C C I I R R E E D D 18

th
International Conference on Electricity Distribution Turin, 6-9 June 2005



DISTRIBUTION TECHNICAL LOSS CALCULATION AND COSTING FOR NETWORK PLANNING AND
DESIGN

C.G. Carter-Brown

Eskom Distribution
South Africa
cartercg@eskom.co.za

S.W. Heunis
Heunis Solutions
South Africa
schalkh@icon.co.za
C.T. Gaunt

University of Cape Town
South Africa
ctg@eng.uct.ac.za

INTRODUCTION

Network plans and designs in Eskom Distribution (the
government owned utility in South Africa) have primarily
been based on initial capital cost considerations. Excessive
generation capacity during the past fifteen years resulted in
low cost electricity and the evaluation and costing of
technical losses was not a priority. The restructuring of the
South African ESI and a forecasted medium term (3 years)
generation shortage are drivers to incorporate technical
loss considerations into distribution network planning and
design. A methodology is required for the calculation and
costing of energy losses in unbalanced distribution networks
supplying stochastic loads.

PRESENT PRACTICES

Loss Load Factor

The Loss Load Factor (LLF) is extensively used [1,2,3] to
calculate the average distribution loss based on the maximum
loss (at peak load). The accuracy and suitability of the LLF
methodology is limited as it is assumed that the loads are
homogenous and vary uniformly at all busbars. The
stochastic nature of loads is not directly accounted for and
the time -of-use (TOU) characteristic of the loss is not known.

Statistical methods for energy loss calculation

The errors in and limitations of the LLF approach arise due to
the stochastic nature of distribution network loads. Herman
and Heunis recognized that "To calculate the resistive losses
in a LV feeder over a period of time, the loads are best
approached as stochastic current signals" [4]. The
stochastic nature of the loads on LV feeders has the largest
impact on expected losses. By describing the load using
statistical parameters (mean, standard deviation, variance and
coefficient of correlation) energy losses can be accurately
calculated over time [4].

Load profiles

Consumer load profiles are used for loss calculation [1,3,5].
However, the loss calculations are based on the mean load
during each time interval. The load variance within time
intervals is ignored, and the errors may be large with
stochastic loads such as in LV networks.
Cost of energy losses

The majority of loss evaluations differentiate between no-load
and load losses and no-load and load loss costs are typically
based on base-load and peak-load generation costs
respectively [2]. The major limitation of this approach is that
the TOU characteristic of the load loss energy is not
considered. Generation costs vary significantly between peak
and off-peak periods. The use of average loss costs may
result in large errors as the load shapes in individual
distribution networks can vary significantly from that of the
total generation system.

PROPOSED METHODOLOGY

Methodology overview

The proposed methodology to overcome the limitations of the
present practices combines the advantages and accuracy of
statistical calculation methods and load profiles, and is
summarised in figure 1.

Fixed loads are assumed constant over the entire lifetime.
Variable loads are described by daily load profiles with the
mean, variance and correlation of load currents between
consumers in each time internal. The TOU statistical
parameters of the load current in each phase/neutral of each
branch are calculated using the fixed loads, consumer
connections and consumer daily profiles. The energy losses
in each TOU period are calculated, and combined with TOU
loss tariff costs to obtain the energy loss cost. The energy
losses and costs in year 1 are adjusted by the consumer load
forecasts and financial forecasts to obtain the total energy
loss and loss cost over the evaluation period.

Calculations and assumptions

Annex A provides details of the loss calculations within the
following process:
Loads are assumed to be constant current, (load current
is independent of changes in voltage).
Load profiles differentiate between weekdays, Saturdays
and Sundays. Seasonal load variations are described by
monthly load profiles (separate profiles for each month)
or annual daily load profiles with adjustment based on
monthly energy consumption.
C C I I R R E E D D 18
th
International Conference on Electricity Distribution Turin, 6-9 June 2005

CIRED2005

Session No 5
The power factor in each daily profile time interval is
described by a mean value (the variance of the power
factor within time intervals is ignored).

FIGURE 1: Overview of loss calculation and costing methodology

The time interval used in the load profiles is dictated
by the TOU energy loss tariff. In this case load profiles
are based on 1 hour time intervals, but any time
interval could be used.
Loads within the same Load Group are summated with
consideration for correlation within the time interval.
Different Load Groups are summated assuming zero
correlation between them.
Loads can be scaled so that load parameters can be
adjusted using energy consumption data.
Networks are limited to radial configurations.
Unbalanced network technologies are directly
supported, including three phase (3-wire and 4-wire),
single phase, bi (dual) phase and Single Wire Earth
Return (SWER).
Energy loss costs are based on TOU long run marginal
generation costs. This improves on the past practice
of splitting losses into peak and base-load
components. The energy loss costs are calculated for
all time periods.
Reactive energy loss costs are based on shunt
compensation costs.
Future costs are represented by net present value
using financial forecasts and the network lifetime.
The energy losses and loss costs are calculated in
year 1in each phase/neutral conductor of each branch.
A composite forecast is derived for each phase/neutral
conductor taking into consideration the load forecasts
of the consumer load groups. The lifetime energy loss
and loss costs are approximated from the year 1 values
and composite forecast. The accuracy of the
approximations has been tested via Monte Carlo
simulation, and gives good results (see RESULTS).
The consumer load shape is assume d to be constant in
future years, and only the magnitude varies.
The TOU loss tariff structure is assumed to be
constant in future years, and only the magnitude of
the tariff components vary relative to the forecast.

If the consumer load shapes or loss tariff structures change
significantly in future years, the TOU losses and loss costs
in the future years should and can be calculated directly in
each future year (not be based on year 1 values).
Load library
Load characteristic:
Time-of-use load parameters,
including mean, standard
deviation, variance and coefficient
of correlation
Load forecast
Tariff library and financial
parameters
Time-of-use generation and
transmission tariffs
Tariff forecasts
Financial parameters (Net Discount
Rate)
Evaluation period (years)
Network topology
Multi branch network model:
Branch length
Branch technology
Branch nominal voltage
Branch resistance
Customer types and numbers
Connection phasing
Technical loss calculation
Time-of-use energy losses
Present value cost of energy losses
C C I I R R E E D D 18
th
International Conference on Electricity Distribution Turin, 6-9 June 2005

CIRED2005

Session No 5

MONTE CARLO SIMULATION

The calculations were tested with South African load
research data using Monte Carlo simulation. The Monte
Carlo process is described in [4], and has been enhanced to
include the mixing of load types, and the calculation of
losses and loss costs in 1 hour time intervals.

The method was tested on a sample radial feeder with the
following characteristics and evaluation parameters:
Three phase 4-wire LV feeder with load connections
shown in table 1. The phase and neutral conductor
resistance between nodes is 0.0479ohm and 0.0778ohm
respectively.
Mixture of three domestic Load Groups (L1, L2 and L3
in table 2 and figure 2).
Evaluation period June and July 2002 (the period of
overlapping load data for the three Load Groups).
The Eskom Wholesale Electricity pricing System for
2003 (WEPS) [6] was used for the TOU tariff for loss
costs. The WEPS tariff rates differentiate between the
time of day, day of week and month of year.
10 year lifetime for energy and loss cost calculation.
Net Discount Rate of 8% for the costing of future
losses.

TABLE 1: Network used to test method (consumer numbers per Load
Group per phase)
Phase A Phase B Phase C
Node
L1 L2 L3 L1 L2 L3 L1 L2 L3
1 3 2 1
2 3 2 1
3 3 2 1
4 3 2 1
5 3 2 1
6 3 2 1

TABLE 2: Summary of load data used to test method
Load Group
(community)
Average monthly
consumption [kWh]
Growth rate
[% per annum]
L1: Antioch 71 10 %
L2: Mfazazana 183 5 %
L3: Welgemoed 1173 3 %

Per unit weekday mean profiles
0
0.2
0.4
0.6
0.8
1
1.2
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time of day (hour)
P
e
r

u
n
i
t

l
o
a
d
L1: Antioch
L2: Mfazazana
L3: Welgemoed

FIGURE 2: Per-unitised weekday mean profiles (June and July 2002) for
each of the three domestic load groups used for testing

RESULTS

Table 3 compares the results of the Monte Carlo simulation
and the calculated values of loss energy and cost for the
sample network and loads over the 10 year evaluation
period. Three methods were tested:
LLF: Average energy losses based on losses at peak.
Average profile: Energy losses are calculated via Load
flow analysis in 1 hour time intervals using mean daily
load profiles for each Load Group.
The proposed method: As described in this paper.

TABLE 3: Comparison of calculated and Monte Carlo simulated
results over the 10 year evaluation period (R61US$). Values in
brackets are differences relative to the Monte Carlo simulation
Result Loss [kWh] Loss cost [R]
Monte Carlo simulation 9720 R 1475
LLF method 4329 (-55.47%) R 489 (-66.85%)
Average profile method 4552 (-53.17%) R 514 (-65.15%)
Proposed method 9863 (+1.46%) R 1485 (+0.68%)

The difference between the Monte Carlo simulation, LLF
and average profile methods is large, and indicate that
these empirical methods are not suitable for energy loss
estimation with stochastic loads such as in LV networks.
The errors in the proposed method are small and well within
practical planning and design requirements. Similar
accuracy is expected with non-domestic loads and mixtures
of domestic and non-domestic loads.

CONCLUSION

A statistically based loss calculation and costing
methodology has been developed for radial distribution
networks. The technical load loss calculation methodology
takes into considerat ion South African network and load
characteristics, and has the following features:
It applies to MV and LV balanced and unbalanced
networks with various load classes described by
monthly and daily load profiles.
The stochastic nature of distribution loads (including
load coincidence) is accurately modelled using
parameters of load profile uncertainty.
Active and reactive energy losses are calculated using
TOU tariffs and reactive compensation costs
respectively.
The errors in empirical energy loss calculations with
stochastic loads are large. The proposed methodology
requires additional effort in the derivation of load
parameters and the computation of losses. The additional
effort is justified due to the large improvement in accuracy.

The loss calculatio n and costing methodology can be used
to enhance network planning and design optimisation
problems, such as the selection of line technology (three
phase, single phase, bi-phase, SWER), voltage level and
conductor size. The methodology should be appropriate for
distribution in most developing countries.

ACKNOWLEDGEMENTS
C C I I R R E E D D 18
th
International Conference on Electricity Distribution Turin, 6-9 June 2005

CIRED2005

Session No 5

The authors acknowledge Eskom Resources and Strategy
Division for the sponsorship of this research. Measured
data from the NRS Load Research programme was used.

REFERENCES

[1] Flinn DG, Hall S, Morris J, 1983, A model for
calculating losses on a distribution system, Modeling
and Simulation. Vol.14. Proceedings of the
Fourteenth Annual Pittsburgh Conference, 127-32
[2] Nickel DH, Braunstein HR, 1981 Distribution
transformer loss evaluation: 1-Proposed techniques,
IEEE Transactions on Power Apparatus and Systems,
vol. PAS-100, no. 2, Feb
[3] Grainger JJ, Kendrew TJ, 1989, Evaluation of technical
losses on electric distribution systems, CIRED 1989.
10th International Conference on Elect ricity
Distribution, vol. 6, 488-93
[4] Herman R, Heunis SW, 2002, Estimation of the annual
resistive loss on LV residential feeders, 7th
International Conference on Probabilistic methods
Applied to Power Systems, 655-60
[5] Oliveira CCB, Kagan N, Meffe A, Jonathan S, Caparroz
S, Cavaretti JL, 2001, A new method for the
computation of technical losses in electrical power
distribution systems, 16th International Conference
and Exhibition on Electricity Distribution, vol. 5, 5
[6] Wholesale Electricity pricing System (WEPS)
Customer Information Brochure 2003: Eskom website:
www.eskom.co.za

AUTHORS
Clinton Carter-Brown has a BScEng degree from the
University of Kwa-Zulu Natal and MScEng degree from the
University of Cape Town. He is a registered professional
engineer and is presently involved in the standardisation
and optimisation of Eskom Distribution network planning
practices.
Trevor Gaunt is a Professor in the Department of Electrical
Engineering at the University of Cape Town, South Africa.
Schalk Heunis has MScEng and PhD degrees from the
University of Stellenbosch and works as a consulting
engineer in Pretoria/Tshwane.
C C I I R R E E D D 18
th
International Conference on Electricity Distribution Turin, 6-9 June 2005

CIRED2005

Session No 5
ANNEX A: ENERGY LOSS CALCULATION AND COSTING EQUATIONS

The Energy Loss [kWh] in time interval t, day d, and month m for a phase or neutral conductor in a branch is given by:

( )
i
V
E E E E
R Loss Energy
m d t LG m d t LG
m d t

+
=
2
, ,
2
, ,
2
, ,
]] [ [ ]] [ [
1000


Where:

E[E[m
2
]]
LG t,d,m
is the expected value of the square of the load mean in time interval t, day d, and month m [kVA
2
]
E[E[s
2
]]
LG t,d,m
is the expected value of the variance of the load in time interval t, day d, and month m [kVA
2
]
i is the time interval duration used in the load data [hours]
V is the branch voltage [volts]
Note: E[ ] is the expected value operator. E[ ] refers to the expected value across a number of consumers.
E[E[ ]] refers to the expected value across a number of consumers and time.

In an unbalanced multiphase network the load contributions can be from up to three angles (in the case of a neutral
conductor in a 3-phase 4-wire network), and:

( ) ( ) ( ) ( )
( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
2
3 3 2 2 1 1
1
, , 3 3
, , 2 2 , , 1 1
, ,
2
3 3 2 2 1 1
, , 3
, , 2 , , 1
] [
] [ ] [
]] [ [
FPF ARCCOS A Angle FPF ARCCOS A Angle FPF ARCCOS A Angle
LG
L
PF ARCCOS A Angle
m d t L L L
PF ARCCOS A Angle
m d t L L L
PF ARCCOS A Angle
m d t L L L
m d t LG
F FS F FS F FS
E N S
E N S E N S
E E
m d t L
m d t L m d t L
+ + +
=
+
+ +
+ +
+

+ +
=

( ) ( )
( ) ( )
( )
( )
( )
( )

+
+
+
+ +

+ + + +
=
LG
L
L L L L
L L L L
L L L L
L L L
L L L L L L
m d t L m d t L
L L L L L L m d t L m d t L
m d t LG
S N S N A A
S N S N A A
S N S N A A
N N S
N N S N N S
E E
S N S N S N E V
E E
1
3 3 1 1 3 1
3 3 2 2 3 2
2 2 1 1 2 1
3 3
2
3
2 2
2
2 1 1
2
1
2
, , , ,
2
3 3
2
2 2
2
1 1
2
, , , ,
, ,
2
cos 2
cos 2
cos 2
1
1 1
] [ ] [
] [ ] [
]] [ [

Where:

E[m
L t,,d,m
] is the average of the Load Group L means in time interval t, day d, and month m [kVA]
V[m
L t,,d,m
] is the variance of the mean values of Load Group L in time interval t, day d, and month m [kVA
2
]
E[s
L t,,d,m
] is the average of the Load Group L standard deviations in time interval t, day d, and month m [kVA]
E[s
2
L t,,d,m
] is the average of the Load Group L variances in time interval t, day d, and month m [kVA
2
]
E[r
L t,,d,m
] is the average of the Load Group L correlation in time interval t, day d, and month m [unitless]
LG is the number of Load Groups supplied by the branch
A
1
, A
2
, A
3
are the fixed angles for load contributions in unbalanced multiphase networks [degrees]
N
1L
, N
2L
, N
3L
are the number of consumers of Load Group L at angles A
1
, A
2
and A
3
respectively
S
1L
, S
2L
, S
3L
are the load scaling factors of Load Group L at angles A
1
, A
2
and A
3
respectively [PU]
PF
L t,d,m
is the power factor of Load Group L in time interval t, day d, and month m
F
1
, F
2
, F
3
are the Fixed Load magnitudes at angles A
1
, A
2
and A
3
respectively [kVA]
FS
1
, FS
2
, FS
3
are the Fixed Load scaling factors at angles A
1
, A
2
and A
3
respectively [PU]
FPF
1
, FPF
2
, FPF
3
are the Fixed Load power factors at angles A
1
, A
2
and A
3
respectively

The Energy Loss Cost [R] in time interval t, day d, and month m is given by:

m d t m d t m d t
EC Loss Energy Cost Loss Energy
, , , , , ,
=

Where:

C C I I R R E E D D 18
th
International Conference on Electricity Distribution Turin, 6-9 June 2005

CIRED2005

Session No 5
EC
t,d,m
is the cost of energy losses in time interval t, day d and month m [R/kWh]
The Lifetime Energy Loss [kWh] over the evaluation period is given by:


= = = =

=
M
m
E
y
DT
d
i
i t
m d t m d y m
Loss Energy D Forecast AE Loss Energy Lifetime
1 1 1
/ 24
, ,
2

Where:

M is the number of months in the year (can be less than 12 for evaluations within a year)
E is the evaluation period (lifetime) [years]
AE Forecast m y is the increase in apparent energy in month m in year y (relative to year 1) [PU]
DT is the number of day types used in the load data (typically 3 i.e. weekday, Saturday and Sunday)
Dd m is the number of days of type d in month m

The Lifetime Energy Loss Cost [R] over the evaluation period is given by:

( )

=
= =
=

=
M
m
DT
d
i
i t
m d t m d t m d
E
y y
y y m
EC Loss Energy D
GDR PU
Tariff PU Forecast AE
Cost Loss Energy Lifetime
1
1
/ 24
, , , ,
1
2

Where:

PU Tariff
y
is the increase in the energy loss tariff (relative to year 1) [PU]
PU GDR
y
is the Gross Discount Rate (relative to year 1) [PU]

AE Forecast
m y
is calculated as follows:

( )

LG
L
y L m L y m
F AE Group Load PU ABS Forecast AE
1

Where:

F
L y
is the increase in apparent energy consumption of Load Group L in year y (relative to year 1) [PU]

( ) ( ) ( )
m L m
m
m L
m L
AE Group Load Angle AE Branch Angle COS
AE Branch
AE Group Load
AE Group Load PU

=


( ) ( )
( ) ( )
( ) ( )

= =
+
+
+

+
+
=
DT
d
i
i t
PF ARCCOS A Angle
m d t L L L
PF ARCCOS A Angle
m d t L L L
PF ARCCOS A Angle
m d t L L L
m d m L
m d t L
m d t L
m d t L
E N S
E N S
E N S
i D AE Group Load
1
/ 24
, , 3 3
, , 2 2
, , 1 1
, , 3
, , 2
, , 1
] [
] [
] [

( )
m
LG
L
m L m
AE Load Fixed AE Group Load AE Branch + =

=1


C C I I R R E E D D 18
th
International Conference on Electricity Distribution Turin, 6-9 June 2005

CIRED2005

Session No 5
( ) ( ) ( ) ( )
( ) ( )

=
+
+ +

+ +
=
DT
d
m d
FPF ARCCOS A Angle
FPF ARCCOS A Angle FPF ARCCOS A Angle
m
D
F FS
F FS F FS
AE Load Fixed
1
3 3
2 2 1 1
24
3 3
2 2 1 1