Enhanced Evolution of Parallel Algorithm Portfolio for Vehicle Routing Problem via Transfer Optimization

Parallel Algorithm Portfolio (PAP), comprising several component solvers with complementary capabilities, emerges as a cutting-edge computational technique for addressing computationally hard problems. Automatic construction of PAPs can develop high-performance PAPs without human intervention. It is...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on evolutionary computation p. 1
Main Authors: Guo, Tong, Mei, Yi, Zhang, Mengjie, Tang, Ke, Cai, Kaiquan, Du, Wenbo
Format: Journal Article
Language:English
Published: IEEE 2025
Subjects:
Artificial intelligence
Evolutionary computation
Feature extraction
Knowledge transfer
metaheuristic
Optimization
Parallel algorithm portfolio
Parallel algorithms
Portfolios
Training
transfer optimization
Vectors
Vehicle routing
vehicle routing problem
ISSN:1089-778X, 1941-0026
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Parallel Algorithm Portfolio (PAP), comprising several component solvers with complementary capabilities, emerges as a cutting-edge computational technique for addressing computationally hard problems. Automatic construction of PAPs can develop high-performance PAPs without human intervention. It is natural to believe that the component solvers share common useful building blocks, thus knowledge transfer among them could be beneficial during the evolution. However, this has been neglected by existing studies. To fill this gap, we propose a transfer optimization framework for automatically co-evolving high-performance PAPs. Specifically, we develop a novel performance-oriented dynamic instance grouping strategy to divide problem instances into groups, each of which is associated with a subpopulation of individuals tasked with evolving a component solver. Additionally, the framework incorporates an adaptive knowledge transfer strategy that automatically identifies when and how to transfer knowledge among instance groups. We conducted extensive experiments on the well-known Vehicle Routing Problem (VRP), a famously challenging NP-hard combinatorial optimization problem. The comprehensive experimental results from three public benchmarks demonstrate that our proposed framework significantly outperforms existing state-of-the-art VRP solvers and automatic PAP construction methods.
ISSN:1089-778X
1941-0026
DOI:10.1109/TEVC.2025.3616385