An Ant Colony Algorithm Based Partition Algorithm for TSP

Optimization algorithms inspired by models of co-operative food retrieval in ants have been unexpectedly successful and become known in recent years as Ant Colony Optimization (ACO).As a novel computational approach, swarm intelligence systems such as ant system have become a hot research domain. This paper proposes a meeting algorithm and a partition algorithm for TSP based on typical ant algorithms. The meeting algorithm improves the ant touring quality, it provides good initial touring results for local optimization on the condition of low iteration times. Combining with a simple parallelization strategy--the partition algorithm, this paper gets some good experiment results on Traveling Salesman Problems.The main idea of meeting algorithm is that there are two ants in a touring instead of one ant in typical ant algorithms. The two ants start a touring from a same city, and choose cities from different directions. By sharing a same tabu list, they meet at the middle of a touring. Experiment results show the two-ant touring is slightly shorter than the one-ant touring. Based on the shorter touring by two ants on the condition of low iteration times, a parallelization strategy is developed. It is the partition algorithm. The whole path is partitioned into several segments, then adapted meeting algorithm is again applied on the segments. Three TSP instances are used in this paper. They are ST70(70 cities), KroB150 (from TSPLIB) and CHC144(Chinese 144 cities). Comparing the best results available, 678.59( from the best path provided by TSPLIB), 26130 (from TSPLIB) and 30354.3 (by GA), some experiment results on the synthesized algorithm of meeting algorithm and partition algorithm are rather better. They are 677.1096, 26127.35 and 30354.3.