The Method And Application Of Agent Coalition Formation In Distributed Setting

Forming coalitions for a group of tasks efficiently in a distributed settingis a practical problem and a hot research topic in agent research. Most ofcurrent solutions for this problem are based on the assumption that there is norelationship between these tasks. However, logical interdependentrelationships may exist between tasks so that transfer costs will be incurredwhen interdependent tasks are assigned to different agents. In order to solvethis problem, a new coalition formation strategy needs to be proposed.In response to these issues, the main efforts of this paper are as follows:1. As all the related work of agent coalition in distributed setting cannotbe applied to the task process with logical interdependentrelationship and transfer cost, meanwhile, exiting distributedMulti-Agent Systems generally have a low speed and accuracy informing coalitions. To solve this problem, a coalition formationstrategy is proposed. In this strategy, it is allowed that agents disclosesome personal cost information to compete for tasks in a controlledway. Experimental results show that, under the labor-based profitdistribution framework, the information disclosure strategy is fasterin forming coalitions and increases coalition net profit compared tothe traditional information non-disclosure strategy.2. Make a deep discussion on the problem of making use of moresimilar tasks to learn the historical coalition information to speed upcoalition formation for new task. For the two key points ofdistribution and multi-task, we proposed two different learningstrategies: learning agent’s cost information and learning task’s structure. We have a detailed description on these two strategies insides of background, algorithm and experiment. At last we made aanalyisis on these two methods’ advantage and disadvantage.3. Using an example of scientific workflow, we in detail discussed thewhole process of agent coalition formation through rounds ofnegotiation. Algorithm’s main idea has been totally expressed in thisexample. Finally in order to reduce the whole budget, we introducethe idea of bilateral negotiation to make an optimization for thisproblem.