You are on page 1of 11

2200 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO.

3, JULY 2005

Lightning Performance Analysis of Overhead


Transmission Lines Using the EMTP
Juan A. Martinez, Member, IEEE, and Ferley Castro-Aranda

Abstract—This paper presents a procedure for the calculation that could detect those parameters for which more accurate in-
of lightning flashover rates of transmission lines using a Monte formation is required and the range of values that can be of con-
Carlo method. The procedure has been implemented in the Alter- cern for each parameter; see Section IX of this paper.
native Transients Program version of the Electromagnetic Tran-
sients Program. Parametric studies using this procedure can also The main contribution of this paper is the Alternative Tran-
be performed to determine the sensitivity of the flashover rate with sients Program (ATP) implementation of a new Monte Carlo
respect to some parameters of the transmission line and the return procedure for calculation of lightning flashover rates of over-
stroke. Some refinements are proposed to decrease the computer head transmission lines. The ATP is a well-known member of
time while preserving the accuracy of calculations. the Electromagnetic Transients Program (EMTP) family; there-
Index Terms—Modeling, Monte Carlo method, overvoltages, fore, its main solution algorithms are common to most electro-
power system lightning effects, sensitivity, statistics. magnetic transients programs [1]. For a summary of ATP capa-
bilities that can be useful for the present work, see [2].
I. INTRODUCTION The paper is organized as follows. Section II includes a sum-
mary on modeling guidelines for representing transmission lines
HE lightning performance of an overhead line can be
T measured by the flashover rate, usually expressed as the
number of flashovers per 100 km and year. Due to the random
in lightning overvoltage calculations. Two critical aspects when
calculating lightning overvoltages in transmission lines are the
representation of footing impedances and lightning strokes. An
nature of lightning, calculations must be based on a statistical analysis of the footing impedance and an introduction to the
approach. A Monte Carlo simulation is a very common method characteristic parameters of the return stroke are provided in
for this purpose. Sections III and IV, respectively. Section V gives a summary
Transmission lines are usually shielded by one or several of methods developed to date for the calculation of lightning
wires; therefore, lightning failures can be caused by strokes to flashover rates in overhead transmission lines. A description of
either a shield wire or a phase conductor, since overvoltages the Monte Carlo procedure and its implementation in ATP is de-
induced by strokes to ground can be neglected. Shielding tailed in Section VI. The application of the procedure to a test
failures cannot be totally prevented, but the number of strokes line is presented in Section VII. Some refinements of the proce-
to phase conductors is usually very low. The flashover rate of dure are analyzed in Section VIII. A parametric study of the test
a transmission line is therefore divided into the backflashover line aimed at determining the relationship of the flashover rate
rate (BFOR) and the shielding failure flashover rate (SFFOR). with respect to some parameters of the line and some variables
To obtain both quantities, an incidence model is required to of the return stroke are detailed in Section IX.
discriminate strokes to shield wires from those to phase con-
ductors and those to ground. II. MODELING FOR LIGHTNING OVERVOLTAGE CALCULATIONS
A Monte Carlo procedure for calculation of lightning
flashover rates can consist of the following steps: generation Several documents have been published to provide modeling
of random numbers to obtain those parameters of the lightning guidelines of power components in lightning overvoltage sim-
stroke and the overhead line of random nature; application ulations [3]–[8]. The following paragraphs summarize models
of an incidence model to deduce the point of impact of every and guidelines to be considered for overhead transmission lines.
lightning stroke; calculation of the overvoltage generated by 1) A transmission line is modeled by two or three spans at
each stroke, depending on the point of impact; and calculation each side of the point of impact. Each span is represented
of the flashover rate. by a multiphase untransposed distributed parameter line
Some of these steps are usually carried out with incomplete section. This representation can be made by using either
information or models of limited accuracy. For instance, the a frequency-dependent or a constant parameter model.
knowledge of the lightning parameters is usually incomplete, or If the second option is chosen, then it is recommended
the incidence model is not accurate enough. Some limitations to calculate parameters at a frequency between 400 and
can be partially overcome by performing a sensitivity analysis 500 kHz [7]. More accurate results are derived when the
corona effect is included in the model.
Manuscript received September 11, 2003; revised January 7, 2004. Paper no. 2) The representation of a line termination is needed at each
TPWRD-00467-2003. side of the above model to avoid reflections that could
The authors are with the Departament d’Enginyeria Elèctrica, Univer- affect the simulated overvoltages around the point of im-
sitat Politècnica de Catalunya, Barcelona 08028, Spain (e-mail: martinez@
ee.upc.edu). pact. This can be achieved by adding a long enough sec-
Digital Object Identifier 10.1109/TPWRD.2005.848454 tion at each side of the line, or by inserting a resistance
0885-8977/$20.00 © 2005 IEEE
MARTINEZ AND CASTRO-ARANDA: LIGHTNING PERFORMANCE ANALYSIS OF OVERHEAD TRANSMISSION LINES USING THE EMTP 2201

matrix at each termination whose values equal the line


surge impedances.
3) Several models with a different level of complexity have
been proposed for representing towers [9]–[13]. The
simplest one is based on a single conductor distributed
parameter line. While a detailed and sophisticated
model can be necessary when analyzing ultra-high
voltage (UHV) lines, a very simple representation can
suffice for single circuits with towers shorter than 30
m [14]. The value of the tower surge impedance ranges
from 100 to 300 [7].
4) Footing impedance modeling is one of the most critical
aspects. A nonlinear frequency-dependent representation
is required to obtain an accurate simulation [15]–[18].
However, the information needed to derive such a model
is not always available. As a consequence, a lumped non-
linear resistance is usually chosen for representing the
footing impedance, although it cannot be always ade-
quate. More details on this aspect are provided in Sec-
tion III.
5) Phase voltages at the instant at which the lightning stroke
impacts the line must be included. For a deterministic
calculation, worst-case conditions should be determined
and used. For statistical calculations, phase voltage mag-
nitudes are deduced by randomly determining the phase
voltage reference angle and considering a uniform distri-
bution between 0 and 360 .
6) A lightning stroke is represented as a current source
whose polarity can be positive or negative. The parame-
ters of the stroke, as well as its polarity, can be randomly
determined according to the distribution density func-
tions recommended in the literature [5], [14], [19]–[21].
See Section IV for more details.
Fig. 1. Variation of the footing resistance. (a) Current through resistance = 34
7) Several approaches have been developed for repre- kA. (b) Soil resistivity  = 500
1m.
sentation of insulators. They are based on a simple
voltage-dependent flashover switch with a random be- where is the footing resistance at low current and low fre-
havior, on a voltage–time characteristic [22], or on the quency, is the limiting current to initiate sufficient soil ion-
leader progression model, see, for instance, [23]. In sta- ization, and is the stroke current through the resistance.
tistical calculations, insulators can also be represented The limiting current is given by
as open switches: lightning overvoltages are obtained
across their terminals; after the Monte Carlo method (2)
convergence is achieved, the overvoltage probability where is the soil resistivity (in m) and is the soil ion-
density function is determined and the risk of failure is ization gradient, whose value is between 300 and 400 kV/m [7],
calculated by comparing this function to the cumulative [16].
probability density function of the insulator strength Fig. 1 shows the variation of as a function of , and .
[24]. The nonlinear behavior of the footing resistance and its strong
dependency is evident from these plots with respect to the soil
III. FOOTING IMPEDANCE resistivity and the lightning current. One can conclude that the
resistance value is greater for small lightning currents, and its
The footing impedances of line towers have a significant ef- variation with respect to is only significant for large values
fect on the peak overvoltages caused by strokes to shield wires. of the soil resistivity.
An accurate representation of this impedance is not easy since
its behavior is nonlinear and frequency dependent. IV. LIGHTNING STROKE PARAMETERS
In this work, the footing impedance is approximated by a non-
A. Return-Stroke Waveform
linear resistance given by [7], [24]
Both the double-exponential and the triangular waveforms
(1) have been frequently used to represent lightning return stroke
currents. Presently, it is assumed that a concave waveform of
2202 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005

factor . Although the three waveforms have the same rise and
tail times, the time intervals between the start of the wave and
the crest are different.

B. Probability Distribution of Return-Stroke Parameters


The statistical variation of the lightning stroke parameters
can be approximated by a log-normal distribution, with the fol-
lowing probability density function [21]:

(4)

where is the standard deviation of , and is the me-


dian value of .
The joint probability density function of two stroke parame-
ters can be expressed as follows:
Fig. 2. Parameters of a return stroke–concave waveform.

(5a)

where

(5b)

(5c)

(5d)

and is the coefficient of correlation.


If and are independently distributed, then and
.
The conditional probability density function of for a given
can be found by a change of variables [21]

(6a)
Fig. 3. Heidler model. Effect of factor n (t = 1:2 s, t = 50 s).
where

the first stroke is a better representation since it does not show (6b)
a discontinuity at .
(6c)
Several expressions have been proposed for such a waveform.
One of the most widely used is the so-called Heidler model. It This new function has a median value , which is the antilog
is given by of and a standard deviation given by (6c).

V. PROCEDURES FOR CALCULATION OF LIGHTNING FLASHOVER


(3)
RATES IN TRANSMISSION LINES

where is the peak current, is a correction factor of the peak Procedures for the calculation of lightning flashovers can be
current, is the current steepness factor, and and split into two main groups.
are time constants determining current rise and decay • Methods based on simplified models and approximated
time, respectively [25]. calculations of lightning overvoltages; they are derived
Fig. 2 depicts the waveform of a concave return stroke. The from guidelines proposed by IEEE [3]–[4], [27], IEC [24],
main parameters used to define this waveform in the present and CIGRE [5].
work are the peak current magnitude , the rise time • Methods based on more rigorous models and calculations;
, and the tail time (i.e., the time interval be- they usually rely on results derived from EMTP-like tools.
tween the start of the wave and the 50% of peak current on tail). The main differences between procedures are related to the
The main difficulty to synthesize a concave waveform is the following issues:
determination of the parameters to be specified in (3) from those • lightning stroke waveform and parameters;
of the return stroke [26]. Fig. 3 shows the effect of • lightning incidence models;
MARTINEZ AND CASTRO-ARANDA: LIGHTNING PERFORMANCE ANALYSIS OF OVERHEAD TRANSMISSION LINES USING THE EMTP 2203

• modeling guidelines and parameter calculation for over-


head lines, towers, footing impedance, and insulator
strength;
• overvoltage calculations;
• calculation of flashover rates, namely, SFFOR and BFOR.
A tradeoff usually exists between computation time and ac-
curacy. Methods of the first group provide a faster computation
but with less accurate results. A well-known tool of this group is
the FLASH program [28]. It was originally based on Anderson’s
work [29] and adapted by the IEEE. This tool is periodically re-
vised and updated.
On the other hand, several approaches have been developed,
taking advantage of EMTP capabilities.
• References [30] and [31] present the application of a
method based on structural reliability theory, whose main
goal is to optimize the number of EMTP simulations
needed to obtain the risk of failure caused by lightning.
• A pioneering work based on the Monte Carlo method and Fig. 4. Diagram of tasks implemented in ATP.
ATP capabilities was presented in [32]; this procedure was
developed only for calculation of the BFOR without in- the rate is determined using a stand-alone environment,
cluding any incidence model. and simulations are halted before the maximum simula-
tion time in case of flashover.
VI. MONTE CARLO PROCEDURE FOR CALCULATION 5) The convergence of the Monte Carlo method is checked
OF LIGHTNING FLASHOVER RATES by comparing the probability density function of all
A. General Procedure random variables to their theoretical functions; the pro-
cedure is stopped when they match within the specified
The following paragraphs describe some important aspects of
error.
the procedure.
1) The calculation of random values includes the parame- B. ATP Implementation
ters of the lightning stroke (peak current, rise time, tail
ATP capabilities were used to develop the procedure as listed
time, and location of the vertical channel), phase con-
below; see Fig. 4.
ductor voltages, the footing resistance, and the insulator
strength. • A multiple-run option is used to perform all of the runs
2) The last step of a return stroke is determined by means of required by the Monte Carlo method.
the electrogeometric model as used by the IEEE Working • A compiled routine has been developed and linked to a
Group [27]; therefore, striking distances to shield wires MODELS section to obtain the values of the random pa-
and phase conductors are assumed equal, while the rameters that must be generated at every run according to
striking distance to ground is smaller. Since a three-di- the probability distribution function assumed for each one.
mensional (3-D) model has been used in this work to • Phase-to-ground voltages across insulators are continu-
define the attractive area, the heights to be specified in ously monitored; when the voltage stress in a single phase
the striking distance expressions are those measured at exceeds the strength, the flashover counter is increased,
the distance from the tower where the vertical channel of and the simulation is stopped.
the return stroke is located. • A report showing the main input and output variables is
3) Overvoltage calculations are performed once the point of printed at every run; the progress of the flashover rate is
impact of the stroke has been determined. The only differ- also reported with a frequency specified by the user.
ence between models for backflash and shielding failure
simulations is the node to which the current source that VII. TEST CASE
represents the stroke must be connected. In this work, only A. Line Configuration
two connecting points (tower, midspan) have been consid-
Fig. 5 shows the tower design for the line tested in this
ered for strokes to either shield wires or phase conductors.
paper—it is a 400-kV line, with two conductors per phase and
Overvoltages caused by nearby strokes to ground are not
two shield wires (Table I).
simulated since their effect can be neglected for transmis-
sion insulation levels.
B. Parameter Specification
4) The overvoltages calculated at every run are compared to
the insulator strength; if the peak voltage at one insulator A model of the test line was created using ATP capabilities
exceeds this random value, the counter is increased and and following the guidelines summarized in Section II.
the flashover rate is updated [32]. Since only single stroke • The line was represented by two span sections at each side
flashes are simulated, this option has several advantages: of the point of impact plus a 3-km section termination at
2204 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005

Fig. 5. A 400-kV line configuration (values within parenthesis are midspan


heights).

TABLE I
LINE CONDUCTORS CHARACTERISTICS

Fig. 6. Distribution of stroke currents that caused flashover. (a) Strokes to


shield wires. (b) Strokes to phase conductors.
each side; each section was modeled as a constant dis-
tributed parameter line, whose values were calculated at
• The return stroke was represented by means of a concave
500 kHz.
waveform, as described in Section IV. In all cases, the cur-
• Towers were represented as lossless single-phase fre-
rent steepness factor to be specified in (3) was 5.
quency-independent distributed parameter lines, whose
The following probability distributions were assumed for
surge impedance value was calculated according to the
each random value:
expression suggested by CIGRE [5]. A value of 134
was estimated for the surge impedance of all towers. • lightning peak current magnitude (only negative polarity):
• The footing impedance was represented as a nonlinear re- log-normal kA, kA;
sistance that behaves according to (1). • rise time: log-normal s, s;
• The strength of insulator strings for negative polarity • tail time: log-normal s, s;
strokes and lines located at sea level was calculated ac- • phase reference angle: uniform, between 0 and 360 ;
cording to the expression proposed by IEC 60 071-2 [24] • insulator flashover: Weibull, kV,
%;
(7) • footing resistance: normal, .
where is the striking distance of the insulator string. The footing resistance parameter is the mean value of the
A correction factor should be used for nonstandard atmo- resistance at low current and low frequency. The value of the
spheric conditions. This is a very simple representation in soil resistivity is 500 m. Random lightning parameters are
which some effects (e.g., effect of the tail time constant independently distributed, that is . The influence of a
[14]), are not included. Some other approaches have been nonzero correlation coefficient is analyzed in Section IX.
proposed; for instance, a factor 605 instead of 700 is sug- The stroke location, before the application of the electro-
gested in [14]. Since a difference of about 15% exists be- geometric model, was generated by assuming a vertical path and
tween both factors, there will be significant differences be- a uniform ground distribution of the leader.
tween the flashover rates derived from both approaches.
• No flashovers other than those across insulator strings C. Simulation Results
(e.g., flashovers between conductors), have been Figs. 6 and 7 show some of the results derived from the base
considered. case, for which a line span of 400 m was assumed. By com-
MARTINEZ AND CASTRO-ARANDA: LIGHTNING PERFORMANCE ANALYSIS OF OVERHEAD TRANSMISSION LINES USING THE EMTP 2205

Fig. 7. Rise-time distribution of strokes to shield wires and towers.


Fig. 8. Optimizing the area of impacts.

paring the two distributions of Fig. 6, one can see that there is a
range of values for every type of failure and a range of peak cur- • Return stroke parameters play an important role in the
rent magnitudes that cause no failure. The procedure is stopped lightning performance of a transmission line. Since only
when the probability density function of all the random variables negative polarity strokes were assumed, more accurate re-
matches their theoretical functions within the specified error. In sults would be derived by assuming that a percentage of
this work, the resulting and the theoretical distributions were return strokes is of positive polarity. A seasonal variation
compared at 10%, 30%, 50%, 70%, and 90% of the cumulative of this percentage could also be considered [21].
distribution functions. More than 10 000 runs were needed to • The return stroke parameters that were included in the
match them within an error margin of 10%. For an error margin study and the way in which they were generated were
of 5%, no less than 30 000 runs were needed. Results shown in motivated by the fact that only the maximum over-
Fig. 6 were obtained after 40 000 runs. The flashover rate, for voltage across insulator strings was of concern. However,
fl/km , was 1.477 per 100 km and year. flashover caused by subsequent strokes could also be
Fig. 7 shows the rise-time distribution of lightning strokes considered [14]. Additional random parameters (e.g., the
to shield wires and towers. It is evident from this plot that the number of strokes per flash and the probability density
probability of failure with rise times above 5 s is negligible. functions for each multiplicity), should be calculated
when other aspects (e.g., line arrester failures) were of
concern.
D. Discussion
• Return strokes with a nonvertical path, when the leader
The following paragraphs are aimed at discussing the limita- approaches ground, have been reported [14]. Therefore, a
tions of the models used in this work and some future work. probability density function for the leader angle could be
• One of the steps that has received more criticism is the ap- considered in future versions of the procedure.
plication of the electrogeometic model, used to determine
the point of impact of a return stroke. Although it has been
adopted by some standards [27], it is recognized that other VIII. REFINING THE PROCEDURE
models (e.g., the Leader Progression model), represent an 1) With the criterion chosen for checking the convergence of
improvement [33]. the Monte Carlo method, one can assume that the number
• The limitations of models used in transients simulations of runs will be fixed for a given convergence error. That is,
are usually due to two reasons: lack of reliable data and this criterion guarantees the convergence of the input vari-
limited built-in capabilities of the simulation tool. Several ables, but since only a small percentage of the randomly
parts of the implemented model are not accurate enough: generated strokes will impact the line, the convergence of
the corona effect was not included in the line span models, the output variables (i.e., the distribution of stroke peak cur-
voltages induced by the electric and magnetic fields of rent magnitudes to shield wires and phase conductors), is
lightning channels to shield wires and phase conductors generally not achieved. This can only be improved by in-
were neglected, footing impedance and insulator string creasing the number of impacts to the line. A very simple
models were too simple. The calculation of induction ef- solution that keeps the above convergence criterion and im-
fects in transmission lines is a new subject for which not proves the solution of the Monte Carlo procedure can be
much work has been performed, see [34]–[35]; however, based on a reduction of the area of impact of return strokes;
they can significantly affect the flashover rate, as reported that is, on decreasing the maximum distance from vertical
in [35]. A new random variable, the return stroke velocity, paths and the line. Locations of vertical path strokes are
must be generated when induction effects are simulated. randomly generated by assuming a uniform ground distri-
Present ATP capabilities are a drawback for an easy and bution. Therefore, only a span length of the line has to be
accurate implementation of corona and induction effects. analyzed. If the area where stroke channels are located is
2206 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005

Fig. 10. Strokes to phase conductors—base case.

The new flashover rate, with m, is 1.466


flashes per 100 km and year. The difference with respect to
the base case is less than 1%; and the number of flashovers
has been increased in 55.3%, as predicted above. Fig. 9
shows the distribution of stroke currents with the new area of
impacts. The range of peak current magnitudes that caused
shielding failure is now different. The sequence of random
numbers was the same in both cases, but the area of impact
was reduced in the second one; therefore, the distances
between vertical channels and the line were changed.
2) Since only a small percentage of randomly generated
strokes reaches the line, not all of the cases have to be
simulated. However, the number of cases to be simulated
can also be decreased without decreasing significantly
the accuracy of calculations by using very simple rules.
The procedure can be trained in order to learn when some
Fig. 9. New distribution of stroke currents. (a) Strokes to shield wires. strokes need not to be simulated as the result can be easily
(b) Strokes to phase conductors. predicted from previous calculations. The rules will be
justified by distinguishing between strokes to phase con-
that shown in Fig. 8, the number of flashovers per 100 km ductors from those to shield wires.
and year, after runs and flashovers, is
A. Strokes to Phase Conductors
(8) The following rules can be used for return strokes that reach
a phase conductor.
Calculations of the base case were performed with
• If the peak current magnitude is larger than the smallest
km. For a sufficiently high number of runs, one can assume
that the probability density function of the return stroke peak current magnitude that caused flashover , the
variables is the same for every horizontal strip with thick- new return stroke will also cause flashover.
ness (Fig. 8). Therefore, the optimum area should be • If the peak current magnitude is smaller than the largest
limited by a distance that corresponds to the max- peak current magnitude that did not cause flashover
imum peak current magnitude as derived from the appli- , the new return stroke will not cause flashover.
cation of the electrogeometric model. The maximum peak This is only partially true, since the flashover caused by a given
current magnitude generated in the previous example was return stroke depends not only on the peak current magnitude
about 400 kA, so the maximum distance should be about but on the voltage and the insulator strength of the phase of im-
500 m, as used in this work. However, the percentage of pact. Due to this fact, a new rule is required, as reasoned below.
strokes with a peak current magnitude exceeding 200 kA Fig. 10 shows the effect of each return stroke that reached the
is very small. If the maximum peak current magnitude is test line after 40 000 runs. One can observe that a clear distinc-
assumed 200 kA, then is reduced to 320 m, and the tion between those strokes that caused flashover from those that
number of impacts to the line will be increased by a factor did not cause it can be established, except for a very few strokes.
of 56%. This approach does not avoid strokes with peak A security margin (e.g., 10%), can be used; that is, every time
current magnitudes above 200 kA, but they will impact at the values of either or are updated during a simu-
less than 320 m from the line. lation, they are, respectively, decreased or increased by a factor
MARTINEZ AND CASTRO-ARANDA: LIGHTNING PERFORMANCE ANALYSIS OF OVERHEAD TRANSMISSION LINES USING THE EMTP 2207

Fig. 11. Strokes to phase conductors after applying the rule.


Fig. 13. Strokes to shield wires after applying the rule.

TABLE II
PERFORMANCE OF THE REFINED PROCEDURE

the coordinates of the terminal points of the curve, and are


derived from the following expressions:
(10a)
Fig. 12. Strokes to shield wires—base case.
(10b)
of 10%. Fig. 11 shows the results and the borders derived after
40 000 runs and using a security margin of 5%. One can observe Therefore, one can assume that for a given rise time
that only about one-third of the strokes to phase conductors were • a return stroke will cause flashover if the peak current mag-
simulated. nitude is on or above the curve;
As the borders are dynamically generated, only those strokes • a return stroke will not cause flashover if the peak current
with a peak current magnitude between the provisional borders magnitude is below the curve.
have to be simulated, since their effect is undetermined and It is obvious from Fig. 12 that this is again partially true, since
cannot be deduced. This guarantees that a wrong conclusion will a non-negligible percentage of strokes above the curve did not
not be derived for any stroke. cause flashover. Therefore, two curves can be again fixed to limit
1) Strokes to Shield Wires: This is a more difficult situation the area of strokes that will be simulated from those for which
since the parameters that have a significant influence are now the above rules can be applied. The security margin was now in-
higher. In addition to the phase voltage, the insulation strength, creased to 15%. Fig. 13 illustrates the effect of these rules when
and the peak current magnitude of the return stroke, the rise time applied to the same cases shown in Fig. 12. About one-third of
of the return stroke and the footing resistance must be included the cases were simulated and 11% of the strokes to shield wires
too. In order to simplify the training of the procedure, it will produced incorrect results.
be assumed that the value of the footing resistance is constant. If the statistical parameters of return strokes are preserved,
Remember that this resistance has a nonlinear behavior. On the the performance of the refined procedure for the calculation of
other hand, the overvoltage caused by a return stroke will not be SFFOR calculation will not be affected when some line param-
the same if the point of impact is the tower or the midspan. eters are changed (e.g., the footing resistance). However, the
Fig. 12 shows the effect of the return strokes that reached the BFOR calculation can show a different performance since the
shield wires after 40 000 runs. The plot includes a curve that can border curves can be different for different values of the footing
be used as a border between strokes that caused flashover from resistance.
those that did not cause it. This curve can be approached by an Table II shows the effect that the approach has had with sev-
equation with the following form: eral cases for which only the value of the footing resistance was
changed. The table shows the number of backflashovers before
(9)
and after applying the above rules, the error in BFOR calculation,
where and are the peak current magnitude and the rise as well as the reduction of the number of simulated cases. It is ev-
time, respectively. If and are ident that the more backflashovers are caused, the more accurate
2208 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005

the new approach will be, although the percentage of simulated


cases remains the same. The calculation error can be decreased by
increasing the security margin, but a tradeoff will always exist
between accuracy and the number of cases to be simulated.

IX. SENSITIVITY STUDIES


Parametric calculations can be very useful to analyze the in-
fluence of some line and stroke parameters, and to determine
what range of values can be of concern. Note that the number of
parameters involved in lightning calculations is very high; how-
ever, for a given transmission line, one does not need to analyze
the influence of all of them, as some can be accurately specified
from the line geometry.
1) Sensitivity studies based on the procedure described
above were performed to analyze the influence that the
median values of the peak current magnitude and the rise
time of the return stroke have on the flashover rate. Plots
of Fig. 14 show the flashover rate, measured per 100 km
and year and using the footing resistance as a parameter.
It is obvious that the rate increases with the peak current
magnitude and decreases with the rise time, but the influ-
ence of the tower footing resistance is not critical for low
values of the soil resistivity and low values of (mean
value of ). All results were deduced after running the
procedure 20 000 times.
2) Another study was performed to analyze the influence that
a nonzero correlation coefficient between the probability
density functions of the peak current magnitude and the
rise time can have on the flashover rate. The generation of
random variables, for which a joint probability distribu-
tion function like that shown in (5) is assumed, has been
based on the conditional probability density function de-
tailed in (6). See [36] for the generation of random vari-
ables tied by a joint probability distribution function with
a nonzero coefficient of correlation.
Two tests were performed; both of them were based on the
probability distribution functions used for the base case detailed
above. First, the values of the peak current magnitude were ran-
domly generated, and the conditional probability distribution
function of the rise time for a given value of the peak current
magnitude was used to obtain rise time values, see (6). Then,
the procedure was inverted, the values of the rise time were ran-
domly generated, and (6) was used to obtain peak current mag-
nitude values.
Fig. 15 shows the results deduced after running both test cases
20 000 times. These plots present the total flashover rate and the
rate of flashovers caused by strokes to shield wires, respectively.
Fig. 14. Sensitivity analysis. Effect of the median values of the peak current
It is evident that results obtained from both test cases match magnitude and the rise time (N = 1 fl/km ). (a) Flashover rate versus peak
very well to each other. Two conclusions can be derived: the current magnitude ( R = 20
). (b) Flashover rate versus peak current
BFOR is very sensitive to the coefficient of correlation between magnitude ( R R
= 100
). (c) Flashover rate versus rise time ( = 20
).
the peak current magnitude and the rise time, while the SFFOR R
(d) Flashover rate versus rise time ( = 100
).
remains practically constant, irrespective of the value of . The
conclusion is that the total flashover rate decreases as the value
of the coefficient of correlation increases [Fig. 15(a)]. This result sity function of the peak current magnitude given the rise-time
is important since values equal or greater than 0.4 have been value.
suggested for [5]. • If the rise-time value is not equal to its median value (2 s),
The following paragraphs are aimed at justifying these re- the median value of the peak current magnitude will be
sults. The reasoning is based on the conditional probability den- different from its original value; see (6b).
MARTINEZ AND CASTRO-ARANDA: LIGHTNING PERFORMANCE ANALYSIS OF OVERHEAD TRANSMISSION LINES USING THE EMTP 2209

ACKNOWLEDGMENT
The second author would like to express his gratitude to the
Universidad del Valle (Cali, Colombia) for the support received
during the preparation of his Ph.D.

REFERENCES
[1] H. W. Dommel, Electromagnetic Transients Program. Reference
Manual. Portland, OR: Bonneville Power Administration, 1986.
[2] J. A. Martinez and J. Martin-Arnedo, “Expanding capabilities of EMTP-
like tools: From analysis to design,” IEEE Trans. Power Del., vol. 18, no.
4, pp. 1569–1571, Oct. 2003.
[3] IEEE Working Group on Lightning Performance of Transmission Lines,
“A simplified method for estimating lightning performance of trans-
mission lines,” IEEE Trans. Power App. Syst., vol. PAS-104, no. 4, pp.
919–932, Apr. 1985.
[4] IEEE Working Group on Lightning Performance of Transmission
Lines, “Estimating lightning performance of transmission lines II:
Updates to analytical models,” IEEE Trans. Power Del., vol. 8, no. 3,
pp. 1254–1267, Jul. 1993.
[5] Guide to Procedures for Estimating the Lightning Performance of Trans-
mission Lines, 1991. CIGRE WG 33-01, CIGRE Brochure 63.
[6] Guidelines for Representation of Network Elements When Calculating
Transients, 1990. CIGRE WG 33-02, CIGRE Brochure 39.
[7] IEEE TF on Fast Front Transients, “Modeling guidelines for fast tran-
sients,” IEEE Trans. Power Del., vol. 11, no. 1, pp. 493–506, Jan. 1996.
[8] A. M. Gole, J. A. Martinez-Velasco, and A. J. F. Keri, Modeling and
Analysis of System Transients Using Digital Programs: IEEE PES Spe-
cial Publication, 1999, TP-133-0.
[9] W. A. Chisholm, Y. L. Chow, and K. D. Srivastava, “Lightning surge
Fig. 15. Sensitivity analysis. Influence of the coefficient of correlation (Peak response of transmission towers,” IEEE Trans. Power App. Syst., vol.
current magnitude = 34 kA, rise time = 2 s, N = 1 fl/km ). (a) Total PAS-102, no. 9, pp. 3232–3242, Sep. 1983.
flashover rate. (b) Backflashover rate. [10] M. Ishii et al., “Multistory transmission tower model for lightning surge
analysis,” IEEE Trans. Power Del., vol. 6, no. 3, pp. 1327–1335, Jul.
1991.
[11] T. Yamada et al., “Experimental evaluation of a UHV tower model for
• According to (6b), the median value of the conditioned lightning surge analysis,” IEEE Trans. Power Del., vol. 10, no. 1, pp.
393–402, Jan. 1995.
peak current magnitude decreases with the given rise-time
[12] T. Hara and O. Yamamoto, “Modeling of a transmission tower for light-
value. The effect is very dependent on the coefficient of ning surge analysis,” Proc. Inst. Elect. Eng., Gen. Transm. Distrib., vol.
correlation; the greater this coefficient is, the greater the 143, no. 3, pp. 283–289, May 1996.
reduction of the median value will be. [13] Y. Baba and M. Ishii, “Numerical electromagnetic field analysis on light-
ning surge response of tower with shield wire,” IEEE Trans. Power Del.,
• The BFOR decreases as increases because most back- vol. 15, no. 3, pp. 1010–1015, Jul. 2000.
flashovers are caused by return strokes with a rise time [14] A. R. Hileman, Insulation Coordination for Power Systems. New
equal or shorter than 2 s (see Fig. 7). York: Marcel Dekker, 1999.
• Direct stroke flashovers are caused by strokes with a peak [15] W. A. Chisholm and W. Janischewskyj, “Lightning surge response
of ground electrodes,” IEEE Trans. Power Del., vol. 14, no. 2, pp.
current magnitude ranging from 14 to 28, but the per- 1329–1337, Apr. 1989.
centage of strokes with these values remains practically [16] A. M. Mousa, “The soil ionization gradient associated with discharge
constant, regardless of the value of . of high currents into concentrated electrodes,” IEEE Trans. Power Del.,
vol. 9, no. 3, pp. 1669–1677, Jul. 1994.
[17] M. E. Almeida and M. T. Correia de Barros, “Accurate modeling of
rod drive tower footing,” IEEE Trans. Power Del., vol. 11, no. 3, pp.
X. CONCLUSION 1606–1609, Jul. 1996.
[18] A. Geri, “Behavior of grounding systems excited by high impulse cur-
This paper has presented a new procedure for lightning anal- rents: The model and its validation,” IEEE Trans. Power Del., vol. 14,
ysis of overhead transmission lines based on new ATP capabili- no. 3, pp. 1008–1017, Jul. 1999.
ties. All results presented in this document have been derived by [19] R. B. Anderson and A. J. Eriksson, “Lightning parameters for engi-
using a single input file. One of the main goals was to analyze neering applications,” Electra, no. 69, pp. 65–102, Mar. 1980.
[20] P. Chowdhuri, Electromagnetic Transients in Power Systems. New
the influence that some parameters can have on the flashover rate York: RSP Press/Wiley, 1996.
of a transmission line. Sensitivity studies can be useful for eval- [21] IEEE TF on Parameters of Lightning Strokes, “Parameters of lightning
uating the influence of every parameter involved in the lightning strokes: A review,” IEEE Trans. Power Del., vol. 20, no. 1, pp. 346–358,
Jan. 2005.
performance and deciding with which accuracy some parame- [22] H. Motoyama, “Experimental study and analysis of breakdown charac-
ters should be specified. teristics of long air gaps with short tail lightning impulse,” IEEE Trans.
The study has been based on a simplified representation of Power Del., vol. 11, no. 2, pp. 972–979, Apr. 1996.
some important parts of the whole model. Future work will in- [23] A. Pigini et al., “Performance of large air gaps under lightning overvolt-
ages: Experimental study and analysis of accuracy of predetermination
clude a more accurate representation of some components (e.g., methods,” IEEE Trans. Power Del., vol. 4, no. 2, pp. 1379–1392, Apr.
footing impedances and insulator strings). 1989.
2210 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005

[24] Insulation Co-ordination, Part 2: Application Guide, 1996. IEC [34] S. Sekioka, T. Ueda, I. Matsubara, and S. Kojima, “Incoming lightning
60 071-2. surge analysis considering return stroke parameters,” in Proc. IPST’99,
[25] F. Heidler, J. M. Cvetic, and B. V. Stanic, “Calculation of lightning cur- Budapest, Hungary, Jun. 20–24, 1999, pp. 383–388.
rent parameters,” IEEE Trans. Power Del., vol. 14, no. 2, pp. 399–404, [35] P. Chowdhuri, S. Li, and P. Yan, “Rigorous analysis of back-flashover
Apr. 1999. outages caused by direct lightning strokes to overhead power lines,”
[26] J. A. Martinez, F. Castro-Aranda, and O. P. Hevia, “Generación aleatoria Proc. Inst. Elect. Eng., Gen., Transm. Distrib., vol. 149, no. 1, pp. 58–65,
de los parámetros del rayo en el cálculo de sobretensiones atmosféricas,” Jan. 2002.
ALTAE, Aug. 18–23, 2003. in Spanish. [36] G. J. Anders, Probability Concepts in Electric Power Systems. New
[27] IEEE Guide for Improving the Lightning Performance of Transmission York: Wiley, 1990.
Lines, 1997. IEEE Std. 1243-1997.
[28] W. A. Chisholm, “The IEEE Flash program: A structure for evaluation of
transmission lightning performance,” Trans. Inst. Elect. Eng. Jpn., vol.
121-B, no. 8, pp. 914–197, 2001.
[29] J. G. Anderson, Transmission Line Reference Book, 345 kV and Above, Juan A. Martinez (M’83) was born in Barcelona, Spain.
2nd ed. Palo Alto, CA: EPRI, 1981, ch. 12. Currently, he is Profesor Titular, Departament d’Enginyeria Elèctrica, Univer-
[30] M. A. Ismaili, P. Bernard, R. Lambert, and A. Xémard, “Estimating sitat Politècnica de Catalunya, Barcelona, Spain. His teaching and research in-
the probability of failure of equipment as a result of direct lightning terests include transmission and distribution, power system analysis, and EMTP
strikes on transmission lines,” IEEE Trans. Power Del., vol. 14, no. 4, applications.
pp. 1394–1400, Oct. 1999.
[31] R. Lambert, E. Tarasiewicz, A. Xémard, and G. Fleury, “Probabilistic
evaluation of lightning-related failure rate of power system apparatus,”
IEEE Trans. Power Del., vol. 18, no. 2, pp. 579–586, Apr. 2003.
[32] G. Furst, “Monte Carlo lightning backflash model for EHV lines. A Ferley Castro-Aranda was born in Tuluà, Colombia. He is currently pursuing
MODELS-based application example,” in EEUG Meeting, Budapest, the Ph.D. degree at the Universitat Politècnica de Catalunya, Barcelona, Spain.
Hungary, Nov. 10–12, 1996, pp. 10–12. He is Profesor Asociado at the Universidad del Valle, Cali, Colombia. His re-
[33] Lightning Exposure of Structures and Interception Efficiency of Air Ter- search interests are in the areas of insulation coordination and system modeling
minals, 1997. CIGRE TF 33.01.03, CIGRE Brochure 118. for transient analysis using EMTP.

You might also like