A Self-Organizing Map Algorithm for the Traveling Salesman Problem

In this article, based on the convex-hull property of the TSP and the neighborhood preserving property of self-organizing map (SOM), an improved SOM neural network is proposed for the traveling salesman problem. The proposed SOM is firstly initialized by the nodes on the convex-full. It then evolves based on a principle of neuron creation and deletion, and yields a near-optimal solution to a problem at last. Simulations are performed on benchmark problems taken from TSPLIB. The simulation results show that the proposed algorithm can obtain near-optimal solutions to these benchmark problems effectively.