分布式环境下面向复杂任务的Agent联盟构建  

Agent Coalition Formation for Complex Task Under Distributed Environment

作　　者：王皓[1] 曹健[1] 

机构地区：[1]上海交通大学计算机科学与工程系,上海200240

出　　处：《计算机工程》2013年第12期216-222,共7页Computer Engineering

基　　金：国家自然科学基金资助项目(61073021);上海市科委基金资助项目(11511500102;10511501503;10DZ1200 200)

摘　　要：在分布式环境下,现有Agent联盟构建算法不能解决带有相互依赖关系和转移成本的任务流程问题。为此,利用Agent协商构建联盟,在协商过程中设定方案发布Agent和参与Agent,并对应设计以成本信息调整和盈利任务争取为主的决策算法。在Agent的反馈信息中加入争取信息,允许参与Agent在多轮协商中采用可控制的信息泄露机制,通过泄露自己的成本信息向方案发布Agent争取可获利的任务,经过多轮协商,形成最优联盟结构。实验结果表明,在按劳分配联盟总收益的模式下,相比传统的信息不泄露机制,该信息泄露机制能够更快地形成联盟,并且具有更高的联盟净收益和Agent平均收益率。In the distributed environment, existing Agent coalition formation algorithm can not solve the problem of task flow with logical interdependent relationships and transfer costs. In order to solve this problem, it uses Agent negotiation to build union. During the negotiation process, two roles are set up： publisher Agent and participant Agent, and two corresponding algorithms are proposed for each of them： focusing on cost information adjustment and high-profit tasks competition respectively. It is innovatively allowed that participant Agent discloses some personal cost information to publisher Agent to compete for tasks in a controlled way. By leaking their own cost information to publisher Agent for profitable task, it forms the optimal coalition structure after several rounds of negotiations. Experimental results show that under the labor-based profit distribution mode of coalition total revenue, the information disclosure mechanism is faster in forming coalitions and increases coalition net profit and Agents＇ average profit rate compared with the traditional information non- disclosure mechanism.

关 键 词：联盟构建 逻辑依赖关系 转移成本 分布式环境 按劳分配 信息泄露 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]