Network Loss Study in PowerFactory™

Albert Pors Harshul Dalal

Technical Services Engineer Network Planner
Asset and Network Planning Asset and Network Planning
Endeavour Energy Endeavour Energy
Sydney, NSW Sydney, NSW
[email protected] [email protected]

Abstract— This paper describes Endeavour Energy’s approach B. Historic Method

to calculating distribution loss factors and its experience in
implementing the loss factor study in DIgSILENT Historically, Endeavour Energy has used a hierarchical
PowerFactory. The paper also discusses performance issues approach to determine the distribution loss factors. In general,
experienced with a computationally intensive methodology for a the method used to calculate series energy losses has been to
large model.
first calculate the loss on the relevant part of the system under
Index Terms - DIgSILENT PowerFactory, Distribution loss peak load conditions using a load flow package. This includes
factor, Network Losses, Load flow, Sensitivity Analysis. Sub-transmission (132-33kV) and distribution (22-11kV).
The associated energy losses as in (1) incorporates the Load
I. INTRODUCTION Loss Factor (LLF) which is the ratio between the
For the past few years Endeavour Energy has developed instantaneous losses incurred at peak load and the average
and utilised an approach to loss factor calculation that includes instantaneous losses over a year as in (2).
performing a year’s worth of 15-minute load flows in
combination with sensitivity load flows to determine the sub-
transmission network loss component [1]. The development of LossesSeries (kWh) = Losses Peak (kW ) × 8760(hours) × LLF (1)
a PowerFactory network model presented an opportunity to
overcome many of the challenges experienced previously in 35040 ( Iterations15− min )  2 
implementing this methodology using another load flow ∑  Load n 
n =1  Load 2 
package.
LLF =  Peak 
(2)
This paper describes Endeavour Energy’s network model, 35040
the method of calculating distribution loss factors and its
implementation in PowerFactory using DIgSILENT These LLFs were traditionally calculated for all Bulk
Programming Language (DPL) scripts. The paper also Supply Points (BSP’s), major generators, transmission
discusses the results obtained and the experience gained substations, zone substations and major customers. These
around performance issues of this computationally intensive LLFs were evaluated using metering data at each location if
approach to loss calculations in PowerFactory. available, or alternatively, SCADA data. Peak losses were
then modeled using location specific demand forecasts for the
II. NETWORK LOSS FACTOR CALCULATIONS
coming year.
A. National Electricity Rules Requirements
Within substations, transformer no-load shunt losses have
The National Electricity Rules [2] require that Distribution been calculated using manufacturer’s data if available as in
Network Service Providers (DNSPs) calculate Distribution (3). Since these losses are independent of the transformer
Loss Factors (DLFs) for the next financial year and submit loading, no LLF is applied.
them to the Australian Energy Market Operator (AEMO)
following approval from the Australian Energy Regulator
(AER). Losses Shunt (kWh) = Losses Shunt (kW ) × 8760(hours ) (3)
The NER require an annual calculation of site specific The main limitations to this approach were:
DLFs for major customers (>10MW or >40GWh), while loss
factors for each tier of the network should be provided to • Manual intervention is required to determine the LLF
calculate the losses attributable to the remainder of the for each customer or network tier, as well as in
customers. calculating the final loss factors. This is generally an
iterative process.
 • The LLF is a function of the instantaneous losses at It is clear that this assumes that both the tier losses, and
peak load and the average losses over one year. This consequently the total losses, will equal zero when no energy
single snapshot of the power system fails to account is supplied by the network. This is not actually the case,
for changes in the network operating point as primarily due to charging currents. Nevertheless, it is a
generation and load vary. reasonable means of apportioning the losses between each
major customer or network tier.
This sensitivity coefficient can be determined from the rate
C. Current Methodology of change in network tier loss with respect to a change in
network load at the current operating point of the system as
In the current methodology, the series losses on the sub- per (6)
transmission network (132kV to 22/11kV substation busbars)
are calculated by incorporating the 15 minute metering data ∂Loss K
from the previous financial year in a load flow routine. In each RateOfChange = (6)
metering interval, one load flow solution is obtained to ∂Load K
determine the total network losses and the accumulated
network losses within each tier. To derive this rate of change in loss, the load consumed by
the specific network tier, or major customer, is perturbed by
10% and the resulting change in loss is noted. This is shown
graphically in Figure (2).

Figure 2

Due to the fact that the network losses (I2R) are not
linearly proportional to the total network tier load, the rate of
change derived will not equal the sensitivity coefficient.
Figure 1 Consideration was given to utilizing a further compensation
However, this is the total system losses incurred at a factor based on the sensitivity of the network tier losses to the
particular voltage level and not the total network tier losses total system load, however from experience with real metering
attributable to that particular customer, or network tier. data this compensation factor had mixed results. Therefore
Therefore, the total losses at each voltage level (or network each individual sensitivity coefficient is simply scaled for each
tier) must be apportioned between the customers connected to network tier such that they sum to unity.
the network. This approach still fails to cater for variations in the
network configuration; however the network model is set up
This is achieved by linearising the network operating point,
to represent the switching configuration which is true for most
which is a function of all the site specific loads or network tier of the year.
pools. This operating curve has K dimensions and may be
linearised by assuming the relationships of (4) and (5) The loss factor calculations have been achieved by
calculating the losses attributable to each major customer, or
load group, within each tier of the network, at each 15 minute
K
Losses (Total ) Series = ∑ Ai Losses (Total ) Series (4) interval including shunt losses. In this case, the loss factor LF
i =1 for tier K can be given by
Losses (i ) Series = Ai Losses (Total ) Series (5)
35040 35040
Where, Ai is the sensitivity coefficient for the particular ∑ Losses( K ) Series + ∑ Losses( K ) Shunt
N =1 N =1
customer or tier and Losses(Total)Series is the total series loss LF = 35040
(7)
incurred within that tier of the network. This equation implies ∑ Energy ( K )
that the sum of these sensitivity coefficients must sum to N =1
unity.
 III. POWERFACTORY NETWORK MODEL factor studies; this meant it was separately maintained which
Endeavour Energy has a sub-transmission model built in alone required significant effort. However, the previous model
DIgSILENT PowerFactory, which covers from the interface had considerably fewer nodes as lines were not sectionalised
with the Transmission Network Service Provider (TNSP) and the substation busbars were generally represented by a
injection point, usually a 330/132kV substation; down to the single node.
Zone Substation (ZS) level busbar, 11kV or 22kV. Upstream The current methodology as previously described requires
of the TNSP injection point is a reduced equivalent National approximately 30 load flows (dependent on number of site
Electricity Market (NEM) model. The 11 or 22kV distribution specific customers) per 15-minute metering data interval
feeders are not captured in this model. which results in a total of 1,051,200 load flows for a full
The model is populated from a network characteristics year’s calculation.
database using a DGS converter tool, which is continuously To address the long processing time that results from the
updated as projects are commissioned. It also makes use of the large number of nodes and number of load flows, the script
PowerFactory ‘Variations’ function so as to capture future includes the ability to break up the runs so sections of the full
network expansion, changes and variations which can be study can be run across multiple computers. For the full study
turned on and off as required. run four computers were utilised.
Substations are represented by their true busbar A comparison of the computational performance between
configurations. Lines are broken up as individual pi equivalent old and new models is given below in Table (1)
sections representing change in construction such as conductor
Run Time Run Time / Non-
changes or tee off points. Where lines are mutually coupled Measure Nodes
(hrs) node (hrs) convergences
the coupling data is entered using the tower type model. Loads Old Model 600 96 0.160 0.23%
and generation are included with values for the forecast New Model 8800 302 0.034 0%
period. Variation 1466% 314% 21%
Table 1
IV. LOSS STUDY SCRIPT DEVELOPMENT
While the overall run time has increased due to the node
To perform the network loss factor study in the count in the new model, a few significant advantages have
PowerFactory sub-transmission network model, a DPL script been achieved which far outweigh the additional run time. By
was developed which implements the previously described using a common model which is regularly updated for
calculation methodology. multiple studies and is linked to a common characteristics
A high level overview of the functions of the script is as database, several weeks of model preparation and updating
follows: can be avoided. The new model is easily adapted for a future
study period using the variations built into the model. The
• Import from file the 15-minute metering data which has confidence in the accuracy of the loss study result has also
been scaled to forecast values for zone substation loads, improved as evidenced for example by zero non-convergent
special customer loads and generator load into chavecfiles load flows. Furthermore, since the model is graphical and
within PowerFactory. provides detailed output run information, the sources of errors
• Identify from an input file the special customers that and divergence are relatively easily found and corrected.
require individual loss factors and note them in a matrix.
VI. FUTURE WORK
• Run the individual load flows, including perturbation load To improve performance of running the loss factor study,
flows, for each 15-minute interval. This results in some areas of further development which may result in
approximately 30 load flows per 15-minute time interval. significant speed improvement have been identified as:
The taps are locked after the first run for each load flow
in the interval. Non-convergent load flows are reported on • Creating a script to remove unnecessary nodes prior
if they occur. to a loss study run. For example nodes between
consecutive line sections could be collapsed without
• Record individual interval results in matrices. loss of accuracy. It was found that speed performance
• Export results matrices to an output file at the end of the could be improved by a percentage approximately
year’s worth of runs. equal to the square root of the percentage reduction
in nodes.
Perturbation was achieved using the “Scaling Factor” of
the load element by setting it to 90% for the individual loads • Utilise multiple instances of PowerFactory running
or load elements in the tier. on individual processor cores. Most PCs these days
have quad-core processors and this would better
V. RESULTS AND PERFORMANCE utilise processor capability.
The PowerFactory model captures the network in a high • Investigating the potential use of the contingency
level of detail as it is used for a wide variety of studies. As analysis functionality in PowerFactory. This
such, the number of nodes in the model is relatively high. The performs a base load flow from which further results
previous model which was text based was only used for loss can be computed without a load flow.
 ACKNOWLEDEGMENT REFERENCES
The authors of this paper wish to acknowledge the work of [1] D. Spoor, C. Howat, "Evaluation of Distribution Loss Factors",
both existing and previous Endeavour Energy employees in Endeavour Energy internal Special Report No. S536, Australia, 2008.
developing the methods described in this paper as well as
those contributing to the development of scripts. [2] National Electricity Rules V56 – Chapter 3, Section 3.6.3, AEMC,
2013, http://www.aemc.gov.au/Electricity/National-Electricity-
Rules/Current-Rules.html.