An Experimental Study of Large-scale Capacitated Vehicle Routing Problems

The recently proposed Scalable Approach Based on Hierarchical Decomposition (SAHiD) has shown its superiority on large-scale capacitated arc routing problems (CARP) in terms of both computational efficiency and solution quality. The main idea of SAHiD is that the underlying Hierarchical decomposition (HD) scheme is able to efficiently obtain a good permutation of tasks for CARP in a hierarchical divide-and-conquer way, where both the number and size of subproblems can be kept in tractable for large-scale problems with thousands of tasks. Motivated by the frequent observations of the similarity between CARP and Capacitated Vehicle Routing Problem (CVRP), the HD scheme and SAHiD algorithm are expected to work well on CVRPs. This paper applies SAHiD to large-scale CVRPs and discovers that SAHiD does not work as well as expected on large-scale CVRP. Possible reasons for this are given after extensive experimental studies. Two directions for improving SAHiD on large-scale CVRP are pointed out.