Jadwal Sholat

Memuat jadwal sholatโ€ฆ

Ilmu Komputer & AI editorial

Open AccessOA2024

Machine Learning to Solve Vehicle Routing Problems: A Survey

This survey systematically reviews machine learning methods for solving NP-hard Vehicle Routing Problems (VRPs), presenting a taxonomy of learning paradigms, solution structures, models, and algorithms. It demonstrates that learning-based methods are competitive with traditional approaches and highlights future directions for modern transportation systems.
Aigerim Bogyrbayeva; M. Meraliyev; Taukekhan Mustakhov; Bissenbay Dauletbayevยท IEEE transactions on intelligent transportation systems (Print)ยท 2024ยท DOI 10.1109/TITS.2023.3334976

The core problem

Vehicle Routing Problems (VRPs) are a class of NP-hard combinatorial optimization problems with wide applications in logistics, transportation, and supply chain management. Traditional solution methods rely on handcrafted heuristics and exact algorithms, which can be computationally expensive for large-scale instances. Recently, the machine learning and operations research communities have shown great interest in solving VRPs using pure learning methods or hybrid approaches that combine learning with traditional heuristics. This survey provides a systematic overview of machine learning methods applied to VRPs, aiming to categorize existing studies and identify future research directions.

Innovation

The authors conduct a systematic literature review of machine learning approaches for VRPs. They propose a taxonomy based on four dimensions: learning paradigms (e.g., supervised, reinforcement, unsupervised), solution structures (e.g., construction, improvement), underlying models (e.g., graph neural networks, sequence-to-sequence models), and algorithms (e.g., pointer networks, transformer architectures). The survey covers studies from both machine learning and operations research communities, including pure learning methods and hybrid approaches that integrate learning with traditional handcrafted heuristics. Detailed results of state-of-the-art methods are compiled and compared.
Introduction
Vehicle Routing Problems (VRPs) are a class of NP-hard combinatorial optimization problems with wide applications in logistics, transportation, and supply chain management. Traditional solution methods rely on handcrafted heuristics and exact algorithms, which can be computationally expensive for large-scale instances. Recently, the machine learning and operations research communities have shown great interest in solving VRPs using pure learning methods or hybrid approaches that combine learning with traditional heuristics. This survey provides a systematic overview of machine learning methods applied to VRPs, aiming to categorize existing studies and identify future research directions.

Why it matters

The authors analyze the advantages of machine learning-based models, especially their ability to exploit the symmetry of VRP solutions, which is often overlooked in traditional methods. They discuss the trade-offs between solution quality and computational time, and the challenges of generalizing to different VRP variants. The survey also outlines future research directions, including:

- Incorporating learning-based solutions into modern transportation systems to address dynamic and stochastic VRP challenges.
- Developing more interpretable and robust learning models.
- Bridging the gap between theoretical advances and practical deployment.

Overall, the survey provides a comprehensive roadmap for researchers and practitioners interested in applying machine learning to vehicle routing problems.

Who should read this

Pembaca Research Digest dan praktisi rumpun ini.

Opening member contentโ€ฆ