Foundations and Trends® in Electric Energy

Systems
Circuit-Based Electromagnetic
Transient Simulation
Suggested Citation: Amritanshu Pandey (2023), “Circuit-Based Electromagnetic Tran-
sient Simulation”, Foundations and Trends® in Electric Energy Systems: Vol. 6, No. 2,
pp 83–118. DOI: 10.1561/3100000038.

Amritanshu Pandey
University of Vermont
[email protected]

This article may be used only for the purpose of research, teaching,
and/or private study. Commercial use or systematic downloading
(by robots or other automatic processes) is prohibited without ex-
plicit Publisher approval.
Boston — Delft
 Contents

Introduction 84

1 Equivalent Circuit Approach for EMT Simulations 86

1.1 Wye-Connected RL Load . . . . . . . . . . . . . . . . . . 88
1.2 Induction Motor Load . . . . . . . . . . . . . . . . . . . . 97

Conclusion and Resources for a Deeper Dive 113

Notation 114

Acknowledgments 116

References 117
Circuit-Based Electromagnetic
Transient Simulation
Amritanshu Pandey
University of Vermont, USA; [email protected]

ABSTRACT
The growing penetration of inverter-based resources and as-
sociated controls necessitates system-wide electromagnetic
transient (EMT) analyses. EMT tools and methods today
were not designed for the scale of these analyses. In light
of the emerging need, there is a great deal of interest in
developing new techniques for fast and accurate EMT simu-
lations for large power grids; the foundations of which will
be built on current tools and methods. However, we find
that educational texts covering the fundamentals and inner
workings of current EMT tools are limited. As such, there
is a lack of introductory material for students and profes-
sionals interested in researching the field. To that end, in
this tutorial, we introduce the principles of EMT analyses
from the circuit-theoretic viewpoint, mimicking how time-
domain analyses are performed in circuit simulation tools
like SPICE and Cadence. We perform EMT simulations for
two examples, one linear and one nonlinear, including in-
duction motor (IM) from the first principles. We anticipate
that by the end of this monograph, the readers will have a
basic understanding of how power grid EMT tools work.

Amritanshu Pandey (2023), “Circuit-Based Electromagnetic Transient Simulation”,

Foundations and Trends® in Electric Energy Systems: Vol. 6, No. 2, pp 83–118. DOI:
10.1561/3100000038.
©2023 A. Pandey
 Introduction

This monograph is a tutorial to help students and working professionals

learn the inner workings of tools that perform electromagnetic transient
(EMT) simulations. The tutorial uses a circuit-theory based approach
to EMT simulation. The tutorial is introductory in that it shows the
workings through two simple examples and expects the reader to build
expertise through more detailed readings. Nonetheless, this monograph
is an excellent first step for anyone interested in the subject, especially
those interested in building their own tools.
Before we dive into EMT models and simulation, it is pertinent to
discuss and differentiate between the two types of commonly performed
time-domain simulations in the power community:

• Transient Stability (TS): Performs time-domain simulations

on balanced networks (only positive sequence components). It does
not include network impedance transient response and models
them moving from one steady state to another instantaneously.
Therefore transient stability frameworks model the network con-
straints with algebraic equations and only model the injection
components and controls with differential algebraic equations. TS
is also sometimes called root-mean-squared (RMS) transient anal-
ysis. These have been traditionally the workhorse for system-wide
time-domain power grid analysis.

84
 85

• Electromagnetic Transient Simulation: Works exactly as

circuit-simulation tools (for instance, SPICE and CADENCE). It
models all three phases, including the transient response for the
network impedance components. The industry demand for large-
scale EMTs is growing rapidly due to the recent events involving
power electronics devices.

This monograph focuses on electromagnetic transient simulation.

These are becoming increasingly common (with significant active re-
search) due to the introduction of inverter-based resources on the grid.
The monograph provides a step-by-step tutorial on performing an EMT
simulation for simple networks from the first principles. It uses concepts
from circuit simulation, Newton-Raphson, and numerical integration
with difference methods. Good references to brush up on these topics
are [9], [4], and [1]. The monograph does not discuss modeling controls
in EMT but provides sufficient background such that it can be tackled
as the next step.
Furthermore, this monograph uses modified nodal analysis (MNA)
and sometimes loop analysis to encapsulate network physics. Other
alternate approaches like Tree Link Analysis (TLA) and Sparse Tableau
Analysis (STA) [9] exist but are not covered in this monograph. In
naive nodal analysis, the currents at each node are summed to zero
(i.e., Kirchhoff’s current law) to satisfy network physics. However, the
method fails when the network includes voltage sources, as the current
through the voltage source is not implicitly known. Therefore, to include
voltage sources, the modified version adds one additional constraint per
voltage source to the set of nodal equations. These additional constraints
give us the currents through the voltage sources as new variables. The
term modified in MNA refers to this modification. Finally, the readers
must note that anytime in this monograph when they come across the
following terms: system matrix, Y matrix, solution matrix, or simple
nodal matrix, these all refer to the same thing: a set of nodal or loop
equations (or sometimes a mix) for the linearized network that is being
evaluated at a given time-step t.
 1
Equivalent Circuit Approach for EMT
Simulations

We will use an equivalent circuit framework [7], [8] for power grids to
develop EMT models for two simple toy networks. These toy examples
are based on a simple 2-bus three-phase power network with different
load configurations. We consider the following load scenarios:

• A linear wye-connected series RL load (see Figure 1.3).

• A non-linear induction motor load (see Figure 1.11).

Note: For both examples, we assume that the load is connected in a

wye configuration and all the neutrals in the network are grounded.
The illustration of the two-bus power network with a generic load
model is shown in Figure 1.1. It includes power system verbiage such as
infinite bus and wye-connected load. For the moment, the infinite bus
(at nodes a0 , b0 , c0 ) can be considered the power source, and the load
bus (at nodes a, b, c) can be considered the power sink.
The illustration in Figure 1.1 is a power system representation. To
perform EMT simulation with an equivalent circuit framework, as a first
step, we represent any power grid-specific elements as electric circuit
elements. The network in Figure 1.1 has three components: (i) an infinite
bus, (ii) a transmission line (a very simplified model), and (iii) a load.

86
 87

Figure 1.1: Power system representation of the 2-bus power network.

Two of these three components (infinite bus and transmission line) are
shared across the two examples. Therefore, we begin with them. The
infinite bus is trivial to model as a circuit element. It is simply an
ideal voltage source with a known magnitude and phase angle. Further,
the frequency of this source is assumed to be either 60 Hz or 50 Hz
depending on the system. For a three-phase network, the infinite bus is
modeled by a set of three independent voltage sources (offset by 120◦ )
connected in a wye formation. The network with an infinite bus replaced
with circuit elements is shown in Figure 1.2. The transmission line model
with R and L values already represents a set of circuit elements and
does not require translation. Note that this model of the transmission
line is greatly simplified to improve the readability of this tutorial.
A more detailed model (see [3]), at a minimum, should include shunt
capacitances and mutual impedances. Longer lines should use lumped
models. Still, the readers will learn to model mutual impedances while
developing the EMT model for an induction motor.

Figure 1.2: Power system representation of the 2-bus power network.

88 Equivalent Circuit Approach for EMT Simulations

Figure 1.3: Power system representation of the 2-bus power network.

The last step is to convert the load component in Figure 1.2 into
an equivalent circuit before getting started with EMT-related steps.
Loads in power systems correspond to various devices ranging from
large induction motors to resistive heating loads. We will translate the
two load models into equivalent circuits in their respective subsections.

1.1 Wye-Connected RL Load

The first example represents the load via a wye-connected series RL

impedance. With series RL impedance as the load component, the
equivalent circuit representation of the 2-bus network is trivial and
is shown in Figure 1.3. Simple observation confirms that all elements
in this equivalent circuit are linear; therefore, to obtain the transient
response, we only need to recursively solve the network as a function of
time. Iterative solves with Newton-Raphson (NR) are only necessary in
the presence of nonlinear components. We will learn about that in the
induction motor (IM) example.
In current EMT tools (for instance, EMTP-RV or PSCAD), a
sequence of steps that are followed to obtain the time-domain response
of a linear power network are as follows:

(1) First, ordinary differential equations (ODEs) corresponding to the

sum of currents at each node in the network are defined.

(2) Second, numerical integration approximation (such as trapezoidal

or backward Euler) is applied to these ODE equations to convert
them from differential algebraic equation (DAE) form to purely
algebraic equation (AE) form.
1.1. Wye-Connected RL Load 89

(3) Third, if the network is linear, the final solution for each time
instance is obtained via a recursive linear solve of the algebraic
equations for each time-tick.

Note that the equations for subsequent time steps are coupled
through terms obtained from numerical integration. So currents at
t = n + 1 will be functions of currents at t = n. We will see this in more
detail while constructing the companion models for various memory
elements. Also, note that linear equations are recursively solved to
traverse through simulation time by choosing a time-step (∆t). The
value of this time step depends on the local truncation error (LTE).
More details on the calculation of LTE can be found in [9]. The general
idea is to maintain a sufficient tradeoff between simulation speed (by
taking larger time steps) and accuracy (by taking smaller time steps).
Now we will learn how to solve the EMT problem from the viewpoint
of the equivalent circuit approach (ECA). To obtain the time-domain re-
sponse of the network with ECA, we apply two (2) circuit-simulation
tricks to enable efficient construction of the linearized nodal equations
(Yt Vt − Jt = 0) for each time step (t) in the transient simulation.
First, rather than defining the ordinary differential equation (ODE)
that captures the KCL constraint (sum of currents) at each node, we
construct specialized symbolic stamps for each component in the network
(e.g., resistance, inductance, etc.). Note that a stamp(s) here refers to
term(s) that are to be added to the solution matrix. We use these stamps
to implicitly construct the system matrix at each time tick. To do so, we
parse through each circuit element in the network (for each time tick),
and we add the stamps corresponding to this element in the solution
matrix. Once completed, the matrix (completed by adding stamps from
each component) will be equivalent to if we constructed the matrix by
adding linearized nodal equations in each row. In this approach, rather
than arduously constructing nodal equations for each node (imagine
a circuit with million nodes), we only have to symbolically construct
stamps for the number of types of circuit elements. Once that is done,
all we need to do is add the stamps for each circuit element into the
solution matrix.
90 Equivalent Circuit Approach for EMT Simulations

Second circuit simulation trick pertains to simplifying the use of

numerical integration of differential terms. Instead of applying numerical
integration to all nodal ODE equations explicitly, we develop and use
companion circuits for time derivative memory elements (e.g., inductance
and capacitance) in the network. By applying these two tricks, the ECA
approach, rather than tackling one nodal or loop equation at a time,
builds modular circuits for various elements with embedded implicit
symbolic linearization and discretization. This allows us to focus on a
limited number of types of power grid elements (generator, load, etc.)
instead of millions of nodal or loop equations.
In ECA, with the two tricks, we follow the following steps sequentially
to enable modular construction and solution of the system matrix at
each time tick without ever having to define the complete set of nodal
equations explicitly:
(1) Represent time-derivative elements in the circuit (e.g., inductor,
capacitor etc.) with their companion circuits.

(2) Develop a specialized stamp for each element in the circuit (with
derivative elements replaced by their companion circuits).

(3) ∀ t ∈ τ = {t0 , . . . , tf inal }, parse through the overall network (gen-

erally a netlist or input file of sorts) and stamp the terms corre-
sponding to each element in a sparse linear matrix (remember at
this point, time-derivative terms are replaced by their companion
models).

(4) Solve the linear problem Yt Vt = Jt , to obtain the solution of state

variables at time t, and update and repeat until final time step.
This approach allows us to implicitly obtain and solve the linearized
nodal equations without constructing them symbolically. This is highly
beneficial for developing heuristics and for solving large-scale networks.
More on that later.

1.1.1 Companion Model for Inductors and Trapezoidal Rule

Returning to the RL-load example, the inductor is the only element
with a time derivative term in this network (see Figure 1.3). Therefore,
1.1. Wye-Connected RL Load 91

all we have to do is construct a companion model for the inductor. We

then replace all the inductors in the network with their companion
models. We now discuss the construction of these companion models.
This tutorial will use trapezoidal integration approximation to con-
struct companion models. Other Euler-based or higher-order numerical
integration methods can also be used following the same approach. Ap-
plying the trapezoidal integration approximation, the companion circuit
for the inductor can be derived as follows (for more details, see [9]):
di
L =v (1.1)
dt
vdt
di = (1.2)
L
Z tn+1 Z tn+1
vdt
di = (1.3)
tn tn L
∆t
(i(tn+1 ) − i(tn )) = (v(tn+1 ) + v(tn )) (1.4)
2L
∆t ∆t
i(tn+1 ) = i(tn ) + v(tn ) + v(tn+1 ) (1.5)
2L 2L
The equation in (1.5) has three terms. The last term ∆t2L describes the
relationship between current and voltage at time tn+1 , across the same
branch, and is a conductance G. The first two terms are constants that
are known from states at prior time-point tn states and therefore are
represented via a constant current source I. Aggregating these elements
into a circuit, we get the equivalent circuit in Figure 1.4, and it is a
companion model for an inductor with trapezoidal numerical integration
approximation.

Figure 1.4: Inductor Trapezoidal Companion Circuit. Reconstructed from Pileggi,

Carnegie Mellon ECE 18-762 notes [9].
92 Equivalent Circuit Approach for EMT Simulations

Figure 1.5: Equivalent circuit with companion circuits (com. ckt.).

With the companion model constructed, we can replace the inductor

elements in Figure 1.3 with the respective companion circuits. Following
this step, we obtain a time-dependent equivalent circuit (see Figure 1.5),
which can be characterized entirely by linear algebraic equations. This
set of equations is repeatedly solved to find the electromagnetic transient
response of the network.
However, as discussed, the trick is to avoid explicitly constructing
the complete equivalent circuit and corresponding nodal equations
(Y V − J) for each node. Instead, we will follow a modular approach.
As we encounter each element in the netlist of the network, we will add
the specialized stamps corresponding to that element into the sparse
system matrix. In this example (see Figure 1.5), there are three ideal
voltage sources, six resistance and six inductor models. Therefore, we will
cover the stamps for these elements, which will be sufficient to construct
the linearized nodal equations for each time-step (Yt Vt − Jt = 0) for
the network in Figure 1.5. By recursively solving these equations over
time from the initial time-step t0 to the final time step tf inal , we can
obtain the time-domain response of this network. Next, we will cover
the stamps for these elements.
Note: We will refer to the sparse linear matrix for the system as
Y from here on. Each time step will have a different system matrix
notated by Yt or sometimes simply by Y . Further, even for each time
step, the matrix can be expressed as the sum of two separate matrices:
Yt = Ytlin + Ytnlin . Terms in Ytnlin change for each iteration of NR,
1.1. Wye-Connected RL Load 93

Figure 1.6: Resistance stamps in the nodal matrix. Redrawn from CMU, ECE,
18-762 notes [9].

whereas the Ytlin , remain fixed. Because this circuit is linear, we do not
have Ytnlin .

1.1.2 Stamp for Resistors

The stamps for a resistance R between node m and n in the nodal
matrix Y are shown in Figure 1.6. The general approach is that when
you encounter resistance in an input file, add the terms defined in
Figure 1.6 to sparse linear matrix Y in the index locations: (m, m),
(m, n), (n, m), and (n, n). Here node m is the from node of the resistor,
and n is the to node of the resistor.

1.1.3 Stamps for an Independent Current Source

An independent current source stamp I between node m and n will add
no additional stamps to matrix Y but instead will add terms to column
vector J as shown in Figure 1.7. The column vector J only consists of
constant terms that are not a function of any state in the network. The
negative sign in the J vector is due to moving the constant number to
the RHS of the equation. However, keep in mind the direction of the
current flow from the independent current source for the signage.

1.1.4 Stamps for Inductors

Inductor consists of time-derivative terms. Therefore, as a first step,
each time we encounter an inductor in the netlist between node m and
node n, we will have to stamp the corresponding companion circuit.
Remember doing so will replace the time-derivative terms with algebraic
expressions. The companion circuit consists of a current source and a
94 Equivalent Circuit Approach for EMT Simulations

Figure 1.7: Independent current source stamps in J vector. Redrawn from Pileggi,
CMU 18-762 notes [9].

resistor in parallel between nodes m and n as seen in Figure 1.4. As we

have already covered stamps for resistance and current source, follow
the steps in the previous subsections to stamp those components. The
key thing to note is that stamp values for inductors in system matrix Y
will change for each time step and must be updated accordingly. This is
because the values of the independent current source in the companion
circuit will be a function of currents and voltages from the last time
tick tn .

1.1.5 Stamps for Independent Voltage Source

Next, we derive the stamps for a voltage source. To include an inde-
pendent voltage source between nodes m and n, nodal equations are
augmented to include an additional row. With an extra row comes an ad-
ditional variable representing the magnitude of current flow through the
voltage source. The additional row (for each voltage source) constrains
the potential difference between node m and node n:

Vm − Vn = VS (1.6)

and the additional current is notated by i. The stamps in the nodal

matrix (Y ) and source vector (J) for independent voltage source are
shown in Figure 1.8.
Now that we have learned how to develop stamps for the resistor,
inductor, current source, and voltage source, we can construct the nodal
matrix (Y ) and source vector (J) for the simple power system network
1.1. Wye-Connected RL Load 95

Figure 1.8: Independent voltage source stamps. Redrawn from Pileggi, CMU- 18-762
notes [9].

in Figure 1.3 for each time tn+1 . To obtain the time-domain response
of the network, the set of linearized equations is recursively solved with
a sparse matrix solver (using LU factorization). Note that to obtain
the network’s response at time t = 0+ , we require the network states at
time tn = 0. For this particular network, these are needed to initiate the
terms in inductor stamps dependent on the prior time step. Not just
that, it also helps dictate the system state at which the time-domain
response begins.

1.1.6 Initialization
We can initialize the linear circuit through AC or DC analysis. In DC
analysis, we replace the sinusoidal voltage sources with DC voltage
sources. In AC analysis, we convert the inductors to an impedance at a
single frequency.
Here we will describe the DC analysis approach to initialization. Let
us assume the network is in the DC state between t = −∞ and t = 0.
At t = 0, it switches to the AC sinusoidal source, and let us assume the
switching is smooth (i.e., DC voltage and AC voltage at time t = 0 are
identical). To know the initial states of the circuit at t = 0, we perform
DC analysis, wherein the following condition holds true for memory
elements (inductors and capacitors):
di
L = 0 @DC (1.7)
dt
dv
C = 0 @DC (1.8)
dt
From these conditions in (1.7) and (1.8), we can infer that during the
DC state, the inductor LDC is short-circuited (VL = 0) and the capacitor
96 Equivalent Circuit Approach for EMT Simulations

Figure 1.9: Corresponding equivalent circuit during DC state. Note that the
sinusoidal voltage sources are set to a fixed DC value in the case of DC analysis.

C DC is open-circuited (IC = 0). For this example, the DC equivalent of

the network in Figure 1.5 is shown via a network in Figure 1.9. With
access to DC equivalent circuit, we stamp the elements in the circuit
into the system matrix Ydc and vector Jdc and solve them using a sparse
linear solver. The solution is used to define the system state at t = 0.

1.1.7 Time-Domain Solution

With access to the initial state at t = 0, we recursively solve a system
matrix (moving forward by ∆t in each recursion) following steps de-
scribed in Figure 1.10 to obtain the time-domain response from t = 0
to t = tf inal . We can dynamically adjust ∆t at each step based on
the trade-off between simulation run-time and local truncation error
(LTE). LTE is calculated by approximating the higher-order terms of
the numerical integration. More details can be found in [9]. Note that
dynamically adjusting LTE is not always necessary as for a purely
linear system changing ∆t will require re-generating the ∆t dependent
companion model terms at each iteration at an additional computa-
tional cost. Practically speaking, a happy medium can be obtained by
periodically checking the LTE throughout the simulation and adjusting
∆t if it is found high.
One facet that this example does not cover is how to handle non-
linear terms. We will learn how to include nonlinear terms in the
following example with IM as an electric load. We will also learn to use
an alternative mathematical form (other than MNA) to describe the
1.2. Induction Motor Load 97

Figure 1.10: Running EMT simulation for linear networks.

Figure 1.11: 2-bus network with induction motor load.

system’s behavior. In addition, we will also learn how to incorporate

non-electrical physics (i.e., IM’s mechanical physics), like rotor speed,
with the circuit-theoretic paradigm, into the mathematical formulation.

1.2 Induction Motor Load

In the second example, we will learn how to obtain the time-domain

response for a non-linear grid component: the induction motor (IM).
To construct a toy power grid network with IM, we will replace the RL
load in the first example with an IM load (see Figure 1.11).
98 Equivalent Circuit Approach for EMT Simulations

Analyzing IM in the abc frame can be difficult. The flux generated

by the three-phase IM in abc frame has time-varying coefficients in its
voltage terms due to the sinusoidal nature of the mutual inductance.
This makes the analysis of three-phase IM cumbersome in the abc
reference frame. However, this undesirable feature can be eliminated by
using dq transformation. dq transformation is a linear transformation.
It can be performed by choosing one of the three reference frames: (i)
synchronous reference frame, (ii) stationary reference frame, and (iii)
rotating reference frame. For more details on dq transformation, see [5].
For derivations in this work, we will use the stationary reference
frame and a power invariant dq transformation. Doing so allows us to
set θ = 0 and simplify further calculations by having a time-invariant
voltage source across the IM. In the stationary frame, abc variables
(both currents and voltages) are converted to dq variables by following
power invariant dq matrix transformation:
[F0dq ] = [Pθ ][Fabc ] (1.9)
where:
√ √ √ 
r 0.5 0.5 0.5
2
[Pθ ] = cos(θ) cos(θ − λ) cos(θ + λ) , θ = 0 (1.10)

3
sin(θ) sin(θ − λ) sin(θ + λ)
λ is the phase difference between phases abc during balanced operation:
2π
λ= rad (1.11)
3
After applying dq transformation to IM and decoupling non-IM abc
components (see left of Figure 1.12) from dq components using controlled
current and voltage sources, the network model in Figure 1.11 is modified
to one in Figure 1.12.
Note that controlled current and voltage sources in Figure 1.12
encapsulate the math behind dq and inverse dq transformation for
the variables. dq transformation is applied to capture abc network
voltages and reflect those for IM sub-circuit in dq frame. Inverse dq
transformation is used to convert the currents consumed by the IM in
dq frame into abc frame (IaIM , IbIM , IcIM ) such that they can be fed into
rest of the network.
1.2. Induction Motor Load 99

Figure 1.12: 2-bus network with induction motor load after dq transformation.

The transformed Figure 1.12 has three components: (i) the infinite
bus represented by a set of wye-connected independent voltage sources,
(ii) the transmission line represented by a series RL circuit, and (iii) an
IM. We have already discussed the construction of stamps and compan-
ion circuits for infinite bus and transmission line elements (reiterating
that in this tutorial the transmission line model is greatly simplified,
see [3] for detailed models like pi model and Bergeron model). Now, we
will learn how to represent the IM physics with an equivalent circuit.
We will also learn how to replace the time derivative terms in the IM
equivalent circuit with an analogous companion circuit.
The generalized equations that represent the electrical component
of IM physics, independent of the reference frame choice in dq transfor-
mation, are expressed via the following KVL equations [5]:

Vds = Rs Ids + pψds + ψqs pθ (1.12)

Vqs = Rs Iqs + pψqs − ψds pθ (1.13)
Vdr = Rr Idr + pψdr + ψqr pβ (1.14)
Vqr = Rr Iqr + pψqr − ψdr pβ (1.15)
100 Equivalent Circuit Approach for EMT Simulations

where p is the differential operator and,

ψds = (Lls + Lm )Ids + Lm Idr (1.16)

ψdr = (Llr + Lm )Idr + Lm Ids (1.17)
ψqs = (Lls + Lm )Iqs + Lm Iqr (1.18)
ψqr = (Llr + Lm )Iqr + Lm Iqs (1.19)

In the expressions above, Lls and Llr represent the leakage induc-
tance of the stator circuit and rotor circuit, respectively. Lm is the
mutual inductance between the rotor and stator circuits. Rs and Rr
are the stator and rotor resistance, respectively. ψ represents the flux
component across each electrical sub-circuit. The nonlinearity in the
electrical part of the IM is due to the speed voltage terms (ψqr ωr , ψdr ωr ).
In addition to the IM electrical part equations in (1.12)–(1.19), the
mechanical IM part is represented by a single differential equation, the
swing equation (1.20):
(TE − TL − Dωr )
pωr = (1.20)
J
where, electrical torque TE is further described by interactions between
the IM currents in the electrical sub-circuit, which also introduces
nonlinearities in the IM model:
3
TE = Lm Np (Idr Iqs − Iqr Ids ) (1.21)
4
TE , the electrical torque, is given in N.m and J, the motor net inertia,
in kg · m2 . Np is the number of poles in the IM. The load torque TL is
generally represented as a polynomial function of rotor speed.
We can further simplify the electrical components’ expressions be-
cause we use the stationary reference frame for dq transformation. In
the stationary reference, θ = 0, β = −θr and pβ = −ωr . Furthermore,
pθ = ωs = 0. With these simplifications and reducing the expressions
in (1.12)–(1.19), the following form for the IM’s electrical part, in the
1.2. Induction Motor Load 101

stationary frame, can be obtained:

Vds − Rs Ids − Ls pIds − Lm pIdr = 0 (1.22)

Vqs − Rs Iqs − Ls pIqs − Lm pIqr = 0 (1.23)
Vdr − Rr Idr − Lr pIdr − Lm pIds + ωr Lr Iqs + ωr Lm Iqr = 0 (1.24)
Vqr − Rr Iqr − Lr pIqr − Lm pIqs − ωr Lr Idr − ωr Lm Ids = 0 (1.25)

where,

Ls = Lls + Lm (1.26)
Lr = Llr + Lm (1.27)
Vdr = 0 (1.28)
Vqr = 0 (1.29)

Note that the rotor windings are shorted in most IM designs, and hence
voltages Vdr and Vqr are set to 0.
Now to construct the equivalent circuit for the IM model, we map
the four KVL expressions in (1.22)–(1.25) and one KCL expression in
(1.20) to Figure 1.13.
One can observe in Figure 1.13 that aside from self-inductance (Ls
and Lr ), the IM equivalent circuit also includes time-derivative terms
˙ ω̇r ) for mutual inductance (Lm ) and rotor speed (ωr ).
(given by p or I/
We have previously learned how to approximate time-derivative terms
for self-inductance by constructing and stamping the corresponding
companion circuits. Here, we will derive the companion circuits for
combined self and mutual inductance elements (see blue elements in
Figure 1.14). We will use a simple two-coil example in Figure 1.15 to
develop the companion circuit for combined self- and mutual inductance.
Later in the section, we will learn how to construct the companion circuit
for time-derivative terms corresponding to rotor speed. We will finish
by discussing how the nonlinear elements in the IM equivalent circuit
are handled and subsequently added to the system matrix.
We construct the companion circuit for combined self and mutual
inductances to approximate its time-derivative behavior by a set of
algebraic equations. This way, we can approximately solve a set of
102 Equivalent Circuit Approach for EMT Simulations

Figure 1.13: Equivalent circuit for IM in the stationary frame.

ODEs parameterized by self- and mutual-inductances by recursively

solving a set of algebraic equations over time. We begin the derivation
by observing a two-coil example in Figure 1.15, which has two pairs of
self- and mutual-inductance (self: L11 or L22 , mutual: L12 ). We focus
on voltage V1 expression in the first coil. It can be represented as a sum
of the voltage across the self and mutual inductances:
dI 1 dI 2
V1 (t) = L11 + L12 (1.30)
dt dt
The voltage expression in (1.30) is an ODE that includes time-derivative
terms. Therefore, we apply the trapezoidal integral rule to approximate
the ODE’s solution. With the trapezoidal integral rule applied, the
voltage across self- and mutual-inductance (1.30) can be represented by
1.2. Induction Motor Load 103

Figure 1.14: Illustration of combined self and mutual inductances in IM.

Figure 1.15: Two coil example.

the sum of the following difference equations, which are purely algebraic
terms:
2L1
V11 (tn+1 ) = (I1 (tn+1 ) − I1 (tn )) − V11 (tn ) (1.31)
∆t
2L12
V12 (tn+1 ) = (I2 (tn+1 ) − I2 (tn )) − V12 (tn ) (1.32)
∆t
104 Equivalent Circuit Approach for EMT Simulations

Figure 1.16: Two coil example companion circuit.

Therefore the total voltage induced in coil 1 (V1 (t) = V11 (t) + V12 (t)) is
given by:
2L1 2L12
V1 (tn+1 ) = (I1 (tn+1 ) − I1 (tn )) − V11 (tn ) + (I2 (tn+1 ) − I2 (tn )) − V12 (tn )
∆t ∆t
(1.33)
and after re-arranging the terms:
 
2L1 2L12 2L1 2L12
V1 (tn+1 ) = I1 (tn+1 ) + I2 (tn+1 ) − I1 (tn ) + I2 (tn ) + V12 (tn ) + V11 (tn )
∆t ∆t ∆t ∆t
(1.34)
The voltage at time tn+1 across the coil 1 is approximated using
(1.34) and is represented as an equivalent circuit in Figure 1.16. The
third term is only dependent on the historical values of the variables and
can be represented by an independent voltage source V1hist . Similarly, a
resistance REQ represents first term as it maps the linear relationship
between the voltage and current across the same branch and a current
controlled voltage source represents the second term, as it maps the
voltage contribution due to current in the second coil.
By replacing the time-derivative terms for self- and mutual-inductance
with the corresponding companion circuits in Figure 1.16, we can solve
the set of ODEs corresponding to the electrical part of IM, recursively
over time. But before, we must learn how to address the time-derivative
term in the mechanical part of the IM model. The variables in the
mechanical part impact the electrical part and vice-versa. Therefore
we need to solve the mechanical part concurrently with the electrical
portion of IM. The approach for solving the mechanical part is trivial.
On close observation, one can see the mechanical circuit of the IM is
analogous to a parallel RC electrical circuit with current sources, where
inertia J is analogous to capacitance C, TL is analogous to a constant
current source I, D is analogous to a conductance G, TE is analogous
1.2. Induction Motor Load 105

to a current controlled current source (see (1.21)), and ωr is analogous

to the voltage at the mechanical circuit node. The only time-derivative
term in the mechanical circuit is the partial derivative of rotor speed
times inertia (J dω
dt ), which is analogous to a current through a capaci-
r

tor. Therefore, we will learn how to construct a capacitor’s companion

circuit to replace the time-derivative term in the mechanical part of IM.
By replacing the time-derivative term, we can represent the physics of
IM’s mechanical part with a single nodal equation that can be solved
recursively over time (assuming we have dealt with nonlinearities).
To construct the companion model for a capacitor, we will use
trapezoidal integration approximation following the same approach we
used for inductors. Applying the trapezoidal integration approximation,
the expression for current/voltage through/across a capacitor at time
tn+1 given values at time tn can be expressed as follows:
dv
C =i (1.35)
dt
idt
dv = (1.36)
C
Z tn+1 Z tn+1
idt
dv = (1.37)
tn tn C
∆t
(v(tn+1 ) − v(tn )) = (i(tn+1 ) + i(tn )) (1.38)
2C
∆t ∆t
v(tn+1 ) = v(tn ) + i(tn ) + i(tn+1 ) (1.39)
2C 2C
∆t
The equation in (1.39) has three terms. The last term 2C describes
the relationship between current and voltage at time tn+1 and is a
conductance G. The first two terms are constants that are known from
prior time-point tn and therefore are represented via a constant voltage
source V . Aggregating these elements in series, we get the equivalent
circuit in Figure 1.17, representing a capacitor’s companion model with
trapezoidal numerical integration approximation. For representing the
speed time-derivative terms in IM, we will use analogous symbols in
the companion model: inertia J for capacitor C and rotor speed ωr for
voltage v.
106 Equivalent Circuit Approach for EMT Simulations

Figure 1.17: Capacitor trapezoidal companion circuit. Redrawn from Pileggi, 18-762
notes [9].

1.2.1 Handling Nonlinear Terms

In the case of IM, we began with a set of models characterized by
differential-algebraic equations (DAEs), which had both differential and
nonlinear terms. As a first step, we built and used companion circuits
to approximate the differential term behavior with algebraic terms.
Next, we linearize the nonlinear terms and solve them iteratively to
approximate the nonlinear behavior.
We linearize the nonlinear terms with first-order Taylor’s approxima-
tion, like in nonlinear power flow analysis. However, instead of linearizing
complete nodal or loop equations (e.g., KCL or KVL), we linearize the
nonlinear terms corresponding to each component modularly by devel-
oping linearized equivalent circuits. Then for each iteration of NR, we
iteratively update the stamps corresponding to these components in the
system matrix Y and resolve.
Let us write out and understand the general form and application
of first-order Taylor approximation. For general nonlinear function f (x)
(x: RN ×1 ), the first order Taylor approximation is given by:
T
f k+1 (x) ≈ f k (x) + f 0 (x) (xk+1 − xk ) (1.40)

With nonlinear f (x) linearized around xk from the k th iteration, the

next iterate can be obtained via a linear solve. We will use the general
theory now to understand the linearization of IM terms.
In the IM model in Figure 1.13, for the electrical part, the non-
linearity stems from four speed-voltage terms (ωr Lr Idr , ωr Lm Ids , ωr Lr Iqr ,
and ωr Lm Iqs ) in the dr and qr sub-circuits (iii, and iv in Figure 1.13).
1.2. Induction Motor Load 107

There are additional nonlinearities in the mechanical sub-circuit (v in

Figure 1.13)
To further explore how to add IM nonlinear terms to the Y matrix,
let us consider the nonlinear terms in the KVL expression (1.25) for
subcircuit (iv) in Figure 1.13:
NL
fqr (ωr , Ids , Idr ) = −ωr Lr Idr − ωr Lm Ids (1.41)

(1.41) is a nonlinear function of the rotor speed (ωr ), direct-axis ro-

tor current (Idr ), and direct-axis stator current (Ids ). The linearized
approximation for this expression for the (k + 1)th NR iteration is:
 ∂f   ∂f 
k+1 k k+1
fqr (ωr , Idr , Ids ) = fqr (ωr , Idr , Ids ) + (ωrk+1 − ωrk ) + (Idr k
− Idr )
∂ωr k ∂Idr k
 ∂f 
k+1
+ (Ids k
− Ids ) (1.42)
∂Ids k

where:

f k (ωr , Idr , Ids ) = −ωrk Lr Idr

k
− ωrk Lm Ids
k
(1.43)
∂f
 
k k
= −Lr Idr − Lm Ids (1.44)
∂ωr k
∂f
 
= −ωrk Lr (1.45)
∂Idr k
∂f
 
= −ωrk Lm (1.46)
∂Ids k
Similar to the treatment of nonlinear terms in (1.25), other nonlinear
terms in the IM are also linearized and stored in symbolic form. These
terms are added (or stamped) in the system matrix Y following the
discussion in next subsection and updated for each iteration of NR.

1.2.2 Stamping Controlled Current Sources

In the IM linearized equivalent circuit, we end up with two new circuit
elements, a current-controlled voltage source (CCVS) and a current-
controlled current source (CCCS). The CCVS originate from linearizing
speed current terms in (1.22)–(1.25) (see terms in (1.42) for Fqr KVL
constraint). The CCCS terms originate while linearizing the TE equation.
We briefly describe how those terms are added to the system matrix.
108 Equivalent Circuit Approach for EMT Simulations

Figure 1.18: Off-diagonal entries for modeling CCVS in IM’s Fqr electrical sub-
circuit.

Figure 1.19: Stamps for CCCS. Reconstructed from Pileggi, Carnegie Mellon ECE
18-762 notes [9].

As we use KVL to describe the IM’s electrical equations, the current-

controlled voltage sources only show up as off-diagonal terms in the
system matrix. For example, the CCVS stamps for Fqr KVL expression
are shown in the system matrix in Figure 1.18. These do not include
the rotor speed dependent terms.
Next, we discuss CCCS stamps, which are in the mechanical equation
and are stamped following the MNA approach in Example 1. The stamps
for a generic CCCS between nodes k and l are shown in Figure 1.19. In
general, to measure the independent current on which the controlled
source depends, we add a zero-valued voltage source (ammeter) between
nodes m and n. Note that adding a voltage source results in an addition
of a row to the Y -matrix, and in return, we get the current value through
the source. Now, as the dependent current source between nodes k and l
is valued β times the measured current i, for the example in Figure 1.19,
we add corresponding terms to KCLs for nodes m and n. However, for
the IM example specifically, we do not need an additional zero-valued
voltage source as IM currents (Iqr , Iqs , Idr , and Ids ) are already variables
and easily accessible.
1.2. Induction Motor Load 109

Figure 1.20: Stamps for VCVS. Reconstructed from Pileggi, Carnegie Mellon ECE
18-762 notes [9].

The IM in dq frame is coupled to the rest of the circuit with

controlled sources as well. These are given by IaIM , IbIM , IcIM and Vds ,
Vqs in Figure 1.12. The first set IaIM , IbIM , IcIM , which map the currents
consumed by the IM in dq frame into abc frame, are modeled via CCCS.
The underlying math defining the relationship between currents in
dq and abc frames is given by inverse dq transformation. The stamps
for these sources are covered in the CCCS description above. The
second set, which maps the abc voltages to dq voltages with linear dq
transformation, is represented via voltage-controlled voltage sources
(VCVS). The stamps for generic VCVS are shown in Figure 1.20. Here,
the voltage source across nodes m and n equals α times the difference
between voltages at nodes p and q. This relationship is captured in the
extra node N + 1 added to the matrix, which also gives us the current
i through the voltage source. The current is added to KCL rows for
nodes m and n. In the IM, the voltage Vds and Vqs will be functions of
VaIM , VbIM and VcIM .

1.2.3 Stamping IM Equations

So far, we discussed addressing nonlinear and differential terms modu-
larly for each component. We also learned how individual components
are added to the system matrix. The next step is to learn how the
various terms come together in the overall Y matrix to satisfy system-
level physics for the network in Figure 1.12. We must also learn the
updation of these terms for each iteration (for nonlinear elements) and
110 Equivalent Circuit Approach for EMT Simulations

each recursion (for each time step forward) to obtain a time-domain

solution for the overall circuit.
Let us begin with non-IM linear components in Figure 1.12. We add
them into the system matrix following the MNA approach described
in Example 1 of this tutorial. We update the companion circuit terms
for memory elements every instance we move forward in time. For the
IM terms, we will use an alternative approach. If we were to stamp the
IM circuit elements using MNA, we would end up with ∼20 additional
nodes and corresponding equations for each IM we encounter in the
network (refer to Figure 1.13). Therefore, to reduce the dimension
of IM equations, in this tutorial, we instead used a combination of
Kirchhoff Voltage Law (KVL)-based loop constraints and KCL-based
nodal constraints to stamp the IM model’s equivalent circuit elements.
With this approach, we will stamp the equations for electrical sub-
circuits (i through iv in Figure 1.13) in 4 new rows of the system matrix
Y following KVL-based loop constraints. Specifically, for each electrical
sub-circuit in Figure 1.13 and (1.22)–(1.25), moving from left to right,
we will add stamps (for voltages) for each element such that the net
sum of voltages in the loop is equal to zero. For the mechanical part (no.
(v) in Figure 1.13), we will add terms to matrix Y (including linearized
TE and companion circuits ∂ω ∂t ) following KCL-based nodal constraint.
r

With this approach, we only add five new variables and corresponding
rows for each instance of IM (instead of ∼20). The variables are Ids ,
Iqs , Idr , Iqr , and ωr .
Recall that before stamping the terms into the system matrix Y , we
have to perform two steps. First, we will replace the circuit elements
with time-derivative terms with corresponding equivalent companion
circuits. Second, we will replace the circuits with nonlinear terms with
their linearized approximation as shown for Fqr in (1.42). For instance,
the nonlinear symbolic stamps for Fqr loop constraint in (1.25) is added
to the Y matrix as shown in Figure 1.21.
Next, to obtain the transient response of the nonlinear IM network,
we will recursively solve the system matrix Y over time, and we will
perform iterations to solve the nonlinear terms. We will update the
linearized terms in each iteration. We will update the terms in companion
circuits only when we recursively move forward in time.
1.2. Induction Motor Load 111

Figure 1.21: Nonlinear IM stamps for Fqr KVL equation in (1.25).

1.2.4 Initialization
Next, we discuss how to initialize the IM network. In general, many
approaches are available to initialize the network and we will discuss
one such approach.
Remember because of dq-transformation on IM variables, two sub-
circuits evolve (see left and right of Figure 1.12). The sub-circuit on
the left without the IM equations is a linear circuit. AC analysis can be
used to obtain the initial condition for these, assuming a rated complex
current draw by IM (which models the Ia , Ib , and Ic in Figure 1.12).
A good guess for a complex current draw by IM can be obtained by
running power flow with IM modeled as a PQ load and calculating
the current from the solution (I = SIM ∗ /V ∗ ). Initialization of the IM
IM
components requires handling the nonlinearities due to speed-flux terms.
With the proper choice of the reference frame in dq-transformation,
time-invariant voltage sources (Vds and Vqs are constants) across the
IM can be obtained. We can then use DC analysis to obtain the initial
conditions for the IM circuit. The Vds , and Vqs voltages are DC-values
in the rotating reference frame. We can short the inductor and open
the capacitors to obtain the steady-state initial conditions with the
source voltages as DC values. The rated voltage at the IM terminals
for dq-transformation can be obtained from the power flow solution
VIM . In reality, with a slightly more involved procedure, we can get
exact initial conditions for the IM circuit by solving a set of nonlinear
equations representing the overall circuit iteratively using NR.
Generally, one must note that initializing large complex EMT net-
works is not trivial, and many commercial tool manuals [10], book
chapters [9], and research papers are devoted to the study of efficiently
initializing the network. In power systems, a common practice is to run
112 Equivalent Circuit Approach for EMT Simulations

load flow and then map the load flow solution from the positive sequence
frequency domain to three time-domain to initialize large power grid
EMT networks. However, with emerging inverter-based resources, this
approach may no longer work.

1.2.5 Nonlinear Time-Domain Simulation

For the nonlinear IM problem, with access to the initial state at t = 0,
we repeatedly solve a system matrix Y (moving forward by ∆t in each
recursion) following steps described in Figure 1.22 to obtain the time-
domain response from t = 0 to t = tf inal . How we update the system
matrix in the nonlinear analysis is different than in the case of linear
analysis in Section 1.1.7. Linearized terms in the system matrix Y
corresponding to nonlinear devices are updated iteratively after each
linear solve. The memory-stamps corresponding to the devices with
differential terms are only updated each time we move forward in time
(recursion). We only move forward in time once the nonlinear stamps
have converged (i.e., no considerable difference in the value of linearized
stamps in subsequent iterations). As in the linear case, we dynamically
adjust ∆t at each step based on the trade-off between simulation run-
time and local truncation error (LTE).

Figure 1.22: Running EMT simulation for nonlinear networks.

 Conclusion and Resources for a Deeper Dive

This monograph covers the fundamentals of building power systems

EMT tools using circuit theory with the aid of two simple examples.
The tutorial is introductory, and it does not include many critical topics
in detail, such as solver initialization, circuit-simulation heuristics for
steep nonlinear models, sparse matrices, new power electronics-based
models, and model reduction. For a deeper dive into the subject of
EMT simulations, I recommend the following references [3], [2], [9], [10],
and [6].
Parameter values for the two examples and a few others are in-
cluded in the .json format and shared in the following publicly avail-
able git repo: https://github.com/amritanshup7/Tutorial-Circuit-based-
Electromagnetic-Transient-Simulation/blob/main/example2_IM/IM_
circuit.json. Anyone interested in EMT tool implementation for the
two examples in Python language can reach out to me at amritan-
[email protected].

113
 Notation

Notations for parameters and symbols

Parameters Description

Simulation Variables and Parameters.

Y Nodal admittance matrix

V Solution vector of system states
J Independent source vector
k iteration count
∆t time-step
t time

DQ Transformation

θ Arbitrary value for stationary, synchronous, or rotating reference frame (rad)

β Rotor angular position wrt to reference frame (rad)
λ 2pi/3 (rad)

Linear Circuit Example

R a , R b , Rc Line resistances for phases abc (Ω)

La , Lb , Lc Line inductances for phases abc (H)
L
Ra , RbL , RcL Load resistances for phases abc (Ω)
LL L
a Lb , Lc
, L
Load inductances for phases abc (H)

Induction Motor Example

Fdr Nonlinear KVL equation for IM’s dr subcircuit

Fds Nonlinear KVL equation for IM’s ds subcircuit
Fqr Nonlinear KVL equation for IM’s qr subcircuit
Fqs Nonlinear KVL equation for IM’s qs subcircuit
Rs Rotor resistance (Ω)
Rs Stator resistance (Ω)
Ls Stator self-inductance (H)
Lr Rotor self-inductance (H)
Lm Magnetizing inductance (H)

114
 115

Parameters Description

Lls Stator leakage inductance (H)

Llr Rotor leakage inductance (H)
ωs Synchronous angular velocity (rad/s)
ωr Angular velocity of the rotor (rad/s)
Idr Current in rotor’s direct axis (A)
Ids Current in stator’s direct axis (A)
Iqr Current in rotor’s quadrature axis (A)
Iqs Current in stator’s quadrature axis (A)
Vdr Voltage across rotor’s direct axis (V)
Vds Voltage across stator’s direct axis (V)
Vqr Voltage across rotor’s quadrature axis (V)
Vqs Voltage across stator’s quadrature axis (V)
ψdr Flux linkages across rotor’s direct axis (Weber)
ψds Flux linkages across stator’s direct axis (Weber)
ψqr Flux linkages across rotor’s quadrature axis (Weber)
ψqs Flux linkages across stator’s quadrature axis (Weber)
J Combined inertia of motor and load (kgm2 )
D Combined viscous friction of motor and load (N · m/(rad/s))
θr Motor mechanical angular position (rad)
TL Motor load torque (Nm)
TE Motor electrical torque (Nm)
Np number of poles
 Acknowledgments

Tim McNamara and Naeem Turner-Bandele have both TA’ed the course
where this material was taught and they have spent a significant amount
of time ensuring that the derivations and explanations in this monograph
are precise and accurate. Tim also provided valuable feedback on the
structure and content. I also want to acknowledge my Ph.D. advisor
Larry Pileggi whose book, lecture slides, and guidance were critical in
compiling this monograph.

116
 References

[1] P. J. Davis and P. Rabinowitz, Methods of Numerical Integration.

Courier Corporation, 2007.
[2] H. W. Dommel, “Digital computer solution of electromagnetic
transients in single-and multiphase networks,” IEEE Transactions
on Power Apparatus and Systems, no. 4, pp. 388–399, 1969. doi:
10.1109/TPAS.1969.292459.
[3] H. W. Dommel, EMTP Theory Book. Microtran Power System
Analysis Corporation, 1996. doi: 10.1137/1.9780898718898.
[4] C. T. Kelley, Solving Nonlinear Equations with Newton’s Method.
SIAM, 2003.
[5] R. J. Lee, P. Pillay, and R. G. Harley, “D, q reference frames for the
simulation of induction motors,” Electric Power Systems Research,
vol. 8, no. 1, pp. 15–26, 1984. doi: 10.1016/0378-7796(84)90030-0.
[6] F. Milano, Power System Modelling and Scripting. Springer Sci-
ence and Business Media, 2010. doi: 10.1007/978-3-642-13669-6.
[7] A. Pandey, M. Jereminov, X. Li, G. Hug, and L. Pileggi, “Unified
power system analyses and models using equivalent circuit formu-
lation,” in 2016 IEEE Power & Energy Society Innovative Smart
Grid Technologies Conference (ISGT), pp. 1–5, IEEE, 2016. doi:
10.1109/ISGT.2016.7781182.

117
118 References

[8] A. Pandey, M. Jereminov, M. R. Wagner, D. M. Bromberg, G.

Hug, and L. Pileggi, “Robust power flow and three-phase power
flow analyses,” IEEE Transactions on Power Systems, vol. 34,
no. 1, pp. 616–626, 2018. doi: 10.1109/TPWRS.2018.2863042.
[9] L. Pillage, Electronic Circuit & System Simulation Methods (SRE).
McGraw-Hill, Inc., 1998. doi: 10.1109/MCD.1998.735799.
[10] PSCAD, “Pscad initialization tool—revision 1, written for pscad
5.0,” November 24, 2021.