CiprianStanciu Genetic Algorithms Applications

Embed Size (px)

Citation preview

  • 8/13/2019 CiprianStanciu Genetic Algorithms Applications

    1/6

    GENETIC ALGORITHMS. APPLICATIONS

    eng. Stanciu Ciprian

    PhD student

    Computer Science and Engineering Department 

    I. INTRODUCTION TO GENETIC ALGORITHMS

    The life has been shaped by evolution during billions of years. Evolution in nature

    inspires computer scientists to implement, analyze, and utilize methods similar to evolution to

    solve optimization and search problems that are computationally hard and complex.

    Optimization is a general tool used in numerous problems of engineering and sciences.Hence there seems to be things shared or sharable beteen different sciences that can be

     beneficial in many ays in the context of optimization and search.

    !enetic "lgorithms, first proposed by #ohn Holland in $%&', are a type of meta(

    heuristic search and optimization algorithms inspired by Darin)s principle of natural

    selection, as stated in *$+. The central idea of natural selection is the fittest survive. Through

    the process of natural selection, organisms adapt to optimize their chances for survival in a

    given environment. andom mutations occur to the genetic description of an organism hich

    is then passed on to its children. -f a mutation proves to be helpful, these children are more

    liely to survive to reproduce. Otherise, if it proves to be harmful, these children are less

    liely to reproduce, so the bad trait ill die ith them.

    "ccording to */+, genetic algorithms are simulations of evolution and, in most cases,

    genetic algorithms are nothing else than probabilistic optimization methods hich are based on

    the principles of evolution. The principles behind the genetic algorithms are described in *0+

    */+ *1+ and presented belo.

    !enetic algorithms or by generating a population of numeric vectors, called

    chromosomes, each representing a possible solution to a problem. The individual components,

    hich are numeric values, ithin a chromosome are called genes. 2e chromosomes are

    created by crossover, this being the probabilistic exchange of values beteen vectors,

    mutation, hich is the random replacement of values in a vector, and clone. 3utation providesrandomness ithin the chromosomes to increase coverage of the search space and help prevent

     premature convergence on a local optimum.

    4hromosomes are evaluated according to a fitness function, ith the fittest surviving

    and less fit being eliminated. The result is a gene pool that evolves over time to produce better

    solutions to a problem. The genetic algorithm5s search process typically continues until a pre(

    specified fitness value is reached, a set amount of computing time passes or until no significant

    improvement occurs in the population for a given number of iterations.

    The ey to finding a good solution using a genetic algorithm lies in developing a good

    model of the problem. " good genetic algorithm model should reduce or eliminate redundantchromosomes from the population and provide genetic algorithm operators that effectively

  • 8/13/2019 CiprianStanciu Genetic Algorithms Applications

    2/6

    improve the population of solutions. The chromosome must accurately represent the problem

    and allo the genetic algorithms operators to or effectively on the chromosomes to generate

     better solutions as the iterative process goes on.

    -n the next section, several applications of genetic algorithms are presented. These

    applications illustrate the vast application area and the great number of domains that are or can be beneficiaries of the use of genetic algorithms.

    II. APPLICATIONS

    6ince the beginning of genetic algorithms, they have originated applications in

    numerous areas. Due to the fact that genetic algorithms are non(problem(specific, their

    application is not restricted by the problems) physical bacground, and hence can be applied to

    many combinatorial optimization problems in different disciplines, as stated in *$+.

     2ext, a number of applications of the genetic algorithms are presented. These

    applications ere chosen to be presented by considering their application domains, and the

    distinctness of these domains, to give an insight over the vast application domain of genetic

    algorithms.

    1. Application !or I"age En#ance"ent an$ Seg"entation

    " survey regarding these types of applications is presented in *'+. "s stated in *'+, the

    first tas of machine vision is to enhance image 7uality in order to obtain a re7uired image

     perception, and this is done by removing noise, amplifying image contrast and amplifying the

    level of a detail. !enetic algorithms are used to construct ne filters, to optimize parameters ofexisting filters, and to loo for optimal se7uence of existing filters. 8ilter optimization using

    genetic algorithms significantly improves color and structural characteristics of the traditional

    color filtering scheme. Optimized filters exhibit acceptable noise attenuation capabilities,

    according to the research results presented in *'+.

    On the sub9ect of image segmentation, excellent results are obtained here parallel

    genetic algorithm is adopted for surface model fitting in three(dimensional space. Parallel

    genetic algorithm as demonstrated to sho that it is a better optimizer than the classical

    genetic algorithm in the problem of image segmentation, according to the same survey.

    %. Application o! Genetic Algorit#" in &inger Print Regitration 

    8inger prints registration confronts ith the problem of aligning to images: the

    reference and sensed images. The ey of image registration is to find the proper transformation

    of one image to another so that each point of one image is spatially aligned ith its

    corresponding point of the other, as stated in *;+.

    !enetic "lgorithms have been used in *;+ to detect and ad9ust the rotation in the finger

     prints images. , ith a reasonable time of execution,

    according to *;+. "n important fact is that the retrieved rotation angle alays less than or e7ual

  • 8/13/2019 CiprianStanciu Genetic Algorithms Applications

    3/6

    to the original rotation angle. "nother important fact that as demonstrated in *;+ is that a

    ider rotation angle produces a loer relative error hen applying genetic algorithms.

    '. E(e Location Uing Genetic Algorit#"s

    "ctive perception ors by systematically organizing the visual tass in such a mannerthat as visual processing progresses in time the volume of data to attend to is reduced and

    computing resources are focused only on salient regions of the image. The adaptive eye

    location approach, described in *&+, sees first here salient things are and then hat their

    identity is.

    Eye location involves, according to *&+, to stages: ?$@ the derivation of the saliency

    attention map, and ?/@ the possible classification of salient locations as eye regions. The

    saliency ?or AhereB@ map is derived using consensus beteen navigation routines encoded as

    finite state automata exploring the facial landscape and evolved using genetic algorithms. The

    classification stage is concerned ith the optimal selection of features and the derivation of

    decision trees for confirmation of eye classification, by location, using genetic algorithms.

    ). Application o! Genetic Algorit#" in Ae"*l( Planning

    "utomatic assembly se7uence planning ?"6P@ plays a vital role in planning

    manufacturing processes for products during product development and, thus, plays an

    important role in computer(aided intelligent concurrent engineering, according to *C+.

    ather than using a single(level genetic algorithm ith fixed genetic operator

     probability setting ?crossover and mutation rates, for example@, in *C+, the genetic operator

     probability setting is dynamically updated to optimize the effectiveness of each type of genetic

    operator during the solution searching process. Thus, an improvement in state(of(the(artgenetic algorithms for "6P is proposed: a multi(level genetic assembly se7uence planner that

    dynamically optimizes the selection of genetic operator probability setting, to improve solution

    7uality and to speed up the searching process for an "6P solution. "ccording to *C+, this

    approach ors ell on all of the tested product structures.

    +. Application o! Genetic Algorit#" in ,Arti!icial Ant- Pro*le"

    -t is often possible to construct an automaton using heuristic methods but in some

     problems such methods consume a lot of time or such construction is even impossible. "n

    example of such a problem is the A"rtificial antB problem. -n *%+, the A"rtificial antB problem

    is described: the game taes place on the surface of a torus ith a specific idth and height

    ith food in some cells. "nt starts moving from a cell labeled A6tartB. The game lasts for a

    specified number of moves and the goal is to !oal of the game is to mae the ant AeatB

    maximum possible 7uantity of AfoodB, ith minimal number of states ?minimal number of

    moves@. !enetic algorithms and genetic programming can be and are applied to construct

    automata in problems of this ind, according to *%+.

    . Application o! genetic algorit#" to real /orl$ reource contraine$ pro0ect

    c#e$uling pro*le"

    The esource 4onstrained Pro9ect 6cheduling Problem ?4P6P@ is concerned ithfinding the schedule hich can be executed, hilst conforming to both precedence relations

  • 8/13/2019 CiprianStanciu Genetic Algorithms Applications

    4/6

    and resource limitations, ithin the minimum time frame. This problem is combinatorial in

    nature and is ell non to be 2P(Hard, as stated in *$=+.

    The research presented in *$=+ has demonstrated the applicability of genetic algorithms

    to optimize the real orld 4P6P. "ccording to the same paper, this extends the functionality

    of existing commercial softare beyond the existing resource leveling capabilities to find theoptimal pro9ect duration ithin the imposed resource constraints.

    . Application o! Genetic Algorit#" !or Po/er S(te" State Eti"ation

    6tatic state estimation for on(line application in a poer system control centre is mainly

    aimed at providing a reliable estimate of the system voltages and flos to be presented to the

    operator and supplied to the database. -n order to do this, the state estimator should be able to

    estimate the best state and should process bad data, according to *$$+. The static state estimator

    ?6E@ is a data processing algorithm for converting redundant meter readings and other available

    information into an estimate of the static state vector. " genetic algorithm based solution

    techni7ue is ell suited, as argued in *$$+, for large netors ith multiple bad data. The

    results have shon that, genetic algorithms approach has provided a good solution to the 6E

     problem.

    2. Application o! Genetic Algorit#" to t#e Opti"i3ation o! Microtrip Antenna

    This application is described in *$/+ and presented briefly next. " microstrip antenna is

    a metallic radiating patch fabricated over a dielectric substrate baced by a metallic ground

     plane and generally used in microave and millimeter(ave fre7uencies. -n the design and

    syntheses of antennas, the goal is to find a radiating structure that meets performance criteria

    that may be gain, input impedance, beam idth or a combination of any of the above parameters. Optimization of gain, using genetic algorithms, can be done for various values of

    dielectric constant and substrate height, as argued in *$/+.

    4. Application o! Genetic Algorit#" to H($rogenate$ Silicon Cluter

    -n the last fe years, there has been a very rapid development in the study of small

    clusters due to the groing importance of these systems in applications lie catalysis and due

    to possibilities of developing nano(electronic devices, as stated in *$0+. "lso, there have been

    attempts to design novel materials using these clusters as basic building blocs, according to

    the same paper. These materials may have very different properties compared to the naturally

    occurring solids. -n *$0+, it is demonstrated that genetic algorithms can be efficiently used for

    computing the ground state configurations of the small, hydrogenated silicon clusters. These

    techni7ues can be easily applied for a large number of related problems and, according to *$0+,

    their importance in materials research is expected increase in the future.

    15. Application o! Genetic Algorit#" in Stoc6 Mar6et Data Mining Opti"i3ation

    -n stoc maret and other finance fields, genetic algorithms have been applied in many

     problems. There have been a number of attempts to use genetic algorithms for ac7uiring

    technical trading rules. One of these attempts is described in *$1+.

    "s shon in *$1+, by applying genetic algorithms, very little precision is lost but a lotof running time is saved. 4onse7uently, this approach can be used in a real analysis system,

  • 8/13/2019 CiprianStanciu Genetic Algorithms Applications

    5/6

    and the results are similar to the best one. -n the mean time, this approach can be used as a

     basic tool for other application, such as raning trading rules.

    11. Application o! Genetic Algorit#" to Te7ture Anal(i

    This application is described in *$'+. " typical goal of texture analysis is to calculatethe orientation distribution function ?OD8@ is calculated from a set of pole figures ?P8@. The

    results of the OD8 calculation, presented in *$'+, sho that the genetic algorithms method can

     be successfully used in the field of texture analysis.

    -n the same paper, it is argued that it is possible to apply genetic algorithms to other

    important problems in the field of crystallographic textures. The example is finding of the best

    textures of materials for a given application.

    1%. Application o! Interacti8e Genetic Algorit#" to &a#ion Deign

    -n *$;+, it is presented a fashion design aid system ith interactive genetic algorithm

    using domain specific noledge. -nteractive genetic algorithms are the same as the basic

    genetic algorithms except the fitness function.

    -n interactive genetic algorithms user assign the fitness to each individual instead of a

    fitness function. This ay the interactive genetic algorithms can percept user)s emotion or

     preference in the course of evolution. 4onse7uently, this techni7ue can be used to solve

     problems concerning art and design, hich cannot be easily solved by the basic genetic

    algorithms.

    III. CONCLUSIONS

    The main purpose of this paper as to give an insight on the large space of application

    domains of the genetic algorithms. 6everal applications from several different domains ere

    described. The main advantages that made the use of genetic algorithms appropriate for so

    many different domains refer to reduced computation time in the context of providing very

    similar outputs to the best solution.

    The genetic algorithms, ith their advantages, made the development of interactive

    applications possible in domains in hich such a possibility ouldn)t have been an option

    otherise.

    Re!erence

    *$+ ao hou, A6tudy on !enetic "lgorithm -mprovement and "pplicationB, " Thesis

    6ubmitted to the 8aculty of the Forcester Polytechnic -nstitute in partial

    fulfillment of the re7uirements for the Degree of 3aster of 6cience in

    3anufacturing Engineering, 3ay /==;

    */+

  • 8/13/2019 CiprianStanciu Genetic Algorithms Applications

    6/6

    *0+ "rthur E. 4arter, ADesign and "pplication of !enetic "lgorithms for the 3ultiple

    Traveling 6alesperson "ssignment ProblemB, Dissertation submitted to the faculty

    of the Kirginia Polytechnic -nstitute and 6tate