Beruflich Dokumente
Kultur Dokumente
Abstract—Recent studies on the effect of arc furnaces in a power Cox and Mirbod in [7] have used a current-source model for
system lack accurate predicting voltage waveform of an arc fur- arc. The source current has been represented by the Fourier se-
nace. This is mainly due to the random nature of the arc length. In ries where the Fourier coefficients change randomly in each pe-
this paper, a novel two-step optimization technique is presented to
identify the arc furnace parameters considering the stochastic na- riod. It takes into account the stochastic behavior of the arc fur-
ture of the arc length. The proposed method is based on a genetic nace in a correct way. However, it is more suited to design filter
algorithm (GA) which adopts the arc current and voltage wave- components [8].
forms to estimate parameters of the nonlinear time-varying model An alternative method to model the arc furnace is based on
of an electric arc furnace. Simulation results are compared with chaos theory [9]. The method has become a modeling issue, sub-
data obtained from two real arc furnace plants. Analyses show that
the proposed method is profitable to identify accurate values of the sequent to recognize the chaotic responses in electric arc fur-
arc furnace parameters which incorporate the stochastic nature of naces [10]. A model based on chaotic dynamics is introduced
the arc length. in [11]. The main disadvantage of the model is that it does not
Index Terms—Arc furnace, genetic algorithm (GA), nonlinear generally have an accepted precise mathematical definition [2].
time-varying arc model, parameter estimation, power quality Furthermore, it exhibits extreme sensitivity in the state trajec-
(PQ). tory with respect to the initial conditions [12].
A frequency-domain analysis method has been proposed in
[13] which represents the arc voltage and current by their har-
I. INTRODUCTION monic components. In this model, it is assumed that the arc fur-
nace draws the maximum power in the fundamental frequency,
which is not always true [2]. The model is simple. However, it
A RC FURNACES used for refining and melting metals
in the steel production procedure are some of the main
causes of power-quality (PQ) problems in electric power sys-
cannot represent the stochastic nature of the arc [14].
In [2] and [15], a power balance method is proposed. The
tems. As the popularity of the arc furnaces increases, PQ prob- model is based on the energy balance equation, which is a non-
lems will be more severe [1]. In order to take precautions to linear differential equation of the arc radius and the arc current.
minimize the adverse effects of arc furnaces, it is necessary to Since this method is based on the experimental formula, which
develop an accurate and easy-to-use electric arc model. How- varies in different furnaces, any change in load requires the ap-
ever, this has so far been quite a challenging task. The reason plication of different models in the simulation process [8].
was the complexity of the electric arc physical phenomena and A voltage-source model is used in [16], where voltage is con-
the randomness associated with the melting stage of arc furnace sidered as square waves with modulated amplitude. The new
operation [2]. amplitude of voltage is generated after every zero crossing of
In [3] and [4], a nonlinear resistance model has been pro- the arc current during the reinitiating process [8]. In [17], a non-
posed which uses numerical analysis to solve differential equa- linear time-varying voltage-source model has been presented. In
tions describing the arc furnace. However, this model does not this model, the arc voltage is defined as a nonlinear function of
consider the time-varying nature of the arc. A time-varying re- the arc length. The time variation of the arc length is modeled
sistance model has been recently presented in [5] which relates by using deterministic or stochastic laws. The deterministic law
the reference resistance of the arc to power consumed by it. The usually assumes a sinusoidal behavior for the arc length.
actual arc resistance can have sinusoidal or bandlimited white Among the literature reviewed, it can be seen that
noise variation around the reference value [6]. time-varying resistance and voltage models not only are
able to perfectly simulate the stochastic nature of the arc length,
Manuscript received July 26, 2009; revised November 24, 2009. First pub- but they also have some advantages to the other methods
lished July 12, 2010; current version published September 22, 2010. Paper no. mentioned before. The advantages are reported in [17].
TPWRD-00562-2009. Several papers have been published which aim to identify
The authors are with the Department of Electrical Engineering, Amirkabir
University of Technology, Tehran 15914, Iran (e-mail: s.m.mousavi@aut.ac.ir; parameters of the arc furnaces. A power balance method was
Hosseinian@aut.ac.ir; askarian@aut.ac.ir; nimamoaddabi@yahoo.com). adopted in [1], together with differential evolution algorithm, to
Color versions of one or more of the figures in this paper are available online estimate the arc furnace parameters. However, the main disad-
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TPWRD.2010.2044812
vantage of the proposed method in [1] is that the time variation
A. Objective Function
The subsection describes the objective function which is used
to estimate the parameters of the nonlinear time-varying arc fur-
nace model introduced in Section II. The main purpose of the Fig. 3. Stochastic nature of arc-length implementation using the band-pass
white noise signal.
procedure is to identify the set of arc furnace parameters be-
longing to the solution space that minimize the global error
between the measured and the estimated voltage
every 10 ms. This duration of time is chosen because of the fact
samples.
that although the arc length has a random value, its value will
In particular, the overall problem can be regarded as an un-
not change at least for a half cycle, as mentioned in the literature
constraint minimization of the estimation error as follows:
[13], [17].
The following proposed method, which is capable of identi-
(4) fying arc furnace parameters, is presented.
C. Proposed Method
Two different objective functions (OFs) have been used in a The main objective of this paper is to identify electric arc
similar optimization process in the literature [18]. According to furnace parameters. Among the parameters introduced in the
the previous works [2], [18], one of them has been reported to previous section, , and are deterministic variables;
be more appropriate than the other for this purpose. Here, this however, has a stochastic nature. The value of arc length
OF is used, which calculates the mean value for the rooted sum varies in subsequent half cycles and its variation is described
of squared errors, as follows: by a uniformly distributed random variable in the range which
is defined by the reference arc length , and maximum arc-
length deviation .
(5) Different optimization algorithms can be used to solve the
aforementioned parameter identification problem. However, the
where and are the estimated and the measured value number of iterations in most of these algorithms is limited. Due
of the arc furnace voltage related to the th arc current sample, to this and many other reasons (such as the heuristic nature of
respectively. is the number of samples per half cycle. modern optimization techniques and probability of trapping into
It is noted that a similar OF was used in [18]. However, the the local optimum), implementing these techniques for various
proposed algorithm in [18] did not consider the stochastic nature half cycles may lead to obtaining different values for the model
of the arc length. The following subsection describes how this parameters. However, as mentioned previously, among the pa-
important criterion is incorporated in arc furnace modeling. rameters, only has a stochastic nature (its value can be different
in various half cycles), and all of the other deterministic param-
B. Arc Furnace Modeling
eters should have a unique value in all of the half cycles of the
The arc furnace model is developed in the power system sim- observation window.
ulation tool, PSCAD/EMTDC, by using the PSCAD component The previously mentioned shortcomings in the existing opti-
builder. Internal to the component, FORTRAN code is written mization algorithms were the main motivation for the authors
to simulate the arc furnace as a voltage-dependent source, base to propose a two-step optimization technique. The proposed
on the model described in Section II. The stochastic nature of method divides the observation window into several half cycles,
arc length is incorporated into the model by generating a white each having the same duration of 10 ms. Then, the parameter
noise signal representing arc-length deviations. The idea is sup- identification procedure is performed based on the data coming
ported by findings reported in [13]and [17]. The white noise from the measured voltage waveform in each period of 10 ms,
signal is created by using a random number generator having separately. It does not mean that the procedure should be com-
a uniform distribution over the interval . is refer- pleted during 10 ms. Rather, the procedure may take some sec-
ence arc length and is maximum arc-length deviation. The onds to complete parameter identification in each half cycle. It
output of the random number generator is fed to a band-pass may also take several minutes for the whole half cycles of the
filter with lower and uppercut frequencies 4 Hz and 14 Hz, re- observation window. However, these are not constraining prob-
spectively (according to the frequency range reported in [20]). lems since the proposed method is based on offline calculations.
Fig. 3 depicts the stochastic variation of arc length that is simu- The steps of the proposed method are described as follows:
lated with a reference arc length of 30 cm and maximum devia- Step 1) First, an observation window is considered on the
tion of 10 cm. As can be seen from Fig. 3, the arc length varies arc furnace voltage waveform available from direct
2862 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 25, NO. 4, OCTOBER 2010
TABLE II Half-cycle intervals are arranged in such a way that the deriva-
TOLERANCE LIMITS FOR THE PARAMETERS OF THE ARC tive of the arc current is positive in odd half cycles, and reversely
FURNACE MODEL USED IN SIMULATION
its value is negative in even half cycles. In Table III, the dash sign
corresponds to parameters which cannot be estimated in some
half cycles. For example, the estimated value for is marked
by a dash in even half cycles. This is because the derivative of
arc current is negative in these half cycles. In the first column
of Table III, a list of arc furnace parameters is presented. In this
table, only some half cycles of the observation window (100 half
cycles) have been selected for more accurate investigation. The
estimated value of each parameter is shown in each half cycle
(columns 2 to 7). Based on the reported data in Table III, there
are some differences between the estimated values of the deter-
ministic variables in different half cycles.
TABLE III The average value of the estimated parameters in all half cy-
VARIATION OF THE ESTIMATED PARAMETERS OF THE ARC cles of the observation window is shown in the eighth column.
FURNACE MODEL USING THE PROPOSED METHOD
Fig. 5 shows how the values of the estimated parameters vary
during half cycles of the observation window. The average value
is depicted by bolted horizontal line in the figure.
As mentioned previously, the variation of the estimated
values for the deterministic parameters is related to the max-
imum generation numbers in GA. The data needed to investigate
the relationship between the generation size and the variances
of the deterministic variables are presented in Table IV. As is
shown in Table IV, variances of all parameters decrease more
or less, as the number of generations increases. Since the results
obtained with the maximum generation number of 100 are in
good agreement with the data presented in [17], it is assumed to
be the base case in the study. Fig. 6 shows the stochastic nature
of the arc length. It can be seen that the reference value of the
arc length is estimated at 30.35 cm and the maximum deviation
to 9.70 cm, which are in agreement with the data mentioned in
[17].
To summarize, Table V provides a comparison between the
results of the proposed method and the actual values of the pa-
subsequently to a trial-and-error procedure. rameters which are reported in [17]. The proposed method is
Population size: 30 performed 1000 times to check validity of the obtained results.
Maximum number of generations: 100 The fourth column in Table V presents the average value of the
Crossover probability: 0.5 relative error between the data obtained from proposed method
Mutation probability: 0.08. over 1000 runs, and actual values. For further investigation, the
For better functionality of the algorithm, the tolerance limit of maximum and minimum relative errors for various parameters
each parameter should be determined. It is noted that the upper are shown in columns 5 and 6 of Table V. From the reported
and lower tolerance limits for the parameters were determined data, it can be seen that the proposed method will result in a
based on data available in the literature [17], [19], and [20]. In maximum relative error of about 5% in the worst case (max-
some cases, when little data were available, the tolerance limit imum variation of arc length). As mentioned before, the accu-
was chosen heuristically large enough. Table II provides the tol- racy of the proposed method depends on the sampling rate. This
erance limits for the parameters used in this study. is because of the evolutionary nature of the algorithm which is
used in this paper. Assuming the observation window size of 100
A. Case I: Arc Furnace Plant Installed in Northern Italy cycles, different numbers of samples are considered. The sam-
pling rate changes between 2, 3, and 5 samples per half cycle.
The summary of results obtained in this case is depicted in The results show that the relative errors of the different param-
Table III. The simulation is carried out by using the data pre- eters decrease, as the sample rate increases. It is noted that in
sented in Appendix A. case of and , the relative error decreases more, with the
In this study, sample rate is selected to be five samples per half increase of sampling rate. It is due to the relatively large tol-
cycle. The observation window contains 100 half cycles. There- erance limit considered for these parameters. Investigating the
fore, step 1 of the proposed method is carried out for each half results shows that five samples per half cycle provide a compro-
cycle, resulting in 100 runs for the whole observation window. mise between the estimation accuracy and running time of the
2864 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 25, NO. 4, OCTOBER 2010
Fig. 5. Variation of deterministic parameters of arc furnace model in half cycles of the observation window: (a) C (in kikowatts).. (b) C (in kilowatts). (c) D
(in kiloamperes). (d) D (in kiloamperes). (e) A(V). (f) B(V/cm).
TABLE IV
VARIANCE OF ESTIMATED VALUES FOR DETERMINISTIC VARIABLES IN HALF
CYCLES OF THE OBSERVATION WINDOW
Fig. 6. Variation of the arc length estimated by using the proposed method.
TABLE V
COMPARISON BETWEEN THE ESTIMATED PARAMETERS OF THE ARC FURNACE
MODEL USING THE PROPOSED METHOD AND REAL DATA PRESENTED IN [17]
Fig. 8. Electric arc furnace voltage waveform over the entire observation
window obtained from the (a) measurement and (b) proposed method.
TABLE VI
IDENTIFIED PARAMETER VALUES FOR A REAL ARC FURNACE IN TABRIZ, IRAN
TABLE VII
FLICKER SEVERITY INDEX FOR THE SIMULATED
AND MEASURED VOLTAGE WAVEFORMS
electric arc furnace. The proposed method has been suggested [5] S. Vardan, E. B. Makram, and A. A. Girgis, “A new time domain
to be applied in modeling other electric systems in which the voltage source model for an arc furnace using EMTP,” IEEE Trans.
Power Del., vol. 11, no. 3, pp. 1685–1691, Jul. 1996.
deterministic and stochastic variables exist together. [6] H. M. Peterson, R. G. Koch, P. H. Swart, and R. Van Heerden, “Mod-
elling arc furnace flicker and investigating compensation techniques,”
in Proc. IEEE Ind. Appl. Soc. Annu. Meet., 1995, vol. 2, pp. 125–128.
APPENDIX [7] M. D. Cox and A. Mirbod, “A new static VAR compensator for an arc
furnace,” IEEE Trans. Power Syst., vol. PWRS-1, no. 3, pp. 110–120,
Aug. 1986.
[8] I. Vervenne, K. Van Reusel, and R. Belmans, “Electric arc furnace mod-
A. Arc Furnace Plant Installed in Northern Italy elling from a power quality point of view,” presented at the 9th Int.
Conf. Electrical Power Quality and Utilization, Barcelona, Spain, Oct.
Parameters of an electric arc furnace plant installed in 9–11, 2007.
northern Italy are presented here. The parameters are used in [9] G. Carpinelli, F. Iacovone, A. Russo, and P. Varilone, “Chaos-based
the first case study in this paper. modelling of DC arc furnaces for power quality issues,” IEEE Trans.
Power Del., vol. 19, no. 4, pp. 1869–1876, Oct. 2004.
Source: An upstream network is modeled by using an ideal [10] P. E. King, T. L. Ochs, and A. D. Hartman, “Chaotic responses in elec-
sinusoidal ac voltage source with an amplitude of 220 kV and tric arc furnaces,” J. Appl. Phys., vol. 76, no. 4, pp. 2059–2065, Aug.
Thevenin inductance of . 1994.
[11] E. O’Neill-Carrillo, G. Heydt, E. J. Kostelich, S. S. Venkata, and
HV/MV transformer: Two windings linear single-phase A. Sundaram, “Nonlinear deterministic modeling of highly varying
transformer. Nominal power: 95 MVA, nominal voltage ratio: loads,” IEEE Trans. Power Del., vol. 14, no. 2, pp. 537–542, Apr.
220/21 kV, short-circuit voltage and losses 12.5% and 0.5%, 1999.
[12] K. Alligood, T. Sauer, and J. Y. Chaos, An Introduction to Dynamical
respectively. Systems. New York: Springer-Verlag, 1997.
MV/LV transformer: Two windings linear single-phase [13] J. G. Mayordomo et al., “A frequency domain arc furnace model for
transformer. Nominal power: 60 MVA, nominal voltage ratio: harmonic power flows under balanced conditions,” in Proc. IEEE 7th
ICHQP, Las Vegas, NV, Oct. 1996, pp. 419–427.
21/0.6 kV, short-circuit voltage and losses 10% and 0.5%, [14] J. G. Mayordomo, L. F. Beites, R. Asensi, M. Izzeddine, L. Zabala,
respectively. and J. Amantegui, “A new frequency domain arc furnace model for
Flexible cable: resistance , inductance iterative harmonic analysis,” IEEE Trans. Power Del., vol. 12, no. 4,
pp. 1771–1778, Oct. 1997.
. [15] Task Force on Harmonics Modeling and Simulation, “Modelling
devices with nonlinear voltage-current characteristics for harmonic
studies,” IEEE Trans. Power Del., vol. 19, no. 4, pp. 1802–1811, Oct.
B. Arc Furnace Plant in Tabriz, Iran 2004.
[16] D. Stade, H. Schan, and S. Kramer, “Modelling of the electrical be-
Parameters of a real electric arc furnace plant installed in havior of arc furnace,” in Proc. 29th Universities Power Eng. Conf.,
Tabriz, Iran are presented here. The parameters are used in the 1994, vol. 1, pp. 125–128.
[17] G. C. Montanari, M. Loggini, A. Cavallini, L. Pitti, and D. Zaninelli,
second case study in this paper. “Arc furnace model for the study of flicker compensation in electrical
Source: Upstream medium-voltage (MV) network is modeled networks,” IEEE Trans. Power Del., vol. 9, no. 4, pp. 2026–2036, Oct.
as an ideal source with an amplitude of 20 kV, which has the 1994.
[18] W. M. Al-hasawi and K. M. EI-Naggar, “A genetic based algorithm
following parameters: 0.0097 p.u., 0.0882 p.u., for voltage flicker measurement,” presented at the IEEE MELECON,
0.0124 p.u., p.u. Moreover, an existing local Cairo, Egypt, May 7–9, 2002.
generator is modeled by using the following parameters [19] J. Sousa, M. T. Correia de Barros, M. Covas, and A. Simoes, “Har-
monics and flicker analysis in arc furnace power systems,” presented at
0.005 p.u., 2.0446 p.u., 0.005 p.u., 0.200 p.u., the IPST’99-Int. Conf. Power Systems Transients, Budapest, Hungary,
0.457 s. Jun. 20–24, 1999.
MV cable: 21 m of XLPE cable, 3 (1 70) mm . [20] G. Manchur and C. C. Erven, “Development of a model for predicting
flicker from electric arc furnaces,” IEEE Trans. Power Del., vol. 7, no.
MV/LV transformer: two windings single-phase transformer. 1, pp. 416–426, Jan. 1992.
Nominal power: 4 MVA, nominal voltage ratio: 20/0.4 kV, [21] G. C. Montanari and M. Loggini, “Voltage-distortion compensation in
impedance 7.5%. electrical plants supplying static power converters,” IEEE Trans. Ind.
Appl., vol. 23, no. 1, pp. 181–188, Feb. 1987.
Flexible cable: 46 m of PVC cable, 3 (3 185) mm . [22] M. Loggini, G. C. Montanari, L. Pitti, E. Tironi, and D. Zanmelli, “The
effect of series inductors for flicker reduction in electric power systems
supplying arc furnaces,” presented at the IEEE/Ind. Appl. Soc. Annu.
REFERENCES Meeting, Toronto, ON, Canada, Oct. 1993.
[1] Y. Wang, Z. Mao, Y. Li, H. Tian, and L. Feng, “Modeling and param- S. M. Mousavi Agah (M’10) was born in Rasht,
eter identification of an electric arc for the arc furnace,” presented at Iran, on December 4, 1984. He received the B.Sc.
the IEEE Int. Conf. Automation and Logistics, Qingdao, China, Sep. and M.Sc. degrees (Hons.) in electrical engineering
2008. from Amirkabir University of Technology (AUT),
[2] O. Ozgun and A. Abur, “Flicker study using a novel arc furnace model,” Tehran, Iran, in 2006 and 2008, respectively, and
IEEE Trans. Power Del., vol. 17, no. 4, pp. 1158–1163, Oct. 2002. is currently pursuing the Ph.D. degree in electrical
[3] E. A. C. Plata and H. E. Tacca, “Arc furnace modelling in ATP-EMTP,” engineering at AUT.
presented at the Int. Conf. Power Systems Transients, Montréal, QC, His areas of research are power quality, power
Canada, 2005. system protection, distributed generation systems,
[4] T. Zheng, E. B. Makram, and A. A. Girgis, “Effect of different arc fur- and probabilistic analysis of power systems. He
nace models on voltage distortion,” in Proc. IEEE 8th ICHQP, Athens, has a lot of experience in protection, control, and
Greece, Oct. 1998, pp. 1079–1085. automation of high-voltage substations in Iran.
MOUSAVI AGAH et al.: PARAMETER IDENTIFICATION OF ARC FURNACE BASED ON STOCHASTIC NATURE OF ARC LENGTH 2867
S. H. Hosseinian was born in Iran in 1961. He re- N. Moaddabi was born in Rasht, Iran, in 1984. He
ceived the B.Sc. and M.Sc. degrees in electrical en- received the B.S. and M.Sc. degrees in electrical
gineering from Amirkabir University of Technology power engineering from Amirkabir University of
(AUT), Tehran, Iran, in 1985 and 1988, respectively, Technology (AUT), Tehran, Iran, in 2006 and 2008,
and the Ph.D. degree in electrical engineering from respectively, where he is currently pursuing the
the University of Newcastle, U.K., in 1995. Ph.D. degree in power electrical engineering.
Currently, he is Assistant Professor in the Elec- His research interests include power quality, power
trical Engineering Department at AUT. His special system protection, power electronics, and distributed
fields of interest include transients in power systems, generation systems. He has a lot of experience in pro-
power quality, restructuring, and deregulation in tection and control design of high-voltage substations
power systems. He is the author of four books in the in Iran.
field of power systems. He is also the author and coauthor of many technical
papers.