Online Learning Search for Stackelberg Game Strategy

Abstract

This paper deals with the public service system design problem, in which we consider more service providers competing for the profit. We assume that the profit of a provider is proportional to transportation performance necessary to satisfy the demands of system users. The considered profit may be enlarged by system reengineering, which consists in changing current positions of some centers operated by the considered provider. Since each center relocation causes some reaction of other providers to maximize their profit, the providers' behavior can be modelled by Stackelberg's game. In this paper, we introduce an approximate solving technique for a special Stackelberg's game played by a leading service provider and other mutually cooperating providers represented by the follower. The strategy of the leading service provider may be sensitive to some parameters. To find their suitable settings, we suggest an online learning algorithm.