DER-Solomon: A Large Number of CVRPTW Instances Generated Based on the Solomon Benchmark Distribution
Primary Area: datasets and benchmarks
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: backward derivation, capacitated vehicle routing problem with time windows, deep reinforcement learning, Solomon benchmark
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: This paper introduces an extended version of the Solomon benchmark applied in the CVRPTW field, DER Solomon, which extends a limited number of 56 instances to an infinite number of instances, making it applicable to deep reinforcement learning.
Abstract: The Solomon benchmark is a well-known resource for researching Capacitated Vehicle Routing Problem with Time Windows (CVRPTW), and has been used by many traditional methods. However, the limited scale of the Solomon benchmark poses challenges to effective utilization by learning-based approaches. To address this, we propose an expanded version with a large set of new instances, called DER-Solomon benchmark, which follows a similar distribution as the Solomon benchmark. First, we analyze the Solomon benchmark and use backward derivation to establish an approximate distribution, from which the DER-Solomon is generated, thereby significantly expanding the size of the benchmark. Next, we validate the distribution consistency between the DER-Solomon benchmark and the original Solomon benchmark using traditional algorithms. We then demonstrate the superiority and reliability of DER-Solomon compared to other similar Solomon-like datasets using state-of-the-art Deep Reinforcement Learning (DRL) algorithms. Finally, we train multiple DRL algorithms using the DER-Solomon benchmark and compare them with the traditional algorithms. The results show that the DRL algorithms trained on the DER-Solomon benchmark can achieve the same level of solution quality as the traditional algorithms on the Solomon benchmark while reducing the computational time by over 1000 times on CVRPTW. All the results demonstrate that the DER-Solomon benchmark is sufficiently excellent, serving as an extension of the Solomon benchmark, which offers valuable tools and resources for further research and solutions to the CVRPTW problem.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 875
Loading