An Improved Branch-And-Price Algorithm To The Capacitated Vehicle Routing Problem With Time Window
Abstract
Motivated by real-life needs to reduce high operational costs and satisfy growing
delivery demand as well as the essence of its NP-hardness, this study aims to construct
an effective solution method to the state-of-the-art Capacitated Vehicle Routing Prob lem with Time Windows (CVRPTW). Even though the dynamics of real-world scenarios
results in a variety of routing problems with specific settings, these problems are of ten easy to tailor to meet their requirements. We believe it is significant to study a
generic problem and improve its solution approach. We thus decided to study the classic
CVRPTW in this project. Two major solution approaches to this problem are exact solution
approach and heuristic-based approach. This study centres on a Branch-and-Price
(BnP) algorithm that combines both exact solution and heuristic-based approaches via
a set-covering formulation and the efficient Column Generation (CG) technique. The
BnP algorithm adopts Iterated Local Search (ILS) to improve initial solutions and feed
them into the Master Problem (MP). After that, the subproblem is treated as a Shortest
Path Problem with Resource Constraints (SPPRC), in which solutions with negative
reduced costs are iteratively generated based on the classic labelling algorithm. This study
provides a detailed explanation of modeling procedure and provides several numerical
examples for illustration. A total of 24 Solomon benchmark instances were tested to attain
computational results of the new solution method. Other heuristic and exact methods
are also implemented for comparison of solution quality and time complexity. Based on
result analysis, it is demonstrated that the meta-heuristic ILS is able to improve initial
solution significantly compared to the well-known greedy heuristic, generating more
meaningful feedback from the MP for the subproblem. In addition, due to small
improvement in labelling approach, the subproblem is also sped up. In the last part of this
study, we discuss potential directions for future research on this topic