Electrical Power and Energy Systems 115 (2020) 105495

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Multi-objective optimization of overhead transmission lines including the T

phase sequence optimization
⁎
Jhair S. Acosta , Maria C. Tavares
University of Campinas, Campinas, São Paulo 13083-852, Brazil

A R T I C LE I N FO A B S T R A C T

Keywords: Since the human density, technologies, industries and services are in continuous grow, the electrical power
Environmental impact consumption increments over the time. To satisfy those power levels the transmission lines should increment
Evolutionary algorithms their power capacity. However, the increment has to be done taking into account that the line projects should use
HSIL lines the lowest possible space to decrease their environmental impact. The present paper proposes the use of opti-
Multi-circuit power lines
mized multi-circuit transmission lines with several circuits and voltage levels on the same tower. A mixed-
Multi-objective optimization
integer non-linear programming (MINLP) model with variables in a permutation based space for optimizing the
surge impedance loading (SIL), right of way (ROW), costs and height of power transmission lines is presented
and solved using evolutionary computation with a multi-objective approach based on the Non-dominated
Sorting Genetic Algorithm II (NSGA-II). The optimization is done by choosing the optimal position, type and
quantity of conductors as well as the phase sequence in multi-circuit transmission lines. To deal with the dif-
ferent types of decision variables, a custom constraint handling is proposed. As a result of the optimization
process, multi-circuit power transmission lines with narrower ROW, lower environmental impact, higher SIL and
similar costs to conventional lines are obtained.

1. Introduction
Within the electricalpower system, transmission lines (TLs) are re-
sponsible for transporting electrical energy from power plants to main
load centers. In this context, TLs should guarantee the transmission of
the required power by the loads. However, because of the increment of
the population, industries and services, the power consumption is
higher, overcoming the power transmission capacity of conventional
and single-circuit TLs. Accordingly, alternatives to increase the TLs
capacity should be adopted. Nevertheless, because of the growing on
density population and the environmental impact of TLs, new TLs
projects should use the land in an efficient way.
The former research to increase TL power capacity appears in 1909,
and proposes to use bundled conductors [1]. The typical distance be-
tween conductors belonging to the same phase was adopted as 0.457 m
in the American standards and 0.4 m in the European standards [2].
Several years later Russian researchers proposed the use of compact
lines with conductors non-uniformly separated inside a circumference
[3]. Those lines were denominated as high surge impedance loading
(HSIL) lines. Based on this concept several researches were developed
[4–6]. The first research to solve the problem from a multi-objective
approach was proposed in [7], increasing the SIL and optimizing the
costs. Until that point the researches were focused on single-circuit TLs.
A second alternative to increase the power transmitted in TLs and re-
duce the environmental impact is to use several circuits in the same
tower. Actually conventional multi-circuit TLs exist in the world. The
largest overhead double-circuit transmission line, the Kita-Iwaki power
line, operating at 500 kV with a height of 108 m and a ROW of 38 m,
was built in 1999. In countries with high population density, such as
Germany, India, Japan and China, double- or quadruple-circuit TLs are
common, meanwhile in rare cases six-circuit TLs are used.
General multi-circuit transmission line behaviour has been char-
acterized in several researches [8,9]. Specific studies for multi-circuit
AC power lines have been developed in the following topics: protection
[10,11], lightning performance [12–16], electromagnetic behavior
[17,18], transient behavior [19], secondary arc current extinction [20]
and optimization of four-legged reactor [21]. As seen in [22], the re-
searches are not focused only in AC TLs, but also in multi-circuit DC
power lines. Currently, some attention has been focused on hybrid AC/
DC multi-circuit TLs [23–25].
Recently, studies on the optimization of multi-circuit TLs, have been
performed [26]. Thus, Multi-circuit TLs operating at 138/230 kV were
optimized using a multi-objective approach based on evolutionary
computing inside a range of fixed conventional topologies.

⁎
Corresponding author.
E-mail addresses: jhairacosta@gmail.com (J.S. Acosta), ctavares@unicamp.br (M.C. Tavares).

https://doi.org/10.1016/j.ijepes.2019.105495
Received 28 January 2019; Received in revised form 3 July 2019; Accepted 18 August 2019
0142-0615/ © 2019 Elsevier Ltd. All rights reserved.
J.S. Acosta and M.C. Tavares Electrical Power and Energy Systems 115 (2020) 105495

The present document goes further by proposing the use of an al- Subjected to:
gorithm and a mathematical model to optimize multi-circuit TLs with 0.45 ⩽ b sb ⩽ B sb (7)
different voltage levels at same tower. We will refer to this lines as
Multi-circuit Multi-voltage Transmission Lines (MCMVTLs). The math- 20Vl ci
df ci ⩾
ematical model and restrictions can be used in TLs above 230 kV. The 10−3δrci 3 (8)
algorithm uses a multi-objective approach, taking into account the
1.667
conductor type, position, number of sub-conductors, circuit geometry, 2.0 ∗ 1.1Vl ci ∗ 2 ∗ 1.15 ⎞
Sdci ⩾ 1.03 ⎛ ⎜ ⎟
tower type, and phase sequence. The mathematical model has four ⎝ 600 3 ⎠ (9)
objective functions: maximize SIL and minimize: costs, ROW and
tower’s height. The optimization process is done through evolutionary Dci1, ci2 ⩾ max(Sdci1 + Sdci2 + tc, max(df ci1, df ci2)) (10)
computing with the NSGA-II [27] that is widely used in multi-objective
problems (however, it could be solved with different optimization al- mVlci ⩽ n scci ⩽ MVlci (11)
gorithms). The optimization was implemented for towers with 1 to 4
MTx = 0 (12)
circuits, operating from 230 to 1000 kV. As a result, enhanced compact
TLs with modified bundles and phase sequence were obtained. Even- − 5 ⩽ hvci ≤ 5 (13)
tually, the trade-off between the SIL, total cost, ROW and tower height
Esoil ≤ 8.33 (14)
is evidenced and represented via Pareto fronts and parallel coordinated
plots. Esupk < Ecritk (15)
The main contributions of this paper are as follows:
b Eci ≤ 10 (16)
1. A new mathematical model to represent the multi-circuit transmis-
→
sion line optimization problem. Psci → is in a permutation based space (17)
2. A new methodology to solve the multi-circuit transmission line
where
optimization problem.
3. Optimization of the phase sequence of each circuit in the TLs. CI , ci Set/Index of circuits
4. New optimized geometries for multi-circuit TLs. K, k Set/Index of conductors
5. Generation of Pareto fronts and parallel coordinated plots for the B, b Set/Index of bundles
multi-circuit TLs decision making process. SILo Sum of the surge impedance loading (SIL) of the CI conventional lines at
their own voltage levels
SILop Sum of the SIL of the CI optimized lines at their own voltage levels
2. Mathematical model Cto Sum of the costs of the CI conventional lines at their own voltage levels
Ctop Sum of the costs of the CI optimized lines at their own voltage levels
An optimal transmission line depends on the specified objectives. Rwo Sum of the individual ROW of the CI conventional lines at their own
Commonly, there exist technical and economical objectives, that in voltage levels
Rwop Total ROW of the optimized tower
most of the cases are in conflict between themselves. Thus, if a technical
Hto Sum of the individual height of the CI conventional lines at their own
optimization is selected (i.e., increase the transmission line capacity or
voltage levels
SIL) the cost will increase. On other hand, if an economical optimiza- Htop Total height of the optimized tower
tion is granted, the technical objectives will have poor values. b sb Bundle size of bundle b [m]
In this research there are modeled three technical objectives and B sb Maximum bundle size of bundle b [m]
one economical objective. The technical objectives are: increase the SIL Vlci Line voltage at circuit ci [kV]
( f1), reduce the ROW ( f3 ), and reduce the tower height ( f4 ), meanwhile df ci Clearance between phases of circuit ci [m]
the economical objective is decrease the costs ( f2 ). df ci1 Clearance between phases of circuit ci1 [m]
The problem to solve in this research is modeled initially as the df ci2 Clearance between phases of circuit ci2 [m]
maximization problem presented in (1). D ci1, ci2 Clearance between phases of circuits ci1 and ci2 [m]
Sdci Disruption distance of circuit ci [m]
max(f1 (x ), −f2 (x ), −f3 (x ), −f4 (x )) (1) Sdci1 Disruption distance of circuit ci1 [m]
Then, the problem is transformed in its pseudo normalized version Sdci2 Disruption distance of circuit ci2 [m]
in (2). Note that according to (3)–(6) f1′ can acquire values between tc Width of the tower column [m]
n scci Wires per bundle at circuit ci
[−1, ∞) , meanwhile f2′ to f 4′ acquire values between (−∞, 1]. Thus, f1′ is
mV l Minimal wires per bundle at voltage Vlci (see Table 1)
focused on the SIL increments whereas f2′ to f 4′ in penalizing the solu- ci
MV l Maximum wires per bundle at voltage Vlci (see Table 1)
tions that are worst than conventional lines in those objectives (costs, ci
MTx X coordinate of the tower mass center [m]
ROW and height). The problem is constrained to the electrical and
hvci Horizontal or vertical distance between central phase and external phases
mechanical restrictions presented in (7)–(17). The restrictions are at circuit ci [m]
properly described in Section 2.2. Restrictions (8) and (9) were taken Esoil Maximum electric field on the soil [kV/m]
from [28], whereas the others were proposed in [7] for single-circuit Esupk Maximum electric field on the surface of k conductor [kV/cm]
TLs to represent the mathematical problem related to TLs optimization. Ecritk Critical electric field of k conductor [kV/cm]
b Eci Difference on the electric field of conductors that belong to the same phase
max(f1′ (x ), f2′ (x ), f3′ (x ), f 4′ (x )) (2) at circuit ci [%]
δrci Air relativity density at circuit ci
With: → Vector containing the phase sequence of circuit ci.
Psci
f1′ = SILop / SILo − 1 (3)
2.1. Objective functions
f2′ = (Cto − Ctop)/ Cto (4)
2.1.1. Surge impedance loading
f3′ = (ROWo − ROWop)/ROWo (5)
The surge impedance loading is a reference for transmission line
f 4′ = (Hto − Htop)/ Hto capacity. Although TLs may operate under heavy loading conditions,
(6)
SIL is associated to TLs power capacity. When a TL transmits a power

2
J.S. Acosta and M.C. Tavares Electrical Power and Energy Systems 115 (2020) 105495

1
δ=
jωμ0 σc (19)
∼ ∼
∼ jωμ0 ⎛ 2h′k ⎞ ∼ jωμ0 ⎛ D′kl ⎞
z kk = log ⎜ + rDCk z kl = log ⎜
2π R ⎟ 2π d ⎟ (20)
⎝ k ⎠ ⎝ kl ⎠

∼ 2h D
Y = jω2πε0 [P]−1 pkk = log ⎛ k ⎞ pkl = log ⎛ kl ⎞
⎜ ⎟ ⎜ ⎟

⎝ Rk ⎠ ⎝ dkl ⎠ (21)
On the above equation rDCk is the internal DC resistance of con-
ductor k.
SIL calculation in multi-circuit TLs. Calculating SIL in multi-circuit TLs
∼
is slightly different from the case of single-circuit TLs. Dimension of Z
∼
and Y matrices depends on the number of circuits (CI). Thus, the re-
∼ ∼
duced Zred and Yred matrices are going to be squared with dimension of
∼ ∼
3·CI . Calculating the sequence component matrices Z012 and Y012 fol-
lows the next equations:
Fig. 1. Images method representation with complex soil.
1 1 1 ⎤
∼ ⎡ 4π 2π ⎡0 0 0⎤
equal to its SIL, the self generated and consumed reactive power are the T = ⎢1 e j 3 e j 3 ⎥ Tm = ⎢ 0 0 0 ⎥
⎢ j 2π j 4π ⎥ ⎣0 0 0⎦
same, so the line has minimum reactive power exchange with the power ⎣1 e 3 e 3 ⎦ (22)
system. At fundamental frequency (50 or 60 Hz) SIL is defined by the
∼
characteristic impedance of the line (z c ) and the rated line voltage (Vl ), ⎡T Tm ⋯ Tm ⎤
as presented in Eq. (18). At the same time z c depends on the series ∼ ⎢ T ⋱ ⋱ ⋮⎥
T2 = ⎢ m ⎥
positive sequence impedance (z1) and the positive sequence shunt ad- ⎢⋮ ⋱ ⋱ Tm ⎥
mittance ( y1) per unity of length, as presented in second term of Eq. ⎢ Tm ∼⎥
⎣ ⋯ Tm T ⎦ (23)
(18).
∼ ∼−1 ∼ ∼
Z012 = T 2 · Zred · T2 (24)
Vl2 V2
SIL = = lz ∼ ∼−1 ∼ ∼
zc 1 Y012 = T 2 ·Yred· T2 (25)
y1 (18)
∼
Note that T2 in Eq. (23) is a squared matrix with dimension of 3·CI .
To obtain the positive sequence parameters z1 and y1 it is necessary The total SIL will be the sum of the SIL of each circuit.
∼ ∼
to find the primitive line matrices Z and Y (with dimensions of K by K,
in which K is the total number of conductors and shield wires) by using 2.1.2. Costs
∼ ∼
the image method represented in Fig. 1. Those matrices Z and Y are When designing a transmission line project it is important to mini-
reduced to 3 by 3 by using the Kron reduction, incorporating earth mize the costs. Determine these costs with precision is complex because
wires effect [29]. Then, a fully transposition scheme is performed. each change on the design affects the costs. Therefore, this research has
Finally, the positive sequence parameters z1 and y1 are obtained by taken into account the most significant costs, trying to have a good
using the Fortescue transformation [30], supposing that the transmis- reproduction of real costs. However, even if the costs are not exact, this
sion line is balanced. will not compromise the comparison, because conventional and opti-
Although positive sequence parameters and SIL are obtained for mized TLs are addressed with the same calculations, so the degree of
balanced TLs, this does not mean that the presented model is only ap- cost optimization is fairly calculated.
plied to fully transposed lines. The total cost Ct (40), related to the proposed TLs, involves the cost
In the image method represented in Fig. 1 imaginary conductors are of: insulators Itc (26); phase wires and shielding wires Wtc (27); support
drawn under the soil. When a return current is taken into account an elements for shielding wires Stc (28); conductor laying Cltc (29); foun-
∼
imaginary soil or depth δ appears [31]. In the case of Z there is a return dations Ftc (30); tower re-positioning Rptc (31); commissioning Cotc (32);
current on the soil and δ appears. It is calculated by the complex depth steel and assembling of towers Totc (33); power losses Pltc (36); corona
∼
Eq. (19) where σc is the soil conductivity. For Y calculation there is no losses Pctc (38); and maintenance of elements Mtc (39) [7]. Note that all
complex soil, then δ = 0 . the costs are given in [USD].
Since only the positive sequence parameters at fundamental fre- Insulator costs. The costs related to insulators in this research are
quency are necessary for the optimization process, there is no need to only the acquisition costs. Maintenance cost are consider as a percen-
properly represent the return impedance or calculate zero sequence tage of the total cost of materials as presented later in (39). The
parameters at this design stage. However, the complete parameter minimum creepage distance in [mm/kV] according the pollution level
equations were considered as above described. Regarding to the is 20 for very light, 24 for light, 28 for medium, and 32 for heavy
shielding wires, they are not relevant for the positive sequence para- contamination.
meter, and the SIL calculation. Therefore, they are not considered The total cost of insulators can be calculated using Eq. (26) that in
during the optimization process. Only when the last population is ob- its inner term takes into account the number of insulators units by
tained the shield wires are incorporated, as explained in Section 4. means of the insulator leakage distance.
∼ ∼
Each one of the elements of Z and Y can be calculated using Eqs. CI
3Vlci Ld
(20) and (21), respectively. Note that the self impedance of a k con- Itc = 1.1Ic Qt ∑
ductor is denoted as ∼ z kk meanwhile the mutual impedance between a ci = 1
ld (26)
conductor k and l is denoted as ∼ z kl . On the other hand, the shunt ad-
where
mittance depends on the potential coefficients matrix P , composed by
the self potential coefficient of the k conductor ( pkk ) and the mutual Itc Total cost of insulators [USD]
potential coefficients between the conductor k and l ( pkl ). Ic Insulator disk cost (21 [USD])
Qt Quantity of towers (935)

3
J.S. Acosta and M.C. Tavares Electrical Power and Energy Systems 115 (2020) 105495

CI Total quantity of circuits in the tower voltage. This means that under normal condition very small corona
Vlci Line voltage of circuit ci [kV]
losses are expected. Note that since the projects are designed for a
Ld Minimal insulator leakage distance of each insulator (445 [mm])
specific life-cycle, the power losses vary along years.
ld Insulator leakage distance (20 [mm/kV])
(1 + Ir )Y − 1 ⎞
Pltc = 8760Pl Lf Cpl ⎛ ⎜
Y
⎟

Wires and shielding wires costs. The cost of the wires and shielding ⎝ Ir (1 + Ir ) ⎠ (36)
wires are related to their acquisition (27), necessary support elements
2 2
(28) and cost of conductor laying (29). Shield wires positioning is ex- CI
⎡ 3 ⎛ ⎛ SILci ⎞ ⎞⎤
plained in Section 4. Pl = ∑ r+ci ⎢
Vl ⎜SILci + 3r+ci ⎜ Vl ⎟ ⎟⎥
ci = 1 ⎣ ci ⎝ ⎝ ci ⎠ ⎠⎦ (37)
K
Wtc = 1.1 ∑ Ll Mk Cw K 1.8
(27) f ⎡ Esupk ⎤
k=1 Pctc = ∑ Cpl ⎢ ⎜⎛0.33 − 0.2⎟⎞ Rk ⎥
k=1
50 ⎣⎝ Ecritk ⎠ ⎦ (38)
Stc = 120Qt Nsw (28)
where
Cltc = (6000K + 3500Nsw ) Ll (29)
P ltc Total cost of power losses [USD]
Pl Power losses [MWh]
Lf Load factor (0.8)
Wtc Total cost of wires and shielding wires [USD]
K Total quantity of conductors in the tower f Industrial frequency [Hz]
Ll Line length (350 [km]) Cpl Power losses cost (36.10 [USD/MWh])
Mk Mass per kilometer of k conductor [ton/km] Y Expected life time [years]
Cw Wire cost (4000 [USD/ton] for phase wires and 2000 [USD/ton] for shielding Ir Interest rate (8 [%])
wires) CI Total quantity of circuits in the tower
Stc Total cost of support elements for shielding wires [USD] r+ci Positive sequence resistance of each circuit ci [Ω ]
Qt Quantity of towers (935) Vlci Line voltage of circuit ci [kV]
Nsw Total quantity of shielding wires SILci SIL of circuit ci [MW]
Cltc Total cost of conductor laying [USD] Pctc Total cost of corona losses [USD]
K Total quantity of conductors in the tower
Esupk Electric field at surface of conductor k
Tower costs. To address the cost of the towers requires evaluating the
Ecritk Critical electric field of conductor k
cost of foundations (30), tower re-positioning (31), system commis-
Rk External radius of conductor k [m]
sioning (32), and material (33) (in this case steel) considering suspen-
sion type towers. Since there are no equations on the literature for
Total costs. The total cost of the project Ct in (40) involves the costs
foundations, tower re-positioning and commissioning, they are con-
explained in the previous sections and the maintenance costs. As seen in
sidered as constants multiplied by the number of towers.
(39) it was assumed that the maintenance cost Mtc is 20% of the costs of
Ftc = 7200Qt (30) the materials.
Rptc = 500Qt (31) Mtc = 0.2 ∗ [(26) + (27) + (28) + (29) + (30) + (31) + (32) + (33)]

Cotc = 1000Qt (32) (39)

Ct = (26) + (27) + (28) + (29) + (30) + (31)
1.1Ct To w
Totc = + (32) + (33) + (36) + (38) + (39) (40) (40)
1000 (33)
K
Vk ⎞
To w = ∑ 0.043Kt yk ⎛⎜Lk2/3 + Tk2/3 + ⎟ 2.1.3. Right of way
⎝ 1.141 ⎠ (34)
k=1 The Right of Way (ROW) or transmission corridor is a set of land
used to maintain, repair, operate and construct TLs. ROW also serves to
(3.28 + DF )2
Kt = 2.89 + maintain safety clearances. It varies according to the voltage level and
400 (35)
to the use of the land in the TL route. It is composed by the wire zone
where that is under the wires plus their maximum swing due to wind, and a
border zone that is set for safety clearances. Since the problem to solve
Ftc Total cost of foundations [USD] in this research is an optimization problem, the ROW only needs to be
Rptc Total cost of reallocation of towers [USD] calculated in the same way for conventional and optimized lines. In this
Cotc Total cost of commissioning [USD]
research the ROW was obtained using the EPRI criteria [28]. Therefore,
Totc Total cost of tower [USD]
To w Weight of the tower [ton] since we are assuming a maximum electric field of 8.33 kV/m at 1.5 m
Ct Steel cost (7000 [USD/ton]) above the soil, the corresponding ROW is twice times the distance from
K Total quantity of conductors in the tower the tower center where the electric field at 1.5 m above the soil is
Kt Auxiliary variable for suspension type towers equal/lower to 1.67 kV/m.
yk Height of k conductor [m]
Lk Longitudinal loads at conductor k [daN]
Tk Transverse loads at conductor k [daN] 2.1.4. Tower height
Vk Vertical loads at conductor k [daN] This objective was created trying to avoid trivial solutions in which
DF Maximum horizontal distance between phases [m] all the circuits are allocated one above the other with phases in vertical
arrangement, produced trying to reduce the ROW. Tower height is
Power losses and corona losses costs All TLs projects present losses due defined in such way that the minimal height has to be equal or higher
to the resistance of the cables and occasionally by the corona effect. than the mid-span clearance dictated by the 2002 NESC [32]. It has to
Power losses (37) depend on the positive resistance of each transmis- ensure that electric field at 1.5 m above the ground (Esoil ) is lower than
sion line (r+ci ), meanwhile corona losses (38) depend on the conductor the maximum allowable values. According to Brazilian standards the
diameter and the electric field in the conductor surface. In the present maximum Esoil is 4.5 kV/m for occasional transit of people and 8.33 kV/
process, the critical electric field was considered as 1.1 pu of rated m for trained workers. Finally, the tower height is calculated as the

4
J.S. Acosta and M.C. Tavares Electrical Power and Energy Systems 115 (2020) 105495

maximum absolute value of y component of all k conductors. cd VLGci

d fci =
δrci 10−3 (43)
2.2. Constraints 3.9210pbci
δrci =
273 + Te (44)
Real TLs projects are subjected to several electrical and mechanical
constraints. Most of them are continuous and linear. However, there are pbci = 10 log(76) − yci /18336 (45)
also integer and non-linear constraints. The harder constraints to ad-
dress are the non-linear, i.e., the electric field on the conductors and on where
the ground.
cd Creepage distance [mm/kV].
VLGci Maximum line-ground overvoltage in circuit ci.
2.2.1. Bundle size δrci Relative density of the air in circuit ci.
The minimal distance of conductors belonging to the same phase is Te Environmental temperature in [° C].
pbci Barometric pressure in circuit ci position.
45.7 cm according to United States standards and 40 cm according to
yci Maximum height of conductors in circuit ci.
European standards. However, no upper limits are given. Thus, in
theory it is possible to have bundles of high dimensions as in [4,5,33],
among others. However, bundles with those dimensions, i.e., almost NESC 2002 allows to use (46) when the switching voltage is known.
8 m, are very big and there is no mechanical studies to ensure that the However, as those distances tend to be lower than the former approach,
towers support those bundles and the loads produced by the wind over (43)–(45) were used in the final calculation.
them. Therefore, in this research the maximum bundle size was set as 1.667
1.15 2 VL − L PU ⎤
1.5 m, aiming to reduce high possible mechanical stresses. Lines of d fci = 1.03 ⎡
⎢
⎣ 700 ⎥
⎦ (46)
1000 kV naturally have bundles of almost 1.2 m, so this limit is near to
existent in actual TLs.

VL − L Maximum phase-phase voltage of the line in [kV]. Commonly VL − L = 1.10Vl .

2.2.2. Clearance distances
Mid-span distance. The minimal mid-span distance intends to ensure
that the personnel and buildings are in a safety distance, not being
affected by the electromagnetic field of TLs. Therefore, the minimal
height of the tower has to be in concordance to the NESC 2002 [32].
The minimal mid-span clearance (MS) in [m] can be calculated using
(41).
MS = MS22kV + 0.01(VLG − 22)
where
MS22kV Mid-span clearance for lines with maximum rms line-ground voltage
between 750 V to 22 kV (5.6 [m])
VLG Maximum rms operating voltage of the line in [kV]
NSCE 2002 allows to use the alternative method in (42) to calculate
the minimal mid-span clearance when the maximum switching-surge
factor (PU) is known. It can be used in transmission lines with line to
ground voltages of 98 kV and above. This method must be used for line
to line voltages higher than 814 kV. However, if the line has a rated
voltage lower than 169 kV, the mid-span calculated with the alternative
approach shall not be less than the calculated in the primary approach
in (41).
V ·PU
MS = 1.15c ⎡ ⎤ V
⎣ 500 ⎦
where
PU Maximum switching-surge factor of the line-ground voltage of the line in
[pu].
Distance to metallic structures (Strike distances). To avoid flashovers
and back flashovers, the conductors have to be separated by a certain
distance from metallic structures and support elements. In similar way
than the mid-span clearance, NESC provides two formulas to calculate
(41) the tower strike distance Sd in [m]. The primary approach in (47) is
used for lines operating in voltages above 50 kV.
Sd = 0.280 + 0.005(Vl − 50) (47)
where Vl is the phase-phase voltage in [kV].
NSCE 2002 allows the alternative method in (48) to calculate the
minimal tower strike distance when the maximum switching-surge
factor is known for line-ground voltages higher than 98 kV. This
method must be used for phase-phase voltage higher than 814 kV.
However, for a line with phase-phase voltage lower than 169 kV, the
mid-span calculated with the alternative approach shall not be less than
the calculated in the primary approach in (47).
1.15V ·PU 1.667
Sd = 1.03 ⎡ ⎤
⎣ 600 ⎦ (48)
where
Maximum peak line-ground voltage of the line in [kV].
(42) PU: Maximum switching-surge factor of the line-ground voltage of the line in
[pu].

V Maximum peak line-ground voltage of the line in [kV]. When working with multi-circuit lines, the circuits can be separated
PU Maximum switching-surge factor of the line-ground voltage of the line in in such way that a metallic structure or a support element can be lo-
[pu]. cated in the middle of them. Fig. 2 illustrates the situation in which a
c Margin of safety: 1.2 for vertical clearances and 1.0 for horizontal clearances. structure with width tc is located in the middle of two circuits. In this
case the distance Dci1, ci2 between circuits ci1 and ci2 has to be (49).
Accordingly, the minimum tower height has to ensure that the mid-
Dci1, ci2 ⩾ max(Sdci1 + Sdci2 + tc, max(df ci1, df ci2)) (49)
span height is higher than MS.
Distance between phases (air clearance). Design of air clearances have where
to be done to avoid flashover and back flashover on TLs. It depends on
the contamination of the environment, because it modifies the creepage Sdci1 Strike distance of circuit ci1 [m].
distance per kV of line-to-ground voltage and the insulators’ power Sdci2 Strike distance of circuit ci2 [m].
frequency voltage strength. The minimum creepage distance is the same tc Width of the structure [m].
as used in insulator cost. Therefore, the minimal distance d fci in [m] df ci1 Clearance between phases of circuit ci1 [m]
between phases of each ci circuit can be calculated using (43)–(45). df ci2 Clearance between phases of circuit ci2 [m]

5
J.S. Acosta and M.C. Tavares Electrical Power and Energy Systems 115 (2020) 105495

Fig. 2. Distance of conductors to tower.

Also, it is possible to assume that there are no support elements or

structures in the middle of the circuits, so Dci1, ci2 has to be
Fig. 3. Image method representation for electric field calculation.
⩾ max ⎜⎛d fci1, d fci2⎞⎟.
⎝ ⎠ 2.2.5. Electric field balance
It is important to mention that the distances d fci and Dci1, ci2 are
checked for the wires in three positions: tower, mid-span and maximum A correct balance on the electric field on the conductors of the same
sag at full load. phase is important to avoid overheating one conductor or to reduce
losses. Therefore, in this research was proposed that the maximum
difference on the superficial electric field of the conductors cannot be
2.2.3. Number of sub-conductors
higher than 10% of the average superficial electric field of that phase.
The quantity of conductors allowed per bundle varies according the
operating voltage level as presented in Table 1. This is not standardized,
2.2.6. Audible noise
they correspond to the limits proposed in this research.
In 500 kV TLs and above audible noise is an important issue.
Normally, it is produced under corona effects and it is measured at the
2.2.4. Electric field
edge of the ROW [28]. Audible noise can be calculated for different
When working with TLs in HV/EHV/UHV it is desired to avoid
weather conditions. However, since heavy rain condition is one of the
power losses due to Corona effect and avoid overcome the maximum
most critical operation point, the current research considers this
electric field at soil level. According to researches made by [34,35], the
weather, using Eqs. (55)–(57) [28].
Corona effect is produced when the electric field on the surface of the
if nb < 3:
conductors overcome the maximum electric field (Ecrit ). Ecrit depends on
the conductor radius and the environmental conditions, and in most of 665
Pab = 20lognb + 44logdb − Emaxb
+ k nb+
the cases has a value near to 20 kV/cm. To calculate the electric field at
any point p it is used the image method represented in Fig. 3 and Eqs. 75.2 − 10logDb − 0.02Db (55)
(50)–(52). This method can be extended to any quantity of conductors if nb ⩾ 3:
and voltage level. Thus, in multi-circuit TLs the superficial charge on
each conductor (∼
665 22.9(nb − 1) db
qk ) will change according to the voltage of each k Pab = 20lognb + 44logdb − Emaxb
+ b sb
+
∼ ∼
conductor and the capacitance matrix (Q = C·V ). 67.9 − 10logDb − 0.02Db (56)
K ∼
qk ⎛ x p − xk x p − xk ⎞
∼
Ex = ∑ ⎜
2πε0 ⎝ rc2
− ⎟ = Erx + jE
ix Patot = 10log ∑ 10 Pab /10
k=1
rI2 ⎠ (50) (57)
b∈B

K ∼
qk ⎛ yp − yk yp + yk where
∼ ⎞⎟ = E + jE
Ey = ∑ ⎜
2πε0 ⎝ rc2
− ry iy
k=1
rI2 ⎠ (51) nb Number of conductors in bundle b
Pab Audible noise power at bundle b in [dB]
Ep = Erx2 + Eix2 + Ery2 + Eiy2 (52) db Diameter of conductors at bundle b in [cm]
Db Distance from bundle b to measured point in [m]
where Emaxb Maximum electric field of at bundle b in [kV/cm]
k nb Auxiliary constant. If nb = 1, then k nb = 7.5 but if nb = 2 , then k nb = 2.6
rc2 = (x p − xk )2 + (yp − yk )2 (53) Patot Total audible noise power in [dB]

rI2 = (x p − xk )2 + (yp + yk )2 (54) 2.2.7. Phase sequence

In above equations ε0 is the vacuum permittivity; x p and yp are the Although in the optimization process we have different phase se-
→
coordinates of the analyzed point p;xk and z k are the coordinates of the quence for each circuit (Psci ), we cannot have repeated phases on the
∼ →
k conductor; ∼qk is the charge at conductor k;Ex is the total electric field same circuit ci. Therefore, Psci must be in a permutation based space. In
∼ →
on the point p in the x axis; Ey is the total electric field on the point p in other words, the inner elements of Psci must be different. As example, in
the y axis; Ep is the effective electric field on the point p; and the sub- →
a quadruple transmission line Psci → ci = {1, 2, 3, 4} could be
indices r and i denote the real and imaginary part of the corresponding
ABC − BCA − CAB − ACB , but not ABC − ABA − ABC − CBA , be-
complex variables.
cause there are two A phases in circuit 2.

Table 1
2.2.8. Auxiliary constraints
Allowed number of conductors according to the voltage level.
An additional constraint hvci was created to control in each circuit
Voltage [kV] 69/138 230 345 440/500 750/765 1000 1150 the vertical or horizontal distance between the central phase and the
1 2 2 3 4 8 10
axis created by the external phases of the same circuit. Thus, hvci does
min nsc
max nsc 2 3 4 6 8 10 12 not allow the central phase to be more than 5 m away from that axis. In
circuits with phases in delta hvci works slightly different, controlling the

6
J.S. Acosta and M.C. Tavares Electrical Power and Energy Systems 115 (2020) 105495

→ Vector formed by the hvci value of each ci circuit

movement of the central phase in a straight line limited by the co- hv
ordinates of the external phases. This variable is presented with more →
r Vector formed by the →
rx ci component of each ci circuit
x
detail in Section 3.2.2. →
ry Vector formed by the →
ryci component of each ci circuit
→
r Vector formed by the →
r component of each ci circuit
xc xcci
2.2.9. Constraints handling →
ryc Vector formed by the →
rycci component of each ci circuit
In order to explore prominent solutions, even if some of them are
outside of the feasible region, the objective function has to be penalized
when a constraint has been overcome. Therefore, the penalization The general procedure used to optimize (59) follows the pseudo
function presented in (58) is proposed. Thus, when the value Cvc of a code presented in Algorithm 1. NSGA-II was modified in order to in-
constraint c ∈ C overcomes the limit Clc of that constraint, the inner troduce diversity to the population by adding random solutions. Note
sum in (58) is negative (e.g., if Ecritk (Cl 9 ) is 20 and Esupk (Cv9 ) is 50, the that since the optimization problem has a large number of constraints,
inner sum of (58) yields to −1.5). Note that the set C of constraints is it is necessary to ensure that the initial population is inside the feasible
composed of the ten proposed constraints presented in (7)–(16). region. From former knowledge of the problem it is known that circular
geometries with small bundles usually meet all constraints [7]. This is
P= ∑ ⎨
(
⎧ Clc − Cvc
Clc ) if Cvc > Clc because circular bundles maintain a better electric field balance on the
c∈C ⎩0 otherwise (58) surface of the conductors than other geometries, and smaller bundles
have lower superficial electric field than bigger bundles. Thus, to gen-
Note that the inner term of the sum in (58) is negative if some
erate feasible individuals, it is often sufficient that all internal elements
constraint is overcome, otherwise it is 0. With this approach, even so-
of → rx , → rxc , and →
ry , → ryc be equal, but it does not imply that
lutions outside the feasible region are explored. Therefore, after several →
rx = ry = rxc = →
→ → ryc .
generations, solutions outside the feasible region could be improved
and migrate to the feasible region. Algorithm 1. Optimization using a modified NSGA-II

2.3. Final model

1: Initialize parameters and limits
2: Po ← N random solutions inside the feasible region
The mathematical model introducing all the restrictions and the 3: it ← 1
penalization function is presented in (59). 4: while it ⩽ max(generations ) do
5: Pa ← N /10 random solutions in the feasible region
max (2) + (58) (59)
6: Po ← Po ∪ Pa
subject to: 7: So← Crossover of Po with 90% of probability
8: So← Mutation of So with 5% of probability
(7), (8), (9), (10), (11), (12)(13), (14), (15), and (16). 9: Po ← Po ∪ So
10: Po← Ranking Po according Pareto dominance
Note that the final model has restrictions of type integer, real, 11: Po← Selection of Po according crowding distance
12: it ← it + 1
complex, linear, non-linear and permutations. Therefore, the problem
to be solved is a Mixed-integer non-linear programming model (MINLP)
in a permutation based space.
The initial population Po is composed of N random solutions gen-
3. Optimization method erated inside the feasible region. Then, the algorithm begins the opti-
mization process. It is composed of the stages of insertion of a popu-
3.1. General algorithm lation Pa of new random solutions (to help to maintain diversity),
crossover (producing descendants), mutation, insertion of population of
The MINLP problem in (59) is solved using the NSGA-II since it is descendants (So ), and natural selection through Pareto ranking and
one of the most representative and reliable evolutionary algorithms crowding distance. The stages are executed until the maximum number
(EAs) to solve multi-objective problems [36]. However, other multi- of generations g is reached, obtaining the final population Po . To reduce
objective algorithms could also be used. Since NSGA-II works with a the computational effort, each time that a new population is created, its
population of solutions it is used a population Po composed of N pos- fitness is evaluated. At the end of each generation the survival popu-
→
sible solutions as shown in (60). Each one of the Sn candidates solutions lation conserves its fitness information. Therefore, in each new gen-
(∀ n ∈ N ) is represented with a chromosome comprised of the variables eration it is not necessary to evaluate the fitness of the total population,
shown in (61). but only the fitness of new members.
→ → Fig. 4 graphically describes the process followed by NSGA-II in the
Po = { S1, …, SN }T (60) natural selection process. Note that the Pareto ranking is applied in a
→ → → → → ⎯→ ⎯ → population composed of an initial population of progenitors/parents, a
Sn = {Tty, →
cty, →
nsc , →
c oty , Ps , h t , C x , C y, D , h v, →
rx , →
ry , →
rxc , →
ryc } (61) population of random solutions and a population of sons. Then, each
where solution is sorted according the Pareto frontier in which they belong.
The final population begins to be filled by adding all the solutions in a
Tty Base tower type Pareto frontier if the size of the new population is lower than the po-
→
c Vector formed by base topology of each ci circuit (ctyci ) pulation limit. The process is repeated until the total number of solu-
ty
→
nsc Vector formed by the number of wires at each ci circuit (n scci ) tions in the current frontier does not fit in the final population. In this
→
c oty Vector formed by the conductor type of each ci circuit (cotyic ). The 100 moment, a selection based on the crowding distance is applied to the
ACSR conductor types were numbered from 1 to 100 as in [7] solutions of the frontier that does not fit the final population. Then, the
→ →
Ps Vector formed by the phase sequence of each ci circuit (Psci ) best elements, according to the crowding distance, join the population
→
ht Vector formed by the height of each ci circuit (htic ) until there is no more empty spaces in the population, whereas the
→ Vector formed by the X center of each ci circuit (Cxic ) remaining solutions are deleted. In the example of Fig. 4 the population
Cx
→ Vector formed by the Y center of each ci circuit (Cyic ) P + S is ranked in five frontiers. Frontiers 1 to 3 enter directly to the
Cy
⎯→
⎯ final population, but since solutions of frontier 4 do not fit in the final
D Vector formed by the Dci value of each ci circuit
population limit, a selection based on the crowding distance is

7
J.S. Acosta and M.C. Tavares Electrical Power and Energy Systems 115 (2020) 105495

feasible region. Therefore, the initial population is created randomly,

but trying to be inside the limits imposed by the restrictions. This is
obtained by using circular bundles smaller than 1.5 m and pre-fixed
tower and circuit geometries. Consequently, most of the solutions in the
initial population are inside the feasible region.

3.2.2. Geometry generation

To help to solve the problem in (59) the conductors’ positions in
each circuit were mathematically represented. Four possible base cir-
cuit types were modeled: vertical in (62), horizontal in (63), delta in
(64) and inverted delta in (65), with Dci = Dci + d fci . The variables
⎯→
⎯ → ⎯→ ⎯ → ⎯→ ⎯ →
( Xaci , Yaci, Xbci , Ybci, Xcci , Ycci ) denote the (x , y ) positions of the con-
ductors in each phase at circuit ci.
The mathematical model of circuits with horizontal configuration is
presented in (62).
Fig. 4. NSGA-II process. ⎯→
⎯ ⎯→
⎯ → →
Xaci = X1ci − Dci Ya = Y1ci
⎯→⎯ ⎯→
⎯ → →
performed on them. Then, only the best elements in frontier 4, ac- Xbci = X2ci Ybci = Y2ci + hvci
cording to their crowding distance, will be part of the final population. ⎯→
⎯ ⎯→⎯ → →
Xcci = − Xaci Ycci = Yaci (62)
Before natural selection process, an uniform crossover for integer
variables, an arithmetic crossover for continuous variables, and an OX The mathematical model of circuits with vertical configuration is
crossover for permutations is performed over Po (progenitors/parents), presented in (63).
i.e., a random pair of progenitors produces two sons with combined ⎯→
⎯ ⎯→
⎯ → →
genetic material, according to their variable type. It was used a cross- Xaci = X1ci Yaci = Y1ci − Dci
over probability pc of 90%. Thus, N sons are produced comprising the ⎯→
⎯ ⎯→
⎯ → →
Xbci = X2ci + hvci Yb = Y2ci
population So . As happens in nature, a genetic mutation can occur with ⎯→
⎯ ⎯→
⎯ → →
low probability in any generation. Therefore, the population So can Xcci = Xaci Ycci = Y1ci + Dci (63)
suffer a mutation with a probability pm of 5%. When a mutation occurs, The mathematical model of circuits with delta configuration is
different Δr obtained by gaussians distributions with standard deviation presented in (64).
of 0.1 are added to the continuous variables
⎯→
⎯ ⎯⎯⎯→ ⎯⎯⎯→ ⎯→ ⎯ ⎯→⎯ ⎯→
⎯ ⎯→
⎯ → →
( ht , Cx , Cy , D , hv , →r,→
r,→
xr ,→r ) , meanwhile random integers
y xc yc Xaci = X1ci − Dci /2 Yaci = Y1ci
Δinteger obtained by gaussians distributions with standard deviation of 1 ⎯→
⎯ ⎯→
⎯ → → Dci 2
are added to the integer variables (Tty, → cty, →
nsc , ⎯→
⎯ ) . The mutations of
coty
Xbci = X2ci + hvci Dci /10 Ybci = Y2ci + Dci2 + ( )2

integer variables was controlled in order to only produce feasible so- ⎯→

⎯ ⎯→
⎯ → →
→ Xcci = − Xaci Ycci = Yaci (64)
lutions. For permutation based variables (Ps ) the mutation is performed
by doing random permutations in each circuit. After mutation, the new The mathematical model of circuits with delta inverted configuration is
Po will be composed of the union of Po (progenitors) and So (sons). At presented in (65).
this point, the population Po has a size of 2N . Then, through the natural ⎯→
⎯ ⎯→
⎯ → → Dci 2
selection process it is reduced to N. At the end, the solutions in Po will Xaci = X1ci − Dci /2 Yaci = Y1ci + Dci2 + ( ) 2
be almost uniformly dispersed in the Pareto front. Since in multi-ob- ⎯→
⎯ ⎯→
⎯ → →
jective problems we have two spaces: (1) the solution space, that in this Xbci = X2 + hvci Dci /10 Ybci = Y2ci
⎯→
⎯ ⎯→
⎯ → →
case is the tower geometry with its characteristics, and (2) the objective Xcci = − Xaci Ycci = Y1ci (65)
space, that is the value of the objective function obtained by each so-
lution, it is possible to observe the influence of a variable or set of where
variables in the final solution. Therefore, the influence of uncertainties
⎯→
⎯ → → →
in a variable or set of variables (represented in the solution space), can X1ci = →
rx cicos ⎜⎛ θci − ϕ1ci⎟⎞ Y1ci = →
rycisin ⎜⎛ θci − ϕ1ci⎟⎞
be observed with the Pareto Front and/or with Parallel Coordinated ⎝ ⎠ ⎝ ⎠
Plots. ⎯→
⎯ → → → → →
⎛
X2ci = rx cicos ⎜ θci − ϕ2ci⎟⎞ ⎛
Y2ci = rycisin ⎜ θci − ϕ2ci⎟⎞
The main advantage of the Pareto approach is that in a single run it (66)
⎝ ⎠ ⎝ ⎠
is possible to find several solutions well distributed on the Pareto front,
helping to choose a posteriori solutions in the decision making process. ϕ1ci = oci π / n scci ϕ2ci = oci π / n scci +
π (1 − oci)
2 (67)
For further explanation of NSGA-II, the original algorithm in [27]
should be consulted. Eqs. (66) and (67) are used to help to maintain the conductors in the
bundles uniformly spaced at the beginning, aiming to obtain a small
3.2. Techniques used to solve the optimization problem electric field unbalance on the wires. Furthermore, if n scci is odd oci = 1,
→ → → → →
otherwise oci = 0 . It is important to clarify that rx ci, ryci, rxcci, rycci , and θci
In the following sections are explained in detail some features used → → →
are vectors with dimension of n scci . Additionally, while rx ci, ryci, rxcci , and
to solve the specific problem of this research. → →
rycci may accept any continuous value, θci is a vector with equally spaced
angles from 0 to 360°. It should be noted that the outer phases have
3.2.1. Initial population been formulated to be symmetrical. Additionally, the variable Dci helps
Commonly, EAs initialize the population randomly with normalized to ensure restriction (8).
values. This allows to explore the whole search space in better form. Since Eqs. (62)–(65) produce circuits with the bundle of the lower
However, since the problem solved in this research has several re- phase centered in (0, 0) , it is necessary to move the circuits to their
strictions, most of the solutions randomly created are outside the position in the tower. The variables Cx ci and Cyci are the displacement of

8
J.S. Acosta and M.C. Tavares
Fig. 5. Proposed tower configuration according to the number of circuits.
each circuit center in a direction (x , y ) . However, since the problem has
many constraints, it is necessary to generate base topologies for the
towers, helping to allocate the solutions in the feasible region. As shown
in Fig. 5 two tower configurations were proposed for double circuit
lines (in blue rectangle), four for triple circuit lines (in green rectangle)
and six for quadruple circuit lines (in red rectangle). The mathematical
model of each tower configuration is not presented here, but it can be
deduced taking into account that the towers must satisfy restrictions
(9), (10) and (14).
3.3. Handling the size of the solutions
Since constraint (11) limits the number of conductors per circuit,
the variables →
rx , →
ry , →
rxc , →
ryc have different dimensions depending on n scci .
However, to vectorize the operators of crossover and mutation to in-
crease the algorithm speed, the solutions should have the same di-
mensions. To solve this problem the treatment described in Algorithm 2
was applied to those variables.
Algorithm 2. Handling of variable dimension
1: Identify the higher n scci
→ → → →
2: To initial rx , ry , rxc, ryc add random numbers until a dimension equal to max(n scci )
is obtained
→ → → →
3: When applying crossover and mutation use the new rx , ry , rxc, ryc with dimension
given by max(n scci )
→ → → →
4: When evaluating the fitness use rx , ry , rxc, ryc with their original dimension given
by n scci
Accordingly, crossover and mutation are performed in solutions
→ → → →
with the same dimension. Thus, rx , ry , rxc , ryc are composed by their
original elements plus random numbers added to complete the desired
dimension. Thus, elements that do not have a contribution to the pro-
blem are added. However, the fitness calculation is performed correctly
by only taking into account the original dimension of → rx , →
ry , →
rxc , →
ryc .
By this simple approach crossover and mutation can be vectorized,
9
Electrical Power and Energy Systems 115 (2020) 105495
improving the computation time. Finally, parallel computing was used
in the fitness calculus. Therefore, instead of evaluating sequentially the
fitness of each solution, several solutions were evaluated in parallel.
The number of solutions to be addressed at the same time depends on
the number of CPUs and of the number of cores in the GPUs in cases
with simple operations.
4. Test and results
In this section are presented some of the TLs optimized with the
proposed mathematical model and methodology. All simulations were
performed in a notebook with 8 Gb of RAM and a quad-core i7 7700HQ
operating at 2.8 GHz with all the processors running in parallel. To
show the efficiency and scalability of the proposed solution, single,
double, triple and quadruple TLs with B sb = 1.5 m were optimized.
Although the algorithm does not restrict the voltage of the circuits,
only standard voltage levels were simulated. Additionally, it was con-
sidered that the position of the circuits in the tower also depends on the
voltage level. Circuits with different voltage levels can be at the same
height. However, a circuit with higher voltage cannot be located below
a circuit with lower voltage. This was considered because the algorithm
is not optimizing the voltage level of the circuits and in practical cir-
cumstances it is not reasonable to design circuits with higher voltage at
the bottom of the tower.
Many combinations of circuits could be performed and optimized.
However, only some selected geometries are presented. They corre-
spond to the best solutions of f = f1′ + f2′ + f3′ + f 4′ after 10 runs of the
algorithm. There are considered different voltages, population size N
and maximum number of generations g. The characteristics of the se-
lected optimized lines are shown in Table 2 and 3, whereas the geo-
metries are presented in Figs. 6–9. From Table 2 it is possible to observe
that increasing the SIL in many cases requires more investment. How-
ever, the ROW and height in most of the cases will be lower than
conventional lines. Only in the case of single circuit lines the height was
higher because the algorithm chose a delta circuit to increase the SIL.
Note that, in overall, the TLs of the examples were improved in ranges
between 67% to 182%. Any combination of circuits will produce sig-
nificant increments, even more if the number of circuits increases,
mainly because of the ROW reduction. However, note that if the circuits
have higher voltage levels (e.g., 2×750–1×1000 kV transmission line),
it is more difficult to increase the SIL, but still better solutions than
conventional lines will be produced.
Note that the major advantage of the optimized multi-circuit TLs is
the ROW reduction ( f3′), obtaining reductions up to 88%, with a good
height reduction ( f 4′), that in the best case was 62%. At the same time it
was possible to obtain SIL increments ( f1′) up to 33%, using bundles
smaller than 1.5 m. However, in most of the cases it was necessary to
make higher initial investments because of the use of higher bundles,
more quantity of conductors, larger conductors, and more steel in the
towers. Thus, the highest investment was 10% higher than in conven-
tional lines.
To demonstrate the repeatability of the results, the variance σ of
f = f1′ + f2′ + f3′ + f 4′ when optimizing 10 times each test system was
calculated. In all cases values lower than 5% were obtained, being
2.76% in the single circuit line, 4.52% in the double circuit line, 3.78%
in the triple circuit line, and 0.68% in the quadruple circuit line,
Table 2
Simulation settings and objective function values.
Line [kV] N g time [s] f1′ [%] f2′ [%] f3′ [%] f4′ [%] f [%]
500 300 500 371 31.4 3.9 43.7 −12.5 66.5
2×500 300 500 527 31.6 −1.6 64.8 41.8 136.2
2×750–1×1000 300 500 997 26.1 9.6 82.6 45.8 164.0
2×230–2×500 300 500 833 32.7 −0.5 87.6 62.1 181.8
J.S. Acosta and M.C. Tavares Electrical Power and Energy Systems 115 (2020) 105495

Table 3
Characteristics of the selected optimized TLs.
Line n scci Conductor bs [m] Phase Sequence

500 kV 4 Flamingo 0.76 ABC

2×500 kV 5/5 Duck 1.12 ABCCBA
2×750–1×1000 kV 5/6/8 Flamingo 1.43 CBABACACB
2×230–2×500 kV 2/2/4/4 Flamingo 1.09 ABCBCACABABC

Fig. 8. Optimized triple circuit transmission line at 2×750 kV and 1×1000 kV

with NSGA-II approach.

Fig. 6. Optimized single circuit transmission line at 1×500 kV with NSGA-II

approach.

Fig. 9. Optimized quadruple circuit transmission line at 2×230 kV and

2×500 kV with NSGA-II approach.

circuits have different conductors. The only restriction is that in each

Fig. 7. Optimized double circuit transmission line at 2×500 kV with NSGA-II
approach.
showing that N and g were properly selected, allowing to obtain similar
results in each run.
No matter the quantity and voltage level of the circuits in the tower,
the algorithm always optimizes all the circuits. Thus, the objective
function of a circuit will not be deteriorated by the optimization of
other circuit. Therefore the proposed solutions are worth even for
towers with circuits belonging to different owners.
It is important to highlight that although in Table 3 all the geo-
metries have the same type of conductor, that was just a coincidence,
not imposed by the algorithm. It is possible to obtain cases in which all
circuit only one type of conductor can exist.
It is important to note in Table 3 that the optimized lines have
different phase sequence when compared to conventional lines that
always have an ABC sequence. Without phase sequence optimization
the objective function values presented in Table 2 will be reduced by
10% approximately. The main reason behind this is that when mod-
ifying the phase sequence it is possible to control the electric field
below the lines (Esoil ). Therefore, with lower Esoil it is possible to reduce
the height of the tower and/or the ROW. The decision will depend on
which value is more advantageous for the fitness. Changing the phase
sequence has others implications, as the transient response. A study is
presented in [37] where no important differences to conventional
double-circuit lines were observed. Conversely, the optimization of the
phase sequence in multi-circuit transmission lines help to reduce the
induced overvoltages and the secondary arc current.
Also, note that in Figs. 6–9 shielding wires were also allocated (by
using the electrogeometric model (EGM)). However, it was performed
after the optimization process ended to all candidate solutions. There-
fore, the lightning shielding analysis can be performed independently

10
J.S. Acosta and M.C. Tavares
Fig. 10. Pareto front for an optimized four circuit transmission line at
2×230 kV and 2×500 kV with NSGA-II optimizing the four objectives.
from the optimization process, without significantly modifying the re-
sults. Shielding wire failures were not evaluated in the present material.
In multi-objective problems there are two spaces: the decision space
and the objective space. The decision space is where the variables are
located and is represented in Figs. 6–9, whereas, the objective space is
the value of the objectives against all the other objectives for each so-
lution. Pareto fronts can be easily visualized in problems with two or
three objectives. However, for more objectives it is necessary to use
tools like parallel coordinate plots for a correct visualization.
Fig. 10 presents the Pareto front of a four circuit transmission line
optimization plotting SIL vs. Costs. Note that trying to represent a four
dimensional Pareto front in two dimensions leads to a non well defined
Pareto front. However, all the solutions correspond to non-dominated
solutions.
To observe the behaviour of all objectives at the same time and their
relationship it is possible to use parallel coordinate plots using the
normalized value of the objective functions of each solution. In this case
the value is normalized according to the available values of the objec-
tives in the population. Therefore, a value of 1 in one objective does not
mean that it was improved by 100%, but means that it has the max-
imum value of the population in a specific objective. Thus, the parallel
coordinate plot allows to see the trade-off between the objectives, e.g.,
in Fig. 11 the worst solution for objective 1 is the best solution for
objective 2. In other words, the solution with lowest SIL is also the
solution with lowest Costs. The same analysis can be implemented to
the other objectives, e.g., in Fig. 11 the solution with the lowest ROW is
also the solution with highest height. With this, the trade-off between
objectives and the existent conflict is evidenced. Graphs with solutions
dispersed on the black bars indicate that the algorithm is properly
finding diverse solutions.
It is possible to create also a parallel coordinate plot of the actual
values of f1′ to f 4′ and see the real enhancement of all objectives. Fig. 12
shows the parallel coordinate plot in this circumstance when optimizing
TLs with 1 to 4 circuits on the same tower. With this graph it is possible
to see the behaviour of the objectives according to the number of cir-
cuits. Accordingly, it is possible to see that single circuit lines present
the worst value of ROW and height, meanwhile the 4 circuit TLs have
Fig. 11. Parallel coordinate plot for an optimized four-circuit transmission line
at 2×230 kV and 2×500 kV with NSGA-II.
11
Electrical Power and Energy Systems 115 (2020) 105495
Fig. 12. Parallel coordinate plot when optimizing 1–4 circuits.
the best values in those objectives. Regarding the other objectives, the
solutions have similar values. However, according to the number of
circuits they are more concentrated in a specific zone, e.g., regarding to
costs, the solutions when using 3 circuits are concentrated in better
values than with other quantity of circuits. However, it is not possible to
conclude that the behavior will be always the same, because it depends
on the voltage level of the circuits. Therefore, if the voltage level
changes it is necessary to generate new graph for the analysis process.
It is important to highlight that the multi-objective optimization via
Pareto approaches like NSGA-II gives a bunch of solutions that can be
used in the decision making process from an a posteriori approach.
Therefore, it becomes a very useful tool to choose an appropriate so-
lution according to the specific needs of the project. For instance, if we
have monetary concerns, we could choose one of the upper solutions in
bar 2 in Fig. 12 and sacrifice one or more of the other objectives. As
another example, if our major concern is the environmental impact we
could choose the upper solutions in bar 3 in Fig. 12. As you can see, the
best solution will change according to each specific project. Finally, the
use of the proposed towers in actual projects still demands further de-
sign to address the physical structure concerns. However, the imposed
mechanical restrictions help to produce feasible tower structures.
5. Conclusions
In this paper, a MINLP model in a permutation based space to solve
the problem related to the optimization of multi-circuit TLs with non-
conventional bundles and phase sequence is presented. The proposed
mathematical model and methodology is flexible and scalable by al-
lowing the inclusion of several circuits and different phase sequences.
The optimization model optimizes multi-circuit TLs while con-
sidering technical, economical and environmental aspects of their op-
eration, installation, and maintenance. Different circuit scenarios and
tower topologies are included to improve and ensure the feasibility of
the proposed towers.
The problem was solved from a multi-objective approach via NSGA-
II and improvements of the computational time of the algorithm were
also considered.
As a result, several solutions were obtained. The objectives trade-off
was evidenced via Pareto fronts and parallel coordinate plots.
Therefore, the decision making process can be supported in those
graphs. As an example, choosing the points with better balance in all
objectives (sum of f1′ + f2′ + f3′ + f 4′), multi-circuit TLs with an overall
improvement of 67 to 182% were obtained, evidencing that: (1) with
more circuits more improvement is obtained, specially because the
ROW reduction; (2) The phase sequence optimization helps the ROW
and/or height improvement because of the electric field reduction; (3)
In most of the cases it is not possible to increase the SIL without rising
the investment compared to conventional lines; (4) In higher voltage
levels the SIL improvement is more difficult due to limitations on the
bundle size; (5) Because of the reduced ROW the optimized lines have
J.S. Acosta and M.C. Tavares
lower environmental impact than conventional lines.
Acknowledgements
The authors acknowledge the financial support provided by the
following institutions: CAPES, CNPq and FAPESP (No. 2015/05626-0
and 2017/20010-1).
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in the
online version, at https://doi.org/10.1016/j.ijepes.2019.105495.
References
[1] Thomas PH. Output and regulation in long-distance lines. Trans Am Inst Electr Eng
1909;XXVIII(1):615–40.
[2] Abetti PA. Bibliography on extra-high-voltage systems. Trans Am Inst Electr Eng
Part III Power Appar Syst 1958;77(3):1517–35.
[3] Alexandrov GN, Podporkyn GV. Improvement of the efficiency of 35 to 220 kV lines.
In: Int. conf. AC DC power transm., no. 5, London, UK; 1991. p. 226–31.
[4] Salari JC. Optimizing transmission lines conductors bundle geometries (in
Portuguese) [Master Thesis]. Brazil: Federal University of Rio de Janeiro; 1993.
[5] Maciel RP. Maximization of Characteristic Power of Transmission Lines using a non-
linear optimization method (in portuguese) [Master Thesis]. Brazil: University of
Campinas; 2013.
[6] Sarmiento JSA, Tavares MC. Enhancement the overhead transmission lines’ capacity
by modifying the bundle geometry using heuristics algorithms. IEEE PES Asia-
Pacific power energy eng. conf., Suzhou, China. 2016. p. 646–50.
[7] Acosta JS, Tavares MC. Methodology for optimizing the capacity and costs of
overhead transmission lines by modifying their bundle geometry. Electr Power Syst
Res 2017.
[8] Wang YH, Meng SM, Tang B. Study on type design of multi-circuit transmission line
tower. Adv Mater Res 2012;594–597:900–3.
[9] Novitskiy A, Westermann D. Interaction of multi-circuit overhead transmission lines
of different voltages located on the same pylons. In: IEEE electr. power qual. supply
reliab., Tartu, Estonia; Jun 2012. p. 1–4.
[10] Agrasar M, Uriondo F, Hernandez JR, Alvarez R. A useful methodology for ana-
lyzing distance relays performance during simple and inter-circuit faults in multi-
circuit lines. IEEE Trans Power Deliv 1997;12(4):1465–71.
[11] Zuo XF, Kuribyashi H. Protection of parallel multi-circuit transmission lines. In:
Proc. 6th lnternauunal Conferr. Adv. Power Spm Control. Oper. hlanagcmenr, no.
November, Hong Kong, China; 2003. p. 630–4.
[12] Liew AC, Thum PC. Comparative studies of lightning performance of a quadruple-
circuit dual voltage 275/132KV transmission line design with Wooden Crossarms.
IEEE Trans Power Deliv 1993;8(4):1973–80.
[13] Zheren Z, Qing Y, Libin Y, Wenxia S, Tao Y. Lightning back-flash performance of
220kV AC quadruple-circuit transmission lines on the same tower. Asia-Pacific
symp. electromagn. compat. APEMC, Beijing, China. 2010. p. 1570–3.
[14] Wang X, Li Z, He J. Impact of coupling ground wire on lightning protection effect of
multi-circuit tower. Int. conf. light. prot., Vienna, Austria. 2012.
[15] Li Z, Yu Z, Wang X, He J. A design of unbalanced insulation to improve the lightning
performance of multi-circuit transmission lines. 31st int. conf. light. prot. ICLP,
12
Electrical Power and Energy Systems 115 (2020) 105495
Vienna, Austria. 2012. p. 0–3.
[16] Zhao Y, Su J, Gu S, Ren H, Zhao C. Influence of different coupling ground wire
erecting mode on lightning withstand performance for multi-circuit transmission
line in a tower. Int. conf. light. prot. ICLP, Shanghai, China. 2014. p. 1036–40.
[17] Guihong F, Yanxin W, Bingyi Z. Study on electromagnetic environment of multi-
circuit transmission lines on same tower. Int. conf. power syst. technol.
POWERCON, New Delhi, India. 2008.
[18] Liu W, Pan M, Liu K, Xu G. Research on electromagnetic character of 500/220kV
mixed-voltage quadruple-circuit transmission line. 12th int. conf. environ. Electr.
Eng. EEEIC, Wroclaw, Poland. 2013. p. 227–31.
[19] Kizilcay M, Neumann C. Backflashover analysis for 110-kV lines at multi-circuit
overhead line towers. Int. conf. power syst. transients, Lyon, France. 2007. p. 4–9.
[20] Yang Q, Sima W, Yang M, Wang R, Yuan T. Secondary arc current of ultra-high
voltage transmission line with a mixed voltage of 1000/500 kV on a single tower.
IET Gener Transm Distrib 2015;9(7):686–93.
[21] Lv XC, Sun QQ, Li QM, Shi W. Multi-objective parameter optimization of shunt
reactors for multi-circuit transmission lines on the same tower. 4th int. conf. electr.
util. deregul. restruct. power technol., Weihai, Shandong, China. 2011. p. 271–6.
[22] Li W, Zhang B, He J, Zeng R, Chen S. Ion flow field calculation of multi-circuit DC
transmission lines. In: Int. conf. high volt. eng. appl., no. 3, Chongqing, China;
2008. p. 16–9.
[23] Wei S, Qingmin L, Liang Z, Li Z, Xinchang L. Research on lightning performance of
AC/DC hybrid transmission lines on the same tower. IET Gener Transm Distrib
2013;7(2):166–74.
[24] Mackow A, Kizilcay M, Malicki P. Lightning performance of overhead hybrid
transmission line. IEEE PowerTech, Eindhoven, Netherlands. 2015.
[25] Malicki P, Mackow A, Kizilcay M. Estimating the lightning performance of a multi-
circuit transmission tower. 33rd int. conf. light. prot. ICLP, Estoril, Portugal. 2016.
[26] Ribeiro MF, Vasconcelos JA, Teixeira DA. Optimization of compact overhead lines
of 138/230kV: optimal selection and arrangement of cables and definition of the
best transmission line tower topology. In: 17th IEEE int. conf. environ. electr. eng.
1st IEEE ind. commer. power syst. Eur. EEEIC/ I CPS Eur., Milan, Italy; 2017.
[27] Deb K, Pratap A, Agarwal S, Meyarivan T. A fast and elitist multiobjective genetic
algorithm: NSGA-II. IEEE Trans Evol Comput 2002;6(2):182–97.
[28] Electric Power Research Institute. EPRI AC Transmission Line Reference Book–200
kV and Above. 3rd ed. EPRI; 2005.
[29] Kron G. Tensor analysis of networks. London: MacDonald; 1939.
[30] Fortescue CL. Method of symmetrical co-ordinates applied to the solution of poly-
phase networks. Trans Am Inst Electr Eng 1918;XXXVII(2):1027–140.
[31] Deri A, Tevan G, Semlyen A, Castanheira A. The complex ground return plane a
simplified model for homogeneous and multi-layer earth return. IEEE Trans Power
Appar Syst 1981;PAS-100(8):3686–93.
[32] IEEE. National electrical safety code 2002 edition. IEEE Std C2-2002; Aug 2001. p.
1–324.
[33] Portela C, Gomes S. Analysis and optimization of non conventional transmission
trunks, considering new technological possibilities. In: Proc. VI SEPOPE Salvador
Bras., vol. SP-092, no. 1, Salvador, BA, Brasil; 1998. p. 1–6.
[34] Starr, Lloyd, Peek. An investigation of corona loss: law of corona and dielectric
strength. J AIEE 1927;46(12):1457–61.
[35] Miller CJ. The calculation of radio and corona characteristics of transmission-line
conductors. Trans Am Inst Electr Eng Part III Power Appar Syst 1957;76(3):461–72.
[36] Yusoff Y, Ngadiman MS, Zain AM. Overview of nsga-ii for optimizing machining
process parameters. Procedia Eng 2011;15:3978–83. cEIS 2011.
[37] Acosta JS, Tavares MC. Transient behaviour of non-conventional multi-circuit
power lines with different voltages levels at the same tower. Int. conf. power syst.
transients, Perpignan, France. 2019.