IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 10, NO.
3, JULY 2019
Distribution System Parameter and Topology
Estimation Applied to Resolve Low-Voltage
Circuits on Three Real Distribution Feeders
Matthew Lave , Matthew J. Reno, and Jouni Peppanen
Abstract—Accurate distribution secondary low-voltage cir-
cuit models are needed to enhance overall distribution system
operations and planning, including effective monitoring and coor-
dination of distributed energy resources located in the secondary
circuits. We present a full-scale demonstration across three real
feeders of a computationally efficient approach for estimating
the secondary circuit topologies and parameters using historical
voltage and power measurements provided by smart meters. The
method is validated against several secondary configurations, and
compares favorably to satellite imagery and the utility secondary
model. Feeder-wide results show how much parameters can vary
from simple assumptions. Model sensitivities are tested, demon-
strating only modest amounts of data and resolutions of data mea-
surements are needed for accurate parameter and topology results.
Index Terms—Electrical engineering, energy management,
resistance.
I. INTRODUCTION
O MODEL, simulate, plan, and operate distribution sys-
T tems with growing amounts of photovoltaics (PV) and
other distributed energy resources (DER), more accurate distri-
bution system models are required [1]. Secondary (low-voltage)
circuits, those which connect customers to service/distribution
transformers, are becoming important to include in distribution
grid models since that is where most DERs are connected and
since they have higher per unit impedances and hence a large
share of the feeder per unit voltage drop than primary circuits [2],
[3]. Inaccurate secondary circuit models can lead to, e.g., errors
in hosting capacity estimates [4], [5] and ineffective choices of
smart inverter volt/var or volt/watt settings. Currently, the ma-
jority of existing utility feeder models do not include secondary
circuits at all. When modeled, they are typically unverified and
represented with limited detail.
Manuscript received July 24, 2018; revised December 2, 2018 and April 9,
2019; accepted May 12, 2019. Date of publication May 20, 2019; date of current
version June 20, 2019. This work was supported in part by the U.S. Department
of Energy’s Office of Energy Efficiency and Renewable Energy under Solar
Energy Technologies Office under Grant 34226 and in part by the Department
of Energy’s Grid Modernization Laboratory Consortium under Project 1.3.10
(Vermont Regional Partnership Enabling the Use of DER). Paper no. TSTE-
00736-2018. (Corresponding author: Matthew Lave.)
M. Lave and M. J. Reno are with the Sandia National Laboratories, Livermore,
CA 94551-0969 USA (e-mail: mlave@sandia.gov; mjreno@sandia.gov).
J. Peppanen is with the Electric Power Research Institute, Palo Alto, CA
94304 USA (e-mail: jpeppanen@epri.com).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TSTE.2019.2917679
1949-3029 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1585
The on-going extensive roll-out of smart meters at utilities
across the United States and the growing number of PV invert-
ers and other modern distribution system sensors are rapidly in-
creasing the available measurement data along distribution feed-
ers. This new measurement data has sparked a growing interest in
developing methods for distribution system topology and param-
eter estimation. Distribution system parameter and topology es-
timation methods have been proposed in various other works, in-
cluding [6]–[11]. The methods, however, typically require volt-
age angle measurements and/or are not demonstrated/tested on
real utility feeders with field measurement data.
Compared to these previous works, including our previous
work ( [12]–[14] ), this paper represents three major advances:
1) The parameter and topology estimation results are vali-
dated using several real transformers under normal oper-
ating condition (Section V),
2) We present a new method to resolve the parameters for
secondary systems with only a single customer by com-
paring them to other secondary systems (Section IV-B).
3) We expand on this new method by also applying it to
secondary systems with multiple customers, to detect
additional resistances that might occur when there is a
node between customers that is not at the transformer
(Section IV-C).
II. DISTRIBUTION SYSTEM SECONDARY CIRCUIT PARAMETERS
AND TOPOLOGY ESTIMATION CONCEPT
The overall objective of distribution system secondary circuit
topology and parameter estimation (DSPE) is to determine the
topology, resistance (R), and reactance (X) parameters of a sec-
ondary circuit (shown in red in Fig. 1) by leveraging AMI mea-
surements of voltage (V ), real power (P ), and reactive power (Q)
(shown in blue in Fig. 1). This paper focuses on typical North
American split-phase secondary circuits that can be perfectly
represented (under appropriate assumptions) with single-phase
equivalent circuits. More details on split-phase secondary circuit
modeling can be found in [4].
We follow the linear regresstion parameter estimation (LRPE)
method for secondary circuits described in [2] and [15]. The
LRPE method utilizes the linear approximation of voltage dif-
ference between two customers connected in parallel:
|V1 | − |V2 | ≈ IR2 R2 + IX2 X2 − IR1 R1 − IX1 X1 , (1)
1586
Fig. 1. Secondary circuit topology and parameter estimation problem [12].
where V1 and V2 are the voltages measured at customers 1 and 2,
IR1 and IR2 are the real currents (IR = P/V ) and IX1 and IX2
are the reactive currents (IX = Q/V ) flowing to each customer,
P is the measured real power, and Q is the measured reactive
power. This linear approximation typically has negligible error
[16]. By convention, we assign customers 1 and 2 such that the
voltage is, on average, higher at customer 1 than customer 2 (i.e.,
mean (V1 ) ≥ mean(V2 )).
Many utilities do not know the secondary circuit topologies.
In [12], we apply linear regression topology and parameter es-
timation (LRTE) algorithm to generate the entire secondary cir-
cuit models, including topology, using only the measurements.
The algorithm processes one secondary circuit at a time (e.g.,
the circuit containing all customer on a single transformer). For
each meter pair on that circuit, the algorithm solves the lin-
ear regression problem based on a slightly modified version
of Eq. (1):
V1 − V2 = IR2 R2 + IX2 X2 − IR1 R1 − IX1 X1 + .
where  has been added to account for a possible offset (bias
error) in the meter voltage measurements. Note this equation
applies to both the situation where the wires between the trans-
former (or some other upstream node) and the two customers
are parallel and when they are in series: for series connections,
IR1 and IX1 will be found to be zero by LRTE. The values of
R and X are recorded as the parameters for that meter pair, but
the value  is not recorded.
The meter pair with the best fit (as determined by the high-
est Pearson correlation coefficient R2 value) when using linear
regression is assumed to be the true topology. The method then
continues to estimate the remaining secondary circuit parame-
ters by proceeding from the tree leaf nodes towards the tree root
node. That is, the two customers found to be the best pair create
a virtual node at their meeting point. This virtual node is then
run through LRTE with the remaining meters on the circuit, with
additional virtual nodes created as needed, until the full circuit
topology is found.
III. STUDY FEEDERS AND MEASURED DATA
Three real distribution feeders were studied, each with
customer-level AMI data. These feeders were selected by the
utility as feeders of interest due to expected future high pene-
trations of DERs. In addition to customer AMI data, the utility
provided feeder models as seen in as seen in Fig. 2: Feeders 1 and
2 are long rural feeders, while Feeder 3 is a short urban feeder.
Included in these feeder models were details of the customers
IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 10, NO. 3, JULY 2019
Fig. 2. Layout of feeders studied. Colors indicate phases (white is 3 phase).
TABLE I
FEEDER PROPERTIES
connected to each service transformer. The feeder models also
included a mapping of the secondary system, though the utility
advised us that the secondary system topology was unverified.
Table I describes the Feeder properties. Both Feeders 1 and
2 have a high number of single-customer transformers, which
made the methodology described in IV-B especially impor-
(2) tant. All Feeders have similar distributions of transformer size
(10 kVA, 15 kVA, and 25 kVA are most common) and customer
max loads (individual customer loads rarely exceed 10 kW on
any feeder).
On Feeder 1, AMI rollout occurred in May 2016, while on
Feeders 2 and 3 AMI data existed since at least November 2015.
Data was available through October 2016, resulting in approxi-
mately 6 months of AMI data for Feeder 1 and 12 months of AMI
data for Feeders 2 and 3. The AMI measurements are recorded
every 15-minutes and include consumed kWh and kvarh, and
time-averaged voltage.
Fig. 3 shows the measurements from 2 customers on the same
transformer. As is typical for AMI data, the power consumption
of each customer is highly variable with large spikes during
the duty cycles of large appliances, electric water heaters, and
air conditioning. Each customers’ voltages, however, are more
consistent, as both customers are connected to the same service
transformer.
IV. METHODOLOGY
To compute the parameters and topology of all customers
on a feeder, we use the three-step process that follows. First, we
determine the topology and parameters of the secondary circuits
of transformers with more than one customer by pairing the
customers on the same transformer with one another. Next, we
pair customers who are the only customers on a transformer
(i.e., the transformer has only one customer connected to it)
with other such customers to derive the parameters for these
single customers. Finally, we pair transformers with multiple
LAVE et al.: DISTRIBUTION SYSTEM PARAMETER AND TOPOLOGY ESTIMATION
Fig. 3. Timeseries showing measured voltage, real power, and reactive power
for two customers on the same transformer.
customers with one another, using the virtual nodes found in the
first step, to determine if there is additional impedance between
the final virtual nodes and the transformers. Combined, these
three steps derive the secondary circuit topology and parameters
for all customers connected to the feeder. The three steps are
described in detail in the following sections.
A. Step 1: Transformers With Multiple Customers
For transformers with multiple customers connected, all pos-
sible pairs of customers on the same transformer are evaluated.
For example, for transformers with 3 customers, all 3 possible
combinations of customers pairs (1 and 2, 2 and 3, 3 and 1) are
evaluated. AMI data for voltage and real and reactive power, are
used to solve for R1 , X1 , R2 , and X2 using Eq. (2).
To understand how well the linear regression fits the data,
we computed the Pearson correlation coefficient (R2 ) of the fit.
This can be seen visually in Fig. 4, which shows the predicted
V1 − V2 using the right side of Eq. (2) with R1 , X1 , R2 , and X2
from linear regression on the x-axis plotted against the measured
V1 − V2 on the y-axis.
Once R2 values were computed for all customer pairs, the pair
with the highest R2 value was assigned as the actual topology.
If this pair was found to be connected in parallel, then a new
virtual node, representing the point where these two parallel
lines connect, was created. The voltage at the virtual node was
found by adding the voltage drop to the measured voltage:
Vvirtualnode = V1 + IR1 R1 + IX1 X1 .
Based on the convention mentioned in Section II, mean(V1 ) ≥
mean(V2 ). Thus, equation 3 ensures that the voltage at the vir-
tual node is, on average, higher than the voltage at each of the
customers. Real and reactive power consumption at the virtual
nodes were found by summing the real and reactive power of
the two customers.
As a second iteration, all remaining customers (those not con-
tributing to the virtual node) were again paired with one another
and also were paired with the virtual node. Again, the pair with
1587
Fig. 4. Relationship between measured V1 − V2 and predicted V1 − V2 based
on R and X values found in linear regression (as listed in the bottom right
of plot).
the highest R2 value was assigned as the actual topology, and
a virtual node was created. This method was repeated until all
customers were included in the topology. We note that this linear
regression is very fast, taking less than 0.1 s on a typical laptop
circa 2016. Because of the iterative nature, the number of linear
regressions that must be calculated increases with the number
of customers. However, total runtime remains modest even for
transformers with many customers: the linear regressions for all
pairs of a 20 customer transformer, for example, can be run in
about 2 minutes. Additionally, this process only needs to be run
once: DSPE derived parameters will continue to be valid unless
there is a system configuration change.
For most customers, the voltage fluctuations depended heavily
on the real current draw and only weakly on the reactive current.
That is, the reactive current was typically only responsible for
explaining a small fraction of the variance in the voltage drop, in
part because the real and reactive power draws were often well-
correlated. Therefore, we have more confidence in the derived
resistance values than reactance values. In some cases, including
when negative reactance is found, it may be best to use the found
resistance and then assume a reasonable X/R ratio to determine
the reactance.
B. Step 2: Transformers With Only One Customer
On transformers with a single customer, Step 1 will not work:
we only have one voltage measurement (at the one customer),
(3)
and so cannot solve for R and X between two voltage measure-
ments using Eq. 2. Instead, we find a nearby transformer that also
has a single customer by finding the shortest latitude/longitude
distance to another single-customer transformer on the same
phase. Because of the high voltage level on the primary system,
the per unit resistance on the primary voltage system between
two nearby transformers is much smaller than the per unit re-
sistance on the low voltage secondary system from transformer
to customer, as per unit resistance is inversely proportional to
voltage squared. For example, on a 12-kV system, 900 feet of
1588
TABLE II
ASSUMED TRANSFORMER RESISTANCES BASED ON kVA RATING [17]
wire has the same per unit resistance as approximately 1 foot
of wire on the low-voltage system. Thus, the primary voltage
side of two nearby service transformers can be assumed to be
identical with little impact to the estimated resistances. Thus,
two customers on nearby transformers (who are each the only
customers connected to their transformers) can be compared in
the same fashion as described in Step 1, as though they are on
the same transformer.
However, following this method means that we have included
the transformer resistance and reactance in the parameter es-
timation. Utilities typically have access to the kVA rating of
transformers, and hence the transformer resistance can be fairly
accurately estimated using the values in Table II, which are based
on National Electrification Association specifications. There are
no United States standards for transformer impedances, but we
used a typical value of 2% impedance for all transformer sizes.
The transformer resistance and reactance are subtracted out from
the estimated parameters to find the customer resistance and re-
actance. An additional benefit of this method is that including
the transformers in parameter estimation can validate or identify
errors in the transformer sizes in the utility models. Specifically,
the one noted time when a negative value resulted for customer
resistance, inspection found this was clearly a typographical er-
ror in transformer kVA, which was labeled as 1 kVA but was in
fact 10 kVA.
C. Step 3: Pair Multi-Customer Transformers With
One Another
The impedances found in Step 1 are not guaranteed to be the
complete impedances of the secondary circuit since the final
virtual node found may not be located at the transformer such
that there may be additional wiring between the virtual node and
the transformer leading to additional impedance. To account for
this possibility, we paired transformers with multiple customers
with one another. This is similar to Step 2, though instead of
using voltage and power measurements at single customers, we
used the calculated voltage and power at the final virtual nodes
found in Step 1.
In most cases, the final virtual node from Step 1 was at
the transformer, such that after subtracting out the transformer
impedance, essentially zero remaining impedance was found.
In some cases, slightly non-zero impedances were found (e.g.,
a resistance of 0.002 Ω), which are likely due to slight varia-
tion of transformer impedance from the assumed values listed
in Table II.
In some cases, though, the final virtual node from Step 1
was away from the transformer. For these cases, the additional
impedance of the wiring between the transformer and the virtual
node must be accounted for and added to the secondary topology
found in Step 1.
IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 10, NO. 3, JULY 2019
D. Data Quality Control
Customers with less than 1-week of data were not considered,
as we found the limited data often led to errant results (see
Section V-E). Measured customer voltages which deviated
by more than 5% from nominal voltage (i.e., V > 252 V or
V < 228 V) were not considered, since these may be unusual
grid states. An additional voltage filter, eliminating the times
of the 1% most extreme voltage measurements was applied.
This eliminated typically abrupt voltage excursions (e.g., the
three voltage dip seen on October 25th in Fig. 3) which also
were likely caused by a grid event and not by power draw by
the customers. Times when real or reactive power was zero or
negative were also eliminated, as they typically indicated no
(e.g., kW = 0) or errant (e.g., kW < 0) measurements.
Additional data challenges included errors in the secondary
model such as incorrect customer locations, incorrect phase la-
bels, incorrect customer to transformer pairings, and data mea-
surement errors including unreasonably large or small values.
Results presented in this paper have been manually inspected
for reasonableness of the model and measurement data. How-
ever, feeder-wide validation of DSPE results was difficult due
to these errors, especially inaccuracies in the secondary model
which were not systematic and were difficult to correct except
through manual inspection.
V. RESULTS
In this section, we present results of DSPE applied to the three
feeders. DSPE-found topology and parameters are compared to
the utility secondary model. For parameters, DSPE resistance is
compared to model distance (following conductors in the model)
using typical resistances per distance for different wire types
(Appendix A in [18]). For example, aluminum #2 American
Wire Gauge (AWG) triplex has a resistance of 0.058 Ω per 100 ft
(round-trip resistance for a 240 V load).
A. Transformers With Multiple Customers
Fig. 5 shows results for a transformer on Feeder 1 with two
customers. DSPE results (Fig. 5a) show a parallel configura-
tion, with similar resistances for the two customers, consitent
with the utility secondary model (Fig. 5b). This relationship be-
tween DSPE resistance and utility secondary model distance
from transformer to customer is plotted in Fig. 5c, and shows
that DSPE results are consistent with the secondary model dis-
tances, assuming #2 AWG wiring, which is commonly used by
this utility.
Fig. 6 shows results for a more complicated connection. Cus-
tomers 394 and 521 were first paired in parallel by DSPE to
form virtual node 1. This virtual node 1 was paired with cus-
tomer 579 in parallel to form virtual node two. Virtual node 3
was created by pairing customers 534 and 630. Finally, it was
determined that virtual nodes 3 and 4 had such little resistance
between them that they were, in fact, the same point. The utility
secondary model (Fig. 6b) confirms this topology. The com-
parison between DSPE resistance and utility secondary model
distance (Fig. 6c) suggests that customers 394 and 521 may be
LAVE et al.: DISTRIBUTION SYSTEM PARAMETER AND TOPOLOGY ESTIMATION 1589

Fig. 5. (a) DSPE results for a transformer with two customers on Feeder 1. (b) Satellite map showing utility secondary model for connections of customers to
this transformer. (c) Scatter plot of the DSPE resistances (y-axis) versus the utility secondary model distance (x-axis). Dashed lines show the resistance of different
wire types.

Fig. 6. (a) DSPE results for a transformer with five customers on Feeder 3. (b) Satellite map showing utility secondary model for connections of customers to
this transformer. (c) Scatter plot of the DSPE resistances (y-axis) versus the utility secondary model distance (x-axis). Dashed lines show the resistance of different
wire types.

Fig. 7. (a) DSPE results for two single-customer transformers Feeder 2. (b) Satellite map showing utility secondary model for these two customers. (c) Scatter
plot of the DSPE resistances (y-axis) versus the utility secondary model distance (x-axis). Dashed lines show the resistance of different wire types.

connected with #2 wire while customers 534, 579, and 630 may
be on lower resistance 2/0 wiring. This highlights the value of
DSPE: resistances from transformer to customer cannot be cal-
culated from secondary topology (or satellite imagery) alone;
accurate wiring type is also required. DSPE, resolves the topol-
ogy and parameters without need for the wire type as input.
B. Transformers With Only One Customer
DSPE results are shown in Fig. 7. The resistances found
through DSPE are consistent with 2/0 and 4/0 wiring. These
may be the actual wire types used as both customers 147 and
1046 are connected through underground wiring, which is likely
higher gauge. However, it is perhaps more likely that customer
1046 is connected using 2/0 wiring and the secondary model
distance is inaccurate (i.e., actual distance is shorter).
C. Pair Multi-Customer Transformers With One Another
Fig. 8 shows the results of pairing two neighboring multi-
customer transformers. In Fig. 8a, we see that for transformer
416, an extra 0.085 Ω of resistance were found between the
transformer and its final virtual node. This is consistent with
the satellite imagery and the utility-provided secondary model
1590
Fig. 8. (a) DSPE results for two transformers on Feeder 1, each with multiple customers. (b) Satellite map showing utility secondary model for these two
transformers. (c) Scatter plot of the DSPE resistances (y-axis) versus the utility secondary model distance (x-axis). Open circles show unadjusted parameters,
while filled in circles show parameters adjusted based on the additional resistance found between the final virtual node and the transformer. Dashed lines show the
resistance of different wire types.
Fig. 9. Distribution of customer resistances (x-axis) across each of the three
feeders.
(Fig. 8b), which show the wiring from customers 325 and 883
meets at a virtual node away from the transformer.
For transformer 415, this method found essentially zero addi-
tional resistance between the transformer and the virtual node,
which is also consistent with Fig. 8b. The found resistance was
actually slightly nonzero (−0.003 Ω), likely due to slight errors
estimating the transformer resistance, but we set the resitance
between the virtual node and the transformer equal to zero if the
DSPE found resistance is less than 0.005 Ω.
Fig. 8c compares the unadjusted and adjusted resistances to
the distances listed in the utility model. Adjusted resistances
were found by adding the resistance found between the virtual
node and the transformer (Fig. 8a) to the previously found resis-
tance in Step 1. Resistances for customers 325 and 883 increased
significantly when adjusted, and all 5 customer resistances are
consistent with #2 wiring.
D. Full Feeder Results
DSPE was run for all customers on Feeders 1, 2, and 3. Sum-
mary results are shown in Fig. 9. On all three feeders, it is most
IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 10, NO. 3, JULY 2019
Fig. 10. Measured voltage, and calculated voltage based on DPSE parameters
and a 100 ft of #2 wire assumption for customer 541 on feeder 1. The resistance
values for each method and the mean absolute errors (MAE) compared to the
measured voltage are shown in the text at the top.
common to have customer resistances around 0.05 Ω. However,
even though this is most common, there are a large number
of customers which vary significantly from this value: resis-
tances of 0.15 Ω were still common. This shows the value of
DSPE on a feeder-wide basis, as it can be used to determine
customer-specific parameters rather than relying on assumptions
which may, on average be accurate but which on a customer-by-
customer basis may be quite inaccurate.
The impact of errors in the resistance and reactance param-
eters when assuming a set value for all customers (e.g 0.05 Ω
resistance which is about 100 ft of #2 wire), is partially depen-
dent on the power draw of each customer. Customers with high
power draw will have large discrepancies between measured
voltage and voltage found based on the resistance assumption.
Two extreme examples can be seen in Figs. 10 and 11. Customer
541 on Feeder 1 (Fig. 10) is a dairy barn with high power draw
(up to 40 kW) and low DSPE resistance between the transformer
and customer meter. Assuming 100 ft of #2 wire instead of us-
ing DSPE results in an overestimation of voltage drop, resulting
in a large underestimation of the customer voltage. Conversely,
customer 1089 on Feeder 3 (Fig. 11) is a mini-golf course with
high power draw (up to 10 kW) that has high resistance between
the transformer and the customer meter, and the 100 ft of #2
wire assumption overestimates the customer voltage.
Building on these examples, Fig. 12 shows the mean absolute
errors for voltage modeled using either DSPE parameters or the
LAVE et al.: DISTRIBUTION SYSTEM PARAMETER AND TOPOLOGY ESTIMATION
Fig. 11. Same as Fig. 11, but for customer 1089 on feeder 3.
Fig. 12. Binned mean absolute errors (x-axis), measured versus modeled volt-
age, for all customers on two-customer transformers. The y-axis shows the num-
ber of customers in each bin. Modeled voltages are either from DSPE parameters
(solid lines) or based on the 100 ft assumption (dashed lines).
100 ft of #2 wire assumption for all customers on two-customer
transformers. Over the three feeders, there are a total of 113
customers for which the 100 ft assumption leads to a larger
than 1 V mean absolute error in modeled voltage, versus only
24 customers with greater than 1 V errors when using DSPE
parameters.
E. Sensitivity to Measurement Errors
We applied filters to the input data to test the sensitivity of
DSPE results to input data quality of voltage, real power, and
reactive power AMI measurements. Fig. 13 shows the sensitiv-
ity to the amount of input data. For transformer 233 (the trans-
former in Fig. 5), the DSPE resistances of both customers when
considering only 2,000 data points was within 5% of the value
found when considering all 17,023 data points. 2,000 data points
corresponds to approximately 21 days of 15-minute resolution
data. Other transformers had similar results: of 25 transform-
ers with two customers that were evaluated for sensitivity to
the amount of data, all but three had converged to less than 5%
difference from full data when 8,000 data points were consid-
ered (less than 3 months of 15-minute resolution data). When
this difference threshold was relaxed to 10%, all 25 test trans-
formers had converged by 6,200 data points (about 2 months of
15-minute data).
Next, we examined the sensitivity of DSPE results to vary-
ing resolutions of voltage, real power, and reactive power mea-
surements, as shown in Fig. 14. The actual resolution of the
1591
Fig. 13. DSPE sensitivity to the amount of data.
Fig. 14. DSPE sensitivity to the resolution of voltage (top), real power
(middle), and reactive power (bottom). The reactive power plot has a special
entry on the far left showing the DSPE resistance found when not using reactive
power measurements at all (“no meas.”).
measurements was 0.1 V, 0.04 kW, and 0.04 kvar; worse resolu-
tions were simulated by rounding the data to different intervals.
The sensitivity to voltage resolution was small. For trans-
former 233, results within 5% of the full resolution case were
obtained for voltage resolutions as poor as 1 V. Of the other two-
customer transformers evaluated, 20 of the 25 needed voltage
resolution of only 1 V to maintain DSPE results with 5% of full
resolution data; and all 25 were within 10% of full resolution
data when using only 1 V resolution voltage.
Real power resolutions of 0.5 kW or less resulted in greater
than 5% deviations in DPSE resistance values versus full
resolution. Over the 25 test transformers, 18 of 25 resulted in
less than 5% deviation from full resolution data when using
0.5 kW resolution power and 23 of 25 resulted in less than 10%
deviation. Reactive power resolutions of worse than 0.5 kvar
similarly led to greater than 5% deviations from full resolu-
tion results. However, accurate DSPE results did not depend on
1592
having reactive power measurements at all. As seen by the dots
in the bottom plot of Fig. 14 when no reactive power measure-
ments are used, DSPE results are very similar (less than 3%
difference) to results when using full resolution reactive power
measurements. Similarly, over the 25 test feeders, the differ-
ence between using no reactive power measurements at all and
using full resolution reactive power measurements was always
less than 10%. The generally high correlation between real and
reactive power means that measurement of reactive power is
not very important to DSPE accuracy and is an important result
- reactive power collection may not be necessary for accurate
DSPE.
Adding stochastic errors to the AMI measurements introduced
stochastic differences in DSPE results. However, for added
stochastic measurement errors ranging from 0% to 10% of av-
erage measured voltage, real power, or reactive power, DSPE
results were still within 5% of the results with no stochastic er-
rors added. Constant bias errors in measurements (i.e., if voltage
measurements are 2 V too high over all timesteps) have no im-
pact on DSPE results, as they simply change the  value in Eqn.
(2), but do not impact the resistance or reactance values.
VI. CONCLUSION
Distribution system parameter estimation (DSPE) results
agreed well with imagery and the utility secondary model for
the customers examined. When applied over entire feeders,
significant variations were seen from the simple assumption
that all customers have 100 ft of #2 wire, demonstrating the
value of DSPE. Sensitivity analysis showed that the DSPE
method is robust: modest amounts of data (around 3 months
of 15-minute resolution data) and measurement resolutions (1
V, 0.5 kW) were sufficient to derive accurate parameters. For
this set of mostly residential customers, reactive power mea-
surements were not found to be necessary, but this may change
for other customers with less correlated real and reactive power
consumptions.
This paper supports the integration of the presented DSPE
method into utility distribution grid management software solu-
tions. The presented DSPE method can be leveraged to identify
and fix errors in the utility geographic information system (GIS)
data. It can also be leveraged to automatically create accurate and
detailed secondary system models based on AMI measurements
while minimizing the need for costly manual labor and field in-
spections. As a result, the presented practical DSPE method can
help utilities to greatly increase the accuracy and detail of their
distribution grid models to enable high DER penetrations.
Future work building on the method presented here should
focus on validation of feeder-wide results (which was difficult
with this data set due to the unvalidated utility model) and on
integration into utility software for wide implementation.
ACKNOWLEDGMENT
Sandia National Laboratories is a multimission laboratory
managed and operated by National Technology & Engineer-
ing Solutions of Sandia, LLC, a wholly owned subsidiary of
Honeywell International Inc., for the U.S. Department of
IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 10, NO. 3, JULY 2019
Energy’s National Nuclear Security Administration under con-
tract DE-NA0003525. This paper describes objective technical
results and analysis. Any subjective views or opinions that might
be expressed in the paper do not necessarily represent the views
of the U.S. Department of Energy or the United States Govern-
ment. SAND2018-8168 J.
REFERENCES
[1] B. Palmintier et al., “On the path to sunshot. Emerging issues and
challenges in integrating solar with the distribution system,” National
Renewable Energy Laboratory, Golden, CO, USA, Tech. Rep. NREL/TP-
5D00-65331; SAND2016-2524 R, 2016.
[2] J. Peppanen, M. J. Reno, R. J. Broderick, and S. Grijalva, “Distribution sys-
tem secondary circuit parameter estimation for model calibration,” Sandia
National Laboratories, Albuquerque, NM, USA, Tech. Rep. SAND2015-
7477, 2015.
[3] J. Peppanen, C. Rocha, J. A. Taylor, and R. C. Dugan, “Secondary low-
voltage circuit models; how good is good enough?” in Proc. IEEE Rural
Elect. Power Conf., Apr. 2017, pp. 17–26.
[4] J. Peppanen, M. Bello, and M. Rylander, “Service entrance hosting ca-
pacity,” in Proc. IEEE 7th World Conf. Photovolt. Energy Convers. (Joint
Conf. 45th IEEE PVSC, 28th PVSEC, 34th EU PVSEC), Waikoloa Village,
HI, USA, 2018, pp. 1451–1456. doi: 10.1109/PVSC.2018.8547340.
[5] S. Jothibasu and S. Santoso, “Sensitivity analysis of photovoltaic hosting
capacity of distribution circuits,” in Proc. IEEE Power Energy Soc. Gen.
Meeting, Jul. 2016, pp. 1–5.
[6] J. Yu, Y. Weng, and R. Rajagopal, “Patopa: A data-driven parameter and
topology joint estimation framework in distribution grids,” IEEE Trans.
Power Syst., vol. 33, no. 4, pp. 4335–4347, Jul. 2018.
[7] Y. Weng, Y. Liao, and R. Rajagopal, “Distributed energy resources topol-
ogy identification via graphical modeling,” IEEE Trans. Power Syst.,
vol. 32, no. 4, pp. 2682–2694, Jul. 2017.
[8] A. M. Prostejovsky, O. Gehrke, A. M. Kosek, T. Strasser, and H. W. Bind-
ner, “Distribution line parameter estimation under consideration of mea-
surement tolerances,” IEEE Trans. Ind. Inform., vol. 12, no. 2, pp. 726–735,
Apr. 2016.
[9] T. A. Short, “Advanced metering for phase identification, transformer iden-
tification, and secondary modeling,” IEEE Trans. Smart Grid, vol. 4, no. 2,
pp. 651–658, Jun. 2013.
[10] R. Singh, B. C. Pal, and R. A. Jabr, “Distribution system state estimation
through gaussian mixture model of the load as pseudo-measurement,” IET
Gener., Transmiss. Distrib., vol. 4, no. 1, pp. 50–59, Jan. 2010.
[11] S. Neshvad, H. Margossian, and J. Sachau, “Topology and parameter
estimation in power systems through inverter-based broadband stimula-
tions,” IET Gener., Transmiss. Distribution, vol. 10, no. 7, pp. 1710–1719,
May 2016.
[12] J. Peppanen, S. Grijalva, M. J. Reno, and R. J. Broderick, “Distribution
system low-voltage circuit topology estimation using smart metering data,”
in Proc. IEEE/PES Transmiss. Distrib. Conf. Expo., 2016, pp. 1–5.
[13] J. Peppanen, S. Grijalva, M. J. Reno, and R. J. Broderick, “Secondary
circuit model generation using limited pv measurements and parameter
estimation,” in Proc. Power Energy Soc. Gen. Meeting, 2016, pp. 1–5.
[14] J. Peppanen, S. Grijalva, M. J. Reno, and R. J. Broderick, “Secondary
circuit model creation and validation with ami and transformer measure-
ments,” in Proc. North Amer. Power Symp., 2016, pp. 1–6.
[15] J. Peppanen, M. J. Reno, R. J. Broderick, and S. Grijalva, “Distribution
system model calibration with big data from ami and PV inverters,” IEEE
Trans. Smart Grid, vol. 7, no. 5, pp. 2497–2506, Sep. 2016.
[16] C. W. Brice, “Comparison of approximate and exact voltage drop calcula-
tions for distribution lines,” IEEE Trans. Power App. Syst., vol. PAS-101,
no. 11, pp. 4428–4431, Nov. 1982.
[17] F. P. Villareal, “Nea specifications on distribution transformers,” NEA
Report, Jun. 2015. [Online]. Available: https://www.google.com/url?sa=
t&rct=j&q=&esrc=s&source=web&cd=9&cad=rja&uact=8&ved=2
ahUKEwiqu-7Gus7iAhXvFzQIHVLpD9wQFjAIegQIBBAC&url=http
%3A%2F%2Fnea.gov.ph%2Findex.php%3Foption%3Dcom_phocadown
load%26view%3Dcategory%26download%3D1231%3Adistribution-
transformer-specifications%26id%3D103%3Aconsultative-session-with-
leadership-associations-of-ecs%26Itemid%3D143&usg=AOvVaw25k3
KK1JmUgDHE9PuwluWT
[18] W. Kersting, Distribution System Modeling and Analysis, 3rd ed. New
York, NY, USA: Taylor & Francis, 2012.