Two phase SOFM
Doo Sung Hwang, Mun Sung Han · 2002
Discusses applications of a self-organising feature map (SOFM) to a travelling salesman problem (TSP) and proposes a procedure which prevents route intersection. Route intersection could hardly be removed if it once occurred in the course of finding an optimal tour for TSP, which forces SOFM to converge to a local minimum in the early stage. The authors divide the SOFM learning process into two phases. Phase one uses the learning rule that includes the relationship between a winner and two nearest neighbors, and phase two uses the Kohonen learning rule. Through simulations on 10, 30, and 100 city problems, the authors test the efficiency of the proposed method.>