C I R E D 22nd International Conference on Electricity Distribution Stockholm, 10-13 June 2013

Paper 0664

COMPARING LOAD ESTIMATION METHODS FOR DISTRIBUTION SYSTEM

ANALYSIS

Robert ARRITT Roger DUGAN

EPRI – USA EPRI – USA
[email protected] [email protected]

power flow characteristics for distribution state

ABSTRACT estimation such as voltage profile, equipment evaluations,
Load allocations for distribution system analysis is one losses, and state of distribution equipment.
area in system modelling where simple, generalized
assumptions are commonly made by distribution planners MODELLING THE TEST CIRCUIT
for lack of better data. Most of the time, the assumptions The analysis described in this paper compares power flow
are based on measurements taken only at the distribution results using four different load allocation methods on the
substation bus or at the feeder head. With widespread same circuit model. [3] The circuit was selected because
application of advanced metering infrastructure (AMI) it had 99% AMI coverage. The results for the load
technologies, loading estimates for distribution system allocation method using AMI demand interval data were
analysis can now be based on actual measurements taken assumed to be the most accurate and serve as the
at individual loads. This can result in greatly improved reference against which the other load allocation methods
accuracy in distribution power flow analysis. Three are compared. The AMI demand interval data consists of
techniques for load allocation are analyzed and 15-minute kW demand readings over the period from 1
compared to the case with actual AMI data for all June 2010 to 18 November 2010.
customers. Selected details of the test circuit and analysis The test circuit is a 13.2 kV residential feeder with only
process are provided. 1% commercial load. There are 1779 individually-
metered customers. There are 10 3-phase loads; the
INTRODUCTION remainder are singl-phase. The peak demand on the
Analysis tools used for distribution systems have become feeder is 5800 kW with a load factor of 46% (ratio of
very accurate in the representation of line models, average load to peak demand). The feeder contains two
regulators, load tapchangers, capacitor switching, etc. [1], capacitor banks, rated 450 and 900 kvar, switched under
[1] However, load estimation is often a weakness. current control.
Distribution system analysis tools typically allocate the This circuit was modelled using the EPRI OpenDSS
load demand measured at the substation to individual program. One capability of this program that is useful for
customer meter points based on the kVA rating of the this analysis is that it can perform sequential power flow
MV/LV transformer. Some programs also have simulations efficiently. Each load can be assigned its own
algorithms for load allocation based on monthly kWh
loadshape or all loads can be assigned the same
billing. With the advent of Advanced Metering
Infrastructure (AMI) metering capable of capturing loadshape. Both options were exploited in this analysis
extensive demand interval data, new opportunities exist comparing different load allocation methods. This
to improve customer loading estimates with distribution program was the main research tool used in a similar kind
system analysis based on actual demand interval data. of analysis for distribution efficiency including nearly 80
These improvements can have a significant impact on distribution feeders from the US and Europe reported
both distribution planning and real-time distribution state previously. [6]
estimation. All service transformers (MV/LV transformers) on the
Efforts are currently underway to use the AMI demand feeder were modelled. While US utilities use a split-
interval data for both defining and verifying distribution
phase 120/240 V transformer, a simplified model was
system load models. Each load can be defined separately
using its own measured loadshape, which should yield used for this analysis. Only one value – total kW – was
the most accurate simulations. Research has been available from the AMI data so it was not possible to
performed on examining the use of AMI for modelling accurately split the load between 120 V windings.
loads in distribution system analysis. [3], [4] This Therefore, the transformers were modelled as simple two-
research was inspired by a study by Kersting and Phillips winding transformers with an LV voltage of 240V for
[5] and builds on that work by examining a larger circuit single-phase loads. Each LV service drop was modelled
with nearly full AMI coverage. This paper provides some using an average length of 31 m since detailed LV circuit
additional details of the simulations. data were not available. Due to the low service voltage,
The results of the load allocations using the actual AMI LV circuits in the US are relatively short compared to
data are compared to simpler, or more traditional, load
allocation methods. This example gives additional insight 400 V LV circuits common in other parts of the world.
into the impact of load allocation assumptions on circuit They typically range in length from 15 to 45 m.
Through many case studies of distribution feeders, we

CIRED2013 Session 5 Paper No 0664

 C I R E D 22nd International Conference on Electricity Distribution Stockholm, 10-13 June 2013

Paper 0664

have found it relatively easy to match the active power demand allocation was updated for each month in
component of the power flow to measured results. Of the simulation. This would be similar to a
course, the line and transformer impedances must be technique that allocates loads based on a
sufficiently accurate to account for the losses. It is often customer’s billing information from the previous
more difficult to calibrate the model to the measured year. The substation load is allocated to each
reactive power flow, particularly on feeders in the US customer in proportion to the specified monthly –
where it is common to have multiple switched power or other period – kWh usage. This technique
factor capacitor banks. Fortunately, it was straightforward distributes the kW loading more heavily to those
to match the reactive power characteristic for the test customers using more energy regardless of
circuit because it serves mostly residential loads. The service transformer size. One assumption inherent
power factor of each load was estimated to be 0.95 in this approach is that the heavier users will be
lagging and was assumed to be invariant for the duration the heavier users throughout the billing period,
of the simulation. By switching on local current the which is obviously not always true.
capacitor switching occurs quite naturally in the  Class Loadshape Allocations – This is a method
simulation. Simulation difficulties with reactive power used in some distribution state estimation tools as
can arise when capacitors are controlled remotely or by well as distribution planning tools. It uses a
less deterministic quantities. combination of the transformer kVA size along
With a full 3-phase circuit model, all MV/LV with historical information about the type of load
transformers, and one LV line modelled for each of the being served. The historical load type information
metered loads, there are a total of 3107 electrical nodes in consists of various class loadshapes for customers
the test circuit. For the AMI case, there were 1779 unique based on season of the year, day of week, and
loadshapes, each with 16416 intervals. A full simulation
holidays. The class loadshape is used along with
requires approximately 4.5 min to execute on a typical
the measured substation demand data to develop
Windows-based modern laptop computer. Thus, this type
the loading level for each load. Knowledge of the
of analysis is not onerous when the data are available.
load class is required.
LOAD ALLOCATION METHODS
RESULTS
The load allocation methods used in this study were:
 AMI Allocation – This is the reference method. The four load allocation methods were compared for
accuracy in the following three types of predictions:
It uses the actual kW demand interval
measurements for each customer. Each customer
1. Prediction of equipment loading
has a unique loadshape.
2. Prediction of service voltage magnitudes
For each of the loads with AMI, the 15-minute 3. Prediction of losses
average kW measurements were used directly to
define the demand value for each interval of the Equipment Loading Evaluation
simulation. The unmetered load values (1% of the Loading evaluations of power delivery equipment were
total load points) were estimated from the conducted by comparing the load current ratings of each
substation kW measurements while compensating line and transformer with calculated operating conditions.
for system losses. The evaluation status for each component was flagged as
 Transformer kVA Allocation – This technique a “exceeds normal rating” when the loading exceeded
is commonly used when better data are 110% of user-defined normal limits. The evaluation
unavailable. It uses the total feeder load status was flagged as “exceeds emergency rating” when
measurements taken at the feeder head and then loading exceeds 150% of user-defined normal limits. This
allocates a portion to each load point based on the allows the feeder operator to determine the extent of the
kVA rating of the service transformer. In the test overload and the location of the overloads on the system.
circuit, a transformer often has multiple loads, The test circuit model showed overloads only with
respect to the normal rating and all of the overloads
individually metered. The load allocated to a
occurred on MV/LV transformers. No overloads were
transformer is allocated to the individual loads for
reported on lines and cables.
the simulation. The loadshape assumed for each Table 1 shows the number of overloads identified with
load is the same as the total feeder load. each allocation method. The reference method, the AMI
 Monthly Usage Allocation – This method uses simulation, reported a total of 51 instances of overloads.
the monthly kWh billing to allocate loads to each Of the other methods, the monthly usage allocation
of the customers and then uses the substation load method reported 27 instances of overload, which was the
measurements develop a loadshape that is closest to the AMI method. The substation allocation, the
assigned to all loads. Each customer’s kW common method that allocates loads based on

CIRED2013 Session 5 Paper No 0664

 C I R E D 22nd International Conference on Electricity Distribution Stockholm, 10-13 June 2013

Paper 0664

transformer kVA rating, identified only 5 instances of cases match closely; however, only the AMI model
overload. None of the overloaded transformers found causes the capacitor bank farthest downstream to change
with the kVA rating allocation method were found to be state due to current exceeding the ON setting. This can
overloaded with the AMI simulation, which is assumed to become a critical modeling issue when performing
be accurate. Not only are few overloads identified, but distribution state estimation.
they are misidentified. Note that in the kVA allocation
model each customer was modeled separately; therefore, Table 3. Minimum customer voltage at feeder peak
the loading on each transformer was proportioned to the demand.
total number of customers connected to it.

Table 1. Number of transformers that exceeded their Minimum Customer Voltage At Peak
normal rating.
AMI kVA rating Monthly Usage Class
Allocation Loadshapes
Number of Transformers Identified that Exceeded their 119.95 121.85 119.86 121.40
Normal Rating
AMI kVA rating Monthly Usage Class
Allocation Loadshapes 126

51 5 27 4 125.8
Capacitor Switching on AMI Allocation
125.6

In addition to identifying transformer overloads, the AMI Voltage

125.4

metering data can also be used to better optimize 125.2

transformer asset utililization. Table 2 shows the
transformer loading at the feeder peak. 71% of the 125

transformers are loaded less than 50%. This information 124.8

could allow system planners to better match transformer
124.6
sizes to loads and reduce no-load, or idling, losses.
124.4
AMI kVA rating Usage Class
Table 2. Transformer loading at feeder peak with AMI
allocation. Figure 1. Voltage Computed at the Substation Bus for
All Cases.
Transformer Loading at feeder peak with AMI
Allocation
Loss Estimates
>75% Loss estimates produced by distribution system analysis
>25% and >50% and
<25% and >100%
<50% <75% are sensitive to the assumed load distribution throughout
<100%
the feeder.
34% 37% 19% 7% 3% The AMI and Monthly Allocation methods yielded more
average losses than the Substation and Class Allocation
methods. Both the LV (secondary) and MV (primary)
systems showed more losses in these two cases, which is
Voltage Estimation shown in Figure 2. The AMI and Monthly Usage
Distribution state estimation and voltage optimization methods each have the same MV line losses; however,
both require better estimates of the feeder voltage at the Monthly Usage case shows more LV losses.
various times of the day. All cases showed approximately the same annual no-load
Table 3 shows the minimum service voltage for all losses due primarily to the fact that there was very little
customers in the circuit at the time of peak loading. The voltage drop on this circuit. Thus, the voltages across the
Monthly Usage Allocation comes closest to the voltages service transformers are approximately the same
predicted by using AMI data. The minimum voltage throughout the simulation period.
predicted by the kVA rating allocation is high by As for peak losses, the AMI and the Monthly Usage
approximately 2 out of 120 V, or 1.7%. On average, the methods again predict higher losses. The AMI method
minimum voltage levels are nearly the same for all the predicts 10% more losses on both the LV and the MV
allocation methods. Rather than providing a basis for a than the traditional Transformer kVA Allocation method.
conclusion on voltage drop, this is more a result of the Again, as in the average case, the AMI and Monthly
test circuit having relatively little average voltage drop. Usage methods have the same MV line losses; however,
Figure 1 displays a time-series simulation of the voltage the Monthly Usage case has more LV losses.
at the substation bus. The voltages computed for all four An accurate representation of the LV losses is
 C I R E D 22nd International Conference on Electricity Distribution Stockholm, 10-13 June 2013

Paper 0664

particularly important for those circuits with long LV overloads are reported because all transformers
lines such as are often encountered in European have the same percentage of loading. The test
distribution systems. circuit actually resulted in a few transformer
overloads due to the way load was allocated for
70
63.86 64.31 transformers with numerous loads. However, the
60
61.02
59.64 predicted overloads did not correlate with the
transformers found to be overloaded from the
50 AMI data.
40
 The under loading of transformers can be
identified with proper load allocation and
kWh

30 transformer sizes may be optimized.

 The AMI and Monthly Usage methods result in
20
nearly the same MV line losses, because the
10 8.45 assumed load distribution is approximately the
4.73
2.93 2.77 same. This suggests that if AMI data were
0
AMI kVA rating Monthly Usage Class
lacking, the Monthly Usage method would be
Primary Losses Secondary Losses better than the other methods.
Figure 2. Annual MV and LV Losses.  The Monthly Usage Allocation method tracks the
voltage predicted by AMI data more closely than
160 the other allocation methods.

143.90 143.51
140 The use of AMI data can give a better indication
131.21
125.87 of the state of controlled system components such
120
as capacitors.
100  General conclusions about the Class Loadshape
allocation method cannot be drawn from this
kW

80
example because 99% of the loads were
60 residential.
40
24.81 REFERENCES
20 13.32 12.03 11.21

0 [1] OpenDSS Computer Program, EPRI Open-Source

AMI Sub Allocate Monthly Usage Class
Distribution System Simulator, 2013, Available on
Primary Losses Secondary Losses
Sourceforge.Net, Ver 7.6.2.
[2] W. H. Kersting, 2007, Distribution System Modeling
Figure 3. Peak MV and LV Losses. and Analysis, CRC Press.
[3] R.F. Arritt, R.C. Dugan, R.W. Uluski, T.F. Weaver,
CONCLUSIONS 2012, “Investigating Load Estimation Methods with
The method that is used for estimating the customer load the Use of AMI Metering for Distribution System
can have a significant impact on distribution system Analysis,” 2012 IEEE Rural Electric Power
analysis. Some of the general conclusions derived from Conference; Milwaukee, WI.
this example include: [4] EPRI, 2011, Tools, Methods, and Modeling for
Advanced Distribution Systems., Palo Alto, CA:
 Loading data from the application of advanced 1021993.
metering infrastructure (AMI) will provide [5] W.H. Kersting and W.H. Phillips, 2008, “Load
Allocation Based Upon Automatic Meter Readings”,
distribution planners with greatly improved
2008 IEEE PES Conference, Paper 08TD0162, April
predictions of actual system performance,
21-24, 2008.
assuming analysis tools are available that can [6] R. Dugan, T. Short, and K. Forsten, 2009, “Modeling
process the data. Energy Efficiency Alternatives For Distribution
 Methods such as allocating based on service System Power Delivery”, Paper No. 1021, CIRED
transformer kVA rating, not only under-report 2009, Prague.
overloads but can also mis-report overloads.
Typically, a kVA allocation only includes a
single load at each transformer location and all
transformers initially are assigned the same
percentage of loading. Few, or no, transformer

CIRED2013 Session 5 Paper No 0664