Evaluation Environment for Robust Planning in Scheduled Transportation

[leenasuhl]

We consider the planning process of vehicle and crew scheduling in scheduled passenger traffic, especially for airlines and public transit. Today there are information systems available that can be used to optimize the resource assignment and to minimize costs under given restrictions. Often very complex legal and contractual restrictions are given, such as nightly rest time, maximal flight time, maximal number of landings of pilots per duty, and so on. However, the optimized schedules often cannot be executed as they were planned, because short-term changes and disruptions affect the traffic environment. This in turn implies increase of costs so that the minimal cost plans cannot be executed in practice.
The ultimate goal is thus to generate schedules that are robust in case of short-term changes and disruptions. This means that their execution is not strongly affected by external changes, so that the total executed cost can be minimized instead of the planned costs. Stability and flexibility are important characteristics of robustness. With stability we mean here that a plan remains stable against external changes as it is able to absorb disruptions, whereas flexibility denotes that there are possibilities of making changes without high cost as a reaction to disruptions.
The project is part of a research program on robust planning for airlines and public transit. In previous projects of the DS&OR Lab an environment for airline traffic simulation has been built that includes components to generate vehicle and crew schedules and to simulate their execution with different control strategies in case of disruptions. The disruptions are expressed as delays according to a simple statistical distribution. However, the simulation results are not fully based on the practical situation, because the delays do not represent the way how delays and disruptions occur in practice. In reality, factors such as bad weather or technical problems often imply several delays in a short time, and on top of that the number of unpunctual flights is multiplied through delay propagation within the traffic network.
The goal of this project is to analyze the realistic delay occurrence and build a model of it as a basis for generating and evaluating robust planning methods. As a starting point there are preliminary studies available carried out with a punctuality data base from Deutsche Lufthansa AG. These preliminary models are able to characterize days with a high or low delay frequency with data mining technologies. Within this project, the use of this information in simulation studies with different delay management strategies is to be analyzed. Methodologically, methods from statistics, mathematical programming as well as heuristic algorithms can be used.