/AGH

[IEEE TKDE | TITS 2023] "Learning Large Neighborhood Search for Vehicle Routing in Airport Ground Handling" | "Neural Airport Ground Handling"

Primary LanguagePythonMIT LicenseMIT

Airport Ground Handling

This repository contains the implementations of our papers studying Airport Ground Handling (AGH) problems:

  • Learning Large Neighborhood Search for Vehicle Routing in Airport Ground Handling

Jianan Zhou, Yaoxin Wu, Zhiguang Cao, Wen Song, Jie Zhang, Zhenghua Chen

Paper    License    Paper

  • Neural Airport Ground Handling

Yaoxin Wu*, Jianan Zhou*, Yunwen Xia, Xianli Zhang, Zhiguang Cao, Jie Zhang

Paper    License    Paper

Note: These works were done in 2021 and 2022, respectively. Based on our experiments, we recommend building upon the code of the second work for further research.

Overview

Learning Large Neighborhood Search for Vehicle Routing in Airport Ground Handling

We propose a learning-based improvement framework to solve large-scale Airport Ground Handling (AGH) instances. Specifically, we leverage the Large Neighborhood Search (LNS) framework, which consists of a pair of destroy and repair operators, to decompose the global (intractable) optimization problem and re-optimize each sub-problem. The operation scheduling in AGH is formulated as a mixed integer linear programming (MILP) model. To mitigate the need of domain expertise, 1) our proposed framework directly operates on the decision variables of the MILP model; 2) we employ an off-the-shelf solver (e.g., CPLEX) as the repair operator to conduct re-optimization. Our method could efficiently solve large-scale AGH instances with hundreds of flights, while CPLEX would simply stuck, even when searching for a feasible solution.

framework

Neural Airport Ground Handling

We propose a learning-based construction framework to solve Airport Ground Handling (AGH) problems in an end-to-end manner. The studied problem is modeled as a multiple-fleet vehicle routing problem (VRP) with miscellaneous constraints, including precedence, time windows, and capacity. It is much more complicated than the simple VRPs (e.g., TSP/CVRP) studied in the major ML conferences. The proposed method could also serve as a simple learning-based baseline for further research on complicated VRPs (e.g., CVRPTW).

framework

File Structure

./
├── Construction_based               # The implementation of TITS paper
├── Improvement_based                # The implementation of TKDE paper
├── LICENSE
├── README.md
└── imgs

Citation

@article{zhou2023learning,
title       = {Learning Large Neighborhood Search for Vehicle Routing in Airport Ground Handling},
author      = {Jianan Zhou and Yaoxin Wu and Zhiguang Cao and Wen Song and Jie Zhang and Zhenghua Chen},
journal     = {IEEE Transactions on Knowledge and Data Engineering},
year        = {2023},
doi         = {10.1109/TKDE.2023.3249799},
publisher   = {IEEE}
}
@article{wu2023neural,
title       = {Neural Airport Ground Handling},
author      = {Yaoxin Wu and Jianan Zhou and Yunwen Xia and Xianli Zhang and Zhiguang Cao and Jie Zhang},
journal     = {IEEE Transactions on Intelligent Transportation Systems},
year        = {2023},
doi         = {10.1109/TITS.2023.3253552},
publisher   = {IEEE}
}