Modeling Multi-Winding Transformers
Modeling Multi-Winding Transformers
Roger C. Dugan
Sr. Technical Executive, EPRI
September 14, 2013
Figure 1 shows the common 3-winding transformer equivalent model used in power
system analysis based on positive-sequence equivalent circuits. This is used in nearly all
(if not all) of the popular transmission system power flow, short-circuit, and dynamics
tools. The popularity of the model likely stems from the fact that it can be constructed
entirely from simple two-terminal R-L branch models. This makes the programming
somewhat simpler, although it does create some issues because a single device (a
transformer) is modeled by three branches. So programs usually contain some “kluge
code” to keep track of branches that are associated with transformers. Also, it requires the
creation of a fictitious node (M) that doesn’t physically exist.
(M)
H L
ZH ZL
ZT
T
Figure 1. Common Model of 3-Winding Transformers used in Positive-Sequence Analysis
This model is so ingrained in the minds of electric power systems analysts that many, if
not most, think the impedances in the model are actually found somewhere in a
transformer. It is not until one encounters an unusual transformer such as a 4-winding
transformer that one discovers that this modeling approach works only for the special
case of a 3-winding transformer.
ZT-H ZL-T
T
Figure 2. Equivalent Model that Does Not require Fictitious Node
The architects of the Common Information Model (CIM) chose to use the values of ZH,
ZL, and ZT to describe a 3-winding transformer. Each winding – H, L, and T – of the
transformer is stored in the database separately, each with a single value of resistance, R,
and reactance, X. While it might be argued that it is acceptable to store the resistance with
the winding model because each winding does indeed have resistance, there is no such
thing as the reactance of a single winding, such as XH, for example. Short-circuit leakage
reactances only exist between pairs of windings. If the architects of the CIM had wanted
to model transformers more correctly, they would have defined a separate object to
contain short circuit tests, of which there would be 3 for a 3-winding transformer. This
defect (in my opinion) in the CIM has caused problems trying to adapt it to distribution
system analysis, which uses more complicated transformer models. As it is, one can only
model 2- and 3-winding transformers.
While not common in the transmission grid, transformers with more than 3
windings show up in various industrial processes and have also been used to
connect inverter-based DG such as fuel cells. They are used to provide isolation
between devices with electronic power converters.
http://sourceforge.net/apps/mediawiki/electricdss/index.php?title=TechNote_Mod
eling_Single-phase,_Center-tapped_Distribution_Transformers
To make the modeling problem for the distribution transformer more interesting,
the mixture of 240 V and 120 V loads makes it impossible to model the
transformer with a single per-unit model. (Try it if you don’t believe me.) That is
one reason why we don’t use per-unit models of lines and transformers in the
distribution analysis tools such as OpenDSS. A more general transformer model is
needed.
While I learned the conventional 3-winding model for power flow in college and have
worked with several computer programs that use it, I learned a different way of modeling
transformers in 1973 and have used it almost exclusively in programs I designed and
wrote myself. At the core of most power systems analysis programs whether for power
flow, short circuit, harmonics, dynamics, or transients, is a system nodal admittance (Y)
matrix. Y may be a plain nodal admittance matrix, an augmented matrix or a component
of another set of equations such as a Jacobian for a Newton-based solution. The general
approach is to first build Y from so-called “primitive” admittance matrices for each
element.
ܫு ܸு
ܫ ൩ ൌ ܻ ൩ ܸ ൩ (1)
்ܫ ்ܸ
This equation relates the transformer’s terminal currents to the node-to-ground voltage at
the terminal. This is the voltage of the positive-sequence network model. Thus, every
element of YPrim can be summed directly into the system Y matrix. There is a direct
mapping of each element of the YPrim matrix to the system Y determined by a terminal-to-
node incidence list.
The YPrim matrix can be constructed directly from short-circuit test data. There is no need
to generate an intermediate equivalent circuit. This method applies not only to a simple
single-phase positive-sequence model but to general n-phase, m-winding transformers.
This is the approach used in OpenDSS.
Here is description of the method adapted from my paper with Surya Santoso presented at
the 2003 IEEE T&D show: (Note: Equation numbering restarts at 1.)
---------------------------------------------Insert----------------------------------------------------
One of the reasons we have adopted this approach is that transformer manufacturers can generally
provide accurate values for the short circuit impedances between windings by test or empirical formulae for
any number of windings.
From these data, the short circuit impedance matrix, ZB, is constructed with one of the windings serving
as the reference (assumed shorted). The process is identical to forming the short circuit matrix for a power
system with the infinite bus as reference. Thus, the designation, ZB, is borrowed from that area of power
system analysis.
The short circuit impedance between windings i and j, which we will designate zSCi,j, is generally
expressed in percent on some voltampere (VA) base, usually that of the first winding. Let zbase be the
Diagonal Elements of ZB
Note that the order of ZB is (m-1) – one less than the number of windings. For a simple 2-winding,
single-phase transformer, ZB is trivial, having only one element.
Once this matrix is computed, any arbitrary transformer connection may be modeled by simply applying
linear algebra to perform power-invariant reference frame changes and the conversion to actual values.
The desired result here is the primitive admittance matrix for the transformer. This is merged with the
primitive admittance matrices for other elements in the system to form a system nodal admittance matrix,
Y, representing the connections to the buses in the system. Therefore, the immediate goal is to transform ZB
into the corresponding primitive admittance matrix, Yprim. The transformation can be written in matrix
notation as shown in Equation (3).
B is defined as an m x (m-1) incidence matrix whose elements are either 1, -1, or 0. It relates currents in
the short circuit reference frame where the first winding is assumed shorted to the currents in the nodal
admittance reference frame on a one-turn or one-volt base designated Y1. This network is essentially a per-
unit value model with no winding connections represented, similar to what one might use for a positive-
sequence power flow.
Y1
Yw
N is either a m x m or a 2m x m incidence matrix whose non-zero elements are the inverse of the number
of turns in the windings (or the voltage rating of the winding, depending on the base of ZB). The matrix is
2m x m if you wish to explicitly represent each terminal of each winding individually at this point. This
matrix relates the currents in the Y1 equation to the actual winding currents. The resulting admittance
matrix in the winding reference frame, YW, is in actual values (S). The voltages and currents in this
reference frame are the voltages across the windings and the currents through the windings.
The A matrix is an incidence matrix, whose non-zero elements are generally either 1 or -1, that relates
the winding currents to the actual transformer terminal currents. A single-phase, two-winding transformer
would yield a 4x4 Yprim matrix because there are 4 conductors available for connecting to something (two
conductors per winding). A three-phase, wye-delta transformer with the neutral terminal explicitly
modeled would yield a 7x7 Yprim matrix. (In OpenDSS it would be a 8x8 because all terminals must have
the same number of conductors and there is a neutral conductor for the delta winding that is not connected
to anything.)
Note that nowhere in this formulation is a three-phase system assumed. There are not necessarily any
values appearing in the incidence matrices as in many formulations found in the literature. When the 3 is
Figure 3.Schematic of transformer model development for a simple wye-delta 3-phase transformer.
Y1 represents a ground-referenced nodal admittance network that will give the proper leakage impedance
model on a one-turn or one-volt base. The N matrix represents the effect of the ideal transformers shown to
obtain actual winding voltages. Finally, the A matrix represents the connections of the winding conductors
to obtain the transformer terminals as depicted by the dashed lines.
A neutral impedance of wye-connected windings can be implicitly included in Yprim by adding the
appropriate admittance to the corresponding diagonal element (Yprim 44 in this case).
----------------------------------------End Insert----------------------------------------------------
See the Appendix for more detail on modeling transformers adapted from an OpenDSS
document. The focus of the Appendix is on how to address core behavior for multiphase
transformer models, but the general principles can be gleaned from the material to apply
to modeling the short circuit impedance characteristics of any transformer.
The main issue with core effects is with core structure of 3-phase transformers. The
existing OpenDSS Transformer model does not represent the transformer core other than
as a linear reactance (if the “%imag” property is greater than zero). The magnetizing
branch is embedded within the matrices as the short-circuit impedance matrix is
computed. Therefore, it is difficult to separate it explicitly from the model. Most of the
time, this is sufficient for distribution system analysis, but there are exceptions. One, in
particular, is when the characteristics of the magnetic circuit in 3-legged core
transformers results in significant impacts on zero-sequence impedances. The present
Transformer model in OpenDSS assumes no magnetic coupling between the phases.
Coupling between phases is accomplished by electrical connections of the windings. So
when it becomes important to model the core effects on the zero sequence impedances,
something special must be done.
1
Kluge = a workaround, a quick-and-dirty solution, a clumsy or inelegant, yet effective, solution to a
problem, typically using parts that are cobbled together. (Wikipedia)
Also, the common split-phase 120/240V residential service transformer is not amenable
to per unit representation when there are both 120V and 240V loads. Which base do you
use?
Both of the issues in the preceding paragraphs have a bearing on the transformer model in
the OpenDSS. As a Power Delivery element, a transformer is ultimately modeled by a
primitive Y matrix that embodies all impedances and winding connections. No attempt is
made to model the nonlinear portion of the magnetizing impedance within the
transformer model; it is modeled as a linear reactance, if specified at all - %imag defaults
to zero. The model is focused on the leakage impedance behavior, which has the most
impact on power flow, harmonics, etc.
Figure A-4 shows the schematic for a Yg-Delta-Yg transformer. If this transformer is
constructed from three single-phase transformers, a shell-type 3-phase transformer, or
even a 5-legged core 3-phase transformer, we generally do not worry about interphase
coupling of the magnetic circuit for the types of analysis performed with OpenDSS. Of
course, the delta winding provides coupling between the phases.
There are 3 short circuit measurements between each pair of windings required to
construct the short circuit reactance matrix as depicted in Figure A-5. The OpenDSS can
take this matrix and construct a full 12x12 primitive Y matrix that represents the
transformer windings including the neutral point.2 (A dummy neutral point is generated
for the delta winding to satisfy an OpenDSS requirement that all terminals have the same
number of conductors. By default, it gets connected to ground and its current is zero.)
X1 2
XSC
X1 3 X 2 3
Figure A-5. Short circuit reactance matrix required to represent a 3-winding transformer (percent or
per unit values).
The OpenDSS transformer model provides the XHL, XHT, and XLT properties as a means
for specifying the reactances of a 3-winding transformer. Alternatively, users may use the
xscarray property, which accepts the short circuit reactances as an array representing the
lower triangle matrix depicted in Figure A-5.
Note that some data sources provide per unit or percent values for XH, XL, and XT to
represent a 3-winding transformer (Figure A-6). Keep in mind these values are for a
model that works for the special case of a 3-winding transformer. In reality, there are no
corresponding physical values for these reactances. A clue to this is that XL frequently
comes out negative.
XL
XH L
H
T
XT
Figure A-6. Three-branch per unit or percent model commonly used in power system analysis
programs.
The reactances arise from the leakage flux between pairs of windings. OpenDSS is
designed to model transformers of nearly arbitrary numbers of windings. All that is
2
R. Dugan and S. Santoso, “An example of 3-phase transformer modeling for distribution system
analysis”, IEEE T&D 2003 Conf Proceedings.
Phantom Windings
If the transformer is constructed around a 3-legged core, there is a complication to
specifying the short circuit impedances between the windings. As shown in Figure A-7,
zero sequence flux is in phase in each leg of the core and has to leave the core to
complete the magnetic circuit. The path is mostly “air core” (i.e., not in steel). Thus, the
magnetizing reactance to zero-sequence flux is considerably less than for other normal
mode fluxes (positive- and negative- sequence, for example). In fact, although it is larger,
it is of the same order of magnitude as the leakage (short circuit) reactances and often
cannot be neglected.
The net effect for a 3-winding 3-phase transformer is like having separate cores for each
phase but having a 4th delta-connected winding. In transformer literature, this is often
referred to as a “phantom” delta winding as depicted in Figure A-8. A 4-winding model is
required to construct a physically-based model of the Yg-Delta-Yg transformer if we
choose not to neglect the error introduced by neglecting, or approximating, the phantom
winding effect. Unfortunately, the phantom winding does not physically exist and a
measurement of the short circuit impedances cannot be made. Measurements can be
made only to the physical windings. However, some transformer design programs can
compute these values (I used to work for a transformer manufacturer and could obtain the
appropriate values from the Engineering Department on special occasions).
X1 2
XSC X 1 3 X23
X
X
1
4
2
4
3
4
Figure A-9. Ideally, one would provide three additional short circuit measurements to the phantom
winding.
Modeling Options
Having a physical delta winding on the core will help with zero-sequence modeling. It
will naturally get the zero-sequence impedance in the ballpark of where it should be. But
the zero-sequence impedance still needs to be a little lower. Options for OpenDSS
modelers include:
As shown in Figure A-10, the physical Tertiary winding is generally wound next to the
core. Then the Low voltage winding and the High voltage winding are wound on top of
the Tertiary in that order. The short circuit reactance is proportional to the physical space
between the winding. The greater the space, the higher the reactance is. Proportions vary
with designs. The default values in the OpenDSS transformer model are XHL=7%,
XHT=35% and XLT = 30%. The Low winding is closer to the Tertiary than the High
winding. Thus, XLT is lower than XHT. In another transformer (from the example in this
document) in which the Low is physically much closer to the Tertiary the values are XHL
= 8.98%, XLT = 7.3%, and XHT=13.32%.
For estimating the short circuit reactance to the phantom winding, we need to come up
with three values as indicated in Figure A-9. One might think of the phantom (4th)
winding as occupying the space inside the Tertiary winding. The Tertiary (3rd) winding is
closest, so X3-4 would be smallest of the three and X1-4 (High to phantom) would be the
largest. Knowing these proportions, we can make some educated guesses and can
generally match test results after a few quick iterations using OpenDSS scripts.
HIGH
LOW
TERTIARY
CORE
Zero Sequence:
XHL = 8.98%
XLT = 7.3%
XHT=13.32%
Both models use a bank constructed of 1-phase units for clarity, although a 3-phase
model could also be used.
X1-4 = 100%
X2-4 = 80%
X3-4 = 75%
This was based on my assumption of the construction of the windings (see above). This
turned out to be a good guess because I had only to drop the last two values slightly to
obtain the desired fault currents.
X1-4 = 100%
X2-4 = 70%
X3-4 = 65%
This took only 4 or 5 iterations using the script on the next page. Note that this script uses
the new XfmrCode object to define the transformer. If you get an error when you run this,
you will need to update your version of the OpenDSS.
New Circuit.4Winding
~ BasekV=161 isc3=1000000 isc1=100000
Solve
solve