On the Robust Shortest Path Problem

Abstract

The shortest path problem is of great importance to the real world in such areas as transportation, network design, telecommunication, etc. The deterministic version of the problem is easily solved. However, in the real world, uncertainty is frequently encountered and must be dealt with. For instance, in a transportation network, the path that has the least travel time when no traffic is present might be prone to accidents and traffic jams, and thus it may lead to a drastic increase in travel time during rush hours. In situations with significant uncertainty, the deterministic approach can be far from sufficient. New and appropriate criteria and models to handle uncertainties along with efficient solution techniques are in need. In this paper, we take a minimax approach to measure the performance under uncertainties. Under our measure, one path is superior to all others only if its worst case performance is better than that of the others. The complexity of the defined models, exact and heuristic solution methods are explored and presented. Analysis is done to examine the performance of the heuristic method. The shortest path (SP) problem in a network with nonnegative arc lengths can be solved easily by Dijkstra's labeling algorithm in polynomial time. In the case of significant uncertainty of the arc lengths, a robustness approach is more appropriate. In this paper, we study the SP problem under arc length uncertainties. A scenario approach is adopted to characterize uncertainties. Two robustness criteria are specified: the absolute robust criterion and the robust deviation criterion. We show that under both criteria the robust SP problem is NP-complete even for the much more restricted layered networks of width 2, and with only 2 scenarios. A pseudo-polynomial algorithm is devised to solve the robust SP problem in general networks under bounded number of scenarios. Also presented is a more efficient algorithm for layered networks. However, in the case of unlimited number of scenarios, we show that the robust SP problem is strongly NP-hard. A simple heuristic for finding a good robust shortest path is provided, and the worst case performance is analyzed.