基于multi-agent的多任务分配问题研究

传统的多agent多任务分配往往采用集中式的分配方法。由于集中式分配方法存在计算量大且易造成通信拥塞的不足,近年来又产生了各种分布式分配算法,但分布式算法容易陷入局部最优,影响分配效果。针对动态环境中任务复杂多样且agent能力差异,提出一种agent通信与结构框架,通过三个层次将集中式与分布式相结合,反馈地引入有利于分配过程的优化。根据任务的时间窗口和agent的代价设计合理的任务回报函数,提出一种分布式任务分配算法,对多任务进行合理分配,同时保证算法的鲁棒性。仿真结果表明,与单纯的集中式和分布式任务分配方法相比,混合系统能够完成全部的任务,且获得与集中式相当的任务收益,对框架下的各个age...

Full description

Saved in:
Bibliographic Details
Published in计算机应用研究 Vol. 31; no. 7; pp. 1980 - 1983
Main Author 刘科 郭小和 周继强 卢永乐
Format Journal Article
LanguageChinese
Published 南昌航空大学飞行器工程学院,南昌,330063%中航工业洪都飞机设计研究所,南昌,330024 2014
Subjects
Online AccessGet full text
ISSN1001-3695
DOI10.3969/j.issn.1001-3695.2014.07.013

Cover

More Information
Summary:传统的多agent多任务分配往往采用集中式的分配方法。由于集中式分配方法存在计算量大且易造成通信拥塞的不足,近年来又产生了各种分布式分配算法,但分布式算法容易陷入局部最优,影响分配效果。针对动态环境中任务复杂多样且agent能力差异,提出一种agent通信与结构框架,通过三个层次将集中式与分布式相结合,反馈地引入有利于分配过程的优化。根据任务的时间窗口和agent的代价设计合理的任务回报函数,提出一种分布式任务分配算法,对多任务进行合理分配,同时保证算法的鲁棒性。仿真结果表明,与单纯的集中式和分布式任务分配方法相比,混合系统能够完成全部的任务,且获得与集中式相当的任务收益,对框架下的各个agent的能力进行了充分利用,验证了通信与结构框架以及所提任务分配算法在动态环境多agent多任务分配过程中的有效性和合理性。
Bibliography:51-1196/TP
level distribution;decentralized;centralized;hybrid system;task allocation;time windows;cooperative control
LIU Ke, GUO Xiao-he, ZHOU Ji-qiang, LU Yong-le ( 1. School of Aircraft Engineering, Nanchang Hangkong University, Nanchang 330063, China ; 2. AV1C Hongdu Aircraft Design & Research In- stitute, Nanchang 330024, China)
Traditional multi-agent multi-task allocation often used centralized allocation method,however,it had many drawbacks,such as require a large amount of computational and cause network congestion easily,so different distributed allocation algorithms were arise in recently. While these algorithms were easy to fall into local optimal and affect distribution effects,this paper proposed an agent communication and structure framework according to diverse complex tasks in a dynamic environment and agent's capacity variance,and through three levels to combine centralized with distributed,the introduction of feedback was in favor of optimal allocation process. It put forward an distributed t
ISSN:1001-3695
DOI:10.3969/j.issn.1001-3695.2014.07.013