Академический Документы
Профессиональный Документы
Культура Документы
1.1 Introduction
A Fault Current Limiter (FCL) is a device which limits the prospective fault current when a
fault occurs in a power system. Fault current limiting devices have been gaining interest in recent
times to combat large fault currents. Also low rating circuit breakers can be used in this case as
they are very cost effective compared to breaker replacement. The term FCL is generally applied
to superconducting devices, whereas non-superconducting FCL devices are typically termed
Fault Current Controllers.
Depending on the function and the main characteristics of each FCL type, we compare to
what extent each type can fulfill the main requirements of an ideal limiter. The major
requirements of a fault current limiter device are:
Low AC loss and voltage drop and low impedance in the normal state
Capability to perform gradual change of impedance from normal state to fault state
and vice-versa.
into commercially viable product have culminated in only a few practical designs and even fewer
working prototypes. Many of these designs have shortcomings (e.g. size, performance,
reliability, recovery under load, or cost) that hinder them from reaching full commercial
potential. Most SFCL designs use the aforementioned quench behavior to limit fault currents
within the first cycle. However, that is often where the similarities end as each SFCL design has
its own methods of sustaining the limiting action once the superconductor becomes resistive.
Described below are brief & simplified overviews of SFCLs, categorized into three types.
The first two types depend on the quenching action of superconductors and the third type uses
DC HTS magnet windings to saturate an iron core.
1.2.1.1 Resistive SFCL
Resistive SFCL utilize the superconducting material as the main current carrying conductor
under normal grid operation. The principle of their operation is shown in Fig1.1
When a fault occurs, the current increases and causes the superconductor to quench thereby
increasing its resistance exponentially. The current level at which the quench occurs is
determined by the operating temperature, and the amount and type of the superconductor. The
rapid increase in resistance produces a voltage across the superconductor and causes the current
to transfer to a shunt, which is a combined inductor and resistor. The shunt limits the voltage
increase across the superconductor during a quench. In essence, the superconductor acts like a
switch with millisecond response that initiates the transition of the load current to the shunt
impedance.
Fig.1.2. Normalized plot of voltage and current in a superconductor at constant temperature and magnetic field.
The above Fig. 1.2 shows voltage across RSC as a function of ration of current through the
device, Iline , to the "critical current" , Ic which is define as the current at which a voltage drop
of 1V /cm is observed along the conductor. The grid characteristic of the resistive SFCL after a
quench is determined by the shunt element. Thus, because the shunt is typically quite reactive, a
resistive SFCL typically introduces significant inductance into the power system during a fault.
During the transition period when current is being transferred from the superconductor to the
shunt, the voltage across the combined element (shown in the normalized plot above) is typically
higher than it is after the current has transitioned into the shunt. The dynamics of this process
depend on the two elements and their mutual inductance. The quench process in resistive SFCLs
results in heat that must be carried away from the superconducting element by the cryogenic
cooling system. Typically, there is a momentary temperature rise in the superconducting element
that causes a loss of superconductivity until the cryogenic system can restore the operating
temperature. This period of time, known as the recovery time, is a critical parameter for utility
systems (which may see multiple fault events occurring close together in time) and is a key
distinguishing characteristic among various SFCL designs.
The principle of the Shielded-Core SFCL and its function is similar to a transformer with a
short-circuited secondary winding. As long as the induced current in the superconducting
cylinder is lower than the critical current, the superconducting cylinder serves as a perfect
magnetic shielding for the primary winding. The flux in the iron core is negligible and the
residual impedance is low. With increasing short-circuit current the current in the
superconducting cylinder exceeds the critical current and the cylinder starts to develop a
resistance.
Although the superconductor in the shielded-core design has to re-cool after a limiting
action just like the resistive type, non-uniform heating of the superconductor (i.e. hot spots) is
easier to avoid through optimization of the turns ratio. A major drawback of the shielded-core
5
technology is that it is approximately four times the size and weight of purely resistive SFCLs.
Although prototypes of shielded-core designs have worked well, their size and weight have
limited grid deployment.
1.2.1.3 Saturable -Core SFCL
Saturable core SFCLs utilizes the dynamic behavior of the magnetic properties of iron
(nonlinear permeability of the iron core, thus no problem of recovery time) to change the
inductive reactance of the ac line. The concept (shown in Fig. 1.4. and Fig. 1.5.) utilizes two iron
core and two AC windings for each phase. The AC windings are made of conventional
conductors, wrapped around the core to form an inductance in series with the AC line. A
constant current superconducting winding on the iron core provides a magnetic bias. . At the
same time, the high temperature superconducting coils are supplied by a DC source & do not
have AC power loss.
Under nominal grid conditions (when the AC current does not exceed the maximum rating
of the local system), the HTS(High Temperature Superconductor) coil fully saturates the iron so
that it has a relative permeability of one. To the AC coils, the iron acts like air, so the AC
impedance (inductive reactance) is similar to that of an air-core reactor. Under fault conditions,
the negative and positive current peaks force the core out of saturation, resulting in increased line
impedance during part of each half cycle. The result is a considerable reduction in peak fault
current.
Essentially, the saturable-core SFCL is a variable-inductance iron-core reactor that has the
impedance of an air-core reactor under normal grid conditions and very high impedance during
fault events. Unlike resistive SFCLs, which may require time between limiting actions to cool
the superconducting components, the saturable-core approach can manage several actions in
succession because the superconductor does not quench. However, a major drawback of
saturable-core SFCL technology is the volume and weight associated with the heavy iron core;
however, manufacturers hope to improve this issue in future prototypes. Zenergy has recently
tested a prototype saturable-core SFCL based on an entirely new design concept that is four
times smaller than its predecessor. GridON, an Israeli-based startup company, is in the process of
developing saturable-core concept intent on reducing size and weight to more accommodating
levels for commercial use.
The main thing to be considered with this approach is the mechanism associated with the
operation of the bypass switch. The other method used, based on this approach, is the current
limiting protector which is mainly used in the distribution systems. In this method the sensing
and the switch operation mechanism are separated from each other.
This circuit consists of silver sand fuse and a copper conductor which is parallel to the fuse
and has s series of notches. The fuse melts upon fault; this is accomplished by high arc voltage
which limits current. The melted fuse and the notches on the copper conductor limit the fault
current. The sensor detects the fault current.
1.2.2.2 Solid State FCLs
Solid-state FCLs utilize actively controlled power electronics to perform or initiate the
limiting action. Typically, the power electronics switch the fault current into a limiting branch or
open the circuit in such a way that the characteristics simulate a circuit breaker. One topology
being developed is the solid-state FCL circuit breaker (SSFCL-CB) shown in the Fig. 1.7. below,
a device that utilizes a diode rectifier bridge circuit with a thyristor-based commutating circuit
that can be switched off during a fault event. SSFCL that utilize thyristors in the bridge circuit
rather than diodes is currently in development which will provide more control of the fault
current through adjustment of the firing angles of the thyristors. EPRI is currently developing an
SSFCL that utilizes thyristors in the bridge circuit rather than diodes.Because the fault current is
8
controlled dynamically through the use of power electronics, the SSFCL is referred to as a faultcurrent controller (FCC).In addition to SSFCL-CB and FCC, designs that use solid-state
switching to direct fault current into inductive coils or detuned LC series circuits arealso being
pursued.
There are advantages and disadvantages associated with each design. Solid-state devices
offer the advantage of additional control of fault currents not available with other technologies
and have the potential to be realized in small geometries. However, issues with solid-state
devices include the reliability of switching electronics to sense fault levels and respond
accordingly as well as possible distortion of the current waveform due to the nature of switching
electronics. To date, no full-scale prototype of a solid-state FCL has achieved grid deployment.
value. Due to switching action there might be a spark across the switch because of the voltage
appearing across one of the capacitors. If one of the components, the capacitor or reactor
becomes faulty, the current limiting action of the circuit will no longer be valid. The cost and
large size are its main drawbacks.
drawback of this approach is that the current to be interrupted depends on the sensor response
and the operating time of the mechanism.
11
Fig. 2.1 shows the structure of the SICSFCL. It is composed of a reactor system, a
cryogenic system, a DC bias system and a monitoring system. The reactor system is the main
functional part. The cryogenic system provides low-temperature circumstance for the
superconducting coil. The DC bias system has three basic functions: fully magnetizing iron
cores during normal condition, cutting off magnetization circuit if a fault current occurs, and
fast recovering from high impedance to low after the fault is cleared. While the monitoring
system carries out the online monitoring of the status of whole system.
12
13
Fig. 2.4 is the schematic diagram of DC bias system, which includes DC source, switch,
current sensor and energy release circuit etc. DC source supplies exciting current for the
14
superconducting coil. Current sensor is used to detect and determine the fault current.The DC
bias system has the following operating characteristics
At the steady state, the DC source output constant exciting current.
The switchers can cut-off DC current in 5 milliseconds once a fault current is
detected. The energy release circuit absorbs the energy of the superconducting
coil in dozens of milliseconds.
After the clearance of fault current, the DC bias system recovers to normal DC
current within hundreds of milliseconds. The SICSFCL turns to low impedance
before the breaker close.
15
Chapter 3
SFCL
3.1 Introduction
There is renewed interest in superconducting FCLs due to the advantages offered by
high-temperature superconductors (HTS) using liquid nitrogen over their low-temperature
counterparts. This predominantly arises from an order of magnitude of savings in cryogenic
equipment costs, and a reduction in operational costs by a factor of up to 100. Liquid
nitrogen systems are also less complex compared to liquid helium systems.
The prototype saturable magnetic core-type FCL considered in this study is constructed
using HTS tapes. The application of HTS tapes in non-quench type SFCLs is most desirable
from commercial and operational viewpoints; they are the most robust, practicable, and
reliable forms available compared to other HTS forms. A description of the design principles
applied to an HTS saturable core-type SFCL is provided. The concepts behind development
of a time-domain model for the SFCL are also presented, followed by simulation results in
relation to the operation of the SFCL for typical medium-voltage (MV) distribution
substation applications.
16
respectively, when a dc bias current Id is imposed. Curve 4 is the i curve of the saturated
core FCL, which is achieved by summation of curves 2 and 3.
Fig.3.1. Simulation circuit for a saturated core FCL; 1,2) AC windings; 3,4) Magnetic Cores; 5)
Superconducting DC coils
In the normal operation, both magnetic cores are driven into saturation by the dc current.
The saturated core FCL works in line segment I, so the impedance of the FCL is very low. When
a short-circuit current occurs, the rapidly increasing ac current drives both magnetic cores out of
saturation alternatively during one ac cycle, hence, the saturated core FCL works in line segment
II, the impedance of the FCL becomes so large that the fault current is limited. However, if the
ac current increases unceasingly, the magnetic cores will be saturated reversely, when the FCL
17
works in line segment III and will lose the capability of limiting current, due to the low
impedance.
18
In a saturated core FCL a DC current carrying winding is used to bias the core. The number
of ampere-turns supplied by the biasing winding needs to be high enough to sufficiently saturate
the core. A separate winding is then used to carry the normal AC load current. During steady
state un-faulted conditions the low current in the AC winding results in the flux density
oscillating through a minor loop and not the full B-H loop. The top right hand corner of the Fig.
2.1 above shows the region where M3 laminated steel saturates. During un-faulted conditions the
flux density is oscillating within this Steady state region. As can be seen in the figure, the
permeability of the material at this field intensity is approximately equal to the permeability of
air; hence, the impedance of the AC winding is equivalent to that of an air-cored inductor.
During a fault event the rising current causes the flux density to oscillate through a much larger
region of the B-H loop. As the core de-saturates and moves into the Faulted state region shown
in the figure, the permeability of the core material increases. Hence, the impedance of the AC
winding also increases, subsequently limiting the fault current.
A schematic diagram of superconducting fault current limiter is shown below in the Fig. 3.4
Here two magnetic cores are required to cater for each half cycle of the AC fault current. Each
core carries a coil in which the AC line current iac flows. An HTS coil carrying a DC bias current is
also wound on each core as illustrated, where the DC bias current Idc is common to both coils.
Magneto Motive Force (MMF) produced by the superconducting dc winding is given by NdcIdc=
2(2W+2h) Hdc where Ndc is the number of turns in the DC winding, Idc is the current in the DC
19
winding, W is the average core width, h is the average core height and Hdc is the magnetic field
intensity due to the DC current. The FCL inductance under normal power system conditions needs to
be as small as possible. Therefore, the normal operating point is selected such that Hdc>>Hknee ,
where the saturation knee point is indicated inFig. 2.5.
When a fault current flows through the ac winding, each core is alternatively taken out of
saturation in each half cycle of as a result of the relatively large MMF caused by the fault current. To
ensure that the FCL offers a large inductance during a fault, the MMF cancelation in each core
should be such that
20
Several figures of merit for a device can be taken from its B-H curve, the most important of
which are the saturation level, the retentivity (or remanence) and the coercivity (or coercive
force). The saturation level (Bsat) is the magnetization level at which all of the domains'
magnetic moments are pointing in the direction of the external field vector. This level, a constant
for the material, represents the maximum field density the material can sustain. The remanence is
defined as the amount of magnetism that remains in the material after the external forces have
been removed. On the B-H curve, it is the point at which the curve crosses the abscissa (H equal
to zero). The coercive force is defined as the external magnetic field required to return the
magnetism of the solid to zero. On the B-H curve, it is the point at which the curve crosses the
ordinate (B equal to zero). These parameters relate physical occurrences within a core to its
magnetic response and provide limits by which mathematical relationships can be derived to
model the magnetic behavior within the core.
21
The relationship between B and H is defined by the permeability () of the material. For
magnetic materials, the permeability () can be related to the permeability of free space (0) by
the relative permeability (r). A particular class of magnetic materials (saturable) can exhibit
highly nonlinear relative permeabilities. The relationship between these quantities is shown in
the following equation
= = 0 = 0 ( + )
( 3.1 )
The parameter M in the equation refers to the magnetism or field intensity within the
material that is contributed by the magnetic domains. The variation in permeability () is a result
of the changing magnetism (M) of the solid as energy from the applied field (H) is absorbed into
the domains. Since the relative permeability (r) is dependent on the ratio of applied and resident
fields (magnetic susceptibility), the field density equation may be expanded into the more
general relationship given by the final relationship, in which the relative permeability is
incorporated into the internal field intensity component (M).
If a sample of magnetic material is examined on a per domain basis, the differential field
strength around any given domain will be somewhat larger than expected due to its proximity to
the remainder of the domains in the material. In effect, a given domain experiences the magnetic
influence of the averaged total magnetism of the solid, since the orientation of any given domain
may be random. At this point, it is appropriate to define effective magnetic field intensity (Heff),
existing within the solid, that is the sum of the applied field (H) and some averaged contribution
from the magnetism (M) of the surrounding domains. The proposed equation adjusts the
percentage of bulk magnetism (M) added to the applied field intensity (H) through the scaling
coefficient alpha () which typically has a value around 103. The modified relationship for the
magnetic field intensity experienced by a single domain is given by the equation,
= +
( 3.2 )
If a magnetic material was able to return all of the magnetic energy that was input, the
resulting magnetization curve would take the form of a single valued sigmoid (equivalent to the
22
center line of the hysteresis loop shown in the model B-H curve in Fig. 3.6. This curve, referred
to as the anhysteric magnetization curve, represents the ideal or lossless magnetization of a
material. The function that was chosen to model this semi-empirical representation was
developed by Langevin. The parameters used to calculate this quantity are the effective field
strength (Heff) given by the above equation, the saturation level (Ms), and the shaping coefficient
gamma (a), which adjusts the slope of the curve according to the magnetic hardness of the
material. The phenomenological representation of anhysteric magnetization (Man) proposed by
Langevin is defined by the equation,
= (coth
( 3.3 )
The Jiles-Atherton model is a physically based model that includes the different mechanisms
that take place at magnetization of a ferromagnetic material. The theory is based on the existence
of magnetic domains, which are separated by domain walls. The motion of these walls,
especially translation, bending, and pinning, is shown to be closely related to hysteresis. The
magnetization M is represented as the sum of the irreversible magnetization Mirr due to domain
wall displacement and the reversible magnetization Mrev due to domain wall bending.
= +
( 3.4 )
The rate of change of the irreversible part of the magnetization is given by,
( )
( 3.5 )
The component of reversible magnetization reduces the difference between the prevailing
irreversible magnetization Mirr and the anhysteretic magnetization Man at the given field
strength. This can be expressed as,
( 3.6 )
= ( )
Where c is a domain flexing parameter, defining the amount of reversible magnetization due
to wall bowing and reversal rotation, included in the magnetization process.
The hysteresis differential equation for reversible susceptibility can be expressed as
= (
( 3.7 )
= (1 )
(1 ) (
= (1 )
(1 ) (
( 3.8 )
( 3.9 )
Where,
(
)
= 0 0
1,
we need to consider,
And,
( 3.10 )
= (1 +
)
( 3.11 )
Hence using the equation we can finally write the following relation,
(1 )
(1) (
(1 )
( 3.12 )
25
1.5e6
1800
1800
1.4e-3
0.14
1.6e6
1000
700
1.4e-3
0.22
26
1.6e6
972
672
1.4e-3
0.14
1.6e6
1010
455
1.8e-3
0.21
27
1.6e6
1085
320
2.0e-3
0.3
1.4e6
1085
50
1.1e-5
0.2
28
1.5e6
2000
4000
0.001
0.1
1.7e6
1000
500
0.001
0.1
The hysteresis loops of a wide range of magnetic materials can be modeled using the
equations given above. Hysteresis parameters a, , k and c can be determined from experimental
hysteresis measurements, and then used to model the hysteresis curves using the theory of
hysteresis. The B-H curves of different sample cores are shown in Fig. 3.7
29
The network circuit diagram is shown in the Fig. 3.8. It consists of a SFCL, a resistive load
R, a variable resistor Rv, an AC voltage source Vs(t) as power supply and a DC magnetizing
current source Id. If we label the cores by numbers 1 and 2, respectively, and suppose that the
30
turns number of AC winding and DC winding on each core are wc and wd, the mean magnetic
path length is l and cross-sectional area of iron core is A, then, according to Amperes circuit
law:
Hence,
= 1
( 3.13 )
+ = 2
( 3.14 )
1
=
2
=
( 3.15 )
( 3.16 )
In the first core the flux B and the current i do not follow the Faradays Law of electro
magnetism as because the increase of current decreases the flux (since we are considering the net
flux to be in the opposite direction of the flux created by the AC current) and hence the emf
induced is positive. Hence it can be written,
1 =
Therefore,
1 = 0
Now if
(1 + 1 )
1
1
0 2
1
= 0
1 +
=
(1 +
)
1
31
( 3.17 )
( 3.18 )
0 2
=
1 = 1 =
1
(1 +
)
( 3.19 )
( 3.20 )
But in the second core the flux B increases as the current i increase and it follows the
Faradays Law of electromagnetism and the emf induced in the second coil is negative to the
supply. Hence,
2 =
Therefore,
2 = 0
(2 + 2 )
2
2
= 0
1 +
2
0 2
2
=
(1 +
)
( 3.21 )
( 3.22 )
Hence,
2 = 2 =
Now based on Kirchhoffs voltage law,
Therefore,
2
1 +
() = 1 + 2 + = 1 + 2 +
32
( 3.23 )
( 3.24 )
() =
Or,
1 2
2 +
+
+
1 2
()
=
2 + 1 + 2
1
( 3.25 )
( 3.26 )
From the above equation the value of the current i is obtained. Using this value we can
numerically work out the voltage u of each winding and obtain the voltage curve through the
simulation program MATLAB. Likewise, based on the rated current and the short resistance, we
can calculate the AC windings voltage under normal operation and the limited current under
fault condition respectively.
33
Chapter 4
4.1 Introduction
The design of power systems often requires the incorporation of fault current limiter
elements such as SISFCL for successful fault limiting applications. The unique, often highly
restrictive constraints placed on the design of these electromagnetic elements by the system
performances and environmental requirements associated with fault limiting applications very
often lead to the need for optimization as part of the design process. While a variety of
optimization techniques are available to the electromagnetic design engineer, many of the
traditional techniques prove unsuitable when confronted with real world applications. This :is
because conventional functional optimization techniques are either based on greedy, local
optimization methods such as gradient methods or consist of random walk solution space
searches. In either case, these conventional techniques are often poorly suited to the task of
optimization when high dimensional, multi-modal functional domains are involved. In addition,
traditional optimization techniques usually require the object function to be, at the very least,
continuous and, in many cases differentiable, placing severe limitations on the form and content
of the object function. This paper discusses a radically different and relatively new functional
optimization methodology known as Genetic Algorithm (GA) optimization that overcomes the
above- mentioned problems of the conventional techniques.
local maxima. However, the tight solution space coupling also places constraints on the solution
domain such as differentiability and/or continuity, constraintsthat can be hard or even impossible
to realize in practice. The global techniques on the other hand, are largely independent of and
place few constraints on the solution domain. This absence of constraints means that the global
methods are much more robust when faced with ill-behaved solution spaces. .In particular, global
techniques are much better at dealing with solution spaces having discontinuities, constrained
parameters, and/or large numbers of dimensions with many potential local maxima.
The
downside to the global methods general independence from the solution space is that they
cannot take advantage of local solution space characteristics such as gradients, during the
search process resulting in generally slower convergence than the local techniques.
Often, however, in design problems convergence rate is not nearly as important as getting a
solution. Having found a solution the ultimate goal in design is to find the best solution or global
maxima. In these applications, global methods are favoured over local methods. Global
techniques either yield global or near global maxima instead of local maxima and often find
useful solutions where local techniques cannot. Global methods are particularly useful when
dealing with new problems in which the nature of the solution space is relatively unknown. Of
the global techniques GA optimizers can readily handle discontinuous and non- differentiable
functions. They are also well suited for constrained optimization problems. Genetic Algorithms
are considerably more efficient and provide much faster convergence than random searches. In
addition, they are easily programmed and readily implemented.
35
Genetic Algorithm (GA) optimizers are robust, stochastic search methods modelled on the
principles and concepts of natural selection and evolution. As an optimizer, the powerful
heuristic of GA is effective at solving complex, combinatorial and related problems. GA
optimizers are particularly effective when the goal is to find an approximate global maxima in
high dimension, multi-modal function domains in a near optimal manner. GA differ from more
conventional techniques in that
They normally operate on a coding of the function parameters (chromosome) rather than
on the parameters themselves
They use simple, stochastic operators (selection, crossover, and mutation) to explore the
solution domain in search of an optimal solution.
Search Space: The term Search Space means all possible solution to the problem.
36
Fig. 1.1 briefly presents a list of some of the commonly encountered terminology in the CA
field as it relates to the optimization problem.
min ()
37
( 4.1 )
such that
( ) 0, i = 1 m
( ) = 0, = + 1 .
( 4.2 )
. =
Where C(x) represents the nonlinear inequality constraints, Ceq(x) represents the equality
constraints, m is the number of nonlinear inequality constraints, and mt is the total number of
nonlinear constraints.
The Augmented Lagrangian Genetic Algorithm (ALGA) attempts to solve a nonlinear
optimization problem with nonlinear constraints, linear constraints, and bounds. In this approach,
bounds and linear constraints are handled separately from nonlinear constraints. A sub problem
is formulated by combining the fitness function and nonlinear constraint function using the
Lagrangian and the penalty parameters. A sequence of such optimization problems are
approximately minimized using the genetic algorithm such that the linear constraints and bounds
are satisfied. A sub problem formulation is defined as
=1
=+1
()2
2
( 4.3 )
= +1
Where
The components i of the vector are nonnegative and are known as Lagrange
multiplier estimates
The algorithm begins by using an initial value for the penalty parameter (Initial Penalty).
The genetic algorithm minimizes a sequence of sub problems, each of which is an approximation
of the original problem. Each sub problem has a fixed value of , s, and . When the sub problem
is minimized to a required accuracy and satisfies feasibility conditions, the Lagrangian estimates
are updated. Otherwise, the penalty parameter is increased by a penalty factor (Penalty Factor).
This results in a new sub problem formulation and minimization problem. These steps are
repeated until the stopping criteria are met.
Each sub problem solution represents one generation. The number of function evaluations
per generation is therefore much higher when using nonlinear constraints than otherwise.
If the individual is infeasible, the penalty function is the maximum fitness function
among feasible members of the population, plus a sum of the constraint violations of
the (infeasible) individual.
Perform reproduction through the fitness weighted selection of individuals from the
population
The base logic of the algorithm is that gens of powerful individuals are based to be carried over
next generation and others are based to be detached in next generation. In natural selection, a
human born, grows, and dies. These stages of human life correspond to the different operators in
the algorithm. GA operators related these stages can be explained as the following
= , + (0,1) ( , , )
( 4.4 )
where the parameters wmin,jand wmax,jshow the minimum and maximum of the variable wj.
40
=1
( 4.5 )
Where Fave, Nk, and Fi represent the average fitness value within population, the number of
individuals within population, and the fitness value of ith individual, respectively.
4.6.3 Selection
A number of selection strategies have been developed and utilized for Genetic Algorithm
optimization. These strategies are generally classified as either stochastic or deterministic.
4.6.3.1 Population Decimation
The simplest of the deterministic strategies is Population Decimation.
In Population
Decimation, individuals are ranked from largest to smallest according to their fitness values. An
arbitrary minimum fitness is chosen as a cutoff pointand anyindividual witha lower fitness than
the minimum is removed from the population. The remaining individuals arethen used to
generate the new generation through random pairing.
()
()
41
( 4.6 )
Fig.4.2. Proportionate Selection represented as a roulette wheel with spaces on the wheel proportional to an
individuals relative fitness.
Fig. 4.2 depicts the Proportionate Selection process as a roulette wheel where individuals are
assigned a space on the wheel that is proportional to their relative fitness. The wheel is spun and
the individual pointed to at the end of the spin is the individual selected.
4.6.3.3 Tournament Selection
A second popular strategy (and perhaps among the most effective for many applications)is
TournamentSelection. In Tournament Selection, a sub-population of N individualsischosenat
randomfromthe population.Theindividuals ofthissub- population competeonthebasisoftheir
fitness.The individual in thesub-population with thehighestfitness wins thetournament and
becomestheselected individual.All of thesub-population members are then placed back into the
general population and the process can be repeated.
4.6.4 Crossover
In this stage, a child is produced to be crossed the parents. New individuals same as the
determined number are produced to be used the crossing method with the scattered parameter
from parents selected via the tournament method.
42
The value of 1 and 0 as gen number of an individual is randomly produced. If the value is 1,
then gen is taken from mother, the value is 0, then gen is taken from father and thus the child is
produced.
Cross: 1 0 1 1 0
Mother: a b c d e
Father: x y z u w
Child: a y c d w
4.6.5 Mutation
The mutation operator provides a means of exploring portions of the solution surface that
are not represented in the genetic makeup of the current population. If p >pmutation , an element
in the string making up the chromosome is randomly selected and changed. In the case of binary
encoding this amounts to selecting a bit from the chromosome string and inverting; it. In other
words, a 1 becomes a 0 and a 0 becomes a 1. If higher order alphabets are used, slightly
more complicated forms of mutation are required.Fig. represents mutation.
Fig.4.3. The mutation operator randomly modifies elements within the chromosome.
Fitness Evaluation
Initial Population
Constraint Check
Selection
Select Parent#1
Select Parent#2
Perform Crossover
(with p = pcross)
Until population
Perform Mutation
Is full
(with p = pmutation)
Fitness Evaluation
Replace Population
Constraint Check
End
44
Fig. 4.5 gives a screenshot of GA solver App available in MATLAB Global Optimisation
Toolbox.
45
Chapter 5
5.1 Introduction
The model of Saturated Iron core Superconducting fault current limiter is optimised by using
Genetic Algorithm. Fault resistance, no. of ac turns (wc), no of dc turns (wd), dc biasing current
(Id) are taken as variable parameters.
(12 + 22 )
46
( 5.1 )
5.2.1.1 Term1
It signifies the peak value ofac current that is obtained when there is a fault in system. It is
obtained by solving the differential equation
()
=
2 + 1 + 2
1
( 5.2 )
0 2
where0 is absolute
permeability, wc is no. of ac turns in each core, A is net areaand l is the mean magnetic path
length.
1
1
and
2
2
= (1 )
(1 ) (
Where =
, +1 if
0, 1 if
( 5.3 )
k is parameter defining the pinning site density of domain walls, Man is anhysteretic
magnetization at a given field strength, Man=Ms (coth(He/a)-a/He ),Ms is the saturation level, c
is a domain flexing parameter, defining the amount of reversible magnetization due to wall
bowing and reversal rotation, included in the magnetization process. is a scaling coefficient,.
He is the effective field strength, He=H+ M.
47
5.2.1.2 Term 2
From JA hysteresis model it is cleared that a voltage is inducted across the biasing coil
during the fault limiting action of SFCL which should be minimum in order to operate the FCL
satisfactorily. Term 2 designates the peak voltage induced across dc coil. It can be formulated as
= 0 (
1
1
2
2
1 +
+
1 +
)
and
1
=
2
=+
( 5.4 )
can be derived as
( 5.5 )
( 5.6 )
5.2.2 Constraints
5.2.2.1 Magnetic field density constraint
Normal operating constraint
For the successful operation of fault current limiter, it should be saturated during normal
operation condition so that it can offer low impedance and does not limit the normal operating
current flowing in the circuit. So, to ensure this the effective ampere turns in both cores should
be enough to place the operating point (Bopt) above the saturation limit (Bsat) . This saturation
level generally depends upon the core materials used.
48
Fault constraint
( 5.7 )
During the fault condition the ampere turns created by the large fault current oppose the dc
ampere turns created by the dc biasing current in such a manner that the operating point comes in
linear zone in BH curve. As a result of it the permeability of the cores increases resulting in
increasing flux linkages which in turns increases the impedance offered by the cores during fault
condition. But if the value of ac ampere turns created by the fault current is high enough, it will
make the net effective ampere turns of the cores to be operated on reverse saturation condition.
So, the fault constraints consist of
And
( 5.8 )
( 5.9 )
According to the law of KVL, the applied voltage is equal to the voltage across SFCL and
the losses on various parts of the circuit. So, the voltage across SFCL ( ) is always less than
the applied voltage ( ).
( 5.10 )
The objective function described earlier is formed in MATLAB. R-K fourth order method is
utilized. Both normal and fault condition are taken into account and constraint function is also
programmed in MATLAB. The objective function and constraint function are optimised using
49
Knowing that the fault resistance is one of the most important quantities of the
saturated core SFCL design, the first idea is to use its value as variable parameter
during the optimization procedureModel 1.
The no of turns in dc windings is effecting the net effective flux which decides the
point of operation in B-H curve, it is selected to use as a variable parameter during
the optimization procedureModel 3.
As dc biasing current provides the necessary bias for operation, the fourth idea is to
use it as a variable parameter during the optimization procedureModel 4..
Among the four variable parameters the next idea is to vary three of them, no of ac
turns, no of dc turns and dc biasing current while fault resistance is kept constant
Model-5.
The last optimization consists of varying all four parameters during the optimization
procedure Model-6.
50
Ms
1.7e6
0.1
500
0.001
1000
Population type
Double vector
Population size
20
Creation function
Constraint dependent
Initial population
Default
Initial score
Default
Initial range
[0;1]
Fitness scaling
Scaling function
Rank
Selection
Selection function
Stochastic uniform
Reproduction
Elite count
Crossover function
0.8
Mutation
Mutation function
Constraint dependent
Crossover
Crossover function
Scattered
Direction
Forward
Fraction
0.2
Population
Migration
51
Constraint
Interval
20
Initial penalty
10
Penalty factor
100
Hybrid function
None
Generations
100
Time limit
Infinity
Stall generations
50
Infinity
Function tolerance
1e-4
Nonlinear constraint
1e-4
parameters
Hybrid function
Stopping criteria
tolerance
52
Id
wd
wc
V0
length
freq
15A
100
20
50V
0.56m
50Hz 45.4e-4
Rn
5ohms
Fig.5.2. IfvsRf
Cases
Rf(ohms)
Case 1
0.2
Case 2
0.4
53
Case 3
0.8
Rfopti
0.9890
54
It is observed that among the cases considered Rfopti=0.9890 ohm gives the optimum value of
peak fault current as well as voltage induced across dc coils. If fault resistance increases further
fault current decreases but net effective flux produced in the cores increases and a condition is
reached when the cores dont come out of saturation in fault condition.. Hence the constraint
function isnt satisfied and optimisation gets stopped.
5.4.2 Model 2
In Model 2, no of turns in ac winding is optimised while keeping fault resistance, dc
winding turns and dc biasing current fixed. The ac winding turns no is optimised between 20 and
50. The values of constant parameters are shown in the table below
Table 5.5. Value of constant parameters for Model 2
Id
wd
Rf
V0
length
Freq
Rn
15A
100
0.1ohms
50V
0.56ms
50Hz
45.4e-4
5ohms
Theoretical optimum value is obtained at wc= 50, at this value the of peak fault current is
29.2A and peak value of voltage across dc coil is 77.72V.The variation of peak fault current with
wc is shownin Fig.5.7.Fig. 5.8describes the variations of both fault current and ac winding turns
with no. of iteration during optimization process.
Fig.5.7. Ifvswc
55
Following Table 5.6 describes four cases including the optimization condition
Table 5.6 Different cases in Model 2 optimisation
Cases
Wc
Case 1
25
Case 2
30
Case 3
35
wcopti
50
Table 5.6 can be represented by different responses described below.From the above
responses it is observed that wcopti=50 give the best fault limiting condition satisfying all
constraints. Though further increase in no of ac winding turns can reduce the peak fault current
but it is uneconomical to keep high no. of ac winding turns beyond the range considered for
optimization because it can impose a limitation on system design. Also it can be noted that as no
of ac winding turns increases beyond 50, cores will not be in saturation region during normal
operating conditions which violets the normal magnetic flux density constraint described in
chapter. However if the no. of ac winding is reduced below 20, cores will get saturated in reverse
direction during fault which will violate fault magnetic flux density conditions.
56
57
5.4.3 Model 3
Model 3 is optimised where the dc biasing current Wd is varying parameter and all the other
parameters are kept constant for the sake of simplicity. The range of optimization is considered
between 80 to 120.The values of other constant parameters are shown in the Table 5.7 below
Table 5.7.Value of constant parameters for Model 3
Id
Wc
Rf
V0
length
freq
Rn
15A
20
0.1ohms
50V
0.56m
50Hz
45.4e-4
5ohms
It is evident from the above figures that peak value of fault current is less affected by the
variation of dc winding turns(as described by Fig. 5.13) but the magnetic characteristics of the
cores are highly affected. Theoretical optimum value is achieved at wd =82 and at this condition
the value of the peak fault current and peak value of voltage induced across dc coil is 99.87A and
187.8V. Fig. 5.14describes the variations of fault current and dc no of turns at every iteration
during optimization process.
Fig.5.13. Ifvswc
Cases
wd
Case 1
100
Case 2
110
Case 3
120
wdopti
82
Fig.5.16 If vs time
59
It is observed from Fig. 5.17that core1 is on the verge of reverse saturation condition during
fault operation. If no of dc turns decreases further peak fault current will decrease slightly but the
net effective flux in fault condition will cause reverse saturation condition in core 1.
5.4.4 Model 4
In Model 4, dc biasing current is optimised while keeping fault resistance, dc winding turns
and no of turns in ac winding fixed. Here the range between which Id is varied is 10 to 20. The
values of other constant parameters are shown in the Table 5.9below
Table 5.9 Value of constant parameters for Model 4
Wc
Wd
Rf
V0
length
freq
Rn
15A
100
0.1ohms
50V
0.56m
50Hz
45.4e-4
5ohms
60
Fig.5.19 Ifvs Id
From the above figures it is evident that if dc biasing current decreases then fault current
also decreases implying less peak fault current. However, the change is very less significant
because dc biasing current decides mainly the operating zone in magnetic hysteresis curve for a
successful fault limiting operation. The optimum value of dc biasing current is obtained at
13.25A and at this point the value of optimum peak fault current is 102.8A and peak voltage is
232.2 V.Fig.5.20. describes the variations of fault current and dc no of turns at every iteration
during optimization process.
Table 5.10 Different cases in Model 4Optimisation
Cases
Id(A)
Case 1
15
Case 2
18
Case 3
20
Idopti
13.25
Fig.5.22vdcvs time
If dc biasing current is further decreased below the optimum point then peak fault current
decreases but net effective flux created in the core will drag the core on reverse saturation
direction.This will violate the constraint condition.
62
5.4.5 Model 5
The dc bias current, the no of ac turns and the no. of dc turns are varied together in model 5
during optimization process. However, the fault resistance is kept constant. The constant
parameters are listed in the Table 5.11below
Table 5.11 Value of constant parameters for Model 5
Rf
V0
length
freq
Rn
0.1ohms
50V
0.56m
50Hz
45.4e-4
5ohms
No. of ac turns, no of dc turns and dc biasing current are varied between 20 to 50, 80 to 150
and 5 to 20 A respectively. At optimum condition the value of Id,wd,wc are 18.018A,80 and 50
respectively. The values of peak fault current and peak value of voltage induce across dc coil
obtained during optimization procedure are 29.31A and61.42Vrespectively.
Table 5.12 Different cases in Model 5 Optimisation
wc
wd
Id(A)
Case 1
25
100
14
Case 2
35
110
15
Case 3
45
120
18
opti3
50
80
18.018
63
It is observed from Fig. 5.27that optimum condition is on verge of saturation limit during
normal condition. Penalty value, scores and best individual during optimization process are
shown in Fig. 5.29below
64
Fig.5.29 a) Penalty Value vs generation, b) current best Individual, c)Best, Worst and Mean scores
5.4.6 Model 6
All four parameters are varied in Model 6 and the overall optimization of the design is
achieved. The ranges used for various parameters are shown inTable 5.13 . Also, different
constant parameters are listed in Table 5.14.
Table 5.13 Ranges of the parameters in Model 6
Parameters
Lower range
Upper Range
Id(A)
10
20
wd
80
120
wc
20
50
Rf(ohms)
0.1
65
freq
V0
length
Rn
50Hz
50V
0.56m
45.4e-4
5ohms
The optimization point is obtained at Id = 18.083A, wd =80, wc =50 and Rf=0.998 ohms. The
optimised values of peak fault current and peak value of voltage induce across dc coils are
23.51A and 36.78V. Penalty value, selection function, scores, expectations of scores and
distances during each generation are described in the following figures
66
FromFig.5.30it is observed that mean penalty is 47.832 which is close to best penalty value.
So, a large no of populations are converging towards optimal solution. It can be also seen
fromFig.
between 5 and 20. But after wards the distance falls to nearly zero indicates that it starts to
converge towards a solution, limited by the tolerance of objective and constraint function.
Table 5.15describes four cases including the optimised one.
Table 5.15 Different cases in Model 6 Optimisation
Id(A)
wd
wc
Rf(ohms)
Case 1
14
90
25
0.4
Case 2
16
100
35
0.6
Case 3
18
110
45
0.8
Opti4
18.083
80
50
0.998
67
Fig.5.36vdcvs time
68
69