Neural networks are attracting an enormous amount of attention in many civil engineering disciplines, including transportation, because they represent a class of robust, nonlinear models capable of learning relationships from data. However, in the development of such models for a particular application, various parameter settings are left to the judgment of the network developer. The net result of poor parameter settings will be slow convergence and/or bad performance on unseen cases. Recently, genetic algorithms have emerged as a potential searching technique to design a neural network model that performs best on a specified task according to explicit performance criteria. Genetic algorithms are search algorithms based on the mechanics of natural selection and natural genetics. In this paper the authors present a genetic algorithm method that evolves a neural network model for the selection of the optimum maintenance strategy for flexible pavements. A hybrid evolutionary-learning system using gradient descent learning as well as a genetic algorithm to determine the network connections weights is described. The developed neural network model has an input vector of seven components and an output vector of seven components. The input vector represents the factors affecting the maintenance strategy selection, whereas the output vector represents the different pavement maintenance strategies available. Brainmaker Professional, a commercially available neural network simulator, was used in the development of the neural network model. The performance of the developed neural network model was validated by testing it using 100 unseen cases. The validation results showed that the system misclassified only six cases with an average error rate of 0.024.


  • English

Media Info

  • Features: Figures; References; Tables;
  • Pagination: p. 70-76
  • Monograph Title: Artificial intelligence and geographical information
  • Serial:

Subject/Index Terms

Filing Info

  • Accession Number: 00714945
  • Record Type: Publication
  • ISBN: 0309061636
  • Files: TRIS, TRB
  • Created Date: Dec 20 1995 12:00AM