1158

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 17, NO. 4, OCTOBER 2002

Flicker Study Using a Novel Arc Furnace Model

Omer Ozgun and Ali Abur

AbstractVoltage flicker and harmonics are the types of powerquality problems that are introduced to the power system as a result of arc furnace operation. Utilities are concerned about these
effects and try to take precautions to minimize them. Therefore,
an accurate model of an arc furnace is needed to test and verify
proposed solutions to this end.
In this paper, an arc furnace model is developed and implemented in a Simulink environment by using chaotic and
deterministic elements. Moreover, the modeling and simulation of
an IEC flickermeter are also performed to evaluate the severity of
fluctuations in the simulated arc furnace voltage.
Index TermsArc furnace, chaos, Chuas circuit, flickermeter,
harmonics, modeling, Simulink, voltage flicker.

I. INTRODUCTION

RC furnaces are used for melting and refining metals. As

the popularity and use of the arc furnace loads in the industry increase, so does the power-quality problems as a result
of this progress. The voltage flicker and the existence of harmonic load currents can be given as examples of adverse effects introduced by arc furnaces. In order to propose solutions
to minimize these adverse effects of arc furnaces, the impact of
these highly nonlinear, time-varying loads on the power quality
of the overall power system should be investigated. Therefore,
obtaining the time response of an electric arc furnace becomes
very important at this point. Hence, developing an accurate and
easy-to-use arc furnace model is a much needed task. However, this has so far been quite a challenging task for several
researchers due to the aperiodic, highly nonlinear, and unpredictable nature of the arc furnaces. Although several studies
have been done in order to build a generic model of an arc furnace, an accurate circuit model to represent the arc furnace operation is still not available.
Historically, there have been two general approaches to the
problem of arc furnace modeling: stochastic and chaotic. In
most of the previous studies, stochastic ideas are used to capture the aperiodic, nonlinear, and time-varying behavior of arc
furnaces [1][3]. It is a fact that frequency modulation (FM)
of the supply voltage that is less than 0.5% can cause voltage
flicker if the frequency of the modulating signal lies within
the 6-to-10-Hz range [3]. Knowing this fact, generating white
noise within the mentioned frequency range and modulating it
with the arc voltage to resemble the behavior of an arc furnace
voltage has been a common approach in these models [4]. Arc
Manuscript received November 2, 1999; revised July 19, 2001. This work was
supported in part by the Texas Higher Education Coordinating Board through
Project ATP-036327-094.
The authors are with the Department of Electrical Engineering, Texas A&M
University, College Station, TX 77843 USA (e-mail: ozgun@ee.tamu.edu,
abur@ee.tamu.edu).
Digital Object Identifier 10.1109/TPWRD.2002.804013

voltage has been obtained either from the simplified characteristics of the arc furnace load, or from the empirical formulas related to the arcing process.
In [3], the arc furnace load is modeled as a voltage source,
and the model is based on the characteristics, using the sinusoidal variation of the arc resistance and bandlimited white
noise variation as well. On the other hand, system-identification techniques are used in [1] to estimate the stochastic characteristics of the generated model. Instead of focusing on the
detailed modeling of - characteristics, a simplified characteristic is chosen and the time-dependent variation of the electric
arc voltage dominates the model. Another model, which consists
of the nonlinear, time-varying resistance where both sinusoidal
and stochastic time variation rules of arc length are considered,
is presented in [2].
Making use of the chaos theory in arc furnace modeling [4]
is rather new compared to the stochastic ideas. Actually, chaos
theory has become a modeling issue for arc furnaces after the
electrical fluctuations in the arc furnace voltage were proven to
be chaotic in nature [5].
In this paper, the arc voltage is simulated by solving the corresponding differential equation, which yields dynamic and multivalued characteristics of the arc furnace load. On the other
hand, a low-frequency chaotic signal is generated via simulation of Chuas well known chaotic circuit. These two signals
are modulated to form the final arc furnace terminal voltage.
The proposed model is connected to the system as a controlled
voltage source. Preliminary results of the model development
were presented in [6]. Finally, in order to evaluate the flicker in
the simulated arc furnace voltage, the IEC flickermeter is implemented based on the IEC 1000-4-15 standard [7] in Matlab
environment by using discrete time methods. In this paper, we
show a flicker study using the developed arc furnace and IEC
flickermeter models.

II. CHAOTIC RESPONSES IN ARC FURNACES

Chaos, also known as the strange attractor, does not generally have an accepted precise mathematical definition. Usually,
from a practical point of view, it can be defined as the bounded
steady-state behavior that does not fall into the categories of the
other three steady-state behaviors (i.e., equilibrium points, periodic solutions, and quasiperiodic solutions [8]). While equilibrium points are zero dimensional and periodic solutions are one
dimensional, strange attractors are more complex, and their dimension is a fraction. A chaotic system is a deterministic system
that exhibits random movement, and it is a nonlinear system that
exhibits extreme sensitivity in the state trajectory with respect
to the initial conditions. With two different initial states arbitrarily close to each other, the trajectories emanating from these

0885-8977/02$17.00 2002 IEEE

OZGUN AND ABUR: FLICKER STUDY USING A NOVEL ARC FURNACE MODEL

two points diverge from each other with a separation, on the average, increasing exponentially [9].
Researchers have long observed the occurrence of chaotic
components in arc furnace load currents. First, the recent U.S.
Bureau of Mines (USBM) work indicates that the electrical fluctuations in electric arc furnaces may be chaotic in nature. Then,
in another project at USBM, after analyzing the current and
voltage waveforms from an experimental electric arc furnace,
it was proven that the arc furnace electrical fluctuations are indicative of a chaotic system [5].
After recognizing the chaotic responses in electric arc furnaces, the first model based on chaotic dynamics is introduced
in [4]. This model employs chaos as a key element to produce
a chaotic response as close to actual data as possible. The basic
feature of this first chaotic model is that it is self-tuning to adjust
the model parameters in order to match the model output with
the artifact readings. The time-scaled Lorenz system is used to
represent the highly varying behavior of the currents in an electric arc furnace. The scaled Lorenz system is tuned to generate
a time series that matches the given real data as close as possible. In summary, at the end of the simulation, the arc current
data are generated for the selected time interval, and injected to
the arc furnace bus as an ideal current source. While this model
can successfully duplicate the artifact data obtained from a real
arc furnace, for most of the harmonic spectra, it is insensitive
to changes in the network. Also, it should be noted that it is
hard to get real data from each kind of arc furnace installation.
Therefore, it would be desirable to develop a model that can be
connected to any bus in the network as a circuit component, and
requires only conventional data available for arc furnaces.

1159

Fig. 1.

Dynamic v i characteristics of electric arc.

Fig. 2.

Typical voltage waveform of electric arc.

III. ARC FURNACE MODEL

The proposed arc furnace model will be developed in two
stages: dynamic, multivalued voltage-current characteristics of
the electric arc are derived in the first part, while a low-frequency chaotic signal is generated by the simulation of Chuas
well known chaotic circuit in order to obtain the arc furnace
voltage fluctuations in the second part.

Here , which stands for the arc radius, is chosen as the state
variable instead of the arc resistance or the conductance. The
arc voltage is then given by
(2)

A. Dynamic Behavior of Electric Arcs

The dynamic, multivalued characteristics of an arc furnace load are obtained by using a general dynamic arc model
in the form of a differential equation derived in [10]. There is
a fundamental difference between this approach and the others
where the electric arc is represented by its static characteristics or some empirical relations. The reason for using the
dynamic arc model is that the model will capture changes in the
characteristics as the operating conditions change in the
power system.
The differential equation that represents the general dynamics
of the arc model is based on the principle of energy conversion.
Therefore, starting from the power balance equation for the electric arc, the following differential equation is derived in [10]:
(1)

where is defined as the arc conductance and given by the following equation:
(3)
It is possible to represent the different stages of the arcing
process by simply modifying the parameters of and in (1).
The complete set of combinations of these parameters for different stages of electric arc can be found in [10]. Here, these
and
, which represent
parameters are chosen as
the refining stage of the electric arc.
Fig. 1 shows the dynamic - characteristics of 250-V, 70-kA
ac electric arc obtained by solving (1) and (2) in time domain.
The simulated characteristics of electric arc match well
with the measured characteristics [10]. The simulated arc
voltage waveform is shown in Fig. 2.

1160

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 17, NO. 4, OCTOBER 2002

Fig. 3. Matlab implementation of proposed arc furnace model.

B. Chaotic Time Variation

The chaotic component of the arc furnace voltage is supplied from the well known chaotic circuit of Chua [11], [12].
Chuas circuit is implemented in Matlab-Simulink environment
by using the new Power System Blockset from Matlab.
It has been shown that in order for an autonomous circuit consisting of resistors, capacitors, and inductors to exhibit chaos, it
has to contain the following components [11]:
1) at least one locally active resistor;
2) at least one nonlinear element;
3) at least three energy-storage elements.
Chuas circuit is the simplest type of circuit that satisfies the
conditions that are listed; moreover, it is the only physical
system for which the presence of chaos has been proven.
These two properties of this circuit motivated its use as a chaos
generator in this work. More detailed information about this
circuit can be found in [11], [12]. Chuas circuit is implemented
(see Fig. 3) by using the new Power System Blockset of Matlab.
One more useful property of this circuit is its ability to generate chaotic signal at any frequency by scaling the values of the
energy-storage elements.
C. Arc Furnace Model
So far, the electric arc voltage is obtained from the simultaneous solution of (1), and the low-frequency chaotic signal
that is generated by the simulation of Chuas chaotic circuit.
Modulation of these two signals produces the final arc furnace
voltage, which is the output of the proposed model. The current
absorbed from the power system bus is injected as the input to
the model. The model behaves as a controlled source, namely
it takes the system current as an input and assigns the terminal

voltage value at each time step. Fig. 3 shows the implementation

of the mentioned arc furnace model in the MatlabSimulink
environment. Therein, chaos generated in the upper portion of
the figure, while the simultaneous solution of (1) is performed
in the lower portion.
IV. IEC FLICKERMETER
Here, an IEC flickermeter, which has already been presented
in the IEC 1000-4-15 International Standard [7], is modeled.
The IEC flickermeter model is implemented digitally in Matlab
environment and basically consists of five blocks as shown in
Fig. 4.
The input to the flickermeter is the system voltage. Block 1
is simply a voltage-adapting circuit that scales the mean rootmean-square (RMS) value of the input voltage to an internal
reference level [13]. Hence, the flicker evaluations can be performed in percent (i.e., independent of the absolute input voltage
level).
Block 2 is basically a squaring demodulator which extracts
the voltage fluctuation. Namely, it separates the modulating
signal from the carrier signal, simulating the behavior of the
lamp.
Block 3 is composed of two filters: the first one filters out the
dc and residual ripple components of the demodulator output,
while the second filter simulates the frequency response to a
sinusoidal voltage fluctuation of a lamp and of the human eye.
Block 4 is composed of a squaring multiplier to represent the
nonlinear visual perception, and a first-order lowpass filter with
a time constant of 0.3 s to represent the built up effect in the
brain. The output of this block represents the immediate flicker
sensation.

OZGUN AND ABUR: FLICKER STUDY USING A NOVEL ARC FURNACE MODEL

1161

Fig. 4. IEC flickermeter implementation.

Finally, block 5 performs the online statistical analysis of the

instantaneous flicker level. Flicker severity level indices are calculated both in short term and long term in this block, and the
results are displayed. The output of this block is divided into
suitable subclasses (at least 64 classes) according to the instantaneous flicker level. First the probability distribution function
(PDF) is formed by accumulating the number of elements at
each level of the flicker, then the cumulative probability function (CPF) can be formed by integrating flicker distribution over
the flicker range. Then, the flicker severity over the short term,
, can be calculated by using the following expression [14]:

TABLE I
PERFORMANCE TESTING

(5)
where is the weighting coefficient, and is the flicker level
exceeded during a particular percent of the observation time.
These values are taken from the cumulative distribution curve
CPF

(6)

is a particular percent of the observation period. At

where
least five points of the CPF should be used while evaluating the
short-term flicker severity. All of the weighting coefficients and
the corresponding time percentages can be found in [7], [14].
In summary, the flickermeter model has two main parts [7]:
simulation of the response of the lamp-eye-brain chain;
online statistical analysis of the flicker signal, and displaying the results.
The first task is accomplished by Blocks 2, 3, and 4, while the
second task is performed in Block 5. More detailed information
about the IEC flickermeter can be found in [7], [13], [14].
Before performing any flicker analysis by using the flickermeter model, it should be verified against standard waveforms
provided in IEC standards [7]. The standard provides the amplitudes and frequencies of the six regular series of rectangular
should
voltage changes for which the flicker severity index
. The test signals and corresponding
values
be
that are obtained by running the flickermeter model are given in
Table I.

Note that in Table I, all

values generated by the developed flickermeter lie within the error tolerance range of IEC
standards.
V. SIMULATION RESULTS
The proposed arc furnace model is tested by using a sample
power system whose Simulink model is shown in Fig. 5. Arc
furnace transformer is connected to an equivalent representing
the power system behind. The point of common coupling
(PCC) corresponds to the primary of the arc furnace transformer. System data for this configuration can be found in the
Appendix.
Fig. 6 shows the furnace current waveform obtained at the
end of simulation. The voltage waveforms, at the secondary and
primary of the arc furnace transformer, can be seen in Fig. 7.
The flicker associated with the simulated voltage waveform
at the PCC, is evaluated by supplying it as an input to the flickermeter. Fig. 8 shows the CPF of the signal permanence in classes
one to 4096. Then the short-term flicker severity index ( ) is
is found as 1.377, a value
calculated by using this function.
exceeding 1.0 and indicating that customer complaints are likely
to occur.

1162

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 17, NO. 4, OCTOBER 2002

Fig. 5. Arc furnace connection to the sample power system.

Fig. 6.

Arc current obtained by simulation.

Fig. 7. Voltage at the secondary and primary of the arc furnace transformer.

Since the production of harmonics is also a major problem

that arc furnaces introduce to the power systems, the proposed

Fig. 8.

CPF of signal permanence in classes one to 4096.

Fig. 9.

Harmonic analysis of simulated voltage waveform.

model should also be tested in terms of the harmonic spectrum.

Although the previous models are able to capture the odd harmonics, they have difficulty in capturing the even ones. Fig. 9
shows a snap shot of the harmonic spectrum of the simulated
voltage waveform. The magnitudes of the harmonics change
from cycle to cycle during the simulation. Note that even harmonics are captured by the proposed model as well.

OZGUN AND ABUR: FLICKER STUDY USING A NOVEL ARC FURNACE MODEL

VI. CONCLUSION
A new model to represent the arc furnace operation has been
introduced. The model provides an arc current with harmonic
spectrum that is consistent with the measured spectrum. It is
built in the time domain and can be readily connected at a specified bus as a circuit component, which takes the system current
as input and assigns the terminal voltage value at each time step
(i.e., it behaves like a controlled voltage source).
Dynamic and multivalued - characteristics are obtained by
solving the corresponding differential equation. In order to represent the flicker effect, a low-frequency chaotic signal is modulated with the arc voltage. The symmetry of the arc - characteristics are therefore destroyed, resulting in even harmonics
as well as odd ones.
To evaluate the flicker severity of the simulated data, the IEC
digital flickermeter model is created in Matlab environment.
Flicker evaluation of the simulated arc furnace voltage waveform is performed using this flickermeter model.
APPENDIX
Parameters of the studied system shown in Fig. 5 are shown
here.
Source: Ideal sinusoidal ac voltage source with
kV and zero phase shift.
amplitude
and inductance,
Zthevenin: resistance,
mH.
Transformer: Three windings linear single-phase transformer.
MVA.
Nominal power:
kV,
Winding 1 parameters:
,
.
V,
Winding 2 parameters:
,
.
Magnetization resistance and reactance:
.
Arc Furnace: (Parameters for corresponding differential
,
,
,
,
equation)
.
nF,
F,
H
(Chuas circuit)
mho.
with a series resistor,
ACKNOWLEDGMENT
The authors gratefully acknowledge the helpful discussions
with Dr. M. P. Kennedy of University College Dublin, Ireland,
and Dr. E. Acha of University of Glasgow, U.K. They also
thank J. Soward of Texas Utilities Co. and Dr. R. Iravani and

1163

M. Graovac of University of Toronto, Toronto, ON, Canada,

for providing data related to typical arc furnaces.
REFERENCES
[1] R. Collantes-Bellido and T. Gomez, Identification and modeling of a
three phase arc furnace for voltage disturbance systems, IEEE Trans.
Power Delivery, vol. 12, pp. 18121817, Oct. 1997.
[2] G. C. Montanari, M. Loggini, A. Cavallini, L. Pitti, and D. Zanielli, Arc
furnace model for the study of flicker compensation in electrical networks, IEEE Trans. Power Delivery, vol. 8, pp. 20262036, Oct. 1994.
[3] S. Varadan, E. B. Makram, and A. A. Girgis, A new time domain
voltage source model for an arc furnace using EMTP, IEEE Trans.
Power Delivery, vol. 11, pp. 16851690, July 1996.
[4] E. ONeill-Carrillo, G. Heydt, E. J. Kostelich, S. S. Venkata, and A.
Sundaram, Nonlinear deterministic modeling of highly varying loads,
IEEE Trans. Power Delivery, vol. 14, pp. 537542, Apr. 1999.
[5] P. E. King, T. L. Ochs, and A. D. Hartman, Chaotic responses in electric
arc furnaces, J. Appl. Physics, vol. 76, no. 4, pp. 20592065, Aug. 1994.
[6] O. Ozgun and A. Abur, Development of an arc furnace model for power
quality studies, Proc. IEEE-PES Summer Meeting 1999, vol. 1, pp.
507511, 1999.
[7] Flickermeter-Functional and Testing Specifications, CEI, IEC Publ.
61 000-4-15, 1997.
[8] H. Chiang, C. Liu, P. P. Varaiya, F. Wu, and M. G. Lauby, Chaos in a
simple power system, IEEE Trans. Power Syst., vol. 8, pp. 14071417,
Nov. 1993.
[9] K. Alligood, T. Sauer, and J. Yorke, Chaos: An Introduction to Dynamical Systems. New York: Springer-Verlag, 1997.
[10] E. Acha, A. Semlyen, and N. Rajakovic, A harmonic domain computational package for nonlinear problems and its application to electric
arcs, IEEE Trans. Power Delivery, vol. 5, pp. 13901397, July 1990.
[11] M. P. Kennedy, Three steps to chaos, Part 1:Evolution, IEEE Trans.
Circuit Syst. I, vol. 40, no. 10, pp. 640656, October 1993.
, Three steps to chaos, Part 2:A Chuas circuit primer, IEEE
[12]
Trans. Circuits Syst. I, vol. 40, pp. 657674, Oct. 1993.
[13] J. W. Smith and S. M. Halpin, Voltage flicker measurement, Part I:
Digital implementation of the IEC flickermeter, Harmon. Transients
Tech Notes, pp. 1932, July 1998.
[14] J. Jatskevich, O. Wasynczuk, and L. Conrad, A method of evaluating
flicker and flicker-reduction strategies in power systems, IEEE Trans.
Power Delivery, vol. 13, pp. 14811487, July 1998.

Omer Ozgun received the B.Sc. and M.Sc. degrees in electrical and electronics
engineering from Bogazici University, Istanbul, Turkey, in 1995 and 1997, respectively. He is currently pursuing the Ph.D. in the Department of Electrical
Engineering at Texas A&M University, College Station.
His current research involves modeling and simulation of events related to
power quality.

Ali Abur received the B.Sc. degree from METU, Turkey in 1979, and the M.Sc.
and Ph.D. degrees from The Ohio State University, Columbus, in 1981 and 1985,
respectively.
Currently, he is a Professor in the Department of Electrical Engineering at
Texas A&M University, College Station, where he has been since 1985. His
current research involves power system state estimation and metering design,
fault identification and location, and modeling and simulation of events related
to power quality.