Raymond R. Hill (Advisor), James T. Moore (Committee Member), George G. Polak (Committee Member), Zhiqiang Wu (Committee Member), Xinhui Zhang (Committee Member)
Doctor of Philosophy (PhD)
The classic vehicle routing problem considers the distribution of goods to geographically scattered customers from a central depot using a homogeneous fleet of vehicles with finite capacity. Each customer has a known demand and can be visited by exactly one vehicle. Each vehicle services the assigned customers in such a way that all customers are fully supplied and the total service does not exceed the vehicle capacity. In the split delivery vehicle routing problem, a customer can be visited by more than one vehicle, i.e., a customer demand can be split between various vehicles. Allowing split deliveries has been proven to potentially reduce the operational costs of the fleet.
This study efficiently solves the split delivery vehicle routing problem using three new approaches. In the first approach, the problem is solved in two stages. During the first stage, an initial solution is found by means of a greedy approach that can produce high quality solutions comparable to those obtained with existing sophisticated approaches. The greedy approach is based on a novel concept called the route angle control measure that helps to produce spatially thin routes and avoids crossing routes. In the second stage, this constructive approach is extended to an iterative approach using adaptive memory concepts, and then a variable neighborhood descent process is added to improve the solution obtained.
A new solution diversification scheme is presented in the second approach based on concentric rings centered at the depot that partitions the original problem. The resulting sub-problems are then solved using the greedy approach with route angle control measures. Different ring settings produce varied partitions and thus different solutions to the original problem are obtained and improved via a variable neighborhood descent.
The third approach is a learning procedure based on a set or population of solutions. Those solutions are used to find attractive attributes and construct new solutions within a tabu search framework. As the search progresses, the existing population evolves, better solutions are included in it whereas bad solutions are removed from it. The initial set is constructed using the greedy approach with the route angle control measure whereas new solutions are created using an adaptation of the well known savings algorithm of Clarke and Wright (1964) and improved by means of an enhanced version of the variable neighborhood descent process. The proposed approaches are tested on benchmark instances and results are compared with existing implementations.
Department or Program
Ph.D. in Engineering
Year Degree Awarded
Copyright 2009, all rights reserved. This open access ETD is published by Wright State University and OhioLINK.