New assessment and prediction for arc furnace flicker

Abstract

This investigation discussed the load characteristics of EAFs as determined by field measurement and found that the electrical characteristic of the EAF load doesn't totally conform to the Gauss distribution probability mode. This paper suggest that the scaling factor-1.8, arrived at herein, is adopted to replace a traditional scaling factor-3.6 when doing the transformation between DeltaV_max and DeltaV_10max. The survey results max demonstrate that variations of active power and reactive power of EAFs are strongly alike. Meanwhile, an DeltaV₁₀ estimate must account for the effect of active power variation. This paper proposes the maximum complex apparent power fluctuation method to overcome the disadvantages of the traditional method. This revised method reduces the differences between the estimated voltage flicker and the surveyed value of the actual EAF operation. Restated, MAPFM can react to the actual voltage fluctuation during the operation of EAF. Furthermore, surveyed results reveal that the revised method can yield more accurate DeltaV₁₀ estimates than traditional method

Full text

Page 1

1

Abstract--
characteristics of EAFs as determined by field measurement and
found that the electrical characteristic of the EAF load doesn’t
totally conform to the Gauss Distribution Probability Mode. This
paper suggest that the scaling factor-1.8, arrived at herein, is
adopted to replace a traditional scaling factor-3.6 when doing the
transformation between
max
V
∆
and
demonstrate that variations of active power and reactive power
of EAFs are strongly alike. Meanwhile, an
account for the effect of active power variation. This paper
proposes the maximum complex apparent power fluctuation
method to overcome the disadvantages of the traditional method.
This revised method reduces the differences between the
estimated voltage flicker and the surveyed value of the actual
EAF operation. Restated, MAPFM can react to the actual voltage
fluctuation during the operation of EAF. Furthermore, surveyed
results reveal that the revised method can yield more accurate
V
∆
estimates than traditional method.

Index Terms-- flicker estimate,
apparent power fluctuation method (MCAPFM)
This investigation discussed the load
max10
V
∆
. The survey results
10
V
∆
estimate must
10
10
V
∆
, maximum complex
I. INTRODUCTION
OLTAGE flicker associated with an EAF is evaluated in
two main ways around the world [1]. The first is flicker
meter, which is the IEC standard and has been established by
the UIE [2]. The other is
10
V
∆
meter, which is established by
the Japanese Technical Committee [3]. Moreover,
method presently used by the Taiwan Power Company (TPC),
and thus is applied herein. Voltage flicker causes sudden
flashes of luminosity in fluorescent lamps and electric lights,
and disturbances in other electrical equipment [4], [5].
Voltage flicker problems have long existed in many of the
distribution areas served by TPC, especially those that include
steel plants that operate arc furnaces [6]-[8].
The trend in recent years has been toward arc furnace
installations of greatly increased capacity. These furnaces
have a correspondingly strong impact on the quality of utility.
10
V
∆
is the

The work was supported in part by the National Science Council of R.O.C.
under grant NSC-93-2213-E-146-004.
J. L. Guan, and H. H. Chang are with the Hwa Hsia Institute of
Technology, Taipei, Taiwan, R.O.C. (e-mail: gjl4127@cc.hwh.edu.tw)
J. C. Gu is with the National Taiwan University of Science and
Technology, Taipei, Taiwan, R.O.C. (jcgu@ mouse.ee.ntust.edu. tw)
M. T. Yang is with the St. John’s & St. Mary’s Institute of Technology,
Taipei, Taiwan, R.O.C. (e-mail:mtyang@mail.sjsmit.edu.tw)

While EAF operates after installation, the utilities and steel
manufacturers must evaluate the effects of EAF on power
systems. Before the EAF is installed, the capacities of
facilities improved in terms of voltage flicker must be
estimated. Recently, several measuring techniques and EAF
models have been developed to evaluate voltage flicker [9],
[10]. The EAF loads of more than ten steel factories have been
extensively surveyed over the last several years. This study
investigates ac and dc EAFs by making field measurements,
with those results indicate that the estimated
obtained using the conventional method is significantly lower
than the surveyed value [6]-[8]. The difference of
V
∆
between the estimated value and the actual value is quite
large.
The criteria presently used to estimate the severity of EAF
voltage flicker involve the maximum reactive power
fluctuation method (MRPFM) [3], [11]. The variation in the
load on EAF during steel manufacture is quite violent, so the
melting of scrap iron is clearly non-linear. In estimating the
severity of voltage flicker, part of line and load parameters are
often ignored since the true parameters hard to obtain.
However, steel factories still experience serious voltage
flicker problems even flicker improving facilities has been
installed, and thus this investigation probes into the
differences between the estimated
of actual EAF operations. Furthermore, an actual coefficient
of V
∆
/
10
V
∆
This investigation discusses the load characteristics of
EAFs as determined by field measurement. The survey results
demonstrate that significant variations in active power and
reactive power are very alike. Furthermore, the
cannot ignore the effect of active power variation [13].
Restated, the significant active power variation of EAF is an
important cause of voltage flicker. The traditional
compensator is completely unable to supply any fluctuating
real power drawn by the furnace. Real power fluctuation
produces phase variations at the critical bus that do in fact
contribute to flicker.
This study proposes the maximum complex apparent
power fluctuation method (MCAPFM) to overcome the
disadvantages of the traditional method. This revised method
reduces the differences between the estimated voltage flicker
and the surveyed value of the actual EAF operation. Restated,
MCAPFM can react to the actual voltage fluctuation during
the operation of EAF. Furthermore, surveyed results reveal
10
V
∆
value
10
10
V
∆
and the survey value
related to EAF operation is obtained [12].
10
V
∆
estimate
New Assessment and Prediction for
Arc Furnace Flicker
J. L. Guan, J. C. Gu, M. T. Yang, and H. H. Chang
V
1-4244-0493-2/06/$20.00 ©2006 IEEE.

Page 2

2
that the revised method can yield more accurate
estimates than traditional method
10
V
∆

II. V
investigation
∆
And
10
V
discusses
and
∆
OF VOLTAGE FLICKER
and
V
∆
, of six AC (namely A1,
This analyses load
characteristics, such as V
A2, A3, A4, A5 and A6) and three DC (namely D1, D2 and
D3) EAF by field surveying. The statistical method is used to
reveal the contents of the measurement data. The points of
field surveying are shown in Fig. 1

POINTS OF FIELD SURVEYING
∆
10
EAF
UTILITY
FILTERS
OTHER LOAD
M.TR.1
M.TR.2
FT
SERVICE
ENTRANCE
POINT
PRIMARY
SIDE

Fig. 1. Single-line diagram of the steel factory
Using the continuous model, the data are sampled during
several days. Furthermore, the frequency spectrums of
sampled voltage data are undertaken. To help clarify the load
changing conditions and operating characteristics of an EAF
load the sampling rate is at a rate of 12 samples/min..
Factories D1 and A1 were selected to plot the curve of line
voltage, V
∆
and
10
V
∆
for a complete steel-making period of
the EAF operation. The complete steel-making period mainly
included meltdown period and refining period. Fig. 2 shows
the line voltage curve of D1 factory around the primary side
of F.T.. When an EAF is not operating, that is the time for
feeding the raw metals into the furnace, the filters remain
connected which causes the reactive power to flow to the
service entrance point and voltage remain in 34.96kV. Fig. 3
presents the V
∆
and
10
V
∆
curves. The
0.073%~5.717% while the
0.107%~12.543%. Fig. 4 shows the line voltage curve of
factory A1 around the primary side of F.T.. Meanwhile, Fig. 5
illustrates the V
∆
and
10
V
∆
curves and shows that the
value is between 0.138%~6.767%, while the
between 0.315%~17.081%.

10
V
∆
value
value is between
is
V
∆
between
10
V
∆

V
∆
value is

Fig. 2. Field voltage measurement result of factory D1.

Fig. 3. Field V
∆
and
10
V
∆
measurement result of factory D1

Fig. 4. Field voltage measurement result of factory A1

Fig. 5. Field V
The survey results indicate that the variation of the EAF’s
voltage is unusually severe. The ten minutes voltage of the
meltdown period from the first section of previous surveying
is now taken, and the probability density spectrum is obtained
via the probability density function (PDF) operation. For
example, Fig. 6 shows a typical voltage probability density
spectrum of factory D1 and A1 respectively during the
meltdown period. Apparently, the normal probability
distribution of the voltage is asymmetrically spread. Owing to
intermittent electric arcs during the steel-making period of an
AC EAF’s operation, the appearance probability of unusually
high voltage appears comparatively higher than unusually low
voltage. It is quite obvious that the spread characteristic of the
EAF voltage is clearly not normal distribution.
∆
and
10
V
∆
measurement result of factory A1

Fig. 6. Typical voltage probability density distribution of D1 and A1

Totally, there are fifteen minutes of voltage data during the
meltdown period that are sampled and used to calculate the
frequency spectrum of V
∆
. Fig. 7 and Fig. 8 show typical
frequency spectrums of V
∆
near one DC EAF and one AC
EAF primary side of F.T. respectively . The bold line indicates
the average value while the plain line displays the sum of the
average value and two standard deviations (σ ). Notably, the
variation of the frequency spectrum is especially severe during
the meltdown period. Clearly, the standard deviation of each
frequency is roughly the same size as its average value,
confirming that the frequency spectrum spread of the EAF is
extremely random.

Fig. 7. Typical frequency spectrum of V
∆
of the DC EAF

Fig. 8. Typical frequency spectrum of V
∆ of the AC EAF

Page 3

3
Besides, four seconds of sampled data for different stages in
meltdown period on F.T. of factories D1 and A1 are selected.
The frequency spectrum curves are plotted in Figs. 9 and 10,
which resemble Figs. 7 and 8, except that the variation is
comparatively high. From the previous analysis, we already
know that the voltage fluctuation (
decreases with frequency, and the standard deviation of each
frequency is extremely high. Careful observation of Figs. 3
and 5 reveals that V
∆
is significantly higher than
the moment of electrodes enter or leave the furnace, and also
when the electric arc is suddenly interrupted. Fig. 11
illustrates the frequency spectrum of V
undergoes a sudden and significant change. Obviously, part of
the low frequency fluctuation is relatively high. Meanwhile,
part of the comparatively low frequency fluctuation do not
respond completely on the
V
∆
coefficient of the frequency is quite small. This is the reason
why the ratio of V
∆
to
10
V
∆
is abnormally high.
n
V
∆
) of an EAF gradually
10
V
∆
when
∆
when an EAF load
10
since the related sensitivity

Fig. 9. Field frequency spectrum of V
stage of meltdown period; pale line: ending stage of meltdown period)
∆
of the factory D1 (black line: initial

Fig. 10. Field frequency spectrum of V
stage of meltdown period; pale line: ending stage of meltdown period)
∆
of the factory A1 (black line: initial

Fig. 11. Field frequency spectrum of V
line: electric arc interrupted suddenly on factory A1; pale line: moment of
electrodes plug-in on factory D1)
∆
of the factories A1 and D1 (black
III. Analysis Of The Scaling Factor Of V
∆
To
10
V
∆

Although the voltage fluctuation of the EAF does not
conform to the normal distribution characteristics, as with the
analysis of large numbers of sampled data, a generalized and
conformable solution can be sought using statistical methods.
This investigation conducts a statistical analysis to utilize
survey data. We assume that the maximal voltage fluctuation
V
∆
within a range of
σ
3
±
. This value is compared with a
real survey value of
max
V
∆
from formula (1). The
value is obtained based on the maximum
survey. However, the process of calculating is based on
formula (2) and a reasonable scaling factor of V
sought.

∑
=
n
max
max10
V
∆

10
V
∆
found in the
∆
to
10
V
∆
is
n
VV
2)(ΔΔ
(1)

∑
n
∆×=∆
2
n
n
)V(aV10
(2)
To make it more clear, only the data in meltdown and
refining period excluding feeding time of the EAF operating,
as Figs. 2~5, are chosen for conducting analysis, and the
statistical results are listed in Table I.
Obviously, the σ value in Table I, column 4 is relatively
small compared with the 3rd column in Table II, particularly
for the DC EAF. Then, the K factor of factory D1 ranges
between 1.25~1.60 and the K factor of factory A1 ranges
between 1.43~1.73. However, it differs significantly from the
Japanese empirical value 3.6.
Generally, utilities must examine the possible impact of
voltage flicker to the power systems before EAF installation.
Conventionally, the
max10
V
∆
value is calculated in the way of
6 . 3/
max
V
∆
[3]. This approach has been applied for many
years. The empirical coefficient 3.6 was suggested by CRIEPI
in 1978 from the statistical survey of three small EAFs in
Japan. However, times have changed. The capacities and
operating technology of EAFs have been significantly
improved. It is hereby recognized that the estimating
philosophy of the flicker may need to be modified. On the
other hand, the spread characteristic of the EAF voltage
doesn’t completely correspond with the Gaussian distribution
Probability Mode as seen in Fig. 6. Consequently, an error is
produced when the maximal voltage fluctuation (
calculated directly by using
σ
3
±
7, the statistical results tell us this truth.
Supposing that the voltage fluctuation ( V
using formula (1). From another perspective, the ratio of
/ VV ∆∆
can be estimated. Furthermore, a statistical analysis
was performed using survey data of Figs. 2~5. Table II lists
the statistical results and reveals that the µ value ranged
between 1.703~1.998 (the 3rd column), while the total average
value was 1.789. Comparison of the µ value in Table II with
the K value in Table I clearly reveals that both are very
consistent.
max
V
∆
) is
( σ
6
). From Table I, column
∆
) is obtained
10
TABLE I
Statistical results of the survey data at a rate of 12 samples/min. measured on
the primary side of F.T. of factories D1 and A1.
∆
calculate survey calculate
V

(kV)
(kV)
σ
6

First Section
Boredown
Second Section
Boredown
Boredown
Last Section
First Section
Boredown
Second Section
Boredown
A1
Refining
Period
V
(%)
10
V
∆
(%)
survey
∆
K value
6
V
∆
Factory/Period
mean
σ
max
V
∆

6 . 3/6σ
max10
V
max10
σ
30.9090.3727.225 12.543
2.007 5.275 1.370
31.0890.3707.140 10.148
1.983 5.717 1.249
D1
31.0560.1893.661 4.145
1.017 2.291 1.598
21.7120.42511.736 17.081
3.260 6.767 1.734
21.7110.3489.628 13.185
2.675 6.723 1.432
21.8150.1865.125 6.791
1.424 3.440 1.490
TABLE II
/ VV ∆∆
The statistical results of
10
of factories D1 and A1
σ
10
/ VV ∆∆
Factory / Period
μ
μmax
μmin

Page 4

4
First Section Boredown 1.837 0.449 3.998 1.355
Second Section Boredown 1.723 0.286 3.262 1.354


D1
Boredown Last Section 1.720 0.130 2.057 1.454
First Section Boredown 1.754 0.324 3.576 1.105
Second Section Boredown 1.703 0.308 2.900 1.089


A1
Refining Period 1.998 0.247 2.903 1.641
IV. ANALYSIS OF EAF LOAD
The sampling rate is adjusted to one sample/cycle and a
total of 3600 samples per minute to help clarify the load
changing conditions and operating characteristics of an EAF
load. Factories D1 and A1 were selected as examples. Figures
8 and 9 show the survey results concerning the primary side of
furnace transformer (FT).
Curves of line voltage, active power and reactive power are
plotted and data are sampled during the initial stage of
meltdown, during which voltage flicker is particularly evident.
Figures 12 and 13 clearly show extreme variations in EAF
load.




Fig. 12. Fields V, P and Q measured results based on a rate of 1sample/cycle
on the primary side of FT of factory D1



Fig. 13. Fields V, P and Q measured results based on a rate of 1sample/cycle
on the primary side of FT of factory A1

Figure 12 presents the survey results concerning factory D1,
and reveals that the active power is between 4MW and 57MW
while the reactive power is between 19MVAR and 72MVAR.
The survey results also show that load control of dc EAF is
achieved using a constant current. Apparent power is
maintained almost constant, while active and reactive power
remains complementary. The ranges of variation of P and Q
are almost consistent, so P
substantially from that used in formula (3). Thus, the estimate
of
10
V
∆
for dc EAF must consider not only Q
∆
can increase
V
∆
quite
∆
but also P
∆ .
SXQPXV
SS
∆×=∆+∆×=∆
22
(3)
Figure 13 shows the survey results of factory A1, revealing
that the active power is between 0 and 28MW while the
reactive power is between 0 and 35MVAR. Clearly, the
estimate of
10
V
∆
for ac EAF must not ignore the effect of
active power variation. In particular, P
voltage flicker when the reactive power has been compensated.
Therefore, this study proposes the MCAPFM of obtaining
the
10
V
∆
value of EAF to overcome the disadvantages of the
traditional method. The method not only considers the reactive
power variation but also the active power variation in
calculating the estimated
10
V
∆
value of dc and ac EAFs. Fig.
14 and 15 show the load curve of dc and ac EAF.
∆
is major cause of

Fig. 4. Load curves of dc EAF

Fig. 3. Load curve of ac EAF
V. COMPARISON OF ESTIMATED AND SURVEYED
∆V VALUES
10
According to the previous analysis, the original estimates
and survey data of
10
V
∆
values of the EAF factories differ

Page 5

5
significantly, the relationship between voltage fluctuation and
active power is quite close when EAF factories typically
install reactive power compensation equipment to suppress the
high voltage flicker. Besides, the authors’ investigation found
that the electrical characteristic of the EAF load does not
wholly match the Gaussian Distribution Probability Mode.
Therefore, the
10
V
∆
value can be evaluated using the
V
∆
/1.8 [12]. Meanwhile, surveyed results reveal that this
way can more accurately estimate
analyzed and studied as follows.
max
10
V
∆
. Finally,
max
V
∆
can be
A. The ac EAF
0max
1 XQ
=

r
CosQQ
θ
2
max
'
max
×≅∆
(Uncompensated) (4)
%) 1 (
'
max max
α−× ≅∆
QQ
(Compensated) (5)
rrCos
θ
SinQP
θ
×≅∆
maxmax
(6)
2
max
2
maxmax
QPS
∆+∆=∆
(7)
scs
MVASSXV
maxmaxmax
∆=∆×=∆
(8)
8 . 1 /
maxmax10
VV
∆=∆
(9)
B. The dc EAF
)(
min
'
max
α
θθβ
SinSinSQ
sFT
−×≅∆
(10)

%)1 (
'
maxmax
α−×≅∆
QQ
(Compensated) (11)
)(
minmaxsFT
CosCosSP
θβθα
×−≅∆
(12)
2
max
2
maxmax
QPS
∆+∆=∆
(13)
scs
MVASSXV
maxmaxmax
∆=∆×=∆
(14)
8 . 1 /
maxmax10
VV
∆=∆
(15)
Factory A1 is selected as an example. The
estimated and the modified value is 1.552%.
max10
V
∆
is newly
MVASB
10
=
；
puXS
01473. 0
=
；
puX285588. 0
0=

pu202793Q. 2285588. 0. 0
2'
max
==∆
(Uncompensated)
pu85920Q
. 0)61. 01 (. 2
max
=−×=∆
(Compensated)
pu691793609P
. 1285588. 0. 0. 0
max
=×=∆

pu897859691S. 1. 0. 1
22
max
=+=∆

%. 2. 1
×
01473. 0
max
794897V
==∆

%. 18 . 1. 2
max10
552794V
==∆

Factory D1 is selected as an example. The
estimated and the modified value is 1.781%.
max10
V
∆
is newly
puXS
00437. 0
=
；
MVASFT
82
=
；
MVASB
10
=

()
puSinQ28. 336.870 . 1
×
182
max
=°−×=∆

()
puSinCosCosP56. 6)1.0(136.8782
1
max
=×−°×=∆
−

puS334. 728. 356. 6
22
max
=+=∆

%052 . 3334. 700437. 0
max
=×=∆V

%781. 18 . 1052 . 3
max10
==∆V

Similarly, the
D3 are also calculated. Table III compares the original design,
modification and cumulative probability value of the actual
field surveys of
10
∆V for the ac and dc EAF factories.
TABLE IV
ORIGINAL DESIGN, MODIFICATION AND CUMULATIVE PROBABILITY VALUE
OF THE ACTUAL FIELD SURVEY OF


∆Q
10max
∆V
(%)
(MVA)
(%)
10max
∆V
values for factories A2, A3, D2 and
10
V
∆
FOR THE EAF FACTORIES
Original
Calculation
Modified
Calculation
Actual Field Survey
(Cumulative Probability)
Factory
max
(MVAR
)
max
∆S

10max
∆V
95%10
∆V
−
(%)
97.5%10
(%)
∆V
−
99.9%10
∆V
−
(%)
A18.58 0.351 18.97 1.552 1.125
1.222
1.546
A27.41 0.162 19.18 0.837 0.665
0.725
0.929

ac
EAF
A3 22.05 0.150 35.90 0.488 0.438
0.476
0.590
D1 32.77 0.398 73.34 1.781 0.820
0.944
1.321
D2 41.86 0.302 69.98 1.003 0.529
0.616
1.006

dc
EAF
D3 38.71 0.350 87.89 1.592 0.948
1.077
1.505

However, a minor discrepancy exists between the modified
results and those of the actual field survey, probably because
of an abnormality in the operation of circuit elements or an
irregularity in the EAF during the steel-making process.
Besides, the assumed PF may be too small to reacting with
the actual ac EAF and the assumed
not match the actual operation of dc EAF. Briefly, the causes
stated above may lead to differences between the results after
modification and those of the actual field survey. In any case,
this study proposes the MAPFM that more accurately
∆V estimate than the conventional method.
αmin
θ
and
max
Q
also may
10
VI. CONCLUSIONS
The EAF loads of steel factories have been extensively
surveyed during the past several years. Meanwhile, these
investigations found that the estimated
the conventional means of estimating criteria is applied than
actual surveyed. The severity of the problem of voltage
flicker caused by EAF loads were under estimated, and so
some factories did not install any flicker compensation
equipment, while others failed to install sufficient
compensation equipment. Both utilities and factories are
confused by this mismatch between the theoretical estimate
10
V
∆
is lower when