Some results on heuristical algorithms for shortest path problems in large road networks

This thesis studies the shortest path problem in large road networks. The classical algorithm for networks with non-negative edge weights is due to Dijkstra and has a worst-case performance of O ( |E |+ |V |log |V |) using a simple priority queue as data structure for temporarily labeled nodes. We present a new, so-called tree heuristic, which is based on the similarity of shortest path trees and which can be used to speed up the shortest path search especially in practical applications like microscopic simulation of traffic or route guidance systems. Instead of searching a path in the original network, the tree heuristic partitions the network into classes of about equal size and constructs a special searchgraph for each class. On a test road network of about one million nodes the tree heuristic outperforms Dijkstra\'s algorithm by a factor of more than three with respect to runtime and about seven with respect to permanently labeled nodes where the found paths can be expected to have a relative error below 1%, if the starting and end node are not too close to each other. We also analyze the A -algorithm with overdo-factor, originally devised for Euclidean networks and derive an interval [1.... 27......,5] from which an optimal overdo-factor should be chosen in practical applications. Finally we give an algorithm which calculates edge tolerances for a shortest path and which can be used to generate reasonable alternative routes to the exact shortest path