IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 21, NO.

1, FEBRUARY 2006 153

A Unified Three-Phase Transformer Model

for Distribution Load Flow Calculations
Peng Xiao, Student Member, IEEE, David C. Yu, Member, IEEE, and Wei Yan

Abstract—This paper provides a unified method to model

three-phase transformers for distribution system load flow cal-
culations, especially when the matrix singularity caused by the
transformer configuration arises. This paper shows that the singu-
larity appears only in certain transformer admittance submatrices
and only in certain transformer configurations. The unified
method presented in this paper can solve the voltage/current equa-
tions in the forward/backward sweep algorithm for various types
of transformer configurations, whether or not the corresponding
admittance submatrices are singular. Comprehensive compar-
isons have been made between the proposed approach and other
methods. Test results demonstrate the validity and effectiveness of Fig. 1. General three-phase transformer model.
the proposed method.
Index Terms—Admittance matrix, load flow analysis, power
distribution, power transformers. A two-block three-phase transformer model was presented
in [4]. As shown in Fig. 1, a series block represents winding
connections and leakage impedance, and a shunt block models
I. INTRODUCTION real and reactive power losses in the transformer core. A similar
model was proposed in [5], where the shunt block is connected
A S the power distribution networks become more and more
complex, there is a higher demand for efficient and reli-
able system operation. Consequently, two of the most important
to the primary side.
The core loss of a transformer is approximated by shunt core
system analysis tools, power flow and short-circuit studies, must loss functions on each phase of the secondary terminal of the
have the capability to handle various system configurations with transformer. Generally, the functions are nonlinear and the coef-
adequate accuracy and speed. Of the several dedicated distribu- ficients should be determined by experiments. Since transformer
tion system load flow methods used in the power industry, the winding connections have little effect on core loss, this paper fo-
forward/backward sweep algorithm [1], with its low memory cuses mainly on the series block, while the core loss block can
and computation requirements and robust convergence charac- be treated as a three-phase load on either side of the transformer.
teristic, has gained the most popularity in recent years. The algo- To include such a transformer model into the forward/back-
rithm is also known to be the branch current-based feeder anal- ward sweep algorithm, specific voltage/current relationships
ysis. Based on the ladder theory for linear circuit analysis, the should be derived. Several approaches have been developed in
forward/backward sweep algorithm can fully utilize the radial the past several years. In [4], fictitious injection current sources
structure of most distribution networks. With minor modifica- were used to resolve the coupling between the primary and sec-
tions, several methods [2], [3] have been proposed to extend its ondary sides, which greatly simplified the admittance matrix.
application in weakly meshed distribution systems. However, this technique faces slow convergence problems when
To take into account the existence of multiphase load and un- employed in the forward/backward sweep algorithm. In [5],
symmetrical feeders in distribution systems, three-phase system voltage and current update equations were developed for three of
representation is generally used. The modeling of three-phase the most commonly used transformer connections based on their
transformers is a vital step in distribution system analysis. Due equivalent circuits. In [9], voltage/current equations were derived
to unbalanced system operations, a complete and accurate three- in matrix form for transformers of the ungrounded wye-delta
phase model is desirable for distribution and inline transformers connection. However, these methods are mainly based on circuit
of various core and winding configurations. analysis with Kirchhoff’s voltage and current laws. Different set
of equations are needed to handle each connection type. Trans-
formers of various winding connections can only be analyzed
Manuscript received August 18, 2004; revised January 8, 2005. This work was on a case-by-case basis, which is not convenient for efficient
supported in part by the Visiting Scholar Foundation of the Key Laboratory of implementation and incorporation of new connection types.
High Voltage Engineering and Electrical New Technology, Education Ministry,
China. Paper no. TPWRS-00450-2004. In this paper, a unified method is proposed to model the
P. Xiao and D. C. Yu are with the Electrical Engineering Department, Uni- diverse distribution transformers into the forward/backward
versity of Wisconsin-Milwaukee, Milwaukee, WI 53211 USA. sweep load flow algorithm. A brief description of the for-
W. Yan is with the Electrical Power Department, Chongqing University,
Chongqing 400044, China. ward/backward sweep algorithm is presented in Section II. In
Digital Object Identifier 10.1109/TPWRS.2005.857847 Section III, the proposed modeling procedure is explained in
0885-8950/$20.00 © 2005 IEEE
154 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 21, NO. 1, FEBRUARY 2006

on estimated bus voltages, incorrect voltage values will result

in incorrect branch currents. In forward sweep, the root voltage
information, together with the branch currents obtained from the
last backward sweep, are used to update the leaf node voltages.
Starting from the root node, the voltages at all the child nodes
are calculated based on the parent bus voltages and the currents
Fig. 2. One section of a distribution network. in the branches. The procedure continues until all the leaf node
voltages are updated.
detail. Extensive computation and comparisons have been done As indicated in Fig. 2, the forward sweep requires that the
to verify the method, and the results are presented in Section IV. branch be modeled in a way that can be obtained when
and are known.
The proposed method for modeling distribution transformer
II. FORWARD/BACKWARD SWEEP ALGORITHM
satisfies the requirements of the forward/backward sweep algo-
Although the forward/backward sweep algorithm can be ex- rithm and can be easily integrated into the algorithm, regardless
tended to solve systems with loops and distributed generation of the types of transformer configurations.
buses, a radial network with only one voltage source is used here
to depict the principles of the algorithm. Such a system can be III. TRANSFORMER MODELING PROCEDURE
modeled as a tree, in which the root is the voltage source, and
A. Construction of the Primitive Admittance Matrix
the branches can be a segment of feeder, a transformer, or other
components between two buses. The starting point for the proposed modeling approach is the
With the given voltage magnitude and phase angle at the root primitive admittance matrix of the transformer, which
and known system load information, the power flow algorithm can be determined according to the transformer winding con-
needs to determine the voltages at all other buses and currents in nections. Depending on the system representation, both per-unit
each branch. The forward/backward sweep algorithm employs values or actual unit values can be used to form the matrix. A
an iterative method to update bus voltages and branch currents. discussion of the advantages and disadvantages of both unit sys-
During each iteration, a backward sweep is performed to update tems was presented in [6].
branch currents, and a forward sweep is performed to update bus
voltages. The algorithm terminates when the voltages converge. B. Conversion to Nodal Admittance Matrix
A brief description of the main steps of the forward/backward The conversion from the primitive admittance matrix
sweep algorithm is shown as follows. to the nodal admittance matrix was discussed in detail in
[8]. Basically, the procedure involves using the node-to-branch
A. Initialization incidence matrix with
Before the first iteration, the voltage at each leaf node, i.e., the (1)
buses that do not have child buses, is given an initial guess value.
The guess values should be as close to the true values as possible The matrix can then be reduced to 6 6 by eliminating
to reduce the number of iterations and to avoid divergence. Typ- the neutral point nodes with Kron reduction. The resulted matrix
ically, the voltage magnitude is set to one per unit, while the is the nodal admittance matrix .
voltage phase angle can be chosen by considering transformer In [10], a systematic approach utilizing symbolic mathemat-
phase shift between the root and the leaf nodes. ical tools was proposed to establish the nodal admittance ma-
trices for various transformer connection types. Minor modifica-
B. Backward Sweep tions are necessary to take into account the off-nominal tap ratio
With voltages at all the leaf nodes given, the backward sweep between the primary and secondary sides of the transformer. The
procedure determines the currents in the branches that connect matrices for the most common transformer configurations
these nodes and their parent nodes and the voltages at the parent were proposed in [4].
nodes. When the voltages at the parent nodes are calculated, Following the same transformer modeling procedures, it
these nodes can then be treated as “leaf” nodes and the calcula- is straightforward to build the nodal admittance matrices for
tion continues until the root is reached. the more particular transformer configurations such as Open
Fig. 2 shows one section of the system, in which bus is the Wye/Open Delta. Once these matrices are determined, they
parent node and bus is the leaf node. Each step of the backward can be used in the forward/backward algorithm. However, this
sweep is to determine the currents flowing through the branch paper does not discuss these more specialized transformer
and the voltages at bus . Therefore, in the backward configurations due to space limitations.
sweep, the branch components need to be modeled in a way that
C. Characteristics of the Submatrices
and can be obtained when and are known.
With the nodal admittance matrix , the transformer
C. Forward Sweep voltage-current relationship can be expressed as
After backward sweep, the currents in all branches are up-
(2)
dated. However, since calculations of these currents are based
XIAO et al.: UNIFIED THREE-PHASE TRANSFORMER MODEL FOR DISTRIBUTION LOAD FLOW 155

TABLE I TABLE II
Y SUBMATRICES FOR COMMON STEP-DOWN TRANSFORMER CONNECTIONS Y SUBMATRICES FOR COMMON STEP-UP TRANSFORMER CONNECTIONS

where the matrix is divided into four 3 3 submatrices: From (7) and (8), it is obvious that both and are
, and . Vectors , and are the singular. Hence, is invertible only for connection,
three-phase line-to-neutral bus voltages and injection currents at and is invertible only for and connections.
the primary and secondary sides of the transformer, respectively. For connection, (3)–(5) can be directly used for forward
As described in the last section, in backward sweep proce- and backward sweep calculations. For connection, only
dure, and are known, while and are to be calcu- (5) can be used in forward sweep. For all other connection types,
lated. From (2), the following can be derived: since the matrices are singular, there is no unique solution to the
above equations. In essence, the singularity in those transformer
(3) configurations arises due to the lack of voltage reference point
(4) on one or both sides of the transformer.

In forward sweep, and are known, and needs to be D. Solving the Singularity Problem
calculated. Similarly To circumvent the singularity issue, it is noted that although
the three-phase line-to-neutral voltages and cannot be
(5)
obtained by solving (3) and (5), the nonzero-sequence compo-
According to (3)–(5), the implementation of the for- nents of the voltages can be uniquely determined. To illustrate
ward/backward sweep algorithm requires the inversion of this point in the backward sweep, rewrite (3) as
submatrices and . However, a close examination (9)
of the matrices for common transformer configurations
shows that these submatrices are often singular. Table I shows Let represent the nonzero-sequence components of ,
the submatrices of for the nine most common step-down i.e.,
transformer connection types, and Table II shows the matrices
for step-up transformers, where (10)

where vector is the zero-sequence voltage on the primary

(6) side; thus

(11)
(7) The product of and is always zero for transformer
configurations other than . This is because is rep-
resented by or in all other transformer configurations
(8) except . From (7) and (8), it can be seen that

(12)
and is the per-unit transformer leakage admittance. For sim-
plification, the leakage admittances of each phase are assumed so (11) can be reduced to
to be identical. For transformers with unbalanced admittances, (13)
their nodal admittance matrices are more complex and do not
take the forms shown in Tables I or II. However, it is proved Equation (13) indicates that the zero-sequence component of
that the singularity of the transformer submatrices remains the does not affect the backward sweep calculation for trans-
same. formers with a singular matrix. The above analysis shows
156 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 21, NO. 1, FEBRUARY 2006

that (3) can be used to calculate both and its nonzero-se-

quence components . However, if is singular, (13) still
cannot uniquely determine , and a modification is needed.
Since does not contain zero-sequence component, it satisfies

(14) Fig. 3. Four-bus example system.

Equations (13) and (14) can be combined as

Equation (20) can then be transformed into
(15)

where is obtained by replacing the last row of with

, while and are the same as and , except
that elements in their last row are set to 0 so that (14) is satisfied.
Now that is not singular, the nonzero-sequence compo- (22)
nents of the voltages on the primary side can be determined by

(16)

Similar results can be obtained for forward sweep calculation E. Modified Forward/Backward Sweep Algorithm
With the above transformation, the equations for transformer
(17) voltage calculation are no longer singular. However, the resulted
transformer voltages and only contain the positive and
where is the nonzero-sequence component of negative sequence components. Thus, zero-sequence voltages
is the same as , except that the last row is replaced with must be added to them to form line-to-neutral voltages. The
, and are obtained by setting the elements in primary-side zero-sequence voltage can be initialized to 0 and
the last row of and to 0, respectively. is updated during the forward sweep. Since voltages are also
Once the nonzero-sequence components of or are cal- calculated when line currents are updated, the secondary-side
culated, zero-sequence components are added to them to form zero-sequence voltage can be obtained directly from the back-
the line-to-neutral voltages so that the forward/backward sweep ward sweep.
procedure can continue. To illustrate the modified procedure, a four-bus example
As an example, consider the backward sweep for an un- shown in Fig. 3 is used.
grounded transformer. According to Table I 1) Initialization: The forward/backward sweep algorithm
begins with all the load information and only the source voltage
known. A guess value is given to the voltages on bus 4.
(18) 2) Backward Sweep: With given, the load currents can
be calculated. If the transformer core loss is modeled on the sec-
and ondary side, core loss functions can be used to determine the
absorbed power and current. Thus, the three-phase currents
that flows through feeder 3–4 can be obtained. Assume that line
(19) charging is neglected, then the currents flowing through the sec-
ondary side of the transformer are equal to . The voltages
on the secondary side of the transformer are
Thus, (13) becomes

where is the line impedance matrix for feeder segment

3–4.
For the -connected transformer, its matrix is not
(20) invertible. According to (16)

The singularity of matrix is evident. However, the matrix

can be changed to nonsingular if one of its rows is replaced with
, i.e., Since only contains the positive-and negative-sequence
components of , an initial value of zero-sequence voltage
is needed to get the transformer primary side voltage
(21)
XIAO et al.: UNIFIED THREE-PHASE TRANSFORMER MODEL FOR DISTRIBUTION LOAD FLOW 157

Next the currents at the ungrounded Wye side can be deter- TABLE III
mined with V DURING ITERATIONS

The backward sweep calculation continues until the source

bus is reached. If the difference between the computed source
voltage and the actual source voltage is not within the
required error limit, a forward sweep is performed.
3) Forward Sweep: The forward sweep begins by setting
to its actual value . Since the currents in feeder sections 1–2
have been calculated in the backward sweep, the voltage at bus
2 can be obtained by

grounded elements that are connected to the same side of the

transformer, such as grounded load, line charging, or other
where is the line impedance matrix for feeder segment grounded transformers. However, due to the limitations of for-
1–2. Note that by updating , the zero-sequence component ward/backward sweep algorithm [11], zero-sequence voltage
is also updated and will be used in the next backward sweep cannot be updated. Therefore, the secondary line-to-neu-
procedure. tral voltages in these conditions are the assumed value. The
The next step is to determine based on the knowledge common practice in the utility industry will seldom see a single
of and . Still, due to matrix singularity, only positive-and phase-to-ground load connected to an ungrounded three-phase
negative-sequence components can be determined source or transformer, so the zero-sequence current is generally
very small compared with load currents. Thus, even if the
calculated line-to-neutral voltage is not real, the corresponding
line-to-line voltage is still accurate.
The voltage can then be calculated by adding zero-se-
quence voltage , which is obtained from the last value of IV. TEST EXAMPLES
in backward sweep. The forward sweep then continues until bus
To verify the proposed method, different transformer configu-
4 is reached.
rations were included in a three-phase distribution load flow pro-
4) Zero-Sequence Voltage Update: As illustrated above, for
gram with a forward/backward sweep algorithm. Two different
transformers with singular or submatrices, zero-se-
sized examples were implemented, and results were compared
quence voltage update calculation cannot be performed from
with other existing methods.
one side to the other. This is due to the fact that zero-se-
quence equivalent circuit is interrupted at a transformer with
ungrounded Y and/or windings. Hence, the zero-sequence A. Model Validation
voltage on one side of the transformer cannot be determined The IEEE four-node test feeder [12] was used to provide a
based on line-to-neutral voltages on the other side, even when simple system for the testing of various three-phase transformer
line current information on both sides is available. In these connections. The one-line diagram of the example is shown in
cases, the zero-sequence voltages on the primary side can be Fig. 3.
updated by the source voltage during the next forward sweep. The two distribution feeder segments have unequal mutual
However, the zero-sequence voltages on the secondary side coupling between the phases. The load is unbalanced with dif-
with a or ungrounded Y winding will not be updated due to ferent kilovoltampere and power factor in each phase.
the lack of voltage reference point on the secondary side, which For an ungrounded transformer connection, the pro-
makes determining the real line-to-neutral voltage impossible. gram terminates after five iterations, and the difference between
To avoid such difficulties, the method in [9] used the line- computed source voltage and specified source voltage
to-line voltage on the or ungrounded Y side of the trans- are within 0.0001 p.u. The -phase line-to-line voltages at the
former and the line-to-neutral voltage in the remaining part of load are listed in Table III.
the system. The method in [5] introduced an arbitrary refer- The results match very well with those listed in [9], where a
ence neutral point to convert the line-to-line voltages to the detailed step-by-step example was given. Tests have been con-
line-to-neutral voltages. The proposed method, using an initially ducted for all other transformer connections, and the results
guessed zero-sequence voltage on the primary side, does not re- match those listed in [12].
quire any special treatment in handling the or ungrounded However, the method mentioned in [9] and [12] requires dif-
Y transformers. With this initial zero-sequence voltage, line-to- ferent sets of equations for different transformer configurations,
neutral voltages can be used throughout the system. which is not convenient for efficient implementation and incor-
The primary-side zero-sequence voltages are updated by the poration of new connection types. The unified method proposed
source voltage. The zero-sequence voltage on the secondary in this paper does not require any special equations to handle the
side with a or ungrounded Y winding are determined by transformers with the or ungrounded Y windings.
158 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 21, NO. 1, FEBRUARY 2006

TABLE IV TABLE VI
ZERO-SEQUENCE VOLTAGES AT BUS 2 LINE-TO-LINE VOLTAGES AT BUS 4

TABLE V
ZERO-SEQUENCE VOLTAGES AT BUS 4

B. Zero-Sequence Voltage Update Fig. 4. IEEE 123-bus example system.

Different transformer configurations were implemented in

the four-bus example to further test the updating behavior of other side together with current information. For other connec-
the zero-sequence voltage. Two different initial values (0 and tions, a close examination reveals that the final values are in the
0.1 p.u.) were chosen for each configuration. If different initial vicinity of the initial value. In other words, zero-sequence volt-
values always converge to the same final value, then it means ages are not updated, even though the algorithm converges. This
the zero-sequence voltage can be correctly updated during the is due to the fact that without additional grounding devices on
iterations for the given type of transformer configuration. the secondary side, the subnetwork is isolated, and the zero-se-
Table IV shows the final zero-sequence voltages on the pri- quence voltage will not be affected by other part of the system.
mary side of the transformer for different connection types. The Table VI shows the corresponding line-to-line voltages at
results show that, for each transformer configuration, the pri- bus 4 for different transformer configurations with different
mary side zero-sequence voltage always converges to a similar initial zero-sequence voltages. The results indicate that even
value, regardless of its initial value. These results indicate that though different initial values may produce different final
the primary side zero-sequence voltage can be updated. This up- zero-sequence voltages, the corresponding line-to-line voltages
date takes place during the forward sweep, and it is due to the are correct when the algorithm converges, which meets the
Y-grounded source voltage. requirement for most load flow analysis.
Table V shows the final zero-sequence voltages on the sec-
ondary side (bus 4) of the transformer. The results indicate that C. Large System Tests
for some transformer configurations, different initial values will The IEEE 123-bus example shown in Fig. 4 is used to demon-
produce different final values. This is true for every transformer strate the transformer models in large systems. The load flow
configuration except and . The results further analysis calculation was performed using per-unit values on a
show that forward/backward sweep algorithm cannot produce basis of 115 kV/4.16 kV/480 V and 10 MVA. There are two
unique secondary line-to-neutral voltages in these cases. It transformers in this system. One is located between nodes 150
should be noted that the initial zero-sequence voltage is and 149, and the other is between nodes 61 and 610.
obtained from the initial leaf-node voltages . Extensive tests have been performed on the system to verify
For and transformers, zero-sequence voltage the validity of the proposed method. Comparisons have been
on one side can be uniquely determined from voltages on the made against transformer modeling approaches developed in [4]
XIAO et al.: UNIFIED THREE-PHASE TRANSFORMER MODEL FOR DISTRIBUTION LOAD FLOW 159

TABLE VII the corresponding zero-sequence voltage cannot be updated.

ITERATION NUMBERS FOR DIFFERENT SYSTEM LOADING CONDITIONS This paper also proves that the line-to-neutral voltage, ex-
cluding its zero-sequence component, can still be used in
the forward/backward sweep. Based on these findings, the
proposed technique separates the zero and nonzero sequence
voltage during the forward/backward sweep, regardless of the
transformer configuration to avoid the singularity problem.
For certain transformer connections, the zero-sequence voltage
on the secondary side cannot be updated during the sweeps.
The proposed method was implemented in two different sized
examples, and tests were conducted to compare with other
existing approaches. The results show the validity and effec-
tiveness of the proposed technique.

REFERENCES
TABLE VIII [1] W. H. Kersting, Distribution System Modeling and Analysis. Boca
ITERATION NUMBERS FOR DIFFERENT R/X RATIOS Raton, FL: CRC, 2002.
[2] G. J. Chen, K. K. Li, T. S. Chung, and G. Q. Tang, “An efficient two-stage
load flow method for meshed distribution networks,” in Proc. APSCOM,
2000, pp. 537–542.
[3] M. H. Haque, “Efficient load flow method for distribution systems with
radial or mesh configuration,” Proc. Inst. Elect. Eng., Gener., Transm.,
Distrib., vol. 143, no. 1, pp. 33–38, Jan. 1996.
[4] T. H. Chen, M. S. Chen, T. Inoue, P. Kotas, and E. A. Chebli, “Three-
phase cogenerator and transformer models for distribution system anal-
ysis,” IEEE Trans. Power Del., vol. 6, no. 4, pp. 1671–1681, Oct. 1991.
[5] M. E. Baran and E. A. Staton, “Distribution transformer models for
branch current based feeder analysis,” IEEE Trans. Power Syst., vol. 12,
no. 2, pp. 698–703, May 1997.
[6] R. C. Dugan, “A perspective on transformer modeling for distribution
system analysis,” in Proc. IEEE Power Eng. Soc. General Meeting, vol.
and [5]. For convenience, the proposed modeling method is la- 1, Jul. 2003, pp. 114–119.
beled model 1 in the tables, and methods in [5] and [4] are la- [7] M. J. Gorman and J. J. Grainger, “Transformer modeling for distribution
system studies part I: Linear modeling basics,” IEEE Trans. Power Del.,
beled model 2 and model 3, respectively. vol. 7, no. 2, pp. 567–574, Apr. 1992.
Table VII compares the results under different loading condi- [8] , “Transformer modeling for distribution system studies part II: Ad-
dition of models to Y and Z ,” IEEE Trans. Power Del., vol. 7,
tions, and Table VIII compares the results under different R/X no. 2, pp. 575–580, Apr. 1992.
ratios. It appears that there is a strong agreement in terms of [9] W. H. Kersting and W. H. Phillips, “A new approach to modeling three-
the resultant bus voltages and branch currents among the three phase transformer connections,” IEEE Trans. Ind. Appl., vol. 35, no. 1,
pp. 169–174, Jan. 1999.
models under all test conditions. [10] M. R. Irving and A. K. Al-Othman, “Admittance matrix models of
From the results, it is observed that as the load increases, the three-phase transformers with various neutral grounding configura-
number of iterations for convergence increases irrespective of tions,” IEEE Trans. Power Syst., vol. 18, no. 3, pp. 1210–1212, Aug.
2003.
the models. However, model 1 and model 2 appear to be less [11] Z. Wang, F. Chen, and J. Li, “Implementing transformer nodal admit-
sensitive than model 3. Results further reveal that the iteration tance matrices into backward/forward sweep-based power flow analysis
for unbalanced radial distribution systems,” IEEE Trans. Power Syst.,
numbers for model 3 are larger than that of model 1 and 2. The vol. 19, no. 4, pp. 1831–1836, Nov. 2004.
major reason is that in model 3, the injection currents of all the [12] W. H. Kersting, “Radial distribution test feeders,” in Proc. IEEE Power
equivalent current sources are calculated based on voltages ob- Eng. Soc. Winter Meeting, vol. 2, Jan. 2001, pp. 908–912.
[13] A. Tan, W. H. Liu, and D. Shirmohammadi, “Transformer and load mod-
tained in the previous iteration, instead of the updated voltages. eling in short circuit analysis for distribution systems,” IEEE Trans.
Even though the computation time needed for each iteration is Power Syst., vol. 12, no. 3, pp. 1315–1332, Aug. 1997.
less for model 3, the smaller amount of voltage/current update
makes the total iteration number much higher.
It can be seen that with increases in R/X ratios, all the models Peng Xiao (S’04) received the M.S. degree in electrical engineering from
North China Electric Power University, Beijing, China, in 2000. He is currently
exhibit poor convergence, although model 1 and model 2 are working toward the Ph.D. degree at the University of Wisconsin-Milwaukee.
less sensitive.

V. CONCLUSION David C. Yu (M’84) is currently a Full Professor with the Department of Elec-
trical Engineering and Computer Science, University of Wisconsin-Milwaukee.
A unified method to incorporate three-phase transformers
into the forward/backward sweep-based distribution load
flow is presented. The singularity issues existing in certain Wei Yan received the Ph.D. degree in electrical engineering from Chongqing
transformer configurations were thoroughly examined. This University, Chongqing, China, in 1999.
paper indicates that the singularity appears only in certain Currently, he is an Associate Professor and Associate Chairman of the
Electrical Power Department, Electrical Engineering College, Chongqing
transformer submatrices and only in certain transformer con- University. His research interests include optimal operation and control in
nections. This paper shows that when the singularity occurs, power systems.