A Particle Swarm Optimization hyper-heuristic for the Dynamic Vehicle Routing Problem

TL;DR

This paper introduces a hyper-heuristic approach that uses a linear model to select the most suitable Particle Swarm Optimization algorithm for the Dynamic Vehicle Routing Problem, improving solution quality on benchmark instances.

Contribution

It proposes a data-driven hyper-heuristic method that predicts the best PSO algorithm for dynamic routing problems based on initial data, enhancing decision-making accuracy.

Findings

The model improved average results by 82% in significant cases.

Two advanced multi-swarm PSO algorithms were used as benchmarks.

The approach effectively selects algorithms based on initial problem data.

Abstract

This paper presents a method for choosing a Particle Swarm Optimization based optimizer for the Dynamic Vehicle Routing Problem on the basis of the initially available data of a given problem instance. The optimization algorithm is chosen on the basis of a prediction made by a linear model trained on that data and the relative results obtained by the optimization algorithms. The achieved results suggest that such a model can be used in a hyper-heuristic approach as it improved the average results, obtained on the set of benchmark instances, by choosing the appropriate algorithm in 82% of significant cases. Two leading multi-swarm Particle Swarm Optimization based algorithms for solving the Dynamic Vehicle Routing Problem are used as the basic optimization algorithms: Khouadjia's et al. Multi-Environmental Multi-Swarm Optimizer and authors' 2--Phase Multiswarm Particle Swarm Optimization.