Artificial Intelligence based Profit-Sharing Algorithm in Multi-Agent Systems

In a dynamic environment, it is quite reasonable to implement functions as a set of rules into an agent to react to unexpected situations, because it is very difficult to have a correct model of the environment. This paper describes a reinforcement studying method to acquire the rules, each of which consists of state and action pair, for coordinating execution of the multiple cranes in a coil-yard of steel manufacture. The cranes are operated asynchronously, based on the decision of each crane’s operator who only knows locally available information in this domain where it is natural to use a decentralized multiagent model. However, decentralization causes other serious problems, such as conflicts among the agents. It is very difficult to design rules to resolve conflicts by means of mathematical analysis, because information is scattered, missions are generated stochastically and it is very hard to execute missions on schedule. In this paper, we, here, give a kind of Profit-sharing strategy, which is based on artificial intelligence. Profit-sharing strategy solves the issue that when conscious aliasing and concurrent studying occur. In a true multi-agent environment, a fact is that movement actions means more under certain circumstance, if you consider the surroundings as MDPs. Since Profit-sharing is an exploitation intensive method, it reinforces meaningful actions remarkably to resolve the conflicts.