SPECIAL ISSUE: ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING TECHNOLOGIES

Solving large-scale routing optimization problems with networks and only networks

A. G. Soroka^a, A. V. Mesheryakov<sup>ab

a</sup> Lomonosov Moscow State University, Faculty of Computational Mathematics and Cybernetics, Moscow, Russian Federation
^b Space Research Institute, Russian Academy of Sciences, Moscow, Russian Federation

Abstract: For the first time, a fully neural approach has been proposed, capable of solving the optimization problem of routes of extremely large dimensions ($\sim$5000 points) with real-world constraints such as cargo capacity, time windows, and delivery sequencing. The proposed solution allows for rapid suboptimal problem-solving for small and medium dimensions ($<$ 1000 points). Meanwhile, it outperforms heuristic approaches for tasks of extremely large dimensions ($>$ 1000 points), thereby representing a state-of-the-art (SotA) solution in the field of route optimization with real-world constraints and extremely large dimensions.

Keywords: vehicle routing problem, reinforcement learning, deep neural networks.

UDC: 517.54

Presented: A. L. Semenov
Received: 05.09.2023
Revised: 15.09.2023
Accepted: 18.10.2023

DOI: 10.31857/S2686954323602014

 English version: Doklady Mathematics, 2023, 108:suppl. 2, S242–S247

Bibliographic databases:

•