Rewriting by Generating: Learn Heuristics for Large-scale Vehicle Routing Problems
The large-scale vehicle routing problem is defined based on the classical VRP with usually more than one thousand customers. It is of great importance to find an efficient and high-qualified solution. However, current algorithms of VRP including heuristics and RL based methods, only performs well on small-scale instances with usually no more than a hundred customers, and fails to solve large-scale VRP due to either high computation cost or the explosive exploration space that results in model divergence. Based on the classical idea of Divide-and-Conquer, we present a novel Rewriting-by-Generating(RBG) framework with hierarchical RL agents to solve the large-scale VRP. RBG contains a rewriter agent that refines the customer division and an elementary generator to generate regional solutions individually. Extensive experiments demonstrate our RBG framework has significant performance on large-scale VRP. RBG outperforms LKH3, the state-of-the-art method in solving VRP, by 2.43% with the problem size of N=2000 and could infer solutions about 100 times faster.
PDF Abstract