Vehicle Utilization in Hub Network Design: Exploiting Economies of Scale in Transportation

TL;DR

This paper introduces a vehicle-based hub network design model that captures economies of scale in freight and parcel delivery, offering a new approach with improved solution methods and applicability to uncertain demand.

Contribution

It develops a quadratic mixed-integer model and a linear reformulation for HNDPv, along with a branch-and-cut algorithm and solutions for stochastic demand, advancing hub network design optimization.

Findings

HNDPv outperforms classical HLP in vehicle utilization and costs.

The proposed algorithm efficiently solves large-scale instances.

The model effectively handles demand uncertainty.

Abstract

We study a vehicle-based hub network design problem (HNDPv) with the main applications in freight distribution and parcel delivery systems, where the economies of scale stem from the effective utilization of vehicles that move consolidated freight. The HNDPv is a generalization of the classical single allocation hub location problem, in which the transportation costs are stepwise functions of the number (and type) of vehicles that move the demand. We present the quadratic mixed-integer programming formulation of the problem and its linear reformulation. Exploiting the special structures of the linearized model, we develop a branch-and-cut method based on Benders decomposition with solely feasibility subproblems. We derive closed-form solutions for the extreme rays of the feasibility subproblems that improve the efficiency of the proposed algorithm through generating stronger feasibility cuts. We also address the HNDPv under demand uncertainty and show the flexibility of our solution methodology in handling the stochastic variant of the problem. To evaluate the efficiency of our models and solution approaches, we perform extensive computational experiments on uncapacitated and capacitated instances of the problem derived from the classical Australian Post dataset. The results show a considerable advantage of using HNDPv compared to the classical HLP with constant discount factors in terms of vehicle utilization and total transportation costs. Our computational experiments also demonstrate the efficiency of our proposed solution method in solving large-scale problem instances.