PROCEEDINGS OF THE IEEE, VOL. 62, NO.

7, JULY 1974 983

Computation of ElectromagneticTransients
HERMANN W. DOMMEL, MEMBER, IEEE, AND W. SCOTT MEYER, MEMBER, IEEE

Invited Paper

AbstToct-Switching operations, faults, andotherdisturbances while others are more readily solved using digital computer
produce surges on transmission lines and oscillations in transformer programs. Thesimulation of initialandtransient recovery
and generator wn
i dn
i gs. Such electromagnetic transients have pri- voltages, as well as lightning surges, probably belongs to the
marily been studied with transient network analyzers since the late
1930%. In recent years, digital computer programs have been de- latter class.
veloped which make simulation by digital computer competitive.The
solution techniques of such programs aredescribed, and theirus-- 11. HISTORICAL REVIEW
OF COMPUTER PROGRAMS
ness is illustrated with practical examples. Possibiiities for further Traveling-wave problems were already studied with graph-
improvements are mentioned. ical methods in the 1920’s and 1930’s, long before digital com-
I.INTRODUCTION puters became available. Basically two techniques evolved,

s WITCHING OPERATIONS, faults, lightning surges,

and other intended or unintendeddisturbancescause

The system must withstand these overvoltages with

namely, Bewley’s lattice diagram technique, and Bergeron’s
method. In the lattice diagram technique [ 6 ] ,a record is kept
of the transmitted and reflected waves; their apportionment
temporary overvoltages and currents in power systems. is computed from reflection and refraction coefficients. The
a certain method of characteristics was probably first used by Lowy
probability, or their effects must be reduced and limited with [ 7 ] and Schnyder [8] for hydraulic surges. Bergeron applied
protective devices. The simulation of transient phenomena is it to a wide variety of problems [9] and strongly advocated
therefore important for the proper coordination of the insula- its use; therefore, i t is often called Bergeron’s method today.
tion, as well as for the proper design of protection schemes. I t does not need reflection and refraction coefficients: instead,
Simulation studies may be needed to investigate interference i t uses linear relationships between current and voltage (so-
in neighboring communication lines, or hazardouscoupling called characteristics) which are invariant when seen by a
effects to personnel, livestock, and equipment. Such simula- fictitious observer who travels with the wave.
tion is alsoneeded to analyze unexpected transient phenomena Both techniques for solving traveling-wave problems were
after their occurrence--such as recent cases of ferroresonance
adopted for computer solutions, the lattice diagram technique
[l] andsubsynchronousresonance [2]-in order t o avoid [lo] as well as Bergeron’s method [ l l ] . I t appears that
similar conditions in the future. Bergeron’s method is better suited for digital computer solu-
The transient phenomena discussed here occur on a scale tions, and most existing general-purpose programs use it.
of microseconds (e.g., initialtransientrecoveryvoltage), In these early programs, lumped elements L and C were
milliseconds (e.g., switching surges), or cycles (e.g., ferroreso- either represented by so-called stub lines [lo], or the differen-
nance). By nature,thesephenomenaare a combination of
tial equationswere transformed into algebraic difference equa-
traveling-wave effects on overhead lines and cables, and of
tions with the trapezoidal rule of integration [ l l ] . With stub
oscillationsin lumped-element circuits of generators, trans-
lines an inductance L is represented by a lossless line with
formers,andother devices. T o distinguishthemfromthe
surge impedance Z = L/At and travel time r =At, a shunt ca-
electromechanical oscillations of generators in transient sta-
pacitance C by an open-ended lossless line with Z = A t / C and
bility studies, they shall be called “electromagnetic transients.”
T =At (At = step widthof the time discretization). No stub line
I t is practically impossible to study electromagnetic tran-
representationseemsto existforseries capacitance.The
sients using hand calculations, except for very simple cases.
choice of the trapezoidal rule almost 15 years ago was fortu-
This complexity led to the development of transient network
nate, because of its absolute numerical stability. I t belongs
analyzers (TNA’s) in the late 1930’s [3]. They are still widely
to a class of implicit integration schemes [12] which are just
used today, and new ones are still being built-[4], [SI. At the
now gaining in importance for the solution of ”stiff systems”
same time, digital computer programs have been developed
(systems having a large spreadof eigenvalues). Many electro-
in recent years which, in many cases, make digital computer
magnetic transients problems are stiff in this sense.
simulations competitive with T N A simulations, especially for
Mostprograms were originally written for single-phase
companies which do not own a TNA, but do haveaccess to a
networks, but have since been extended to multiphase con-
digital computer. This paperdiscusses digital computer meth-
figurations.Someprogramsarerestrictedtothreephases,
ods only,’ and no attempt is made to assess the advantages
while others permit any number of phases, and can then be
and disadvantages of computer programs versus TNA’s. Cer-
used to study two-pole HV dc lines as well, without program
tain problems can be studied more economically on a TNA,
modification [131, [ 141.
Both Bergeron’s method and the lattice diagram technique
Manuscript received January 1, 1974. are onlyefficient for lossless or distortionless tines. Yet propa-
H. W. Dommelwaswiththe Bonneville Power Administration, gation on overhead lines is far from distortionless, and better
Portland, Oreg. 97208. He is now with the Electrical Engineering Depart-
ment, University of British Columbia, Vancouver, B. C., Canada. approximations for line losses had to be found. Reasonable
W. S. Meyer is with the Methods Analysis Unit, Bonneville Power accuracy is often obtained by lumping resistance a t one or
Administration, Portland, Oreg. 97208.
Analog and hybrid computations are discussed in a companion more points along the line [IS]. This technique is generally
paper by P. C. Krause et d.,this issue, pp. 994-1009, acceptable if the resistance is constant, and small compared
 984 PROCEEDINGS OF THE IEEE, JULY 1974

with the surge impedance. But resistance of the ground path

is highly dependent on frequency, and some programs have
been modified so as to handle this frequency dependence by
use of precalculated weighting functions.
Most general-purpose programs solve transients problems
directly in the time domain.I t is doubtful whether Laplaceor
transformation
Fourier
techniques
are useful for general-pur- 1
pose programs. Such programs must be able to handle switches
which operate as a function of certain variables (e.g., opening Fig. 1. Illustrativesingle-phase network
in the vicinity of node 1.
when current goes through zero; closing of gaps when voltage
reaches a certain value) as well as nonlinear elements such as
lightning arresters and saturable reactors. Fourier and at
La- Here the value1 1 4 i s determined from the values computed
place transformation techniques may well be useful, however, the preceding time step
for special studies, such as wave propagation on a single line.
Superposition techniques have recently been proposed to
reduce solution times in cases where many switching opera-
tions must be simulated so as to obtain statistical distribu-
tions [16].The basic idea is t o precompute a few responses
with any of the available solution techniques, and then to The same equations (5) and (6)would have been obtained if
superimpose these responses for each switching operation. the trapezoidal rule of integration had been applied to the
integral in
111. DESCRIPTION OF NODALSOLUTION METHOD
FOR SINGLE-PHASE NETWORKS
The method to be described here [IS] can solve any net-
work consisting of resistances, inductances, capacitances, and
transmission lines with distributed parameters. Fig. 1 will be Hence the technique canbe referred to as either the trapezoi-
used to explain the technique for these linear elements first. dal rule of integration, or approximation with central differ-
I t represents the details of a larger system around node 1. ence quotients. The branch equation for the capacitance is
Suppose that voltages and currents are just being computed derived analogously. I t becomes
at time t . This implies that thevalues at preceding time steps
t -At, t - ZAt, , are already known. Since the sum of the
currents flowing awayfromnode 1 throughtheconnected
branches must be equal to the injected current il, i t follows
with the value 11-,again known from the values computed in
that
the preceding time step
il-z(t) + i1-d) + il-dt) +i d ) = il(t). (1)
If node 1 has no current source connected to it, then& ( t ) = 0.
The behavior of any arbitrary network canbe sufficiently de- = --. { ~ ( -t At) - ~ 4 ( -
t At) ] - i l - 4 ( t - A t ) . (9)
scribed by a system of node equations for all nodes, similar At
to (1) for node 1. However, to solve these equations, i t is first
Nextthebranchequation for transmission line 1-5 is
necessary to express the branch currentsi ~ z etc.,
, as functions
needed. I t can easily be derived if losses are first neglected, to
of the node voltages, which are the state variables of the nodal
be approximated later on. In this case, the wave equations
formulation. The easiest branchis the resistance, for which
I

The differential equation

for
the
inductance
with L’ and C’ being the
distributed
inductanceand
capaci-
tance per unitlength,havethe well-known solution, first
di1-8 given by d’Alembert,
- va =: L- (3)
dt i = F ( x - at) + f ( x + at)
u = Z . F ( x - at) - Z * f ( x + a t ) .
is replaced by a central difference equation ( 10)
~ l ( t-
) v,(t) + ul(t - Af) - ~ o ( t- At) Here F andf are arbitrary functions, to be determined from
2 problem boundaryandinitialconditions;the newly intro-
il-a(t) - il-s(t - At) ducedparameters Z and a are defined by
= L (4)
At
impedance
surge z=&
which can be rewritten as the desired branch equation
At
2L
DOMMEL AND MEYER: COMPUTATION OF TRANSIENTS 985

ThetermF(r-at)in (11) can be interpreted as a wave Note that this is simplya linear algebraic equation in the un-
traveling at velocity a in the forward direction, whilef(r+at) known node voltages, with the right-hand side being known
is a wavetravelingintheoppositedirection.Thedesired from previously computed steps.
branch equation is derived by multiplying i in (10) by Z and For a general network with n nodes, a system of n such
adding i t t o u linear equations can be formed

u + z-i = 2 . Z . F ( z - ut). (11)

Note that the valueu+Zi in (1 1) does not change whenx-ut where
d& not change. This can be interpreted as follows. Let a [GI nodal conductance
matrix,
fictitious observer sit on the line and travel along the line with[v(t)] column vector of the n node voltages,
wave velocity a. Then r-at and consequently o+Zi will re- [i(t)] column vector of current sources,
main constant for him all along the line. With the travel time [I] columnvector of ‘past history”terms.
(the time fora wave to travel from the sending to receiving
the
end) being If the network contains voltagesources connected to ground,
then (16) can be partitioned into part “A” withunknown
line length voltages, and part ‘B” with known voltages. The unknown
7 =
U voltages are found from

a fictitious observer leaving node 5 at time t--7 will see the

expression ~ , ( t - 7 ) + Z . i ~ ~ ( t - 7when
); he arrives time 7 later
at node 1, he will see the expression ~ ~ ( t ) + Z ( - i ~ - ~ Since
(t)). The actual computation in the transients program proceeds
both must be equal as follows: The matrices [GAA] and [GAB] arebuilt,and
[GAA]is triangularized outside the time-steploop, or whenever
o&) + Z * ( - i l - S ( t ) ) = os(t - + Z - i c 1 ( t -
7) 7). (12) changes in switch positionsor changes due topiecewise-linear
representations of nonlinearities takeplace. At each time step,
The negative sign on il4 is due toreversal of reference direc- the vector on the right-hand side of (17) is computed, and
tion compared with icl. Finally, (12) can be rewritten in the then the system of linear equations is solved for [ ~ ( t ) using
]
form of the desired branch equation theexistingtriangularizedconductancematrix.Thereafter,
the terms I of ( 6 ) , (9), and (14) are updated at time t ; they
1
il-o(t) = --.o1(t)
Z
+I M ( t - 7) (13)
are needed in the next time step for lumped inductances and
capacitances and after the elapseof travel time for transmis-
sion lines, to form [IA]for subsequent time steps.
where the term114 is again known frompreviously computed I t should be pointed out that many elements in [GAA]are
values : zero. This “sparsity” is exploited in the solution process by
using orderedeliminationtechniques [17]. Inparticular,
1 [ G d ] is strictly diagonal in systems where lumped elements
11--5(f - 7 ) = - - og(t - 7 ) - i c l ( t - 7 ) . (14)
appear only from node to ground or from nodes of subset A
Z
to source nodes of subset B. This is because lines with dis-
For example,if 7 = 1 ms and At= 100 ps, then the valueneeded tributed parameters contribute only to thediagonal elements.
would be that of 10 steps back in time. Therefore, a record of In sucha special case,-& equations can be solved separately,
the past history would have to be kept over the last 10 time node by node. Some programs are restricted to this special
steps. topology, or use stub-line representations to obtain such re-
Equation (13) provides an exact solution for the lossless stricted topology. However, sparsity-solution methods auto-
line a t i t s terminals, and is the basis of Bergeron’s graphical matically exploit the special feature of a diagonal matrix, and
method [9]. This relation is also valid for distortionless lines at the same time accept off-diagonal elements without any
if I1-&(t-7) in (13) ismultipliedbyanattenuationfactor restrictions.
exp (- R’r/2Lf). The factor 3 is used to assign the losses de- The past history terms are simply preset to zero if the
fined by R’ to the shunt conductance as well as to the series simulation starts fromzero initial conditions. For cases which
resistance [43]. Overhead lines are not distortionless,however, start from linear ac steady-state conditions, a subroutine must
and a better approximation is obtained by adding lumpedre- be used to compute the ac steady-state solution. The past
sistances R / 2 at the ends, or by adding R/4 at the ends and history tables can then be preset with the correct values.
R / 2 in the middle [IS]. This provides sufficient accuracy for
many cases. IV. EXTENSIONTO MULTIPHASE KETWORKS
Upon the insertion of (2), (5), (8), and (13) into ( l ) , the The preceding method can be extended to multiphase net-
node equation is obtained in its final form, works byformally replacing scalarquantitieswithmatrix
quantities. This generalization is straightforward for coupled
1 At 2C 1 At
R
+ -+ __ +
2L At Z
.o1(t) - - o*(t) - -
R
2 L v&)
lumped inductances. The program of [15] has an input option
for multiphase Pi-circuits, which are solved with the matrix
equivalent of the scalar equations (2) and (5) for the series
impedance matrix andof (8) for the shunt capacitance matrix.
Cascadeconnections of suchmultiphasePi-circuitscan be
used to model lines with any number of phases, e.g., parallel
untransposed lines on the same rightof way. Pi-circuits must
986 PROCEEDINGS OF THE IEEE, JULY 1974

be used on network analyzers, but digital computers offer a

betteralternative which avoidsthe problem of cutoff fre-
quency.
To solve multiphasetransmission lines with distributed
parameters, the coupled equations in the phase domain must
first be transformed into decoupled equations in the modal
domain. Each mode is then described by the same equations
Fig. 2. Transformation of three-phase line
which were derived for single-phaselines. For each mode there from phase to medal domain.
is an equation of the form (13), so that for a three-phase line,
as shown in Fig. 2,
tween phases are equal. The series impedance and shunt ca-
1
ilO-kO) = - na(t)
z.
+ Ila-k(t - 7.)
pacitance matrices of a completely balanced line thus have the
form

1
Z,’ Z,’
Z,’
Z.’
Z,’
Z,’ * ’

* *
Z,’
Z,’ 1
I : -1
1 [Z’] = Z,’ 2,’ z,’ Z,’
ile--b(O = - VlC(t)
Z,
+ Ile-zc(t - 70) (18)
* *

where past history terms are computed from

modal quantities, Z,’ Z,’ Z,’ * ’ * z,‘
1 Y,’
Ila-k(t - 7.) = - - ~ % ( -
t 7). - ik--1o(f- 70) (19)
Za Y,’
with similar expressions for modes b and c. Next (18) is trans- Ym’
formed back into the phase domain, giving
[il-BPt-=] = [zpq-’.[ U p q + [I1_2ph4=] (20) Ym’
where For completelybalanced lines, thereare a number of
simple transformation matriceswhich decoupie the line equa-
tions. One such matrixfor three-phase lines is the transforma-
tion matrix of a,8, 0 components. I t is well suited for tran-
sients studies because its elements are real, and though less
The transformation matrices [S]and [Q] are defined in the important, it is identicalfor voltages and for currents. Trans-
Appendix, and are assumed to be real. Such real transforma- formation to a,0, 0 components is therefore used in many
tionmatricesareapproximations,butreasonablyaccurate three-phase transients programs.
[18]. The matrix [Zmde]--l is the inverse of the diagonal ma- Another simple transformation, which has the same ad-
trix of modal surge impedances as defined in the Appendix. vantage as a,0, 0 components, plus the added advantage that
At this point,(20) can be entered into (17) in a way similar to i t can be used for any number of phases, is [19]:
that used for single-phase line equation (13). The difference is
that a 3 X3 conductance matrix [Zphrse]--l now contributes 9 -1 1 1 ... 1 -
elements instead of 1-9 elements to therows and columns of
1 1-M 1 1 * e ’

nodes l A , lB, lC, and analogously to those of nodes 2A, 2B,

2C. Also, a vector [I1--ghass] with 3 components is entered into [SI = [Q] = 1 1 1 --M . . . 1 (23)
the right-hand side, instead of a scalar value.
If a systemwiththree-phase lines hasonlyshunt-con-
nected lumped elements (node to ground) or lumped elements -1 i 1 . . . 1 - M,
connecting nodes of subset A to nodesof subset B , then [ G A A ]
in (17) becomes block diagonalwith 3x3 submatrices as where M is the number of phases. Its inverse is also simple
diagonal elements. In this particular case, subsets of 3 nodes ’1 1 1 ... 1-
-in practice,the 3 phases of a busbar-could besolved
separately,busbar by busbar. As inthesingle-phase case, 1 -1 0 ... 0
sparsity solution techniques automatically exploit this special 1 0 -1 0
structure of the matrix.

V. BALANCED MULTIPHASE LINES

-1 0 0 ... - 1,
Multiphase lines with distributed parameters are said to
be Yelectromagnetically balanced”if the self-impedancesof all Relation (24) provides an interpretationfor the modal quanti-
phases are equal among themselves, andif all mutual imped- ties: The first mode is defined by a loop formed by all phases
ances between phases are equal. Suchlines are similarlycalled connected in parallel, with ground providing the return path.
‘electrostatically balanced” if the capacitances to ground of The 2nd, . , Mth mode is defined by a loopformedby the
all phases are equal to each other, and if all capacitances be- first phase, with the return path through the 2nd, . . . , Mth
WMMEI.AND MEYER: COMPUTATION OF TRANSIENTS 987

phase, respectively. Thematricesin (23) and (24) arenot

normalized, but thiscould of course easily be done.
The parameters of a balanced line in the modal domain
are, for thefirst (ground return) mode,

andfor modes 2, ,M,

For balanced three-phase lines, the parameters of the first

Fig. 3. Phase configuration of two three-phase
mode (25) are identical with the zero sequence parameters, circuits on 'the same tower.
and those of the other d es in (26) are identical with the
positivesequenceparameters of symmetricalcomponents,
which have been used for more than 50 years to decouple
steady-state line equations. Symmetrical components are im-
practical for transients studies, however, because the trans-
formation matrix is complex.
The assumption of balanced matrices is primarily helpful
A two-pole HV dc line
for single two- and three-phase circuits. Fig. 4. Transposition scheme of a seneacornpensated
double-circuit line, 267 km long.
has balanced matricesu priori, provided the conductorsof the
two poles are identical, and are supported at equal heights
above the ground. The matrices of three-phase lines become
approximately balanced when the line is transposed (a prac-
tice which has usually been followed, except recently in some
cases of UHV lines). But even for an untransposed three-phase
line, acceptable results can often be obtained with balanced
matrices. An exception is the case of lines having shunt re-
actors, where the electrostatic unbalance may have a notice-
abie influence on the oscillating trapped charge[20].
LINES
VI. PARALLEL
Theassumption of balancedmatricesis less useful for
parallel circuits, and, in general, eigenvector matrices must be
found for eachparticular case (see Appendix).Sometimes
simplificationsmm be made [21]. For example, a double-cir-
cuit three-phase line is particularly simple if the circuits areso Fig. 5. Receiving-end voltage of one phase of arcuit I, using Pi-circuit
close together that the average distance betweenphases of representation.Solid curve for balanced matrices,dotted curve for
actual transposition.
different circuits is almost equal to the average distance be-
tween phases within one circuit, as in the exampleof Fig. 3.
Switching surges on such a double-circuit line, with the
transposition scheme of Fig. 4, were simulated for the closing
of circuit breaker 11. The line was first modeled as a cascade
connection of 15 six-phase Pi-circuits. The dotted curve in
Fig. 5 shows the receiving-end voltage on one phaseof circuit
I, for the actual transposition scheme; unbalanced matrices
for the actual tower configuration were used for the Pi-cir-
cuits, and the transpositions were made through appropriate
connections of the Pi-circuits. The solid line in Fig. 5 is ob-
tained if the diagonal and off-diagonal elements of the ma-
trices are averaged out, respectively, to make theline appear
balanced. The average difference in the peak values during
the first two cydes, at the 6 receiving-end nodes, is 7 percent
(based on 2.0 pu). Another comparison was made between
lumped and distributed parameter representations, with bal-
Fig. 6 . Receiving-end voltage of one phase of arcuit I, using balanced-
anced matrices in both cases (Fig. 6 ) ; the average difference matrix representation. Solid curve for distributed parameters, dotted
is about 10 percent (based on 2.0 pu). At least for this case, curve for lumped Pi-circuits.
the difference between actual transposition and balanced ma-
tricesis less than the difference betweenlumpedand dis-
988 PROCEEDINGS OF THE IEEE, JULY 1974

tributedparameter representation-or, fromanother view-

point, between digital computer simulation with distributed $
C
parameters and T N A results. Solution times for the program
of [IS] on a CDC-6400were 14 ms per timestepfor dis- !
tributed parameters, and about ten times as much using Pi-
!
Fig. 7. Transposition scheme for double-arcuit line,
circuits. produang coupling in zero sequence only.
Using balanced matrices for double-circuit lines implies
that a) the two circuits are identical in conductor type and
tower configuration, b) that each circuit is transposed within from the positive sequence of one circuit to the positive se-
itself, c) that coupling between the circuits exists only in the quence of the other circuit, or from the positive sequence of
zero sequence,andd)thattheaveragedistancefromone one circuit to the negative sequenceof the other circuit.
phase of one circuit to a phaseof the other circuit is approxi-
VII. FREQUENCY
DEPENDENCE
OF DISTRIBUTED
mately equal to the average distance between phases within
PARAMETER
LINES
one circuit.
If assumptions a)*) are maintained but assumption d) is Bergeron's method provides an easy solution for lossless
dropped, then the double-circuit linewould be described by a and distortionlesslines, as well as an approximate solution for
series impedance matrix of the form lossy lines by lumping resistancea t discrete locations. Butfor
more accurate modeling of thedistributedresistance,and
-z: Z,' Z,' z,' z,' z,'- especially €or cases where line parameters R' and L' depend
Z,'
Z*'Z,' z,' z,' z,' on frequency, more sophisticatedtechniquesarerequired.
This is particularly true for the ground-return mode, wherein
Z,' Z,' Z,' z,' z,' z,' the impedance of the earth is highly frequency-dependent.
[Z'] = (27) Table I shows such variation for a typical 500-kV overhead
line, as found from Carson's formula [22].
z,' z,' z,' Z,'Z'' Z,' The increased damping a t higherfrequencies produces
z,' z,' z,' Z,' Z'' Z', smoothing and rounding of sharp corners of traveling waves,
as well as general broadening and flattening of pulses which
-z,' z,' z,' Zm'Z,' Z*'-
travel down the line.
and a similar structure for [C']. This provides a reasonable In spite of such parameter variations with frequency, the
approximation for a wider class of double-circuit lines than ac steady-state phasor equations (37), (38) of the Appendix
(22) does, and in terms of data specification, corresponds to still hold for any fixed frequency, and are still linear. Super-
the case where both circuits aredefined by identical zero and position maytherefore be applied, and could be used with
positive sequence parameters, and by zero sequence coupling Fourier transformation theory to obtain the total response
between the two circuits. Such an assumption is sometimes from the responses a t each frequency. Thus a rigorous theo-
used inTNAstudies.Thetranspositionscheme of Fig. 7 retical solution to the problem exists.
would produce the matrix form (27) provided bothcircuits I n Fourier transformation, the frequency spectrum of the
have identical conductors and the tower configuration is sym- outputfunction is obtained by multiplyingthefrequency
metrical. Line equations with matrices of the form of (27) are spectrum of the input function by the transfer function. This
.ansformation matrices multiplication can be converted to the time domainby means
of the convolution theorem, which makes i t possible to solve
- 1 2 2 1 0 0 the problem in the time domain rather than in the frequency
1 - 4 2 1 0 0 domain. a necessity,
Solution in the time domain is practically
for reasons indicated in Section 11. Early applications of the
1 2 - 4 1 0 0
Fourier transformation to power-line problems are found in
[23],[42]. Theconvolutionformulation of [23] was later
modified by a transformation of variables [24] which greatly
1 0 0 - 1 2 2
improved the numerical efficiency, and made the procedure
1 0 0 - 1 - 4 2 completelycompatiblewiththedigitalprogram of [IS].
-1 0 0 -1 2 -4 Defining
bk = vk - zlik b, = Vm - Zli,
1 1 1 1 1 '
.:-1 0 0 0 0
fk = vk + Zlik fm = Dm + zlim
Z1 = lim Z ( j o ) (30)
1 0 - 1 0 0 0 -..I D

for a transmission line from K to m, the integrals needed a t

11 1 1 -1 --11 each time step are
~

'0
Lo
0
0
0
0
1
1
-1
0
0
-1,
bk(t) = JOm { al(u)fm(t - u) + &?(u>fk(t-). ]

Other transposition schemes for double-circuit lines pro-

duce coupling in addition to that in the zero sequence, e.g.,
bm(t) = J 0
O0 { al(u)fk(t - u) + az(u)frn(t - u)1 du (31)
DOMMEL AND AfEYER: COMPUTATION OF TRANSIENTS 989

TABLE I
PUTRRS OF GROUNDRETURNMODE

Fig. 8. Compensationmethod for a single nonlinear or time-varying

where a,(u) and a,(u) are weighting functions which are pre- resistance. The nonlinear or time-varying characteristic at time t is
indicated by ( l ) , the Th6venin characteristic by (2).
calculated by inverse Fourier transformation [25]. I t is the
simple nonrepetitive form of these weighting functions (see
[25, fig. 51) which makes the numerical integration of (31)
easy. This computation must be performed at each time step, ELIMINATION
IN UPPER PART
so bk and b, are known. Then the definitions of (30) provide
the two linear algebraic equations desired. For the line 1-5 ELIMINATION
EOUIVALENT IN LOWER PART
of Fig. 1, (13) would be replaced by (a) (b)
4 ) - Zlil--6(f) = bl(t) (32) Fig. 9. Reduction for network equivalent.

which is identical in form to the conventional representation

derived before. unknown branch current ib is then found from the intersec-
The preceding comments apply to single-phaselines. Pro- tion point, which when inserted into (33), produces the com-
vided one assumes a multiphase line to be batanced, as per plete solution. Not only is such a procedurevery efficient
Section V, there exists a modal transformation independent of numerically, but convergence is certain, provided only that
frequency, such as (23). Frequency dependence as previously the branch characteristic is sufficiently regular.
described can then be applied just to the ground-returnmode This technique for nonlinear resistances can also be used
(25) without complication, exactly as i t was in the single- for nonlinear inductances, since nonlinear flux-current char-
phase case. But for the untransposed line, the process becomes acteristicscan be transformed to vi characteristics by ex-
considerably more complicated. As explained in the Appendix, pressing the flux as the integral of the voltage, and thenusing
eigenvalues and eigenvectors must generallybe calculated a t the trapezoidal ruleon this integral [26].
different frequencies, and the scalar multiplicationsa(u)f(t-u) The compensation method can also be used for systems
of (31) arechangedto matrix-vector products[24].The with more than one nonlinearity, provided they are separated
addedcomputationandprogramming associated withthis from one another by a distributed-parameter line. The ends
general case may not be justifiable, if approximate, frequency- of suchdistributedlinesareudisconnected” byBergeron’s
independent matrices [18] give acceptable results. method,therebyproducingthe possibility of disconnected
subnetworks, with one nonlinear or time-varying element al-
VIII. NONLINEAR ELEMENTS AND SWITCHES
lowed for each.
The solution techniques discussed so far all apply to time- If more than one nonlinear branch existsper disconnected
invariant linear networks only. But the representation of cir- subnetwork, the compensation procedure could still be used,
cuitbreakers,transformer.orreactorsaturation,nonlinear though complications enter. For L nonlinear elementsin a
resistance (lightning arresters, arcs, etc.), and the like all re- given subnetwork, the preceding scalar .z* is replaced by an
quire extensions of the basic procedure. L X L square matrix, and an iterative, simultaneous solution
Little can be said in general about the modeling and solu- of L nonlinemequations would thus be required.
tion of systemswitharbitrary nonlinearities. Yet iterative A related but distinct techniqueis that of network equiva-
solution schemes such as Newton’s method can often be used lents [26], wherein elimination of all voltages for nodes not
successfully. Solutionsingularities,uncertain convergence, incident to nonlinear or time-varying elements is performed
etc., all pose possible complications, however, depending on (see Fig. 9). In this way, an equivalent systemof nonlinear or
the particular typeof nonlinearity. time-varying equations among the remaining nodes is pro-
Practical numerical solutiontechniqueshavethus been duced.Anyiterationduetothenonlinearities,orchanges
tailored to thespecific nonlinearities in question. For example, withtime, need only be appliedtothisreducedsystem of
the compensation method can readilybe applied if the system equations. The opening and closing of an ideal switch ean be
contains only one nonlinear or time-varying resistance (see handled in this way, sincea switch can be regarded as a time-
Fig. 8). Superposition then gives the total network solution varying linear branch; when closed, columns and rows of the
[ v ( t ) ] as being equal to the value [ e o ( t ) ] found with the non- reduced matrix which correspond to the terminalnodes of the
linear branch (k, m) omitted, plus the contribution produced switchmustsimply be added together. For other types of
by branch current i h : nonlinearities, this reduced system may have tobe triangular-
ized more often, perhaps several times per time step if New-
ton’s method is used.
Vector [z] is a property of the linear network only, found by Since very detailed characteristics are not always needed,
a single repeatsolutiononce [GI has been triangularized. it is often possible to use piecewise-linear representations for
Forming v b = u k - v , from (33) gives the straight line of Fig. them.Saturablereactorsandmagnetizinginductances of
8(b), which, of course,representstheTheveninequivalent transformers can normally be represented in this way, often
load line of the linear network, having slope Zth=z&-&,. The with sufficient accuracy using only two or three slopes[ 1 I n 1.
990 PROCEEDINGS OF THE IEEE, JULY 1974

such cases no iteration is required, and the reduced conduc- the diagonal elements. Upon inversion,
[ Y ] - 1 = [ R ] +~ w [ L ] .
tance matrix need only be retriangularized a t times of tran-
sition from one slope to the other.
Matrices [R] and [L] canthen be used torepresentthe
IX. LIGHTNING ARRESTERS coupled windings.
Older types of lightning arresterscould be modeled simply If saturationinthemagnetizingimpedanceisto be
by a nonlinear vi characteristic in series with a switch which modeled, then lumping it as a nonlinear shunt inductance a t
closed when the voltage across it reached the value of the oneterminalissometimesaccurateenough.Thisapproach
sparkovervoltage.
But
modern lightning
arresters
are was used inarriving at thecurves of Fig. 12. Betterac-
equippedwithcurrent-limitinggaps.Thevoltagebuildup curacy mightbe obtained by connecting parts toall terminals.
across these gaps is important, and can be approximated by a Saturation curves are normally only available as urn. versus
voltage source with a fixed voltagetime characteristic [27], & characteristics,though for thesimulationone needs
[28]. More sophisticated models with dynamic effects have flux-current characteristics. One curve can be converted into
recently been developed, whichalso reproduce the loops in the other point-by-point, provided hysteresis effects and fre-
the vi characteristics and. their differing shapes for differing quency dependence are ignored. The modeling of hysteresis,
wave fronts [29]. In principle i t is also possible to model the frequency dependence, and capacitive coupling between layers
sparkover voltage as a function of wave front steepness (see of windings and to the tank of the transformer is beyond the
discussion of [29]) though a workable definition-of steepness scope of this paper, and is an appropriate subject for further
may be difficult for arbitrary switching surges. research. Thedifficulty seems tobe not so much “how well the
model corresponds to reality,” but rather that “the reality
TRANSFORMERS,
X. GENERATORS, A N D LOADS
itself is not well known” [32].
The type of models used for generators and transformers Practically no data are available on composite loads as
dependsontheparticularproblem.Forswitchingsurge seen fromthe high-voltage side of thetransformers,even
studies, it is normally not necessary to represent the supply though their representation seems to be important in certain
network in great detail. If the line is fed from a number of studies where ferroresonance or strong dynamic overvoltages
interconnected power plants, then it is usually sufficient to are present [33].
represent the generators by balanced three-phase voltages E”
behind subtransient reactancesXa”.For longer-duration “dy- XI. COMPARISON WITH FIELD TESTS
namic overvoltages,” the assumptionof E” behind X d “ is not The usefulness of digital computer and TNA simulations
sufficient, and more sophisticated generator models are must ultimatelybe proved by comparing the results with mea-
needed [SO]. surements obtained from field tests. This not only checks the
The impedances of the generator and its step-up trans- correctness of the algorithms, but also the adequacy of the
former can be lumped into oneThCvenin impedance matrix in models. A careful error analysis of the measurements is often
cases of high-side switching. Normally, the zero sequence im- essential if differences have to be explained, sincemeasure-
pedance af this equivalent will be different from that of the ments as well as simulation results can be in error.
positive sequence. But for low-side switching studies, as well Comparisons between field testsandsimulationshave
asforothertypes of problems,thetransformermust be been made for switching surgeproblems, with good agreement
modeled in more detail. The first step toward such sophistica-reported €or both the TNA [34] and the digital computer
tion is the accurate representation of winding connections and [35]. C I G R E (ConfCrence Internationale des GrandsRCseaux
leakage inductances, with resistance and magnetizing imped- Electriques B Haute Tension) Working Group 13.05 has com-
ance ignored. Sucha model can easily be,derived from thewell pared switching surge simulations for line energizations from
known steady-stateequivalents.Forexample, for a two- an inductive source ina specified network. Agreement among
winding transformer, T N A simulations, as well as among digital computer results
from several sources, was found to be within about 55 per-
cent [33].
Fig. 10 shows a comparison between the measured and
computed shape of an impulse on an HV dc line a t a point
If resistance is ignored, then obviously 13.2 km from the sending end. The impulse was created by
discharging a capacitor bank into the two paralleled poles of
[LI-~= jw[ -tyl
-tY t2Y
the line. Ground provided the return path, so the ground-
return mode alone was excited. The good agreement shown
demonstrates the usefulness of Carson’s formula [22] as well
This matrix can thenbe used directly in the matrix equivalent as the technique of Section VI1 for handling frequency de-
of ( 5 ) to model the two coupledwindings. The term At/2L pendence.
would, of course,bereplaced by At[L]-l/2. While [L]-1 is Following currentinterruptionin a circuitbreaker, a
singular, there is no difficulty as long as its inverse is not voltage builds up across the contacts. The shape of this “tran-
used. Three-phase transformer banks made upof single-phase sient recovery voltage” depends upon the characteristics of
units have been represented this way [31]. the network on both sides of the circuit breaker; if i t rises
The next step toward greater detail is the inclusion of a faster than the increase in dielectric strength, restriking will
linear magnetizing impedance. In this case, one can establish occur. Fig. 11 compares the measured and computed transient
the Y-matrix of the well-known steady-stateequivalents, recovery voltage for theinterruption of a line-to-ground
with the magnetizing admittance normally added evenly to fault [36].
DOMMEL A N D MEYER: COMPUTATION OF TRANSIENTS 991

general-purpose transients programs is needed for the study

of dynamic overvoltages, as in the case of load rejection. Such
representation could also be used to study the various damp-
ing effects of faults close to generators, in order to establish
guidelines for the representation of damping in transient sta-
bility studies. I t appears that transients programs will con-
tinue to increase in sophistication and capability, and will
play an increasingrole in power system analysis anddesign in
the future.
APPENDIX
MODALTRANSFORMATION
OF LINE EQUATIONS
Fig. 10. Impulse in ground-return mode of HVdc line. Curve 1-
Sending-end voltage. Curve 2-Measurement at 13.2 km. Curve 3- The steady-state behavior of an M-phase transmission line
Digital simulation a t 13.2 km.
at a discrete frequency is described by the phasor equations

where
[ V ] vector of voltagephasors VI, -
, VM,
[ I ] vector of currentphasors 11,. , IM,
Fig. 11. Transient recovery voltage. Solid curve for measurements, [Z’] series impedance matrix per unit length,
dotted curve for digital simulation
[ Y’] shunt admittance matrix per unit length.
For overhead lines, the conductance(real) part of [ Y‘] can be
ignored; then
[ Y’] = jW[C’] (39)
where [C‘] is the inverse of the matrix of Maxwell’s potential
coefficients [ 3 7 ] . The elements of [Z’] arecomputedfrom
Carson’s formula [ 2 2 ] or Pollaczek’s formula [ 3 8 ] . Both take
-4001 the ground return into account, and were developed 40 years
ago. Questions havebeen raised as to whether the assumptions
Fig. 12. Harmonics following load rejection. Solid curve for
measurements. dotted curve for simulation. in these formulas are justified for power lines, however [ 3 9 ] .
Differentiating (37) a second time with respect to x , and
replacing [ d I / d x ] with (38) gives
Fig. 12 shows the measured and computed steady-state
voltage at the sending end of a transmission line following
load rejection [ 3 6 ] . A strong seventh harmonic appeared in
this case. The harmonics were caused by a voltage rise a t t h e
sending end, which in turn drove the magnetizing impedance and similarly
of the step-up transformer into saturation. The computed
results were obtainedby using theresults of a linear ac
steady-state solution as initial conditions. In this way, the
only disturbance is that due to the deviation between linear
These relations can now be diagonalized* using a modal linear
andnonlinearmagnetizingimpedance;thesolutionthen
transformation
settlesdowntothenonlinearsteady-statesolutionhaving
harmonics very rapidly, in about 2 cycles.

XII. CONCLUSIONS which produce the result

During the last decade, much progress has been made in
solvingelectromagnetictransientsproblemsonthedigital (43)
computer.Forcertainapplications,general-purposedigital
programs have become competitive alternatives to transient
network analyzers. Yet more research is still needed to de-
velop better models (more-accurate transformer representa-
tions, for example), which in manycases will require extensive
field and laboratory measurements on the devices in question, * I t is usually aasumed that the matrix products in (40)and (41) can
always be diagonalized. The authors have not seen any proof of this,
Theincorporation of more-detailed generator models into however.
992 PROCEEDINGS OF THE IEEE, JULY 1974

where [yrnde*] is a diagonal matrix. Its diagonal elements are lowing are approximate equivalent parameters:
the squares of the modal propagation constants. In (42),[SIis
the matrix of eigenvectors of the matrix product [Z’] [ Y’], surge impedance with losses ignored
while [Q]is the matrix of eigenvectors for the reverse product d;
[ Y’] [Z’].
A reliable solution technique for finding eigenvalues zmde-i = - d- R e { ~y~-;~]
and eigenvectors is the QR transformation [4O]. W
Once [SI has been found, [Q]can be computed from the travel time with losses ignored
simple general relationship
length
7mode-i = -d- R e { Trnodc-?}
W

Or, if the conductance part of [I“]is zero, one can alterna- resistance per unit length
tively use
di
Rmode-i = - Im { ~ r n ~ d e - i ’ ] . (5 1)
W

In thepreceding relations, [Dlis an arbitrarydiagonal matrix Here di is the ith componentof diagonal matrix [Dl, arbitrary
of full rank, and can be chosen so as to normalize the columns though assumed to be real. Sincey m d + i Z is a complex number
of [Q].Relation (45) is valid €or any square matrices [Z’] and slightly less than M O O , note that Im (yrnde--if}is positive and
[ Y’],as long as their product is diagonalizable. Since these Re { ”/mde-iz } is negative.
matrices are symmetric, there is another relationship of in-
terest [41]: REFERENCES
(47) [l] E. J. Dolan, D. A. Gillies, and E. W. Kimbark, “Ferroresonance in a
transformer switched with a n E H Vline,” I E E E Trans. Power A p p .
Syst., vol. PAS-91, pp. 12731280, May 1972.
where “t” is used toindicatetransposition.Thisequation [2] J. W. Ballance and S. Goldberg, ‘Subsynchronous resonance in
seems to be valid only if all eigenvalues are distinct. Since a series compensated transmidon lines,” I E E E Trans. Power A p p .
balanced M-phase line (Section V) has M - 1 multiple eigen- Syst., vol. PAS-92, pp. 1649-1658, Sept./Oct. 1973.
H. A. Peterson, Transients i n PowerSystmrs. New York:Dover,
values, (47)will not be used. I t is not valid, for instance, for 1966 (reprint of 1951 ed.).
symmetricalcomponentsappliedtobalancedthree-phase J. K. Dickson, D. E. Hedman, W. A. Lewis, R. M. Webler, and S. N.
Talukdar, ‘A new TNA,” presented a t t h e I E E E 1973 Summer
lines, where Power Meeting, Vancouver, B. C., Canada,July 15-20. Papers

[
C 73 3-2 and C 73 391-0.
1 1 1 G. L. Wilson and J. G. Kassakian, ‘Effects of zero sequence Model-
ing and transposition on switching surge overvoltages,” presented
[SI = [Q] = 1 a2 a ] . (48) at the IEEE 1973 Summer Power Meeting, Vancouver, B. C.,
Canada, July 15-20. Paper T 73 445-4.
1 a a2 L. V. Bewley, Traveling Waves on TransmissionSysUms. New
York: Dover, 1963 (reprint of 1951 ed.).
The solution of (43),(44)in the modal domain for modes R. Lljwy, Druckschwankungen in Bhuckrohrkitungm (pressure vari-
1, . . ,M is identical with the solution of M separate single- ationsin pressure pipe lines), Vienna, Austria:Springer,
p. 84.
1928,
phase equations, for which known solution procedures exist, 0. Schnyder, ‘Druckstijsse in Pumpensteigleitungen” (pressure
such as Bergeron’s method. impulses in pumped pipe lines), Schweiz. Bauztg., vol. 94, no. 22, pp.
271-273; no. 23, pp. 283286,1929.
There are two major difficulties in applying modal trans- L. Bergeron, D u Coup de Bel& en Hydraulique au Coup de Foudrc en
formations to the solution of electromagnetic transients. One ElcchicW. Paris,France:Dunod, 1949 (Transl.: WaterHammcr
is the frequency dependenceof matrices [SIand [ Q ] ;the other i n Hydraulics and Wave Surges in Electricity; translating committee
sponsored by ASME. New York: Wiley, 1961).
is the fact that elements of thesematricesare, in general, L.0. Barthold and G. K. Carter, “Digital traveling-wave solutions.
complex. I t hasbeen shown,however, thatacceptableac- 1-Single phase equivalents,” AZEE Trans. vol. 8 0 , pt. 111, pp. 812-
820, Dec. 1961.
curacycan be obtainedwithapproximatetransformation W. Frey and P. Althammer, “The calculation of transients on lines
matrices which are real and independent of frequency [ 183. by means of a digital computer,” Brown h e r i Rev., vol. 48, pp.
Bergeron’smethodcan be usedwith approximate real 34&355,1961.
C. W. Gear, N u w i c a l Inifial Value Problems in Ordinary Differcn-
frequency-independent transformation matrices
after
the Lial Equations. Englewood Cliffs, N. J.: Prentice-Hall, 1971.
surge impedances and travel times have been defined in the N. G. Hingorani, “Transient overvoltage on a bipolar HVDC over-
modal domain. To determine the appropriate definitions, (37), head line caused by DC line faults,” I E E E Trans. Power A p p . Syst.,
vol. PAS-89, pp. 592602, Apr. 1970.
(38)are transformed into the modal domain, assuming that W. F. Long, ‘A study on some switching aspects of a double circuit
(46)is used for [Q],giving HVDCtransmissionline,” ZEEE Trans. Power A p p . Syst., vol.
PAS-92, pp. 734-74?, Mar./Apr. 1973.
H.W.Dommel, Digitalcomputersolution of electromagnetic
transients in single- and multiphase networks,” I E E E Trans. Power
A p p . Syst., vol. PAS-88, pp. 388-399, Apr. 1969.
P. C. Kalra and E. K. Stanek, “An indirect technique utilizing the
lattice approach and superposition for computing switching surges,”
and IEEE Trans. Power A p p . Syst., voL PAS-92, PP. 916-925, May/

- [%I = jw[D]-l[vm&].
June 1973.
W. F. Tinney and J. W. Walker,
equationsbyoptimally
‘Direct solutions of sparse network
ordered triangularfactorization,” Proc.
ZEEE, V O ~ .55, pp. 1801-1809, NOV.1967.
P. C. Magnuson, “Traveling waves on multi-conductor open-wire
Here both equations have diagonal matrices, so we can define lines-A numerical survey of the effects of frequency dependence of
modal composition,” ZEEE Trans. Pown A p p . Sysf., vol. PAS-92,
the modal surge impedances and travel times by analogy with pp. 999-1008, May/June 1973.
single-phaseequations. For mode i(i= 1, . , M ) , the fol- H. Karrenbauer, ‘Ausbreitung von Wanderwellen bei verschiedenen
WMMEL AND MEYER: COMPUTATION OF TRANSIENTS 993

AnordnungenvonFreileitungenimHinblickaufdieForm der transformermodels for the solution of wavy propagation in dis-

EinschwingspannungbeiAbstandskurzschliissen”(propagation of tributed-parameter systems,” ZEEE Trans. Power A p p . Syst., vol.
traveling waves for various overhead line tower configurations, with PAS-90, pp. 1000-1006, May/June 1971.
respect to the shape of the transient recovery voltage for short line [32] A. Hauspurg, G.S. Vassell, G. I. Stillman, J. H. ChaLkow, and J. C.
faults), Ph.D. dissertation, Munich, Germany,1967. Haahr, “Overvoltages on the AEP 765-kV system, ZEEETrans.
[20] A. Clerici and M. Nogarole, “Influence of linetranspositionson Power A p p . Syst., vol. PAS-88, pp. 1329-1342, Sept. 1969.
reenergization overvoltages-Number of operationsfor
reliable [33] A. Cleria, “Analog and digital simulation for transient overvoltage
overvoltage statisticaldiatributions,” ZEEETrans. Power A p p . determinations,” Elcctra, no. 22, pp. 111-138, May 1972 (published
Syst., vol. PAS-92, pp. 25-30, Jan./Feb. 1973. for CIGRE members only).
[21] S. Virmani, D. K. Reitan, and A. G. Phadke, “Computer analysis of [34] A. C. Legate, “Comparison of field switching surge measurements
switching transients on transmission lines-Single lines and lines on a withtransient networkanalyzermeasurements,” ZEEE Trans.
common right-of-way,”ZEEE Trans. Power A p p . Syst., vol. PAS-90, Power A p p . Syst. vol. PAS-90, pp. 1347-1354, May/June 1971.
pp. 1334-1346, May/June 1971. [35] F. Blanchon, C.Dubanton, and J. Roux, ”DCtermination numerique
1221 J. R.Carson, “Wave propagation in overheadwireswithground des
surtensions d’enclenchement dansun
reseaucomplexe de
return,” Bell Syst. Tech. J . vol. 5, pp. 539-554, 1926. transport d’energie” (digital computation of switching surge over-
[23] A. Budner, “Introduction of frequency-dependentlineparameters voltages in a complex power transmission network), E.D.F. Bulktin
into an electromagnetic transients program,” ZEEE Trans. Power dela Direction des Eludes et Recherches, Sene B, no. 3, pp. 73-80, 1971.
A p p . Syst., vol. PAS-89, pp. 88-97, Jan. 1970. [36] C. A. F. Cunha and H. W. Dommel, “Computer simulation of field
(241 J. K. Snelson, ”Propagation of travelingwavesontransmission tests on the 345-kV Jaguara-Taquaril line,” presented at the 2nd
lines-Frequencydependentparameters, * ZEEETrans. Power Nat. Seminar on Generation and Transmission of Electric Power,
A p p . Syst. vol. PAS-91, pp. 85-91, Jan./Feb. 1972. Belo Horizonte, Brazil, Sept. 1973.
[ 2 5 ] W. S.Meyer and H. W. Dommel, “Numerical modelingoffrequency- [37] M. H. He=,“Electromagneticandelectrostatictransmission-line
dependent transmission-line parameters in a n electromagnetic parameters by digital computer,” ZEEE Trans. Pown A p p . Syst.,
transientsprogram,” presented a t the IEEE 1974 Winter Power vol. 82, pp. 282,291, June 1963.
Meeting, New York, N. Y., Jan. 27-Feb. 1. Paper T 74 080-8. [38] F. Pollaczek, “Uber das Feld einer unendlich kngen wechselstrom-
[26] H. W. Dommel,“Nonlinearandtime-varyingelementsindigital durchfloseenenEinfachleitung”(aboutthe field of an infinite ac
simulation of electromagnetic transients,”ZEEE Trans. Pc-wer APP. exated single-phase line), Elek. Nachrichtmtech., p. 339, 1926.
Syst. vol. PAS-90, pp. 2561-256T, Nov./Dec. 1971. [39] R. Pelissier, ‘Propagation des ondes ClectromagnGtiques guid6es par
[Zf] E. R. Taylor and S. M. Merry, “Switching surge duty on modern unelignemdifilaire”(propagation of electromagneticwaveson
arresters,” ZEEE Trans. Power A p p . Syst., vol. PAS-90, pp. 1 1 0 3 multiconductorlines), Rev. Get. Elcc., vol. 78, pp. 337-352, Apr.
1111, May/Juw 1971. 1969, pp. 491-506, May 1969.
[28] G. Carrara, A. Cleria, G. Mazza, and A. Taschini, “TNAand [40] J. G. F. Francis, “The QR transformation,” C m p u t . J.,vol. 4, no.
laboratyy study of the switching-surgebehavior of lightning ar- 3, pp. 265-271, 1961; and vol. 4, no. 4, pp. 332-345, 1962.
resters, ZEEETrans. Power A p p . Syst., vol. PAS-88, pp. 1449- [41] L. M. Wedepohl, ”Application of matrix methods to the solution of
1462, Oct. 1969. traveling-wave phenomena in polyphase systems,” Proc. Znst. Elcc.
[29] D. P. Carroll, R. W. Flugum, J. W. Kalb, and H. A. Peterson, “A Eng., vol. 110, pp. 2200-2212, Dec. 1963.
dynamicsurgearrester model for use in power systemtransient [42] A. SemlyCn and E. Wagner, “Beitrag zum genaueren Berechnen der
studies”(withadiscussion by G. Karady), ZEEETrans. Power Schaltspannungen in Hochspannungsleitungen nach dem Bergeron
A p p . Syst., vol. PAS-91, pp. 1057-1067, May/June 1972. Verfahren” (contribution to themore accurate calculation of switch-
[30] R. T. Byerly and T. M. McCauley, “Mathematical models and com- ingovervoltages in highvoltagelinesusingBergeron’smethod),
putingtechniques forgeneratorandtransmissionlinevoltages Elektrotech. 2.Ausg. A . , vol. 90, pp. 436440, Aug. 1969.
following load rejection,”in Proc. 6th ZEEE PICA C m f . , May 1969, [43] T. F. Lauber,“Determiningmaximumovervoltagesproducedby
pp. 410-425. EHV circuitbreakerclosure,” ZEEETra*s. Power A p p . Syst.,
[31] W. E. Feero,J. A. Juves, and R. W.Long,“Circuitbreaker and vol. PAS-87, pp. 10261032, Apr. 1968.