An Adaptive Hybrid Genetic and Large Neighborhood Search Approach for Multi-Attribute Vehicle Routing Problems

TL;DR

Empirical findings showcase that AHGSLNS not only competes effectively with ALNS under varying parameters but also exhibits superior performance in terms of convergence and stability.

Abstract

Known for its dynamic utilization of destroy and repair operators, the Adaptive Large Neighborhood Search (ALNS) seeks to unearth high-quality solutions and has thus gained widespread acceptance as a meta-heuristic tool for tackling complex Combinatorial Optimization Problems (COPs). However, challenges arise when applying uniform parameters and acceptance criteria to diverse instances of the same COP, resulting in inconsistent performance outcomes. To address this inherent limitation, we propose the Adaptive Hybrid Genetic Search and Large Neighborhood Search (AHGSLNS), a novel approach designed to adapt ALNS parameters and acceptance criteria to the specific nuances of distinct COP instances. Our evaluation focuses on the Multi-Attribute Vehicle Routing Problem, a classical COP prevalent in real-world semi-automated storage and retrieval robotics systems. Empirical findings showcase that AHGSLNS not only competes effectively with ALNS under varying parameters but also exhibits superior performance in terms of convergence and stability. In alignment with our dedication to research transparency, the implementation of the proposed approach will be made publicly available.

Figures (3)

• Figure 1: System and schematic layout of a warehouse section with three static stations (See more details in https://www.hikrobotics.com/en/mobilerobot/CTU.)
• Figure 2: Illustration of MAVRP in CTU system
• Figure 3: Solution Representation