ac power flow equations,
called DistFkw equations. They substitute for the conventional ac
power flow equations. To summarize the idea, consider a 3-0, balanced
radial distribution feeder with n branches/nodes, I laterals, and nc shunt

0885-8977/89/01oo-O725$01.WO 1989 IEEE

Authorized licensed use limited to: UNIVERSIDADE DO PORTO. Downloaded on October 28, 2009 at 23:22 from IEEE Xplore. Restrictions apply.

726

capacitors placed at the nodes of the system. In Fig.2, one-line diagram

of such a network is shown.

These bounds constitute a set of functional inequality constraints of the

form,
H'(x')IO

i =0,1,. . . .nt

(7)

We will consider two different types of capacitors and represent

them as follows:
i) Fixed Capacitors : They will be treated as reactive power sources with
the constant magnitude at al load levels, i.e..
"o=ul=

Fig.2 : One line diagram of a Distribution Feeder

It can be shown that power flow through each branch in the lateral can
be described by the following recursive equations.

... =uM

(8.i)

ii) Switched Capacitors: It will be assumed that the settings (capacities)

of a switched capacitor, U: can be changed/contmlled at every load level
qnsidered. Therefore, for each capacitor, there are nt+l settings,
U: i = 0,l. . . . ,nt to be determined. We will also assume that the setting
of a capacitor for the peak load, U: will be bigger than the ones for other
lower load levels U:. Hence, the sizes (nominal capacities) of capacitols
will be determined by uo and the relationship between the size and and the
settings of a capacitor will be as follows.
(8.ii)

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

branch k+l connecting node k and node k+l,

v k :bus voltage magnitude at node k ,
Qck:reactive power injection from capacitor at node k.
Eq.(2), called the brnnchjbw equation, has the following form
4+1

= fk+l(xk

nl

ke

(3.i)

&k+d

Where, X k = [ p k Qk v?lT and 4 + 1 = e&+,.

Note that if there is no capacitor at node k, then uk does not appear in
Eq(3.i). By abusing notation, we will simply use U as an nc dimensional
vector containing the nc capacitors i.e.,
UT

= [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)

In addition to the branch flow equations of (3.i). there are terminal

conditions to be satisfied for each lateral (counting the main feeder as the
O'th lateral). For example, for lateral k shown in the figure, we have the
following terminal conditions:
(i) at the branching node k where the lateral is connected to the main
feeder, we define a dummy variable VkO and let
XkO,

= 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)

Where, x = [ x f . . . x[$lT ,xk = [ x & . . . x ; l T .

DistFlow equations can be used to determine the operating point, x
of the system for a given load profile, PL;,Q L ~ i = 1 , . . . , n and the
capacitor settings, U . We prefer to use DistFlow equations over conventional ac power flow equations because the special structure of the DistFlow equations can be utilized to develop a computationally efficient and
numerically rbbust solution algorithm. The details of such a solution algorithm are presented in [ 141.
For the capacitor placement problem since there are nt different load
profiles to be considered, the overall DistFlow equations are

Figure 3 : Capacitor cost functions

Such a function is not easy to handle within this formulation framework;
therefore, it will be approximated by a linear function with a fixed charge
as shown in Fig3 by a dotted line. This function can be formulated by
using a decision variable e E (0.1) as
f ( u o ) = c . e +rc.uo

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

G' (xi ,d)

=0

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.

III. DECOMPOSITION OF THE PROBLEM

The problem formulated in the previous section is a non-liiear,
mixed integer programming problem of the following form,
min {f,(e.u)
e.u

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)

base problems can be solved by the efficient solution method developed

for the sizing problem, PS in [14]. Finally, in the last subsection. we will
come back to the general slave problem and consider the types as well as
the settings of the capacitors.
4.1 Fixed Capacitor Problem
When all the capacitors to be placed are of fixed type, the problem
becomes
nl

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-

This general form of fixed capacitor problem, Pfx is a base problem

because it is essentially the same as the sizing problem, PS. Here, because
of the consideration of more load levels, the objective function comprises
more power loss terms and the constraint set is bigger due to extra DistFlow equations and the corresponding voltage regulation constraints. But
this does not change the structure of the problem very much. Therefore,
the solution methodology developed for the sizing problem can readily be
applied for this problem with small modification.
Note that for this general case, existence of different load levels
makes it possible to have both lower bound voltage violation at peak load
level and upper bound voltage violation at light load level for a given control U . This is especially true when the load levels are diverse. Such
cases are handled in the Phase I - Phase I1 type solution algorithm of the
sizing problem by considering both the lower and the upper voltage constraints in calculating the search direction (for details, see the solution
algorithm for the base problem in [ 141).
4.2. Switched Capacitor Problem
When all the capacitors to be placed are of the switched type. the
slave problem can be re-written as follows.

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)

and is called the musterproblem. These problems can be characterized as,

Master Problem (MP) :Integer Programming Problem
Slave Problem (SP) :Non-linear Differentiable Optimization Problem
The solution for this decomposed problem requires an efficient solution algorithm for the slave problem and a search procedure over E , the set
defined by all the possible decisions, for the master problem. In the next
two sections. we'll discuss and develop such solution schemes.

IV. SLAVE PROBLEM

As indicated in the previous section, the slave problem assumes that
the capacitors are placed and it is a special case of the capacitor placement
problem. The problem is a non-linear differentiable optimization problem
with quite a large number of equality and inequality constraints. It is still
not easy to solve the slave problem. Our aim here is to study the special
features and the structure of the problcm and to develop an efficient solution methodology by exploiting these features.
For the slave problem, the type of capacitors (fixed or switched), and
the number and diversity of load levels are important parameters that
determine the structure and the size of the problem. Therefore, we first
relax the parameter "type" by assuming that they are given. Then the
problem becomes a sizing problem (i.e., determining the capacitor settings
) and we need to consider two cases; one with fixed type capacitors and
one with switched type capacitors. In the following first two subsections,
we will show that for these two cases the problcm is either of the base type
problem or it can be decomposed into the base type subproblems. The

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

fi(U? = min k, Tipi(xi)

S.t

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)

We simply use 0' in evaluating f i@) and Vfi @) as follows:

f;(if)
= k, Tipi (2' ,tii)

fixed capacitors. Therefore, we propose the following heuristic selection

scheme by using the solution methodology developed for the switched
capacitor problem.
Step 0 :Assume all the capacitors are of the switched type.
Step 1 :Solve the problem considering only the switched capacitors and keeping the fixed capacitor power injections as constant at their nominal settings, i.e., as loads.
Step 2 : Using the results of step 1, check the settings of
switched capacitors at light load level (level nr ). For the ones
with nonzero light load settings, (i.e., u t # 0 ) assign that portion of capacitor as fixed type.
Step 3 :If any h x i g occurred, then go to step 1; otherwise,
stop.
V. MASTER PROBLEM
We follow a general approach in solving the master problem, which
is an integer programming problem, and first construct a decision graph
and then develop an efficient search scheme to place the capacitors (i.e., to
determine their numbers and places).
Let there be nc initially given candidate capacitors. Then, all possible decisions about choosing the capacitors for placement among the candidate capacitors can be m g e d as a decision graph assuming one decision is made at a time. Such a graph is shown in Fig5 for nc = 3.

(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

Figure 5 :Decision graph for 3 variable case

In the figure, each node represents a particular decision,
e = [el e2 e 3 IT ; ei E { O J J . Where, ei = 0 means that the capacitor is
not chosen, and ei = 1 means it is still a candidate. A branch f" a node
to another indicates how the transition can be achieved: taking out the candidate capacitor whose number is shown on the branch. Such a relationship is indicated in graph theory by calling a node and all the nodes
emanating from it. aparent and its children, respectively.
Search starts from the root which corresponds to thc case where all
candidate capacitors are chosen to be placed at the designated nodes of the
system. Then there are two possible search techniques, depth-frrstsearch
and breudth-first search, that can be employed to get a local optimum
[17]. Here, rather than employing these general search techniques directly,
the children of a given nodebarent) are sorted first according to their contribution to the objective. We propose a sorting procedure which works
sort of like a "steepest descent" approach in discrete case. The procedure
is

1
1
converged

Figure 4 :Block diagram of SW Capacitor Algorithm

- 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

new objective, f:(u) (This requires only DistFlow

solutions).
- let the contribution of the capacitor k to the objective be
A f =f,"-f 0 .
- sort the existing capacitors by using the Af ,"s

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

Table 3: Test run results for TS1 - fixed capacitor placement

System
TS1.
TS2

Table 1 : Network Data of TS2

- B r . P a r . - Rv.

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.

Although there are regulators at the substations, voltage profiles of

both systems are below lower voltage limits at the peak load level
before the capacitor placement. The solution for TSl indicates that
the capacitors are needed to be used for raising the voltage profile of
the system at the peak load level ( Vmh = 0.9 P.u.) as well as for loss
reduction. The solution for TS2 corresponds to the unconstrainted
optimal point ( Vmh=0.9O7 P.u.); indicating that maximum loss
reduction is achieved.
2. Switched Capacitor Case

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

Table 4: Test run results for TS2 - Fixed Capacitor placement

We have the followmg comments/observations about these test
RSults.
1. Search Procedure :
0 The order of search is about nc (number of capacitors placed);
which is much less than the worst case bound of nc '.
The search converges to the global optimal point in both tests;
although in test 1, for example, there are some other suboptimal
solutions with revenues close to each other.
0 As search goes further down to higher levels, more capacitors are
taken out to increase the revenues. This causes less energy loss
reduction and lower voltage profile. This also shows the sensitivity
of the optimal point with respect to the cost figures k, ,r, ,c.
2. The slave problem fixed capacitor problem:
0 The convergence of slave problem at the root (the initial one) gets
slower as the number of load levels increase (the number of iterations is in the order of 2-3 nc ). But the other slave problems converge much faster (in the order of 1-2 nc ); mainly because it is
easier to find a good initial point for them by simply using the results
of the parent node.

rev.

AE

vmi,

39180
785
.go9

QZ

Q,'

Q,"

197
620

330
1240

330
1700

bus
18
52

We have the following comments about the test results:

0 The performed test runs indicate very good convergence characteristics; the number of iterations between the main problem, SW,
and the subproblems, SWi is usually one or two.
0 When the fixed capacitor and the switched capacitor test results are
compared, it is Seen that: (i) switched capacitor placement yields
Egher revenues and higher capacitor sizes, especially when the cost
data is the same for both cases and the load variations are diverse.
(ii) the voltage profile is higher ( Vmh is higher) and voltage regulation is better for the switched capacitor case. This is because of the
fact that for the switched capacitor case better compensation is
achieved by adjusting the value of capacitors as the load changes.
The final point to be noted about the overall test results is about the
effect of the regulators on the solution; (i) feasibility becomes less of a
problem, (ii) no upper limit voltage violation has been observed for the
given test systems, although the load levels were quite diverse. (iii) as
exemplified here by test run for TSl, capacitors can be used together with
voltage regulators to keep the voltage profile of the system within defined
limits.

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-

dology have been developed for general capacitor placement problem on

radial distribution systems.
The general capacitor placement problem consists of placing the
capacitors (determining their number and the locations) and determining
their types and sizes. The objective is peak power and energy loss reduction while keeping the cost of capacitors at a minimum.
The proposed formulation is comprehensive in the sense that: (i) it
considers all the variables of the problem stated above, (ii) it uses the ac
power flow equations to represent the system, (iii) voltage constraints are
taken into account.
A solution method has been developed for this general problem by
decomposing the problem into two hierarchical levels. The top level problem. called the master problem, is an integer programming problem and is
used to place the capacitors (determine their number and locations). An
efficient search scheme has been developed for the master problem. The
second level problem, called the slave problem, is used by the master
problem as a subroutine. This problem is further decomposed into two
levels: at the top level, the problem consists of determining the type of
capacitors and at the bottom level, the problem is to determine the capacitor sizes once the capacitors are placed with their types assigned. These
slave problems, called the fixed and switched capacitor problems, are
shown to be either the base type or can be decomposed into base type
problems. The base type problem is a capacitor sizing problem and is
solved by using an efficient phase I - Phase I1 type solution method
presented in [ 141.
Test results are presented for the proposed solution scheme. They
indicate. that the method is computationally efficient and the decomposition scheme performs well.
Although not implemented, it is also shown that the formulation and
the solution methodology presented in this paper can be generalized to
include the voltage regulators in the problem.
Acknowledgements
We thank Mr. Wayne Hong and Dr. Dariush Shirmohammadi of
Pacific Gas and Electric for their helpful discussions. This research is supported by TUBITAK-TURKEY and by National Science Foundation
under grant ECS-8715132.

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.

M. Ponnavaikko and K. S. Prakasa Rao, "Optimal Choice of Fixed

and Switched Shunt Capacitors on Radial Distribution Feeders by
the Method of Local Variations", IEEE Trans. on Power Apparatus
andsystem, vol. 102, pp.1607-1614, June 1983.
T. H. Fawzi. S. M. El-Sobki, and M. A. Abdel-Halim, "A New
Approach for the Application of Shunt Capacitors to the Primary
Distribution feeders", IEEE Trans. on Power Apparatus and Systems, vol. 102, pp.10-13, Jan. 1983.
M. Kaplan, "Optimization of Number, Location, Size, Control Type,
and Control Setting of Shunt Capacitors on Radial Distribution
Feeders", IEEE Trans. on PAS, vol. 103, pp.2659-2665, Sept. 1984.
J. J. Grainger, and S. H. Lee, "Optimum Size and Location of Shunt
Capacitors for Reduction of Losses on Distribution Feeders", IEEE
Trans. on Paver Apparatus and Systems, vol. 100, pp. 1105-1118,
March 1981.
J. J. Grainger, and S. H. Lee, "Capacity Release by Shunt Capacitor
P h x t w U on Distribution Feeders: a New Voltage Dependent
Model", IEEE Trans. on Power Apparatus and Systems. vol. 101.
pp. 1236-1244, May 1982.
J. J. Grainger, S.H. Lee, and A. A. El-Kib, "Design of a Real-Time
Switching Control Scheme for Capacitive Compensation of Distribution Feeders", IEEE Trans. on Power Apparatus and Systems, vol.
101, pp. 2420-2428, August 1982.
J. J. Grainger, S. Civanlar, and K. N. Clinard, L. J. Gale, "Optimal
Voltage Dependent Continuous Time Control of Reactive Power on
Primary Distribution Feeders", IEEE Trans. on Power Apparatus
andsystem, vol. 103, pp. 2714-2723, Sept. 1984.
A. A. El-Kib, J. J. Grainger, and K. N. Clinard, L. J. Gale, "Placement of Fixed and/or Non-Simultaneously Switched Capacitors on
Unbalanced Three-phase Feeders Involving Laterals", IEEE Trans.
on Paver Apparatus and Systems, vol. 104, pp. 3298-3305, Nov.
1985.
S. Civanlar and J. J. Grainger, "Volt/Var Control on Distribution
Systems wilh Lateral Branches Using Shunt Capacitors and Voltage
Regulators: Part I, Part 11, Part III", IEEE Trans. on Power
Apparatus andsystems ,vol. 104, pp. 3278-3297, Nov. 1985.
M. E. Baran and F. F. Wu, "Optimal Sizing of Capacitors Placed on
a Radial Distribution System".
submitted to IEEE PES winter
meeting, 1988.
A. M. Geoffrion, "Elements of Large Scale Mathematical Programming, Part I: Concepts", Management Science, vol. 10, pp. 652-675.
July 1970.
A. M. Geoffrion, "Generalized Benders Decomposition", JOTA, vol.
10, pp. 237-260, April 1972.
E. M. Reingold and W. J. Hansen. Data Structures. Little Brown and
Comp., 1983.

Appendix: Decomposition of the Switched Capacitor Problem

We shall apply decomposition techniques to the switched capacitor
problem introduced in section 4.2. We present first a general scheme,
which leads to the Benders Decomposition, and then a simpler, heuristic
based decomposition scheme for the special switched capacitor problem.
Decomposition
A general decomposition scheme is given in [15]. To adopt the
derivation for the switched capacitor problem, Psw , we first re-group the
variables and constraints as follows,
nf

U 1

U = { U I Eq.(16) is satisfied )

U o ={ uo I Eq.(15) is satisfied )

Similarly, we partition the objective function as,

krCI

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,

Where, F(uo,u) comsponds to the coupling constraints of Eq.(17).

We start the decomposition by projecting (purtitioning) the problem
onto space of uo alone as follows.
(a.2)
We assume that the subproblem in the braces,
=

f s (U?

:$

i f r (u)-F(uo*u) 5 0 1

can be evaluated for a given 'U as an optimization problem with respect to

the variable U. Then the main problem becomes
W

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 solution algorithm involves the iterative solution of the main

problem sw& and the subproblems, SW;.After each iteration, a new set
of constraints of the type (a.5.i) is added to Swab until the solution converges.
It is noted that the Benders decomposition has the following features.
(i) The main problem, SW, is not a base type problem. Additional constraints of (a.5.9, called the cuts, are difficult to handle with the solution
algorithm used by the bask problem; one has to identify which one these
constraints will be binding during the solution of web.
(ii) The contribution of the subproblems to the objective of the main problem, fJi(u? is approximated by linearizing this term around the previously calculated points, uoJ. To see this, let the binding constraint for the
subproblem i be the k'th one in Eq.(aS.i). Then the solution for the
correspondingyi will be

I 5 UEU s.t. F(u0,b)I O }

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.

Therefore, the approximation is good only if the actual solution point,'U is

close to the calculated point U* in solving Sw&.
A Heuristic Based Decomposition Scheme
Note that the main problem of (a.3) need not be transcribed into
Benders form, SW, if One can estimate the binding constraints in the sets
U' -'U L; 0. We develop a solution algorithm based on this principle. The
solution scheme uses the sizing algorithm to solve both the main problem
and the subproblems and it also uses a better estimate for the contribution
of the subproblems to the objective of the main problem. The details of
the algorithm is given in section 4.2.

Justification of the Conjecture

U, c

Figure a.1 : Set hOjeCtiOn

Now we generalize the projection by rewriting the problem (a.3) as follows.

min { f,,,(u") +f,(u")

m0

U'E

nv 1

(a.4)

U0

For the switched capacitorproblem, we have a conjecture that

U, c v.
Justification of this conjecture will be discussed later.
Assuming that this conjecture holds for the general case, we can use
the partitioned problem (a.3) rather than (a.4) for the main problem, SW,.
The subproblem f,(U? can further be decomposed into nt subproblems of
the following form.
f s i (U"

The idea behind this conjecture is as follows. Let uo E U,. This

means that 'U amount of reactive power compensation from the capacitors
is enough to satisfy the voltage constraints for the peak load. But this
amount of compensbtion must suffice to have a feasible point for lower
load levels too because the lower the load the higher the voltage profile
will be. This observation is due to the strong coupling between the reactive power flow and the voltage profile of the system.
Note that this conjecture was also the underlining idea behind the
assumption made when formulating the problem in section 2; where, it
was assumed that the capacitor size will be the capacitor setting for the
peak load level, uo and capacitor settings for all the other lower load levels, U' will be smaller than.'U

= 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

fo(u'.Y) =f d (U? + Cyi

s.t.

yi + 3 c " r ~ u o - u o ~ ~ ~ ~j ,=i I~, .u.o. .~p ~ (a.5.i)

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.

1) Select the fixed capacitors. This frequently is simply the selection of

the optimal location of the one capacitor bank needed at minimum

load.
2) Increment the load in steps. At each load level ask the question, Is
more capacitance needed? If so, then determine optimal switched
capacitor locations and sizes based primarily on loss and voltage
profile improvement.
3) Once peak load is reached, decrement the load to determine switch-off
load levels.
This very simplistic method of locating and sizing capacitors incrementally tends to distribute smaller size banks over the feeder where they are
needed and when they are needed. Generally, no more than one capacitor is
added at a load level and the search for optimum location is trivial. My
simulations of the feeder over a years time considering daily and seasonal
load cycles indicate that it is difficult to significantly improve upon this
method economically. One can always find a solution that appears to be a
few percentage points better, but I do not think that the basic data are known
with sufficient accuracy to quibble over a small difference in an off-line
analysis. It would require on-line control to take advantage of the small
gains possible. We have implemented this method in an interactive program
that uses the above algorithm to get close to the most economical solution.
Then the user can tweek the solution interactively taking into account
practical considerations. The whole process takes but a few minutes using a
personal computer (E).
Differing approaches to the problem of capacitor size and placement will
yield differing optimal solutions. 1 suspect that none are truly optimal
and most approaches that consider load variation are generally adequate
(methods that optimize only peak load sometimes give poor results when the
entire load cycle is considered). I would hesitate to defend one approach
over another too strongly because feeder load varies somewhat randomly
and it would be difficult to prove which is more optimal. However, I think
that methods like I have described, which are based on how the feeder
operates, are apt to be more optimal more of the time. They are also simple
to program and the programs execute quickly. Therefore, I question the
practicality of abandoning the simpler approach in favor of a more
sophisticated method like the authors have presented.
Reference
[l] T. Frantz et al., Load behavior observed in LILCO and RG&E

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

This article was downloaded by: [B-on Consortium - 2007]

On: 29 October 2009
Access details: Access Details: [subscription number 908038079]
Publisher Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

Electric Power Components and Systems

Publication details, including instructions for authors and subscription information:
http://www.informaworld.com/smpp/title~content=t713399721

Simple and Efficient Computer Algorithm to Solve Radial Distribution Networks

Rakesh Ranjan a; D. Das b
a
Multi Media University, Ayer Keroh Lama, Melaka, Malaysia. b Electrical Engineering Department, Indian
Institute of Technology, Kharagpur, India.
Online Publication Date: 01 January 2003

To cite this Article Ranjan, Rakesh and Das, D.(2003)'Simple and Efficient Computer Algorithm to Solve Radial Distribution

Networks',Electric Power Components and Systems,31:1,95 107

To link to this Article: DOI: 10.1080/15325000390112099
URL: http://dx.doi.org/10.1080/15325000390112099

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf
This article may be used for research, teaching and private study purposes. Any substantial or
systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or
distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses
should be independently verified with primary sources. The publisher shall not be liable for any loss,
actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly
or indirectly in connection with or arising out of the use of this material.

EMP 31(1) #2689

Electric Power Components and Systems, 31:95107, 2003

c 2003 Taylor & Francis
Copyright

1532-5008/03 $12.00 + .00
DOI: 10.1080/15325000390112099

Simple and Ecient Computer Algorithm to

Solve Radial Distribution Networks
RAKESH RANJAN

Downloaded By: [B-on Consortium - 2007] At: 03:04 29 October 2009

Multi Media University

Ayer Keroh Lama
Melaka, Malaysia

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

radial distribution networks, load ow, circuit model

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)

total number of the load

total number of the branch (LN 1 = NB 1)
real power load of ith node
reactive power load of ith node
voltage magnitude of ith node
resistance of the branchjj
reactance of the branchjj
impedance of the branchjj
current owing through branchjj
total reactive power load fed through node m2
total reactive power load fed through node m2
voltage angle of the node m2
reactive power loss of branchjj
reactive power loss of branchjj

Manuscript received in nal form on 16 January 2002.

Address correspondence to R. Ranjan. E-mail: rakesh.ranjan@mmu.edu.my

95

96

R. Ranjan and D. Das

IS (jj)
IR(jj)
PLOSS
QLOSS

sending end node of branchjj

receiving end node of branchjj
total reactive power loss
total reactive power loss

Downloaded By: [B-on Consortium - 2007] At: 03:04 29 October 2009

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.

Algorithm for Radial Distribution Networks

97

Figure 1. Sample radial distribution networks.

Downloaded By: [B-on Consortium - 2007] At: 03:04 29 October 2009

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.

3. Mathematical Model of Radial Distribution Networks

A mathematical model of radial distribution networks can easily be derived from
Figure 2.
I(jj) =

|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.

Figure 2. Electrical equivalent of one branch of Figure 1.

98

R. Ranjan and D. Das

Downloaded By: [B-on Consortium - 2007] At: 03:04 29 October 2009

Table 1
Branch number, sending end node, and receiving end
nodes of Figure 1
Branch no.
(jj)

Sending end node

ISS (jj)

Receiving end node

IRR(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

From equations (1) and (2) we get

|V (m2)| =
where

{B(jj) A(jj)}

A(jj) = P (m2) R(jj) + Q(m2) X(jj) 0.5 |V (m1)|2

B(jj) = {A2 (jj) {Z 2 (jj) (P 2 (m2) + Q2 (m2))}

(3)

(4)
(5)

Since the substation voltage magnitude |V (1)| is known, for jj = 1, (see

Table 1), m1 = ISS (jj) = ISS (1) = 1 and m2 = IRR(jj) = IRR(1) = 2.
A(jj) = A(1) and B(jj) = B(1) can be computed using equations (4) and (5), if we
know P (m2) = P (2) and Q(m2) = Q(2). After that |V (2)| can easily be computed
from equation (3). Similarly for jj = 2, (see Table 1), m1 = ISS (jj) = ISS (2) = 2,
m2 = IRR(jj) = IRR(2) = 3. A(jj) = A(2), B(jj) = B(2) can be computed
using equations (4) and (5), if we know P (m2) = P (3) and Q(m2) = Q(3) and
hence |V (3)| can easily be computed by using equation (3). In general, for jj =
1, 2, . . . , LN 1, voltage magnitude of all the nodes can easily be computed by using
equations (4), (5), and (3), if we know P (m2) and Q(m2) for m2 = 2, 3, . . . , NB .
Computation of P (m2) and Q(m2) (m2 = 2, 3, . . . , NB ) is explained in section 4.
Real and reactive power losses in the branch jj are
LP (jj) =

R(jj) (P 2 (m2) + Q2 (m2))

|V (m2)|2

(6)

LQ(jj) =

X(jj) (P 2 (m2) + Q2 (m2))

|V (m2)|2

(7)

4. Computation of P (m2) and Q(m2)

A computer logic is presented in this section for automatic computation of P (m2)
and Q(m2). For the requirements of the computer logic presented, in Table 1, branch

Algorithm for Radial Distribution Networks

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)

Downloaded By: [B-on Consortium - 2007] At: 03:04 29 October 2009

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

R. Ranjan and D. Das

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)

Downloaded By: [B-on Consortium - 2007] At: 03:04 29 October 2009

= QL(3) + QL(4) + QL(6) + LQ(3) + LQ(8)

Proposed computer logic will again check whether nodes 4 and 6 are connected
with any other node. It is seen that (Table 1) node 4 is connected with node 5 and
node 6 is connected with node 7, i.e., i = 4, IR(4) = 5 and i = 9, IR(9) = 7.
Therefore,
P (3) = P (3) + PL(5) + PL(7) + LP (4) + LP (9)
= PL(3) + PL(4) + PL(6) + LP (3) + LP (8)
+ PL(5) + PL(7) + LP (4) + LP (9)
and
Q(3) = Q(3) + QL(5) + QL(7) + LQ(4) + LQ(9)
= QL(3) + QL(4) + QL(6) + LQ(3) + LQ(8)
+ QL(5) + QL(7) + LQ(4) + LQ(9)
Exact computation of P (3) and Q(3) is complete because nodes 5 and 7 are not
connected with any other node. Similarly, computer logic will compute exact load
fed through each node. It is to be noted here that if the receiving end node of any
branch is an end node, then total load fed through that node is the load of node
itself. For example, consider node 5 of Figure 1, which is an end node. Therefore,
P (5) = PL(5) and Q(5) = QL(5). The proposed algorithm will also identify the
end nodes. It is worth reporting that node numbering of the feeder is arbitrary and
the algorithm will automatically compute the exact load fed through all the nodes.
Initially, if LP (jj) and LQ(jj) are set to zero for all jj, then the initial estimate
of P (m2) and Q(m2) (m2 = 2, 3, . . . , NB ) will be the sum of the loads of all nodes
beyond node m2 plus the load of m2 itself.

5. Load Flow Computation of RDN

Once P (m2) and Q(m2) are computed by the above logic (section 4), voltage
magnitudes of all the nodes can be easily computed using equation (3). Further real
and reactive power losses are obtained using equations (6) and (7). The complete
load ow algorithm is shown in the form of a ow chart in Figure 3.
The convergence criterion of the algorithm is that if, in successive iteration,
the dierence of real and reactive power delivered from the substation is less than
0.1kW and 0.1kVAr, respectively, the solution has converged.

Downloaded By: [B-on Consortium - 2007] At: 03:04 29 October 2009

Algorithm for Radial Distribution Networks

(a)
Figure 3. (a) Flow chart of load ow technique (continues).

101

R. Ranjan and D. Das

Downloaded By: [B-on Consortium - 2007] At: 03:04 29 October 2009

102

(b)
Figure 3. (b) Flow chart of load ow technique.

Algorithm for Radial Distribution Networks

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.

Downloaded By: [B-on Consortium - 2007] At: 03:04 29 October 2009

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

R. Ranjan and D. Das

Downloaded By: [B-on Consortium - 2007] At: 03:04 29 October 2009

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

Total real power loss = 224.9606 kW; total reactive power

loss = 102.147 kVAr; minimum voltage observed at node 65,
|V65 | = 0.90919 p.u.

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

Total real power loss = 210.998 kW; total reactive power

loss = 143.032 kVAr; minimum voltage observed at node 18,
|V18 | = 0.90377 p.u.

Algorithm for Radial Distribution Networks

105

Table 4
Comparison of relative speed and memory
Method

Relative memory

Relative CPU time

Proposed method
D. Das et al. [12]
Baran and Wu [9]

1
4
6

1.0
1.5
2.9

Downloaded By: [B-on Consortium - 2007] At: 03:04 29 October 2009

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.

Downloaded By: [B-on Consortium - 2007] At: 03:04 29 October 2009

106

R. Ranjan and D. Das

[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

Figure 4. 69-node RDN [9].

Algorithm for Radial Distribution Networks

107

A.2. Data for 33-Node Test Systems [14]

Downloaded By: [B-on Consortium - 2007] At: 03:04 29 October 2009

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