Heuristic methods for variations of the Vehicle Routing Problem

Hamm, Rebecca and Boyaci, Burak and Kheiri, Ahmed (2025) Heuristic methods for variations of the Vehicle Routing Problem. PhD thesis, Lancaster University.

Text (2025hammphd)
2025hammphd.pdf - Published Version
Restricted to Repository staff only until 30 September 2026.
Available under License Creative Commons Attribution-NonCommercial-NoDerivs.
Download (4MB)

Abstract

Using efficient routes for vehicles impacts society, both environmentally and economically. This led to the research of the Vehicle Routing Problem (VRP). Due to its large real-world application, many variations of and methodology for the VRP have been studied. This thesis contains three works considering VRP variations. Two of these works study variations where location decisions are also made. This problem is the Location-Routing Problem (LRP). Due to the dependency between these decisions, solving them separately results in sub-optimal solutions. We explore methods of solution to these problems. In the first work, a VRP describing a commercial waste collection problem is addressed. A realistic problem is presented considering time-dependent travel times. A constructive heuristic, which utilises graph partitioning to group nearby customers is used. This heuristic uses a mechanism to prioritise remote customers. This method outperforms the previously used method on half the instances. The second work presents a two-echelon LRP which describes the distribution network of a large retailer. This problem considers routing decisions and the locations of cross-docking sites using existing sites and potential locations. A commercial solver is used to solve small instances. An ALNS heuristic is implemented to find solutions for larger instances and to improve on these solutions selection hyper-heuristics are developed. The selection hyper-heuristic was shown to be the best performing method on most instances. Within the third work, a Memetic Algorithm approach to a two-echelon LRP is explored. This algorithm is tested on three sets of instances from the literature. The memetic algorithm is a hybridisation of a genetic algorithm and selection hyper-heuristic. Multiple crossover and initial solution methods are considered. The method of sharing LLHs scores between the population is studied. Solutions reported are on average within 0.978% of the best-known solution and a best-known solution for an instance is found