TRANSPORTATION NETWORK DESIGN USING A CUMULATIVE GENETIC ALGORITHM AND NEURAL NETWORK

Currently available algorithms for finding optimal solutions to the discrete transportation network design problem are deficient in two ways. First, their computing time requirements are very large, which makes them infeasible for processing large networks. Second, they cannot process multiple criteria simultaneously--that is, only one objective value can be optimized in one run and, therefore, only one final solution can be obtained. A neural network in the optimal solution search process to replace the trip assignment algorithm for the computation of total travel time is employed. Before a neural network is used, it must be trained and tested with solutions obtained from a user-equilibrium trip assignment model. Experiments show that the trained neural network can predict total travel times quickly and accurately. Next, this neural network is used in combination with a genetic algorithm to search for optimal network designs. The original genetic algorithm did not work well for the problem. However, an analysis of its results suggested improvements that led to the creation of a very powerful search algorithm: the cumulative genetic algorithm. Experiments show that the cumulative genetic algorithm can seek and find system optimal designs extremely fast, using two criteria simultaneously. A full set of optimal solutions can be obtained to construct a trade-off curve for the two criteria. This trade-off curve, composed of optimal solutions, is the boundary of one side of the entire solution space.

Language

  • English

Media Info

  • Features: Figures; References; Tables;
  • Pagination: p. 37-44
  • Monograph Title: Transportation planning, programming, land use, and applications of geographic information systems
  • Serial:

Subject/Index Terms

Filing Info

  • Accession Number: 00626919
  • Record Type: Publication
  • ISBN: 0309054036
  • Files: TRIS, TRB
  • Created Date: Feb 22 1993 12:00AM