A comparative study of machine learning approaches for large-scale vehicle routing

• Main Author: Chan, Ray
• Other Authors: Zhang Jie
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Large-scale vehicle routing
• Online Access: https://hdl.handle.net/10356/175066
• Institution: Nanyang Technological University

With the surge in demand for delivery services, accelerated by the COVID-19 pandemic, and the proliferation of data, Vehicle Routing Problems (VRP) have grown in size and complexity. Despite proven effective solvers for smaller-scale problems, addressing large-scale VRPs remains a challenge. This project delves into the analysis, comparison, and exploration of two novel approaches aimed specifically at tackling Large-scale Vehicle Routing Problems. Aiming to provide insights into their effectiveness, efficiency, and applicability in real-world scenarios. We investigate the performance of the Neural Combinatorial Optimization (NCO) model and compare it with methods like Lin-Kernighan Heuristic (LKH3) and Learning to Delegate. Our analysis indicates that the NCO model consistently outperforms the benchmark algorithm LKH3 and Learning to Delegate for smaller datasets but exhibits diminishing performance for larger datasets, suggesting scalability limitations. Furthermore, NCO demonstrates stronger performance on Learning to Delegate datasets compared to its own, suggesting robust generalization capabilities.