Solving the Traveling Salesman Problem with a Multi-Agent System

We propose a good solution to the traveling salesman problem (TSP) of N cities with fixed resource using a Multi-Agent System. Our strategy involves a partition of the cities into m groups and assign a single agent to search for the shortest path that covers all the cities in the assigned group using the best TSP solver available. Our method is practical as we cannot find the exact solution by exhaustive search when N is large but the computational resource is limited. The theoretical problem needed to solve is to optimize the number m
^*
of agent that produces the smallest total cost, assuming that the hiring of additional agent incurs a fixed cost c
_s
and a cost of time c
_t
to covering all N cities. We find an optimal number m
^*
that is a function of N, c
_s
and c
_t
, as well as the computational powered provided. This strategy in solving traveling salesman problem with finite computing resources is tested on a random point set pattern on the plane and a real case "nrw1379" in the TSPLIB library and the model TSP solver used is based on genetic algorithm.