Diploma Thesis

Cyclists often have multiple criteria in mind when planning a route. They do not only consider the length of the route, but also the (perceived) safety or the environment of the route. Cycleways through park are often more attractive than heavily used roads. But not every cyclist weights these criteria the same way and even for one cyclist the weighting may depend on his current situation. In a hurry travel time might be the most important aspect, but even then memight make consider howmuch he will compromise his safety. In particular he will likely accept a slightly longer travel time, if the alternate route is perceived as much safer. Another often regarded criterion is the elevation difference of the route. We consider this five criteria in our approach. As basis for our algorithm we use A* search. To get suitable potentials we use a reverse Dijkstra search starting a the target node for each criterion. The big drawback of performing a multi-criteria search is that the number of routes may grow very quickly, which leads to a long query time as well as to a route set that does not help the user much in his rotue planning. Therefore we filter the routes during the computation to get a small amount of routes to help the user. To do this we use a variation of Pareto-dominance, in which a route are discarded if they would be dominated assuming it would be worse by an ε. Using this heuristic the number of found routes can be reduced by an significant amount and therefore the running time is reduced accordingly. AS a minor drawback the set of routes found with this approach depends on the order in which the routes are found. Therefore we tested and compared several difference order, which we used for our search. Out of the evaluated orderings the Euclidean distance to the potential of the source node resulted in the fewest routes, if we keep the same ε. Compared to the lexicographic order this resulted in 57 % fewer routes on average on a small graph. On a big graph it even reduced the number of routes by 90 %.