Uma resolução do problema do caixeiro-viajante por mapa auto-organizável com aprendizado winner takes all

Abstract

A one-dimensional self-organizing map with winner takes all learning was used tosolve the traveling salesman problem. In this implementation, each neuron represents a vertex. Afterexecution, the order of neurons indicates the route that represents the solution. Simulations wereconducted with six instances of the TSPLIB base of sizes from 51 to 1379 vertices. A learning rate of0.8 and 3000 training epochs in all executions were used. This approach proved to be efficient andconsistent. However, the solutions are not better than those obtained by other researchers due to theuse of a learning technique with lower computational cost, the use of the same parameters for allinstances and the absence of an optimization algorithm for local search. Consequently, a solutionwith low computational cost was obtained.