Master Thesis MSTR-2020-38

BibliographyParga Cacheiro, Dominic: Learning metrics for balancing load in street-networks.
University of Stuttgart, Faculty of Computer Science, Electrical Engineering, and Information Technology, Master Thesis No. 38 (2020).
47 pages, english.
Abstract

Especially in daily rush-hour-scenarios, a street-network requires enough capacity to support the amount of drivers. On the other hand, a street-network of too much capacity would be inefficient outside of rush-hour-scenarios. Hence, the overload during rush-hour should be spreaded over the network to reduce the impact of the overload (\eg\ traffic-jams). To improve the distribution of routes, alternative routes in multicriteria settings are computed. To support multicriteria settings, Dijkstra's algorithm is combined with personalized routing. Here, these multicriteria or multiple metrics, \eg\ travel-time or travel-distance, can be reduced to scalars by applying preferences to them. Resulting routes or paths are called personalized paths. However, many previous approaches for computing alternative routes need too much parameter-tuning or simply lack in their computational complexity, their needed runtime or in the diversity of their found routes. Therefore, this thesis presents a combination of enumerating personalized paths with the creation of a new penalizing metric. The goal is to compensate other popular metrics with this new metric, which then improves the spread of many routes in the network. This new metric can be processed by every routing-algorithm, that is capable of dealing with multicriteria-routes. By enumerating personalized paths with this new penalization, found routes can be distributed successfully over the network. In addition, user-provided tolerances for preferred metrics are hold. In the end, some experiments on street-networks from OpenStreetMap are presented. Here, results are compared between routes from Dijkstra (with personalized routing) and the enumeration of personalized routes. To speed the route-queries up significantly, the underlying graphs of the networks are contracted via contraction-hierarchies before the searches happen. Here, the contraction is realized by a linear program to improve the performance with multicriteria contractions.

Department(s)University of Stuttgart, Institute of Formal Methods in Computer Science, Algorithmic
Superviser(s)Funke, Prof. Stefan, Barth, Florian
Entry dateDecember 17, 2020
   Publ. Computer Science