7.

TDO using Evolutionary Algorithms

CHAPTER 7: TRANSFORMER DESIGN OPTIMIZATION USING

EVOLUTIONARY ALGORITHMS

7.1 Optimization:

Optimization is the process of making. something better. Problems in

engineering, economics and physical and social sciences. Entail the
optimization (minimization or maximization) of.an objective function subject to
certain constraints. (restrictions and tradeoffs).

7.1.1 Transformer Design Optimization:

Primarily the transformer design conclude by looping the overall cost

minimization including cost of raw materials, cost of labor and capitalization of
losses. The reduction must be considered as constraints which may be
imposed on the transferred power, winding geometry, Losses, % Z, flux
density, tank Dimensions etc. Since the cost and constraints are generally
non-linear functions of the design variables, a non-linear constrained
optimization method is required to apply(98). The suitable method is needs to
developed after detailed study and to be applied for the transformer design
optimization. For simplicity only” major cost components and constraints can
be considered. ”

A transformer should perform satisfactory without overheating or without

damaging itself at the time of certain abnormal events like lightning stroke or
short circuits fault. Transformer is a simple engineering and static device, it
must have inequitable lifecycle (20-25 years),if it functioned at its rated
conditions. Even though satisfying the basic requirements of long life working
still has a wide”room for development in possible designs. The transformer
manufacturers therefore find it in its best economic interest to choose, ”within
the limitations imposed by the constraints, that combination of design
parameters which results in the lowest cost unit. The costs and constraints
can be stated logically in terms of the design variables. The mathematical
theory of optimization with constraints can also applied to Transformer design
optimization problem. Optimization is a huge area of mathematics which
comprises major specialized sectors such as linear programming,

Chiragkumar N Parekh 78
 7. TDO using Evolutionary Algorithms

unconstrained optimization, linear or non-linear equality or inequality

constrained optimization etc.TDO falls into non-linear equality and inequality
constrained”optimization(98). In this area, there are no algorithms or iteration
schemes which guarantee that a global”optimum (say Cost) will be found.
Most of the algorithms are centered to a local optimum with varying degrees
of efficiency. These are also not guaranteed if one starts the iterations too far
from a local optimum. Some insight is therefore usually required to find a
suitable starting point for the iterations. The past experience for the same can
be a milestone or work as a guide. There is a branch of optimization called
geometric programming which guarantees” convergence to a global minimum.
It provided the function to be minimized called the objective or cost function
and the constraints are expressible in a” certain way. This method is very
powerful if the problem functions conform to this type. ”
While designing a transformer, we generally hope for cost minimization. So
our objective function f(x) is mostly therefore a cost function. It comprises
many terms like material costs, cost of losses, labor cost and overhead costs.
These all costs including constraint functions needs”conveyed in terms of a
set of design variables. Here a 2 winding, 3 phase, core-form power
transformer is considered to perform the TDO. ”

Here the non-linear, ”twice differentiate objective or constraint functions is

considered. In practice it is found that these functions are often of the form”
𝑎 𝑎 𝑎 𝑎
𝐹(𝑥) = ∑𝑚 𝑖1 𝑖2 𝑖𝑛 𝑚 𝑛
𝑖=1 𝑏𝑖 𝑋1 𝑋2 … . . 𝑋𝑛 = ∑𝑖=1 𝑏𝑖 ∏𝑗≠𝑘 𝑋𝑗
𝑖𝑗
[1]

where the coefficients b, and the exponents aij are arbitrary constants. This
differs from the polynomials considered”earlier where the coefficients bi all
had to be positive. The k-th component of the gradient vector of this function
is given by”

𝜕𝐹(𝑋) 𝑎 −1 𝑎
= ∑𝑚
𝑖=1 𝑏𝑖 𝑎𝑖𝑘 𝑋𝑘
𝑖𝑘
∏𝑛𝑗≠𝑘 𝑋𝑗 𝑖𝑗 [2]
𝜕𝑋𝑘

The k, q-th entry of its Hessian matrix, where k≠q, is given by

Chiragkumar N Parekh 79
 7. TDO using Evolutionary Algorithms

𝜕𝐹(𝑋) 𝑎 −1 𝑎 −1 𝑎
= ∑𝑚
𝑖=1 𝑏𝑖 𝑎𝑖𝑘 𝑎𝑖𝑞 𝑋𝑘
𝑖𝑘
𝑋𝑞 𝑖𝑞 ∏𝑛𝑗≠𝑘 𝑋𝑗 𝑖𝑗 [3]
𝜕𝑋𝑘 𝜕𝑋𝑞
𝑗≠𝑞

When k=q, the diagonal entry of the Hessian matrix is”

𝜕2 𝑦 𝑎 −2 𝑎
= ∑𝑚
𝑖=1 𝑏𝑖 𝑎𝑖𝑘 (𝑎𝑖𝑘 − 1)𝑋𝑘
𝑖𝑘
∏𝑛𝑗≠𝑘 𝑋𝑗 𝑖𝑗 [4]
𝜕𝑥𝑘2

Above listed formula must be handled carefully when a variable or an

exponent disappears. Further, when a variable is negative and the respective
exponent is non-integer, the expression stands as imaginary. It is also
observed that during the iterations even if the variables are constrained to be
non-negative, the exponent is simply rounded off to the nearest integer. It
permits the iteration process of the equation. Finally with minimum approach,
the suitable variables will become non-negative so that the rounding”come to
be redundant. In case, If a variable is estimated to be negative” at the
minimum, then it should occur with an integer exponent” in all the functions. ”

7.1.2 Transformer Design Optimization Process: (Design Variables)

”The considered basic design variables are as following: ”

(1) Flux density B in Tesla”
(2) LV winding current density is Js A/ mm2”
(3)Core radius Re in mm”
(4) HV to LV hi-lo gap is g mm”
(5) The LV Mean radius is Rs mm”
(6) The LV Mean radius is Rp mm”
(7) Height of LV winding He in mm”
(8) Radial build of the secondary (LV) winding ts in mm”
(9)Radial build of the Primary (HV) winding tp in mm”
(10) Core Weight is Mc in kg”
(11) Tank Weight Mt in kg

The weight of core and tank can be presented in terms of the other” design
variables. Due to expressing some of the material and labor costs and losses
in terms of them, their inclusion in basic set of design variables is found

Chiragkumar N Parekh 80
 7. TDO using Evolutionary Algorithms

convenient. Their correlation on the other variables is going to be expressed

in terms of equality constraints. The magnitude for the above”variables are
kept in the range of about 1 to 100. Same are used internally in the computer
optimization program. For the input and output concerned related to user the
units are a matter of familiarity and can differ from the above. Here, the
primary winding height is not taken as a design variable. In practice, It is
usually” taken to be 10 mm shorter than the secondary winding.

So, Hp=αHs, [5]

Hp is the primary winding, α is a fraction ≈0.95. The gaps considered by user

is represented by gc and go, which are kept fixed and entered, gc depends on
whether a tertiary or tap winding is present” under the LV winding and go
depends on the phase to phase voltages. Core window height is represented
by H and T is stand for the window” width. The maximum stack height is
represented by X ≈2Rc. These” are expressible in terms of the other
variables. ”

7.1.3 Transformer Design Optimization(Cost Function):

The cost of copper is one of the most important component for the TDO.
Same is worked out with reference to the weight of copper. In terms of the
design” variables, the weight of the 3 secondary” windings (for 3 phases) is”

”𝑀𝑠 = 3𝑑𝑐𝑢 uɳ𝑠 (2𝜋RsHs 𝑡𝑠 ) [6]

Where 𝑑𝑐𝑢 is for copper density and ɳ𝑠 is for the fill factor i.e. copper cross
section area of winding. The weight of winding to be multiplied by a Rs./kg to
reach at the cost to include in the objective function. This cost will comprise
the cost of the raw copper and the cost” of paper insulation covering plus any
overhead or storage” costs. The fill factor is being achieved separately by
selecting a conductor size which meets”constraints on geometrical
parameters of and the number of turns.”If a regulating or tertiary winding is
there, its cost is determined as some fraction of the secondary winding’s cost.
Now the weight of the primary winding is similarly as expressed for secondary
winding which is as following.

Chiragkumar N Parekh 81
 7. TDO using Evolutionary Algorithms

𝑀𝑝 = 3𝑑𝑐𝑢 ɳ𝑝 cx(27c𝑅𝑝 ℎ𝑝 𝑡𝑝 ) [7]

Here ɳ𝑝 is the fill factor of primary winding. Similarly secondary same is also
gets multiplied by a Rs/kg. The weight and cost will be differ from secondary
winding because of conductor cross-section area and total numbers of turns
of Primary winding. The weight of the core1000Mc in kg and same is
multiplied by a Rs/kg for the core cost calculation. This Rs/Kg also includes
overhead and storage costs. The tank cost is similarly worked out 1000Mt in
Kg and multiply by Rs/Kg for getting tank cost. Mc and Mt are being
expressed in terms of the other variables by means of equality constraints. ”

Apart from the main raw material like copper and CRGO, other “material like
oil, Mild Steel, PCPB insulation such as cylinders, key-spacers, ”and lead
support structure, Copper flexible, bushings, tap-gear, coolers etc are also the
part of the cost optimization of transformer. Practically the cost working makes
influence on Load and no-load losses of the transformer and same is”included
in the cost function. For customer the running cost is also that much important
as initial cost as it is going to operate many years. So the efficiency which
directly depends on the losses of the transformer is also a most important
portion for the TDO. The load losses include the resistive (Ohmic) loss plus
the losses due to stray flux imposing on metallic components like tank wall,
yoke clamps , pressing bolts etc. The losses in the secondary” winding can be
expressed as

𝑊𝑠 = 𝑃𝑐𝑢 (1 + 𝑒𝑐 𝑓𝑠 )𝐽𝑠 𝑉𝑠 ) [8]

Here“𝑃𝑐𝑢 is stand for the copper resistivity at the applicable temperature,𝑒𝑐 𝑓𝑠 is

the eddy current factor which is related to stray flux and depends”on the
conductor geometry.” Vs is the copper volume which can be expressed”

𝑉𝑠 = 3ɳ𝑠 (2𝜋RsHs 𝑡𝑠 ) [9]

On this base we can calculate the loss of secondary winding (resistive+ stray).
Js is in Amp/mm², ɳ𝑠 needs to be in appropriate units to get ”Ws in kW. By

Chiragkumar N Parekh 82
 7. TDO using Evolutionary Algorithms

multiplying the calculated loss (kW) with the cost of load losses in RS/kW, we
will get the cost capitalization. Similarly, the primary winding’s losses are
given by”

𝑊𝑝 = 𝑃𝑐𝑢 (1 + 𝑒𝑐 𝑓𝑝 )𝐽𝑝 𝑉𝑝 ) [10]

Where, the copper volume is

𝑉𝑝 = 3ɳ𝑝 (2𝜋𝑅𝑝 𝐻𝑝 𝑡𝑝 ) [11]

Jp is the Primary winding current density .As the ampere-turns of the primary
and secondary” are equal under balanced conditions, we can obtain Jp in
terms of Js by”

𝐽𝑠 ɳ𝑠 𝑡𝑠
”Jp = [12]
𝛼 ɳ𝑝 𝑡𝑝

After that, multiply Wp by the Rs/kW for cost of load losses before adding it to
the cost function.”The stray losses are also part of the same is omitted. ”
The no-load losses are termed as the core losses. Practically the CRGO base
mill or CRGO supplier feeds a curve or polynomial expression for the core
losses in W/kg in terms” of the flux density,

Core Loss (W/kg) = ao + a1B +a2B2 + …….ak Bk [13]

The core loss will be achieved by multiplying this by the core weight in kg,
1000Mc. There are numbers of factors like non-uniform flux in the comers,
building stresses, frame factors etc are the responsible for higher core losses
than calculated one. So same is to be considered while making cost
calculation. Labor costs like winding manufacturing cost, Core staking cost,
Cost for CCA and final fitting are an important part of the cost of a transformer
since power transformers are usually custom made.

As can be seen, the cost function can get rather lengthy. ” However, all the
terms mentioned above are”expressions of the form given by following
formula [1]. ”

Chiragkumar N Parekh 83
 7. TDO using Evolutionary Algorithms

7.1.4 Transformer Design Optimization (Equality Constraints):

In the process of design finalization, the equality constraints must be
satisfied closely. Undoubtedly the rated capacity or MVA is the most important
constraint which is asked by the customer. It is needed to express it in terms
of” the chosen design variables. Here it is worked in terms of phase
quantities. The power transferred per phase” P, is given by P=MVA / 3.

This is also given by”

P=VsIs where Vs is the rms secondary phase voltage in kV and Is is the

rms secondary phase” current in kA.

Using Faraday's law, ”

10−3
Vs= 𝑁𝑠𝐴𝐹𝑒 2𝜋𝑓𝐵
√2
[14]

Where
B = The peak” flux density in Tesla.
Ns= The number of secondary turns. ”
AFe,= The area of the iron in a cross-section of the core steel in cm2,
f = The frequency in Hz.
Now, Expressing AFe in terms of our design variables, we have”

AFe = ncc𝜋𝑅𝑐2 [15]

nc= fill factor for the core steel( i.e. the fraction of actual steel in” a circle of
radius Rc ).

The current Is is given in terms of the current density by substituting these

into the expression for” power, we see that Ns cancels out and we find”

𝐽𝑠 ƞ𝑠 ℎ𝑠 𝑡𝑠
Is= [16]
𝑁𝑠

P= (10-3√2𝜋 2 𝑓) csB𝑅𝑐2 𝐽𝑠 ℎ𝑠 𝑡𝑠

Chiragkumar N Parekh 84
 7. TDO using Evolutionary Algorithms

Divided by P to get this to be or order 1 and expressed it in the form of our

standard equality constraints 𝐶𝑖 (x)=0,

10−3 √2𝜋2 𝑓
( ) csB𝑅𝑐2 𝐽𝑠 ℎ𝑠 𝑡𝑠 − 1 = 0 [17]
𝑃

This is an expression of the form [1].Note that

This is an expression of the form [1]. Note that [1] is of this form with all the
ail=0, bl=1, and m=1. We have kept the numerical constants in [18] in order to
reduce errors.

Per unit reactance between windings is also the next most important equality
constraints constraint which is being specified by the customer. A simplified
expression for the same is as following.

(𝑉𝐼)𝑏 𝑟 2 (𝑅𝑠 𝑡𝑠 +𝑅𝑝 𝑡𝑝

”X(p.u)= 7.5606 x10-5(𝑉⁄ 2 + 𝑅𝑔 𝑔)” [18]
𝑁 )𝑏 (ℎ+𝑠) 3 3

Where
r = the co-ratio (=1 for a two winding transformer). ”
Rg= The mean radius of the gap between the primary and secondary”
𝑅𝑠 +𝑅𝑝 𝑡𝑠 −𝑡𝑝
windings and is given by𝑅𝑔 = +
2 4
[19]
H is the average height of two windings

1+𝛼
ℎ= ( ) ℎ𝑠” [20]
2

”S is a correction factor for fringing flux”

𝑡𝑝
𝑠 = 0.32 (𝑅𝑝 + - Rc ) [21]
2

”(𝑉𝐼)𝑏 is the base MVA/phase and(𝑉⁄𝑁)2𝑏 is the base kV/turn. ”

𝑉𝑠
= (10−3 √2𝜋 2 (0.0254)2 𝑓)ƞ𝑐 𝑅𝑐2 𝐵
𝑁
[22]

Chiragkumar N Parekh 85
 7. TDO using Evolutionary Algorithms

”converting to standard form, we obtain

7.5606𝑥10−5 𝑃𝑟 2 1 𝑅𝑠 𝑡𝑠 𝑅𝑝 𝑡𝑝 𝑅𝑠 𝑔 𝑅𝑝 𝑔 𝑡𝑠 𝑔
( ) 𝑥 { + + + +
(10−3 √2𝜋 2 (0.0254)2 𝑓)𝑋 (𝑃. 𝑈. ) ƞ𝑐 𝑅𝑐2 𝐵 2 ℎ𝑠 3 3 2 2 4
𝑡𝑠 𝑔 1+𝛼 𝑅𝑝 𝑡𝑝 𝑅𝑐
− }−( − 0.032 − 0.16 + 0.32 ) = 0
4 2 ℎ𝑠 ℎ𝑠 ℎ𝑠
[23]
This is related to equation [1] and has been rightly regularized to order” of
magnitude 1. ”The constants are again kept explicit to reduce errors.

We treated Rp as an independent variable thus far since it appears in many

formulas. However, it can be communicated in terms of other design variable as

𝑡𝑠 𝑡𝑝
Rp = Rs + 2 + g + 2 [24]

Finally we get the equality constraint as following

𝑅
Mc=10-3dFeƞ𝑐 𝜋𝑅𝑐2 (3𝐻 + 4𝑇 + 6𝑋 + 0.30235 ƞ 𝑐) [25]
𝑐

Now the core weight related equality constraint is worked here. The
parameters like window height H, window width T and the maximum sheet
width X which make up the core stacks are consider for the same. This is
given by following equation.

ƞ 𝑅 𝑅
(10-3dFeπ) 𝑀
𝑐 𝑐
{3ℎ𝑠 + 8𝑅𝑝 +}4𝑡𝑝 + 4𝑅𝑐 + 0.30235 ƞ 𝑐 + 3𝑆𝑙𝑎𝑐𝑘 𝑠 + 8𝑔0 }-1=0
𝑐 𝑐

[26]

The correction factor for the joints is represented by last term in equation.
This is the density” of the core iron in kg/cm2. In terms of the basic design
variables”,

H=h + stack s [27]

where slacks is a slack distance in the window. This is depends on the BIL
of the winding .This is a constant for the unit” under consideration

Chiragkumar N Parekh 86
 7. TDO using Evolutionary Algorithms

𝑡𝑝
T = 2 (𝑅𝑝 + + 𝑔0 − 𝑅𝑐) [28]
2

Now, X is going to be taken here as 2Re. Although a ”more exact formula can
be used for greater accuracy. Thus we obtain, in standard normalized form,

ᶯ𝑐𝑅𝑐² 𝑅𝑐
(10-3dFeπ ) {3ℎ𝑠 + 8𝑅𝑝 + 4𝑡𝑝 + 0.30235 + 3slacks + 8𝑔𝑜 } - 1 = 0 [29]
𝑀𝑐 ᶯ𝑐

The tank related equality expression can also be worked. The tanks dimensions
are fixed not only by Core and Coil geometry but also the clearances based
on BIL, tap leads make grate influences.

Inequality Constraints

The consideration of inequality constraint on the mean radius of the LV

winding will be suitable as it must not drop below a minimum value given
by”

𝑡𝑠
Rs≥Rc + gc + 2 [30]

For further more clarity it is expressing in following standard form

𝑅𝑠 𝑡𝑠 𝑔𝑐
- - –1≥0 [31]
𝑅𝑐 2𝑅𝑐 𝑅𝑐

The hilo gap between HV-LV winding must be maintained as per specified
BIL. This minimum gap grum, leads to the inequality”

➢ g ≥ gmin [32]

Or in standard form
𝑔
➢ –1≥0 [33]
𝑔𝑚𝑖𝑛

Due to overvoltage the over fluxing happens and that leads the core to the
saturation means the flux density B will be saturated. The maximum value
Bmax leads to the inequality in standard form”

Chiragkumar N Parekh 87
 7. TDO using Evolutionary Algorithms

𝐵𝑚𝑎𝑥
–1≥0 [34]
𝐵

The current density is to be within limit. Here, Js which we call Jmax. This is
based on cooling considerations. Same is represented in standard form as
following

𝐽 𝑚𝑎𝑥
–1≥0 [35]
𝐽𝑠

It may be necessary”to limit the tank height for shipping purposes, the tank
height can be expressed as”

H+2X+slackt ≤ Hmax [36]

Where

Hmax= The ”maximum tank height

slackt= The vertical slack of the core. It will depend on the clamping
structure, the presence of leads, etc, slack t” will be a constant”for a given
unit. H is given by above equation and, taking X-2Rc, we get”

hs+4Re+slacks+slackt≤ Hmax [37]

It is converted in standard form as following

𝐻𝑚𝑎𝑥−𝑠𝑙𝑎𝑐𝑘𝑠−𝑠𝑙𝑎𝑐𝑘𝑡 ℎ𝑠 4𝑅𝑐
( 𝐻𝑚𝑎𝑥
) - 𝐻𝑚𝑎𝑥 - 𝐻𝑚𝑎𝑥≥ 0 [38]

The”quantity in parentheses in the above formula is a constant for the unit

under consideration.

The other inequalities creates the forces on the windings which are
restricted by the strength and elongation characteristics of copper
conductor, the thermal and cooling capacity of the winding ”depends on
current density and fill factor, and the high voltage withstand (impulse”
strength) depends on the type of conductor used and voltage level, etc.

Chiragkumar N Parekh 88
 7. TDO using Evolutionary Algorithms

These inequalities are being considered for determining Winding

temperature gradients, winding stresses, and impulse voltages. Designers
keeps some margins on the” initial design so that during finalizing the design
any difficulties are omitted.

This”chapter deals with modern. design optimization of core type Power

transformers. Different optimization techniques” such as Genetic algorithms
(GA), Particle Swarm Optimization (PSO), Teaching and Learning based
Optimization (TLBO) ”are employed to solve important transformer. design
problems. Here 15 MVA, 66/11 kV transformer rating is considered”and
respective design optimization carried out. ”

7.2 Genetic Algorithm for TDO

• An algorithm is a set of instructions .that is repeated to solve a

problem.
• A genetic algorithm conceptually .follows steps inspired by the
biological processes of evolution.
• Genetic Algorithms follow the idea of. Survival of the fittest.
Better and better solutions .evolve from previous generations .until a
near optimal solution is obtained.
• A genetic algorithm is an iterative procedure that .represents its
candidate solutions as strings of genes .called chromosomes.
• The method learns by producing offspring .that are better and better
as measured .by a fitness function, which is a measure of the
.objective to be obtained (maximum or minimum). Generally applied
to spaces which .are too large.
7.3 Genetic Algorithm Concept:

GA is a method for solving optimization problems .based on natural selection,

the process that drives .biological evolution(41). The GA repeatedly modifies
a population.of initial solutions. At each step, it selects individuals .at random
from the current .population to be the parents. It then uses the parents to.
produce children for the next. generation. “Crossover and mutation operators
are applied to the .parents to generate new children. Over successive

Chiragkumar N Parekh 89
 7. TDO using Evolutionary Algorithms

generations, the population .evolves toward an optimal solution. In contrast to

more traditional numerical .techniques, the parallel nature of the stochastic
search done by GA often makes it very .effective in finding the global
optimum. GA is less susceptible to getting .stuck at local optima than gradient
search methods. Also, GA is much less sensitive to .initial conditions and is
widely used in various optimization .problems. Genes are the basic building
blocks of genetic algorithms. A gene is a binary encoding of a parameter. A
chromosome in a computer algorithm is an .array of genes. Each
chromosome .has an associated cost function, assigning .a relative merit to
that chromosome. The algorithm begins with a large list .of random
chromosomes. Cost functions are evaluated for each chromosome. The
chromosomes .are ranked from the most-fit to the least-fit according to .their
respective cost functions. Unacceptable Chromosomes are discarded leaving
a superior species subset of the original list. Genes that survive become.
parents, by swapping some of their genetic material to produce. two new
offspring. The parents reproduce enough to offset the discarded.
chromosomes. Thus, the total number of chromosomes remains constant for
each iteration. Mutations .cause small random changes .in a chromosome.
Cost functions are evaluated for the offspring and the mutated .chromosome,
and the process is repeated”. The algorithm stops after .a set number of
iterations, or when. an acceptable solution is obtained .

Here, 30 runs of the GA algorithm are performed. and the best solution is
chosen as the optimum one. The population type is bit string of. size equal to
20. A random initial population.is created, that satisfies the boundaries
and.linear constraints of the optimization problem. Rank fitness. scaling is
employed, scaling the raw scores based. on the rank of each individual, rather
than its score. Stochastic. uniform selection function. is used, which lays out a
line in which each parent. corresponds to a section of the line. of length
proportional to its expectation. The algorithm moves. along the line in steps of
equal. size, one step for. each parent. At each step, the algorithm allocates a.
parent from the section it lands on. ‘The first step is a uniform random.number
less than the step size. As far as mutation and crossover. functions are
concerned, the first one. is adaptive feasible (it randomly generates directions

Chiragkumar N Parekh 90
 7. TDO using Evolutionary Algorithms

that are adaptive with respect to the last. successful or unsuccessful

generation - a step length is chosen along each. direction so that linear
constraints and bounds are satisfied) and the second one is. scattered (it
creates a random binary vector, selects the genes where the. vector is a 1
from the first parent, and the genes where the. vector is a 0 from the second
parent, and combines. the genes to form the child.). The limit for the fitness.
function is set to 0.5, while the maximum number. of generations (iterations) is
equal to 500.

This chapter deals with modern. design optimization of core type Power
transformers. Four methods are presented. that solve important transformer.
design problems. Genetic algorithms, Particle Swarm Optimization, .Teaching
and Learning based .Optimization and NSGA.

7.4 GA Applied for Power Transformer Design Optimization:

Now a days there is tough competition in the market. The optimized design
will only help .the manufacturer to compete with. market and grab the orders.
So every transformer manufacturer has .focused on TDO, which shape the
transformer .of high efficiency with low cost. There are several .methodology
for design optimization. Among them The GA base optimization .is very
simple, reliable, efficient and robust. Here for GA based TDO several
variables .like K-Factor, flux density (Bm), Conductor geometry, numbers. of
coils and constraints .like NLL, Load Loss, %impedance, total cost etc. are
.considered.”

The “Genetic Algorithm generates billions of .possibilities to satisfy the

constraint and .provide one most nearest optimum output. As it is using
random number it happens to have different output. at every iteration. But
these all are very closer with each other. Here GA based 15 MVA 66/11 kV
transformer design optimization. carried out. Preliminary only 2 variables. viz.
Bm & K factor were considered for NLL .optimization. And Now it is expanded
for .bunch of above listed variables.”

Chiragkumar N Parekh 91
 7. TDO using Evolutionary Algorithms

7.5. Optimization Result of a Power Transformer using GA

The following input data are consider for Core type power transformer,
Q= 15000 KVA, V1=11000 volts, V2=66000 volts, 3-ph, f=50 Hz, vector Group
Dyn11

Figure 7.1 GA Toolbox for TDO and Respective Results

Chiragkumar N Parekh 92
 7. TDO using Evolutionary Algorithms

Figure 7.2 GA Toolbox TDO Results in Graphical Form.

There are many types of selection operators and .mutation techniques as

discussed earlier. Some of the design .results using different mutation
techniques.and selection operators.are as shown below table:

Chiragkumar N Parekh 93
 7. TDO using Evolutionary Algorithms

Table 7.1 TDO Result Using GA Toolbox For Different Mutation

Techniques and Selection Operators

Sr. Selection Roulette wheel Stochastic selection Tournament

No. techniques > selection selection
Mutation Constraint Adaptive Constraint Adaptive Constraint Adaptive
techniques > dependent feasible dependent feasible dependent feasible
1. k-factor 0.442 0.437 0.426 0.415 0.451 0.422
2. Bm (wb/m²) 1.613 1.652 1.639 1.558 1.601 1.598
3. DelLV(A/mm2) 2.21 2.43 2.471 2.454 2.524 2.561
2
4. DelHV(A/mm ) 2.751 2.673 2.614 2.531 2.518 2.617
5. Stiffener 2.41 2.23 2.1 2.42 2.09 2.482
6. No-load 10.44 10.57 10.31 10.56 10.40 10.53
losses (kW)
7. Load losses 66.73 67.75 68.41 66.29 65.95 68.36
(kW)
8. %Z 9.49 9.22 9.48 8.75 9.11 9.52
9. %η 99.17 99.32 99.19 99.21 99.29 99.20
10. GLV 10.5 9.45 9.36 9.04 10.29 9.46
11. GHV 21.77 22.43 21.98 22.44 21.79 21.83
12. Deflection 8.43 7.35 6..98 7.88 6.95 7.41
(mm)
13. Wcore (kg) 9170.4 8839.4 8682.0 9725.9 8870.8 8170.5
14. Wcu (kg) 3825 3772 3782.6 3490.5 3688.9 3745.1
15. Wconction (kg) 36.01 35.95 36.25 35.77 35.49 35.21
16. Winsu (kg) 280.49 266.62 264.32 269.31 273.33 263.96
17. Wframe (kg) 752.03 747.81 738.29 748.41 724.38 757.64
18. Wtankplate (kg) 1297.95 1185.56 1198.41 1249.42 1190.40 1105.32
19. Wstiffener (kg) 427.66 414.64 285.29 332.50 372.63 450.43
20. Ccore (Rs.) 1258595 1356435 1317397 1394642 1401613 1299709
21. Ccu (Rs.) 2101301 2059718 1988810 2066809 2119721 2055798
22. Coil (Rs.) 384755 361860 361722 379976 359890 362731
23. Cinsu (Rs.) 43974.1 41976.53 41882.58 42591.11 42791.11 43150.19
24. Ctank (Rs.) 118172.2 120984.1 114303.5 111645.2 113829.2 109832.1
25. Total cost 6205692.2 6230566 6195511 6290520 6304753 6348652
(Rs.)

As show in table 7.1 using GA in optimization .toolbox, there are various

mutation techniques and selection .operator are used for result comparison.
Here, as the .objective function is cost minimization so if we select .a design

Chiragkumar N Parekh 94
 7. TDO using Evolutionary Algorithms

as per cost then we can select a roulette wheel-selection and adaptive

feasible mutation .technique. Analysis of both the proposed techniques .for a
given power transformer is below.

Case-I

80
60
40
20
0
Closs (kW) Deflection(mm) Cost (lac)

ESM GA

Figure 7.3 Comparison of ESM and GA (Case-I)

Table 7.2 Comparison of ESM and GA (Case-I)

ESM GA

Closs (kW) 9.152 10.06

Deflection (mm) 8.9 6.9

Cost (lac) 64.74 61.9

From the above figure 7.3, it is clear that .the no load losses obtained by
using ESM method .are less than that of GA but correspondingly .the
deflection is greater. in ESM. Due to less deflection maintenance, .cost of tank
reduced. On welded joint .oil leakage can be avoided.

Chiragkumar N Parekh 95
 7. TDO using Evolutionary Algorithms

CASE-II

100

ESM
GA
0
Gradient of Loadloss (kW) Cost(lakh)
HV

Figure 7.4 Comparison of ESM and GA (Case-II)

Table 7.3 Comparison of ESM and GA (Case-II)

ESM GA
Gradient 22.79 22.70
Load losses 67.60 67.17
Cost 64.17 61.9

From the above figure 7.4, it is clear that. the load losses obtained .by using
ESM method are less than that .of GA but correspondingly the cost obtained
by using ESM .method are high than that of GA so manufacturing .cost is high
as well as operating .cost is high in ESM.

CASE-III

100

50 ESM
GA
0
Closs (kW) Deflection Cost (lakh)
(mm)

Figure 7.5Comparison of ESM and GA (Case-III)

Table 7.4 Comparison of ESM and GA (Case-III)

ESM GA
No-Load losses (kW) 9.991 10.06
Deflection (mm) 7.9 6.9
Cost (lac) 61.36 61.9

Chiragkumar N Parekh 96
 7. TDO using Evolutionary Algorithms

From the above figure 7.5, it is clear that the manufacturing cost obtained by
using ESM method are less than that of GA but correspondingly the deflection
obtained by using ESM method are high than that of GA. Ultimately it will
affect to the transformer tank.

Table 7.5 Analysis of ESM and GA Technique for TDO

ESM GA

This method provides all the possible Optimization toolbox gives the final
.variables which are quite a large in value .of variables which satisfy .the
.number. constraint values.

Manual selection of optimum design It gives the value of variables .where

based on .application point of view. the fitness .function is minimum.

Numbers of iterations are more. Numbers of generations are less.

It is time consuming and tedious Less time is required to get optimum

method. value.
It is less accurate. It is more accurate.

In previous method, optimization toolbox is used to optimize. design of

transformer .using genetic algorithm. Now, same approach .is used in
MATLAB but variable as well as constraints .are increased, made a program
and optimized without .use of optimization toolbox.

The list of variable and constraint .are listed in Table 7.6 and Table 7.7
respectively.

7.6 Encoded Genetic Algorithm for Power Transformer:

“A self-developed Binary encoded GA code is applied with. population size
20, generations 250, mutation. probability 0.01, crossover probability 1, and
chromosome. length 48. The constraints are handled using penalty. function
method. The ranges for variables.were specified and each variable like
K-factor, Flux density,. HV & LV Winding and conductor geometry.allowed to
vary. Other customer specifications such as cost, .no load loss, load loss, etc.
became constraints. for the problem stated. The code was run several times
and the respective .result retrieve was the best among them.”

Chiragkumar N Parekh 97
 7. TDO using Evolutionary Algorithms

‘15 MVA Transformer Design was made. by encoded GA Optimization which

comprises Multiple Variables & constraints .to obtain Optimum NLL-Load.
Loss-% Impedance & Total Cost. The variables with respective. ranges are
shown in Table 7.6. The program .pattern is shown in. Figure 7.6 and Figure
7.7 informs about result output.’

Table 7.6 List of Variables, Range and Iteration Possibility for GA-TDO
Variable Range Possibilities
K 0.40 to 0.55 16
Bm 1.55 to 1.70 16
HVbht 8.1 to 16 80
HVbth 1.5 to 4.5 31
HVapc 1 to 2 2
HVrpc 1 to 4 4
LVbht 8.1 to 16 80
LVbth 1.5 to 4.5 31
LVapc 1 to 2 2
LVrpc 1 to 16 16
HVdisc 50 to 70 OR 30 to 50 11
(even)
LVdisc 20 to 40 (even) 11

Total Possibility: (16*16*80*31*2*4*80*31*2*16*11*11) = 4.87718 x 10 13

7.6.1 Constraints

g(1,:)=Ctotal./7500000-1;
g(2,:)=LoadLoss./70000-1;
g(3,:)=8.7./Z-1;
g(4,:)=Z./10.45-1;
g(5,:)=99./effi-1;
g(6,:)=0.4033./k-1;
g(7,:)=k./0.55556-1;
g(8,:)=1.55./Bm-1;
g(9,:)=Bm./1.7-1;
g(10,:)=Closs./10000-1;

Chiragkumar N Parekh 98
 7. TDO using Evolutionary Algorithms

(A)

(B)

Figure 7.6 GA TDO Coding for 15 MVA Transformer

(A) & (B)

Chiragkumar N Parekh 99
 7. TDO using Evolutionary Algorithms

7.6.2 GA-TDO Result :

The Encoded GA-TDO program was executed several. times to get the
optimum. design. As it was following the rendom .numbers every time,It was
producing the new. optimum design.All were complying the asked
constraints.The acquired. design were printed and summarised in tabular
form.The same. is shown in Table 7.7.

Figure7.7 GA-TDO Result for 15 MVA Transformer.

Chiragkumar N Parekh 100

 7. TDO using Evolutionary Algorithms

Table 7.7 Result Summary of GA-TDO Coding Program

Variables and
Result 1 Result 2 Result 3 Result 4 Result 5
Constraints

K factor 0.41 0.42 0.423 0.40 0.41

Bm(Wb/ m²)) 1.56 1.57 1.55 1.56 1.52

HVCon. Ht.(mm) 11.30 10.8 8.1 9.0 10.4

HVCon.thk.(mm) 4 3 1.8 3.6 3.1

ApcHV 1 1 1 1 1

RpcHV 1 1 2 1 1

LVCon. Ht.(mm) 16 16 13 14.4 13.2

LVCon.thk.(mm) 1.6 3.6 2.2 2.8 2.5

ApcLV 2 2 2 2 2

RpcLV 5 3 6 5 4

HV Discs 66 68 60 64 62

LV Discs 40 34 40 32 38

Load Loss( kW) 68.06 68.64 67.9 69.56 68.29

No Load Loss( kW) 9.20 9.62 9.72 9.58 9.34

%Impedance(%Z) 10.39 8.85 10.01 9.92 10.02

%η 99.24 99.39 99.10 99.28 99.34

GLV 9.70 10.22 9.75 9..58 10.10

GHV 20.76 21.07 20.27 21.13 21.08

Deflection (mm) 7.92 7.64 6.91 7.18 6.99

Cost (INR-Lacs) 58.75 60.96 60.527 59.992 60.415

Chiragkumar N Parekh 101

 7. TDO using Evolutionary Algorithms

7.7 Particle Swarm Optimization (PSO)

• PSO is a robust stochastic .optimization technique based on the

movement .and intelligence of swarms.
• PSO applies the concept of social .interaction to problem solving.
• It was developed in 1995 by James .Kennedy (social-psychologist)
and .Russell Eberhart (electrical engineer).
• It uses a number of agents (particles) that .constitute a swarm
moving around in the search .space looking for the best solution.
• Each particle is treated as a point .in a N-dimensional space which
adjusts .its “flying” according to its own flying experience .as well as
the flying experience .of other particles.
• Each particle keeps track of .its coordinates in the solution space
which are .associated with the best solution (fitness) that has
achieved so far .by that particle. This value is called .personal best,
pbest.
• Another best value that is tracked by .the PSO is the best value
obtained so far by any .particle in the neighborhood of that .particle.
This value is called gbest.
• The basic concept of PSO lies in accelerating each particle toward
its pbest and .the gbest locations, with a random.weighted
acceleration at each time step.
• Each particle tries to modify .its position using the following
information:”
 the current positions,
 the current velocities,
 the distance between the. current position and pbest,
 the distance between the current. position and the gbest.
• The modification of the particle’s. position can be mathematically
modeled according the following equation:

Chiragkumar N Parekh 102

 7. TDO using Evolutionary Algorithms

Vik+1 = wVik +c1 rand1(…) x (pbesti-sik) + c2 rand2(…) x (gbest-sik)…[1]

Where,
vik : velocity of agent i at iteration k,
w: weighting function,
cj : weighting factor,
rand : uniformly distributed random number between 0 and 1,
sik : current position of agent i at iteration k,
pbesti : pbest of agent i,
gbest: gbest of the group.

7.7.1 Encoded Particle Swarm Optimization (PSO) Technique for Power

Transformer:
“A self-developed PSO code was applied with population size 70, generations
700. The constants c1 and c2 are chosen as 1.49445 as suggested in the
literature. Similarly, inertia weight w is selected as 0.4. The constraints were
handled using penalty function method. The ranges for variables were
specified and each variable like K-factor, Flux density, HV & LV Winding and
conductor geometry is allowed to vary. Other customer specifications such as
cost, no load loss, load loss, etc. became constraints for the problem stated.
The code was run several times and the respective result retrieve was the
best among them.”

15 MVA Transformer Design was made by .encoded PSO method which

comprises Multiple Variables &.constraints to obtain Optimum NLL-Load
Loss-%Impedance &.Total Cost. The program pattern is shown. in Figure 7.7
and Figure 7.8 informs about result output.’

Chiragkumar N Parekh 103

 7. TDO using Evolutionary Algorithms

7.7.2 Case Study:

(A)

(B)
Figure 7.8 PSO-TDO Coding for 15 MVA Transformer (A) & (B)

Chiragkumar N Parekh 104

 7. TDO using Evolutionary Algorithms

7.7.3 PSO-TDO Result :

The Encoded PSO-TDO program was executed several times to get the
optimum design. As it was following the rendom numbers every time,It was
producing the new optimum design.All were complying the asked
constraints.The acquired design were printed and summarised in tabular
form.The same is shown in Table 7.8

Figure 7.9 PSO-TDO Result of 15 MVA Transformer

Chiragkumar N Parekh 105

 7. TDO using Evolutionary Algorithms

Table 7.8 Result Summary of PSO-TDO Coding Program

Variables and
Result 1 Result 2 Result 3 Result 4 Result 5
Constraints

K factor 0.41 0.421 0.406 0.441 0.434

Bm(Wb/ m²)) 1.55 1.54 1.52 1.58 1.56

HVCon. Ht.(mm) 8.5 11.2 7.9 9.4 9.8

HVCon.thk.(mm) 2.3 2.9 2.7 1.8 3.6

ApcHV 1 1 1 1 1

RpcHV 2 1 1 2 2

LVCon. Ht.(mm) 16 14.5 12.8 14.2 15.4

LVCon.thk.(mm) 1.5 2.7 1.8 2.1 2.0

ApcLV 2 2 2 2 2

RpcLV 6 4 5 7 5

HV Discs 67 66 70 68 64

LV Discs 40 40 40 38 36

Load Loss( kW) 69.90 68.29 66.56 68.26 68.76

No Load Loss( kW ) 9.70 9.72 9.79 9.67 9.46

%Impedance(%Z) 10.40 8.79 8.95 9.56 9.11

%η 99.15 99.24 99.22 99.19 99.26

GLV 8.92 9.46 9.22 9..68 9.39
GHV 20.24 20.82 20.38 21.01 20.71
Deflection (mm) 7.80 7.84 6.49 7.23 6.89
Cost (INR-Lacs) 62.80 61.85 60.93 62.31 62.00

Chiragkumar N Parekh 106

 7. TDO using Evolutionary Algorithms

7.8 Teaching Learning Based Optimization (TLBO) Algorithm

Teaching-learning is an important process where every individual tries to

learn something from other individuals to improve themselves. It
simulates the traditional teaching learning phenomenon of a classroom.
The algorithm simulates two fundamental modes of learning.
Through the teacher (teacher phase) and Interacting with other learners
(learner phase).”
The influence of a teacher on the output of learners in a class. The
algorithm mimics the teaching–learning ability of teacher and learners in
a classroom.
“TLBO methodology is more accurate, does not require any derivative
and follows the entire path to find its solution.
All the evolutionary and swarm intelligence based algorithms require common
controlling parameters like population size and number of generations.
Besides the common control parameters, different algorithms require their
own algorithm specific control parameters. For example, GA uses mutation
rate and crossover rate; PSO uses inertia weight, social and cognitive
parameters; ABC uses number of bees (employed, scout and onlookers) and
limit; HS requires harmony memory consideration rate, pitch adjusting rate
and the number of improvisations; ACO requires exponent parameters,
pheromone evaporation rate and reward factor; etc. Sometimes, the difficulty
in the selection of algorithm specific control parameters increases with
modifications and hybridization. The proper tuning of the algorithm- specific
parameters is a very crucial factor which affects the performance of the
optimization algorithms.”
‘The improper tuning of algorithm-specific parameters either increases the
computational effort or yields the local optimal solution. Considering this
aspect, we have recently introduced the Teaching-Learning-Based
Optimization (TLBO) algorithm which does not require any algorithm-specific
control parameters. TLBO requires only common controlling parameters like
population size and number of generations (and elite size, if considered) for
its working. Thus, TLBO can be said as an algorithm-specific parameter-less
algorithm.’

Chiragkumar N Parekh 107

 7. TDO using Evolutionary Algorithms

TLBO is a teaching-learning process inspired algorithm based on the effect of

influence of a teacher on the output of learners in a class. Teacher and
learners are the two vital components of the algorithm and describes two
basic modes of the learning, through teacher (known as teacher phase) and
interacting with the other learners (known as learner phase). The output in
TLBO algorithm is considered in terms of results or grades of the learners
which depend on the quality of teacher. So, teacher is usually considered as a
highly learned person who trains learners so that they can have better results
in terms of their marks or grades. Moreover, learners also learn from the
interaction among themselves which also helps in improving their results.
TLBO is population based method. In this optimization algorithm a group of
learners is considered as population and different design variables are
considered as different subjects offered to the learners and learners’ result is
analogous to the ‘fitness’ value of the optimization problem. In the entire
population the best solution is considered as the teacher. The working of
TLBO is divided into two parts, ‘Teacher phase’ and ‘Learner phase’. Working
of both the phases is explained below.”
7.8.1 Teacher Phase
This phase of the algorithm simulates the learning of the students (i.e.
learners) through teacher. During this phase a teacher conveys knowledge
among the learners and puts efforts to increase the mean result of the class.
Suppose there are ‘m’ number of subjects (i.e. design variables) offered to ‘n’
number of learners (i.e. population size, k=1,2,…,n). At any sequential
teaching learning cycle i, Mj,i be the mean result of the learners in a particular
subject ‘j’ (j=1,2,…,m). Since a teacher is the most experienced and
knowledgeable person on a subject, so the best learner in the entire
population is considered as a teacher in the algorithm. Let Xtotal-kbest,i is the
result of the best learner considering all the subjects, who is identified as a
teacher for that cycle. Teacher will put maximum effort to increase the
knowledge level of the whole class, but learners will gain knowledge
according to the quality of teaching delivered by a teacher and the quality of
learners present in the class. Considering this fact the difference between the
result of the teacher and mean result of the learners in each subject is
expressed as,

Chiragkumar N Parekh 108

 7. TDO using Evolutionary Algorithms

Difference_Meanj,i= ri (Xj,kbest,i - TFMj,i) [1]

Where, Xj,kbest,i is the result of the teacher (i.e. best learner) in subject j. TF
is the teaching factor which decides the value of mean to be changed, and ri
is the random number in the range (0, 1). Value of TF can be either 1 or 2.
The value of TF is decided randomly as,
TF = round (1+rand(0,1){2-1}) [2]

TF is not a parameter of the TLBO algorithm. The value of TF is not given as

an input to the algorithm and its value is randomly decided by the algorithm
using Eq. [2].
Based on the Difference_Meanj,k,i, the existing solution is updated in the
teacher phase according to the following expression.
X'j,k,i=Xj,k,i+Difference_Meanj,k,I [3]

whereX'j,k,i is the updated value of Xj,k,i. X'j,k,i is accepted if it gives better

function value. All the accepted function values at the end of the teacher
phase are maintained and these values become the input to the learner
phase.

7.8.2 Learner Phase

This phase of the algorithm simulates the learning of the students through
interaction among themselves. The students can also gain knowledge by
discussing and interacting with the other students. A learner will learn new
information if the other learner has more knowledge than him or her. The
learning phenomenon of this phase is expressed below.
Randomly two learners P and Q are selected such that X'total-P,i ≠ X'total-Q,i
(where, X'total-P,I and X'total-Q,i are the updated values of Xtotal-P,i and
Xtotal-Q,i respectively at the end of teacher phase)
X''j,P,i = X'j,P,i + ri (X'j,P,i - X'j,Q,i), If X'total-P,i >X'total-Q,I [4]
X''j,P,i = X'j,P,i + ri (X'j,Q,i - X'j,P,i), If X'total-Q,I >X'total-P,I [5]
(Above equations is for maximization problem, reverse is true for minimization
problem)
X''j,P,i is accepted if it gives a better function value.

Chiragkumar N Parekh 109

 7. TDO using Evolutionary Algorithms

7.8.3 Encoded TLBO code Technique for Power Transformer:

‘The TLBO code was applied with class size 50 and generations 500. The
constraints were handled using penalty function method. The ranges for
variables were specified and each variable like K-factor, Flux density, HV &
LV Winding and conductor geometry is allowed to vary. Other customer
specifications such as cost, no load loss, load loss, etc. became constraints
for the problem stated. The code was run several times and the respective
result retrieve was the best among them.’

15 MVA Transformer Design was made by encoded TLBO method which

comprises Multiple Variables & constraints to obtain Optimum NLL-Load
Loss-%Impedance & Total Cost. The program pattern is shown in Figure 7.10
and Figure 7.11 informs about result.’

Chiragkumar N Parekh 110

 7. TDO using Evolutionary Algorithms

Figure 7.10 TLBO-TDO Coding for 15 MVA Power Transformer

7.8.4 TLBO-TDO Result:

The Encoded TLBO-TDO program was executed several times to get the
optimum design. As it was following the rendom numbers every time, It was
producing the new optimum design. All were complying the asked
constraints.The acquired design were printed and summarised in tabular
form.The same is shown in Table 7.9

Figure 7.11 TLBO-TDO Result for 15 MVA Transformer

Chiragkumar N Parekh 111

 7. TDO using Evolutionary Algorithms

Table 7.9 Result Summary of TLBO-TDO Coding Program

Variables and
Result 1 Result 2 Result 3 Result 4 Result 5
Constraints
K factor 0.409 0.417 0.413 0.42 0.406

Bm(Wb/m²) 1.55 1.55 1.52 1.53 1.52

HVCon. Ht.(mm) 12.1 9.9 9.5 10.2 13.1

HVCon.thk.(mm) 2.4 1.7 2.4 2.2 2.7

ApcHV 1 1 1 1 1

RpcHV 1 2 1 1 1

LVCon. Ht.(mm) 12.2 10.5 11.8 12.3 9.8

LVCon.thk.(mm) 2.4 2.4 2.7 2.0 1.8

ApcLV 2 2 2 2 2

RpcLV 7 4 5 5 6

HV Discs 68 68 66 68 70

LV Discs 40 40 38 36 40

Load Loss( kW ) 69.90 67.94 67.71 69.18 68.44

No Load Loss( kW ) 9.04 9.62 9.48 9.67 9.39

%Impedance(%Z) 10.44 8.99 9.10 9.42 8.88

%η 99.09 99.24 99.22 99.36 99.41

GLV 9.82 9.54 9.66 9.23 9.89
GHV 20.79 20.63 20.52 21.01 20.93
Deflection (mm) 7.55 7.43 6.88 6.94 6.80
Cost (INR-Lacs) 62.71 61.87 62.56 62.02 62.15

Chiragkumar N Parekh 112

 7. TDO using Evolutionary Algorithms

7.9 Comparison of Different Optimization Techniques Result of Power

Transformer Design Optimization:
Binary encoded GA, PSO and TLBO program was developed for Power
Transformer design optimization among to utilize various variable to vary and
satisfy the listed various constraints. All techniques was performed
independently and the respective results were summarized in table for
comparison. Table 7.10 is the summary of results by all three techniques
including conventional design output.

TABLE 7.10 Result Comparison of Conventional TDO, GA-TDO ,PSO-

TDO & TLBO-TDO
Variables and TDO- GA-TDO PSO-TDO TLBO-
Constraints Conventional TDO
Design
Methodology
K factor 0.44 0.42 0.44 0.40
Bm(Wb/ m²)) 1.54 1.57 1.55 1.555
HVCon. Ht.(mm) 12.8 10.8 8.1 9.0
HVCon.thk.(mm) 3.3 3 1.8 4
ApcHV 1 1 1 1
RpcHV 1 1 2 1
LVCon. Ht.(mm) 11.8 16 16 14.4
LVCon.thk.(mm) 3.3 3.6 1.9 3.8
ApcLV 2 2 2 2
RpcLV 8 3 6 3
HV Discs 60 66 74 64
LV Discs 40 34 40 34
Load Loss( kW ) 72.114 70.645 69.9 69.99
No Load Loss( kW ) 9.49 9.486 9.72 9.09
%Impedance(%Z) 9.5 9.78 10.34 10.44
%η 99.21 99.37 99.24 99.16
GLV 8.86 9.17 9.54 8.93
GHV 21.20 20.75 21.07 21.11
Deflection (mm) 7.32 7.46 7.08 6.97
Cost (INR-Lacs) 63.41 60.11 62.38 61.87

Chiragkumar N Parekh 113

 7. TDO using Evolutionary Algorithms

12

10

8
CD-TDO

6 GA-TDO
PSO-TDO

4 TLBO-TDO

0
K-factor Bm %Z

(A)
80

70

60

50
CA-TDO

40 GA-TDO
PSO-TDO
30
TLBO-TDO

20

10

0
%Z Load Loss(kW) NLL(kW) Cost (INR-lakh)

(B)

Figure 7.12 Graphical Form of Result Comparison (Conventional TDO,

GA-TDO,PSO-TDO & TLBO-TDO.(A) Comparison for K-factor, Bm & %Z
(B) Comparison for Losses, %Z & Cost of the Transformer.

Chiragkumar N Parekh 114

 7. TDO using Evolutionary Algorithms

7.10 Conclusion

In this chapter, the main objective was to develop and compare the
performance of different evolutionary optimization techniques for transformer
design optimization problem. The range of design variables have been
chosen as per industrial code of practice. Optimal design dimensions and
performance parameters of transformer have been obtained by varying the
control parameters of GA, PSO and TLBO. Among all the three techniques, it
was observed that GA has better robustness as compared to other
techniques. The proposed method is very effective as GA is more likely find
the global optimum because of their population based approach. A saving of
4.00 % obtained by GA Based TDO, which can be very helpful reference for
transformer designer to present strongly during tendering and order finalizing
stage.”

Chiragkumar N Parekh 115