IEEE Transaction on Power Apparatus and Systems, Vol. PAS-103, No.

8, August 1984 2043

CALCULATION OF THE SENSITIVITY OF POWER CABLE AMPACITY
TO VARIATIONS OF DESIGN AND ENVIRONMENTAL PARAMETERS

M.A. El-Kady, Senior Member, IEEE

Ontario Hydro, Toronto, Canada

Abstract The problem of evaluating power cable resistivity and ambient temperature. However, the
loading performance subject to variations of the ther- thermal model of [5] was based on an approximate analy-
mal circuit parameters is formulated and solved by a sis [1] which suits only those applications in which
fast sensitivity technique based on the finite element detailed thermal circuit structure and boundary condi-
method. The technique provides useful sensitivity tions are not essential. Therefore, the study has been
information which can be accessed and manipulated by continued in Ontario Hydro and has resulted in the
the cable designer or operator to simulate, in a simple development of a sensitivity methodology based on the
and efficient manner, various soil, ambient, loading finite element model in which various cable environ-
and boundary variations. The sensitivity technique has mental parameters are modelled accurately including the
been applied to a variety of practical cable systems heat transfer mechanisms at boundaries, backf ills, duct
including direct-buried and pipe-type cables as well as banks and other media surrounding the cable. The study
cables in duct banks. The technique reduces signifi- included the consideration that many utility engineers
cantly the analysis and computations via identifying may not have the computer resources necessary to carry
the important parameters which most affect the ampacity out a complete finite element analysis each time a
and it simplifies the thermal model by eliminating the change in the thermal circuit is simulated. In addi-
low-sensitivity (non-important) parameters. Because tion, the technique applies both to the design phase
the sensitivity values depend on the particular cable and to the operational aspects of power cables buried
system and soil and boundary conditions and can vary in complex media of soils, heat sources and sinks and
significantly from one situation to another, the gener- variable boundary conditions. The sensitivity tech-
alization of the results is not possible. Therefore, nique stores, in a compact, factorized form, the finite
the paper describes the sensitivity technique and its element tableau of a given cable system at its nominal
direct implementation by illustrating its use for a status (where thermal parameters assume base-case
variety of cable systems. values). Using stored sensitivity coefficients, the
designer and/or the operator can obtain results for a
INTRODUCTION range of parameter variations without repeating the
thermal analysis of the cable. Because the simulation
Over the past 25 years, various numerical tech- procedure for various parameter variations does not
niques have been developed to calculate the permissible involve repeated finite element analyses, the results
loading of a power cable subject to a maximum cable can be obtained by a very fast and compact scheme and
temperature determined by the thermal characteristics displayed in an efficient, straightforward manner.
of its insulation. The maximum cable ampacity is a Since only one nominal finite element analysis is
function of all internal and external cable system required, larger finite element grids can be employed
components which comprise the thermal circuit of the allowing even more accurate modelling.
cable and its boundaries. To date, a variety of compu-
tational procedures are available which represent the This paper describes the analytical and computa-
thermal circuit with various levels of complexity tional aspects of the sensitivity methodology and
starting from the simple, fast but approximate heat presents illustrative results for a variety of cable
transfer models [1,2] to the rigorous, more accurate systems.
finite element models [3,4]. In conventional tech-
niques, the cable ampacity is based on assumed fixed FORMULATION
values of cable thermal circuit parameters including
soil resistivity and heat transfer coefficients at the In this section a brief summary of the analytical
boundaries. The parameters of the thermal circuit of background of the sensitivity evaluation in the finite
power cable are subject to geographical and seasonal element model is given. The derivations associated
changes which affect the allowable loading level of any with the finite element tableau formulation are summa-
particular cable. Therefore, sensitivities of the rized in the Appendix to provide a complete mathemati-
cable ampacity with respect to fluctuations in cable cal formulation and to facilitate direct implementation
circuit parameters are needed to assess the effects on of the technique.
the permissible cable loading caused by these fluctua-
tions without repeating the whole thermal analysis for Finite Element Model
each possible parameter change.
The thermal field in the cable medium is governed
The concept of cable ampacity sensitivity analysis was by the differential equation
described in a previous paper [5] where expressions
were derived for first-order sensitivities of the cable
ampacity with respect to variations in soil thermal div (k grad T) = -q + C T,
aT
(1)

where T denotes the temperature at any point, k and C

represent, respectively, the thermal conductivity and
capacity, q is the heat generation per unit time and T
denotes the time. In steady state thermal analysis of
84 WM 197-0 A paper recommended and approved two-dimensional (x,y) media, Equation 1 reduces to
by the IEEE Insulated Conductors Committee of the
IEEE Power Engineering Society for presentation at a (k a) + a (k iy)
the IEEE /PES 1984 Winter Meeting, Dallas, Texas, ax ay
= -q . (2)
aX ay
January 29 - February 3, 1984. Manuscript submit-
ted September 1, 1983; made available for printing For any homogeneous region of a given thermal
November 18, 1983.
conductivity and heat generation rate, Equation 2 can
0018-9510/84/0800-2043$01.00© 1984 IEEE
2044
be solved for the temperature at any point (x,y) in the parameters. For example, for a typical triangular
region subject to specified boundary conditions. The element of the finite element mesh whose vertices are
power cable thermal circuit includes various regions of i, j and m, the entries of the matrix H at rows and
complicated shapes having different values of thermal columns corresponding to these vertices are functions
conductivity and heat generation and Equation 2 should, of the thermal conductivity of the soil portion associ-
then, be solved for the overall cable medium. The ated with element ijm. In the finite element analysis,
finite element method exploits the theory that the the following cable parameters are independent and can
solution of (2), namely T(x,y) is that which minimizes be freely specified
the functional
(i) The cable current (loading) I in amperes.
(ii) The ambient temperature Ta at all or a part of
0F = Jrt yL(aT) + (T|) ] -qT)dxdy (3) soil surface in 0C.
(iii) The thermal conductivity k of any portion of the
whose Euler-Lagrange differential equation [6,71 is cable circuit (including backfill, duct bank) in
given by (2). This theory converts the direct solution W/(°C m).
of (2) into the minimization of (3). (iv) The heat flux coefficient Q at certain circuit
boundaries in W/m2.
Unlike the finite difference method which approxi- (v) The heat convection loss coefficient a at
mates the differential Equation 2 by discretization, certain circuit boundaries in W/(mLO C).
the finite element method approximates the solution of
(3), hence, of (2). In doing this, the cable medium is Denoting by p any of the above parameters, it is
partitioned into small elements, normally triangles, shown in the Appendix that the sensitivity vector of
forming a mesh as shown in Fig. 1. The minimization of the node temperatures Tv with respect to parameter p
(3) is performed over the finite element mesh is the solution of
yielding [3,41 a set of sparse linear equations of the
form aT
H
-v ap
v
b
-P (6)
H T = b , (4)
where H is the heat conductivity matrix, T is a vector where b is a right-hand-side vector which depends on
of temperatures at the finite element mesh nodes and b the parameter p under consideration. The various
,is a vector which normally contains the heat generation structures of (6) are given in the Appendix. The
associated with each node. Both the matrix H and the sensitivity vectors (aTV/ap), evaluated for various
vector b are adjusted to accommodate the boundary cable parameters, can be used to assess relative
conditions of the thermal circuit. The final structure effects of parameter variations on cable circuit
of H and b of (4) are given in the Appendix. Once the temperatures.
coefficient matrix and the right-hand-side of (4), are
formulated for given cable loading, circuit parameters Sensitivity Manipulation and Ranking
and boundary conditions, the equations can be solved
for the unknown node temperatures. The coefficient matrix of (6) is the same as (5).
In practical application, a decomposed form [8] of Hv
(by triangular factorization) is calculated only once
and stored as basic information for the cable system.
Using this decomposed form, the solution of (6) is
obtained by simple forward and backward substitutions
for any defined right-hand-side vector b associated
with a cable circuit parameter p. By examining these
sensitivity values for various parameters of the cable
circuit the more important parameters associated with
higher sensitivities can be identified and ranked
according to their relative effects on cable tempera-
ture and, hence, on cable ampacity.

As will be illustrated in the next section, the sensi-

tivity of a general performance index f, expressed in
terms of some or all nodal temperatures as well as some
or all cable parameters, is given by

df af aT af
(7)
Fig. 1 A sample of a finite element mesh dp aTv ap

Sensitivity Evaluation where t denotes vector transposition and (df/dp) repre-

sents the sensitivity (total derivative) of function f
As outlined in the Appendix, when the boundary with respect to parameter p.
conditions of the thermal circuit are imposed,
Equation 4 is replaced by the general form Effects of Parameter Variations

H T = b - H T (5) In addition to sensitivities of cable circuit tem-

peratures (and general performance indices) evaluated
by Equation 6 (and Equation 7), the changes of cable
where TB
is a vector containing the specif ied temper-
atures (isothermals) at some nodes of the finite
temperatures due to assigned variations in parameters
may also be required. These changes can be calculated
element mesh, contains other variable (unknown)
!v (or estimated) by exact (or approximate) methods as
temperatures, the matrices HV and Rs and the vector follows.
.v are, in general, functions of the cable circuit
 2045
First-order approximation method:
GROUND SURFACE, I-a -- 20C
The change in the temperature vector Tv re- X.
77//-77
. . . . . . . . . . . . . . . . . . . . ..

sulting from changes in some or all parameters is

,...... . . . . . . . .. . . . . . . . .
. . . . . . .....
............. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..
...
..........................................
given, to a first-order approximation, by NATIVE

t---*-----e---.-.-.*.
~~~~~~~~. . . . . . .
........................................... ...... ::::::::::::::|NAT|

aT
*-..............Lr_xu ................................................-.-.-.-.-.-.-
KF BAC
,.-...-..................................... ,,
LSI
.............................................,_

............................................ I~~~~~~~~S I
AT= Ep-V Ap . (8) 1.06 m .............................................

1
...................... .............................*

............................................. ...B.D.IN .

The first-order approximation method has the I. .-C-ABLE BED1)ING :1

advantage that it requires only the sensitivity vectors
(aTv/ap) already calculated and stored. 1.>. - @ 0.33m
However,
this method is restricted to small variations in param-
eters to maintain sufficient accuracy. For the typical
cable systems analyzed during the course of the study, I_ ~~~~~~~0. 1 8ml I
very good accuracy (±10C) was obtained for changes of 1. 246m
up to 15% in cable loading, ambient temperature, heat
transfer coeff icients and thermal conductivity of Fig. 2 Six 33 kV direct-buried cables
limited portions of the soil. The accuracy is reduced
significantly when larger variations are considered for 21 in depth. The finite element formulation exploits
m
soil conductivity. anefficient node reordering scheme which resulted in a
compact variable banded form of the coefficient matrix
Exact method: .v of (6) with half bandwidth ranging between 3 and
22.
Some cable circuit parameters, for example cable
loading I, are represented only in the right-hand-side Figure 3 shows the 50°C-isothermal contour calcu-
of Equation 5. Changes in the temperature vector Tv lated by the finite-element technique for the nominal
due to variations in these parameters can be calculated values, of the ambient temperature and soil thermal
exactly (without approximation) by defining a new right resistivities indicated above and for a cable loading
hand-side to include the new parameter values and by of 900 A per cable. The cable surface temperature Ts
solving the linear equations using the factorized at point P5 shown in Figure 3 on the surface of one
matrix Hv already available. When large changes are of the two middle cables is Ts = 64.3°C. Table I
assigned for parameters represented in the coefficient shows the sensitivities of the temperature Ts with
matrix Iv, the matrix must be reformulated and fac- respect to various parameters of the thermal circuit.
torized again.
(in)

APPLICATIONS 1.4 1.2 1.0 0.8 0.6 0.40.2 0.0 0.2 0.4 0.60.8 1.0 1.2 1.4
U.U
The temperature sensitivities with respect to var- 0.2
iations in the cable thermal parameters, as calculated BACKFILL, Pf
in the previous section, represent the foundation of a 0.4
variety of subsequent applications concerning the cable 50°C-ISOTHERM FORI 50°C-ISOTHERM
performance indices of interest to cable designers and 0.68 Pf=0.650Cm1W FOR pf=0.8Cm/W
operators. This section presents some useful applica- E~ 0.8-

tions of the sensitivity techniques to a variety of 1.0'

practical cable systems. It should be noted that the
sensitivity values tabulated in this section are appli- 1.2
cable only to the specific cable systems, soil and
boundary configurations under consideration.
1.4 1.4 \ \ ~~~~~~POINTPs|
1.6
1.6- \ < (ON CABLE SURFACE) /
Soil and Ambient Variations 1.8

The relative effects of variations of various soil

and ambient parameters can be evaluated in a fast, Fig. 3 Effect of backfill thermal resistivity on the
direct manner. Figure 2 shows the arrangement of six 50'C-isothermal contour
33 kV, direct buried 1267 mmZ copper conductor cables
having impregnated oil paper insulation and aluminum TABLE I
sheath. The cable outer diameter is 0.070 m, trench SENSITIVITIES OF CABLE TEMPERATURE
width is 1.246 m and the cable bedding extends 0.120 m (Nominal Value: T. = 64.30C)
above and below the cable surface. The cables are
buried at a distance of 1.060 m between the cable Nominal Sensitivity
center and the ground surface which represents an iso- Parameter, p Value aT/ap
thermal boundary of the cable thermal circuit. The
base-case (nominal) value of the ground surface temper- Cable bedding thermal 1.053 W/(°C m) -11.7
ature is 20°C and the base-case values of the thermal conductivity, kb
resistivities of cable bedding, trench backfill and
native soil are, respectively, 0.95°C m/W, 0.8°C m/W
and 1.2°C m/W and they are all subject to variations Backfill thermal 1.250 W/(°C m) -18.2
about their nominal values. For this cable system, we conductivity, kf
wished to determine the 50°C-isothermal contour (which
has some significance concerning the soil dry-out char- Native soil thermal 0.833 W/(°C m) -9.8
acteristics) and to evaluate the sensitivities of the conductivity, ks
cable surface temperature with respect to variations of
soil thermal resistivities. In the thermal analysis, a
finite element grid containing 1040 nodes was created Ground surface 20.0°C 1.0
which extends 28 m on each side of the cable trench temperature, T
(only half of the grid is used because of symmetry) and
2046
The sensitivity values of Table I show that varia- shows the 15°C, 25°C, 30°C and 40°C-isothermal contours
tions of the backfill thermal conductivity will have calculated by the finite element technique and Table II
the greatest effect on the cable temperature. The shows the sensitivities of the pipe temperature at
value of aTs/akf, which for this example denotes point Pe shown in Figure 5 with respect to variations
the sensitivity of the middle cable temperature T. in various soil and boundary parameters.
with respect to the backfill thermal conductivity kf
is -18.2oC2 m/W compared to the value of aT5/aks TABLE II
-9.8oCL m/W for variations in the native soil thermal
conductivity ks. In addition to showing the relative SENSITIVITIES OF PIPE TEMPERATURE
effects of variations of various parameters, the sensi- (Nominal Value: Te - 33.3°C)
tivity values of Table I can be used to estimate
(approximately) the changes in the cable temperature
T. due to variations of one or more of the parameters Nominal Sensitivity
as described in the previous section. For example an Parameter, p Value aTe /ap
increase in kf by 0. 288 W/°C m from the nominal value
of 1.250 W/(°C m), which is equivalent to reducing the Limestone thermal 2.5 W/(°C m) -8.2
thermal resistivity pf from 0.8 to 0.65°C m/W, would conductivity, k1
result in a change in the cable temperature Ts by
(-18.2) x (0.288) -5.2°C approximately, that is
=

Ts = 64. 3 - 5. 2 = 59. 10 C. Figure 3 shows the effect Clear stone thermal 1.0 W/(OC m) -0.13
of reducing the thermal resistivity of the backf ill conductivity, kd
from 0.8 to 0.65°C m/W on the 50°C-isotherm.
Concrete wall thermal 2.0 W/(°C m) -1.10
Variations of Complex Boundaries and Soil Parameters conductivity, kc
The finite element sensitivity technique can be
applied to complex cable configurations and boundaries Native soil thermal 0.67 W/(°C m) -0.92
to assess the effects of variations of various soil and conductivity, ks
boundary parameters. Figure 4 shows two 115 kV/
633.4 mm2 single circuit pipe-type cables buried in a Convection heat 5.0 W/(°C m; ) -0.0077
coef f icient, a
GROUND SURFACE, Ta = 200C
(ISOTHERMAL Ground surface 20.0°C 0.68
3.0m 1.2m temperature, Ta
............. .. .-.--
... LMTOE..............
- .w.*.... ..

11m...-: :;:::-:: SCREENINGS -CONCRETE The sensitivity results of Table II show that varia-
RETAIN ING tions of the thermal conductivity of the backfill
WALL (limestone screenings) would affect the cable tempera-
IX *& r : ..

ture Te more than variations of other soil portions.

NATIVE tCONVJECTIVE For example the temperature Te is nine times more
SOIL ........
BOU INDARY sensitive to variations of limestone thermal conduc-
PERFORATED tivity than variations of native soil thermal conduc-
UK<AIN (10V-U) . n tivity. A decrease of 0.83 W/(0C m) in limestone ther-
mal conductivity which is equivalent to an increase in
CLEAR STONE thermal resistivity from 0.4 to 0.6°C m/W would result
BEDDING in an increase in Te of (-8.2) x (-0.83) 7°C
approximately. The effects of other parameter varia-
tions can be estimated in a similar way using the
Fig. 4 Two pipe-type cables in a complex medium
sensitivity values of Table II.

complex medium of soil and boundaries. A 150 mm dia- Calculation of Maximum Cable Ampacity
meter perforated sub-drain (average temperature of
10°C) in a clear stone bedding is located in the vicin- Unlike the traditional Neher-McGrath analysis [11
ity of one cable circuit. A concrete retaining wall is which calculates the maximum cable ampacity for a given
located on one side of a cable circuit and its outside maximum conductor temperature, the finite element
represents a convective boundary with heat convection analysis calculates the temperature for a given cable
loss coefficient of 5.0 W/(0C m2). The ground surface current. Therefore, the actual conductor temperature
represents an isothermal boundary of 20°C. Figure 5 associated with a given cable loading I cannot be known
until a finite element analysis is completed; the maxi-
mum ampacity Imax which produces a specified maximum
conductor temperature may conventionally be obtained by
a trial and error procedure which may require many
finite element analyses. Alternatively, the sensitiv-
ity technique can be employed for a much faster and
more efficient computation, which does not require any
a priori information except the temperature sensitivi-
ties already available.

Figure 6 shows two 132 kV/1267 mm1 copper conduc-

tor, pipe-type cables buried in one trench of backfill
thermal resistivity pf = 0.6°C m/W. The native soil
thermal resistivity is 1.2°C m/W and the ground surface
represents a natural convective boundary with a heat
convection loss coefficient of 5.0 W/(m ° C) and the
Fig. 5 Isothermal contours ambient temperature is 250C. The maximum conductor
 2047
CONVECTIVE BOUNDARY,a= 5.0W /(m2 oC) For the purpose of checking the accuracy, the
finite element analysis was repeated for a cable
loading of 883 A instead of 850 A (this requires only
an adjustment of the right-hand-side of the equations
while the same factorized coefficient matrix is used).
The average temperature at the conductor surface was
85.1°C (compared with 85°C above) indicating very good
accuracy of the sensitivity analysis. However, it
should be noted that the change in the cable loading
was only 33 A (or 3.9%). For larger changes, the
accuracy of the use of first-order sensitivities is
reduced. A subsequent study performed for the cable
system under consideration showed that the accuracy
remained within +0.5°C for changes in cable loading of
up to 100 A.

Fig. 6 Two 132 kV pipe-type cables Sensitivities of Equivalent Thermal Resistance

The total equivalent thermal resistance Re per
temperature is specified as 85°C. Since the conductor unit length between the cable conductor and a deMined
temperature is required in the ampacity calculation, ambient is a very important quantity which is used by
the, finite element grid is formulated to cover the utility engineers for various ampacity calculations.
internal cable media in some detail where the conductor This quantity is defined as
and insulation zones are represented (for each of the
three cables) by their individual thermal resistivities T - T
and amounts of heat generated. Also, the gas zone in R = c a
(10)
each pipe is represented by a conductive medium of
thermal resistivity of 36°C m/W. The temperature value
eq n I2 R
ac
Tc = 80.7°C at a point on the conductor-insulation
interface of one cable was calculated in the finite where n denotes the number of conductors carrying a
element analysis based on a nominal (base-case) current load current I each and Rac is an equivalent ac
of I = 850 A per cable. Table III shows the sensitivi- electrical resistance [1] which may include factors to
ties of the conductor temperature Tc with respect to accommodate losses in various parts of the thermal
the cable current I as well as other parameters of the circuit.
thermal circuit.
In the present study, as would be typical in most
TABLE III cable systems, information concerning the change of
Req due,to various parameter variations is useful for
SENSITIVITIES OF CONDUCTOR TEMPERATURE the utility engineer. especially when power cables are
-(Nominal Value: Tc = 80.70C) buried in complex media with complicated boundary con-
ditions. Such information becomes even more attractive
if, as in this analysis, no additional calculations are
Nominal Sensitivity necessary; only the already available sensitivities are
Parameter, p Value aT c /ap needed. Since the temperature at the conductor Tc of
(10) is a function of both Ta and I as well as other
Cable current, I 850 A 0.132 soil and boundary parameters, the expected change in
Req due to a change Ap in any parameter p is given by
Backfill thermal 1.67 W/(°C m) -32.43
conductivity, kf aR I aT
AReq (a +- ac2
22 c)) APp - (11)
n I :R
Native soil thermal 0.83 W/(°C m) - 9.60 ac
conductivity, k
Figure 7 shows two circuits of 115 kV/633.4 mm'
Convection heat 5.0 W/(0C m'-) -0.0043 double layer segmental copper conductor, LPOF cables in
coefficient, ac a concrete duct bank. The duct bank contains twelve

Am-bient temperature, Ta 25.0C 1.0 GROUND SURFACE, Ta 200C

i!-_..._-,-_,...,,,_.,,
, I , -,,&,~p
."- _ ! _ o _ i //fms77Mr-/=737-
_,- ,, #, -

NATIVE SOIL
From Table III, the sensitivity of the conductor
temperature Tc with respect to the cable loading is Ps = 1.3 OC m/W GROUND CABLE
aTC/aI 0.132°C/A evaluated at the nominal values CONCRETE
of I = 850 A and Tc = 80.7°C as obtained before.
Hence, the maximum ampacity Imax associated with Pc= 0.60C m/W
Tc(max) = 85°C is obtained approximately from the
first-order differential expression, 0.1m TRANSITE
DUCT m
aT
T
c(max) -(I max -850)
-Tc al (9) DUCT #4
or
85 - 80.7 0.132 (I max - 850)
which yields I max = 883 A. Fig. 7 Two cable circuits in a duct bank
2048
ducts, four of them are empty and two ducts contain TABLE V
ground cables. The thermal resistivity of the concrete
is 0.60C m/W (nominal value) and of the native soil is COMPUTER REQUIREMENTS FOR VARIOUS EXAMPLES
1.3°C m/W. The ground surface represents an isothermal
boundary of 20°C. The conductor temperature value
Tc = 7800 was calculated by the finite element analy- No of
sis for the cable in duct #4 for a nominal value of the Example grid CPU Memory
current I - 780 A. The equivalent thermal resistance nodes (s) (M bytes)
calculated by Equation 10 is Req 2.580C m/W.
Table IV shows the sensitivities of Req with respect 6 direct-buried cables 1040 11 0.058
to variations in some thermal circuit parameters. 2 pipe cables in complex medium 2170 24 0.143
2 pipe cables in backfill 1740 20 0.108
TABLE IV 2 circuits in a duct bank 2630 31 0.174
SENSITIVITIES OF EQUIVALENT THERMAL RESISTANCE
(Nominal Value: Req = 2.580C m/W) CONCLUSIONS
The work presented in this paper was intended to
Nominal Sensitivity provide the utility engineer with an easy and fast sen-
Parameter p Value aR lap sitivity methodology, based on the finite element
.eq. model, to assess the cable thermal performance subject
to variations in the cable thermal circuit parameters.
Native soil thermal 0.77 W/1(°C m) -00.57
The sensitivity methodology suits both power cable
conductivity, k
design and operation applications. The sensitivity
information is useful not only in evaluating the cable
Duct concrete thermal 1.67 W/(°C m) -0.39 ampacity subject to various parameter- changes but also
conductivity, k c
in determining the important and non-important param-
eter variations in terms of their relative effects on
Ground surface 200C 0.0
the cable temperature and ampacity. The applications
temperature, T
of the technique to a variety of cable systems have
a shown that the sensitivity values associated-with vari-
ous parameters depend on the particular cable system
and soil and boundary conditions and can vary from one
The sensitivity results of Table IV show that a situation to another. Nevertheless, the sensitivity,
change in the ambient temperature does not affect technique is general and can be applied to any cable
Req. Variations in the native soil thermal conducti- system configuration. As demonstrated in the paper,
vity k5 have bigger effects on Req than variations the technique reduces significantly the amount of
in the concrete thermal conductivity kc. The sensi- analysis and computations and it avoids numerous finite
tivity aReq/ak8 is -0.57 (°C m/W)2, therefore a de- element analyses required to simulate various parameter
crease in the native soil thermal conductivity k5 by changes.
0.21, from k5 0.77 to k5 = 0.56 W/(°C m), which
=

is equivalent to an increase in thermal resistivity When first-order sensitivities are employed to

from ps = 1.3 to p. = 1.8°C m/W would increase the predict the change in a cable performance index due to
value of Req by (-0.21) (-0.57)
x 0.12°C m/W, assigned variations in thermal parameters, better
approximately, that is Req = 2.70°C m/W. accuracy is obtained when the assigned parameter
changes are relatively small. However, it was observed
COMPUTER REQUIREMENTS that for almost all cable systems considered during the
course of the study, the maximum error in the conductor
The computer resources (computational time and temperature remains within ±1.OC for changes in the
memory) required to carry out the sensitivity analysis thermal parameters discussed in the paper of up to
of a particular cable system depend mainly on the size 15%. Because the exact finite element model is always
of the -finite element grid. Complex cable systems with simulated the only source of error is due to the first-
various soil portions and boundary surfaces require order approximation. However, such an error may be
larger finite element grids to attain better modelling significant especially when large changes in soil and
accuracy. The sparse matrix algorithms employed by the backfill thermal resistivity are considered. In this
sensitivity technique reduce significantly the computa- case a new finite element analysis may be necessary.
tional time and storage required which are approxi- On the other hand, exact results can be obtained for
mately proportional to the number of nodes of the large variations in some parameters such as the cable
finite element grid. Table V shows typical values of loading by performing only forward and backward substi-
computational time (in CPU-seconds) and storage (in M tutions for the coefficient matrix as described in the
bytes of memory) required for both the finite element paper. In general, only one finite element analysis
solution (to calculate base-case temperatures) and the associated with the nominal cable conditions is re-
subsequent sensitivity analysis of various examples quired, hence, a very large finite element grid may- be
considered in the previous section. The calculations analyzed using advanced matrix decomposition techniques
were performed on a UNIVAC-1100 computer at Ontario to attain even better modelling accuracy. The material
Hydro. The matrix factorization of the last example (2 presented in this paper concerns the steady-state ther-
cable circuits in a duct bank) was performed on a mal analysis. The concepts stated, however, represent
CYBER-170 computer at McMaster University. the basis for possible applications to the cyclic and
transient loading conditions provided that the appro-
priate forms of the finite element tableau are
For small and medium sized finite element grids, formulated.
the computerized sensitivity algorithm can be imple-
mented on small desktop microcomputers (for example, ACKNOWLEDGEMENTS
IBM-8086-CP/M) with sufficient accuracy of results.
The author would like to acknowledge the help and
support provided by the Canadian Electrical Association
(CEA) which is sponsoring the developments of the
 2049
sensitivity technique (Contract #138D-375). The author
also thanks other departments of Ontario Hydro and f x
outside clients for their contributions in supplying [x m -
X.]
i
data information and in requesting specific cases of b
study. The contributions by the Stations and
a = xi - xm = YmF Yml (A2)
Transmission Design Department and Canada Wire and - x
x.
Cable Limited are particularly appreciated. The help ii Yi - yi
i

provided by Mr. R. Ganton, Mr. D. Horrocks and

Mr. N. Tata during the course of developing the soft-
ware is gratefully acknowledged. The matrix factoriza- In formulating the overall heat conductivity
tion of some examples of the study was performed on a matrix H, the entries of the various matrices He must
CYBER-170 Computer at McMaster University and was be summed in the appropriate ijm rows and columns of
supported by the Natural Sciences and Engineering H. When a constant heat flux Qe per unit surface or
Research Council of Canada. a convection loss with coefficient ae is assumed at
edge ij of a boundary element e, the entries Hii and
REFERENCES H-; of the matrix H are increased by aede/3, de
denoting the edge length and the entries Hij and
[1] J.H. Neher and M.H. McGrath, "The calculation of Hji are increased by aede/6.
the temperature rise and load capability of cable
systems", AIEE Trans. (Power Apparatus and right hand side vector b of Equation 4 is
The
Systems), Vol. 76, 1957, pp. 752-772. formulated in a similar ordered way by summing (in
appropriate ijm locations) the 3-component vectors be
[2] L.H. Fink and J.J. Smerke, "Control of the thermal associated with various elements e of the form
environment of buried cable systems, Part II",
AIEE Trans. (Power Apparatus and Systems),
Vol. 77, 1958, pp 161-168.
[1A
eA
(A3)
[31 N. Flatabo, "Transient heat conduction problems in b-e 3

power cables solved by the finite element method",

IEEE Trans. Power Apparatus and Systems,
Vol. PAS-92, 1973, pp. 56-63. where qe is the heat generated within element e.

[4] M.A. El-Kady, "A purturbed finite element model In the case of a boundary heat flux Qe or a
for sensitivity calculations of thermal fields", convection loss at edge ij, the entries i and j of b
SIAM National Meeting (Denever, CO, June 1983). are increased by de( eTa + Qe)/2 where Ta
denotes the ambient temperature associated with the
[5] M.A. El-Kady, "Optimization of power cable and convection loss through the element edge ij.
thermal backfill configurations", IEEE Power
Apparatus and Systems, Vol. PAS-101, 1982,
pp 4681-4688. Finite Element Sensitivities
[61 O.C. Zienkiewicz, The Finite Element Method. New When isothermal boundaries with specified tempera-
York: McGraw-Hill, 1977. tures at certain edge points are considered, Equation 4
must be processed further into the appropriate form
[7] A.R. Mitchell and R. Wait, The Finite Element
Method in Partial Differential Equations. New
(5), namely
York: Wiley, 1977.
H T = b H T
[8] K.J. Bathe and E.L. Wilson, Numerical Methods in -v -v -v
-
-s -s' (A4)
Finite Element Analysis. New Jersey: Prentice-
Hall, 1976. where T5 is the vector of specified node tempera-
tures, _Iv contains other variable (unknown) tempera-
tures, iv and Hs are general heat conductivity
matrices, adjusted to accommodate other boundary condi-
tions as described before.
APPENDIX
The vectors and matrices of (A4) are, in general,
DERIVATION OF SENSITIVITY EXPRESSIONS functions of the cable circuit parameters. In the
following derivations a typical finite element model is
Finite Element Tableau considered to illustrate the sensitivity derivation.
Similar procedures can be followed for other, more
The heat conductivity matrix of Equation 4 is complicated models as well as for other cable circuit
formulated [3,51 by an ordered sum of small 3 x 3 parameter variations.
submatrices associated with various triangular elements
e in the finite element mesh of the form
(i) Sensitivities of cable loading:
ke t t
He 4Ae (aeae bee) (Al) The cable current I appears in those entries of
the right-hand-side vector bv which correspond to the
finite elements containing heat generation due to the
where t denotes transposition of vectors and Ae and cable current I (see Equation A3). The heat generation
ke are, respectively, the area and thermal conduc- rate is proportional to I4 with proportionality factor
tivity associated with element e of vertices i,j ,m. depending on the size of the element and its electrical
The values of a and b depend only on the x,y dimensions properties. Equation A4 can then be written in the
of the element and are given by form
2050
which can be solved for the required temperature sensi-
tivities. Note that k can be associated with any part
-I2 r + c in the cable thermal circuit (for example, backfill,
Hv Ts - _ = (A5) protective layer, native soil, etc) by appropriate
.~vIv
[I.r Jb definition of the matrices RV, Cv, R& and C' of
(A9).
where r is a proportionality vector (assumed to be
independent of temperature for simplicity of calcula- (iv) Sensitivities of boundary heat flux coefficient:
tions ) and c is constant vector, hence
When some boundary conditions of a specified heat
flux, Q are considered, the corresponding elements .of
aT2 I r the vector by constitute linear functions of Q. Let
H -v L (A6)

L -I
Q r + c
(All)
bh r
Using the already stored factorized form of Hv, the L -°o-
sensitivity- of the temperatures at any point with
respect to the cable loading can be calculated. hence, from (A4)

aT-v r]
(ii) Sensitivities of ambient temperature: H -

The ambient temperature Ta represents a part (or

-v 3Q [.a (A12)

all) of the specified temperature vector T, of (A4).

It also appears in a few entries of the vector bv
associated with heat convection loss as described which provides the temperature sensitivities with
before. We write respect to variations in parameter Q.

(v) Sensitivities of boundary heat convection

[a -:] (A7) coefficient:-
-bv = Tar- + C

The heat convection loss coefficient a at a speci-

fied boundary is implied, in general, in both matrices
where u is a vector of unity components and r is essen- Hv1 and Hj as well as in the vector bv of (A4).
tially zero vector except for a few entries associated We write
with heat convection boundaries. Then, from (A4)

-v
Hv
aT
S r

from which the required sensitivities are calculated.

[~] (A8) H
-V
=

then, from
R + -V
a -V

(A4)
H
C , -s = C , -V
R + -s
a -S b = c
r +-
a- (A13)

aT (A14)
T + -r R T
(iii) Sensitivities of thermal conductivity: R-v-v -
Hv Da
-
S-
-

Let k be the thermal conductivity of certain portion of

the cable system. In general, a few entries of the which yield the temperature sensitivities with respect
matrices Hv and H. of (A4) would depend on k in a to changes in the heat transfer coefficient
at the
linear relationship as indicated by Equation. Al. We boundaries. -
write

H = R + -v
k -v H
C , -s = R +0
k -8 -s (A9)
-v

where Rv and Rs are mostly zero matrices with a few

non zero elements in certain locations. Then from (A4)

aT
H - = - -Rs- T
R T-V (AIO)
-V ak -V 8S