The Application of the Ant Memory System on the Vehicle Routing Problems

• Main Authors: Jing-Tang Lai, 賴敬棠
• Other Authors: Da-Jie Lin
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/85059338282676640005

碩士 === 逢甲大學 === 交通工程與管理所 === 98 === Transportation cost is a key component of Logistics cost, hence reducing daily operating costs such as VRP (Vehicle Routing Problems) have been of major concerns. VRP is a problem focusing on reducing total distance that commercial vehicles travel to serve customers. It is a NP-Hard problem and solving a VRP is very time consuming so exact solutions won’t be generated in a timely fashion. In order to overcome this efficiency issue, different heuristic algorithms have been applied in VRP and different results have been achieved. Ant System (AS) is one meta-heuristic algorithm that was based on the behavior of ants. AS performs well in solving various NP-Hard problems and several modified versions have been developed, including Ant Memory System (AMS). AMS was developed by author and very good results have been obtained when it is applied in TSP (Travelling Salesman Problems). The design of memory boxes in AMS is able to avoid accepting local optimum that was found in the solving process. This study plans to find out the effectiveness of the AMS in solving VRP. AMS was modified for VRP and some local search mechanisms such as 2-opt were adopted to develop a new approach. International benchmark problems were used to test the efficiency of the modified algorithm. Meanwhile a tool has been developed to observe the iteration in the solving process visually and that would give us significant insights on improving AMS. Commonly used the benchmark problems were tested in this study and the result is compared with the results from other revised algorithms. The comparisons show that AMS provides better approximation solution than the other algorithms. In the random problem C1 ~ C10 average error of 0.98%, in the cluster problem C11 ~ C14 average error of 0.55%.