Deep Learning-based Heuristic Construction for Routing Problems with Dynamic Encoder and Dual-Channel Decoder Architecture

Jianlin Chen, Jianping Luo

Keywords: Routing problem, Combinatorial optimization, Heuristics, Deep learning

Abstract: The routing problem is a classic combinatorial optimization challenge. Constructing heuristics using deep learning models presents a promising approach for its resolution. In this paper, we propose a novel model with a dynamic encoder and dual-channel decoder (DEDD) architecture to learn construction heuristics for the routing problem. The dynamic encoder en-codes the node features of the decomposed sub-problems at each selection step, thereby obtaining more accurate node em-beddings. The dual-channel decoder facilitates more diverse node selections at each step, increasing the probability of the model identifying optimal solutions. Additionally, we design an effective node selection strategy to assist the model in choosing nodes at each step. Experimental results on the Traveling Salesman Problem (TSP) and the Capacitated Ve-hicle Routing Problem (CVRP) with up to 1000 nodes demonstrate that the solutions generated by the DEDD model are nearly optimal, underscoring its efficacy.

Primary Area: optimization

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.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

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: 2930