Application of genetic programming and artificial neural networks to improve engineering optimization

Created by W.Langdon from gp-bibliography.bib Revision:1.3872

@PhdThesis{Gongtao_Wang:thesis,
  author =       "Gongtao Wang",
  title =        "Application of genetic programming and artificial
                 neural networks to improve engineering optimization",
  school =       "Lamar University",
  year =         "1998",
  type =         "Doctor of Engineering",
  address =      "Texas, USA",
  month =        dec,
  keywords =     "genetic algorithms, genetic programming",
  URL =          "http://phdtree.org/pdf/25254058-application-of-genetic-programming-and-artificial-neural-networks-to-improve-engineering-optimization/",
  URL =          "http://search.proquest.com/docview/304558089",
  size =         "105 pages",
  abstract =     "The mathematical models of many engineering problems
                 are very complex and computationally intensive. These
                 complex models are repeatedly used to solve problems.
                 Each optimization process is almost equally burdensome.
                 One solution is to use a response surface model (RSM)
                 to simulate the computationally burdensome model.
                 Several researchers have tried to use conventional
                 regression to simplify computationally intensive
                 optimization models. In most reported efforts of this
                 kind, a quadratic RSM is created from the data
                 collected from previous operations of the
                 computationally intensive model. Optimization is then
                 performed on this simplified model rather than on the
                 complex one. The original model is then consulted at
                 the proposed optimum to verify that all constraints are
                 satisfied. There are two fundamental problems with this
                 approach. The first is that these methods will be
                 inherently inaccurate whenever the underlying function
                 is not quadratic. The second is that it can not recall
                 what was learned about the shape of the design space
                 after an optimization is completed. This research will
                 combine Genetic Programming with Neural Networks to
                 create an RSM. A new way to perform regression is
                 described. This method can discover the underlying
                 simple functional form of a computationally intensive
                 optimization model over the entire design space. A more
                 accurate regression model will be built by using this
                 proper functional form. This RSM can then be saved from
                 one optimization run to the next to serve as a memory
                 of the global and local shape of the design space. This
                 field study develops a new method, which merges Genetic
                 Programming and Neural Networks into an integrated
                 system to perform regression. Experiments are then
                 carried out to compare the existing methods with the
                 developed method when used for optimization. The
                 experimental data shows that the new method is
                 effective if the optimization model is computationally
                 intensive and a set of historical data is available. If
                 the two conditions are satisfied and the same optimum
                 is reached, the computational efficiency improved
                 94.5percent by using the RSM obtained with the
                 developed method, as opposed to optimizing the original
                 model. Compared to using a quadratic RSM, the
                 efficiency is improved 76percent, as the same optimum
                 is reached.",
  notes =        "Supervisor: Victor Zaloom

                 UMI Microform 9938783",
}

Genetic Programming entries for Gongtao Wang

Citations