Inclusion of frequency-dependent parameters in power

transmission lines simulation using harmonic analysis

and proper generalized decomposition
Muhammad Haris Malik, Domenico Borzacchiello, Francisco Chinesta, Pedro
Diez

To cite this version:

Muhammad Haris Malik, Domenico Borzacchiello, Francisco Chinesta, Pedro Diez. Inclusion of
frequency-dependent parameters in power transmission lines simulation using harmonic analysis and
proper generalized decomposition. International Journal of Numerical Modelling: Electronic Net-
works, Devices and Fields, Wiley, In press, �10.1002/jnm.2331�. �hal-01830008�

HAL Id: hal-01830008

https://hal.archives-ouvertes.fr/hal-01830008
Submitted on 5 Jul 2018

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
Inclusion of Frequency Dependent Parameters in Power
Transmission Lines Simulation using Harmonic Analysis
and Proper Generalized Decomposition
M.H. Malika , D. Borzacchielloa,∗, F.Chinestaa , P.Diezb
a
Institut de Calcul Intensif (ICI), École Centrale de Nantes, 1 rue de la Noë, BP 92101,
44321 Nantes cedex 3, France
b
Laboratori de càlcul numèric (LaCàN),C2 Campus Nord, Polytecnic University of
Catalunya, Calle Jordi Girona, 31, 08034 Barcelona, Spain

Abstract

This study presents the application of Proper Generalized Decomposition

on transmission line models involving frequency dependent parameters. Fre-
quency dependence of parameters can be due to a number of reasons including
phenomena such as “Ground Return Effects”, “Proximity Effects” and “Skin
Effects”. In the current study, simplified methods of skin effects based on
Bessel functions are used in order to showcase the method. Although we
implement our study using a specific model for skin effects to demonstrate
the effectiveness of the proposed method to accomodate frequency depen-
dent effects in PGD, the present work is not meant to discuss the merits
of any particular skin effects model. The method can easily accommodate
other effects which induces frequency dependence in the transmission line
parameters. In time domain modeling, the parameters are assumed constant

∗
Corresponding author
Email addresses: [email protected] (M.H. Malik),
[email protected] (D. Borzacchiello),
[email protected] (F.Chinesta), [email protected] (P.Diez)

Preprint submitted to Elsevier September 26, 2017

and these models prove inefficient when incorporating these parameters as
function of frequency. Therefore, a frequency domain simulation is imple-
mented using harmonic analysis. PGD presents a separated representation
and provides a quick and accurate solution for such problems.
Keywords: Skin Effect, Frequency Dependent Parameters, PGD,
Transmission Lines, Fractional Derivative

1. Introduction

Accurate modeling of transmission lines is important in the simulation

and subsequent transient analysis of power grids. In this framework, fre-
quency dependence of the line parameters should be accounted for [1]. The
parameters of transmission lines in practice are highly frequency dependent
and it is necessary to accommodate the frequency dependence over the whole
frequency range of interest.
AC transmission lines are affected by the frequency of the AC voltage
which produces a secondary effect on the line resistance and inductance. This
effect is related to the non uniform distribution of the current which tends to
concentrate more towards the outer surface of the line away from the center as
the frequency increases [2]. Skin effect results in the reduction of the effective
cross-section of the line used for the transmission of current and hence the
resistance increases. This concentration of current at the surface has an
inverse relation with the internal inductance of the line, as the frequency
increases the inductance decreases [3, 4]. There are other phenomena like
ground return and proximity which also have an effect on the transmission
line parameters resulting in the frequency dependence of these parameters [3].

2
If the phenomena which produce frequency dependence are ignored and the
parameters are assumed to be constant, this results in a magnification of the
higher harmonics [5]. The wave shapes are distorted and at higher frequency
the peaks have higher magnitudes than that are observed in practice. There
are a wide range of frequencies present in the signals during the transient
phase and this can be modeled with the frequency-dependent parameters.
The challenge in time domain modeling is to incorporate the frequency
dependence in an efficient manner and, as such, several formulations have
been proposed. As it will be discussed later in this paper, many of these
result in profound modifications of the original model that are sometimes
difficult to interpret in the light of their physical meaning. Accounting for
frequency related effects is obviously more natural when the governing equa-
tions are formulated in the frequency domain rather than in time. However,
this implies that incremental time stepping strategies for transient simula-
tion can no longer be applied and harmonic components have to be solved
individually, each requiring the solution of a different problem.
In this work we propose a robust parametric solver that allows to deter-
mine at once all the harmonics in a given frequency range. While this step
is performed “off-line”, the actual time response can be efficiently computed
“on-line”. The rest of this paper is organized as follows, Section 2 presents
the literature review. Section 3 presents the method proposed in the current
study to deal with the frequency dependent parameters in the frequency do-
main. In the Section 4 we present the skin effects model for the transmission
lines and in Section 5 we present the distributed parameters model of trans-
mission lines that we adopted in this study. Finally, we present the results

3
of our proposed approach in the Section 8.

2. Transmission Line Modeling

The frequency dependence of the parameters is the consequence of the as-

sumption made in the derivation of Telegrapher’s equations from the Maxwell’s
equations [6]. It is assumed that the conductor carries a uniform current in a
uniform time-harmonic magnetic field. Furthermore, it is assumed that the
curvature of the wire is negligible and there is no field variation in the wire
direction i.e. x-axis in this case. It is considered that the current flows in
x-direction only [7]. These assumptions simplify the 3-D Maxwell’s equation
to a manageable 1-D equation known as Telegrapher’s equation given by:

∂v(x, t) ∂i(x, t)
= −Ri(x, t) − L (1)
∂x ∂t
∂i(x, t) ∂v(x, t)
= −Gv(x, t) − C (2)
∂x ∂t

x ∈ [0, l] ; t ∈ [0, ∞]

where, v, i are the voltage and current, R, L, G and C are the resistance,
inductance, conductance and the capacitance per unit length respectively.
Eliminating i and combining the eqs. 1 and 2 yields

∂ 2 v(x, t) ∂v(x, t) ∂ 2 v(x, t)

LC + (LG + RC) + RGv(x, t) = (3)
∂t2 ∂t ∂x2
In order to retain the correct physical behavior, the model parameters like
resistance and inductance are often considered as function of the frequency
in order to compensate for the simplifications introduced by the assumptions

4
in the derivation of the 1-D model. The other scenario in which these param-
eters are kept constant will then require the 3-D partial differential equations
to be solved.
Several numerical methods have been proposed for the solution of fre-
quency dependent models [8]. An overview is listed in the Table 1.
There are mainly two families of solution procedures, i.e., a time-domain
simulation and a frequency-domain simulation. Both families of methods
have their pros and cons and have utility suitable to certain conditions. Time-
domain modeling is delicate and some numerical instabilities and accuracy
issues have been reported in the literature [5, 9]. But, the advantage it
provides is the straightforward compatibility with models for components of
the power system.
Most of the studies, reported in Table 1,avoid convolutions and developed
intelligent models to circumvent the difficulties in time-domain modeling by
modifying the formulation of the equations in time. For example, in the
study by Semlyen and Dabuleanu [10], the authors have used approximations
based on the exponential functions. The study by Chu-Sun et al [11] uses
equivalent circuit with each section is a combination of equivalent resistances
and inductances to include the skin effects. Some recent studies like Marques
da Costa et al [3] used fitting methods to include skin effects in lumped
parameters model of transmission line.
On the other hand, research studies in favor of frequency-domain model-
ing point out the ease of formulation. Instead of time convolutions, we can
readily multiply the harmonic response of the system with the desired input
function. Then obtaining the solution in time-domain is simply just a mat-

5
ter of inverse transformation. Wilcox and Condon [12] proposed an inverse
transformation method based on ARMA (Auto-Regressive-Moving-Average)
fitting method to obtain time-domain model, where ARMA is a statistical
model used in several fields to fit model to observed data. While other studies
like by Kurokawa et al [1] used modal transformation matrix. In frequency-
domain analysis, as described in the study of Meyer and Dommel [13], we
have to use linear superposition of all the frequency components. The solu-
tion is performed for every individual frequency present in the transients, as
the unit impulse response in time, “translates” as the unit heaviside response
in frequency-domain.
In the current study, we present a simple yet efficient frequency-domain
method based on the Proper Generalized Decomposition for the solution of
transmission line including the frequency-dependence of parameters due to
skin effects. We propose a method based on PGD’s separated representation
as a monolithic solver for all the frequencies.

3. Proposed Approach

Refering to eq. (3), the second term is referred as the dissipation term
and the last term is the dispersion term. For a lossless transmission line we
assume G = 0. By introducing a suitable finite element basis, the function
v(x, t) can be approximated as
n
X
v(x, t) ≈ vi (t)φi (x). (4)
i=1

Then, imposing Galerkin orthogonality of the residual of equation 3 on the

basis functions φi (x) leads to the following system of ordinary differential

6
equations in time

[M ]{v̈(t)} + [C]{v̇(t)} + [K]{v(t)} = {f (t)} (5)

where, [M ], [C] and [K] are referred to as the mass, damping and stiffness
matrices by analogy with structural dynamics. The unknowns of the prob-
lems are arranged into the vector {v(t)} = {v1 (t), v2 (t), . . . , vn (t)}. There-
fore, {v̇(t)} and {v̈(t)} indicate the first and second time derivatives of vector
{v(t)}.
Time integration of Eq. 5 requires the solution of a n×n algebraic system
for every time increment for nonlinear or time dependent matrices. Modal
analysis provides an efficient strategy to reduce the numerical complexity of
the system by transforming it into a set of n uncoupled equations.
The generalized eigenvalue problem to be solved for the system of eq. (5),
with damping [C] removed,

λ2 [M ] + [K] {Vi } = {0}

(6)

where, the vectors {Vi } are the eigenvectors and can be grouped in a ma-
trix [Λ] = [{V1 }, {V2 }, · · · , {Vn }]. The matrix [Λ] can be used to diagonalize
the matrices [M ] and [K] as:

[Λ]T [M ][Λ] = [m]

(7)
[Λ]T [K][Λ] = [k]

The system presented in Eq. (5) involves damping, there are generally
three ways to proceed further in cases of non-zero damping. The first method
is to use the direct time integration which we already explained could require

7
solution of n×n algebraic equations. The second method is to setup and solve
a different complex eigensystem in which eigenvalue problem diagonalizes two
matrices that contains [M ], [C] and [K] as sub-matrices [14]. The final and
the easiest method is to continue with the diagonalization presented in eq. (7).
For that purpose, we need the damping matrix [C] in a linear combination
of the matrices [M ] and [K] such that [C] = a[M ] + b[K]. This is usually the
case for the transmission line models because the parameters are constant
along the transmission line. Therefore it can also be diagonalized using the
same matrix [Λ].

[Λ]T [C][Λ] = a[m] + b[k] (8)

Also, the variable of interest is transformed into the modal basis as:

{U } = [Λ]T {V } (9)

This diagonalization gives us an uncoupled ordinary differential equations

that can be solved separately. Writing the equations in indicial notation,

mi üi + (ami + bki )u̇i + ki ui = 0 (10)

The solution of eq. (9) is straightforward and once we have the solution
in terms of modal basis, we can transform the solution back to the physical
basis.

{V } = [Λ]{U } (11)

The case we address here is when the parameters resistance R(ω) and

8
inductance L(ω) becomes frequency dependent because of skin or other ef-
fects. The problem where we have frequency-dependent parameters can not
be solved needs special attention. Studies such as the one performed by
Wilcox and Condon used the modal analysis but modeled the system with
frequency-dependent parameters first in frequency-domain and careful ma-
nipulation of the system itself is required, otherwise the transformation from
frequency-domain model to time-domain model may cause the loss of causal-
ity as seen by Crandall [15] in the context of structural dynamics. In his work,
Crandall discusses the effect of frequency-dependent parameters on time in-
tegration and introduced the concept of non-equations. The better option is
to solve the harmonic analysis problem for every frequency ω. Transforming
the problem from time-domain to frequency-domain using Fourier transfor-
mation of the unknown variable vi (t) and forcing function f (t) that write

Z ∞
V̂ (ω) = v(t)e−ιωt dt
Z−∞
∞ (12)
−ιωt
F̂ (ω) = f (t)e dt
−∞

Therefore, the equation (5) after transformation into frequency-domain

is given as:

−ω 2 [M (ω)] + ιω[C(ω)] + [K(ω)] V̂ (ω) = F̂ ∀ ω

(13)

Now, the harmonic analysis problem requires solution of n systems of

linear equations where n represent the number of discrete frequencies consid-
ered. A similar problem involving fractional damping was discussed in [16].
The use of proper generalized decomposition (PGD) as a paramteric solver

9
as presented in earlier studies by Chinesta et al. [17] and [18] is considered
here as well. PGD provides a strong tool as a single monolithic solver as
compared to solving the harmonic problem using eq. 13 individually for ev-
ery frequency. PGD has already been implemented for the parametric power
flow problem [19] and an error estimation strategy has also been developed
[20].Note that if we solve the linear problem of eq. 13 we have to compute the
matrices for every single frequency as they are no longer constant compared
to the case where the constant parameters was considered and depend upon
the frequency. Therefore, in this study, we present the Proper Generalized
Decomposition (PGD) as a parametric solver, a method capable to solve the
harmonic analysis.

4. Skin Effects Model

Skin Effect results in frequency dependence of the transmission line pa-

rameters. This phenomenon is the result of unequal distribution of current
across the cross-section of the transmission line, this is more pronounced in
long transmission lines. The parameters that become functions of frequency
are resistance and inductance. Inductance of a conductor is related with
the flux linkage, the flux linkage in a conductor happens internally and ex-
ternally. Internal flux linkage is due to the self current of the conductor
while the external flux linkage is due to the presence of adjacent conductors.
Therefore, total inductance of a transmission line is the sum of inductances
due to both internal and external flux linkages. The evaluation of external
inductance is based on the assumption that the total current flow is on the
surface of the transmission line, i.e., external inductance is maximum when
the skin effect is maximum. However, the external inductance is the result of

10
the current flow in the neighboring transmission line and therefore remains
constant. In the current case, we are dealing with the single transmission
line assuming no adjacent transmission lines and in the succeeding discussion
wherever inductance is mentioned it is assumed internal inductance unless
explicitly mentioned.
Typically transmission lines are bundled aluminum strands twisted to-
gether with one kind of transmission line consists of steel strands in the core
to increase the strength of the transmission line. However, for the sake of sim-
plicity the discussion is based on a solid cylindrical conductor to effectively
demonstrate the model.
Several researchers have presented simple models to include skin effects in
the transmission line models [9], [21] and [2]. Since, our focus in the current
study is not to model the skin effects but rather use the existing model and
present a solution that is able to perform transient simulations with skin
effects included. Therefore, we present here some of the simplified models
and use them in our study.
For a transmission line with large radius compared to the skin depth [22],
δ << 4r where skin depth δ is calculated as (14).

 1/2
2
δ= (14)
ωµσ
where, r is the radius of the cross-section of the conductor, ω is the angular
frequency and related to frequency in Hz as ω = 2πf , µ is the magnetic
permeability and σ is the electrical conductivity of the conductor.
The resistance R(ω) and inductance L(ω) are given in the study of Lucht
[22] based on the expansion of Bessel functions with the assumptions listed

11
here. These simplified functions for resistance and inductance for high fre-
quencies is given as:

1 1
R(ω) = , L(ω) = (15)
σ(2πr)δ σ(2πr)δω
In the technical report by Phil Lucht [22], a low frequency limit is de-
fined, above this frequency resistance and inductance become the functions
of frequency, below this limit the resistance and inductance are given by

1 µ
R= , L= (16)
σπr2 8π
The low frequency resistance is the same as the DC resistance of a con-
ductor in per unit length given by

ρ
R= (17)
πr2
where ρ is the resistivity of the conducting material.
The article by Monteiro et al [23] presents a simplified formulation for the
calculation of skin effects on the transmission line wires. The formulation is
based upon the Fourier transformation of the Maxwell’s equation [24].The
methodology presented by Monteiro et al is simple in terms of its application
without the need to use Bessel functions and also presented good accuracy
as it is directly developed from the Maxwell’s wave equations.

P∞ 2 2 2
k=1 (Rk /(Rk + ω L ))
R(ω) = P∞ (18)
( k=1 Rk /(Rk2 + ω 2 L2 )) + ω 2 ( ∞
2 P 2 2 2 2
k=1 L/(Rk + ω L ))
P∞ 2 2 2
k=1 (L/(Rk + ω L ))
L(ω) = P∞ (19)
( k=1 Rk /(Rk2 + ω 2 L2 )) + ω 2 ( ∞
2 P 2 2 2 2
k=1 L/(Rk + ω L ))

12
 (a) Step Response (b) Sine Response

Figure 1: Frequency dependent resistance and inductance of a conductor

where
ξk2 µ (2k − 1)π π
Rk = ; L= and ξk = + (20)
4πσr2 4π 2 4
The units of Rk is Ω.m and L is H/m. The units of electrical conductivity
σ is S/m and of the magnetic permeability µ is H/m. The term ξk is the
k th Bessel function root and is directly proportional to k. As mentioned
previously, the transmission line in this study is considered to be a solid
cylindrical copper wire, its radius and the electrical properties of the copper
are listed in Table 2.
A comparison of different models was performed and the results were
compared with the results from the study of Monteiro et al [23]. The results
for the resistance and inductance with respect to the frequency is presented
in the Figures 1a and 1b.

13
5. Distributed Parameters Model
5.1. Time Domain Simulation

We start with the time domain distributed parameters (DP) model. The
electrical transmission lines provide the path for the voltage and current
waves to propagate, the dynamics of the propagation is described by a dis-
tributed parameters model. This model depends upon the considerations
taking into account like the losses and the frequency dependence of the pa-
rameters.
The distributed parameters model for a lossy transmission line with con-
stant parameters can be written in the time domain as:

∂v(x, t) ∂i(x, t)
= −L − Ri(x, t),
∂x ∂t (21)
∂i(x, t) ∂v(x, t)
= −C
∂x ∂t
for 0 ≤ x ≤ l, where v(x, t) and i(x, t) are the voltage and current in
the line respectively, and the resistance, the inductance and the capacitance
are represented by R, L, C respectively and these constants are positive and
independent of frequency. The parameters R, L, C are given in per unit
length and in distributed parameters model are used without multiplication
with the length of the line. However, these parameters must be multiplied
by the length of the line in case one is using any lumped circuit model
for transmission lines e.g. pi-equivalent model. The effect of transverse
conductance G of the line is neglected in this model. The interconnection
between the transmission line, the load and the generator is performed using
the Kirschhoff’s laws and the boundary values.

14
 v(x, t)|x=0 = V0 (t)
(22)
i(x, t)|x=l = 0

The term V0 is the voltage source assumed to represent a generator, the

boundary condition given above assumes that the voltage source is ideal, that
is, the impedance is zero (Z0 = 0) and the impedance at the receiving end
(x = l) is infinite (Zl = ∞).
The equation is converted to a single variable second order differential
equation in time and space.

∂ 2v ∂ 2v ∂v
2
− LC 2
− RC =0 (23)
∂x ∂t ∂t
The equation (23) is then discretized in time and space, using finite dif-
ference scheme for the time integration and finite elements for the space.
For a copper conductor wire, we can calculate the resistance and induc-
tance from the equations (16), these values are listed in the table 3. Solving a
space time problem for these values, with a boundary condition representing
a step as an input voltage V0 .

0 for t 6 0
V0 = (24)
1 for t > 0

Also, a sine function with a frequency of f = 100Hz was used as an input

voltage for the same system

 0 for t 6 0
V0 = (25)
sin(2π100t) for t > 0

15
 (a) Step Response (b) Sine Response

Figure 2: Time simulation of constant parameters model in space-time domain

The results of time simulation for the receiving end voltage x = l is

presented graphically in the Figures 2a and 2b.

6. Frequency-Domain Model

Skin effects phenomena results in frequency dependence of DP model,

direct time integration of which is difficult to perform. But, the problem in
the frequency domain becomes a linear second order differential equation,
where the parameters resistance and inductance are now a function of fre-
quency. Fourier transformation of the time domain distributed parameters
model gives the equation in the frequency domain as:

∂ V̂ (x, ω) ˆ ω).(R + ιLω),

= −I(x,
∂x (26)
ˆ ω)
∂ I(x,
= −V̂ (x, ω).(ιCω)
∂x
Combining the two equations given in (26), we get a second order ordinary
differential equations for voltage.

16
 ∂ 2 V̂ (x, ω)
+ (LCω 2 − ιRCω)V̂ (x, ω) = 0 (27)
∂x2
Corresponding boundary conditions in the frequency domain is given as


V̂ (x, ω) = V0

x=0
(28)

∂ V̂ (x, ω) 
=0
∂x 

x=l

7. Proper Generalized Decomposition for Constant Parameters Model

The problem presented in Sec. 5 can be solved using PGD that assures
the separated representation,

m
X
V̂ (x, ω) = Xi (x).Oi (ω) (29)
i=1

For the details about PGD formulation and its advantage, one can refer
to several publications for example, [18, 17]. In general, PGD methods com-
prises of an offline and an online part. In the offline stage of the solution,
the problem is formulated in the separated representation and the solution
database is created. The online part then is just to compute the solution
for the particular problem by multiplying the separated coordinates for each
enrichment mode.
Once, a solution using PGD with the boundary conditions given in (28)
is obtained, it is straightforward to get the desired time response using the
discrete inverse Fourier function available in MATLAB. Using the separated
representation of (29), the DP model can be expressed in separated repre-

17
sentation. In order to develop a PGD formulation, first we must have a weak
formulation of the problem.
Multiply equation (27) with a test function δV ∗ (x) where, δV represents
the test function and δV ∗ (x) represents the complex conjugate.

Z 2 Z
∗∂V̂
δV dx dω − δV ∗ f (ω)V̂ dx dω = 0 (30)
Ωx ×Ωω ∂x2 Ωx ×Ωω

where, f (ω) = (LCω 2 − ιRCω).

Here, we briefly describe the PGD algorithm, for details the interested
readers can refer to the literature [17, 18]. PGD follows the greedy algorithm
approach, i.e., one mode is evaluated at a time. The mode is assumed to be
converged after sufficient iterations result in no further improvement. This
procedure of adding modes is termed as the enrichment process. Assuming
that n − 1 initial modes are already converged, we define the nth mode to be
evaluated as:
n−1
X
V̂ (x, ω) = Xi (x).Oi (ω) + Xnp (x)Onp−1 (ω) (31)
i=1

We assume the test function δV ∗ (x, ω) in the separated form as

δV ∗ (x, ω) = O∗ (ω)δX ∗ (x) + X ∗ (x)δO∗ (ω) (32)

In eq. (31), the superscript p indicates the number of fixed point itera-
tions. For individual modes evaluation, we apply the fixed point loop which
is to find solution for each coordinate space alternatively.
Since, the problem is now separated in the space X(x) and the frequency
O(ω), we have to solve two problems using FEM once for the space domain

18
and then for the frequency domain. The weak formulation for the space only
problem is:

dδXn∗ dXnp p−1∗ p−1

Z Z
On On dx dω + f (ω)δXn∗ Xnp Onp−1∗ Onp−1 dx dω =
Ωx ×Ωω dx dx Ωx ×Ωω
Z n−1 ∗ Z n−1
X dδXn dXi p−1∗ X
− On Oi dx dω − f (ω) δXn∗ Xi Onp−1∗ Oi dx dω
Ωx ×Ωω i=1 dx dx Ωx ×Ωω i=1
Z
+ ˆ
δXn∗ Onp−1∗ I(ιLω + R) dω
Ωω

(33)
and for the problem to be solved in the frequency domain, the weak formu-
lation will be

dXnp∗ dXnp ∗ p
Z Z
δOn On dx dω + f (ω)Xnp∗ Xnp δOn∗ Onp dx dω =
Ωx ×Ωω dx dx Ωx ×Ωω
n−1 n−1
dXnp∗ dXi ∗
Z X Z X
− δOn Oi dx dω − f (ω) Xnp∗ Xi δOn∗ Oi dx dω
Ωx ×Ωω i=1 dx dx Ωx ×Ωω i=1
Z
+ ˆ
Xnp∗ |x=0 δOn∗ I(ιLω + R) dω
Ωω

(34)
The tolerances selected for the stopping of the fixed point algorithm εp
and the tolerance for termination of enrichment process εn , along with the
number of modes n in both the constant parameters and frequency-dependent
parameters are listed in Table 4.
The boundary condition in eq (28) is an equivalent of having an impulse
function at the source which gives the impulse response or the transfer func-
tion Ĥ of the transmission line. The voltage impulse response at the receiving
end is given as:

19
 n
X
Ĥ(x = l, ω) = Xi (x = l)Oi (ω) (35)
i=1

Once, this is done, the time response to any arbitrary condition at the
source node is then straightforward multiplication of the voltage function in
frequency domain with the particular solution of the PGD at the receiving
node. Then the frequency response to an arbitrary function Vˆ0 is given as

V̂ (x = l, ω) = Ĥ(x = l, ω) · Vˆ0 (36)

The input voltage V̂0 is usually a step function or a sinusoidal function

in time. In practice, these functions are transformed into frequency-domain
using the fast Fourier transform algorithm to compute the discrete Fourier
transform. Similarly, the response V̂ is inverse transformed to time domain
using the same fast Fourier algorithm using discrete samples in the frequency
interval.

7.1. Verification of PGD results

In this section, we first present the time response using inverse fourier
transform of the PGD solution particularized at the receiving end.
The PGD solution in terms of voltage on a space-frequency domain is
presented in Figures 3a to 3c.
The results from Figures 5a and 5b shows good correlation between the
solution from time integration and the PGD solution. Therefore, it is estab-
lished that the PGD solution is an accurate representation of the DP model
of transmission lines. The difference between the two curves of in the Fig-
ure 5a is due to the fact that the time simulation is carried out using 1st

20
 (a) Absolute Value (b) Real Value

(c) Imaginary Value

Figure 3: Voltage in space-frequency domain with constant parameters

(a) Step Response (b) Sine Response

Figure 4: Time response by inverse Fourier of the PGD solution

21
 (a) Step Response (b) Sine Response

Figure 5: Comparison of time response between PGD solution and time integration

order backward difference scheme. The blue curve of time simulation tends
to approach the red curve of PGD solution as finer time discretization is
used but even then spikes observed in the PGD solution are not captured
by the time simulation. In comparison, PGD based approach uses frequency
domain which by default is a higher order approximation and hence able to
capture a wider range of dynamics.

8. PGD for frequency dependent parameters

The benefit of solving DP model in the frequency domain in the sepa-

rated representation of PGD is the easy, fast and accurate solution with the
option of a parametric solution. PGD methods provide the best option for
parametric problems both in terms of computational efficiency and ease of
solving the problem. In the case of skin effects, the equation (27) now has
parameters resistance R and inductance L as a function of frequency ω. As
explained earlier time integration of such problems is almost impossible to
achieve and hence PGD provides real accurate option to solve such problems.

22
The problem is similar to the case of fractional derivatives which is similar
to the fractional RLC circuit [25].
Eq (27) with R and L as functions of frequency reads as:

∂ 2 V̂ (x, ω)
2
+ (L(ω).C.ω 2 − ιR(ω).C.ω)V̂ (x, ω) = 0 (37)
∂x
The increase in the resistance with the frequency and a simultaneous
decrease in the inductance results in a system more damped than the system
with constant resistance and inductance. This effect can be noticed in the
Figures 6a to 6c showing the voltage in the space-frequency domain, there is
only one peak visible with a much less amplitude.
The effect of increased damping due to the increase in resistance can
be compared in the receiving end voltage for the frequency dependent and
constant paramters solution given in the Figures 7. In Figure 8, the two re-
sponses from the constant parameters and frequency dependent parameters
are plotted on top of each other for a direct comparison. The increased damp-
ing removes almost all the oscillations observed in the constant parameters
solution of Figures 4a and 4b. The difference between the constant parame-
ters model and frequency dependent parameters solution are very different.
Therefore, just to elaborate the effect of different values of the parameters
we used the constant parameters model with parameters values evaluated at
60 Hz and 120 Hz and plotted in the Figure 9.

8.1. Parametric Solution

The advantage of PGD based harmonic analysis is that any parameter can
be introduced as a model extra-coordinate into the separated representation.

23
 (a) Absolute Value (b) Real Value

(c) Imaginary Value

Figure 6: Voltage in space-frequency domain with skin effects included

(a) Step Response (b) Sine Response

Figure 7: Time response from PGD solutions for frequency dependent parameters

24
 (a) Step Response (b) Sine Response

Figure 8: Comparison of response between constant and frequency dependent parameters

(a) Step Response (b) Sine Response

Figure 9: Comparison of response for constant parameters with parameter values at 60

Hz and 120 Hz

25
As an initial step, we introduce length of the transmission line as a new
dimension for the parametric problem. The solution now is given as:

n
X
V̂ (x, ω, l) = Xi (x)Oi (ω)Di (l) (38)
i=1

where Di represents the modes of the dimension introduced to represent

length.
The mathematical formulation of the problem remains similar and there
is no significant additional computational cost to the problem. For the case,
we have reported in this study, the new dimension of transmission line length
is considered for a range (200, 300) km. The discretization for the space and
frequency have been kept the same as in previous case. The discretization
and the other information valuable to gauge the PGD solution is listed in
Table 5.
The advantage of parametric solution is presented in Figure 10 where the
variation of voltages along the transmission line with different lengths for a
frequency of 60 Hz is plotted. Another advantage of such a solution is in the
sensitivity or uncertainty analysis, in such an analysis a fixed point is selected
and a variation of usually 5% in the independent variable and its effect on
the variable of interest is studied. With a PGD solution, we already have a
complete solution as a curve and the derivative can be easily evaluated for
the sensitivity analysis.

9. Conclusions

The work presented in this study, shows the effectiveness of frequency-

domain methods for problems involving frequency-dependent parameters.

26
Figure 10: Variation of voltage along the transmission line for different lengths at 60 Hz

The main contribution of the study presented in this paper is the develop-
ment of a reduced order model able to include phenomenon like skin effects
that cause the parameters to become functions of frequency. The method is
equally applicable for the simulation of transmission line models including
the ground return and proximity effects which has a similar effect on the
parameters, i.e., the frequency dependence. In particular, harmonic analysis
using PGD is fast, accurate and simple method for this kind of parametric
problem. In this paper, we have demonstrated the method of PGD, con-
structing the database of the parametric solution offline and particularize
the solution online for a specific set of parameters online in real time. PGD
is an a-priori model order reduction method and is particularly effective to
obtain solutions online from the pre-computed database of solutions. With
the ability of including parameters such as the length of the transmission
line or material properties of the conducting material as one of the coor-
dinates of the separated representation, it truly provides the capability of
having real time solutions using not so power platforms. In contrast to the

27
method presented in the current study, there are solution methods available
that include the frequency-dependence of parameters and simulate the model
in time-domain, but these methods are complicated to implement and since
they use time integration it is computationally costly and time consuming.
PGD is already established as a fast and accurate method for reduced order
modeling. Another potential application of this method could be fractional
RLC circuits.

References

[1] Sérgio Kurokawa, J. Pissolato Filho, M.C. Tavares, C.M. Portela, and
A.J. Prado. Behavior of Overhead Transmission Line Parameters on
the Presence of Ground Wires. IEEE Transactions on Power Delivery,
20(2):1669–1676, apr 2005.

[2] M. Admane, M. Sorine, and Q. Zhang. Simulation of electric transmis-

sion lines with skin effect based on a reduced order differential model. In
Reliability in Electromagnetic Systems, 2007 IET Colloquium on, pages
1–5, May 2007.

[3] Eduardo Coelho Marques da Costa, Sérgio Kurokawa, Afonso José

do Prado, and José Pissolato. Proposal of an alternative transmission
line model for symmetrical and asymmetrical configurations. Interna-
tional Journal of Electrical Power & Energy Systems, 33(8):1375–1383,
oct 2011.

[4] Dierk Bormann and Hanif Tavakoli. Reluctance Network Treatment

28
 of Skin and Proximity Effects in Multi-Conductor Transmission Lines.
IEEE Transactions on Magnetics, 48(2):735–738, feb 2012.

[5] J. R. Marti. Accurate Modeling of Frequency-Dependent Transmission

Lines in Electromagnetic Transient Simulations. IEEE Transactions on
Power Apparatus and Systems, PAS-101(1):147–157, 1982.

[6] C.A. Nucci, F. Rachidi, and M. Rubinstein. Derivation of telegrapher’s

equations and field-to-transmission line interaction. In Electromagnetic
Field Interaction with Transmission Lines, volume 29, pages 3–22. WIT
Press, feb 2008.

[7] Hajime Igarashi. Semi-Analytical Approach for Finite Element Anal-

ysis of Multi-turn Coil Considering Skin and Proximity Effects. IEEE
Transactions on Magnetics, 02(c):1–1, 2016.

[8] V Maló Machado, M Eduarda Pedro, J.A. Brandão Faria, and D. Van
Dommelen. Magnetic field analysis of three-conductor bundles in flat
and triangular configurations with the inclusion of proximity and skin
effects. Electric Power Systems Research, 81(11):2005–2014, nov 2011.

[9] Kyung Suk Oh. Accurate transient simulation of transmission lines

with the skin effect. IEEE Transactions on Computer-Aided Design of
Integrated Circuits and Systems, 19(3):389–396, mar 2000.

[10] A. Semlyen and A. Dabuleanu. Fast and accurate switching transient

calculations on transmission lines with ground return using recursive
convolutions. IEEE Transactions on Power Apparatus and Systems,
94(2):561–571, Mar 1975.

29
[11] Chu-Sun Yen, Zvonko Fazarinc, and R.L. Wheeler. Time-domain skin-
effect model for transient analysis of lossy transmission lines. Proceedings
of the IEEE, 70(7):750–757, 1982.

[12] D.J. Wilcox and M. Condon. A new transmission-line model for time-
domain implementation. COMPEL - The international journal for
computation and mathematics in electrical and electronic engineering,
16(4):261–274, dec 1997.

[13] W. S. Meyer and H. W. Dommel. Numerical modelling of frequency-

dependent transmission-line parameters in an electromagnetic transients
program. IEEE Transactions on Power Apparatus and Systems, PAS-
93(5):1401–1409, Sept 1974.

[14] R R Craig Jr. and A J Kurdila. Fundamentals of Structural Dynamics.

John Wiley & Sons, Ltd, New Jersey, 2006.

[15] S.H. Crandall. The role of damping in vibration theory. Journal of

Sound and Vibration, 11(1):3 – IN1, 1970.

[16] C. Germoso, A. Fraile, E. Alarcon, J.V. Aguado, and F. Chinesta. From

standard to fractional structural visco-elastodynamics: Application to
seismic site response. Physics and Chemistry of the Earth, Parts A/B/C,
pages –, 2016.

[17] F. Chinesta, A. Ammar, A. Leygue, and R. Keunings. An overview of the

proper generalized decomposition with applications in computational
rheology. Journal of Non-Newtonian Fluid Mechanics, 166(11):578–592,
2011.

30
[18] F. Chinesta, A. Leygue, F. Bordeu, J. V. Aguado, E. Cueto, D. Gonza-
lez, I. Alfaro, A. Ammar, and A. Huerta. PGD-Based Computational
Vademecum for Efficient Design, Optimization and Control. Archives of
Computational Methods in Engineering, 20(1):31–59, 2013.

[19] R. Garcı́a-Blanco, P. Dı́ez, D. Borzacchiello, and F. Chinesta. A reduced

order modeling approach for optimal allocation of distributed generation
in power distribution systems. In 2016 IEEE International Energy Con-
ference (ENERGYCON), pages 1–6, April 2016.

[20] R. Garcı́a-Blanco, D. Borzacchiello, F. Chinesta, and P. Dı́ez. Monitor-

ing a pgd solver for parametric power flow problems with goal-oriented
error assessment. International Journal for Numerical Methods in En-
gineering, pages n/a–n/a, 2016. nme.5470.

[21] O.M.O Gatous and J Pissolato. Frequency-dependent skin-effect for-

mulation for resistance and internal inductance of a solid cylindrical
conductor. IEE Proceedings - Microwaves, Antennas and Propagation,
151(3):212, 2004.

[22] Phil Lucht. Transmission Lines and Maxwell’s Equations. Technical

report, 2014.

[23] José Humberto A Monteiro, Eduardo Coelho M Costa, André Jinno G

Pinto, Sérgio Kurokawa, Omar Mohamed O Gatous, and José Pissolato.
Simplified skin-effect formulation for power transmission lines. IET Sci-
ence, Measurement & Technology, 8(2):47–53, mar 2014.

31
[24] Robert A. Chipman. Theory and Problems of Transmission Lines.
McGraw-Hill, 1968.

[25] Gómez-Aguilar José Francisco, Rosales-Garcı́a Juan, Guı́a-Calderón

Manuel, and Razo-Hernández José Roberto. Fractional RC and LC
electrical circuits. Ingenierı́a, Investigación y Tecnologı́a, 15(2):311 –
319, 2014.

[26] M. Davila, J. L. Naredo, P. Moreno, and A. Ramirez. Practical imple-

mentation of a transmission line model for transient analysis considering
corona and skin effects. In Power Tech Conference Proceedings, 2003
IEEE Bologna, volume 2, pages 6 pp. Vol.2–, June 2003.

[27] B. Gustavsen, J. Sletbak, and T. Henriksen. Calculation of electromag-

netic transients in transmission cables and lines taking frequency de-
pendent effects accurately into account. IEEE Transactions on Power
Delivery, 10(2):1076–1084, apr 1995.

[28] S Kim and D.P. Neikirk. Compact equivalent circuit model for the
skin effect. In 1996 IEEE MTT-S International Microwave Symposium
Digest, volume 3, pages 1815–1818. IEEE, 1996.

32
Authors Phenomena Methods
Time Domain Modeling
Semlyn and Approximation of time variable
Dabuleanu Ground return effect characteristic admittance using expo-
[10] nential functions
Weighting Functions,
J. R. Marti [5] Ground return effect
Characteristic Impedance
Equivalent Circuit using parallel
Chu-Sun et al [11] Skin effect
equivalent resistors and inductors
Suk Oh [9] Skin effect Difference Approximation Method
Ground return,
Marques da Fitting Procedure for state space
skin and corona
Costa et al [3] lumped parameters models
effects
Corona and Skin Differential Integral Equation
Dávila et al [26]
Effects (Radulet Line Equations)
Frequency Domain Modeling
Meyer and Linear Superposition
Ground return effect
Dommel [13] of all frequency components
Modal Analysis,
Wilcox and Conversion to time-domain using Auto-
Skin Effects
Condon [12] Regressive-Moving-Average (ARMA)
as fitting method
Frequency de-
pendent effects Travelling wave model,
Gustavsen et al [27]
of underground Recursive Convolutions
cables
Clarke’s matrix as the
Kurokawa et al [1] Ground return effect
modal transformation matrix
Skin and Prox- Magnetic Field Analysis using
Machado et al. [8]
imity Effects Bessel function development
Bormann and Skin and Prox-
Tavakoli [4] imity Effects Reluctance networks
Gatous and Simplified expressions of R and L
Pissolato [21] Skin Effects based on Bessel Functions
Kim and Compact Circuit Model
Neikirk [28] Skin Effects in form of Ladder Circuit
Admane et al Reduced order differential model
[2] Skin Effects based on Kim’s Ladder Circuit
Table 1: Literature review
33
Parameter Value Units
Radius (r) 30 mm
Length (l) 300 km
Electrical Conductivity (σ) 5.96 × 107 S/m
Magnetic Permeability (µ) 1.26 × 10−6 H/m
Table 2: Copper conductor wire properties

34
Parameter Value Units
Resistance (R) 5.7603 × 10−3 ohm/km
Inductance (L) 5.0980 × 10−5 H/km
Capacitance (C) 1.00 × 10−8 F/km
Length (l) 300 km
Space Discretization (∆x) 1.0 × 10−1 km
Total Time (T) 3.0 × 10−2 sec
Time Step (∆t) 1.0 × 10−5 sec

Table 3: Properties of a copper wire

35
Entity Constant Parameters Frequency Dependent
εp 10−10 10−10
εn 10−10 10−10
n 40 14
Residual 5 × 10−16 4 × 10−19
Computation time 250 sec 50 sec
Table 4: PGD criteria

36
Entity Parametric Frequency Dependent PGD
∆l 0.05
εp 10−10
εn 10−10
n 67
Residual 1 × 10−7
Computation time 700 sec
Table 5: PGD Criteria for parametric solution

37