Disruption management in vehicle routing and scheduling

Zambirinis, Sofoclis and Eglese, Richard (2019) Disruption management in vehicle routing and scheduling. PhD thesis, Lancaster University.

Text (2019zambirinisphd)
2019zambirinisphd.pdf - Published Version
Available under License Creative Commons Attribution-NoDerivs.
Download (1MB)

Abstract

Traditionally, people in modern business environments have been focusing on planning: creating detailed and complete schemes for actions that will lead to gains of the highest value. There is no doubt that constructing a thorough plan before taking actions is extremely important and usually a prerequisite element of success. However, no matter how perfect or optimal a plan is, during the execution phase, several unanticipated events may disrupt the system and force the plan to deviate from its intended course, or even make it infeasible. How should we cope with disruptions in a timely manner? How can we reach the original goals and at the same time minimize the negative impact which was caused by the disruptions? These are amongst the essential topics examined by the field of Disruption Management. Disruption Management has been applied by researchers to optimization problems arising in a wide range of applications, including airline scheduling and production management. In our research we focus on disruption management in vehicle routing and scheduling for road freight distribution, after having recognized several gaps in research in this specific domain. In this thesis we present the following three problems: (1) the disrupted Vehicle Routing Problem with customer-specific orders and Vehicle Breakdown, (2) the Delayed Traveling Salesman Problem with Time Windows, and (3) the Single-Commodity Delayed Vehicle Routing Problem with Time Windows. The second and third problems have never been studied before, to the best of our knowledge. The first one has been studied before under different assumptions (i.e. with non customer-specific orders), which differentiates substantially the problem from the one proposed here. For each problem we present at least one exact mixed-integer linear programming formulation (single-objective or multi-objective), which can be implemented in an optimization solver (e.g. Cplex or AIMMS) and solve small instances to optimality. Due to the fact that the problems under study are computationally hard, for each problem we also propose at least one heuristic algorithm, which is capable of solving larger instances in short time. The heuristics described in this thesis are all based on Tabu Search. We present several variants of problems 2 and 3, which are solved using both single-objective and multi-objective optimization approaches: the Weighting Method, the Lexicographic Approach, and the Epsilon Constraint Method. For each one of the three problems under study, we have constructed a dataset of test instances, which we solved using different approaches. Comparisons of the results of the exact and heuristic methods are provided for each problem.