Professional Documents
Culture Documents
Barrier Optimization Algorithm Applied To Calculation of Optimal Loading of Dissimilar Cables in One Trench
Barrier Optimization Algorithm Applied To Calculation of Optimal Loading of Dissimilar Cables in One Trench
⎣ 1
RT + nR (1 + λ )T + nR (1 + λ + λ )(T + T4 ⎦
) Ip = ⎢
1 2 1 2 3
⎥ (5)
where ⎣ RT1 + nR(1 + λ1 )T2 + nR (1 + λ1 + λ2 )(T3 + T4 ) ⎦
∆θ = θ c − θ amb , the latter representing ambient temperature. with all parameters in this equation pertaining to cable p.
Wd (W/m) represents dielectric losses per unit length.
λ1 , λ2 are the sheath and armor loss factors.
R is the ac resistance per unit length of the conductor at
maximum operating temperature (ohm/m).
T1 , T2 , T3 and T4 are the thermal resistances where T1 is the
thermal resistance per unit length between one conductor
and the sheath, T2 is the thermal resistance per unit length
of the bedding between sheath and armor, T3 is the thermal
resistance per unit length of the external serving of the
cable, and T4 is the thermal resistance per unit length
between the cable surface and the surrounding medium,
(Km/W).
The expression for the external thermal resistance of an Fig. 1 Cables buried underground and their images
isolated cable is given by with respect to the earth’s surface
T4 =
s
2π
ρ
ln u + u 2 − 1( ) (2) This calculation is performed for all cables in the group, and
is repeated where necessary to avoid the possibility of
where ρ s (Km/W) is the thermal resistivity of the soil and overheating any of the cables.
u = 2 L / De with De = external diameter of the cable, (mm)
Let θ ep denote the external temperature of cable p in
and L = depth of burial of the centre of the cable, (mm).
isolation. Substituting (3) into the right-hand side of (4) and
Standard treatment of unequally loaded applying equation (2), the following general expression for
cables the external thermal resistance of cable p is obtained:
⎛ ' ⎞
θ ep + ∆θ p − θ amb ρ s ⎜
( )
The method suggested for the calculation for ratings of a 1 q d pk ⎟
group of cables set apart is to calculate the temperature rise T4p =
Wp
=
2π ⎜⎜
ln u + u 2 − 1 + ∑
W p k =1
Wk ln
d pk ⎟⎟
at the surface of the cable under consideration caused by ⎝ k≠ p ⎠
the other cables of the group, and to subtract this rise from (6)
the value of ∆θ used in the equation (1) for the rated When a group of identical, equally loaded cables is
current. An estimate of the power dissipated per unit length considered, the computations can be much simplified. In
of each cable must be made beforehand, and this can be this type of grouping, the rating of the group is determined
subsequently amended as a result of the calculation where it by the ampacity of the hottest cable. It is usually possible to
becomes necessary. decide from the configuration of the installation which cable
will be the hottest, and to calculate the rating for this one. In
The temperature rise ∆θ kp at the surface of the cable p cases of difficulty, a further calculation for another cable may
be necessary. The method is to calculate a modified value
produced by the power Wk watt per unit length dissipated in of T4 which takes into account the mutual heating of the
cable k is equal to group and to leave unaltered the value of ∆θ used in the
ρs
'
d rating equation (1).
∆θ kp = Wk ln pk (3)
2π d pk
' When the losses in the group of cables are equal, equation
The distances of d pk and d pk are measured from the centre
(6) simplifies to
th
of the p cable to the centre of cable k, and the centre of the
ρ ⎧⎪ ⎡⎛ d p' 1 ⎞⎛ d p' 2 ⎞ ⎛ d pk ⎞ ⎛ d pq ⎞ ⎤ ⎫⎪
( )
' '
reflection of cable k in the ground-air surface, respectively T4 = s ln ⎨ u + u 2 − 1 ⋅ ⎢⎜ ⎟⎜ ⎟ ⋅⋅⋅⎜ ⎟ ⋅⋅⋅⎜ ⎟⎟ ⎥ ⎬ (7)
(see Fig. 1). Thus, the temperature rise ∆θ p above ambient 2π ⎪
⎩ ⎢⎣⎜⎝ d p1 ⎟⎜ ⎟ ⎜ ⎟ ⎜
⎠⎝ d p 2 ⎠ ⎝ d pk ⎠ ⎝ d pq ⎠ ⎥⎦ ⎭⎪
th
at the surface of the p cable, whose rating is being There are (q-1) factors in square brackets, with the term
'
determined, caused by the power dissipated by the other (q- d pp
1) cables in the group, is given by excluded.
d pp
∆θ p = ∆θ1 p + ∆θ 2 p + ... + ∆θ kp + ... + ∆θ qp (4)
with the term ∆θ pp excluded from the summation.
FORMULATION OF THE OPTIMIZATION
PROBLEM
The value of ∆θ in (1) for the rated current is then reduced
by the amount of ∆θ p and the rating of the pth cable is This section describes the proposed solutions for optimal
loading of dissimilar cables. Let us consider the system of
determined using a value of T4 corresponding to an isolated ‘q’ cables. Example of such system is shown in Fig. 2:
Return to Session
⎡ ⎤
ρs ⎢ q ⎛ d′ ⎞
∑ ⎜ n jWdj ln d pj ⎟⎟⎥⎥
2π ⎢ j =1 ⎜⎝ pj ⎠
- ⎣⎢ j ≠ p ⎦⎥ (12)
R pT1 p + n p R p (1 + λ1 p )T2 p + n p R p (1 + λ1 p + λ2 p )(T3 p + T4 p )
In what follows we will rewrite the optimization problem (8)
without the last equality constraint, which is understood to be
enforced. The optimization problem (8) can now be rewritten
as
Minimize − I1 − I 2 − − I q
1 2 c12 2 c1q 2
subject to I1 + I2 + … + Iq ≤ 1
d1 d1 d1
Fig. 2: General system including ‘q’ cables. Some c2 q 2
c21 2 1 2
cables are equally loaded, some are unequally loaded I1 + I 2 + … + Iq ≤ 1
and some have preset fixed currents. d2 d2 d2
(13)
Let’s assume that the operating temperatures of the cables
c p1 cp2 1 2 c pq
are θ1 ,θ 2 ,...,θ q and the maximum allowed operating I12 + I 22 + … + Ip +… + I q2 ≤ 1
dp dp dp dp
temperatures are θ1max ,θ 2 max ,...,θ q max . The conductors’
temperature rises above ambient temperature are
∆θ1 , ∆θ 2 ,..., ∆θ q , where ∆θ i = θi − θ amb . Also, let’s assume cq1
I12 +
cq 2
I 22 + … +
1 2
Iq ≤ 1
that the cables are laid in a uniform medium; system of dq dq dq
cables laid in a non uniform medium can be treated by In a matrix form this can be expressed as shown in (14).
modifying the value of T4 [7]. By calculating the ampacities Solution to this optimization problem is discussed in the
of a group of cables we are trying to find the maximum Appendix.
currents that can be carried by all cables without any of the Minimize − I1 − I 2 − − I q
cable temperatures exceeding its maximum allowed value. ⎡ 1 c12 c1q ⎤
This may be expressed mathematically as an optimization ⎢ ⎥
⎢ d1 d1 d1 ⎥
problem as shown below.
⎢ c2 q ⎥ ⎡ I12 ⎤ ⎡1⎤
Maximize I1 + I 2 + + I n ⎢
c21 1 c23
⎥
θ1 ≤ θ1max , θ 2 ≤ θ 2 max ,...,θ q ≤ θ q max ⎢ d2 d2 d2 d 2 ⎥ ⎢⎢ ⎥⎥ ⎢ ⎥
⎥ ⎢ ⎥ ⎢⎢ ⎥⎥ (14)
subject to (8)
⎢
I ia = I ib = I ic ; i = 1,..., k subject to ⎢ ⎥⎢ ⎥ ≤ ⎢ ⎥
⎢ cq −1, q ⎥ ⎢ ⎥ ⎢ ⎥
where k is the number of circuits composed of single core ⎢ ⎥
cables per phase. The last constraint states that the system ⎢ d q −1 ⎥ ⎢⎣ I q 2 ⎥⎦ ⎢⎣1⎥⎦
is balanced. ⎢ cq1 cq , q −1 ⎥
⎢ 1 ⎥
⎢⎣ dq dq d q ⎥⎦
All constraints in (8) will now be modified to be expressed as
functions of the currents. For the cable p, we have
θ p ≤ θ p max and therefore, ∆θ p ≤ ∆θ p max . Thus, using (5), we APPLICATION OF THE BARRIER METHOD
can write The optimization problem (14) can be written in the following
∆θ p max − Wdp ⎡⎣0.5T1 p + n p (T2 p + T3 p + T4 p ) ⎤⎦ − ∆θ p form:
I ≤
2
(9)
R pT1 p + n p R p (1 + λ1 p )T2 p + n p R p (1 + λ1 p + λ2 p )(T3 p + T4 p ) Minimize f 0 (x)
p
(15)
Expressing Wk in (3) in terms of the current I k and then subject to f i (x) ≤ 0 i = 1,…, n
⎡ 2ci1 ⎤ ⎛1⎞
⎢ d 0 0 ⎥ Minimize f 0 (x) + ⎜ ⎟ φ (21)
⎢ i ⎥ ⎝t⎠
⎢ 0 ⎥ The algorithm used to solve this problem is described in the
∇ 2 f i ( x) = ⎢ ⎥ (17) Appendix.
⎢ 0 ⎥
⎢ 2cin ⎥
⎢ 0 0 ⎥ COMPARATIVE STUDIES
⎢⎣ d i ⎥⎦
First we observe that by definition cij is always positive. The numerical studies presented in this section will illustrate
the efficacy of the proposed algorithm. Two computational
On the other hand, d i is also positive in spite of the procedures will be used. One, based on the IEC 60287
subtraction in the numerator. The reason for this is that the iterative procedure method outlined in Section II and the
terms that are being subtracted represent the temperature other using the optimization algorithm presented in Section
increase in the conductor due to dielectric losses of the ith III. Two different cable types will be used in this analysis as
cable and the remaining cables in the same trench, which shown in Figure 3 and the computed parameters
are in the order of a few degrees, usually very small values. summarized in Table 1.
Therefore, the Hessian above is non-negative. Thus, all
inequality constraint functions are convex and the
optimization problem (14) is convex.
The optimization problem (14) is equivalent to the following
one [10]
n
Minimize f 0 (x) + ∑ I − (fi (x)) (18)
i =1
(b)
Fig 3. Cables considered in the numerical example.
(a) 10 kV-300 mm2 single core, single point bonded
2
circuit (b) 10 kV-300 mm three-core. The graphs
obtained with a CYMCAP [8] program.
Table 1 Computed parameters in the numerical
Fig. 6: Dashed line shows I− whereas the solid lines example (non zero values shown)
show Iˆ− (u ) for t = 0.5, 1, 2. A larger value of t Computed Single core Three core
provides a better approximation. parameter cable cable
T1 ( K .m / W ) 0.214 0.325
The function,
n T2 ( K .m / W ) 0.109 0.042
φ = −∑ log( − f i (x)) (20) T4 ( K .m / W ) *
i =1
0.751 0.637
is called the logarithmic barrier for problem (18). Its domain R (Ω / km) at 90 C 0.0763 0.0790
contains the set of values, x that satisfy the inequality
constraints, f i ( x) ≤ 0 , strictly, i.e. f i ( x) < 0 . *The external thermal resistance is for a single circuit in
isolation with the cable 1m underground, soil thermal
Therefore (18) becomes, resistivity of 1 Km/W and soil ambient temperature of 15°C.
Return to Session
REFERENCES
[1] Kennelly, A.E., (1893) “Minutes, Ninth Annual Meeting,
Association of Edison Illuminating Companies”, New
York, NY.
[2] Mie, G., (1905), “Über die Warmeleitung in einem
verseilten Kable”, Electrotechnische Zeitschrift, p. 137.
[3] IEC Standard 60287 (1969, 1982, 1994), "Calculation of
the Continuous Current Rating of Cables (100% load
factor)", Third edition 1994-1995.
[4] IEC Standard, (1985), "Calculation of the Cyclic and
Emergency Current Ratings of Cables. Part 1: Cyclic
Fig. 5 Six cables with two circuits with fixed current. Rating Factor for Cables up to and Including 18/30 (36)
kV", Publication 853-1.
The results are summarized in Table 3.
[5] IEC Standard, (1989), "Calculation of the Cyclic and
Table 3 Comparative results for dissimilar cables Emergency Current Ratings of Cables. Part 2: Cyclic
Rating Factor of Cables greater than 18/30 (36) kV and
IEC Iterative approach Optimization
Emergency Ratings for Cables of All Voltages",
Conductor Conductor
Ampacity Temperature Ampacity Temperature Publication 853-2.
(A) (A) [6] IEEE (1994) Standard 835, "IEEE Standard - Power
(o C ) (o C )
Cable Ampacity Tables", IEEE Press.
I1 = 200 θ1 = 56.3 I1 = 200 θ1 = 59.6 [7] G.J. Anders, Rating of Electric Power Cables, IEEE
I 2 = 564 θ 2 = 77.0 I 2 = 667 θ 2 = 88.0 Press, New York, 1997.
I 3 = 564 θ 3 = 72.0 I 3 = 667 θ 3 = 83.2 [8] CYMCAP for Windows, CYME International T&D,
I 4 = 564 θ 4 = 81.1 I 4 = 667 θ 4 = 90.0 November 2006.
[9] KNITRO 5.1 Optimization Software, Ziena Optimization
I 5 = 527 θ 5 = 90.0 I 5 = 405 θ 5 = 74.7
Inc., http://www.ziena.com/knitro.htm.
I 6 = 100 θ 6 = 53.4 I 6 = 100 θ 6 = 54.3 [10] S. Boyd and L. Vandenberghe, Convex Optimization.
Total Total Cambridge: Cambridge University Press, 2004.
2519 A 2705 A
APPENDIX
We can observe that, in this case, the optimization algorithm
gives a total ampacity which is 186A (7.4%) greater than the Algorithms for the Barrier Optimization Method
total ampacity obtained with the IEC method. Another
interesting observation is that the ampacities of cables 2 to 5
Introduction
(cables with non fixed currents) are much closer to each We recall that the optimization problem we are solving is
other for the IEC case than for the optimization case. given by (21). Since (21) is now obtained through
approximating the indicator function I − by Iˆ− then its
To test the accuracy of the algorithm presented here, a solution will be an approximation to the solution of the
commercial optimization problem solver called KNITRO [9] original problem (15). However, as t increases this
was used to find the solution to the last optimization approximate solution improves. Newton’s method [10] will
problem. The same results were obtained with both be used to solve (21). Unfortunately for very large t, the
algorithms. objective function in (21) is difficult to minimize using the
Newton’s method since its Hessian varies rapidly near the
CONCLUSIONS boundary of the feasible set. This problem can be mitigated
by solving a sequence of problems, each having the form of
This paper presented an optimization algorithm for the
(21), starting with small t and increasing it at each step. The
analysis of loading of dissimilar cables. The proposed
solution at each step is used as the initial condition for the
approach applies a Barrier Optimization method to solve an
next step. As t grows very large the approximate solution
optimization problem with linear objective function and
becomes very accurate [10].
quadratic constraints. Numerical results were obtained for
two cases, one with identical, but unequally loaded cables,
This method of approximating the indicator function by a
and the other with dissimilar cables. In each case, the
logarithmic barrier and solving the resulting optimization
ampacity computed with the new algorithm was higher than
problem through a sequence of steps is called the barrier
the results obtained using an iterative approach.
method [10].
Return to Session
⎣ ⎦ ⎣ ⎦ ⎢ ⎥
r ( x* )
⎢ ⎥
⎢ ⎡ 1 c1n 1 c2 n 1 1 ⎤⎥
b. Compute damping factor tin for updating x* : ⎢ −2 xq ⋅ ⎢ ⋅ + ⋅ +…+ ⋅ ⎥⎥
⎣⎢ ⎣⎢ f1 ( x) d1 f 2 ( x) d 2 f q ( x) d q ⎦⎥ ⎦⎥
x = x*
tin = 1
( )
while norm r ( x* + tin ∆x* ) ≥ (1 − α ⋅ tin ) ⋅ norm r ( x* ) ( ) The Hessian of f 0 ( x) evaluated at x* is ∇ 2 f 0 ( x* ) = 0 . The
tin = β ⋅ tin Hessian of φ ( x) evaluated at x is ∇ 2φ ( x* ) and is given by:
*
c. Update x = x + tin ∆x .
* * *
⎡ ∂ ⎛ q 1 ∂f ⎞ ∂ ⎛ q 1 ∂f ⎞ ⎤
⎢ ⎜∑ ⋅ i⎟ ⎜∑ ⋅ i ⎟⎥
d. Stopping criterion: stop if norm r ( x ( *
)) ≤ ε in . ⎢ ∂x1 ⎝ i =1 f i ( x) ∂x1 ⎠ ∂xq ⎝ i =1 f i ( x) ∂x1 ⎠ ⎥
⎢ ⎥
2. Update x = x (t ) * ∇ 2φ ( x* ) = ⎢ ⎥
⎢ ⎛ q ⎥
3. Stopping criterion: stop if n / t < ε out ⎢ ∂ ⎜∑ 1 ∂f ⎞ ∂ ⎛ q 1 ∂f i ⎞ ⎥
⋅ i ⎟ ⎜∑ ⋅ ⎟
4. Increase t = µ ⋅ t ⎢ ∂x1 ⎜⎝ i =1 f i ( x) ∂xq ⎟⎠ ⎜ ⎟
∂xq ⎝ i =1 f i ( x) ∂xq ⎠ ⎥⎦ *
⎣ x= x
In the above algorithm, the norm function is the Euclidean .
norm. ∇ is the gradient operator and ∇ 2 is the Hessian
operator. ∆x is used in updating x . The variation in x
due to adding ∆x to it is dampened by the damping factor,
tin . α is used in the condition statement for finding a
suitable tin . β is the factor used in decreasing the value of
tin to a suitable value. Based on experience with the
algorithm, good choices for α and β that result in suitable
values for tin are 0.3 and 0.7, respectively.