Decentralized problem solving using the double auction market institution

摘要

Decentralized decision making is an important problem in distributed artificial intelligence (DAI) and multi-agent systems (MAS). Given a multi-agent system, the actions or choices made by one agent can affect the actions or choices that can be made by other agents. Wellman (Artificial Intelligence Research, 1, 1–23, 1993) has recently proposed the approach of market-oriented programming for decentralized decision making. By transforming the decentralized decision making problem into a computational economy, market-oriented programming draws on the theory of general equilibrium to establish the existence of a competitive equilibrium. The competitive equilibrium of the computational economy represents the market solution to the original problem. General equilibrium theory, however, is institution free and provides no information about the dynamic process by which the competitive equilibrium is found. Wellman uses a variant of the market institution known as tâtonnements. Tâtonnement's requirement of strict synchronization of the individual agents restricts market-oriented programming to purely atemporal situations (problems that have a stationary environment and hence a stationary equilibrium). In this paper we suggest the use of the continuous double auction as a general framework for market-oriented programming. By using the continuous double auction, market-oriented programming can also be applied to problems where the environment, and therefore the equilibrium, is continuously evolving with time.