Summary: | 碩士 === 國立中正大學 === 資訊工程研究所 === 107 === The traveling salesman problem (TSP) is a classic combinational problem, related to realworld
applications, such as VLSI design problems, transportation problems, and printed
circuit board drilling problems. The goal of TSP is to find the shortest route that goes
through every city once. The 2-opt operator is a popular local search method for TSP.
Basically, 2-opt eliminates two edges from the tour and forms a new tour by reconnecting
the two additional edges. If the distance of the new tour is better than the old tour, the
old tour replaced by the new tour. However, the neighborhood size will increase rapidly
when dealing with large scale TSP instances. Therefore, the running time significantly increases.
In this study, we propose an efficient segmented local search framework and apply
the proposed framework to 2-opt (seg2opt) and Lin Kernighan heuristic (segLKH). The results
show that the running time of seg2opt implemented with GPU can achieve 3129 times
faster than the 2-opt implemented with CPU on instance usa13509. The convergence speed
of seg2opt and segLKH are faster than 2-opt and Lin Kernighan heuristic on six TSPLIB
instances, respectively. In addition, the seg2opt obtains the best search performance among
2-opt, Lin Kernighan heuristic, and segLKH. In conclusion, this proposed framework increases
the convergence speed with the same solution quality.
|