摘要
本文对于多Agent系统在仿真系统中的应用,特别是在对实时性有一定要求的仿真系统中必须要满足的时间和空间复杂度要求,提出了一种分析和解决的方案,探讨了使用JAVA语言的多线程技术实现多Agent系统的方法,并构筑了一个电信网络仿真系统。
The implementation of Multi-Agent system has become an improtant research direction in AI. In this paper we present an approach to apply the Multi-Agent system into emulation system and describe some methods that satisfy the requests of spatio-temporal complexity in real time Multi-Agent System.
In emulation system, the Multi-Agent system has four outstanding characteristics: collaboration, distribution, concurrency and real-time. According to these characteristics, their corresponding methods are introduced. (1)Combining hierarchical control structure with centralized control structure. Agents are partitioned into separate groups according to the real-time characteristic, and the collaboration between agents in the same group is implemented by centralized controls. (2)Dynamic topology structure. For the various living time of agents, dynamic topology agent society can build and release such agents dynamically to reduce the occupation of resources. (3)Mixed MAS structure. Agents are grouped into different classes and communicate with each other by communication protocol. Every agent may consist of many kinds of sub-agents and process distributing transaction inside it. (4) Implementation of real-time message transfer using message pools, message selector and message interpreter. Actually real-time mutual communication is the key of real-time emulation system.
In this paper an emulation system simulating a telecom network is carried out using Java. It can be viewed as a Multi-Agent society and every switch in this network is an independent, concurrency, distributive and collaborative agent. In our emulation agent society, agents are grouped under four sets: Interface agent, Switch agent, sub-Agent and Forecaster agent. The Switch agent is the kernel of multi-agent system. Based on feedback message from Forecaster agent, it makes decision by itself. The Sub-Agent is a kind of controlled agent by the Switch agent. As soon as Switch agent need to select the best bypass route, it build a sub-Agent to fulfill this task. The Interface agent maintains the interface of emulational system and sharing message in Agent society. The Forecaster agent adopts neutral networks to forecast the future load on certain route by learning the bypass actual loads on every route in the network. The messages in our system can be grouped into three kinds: Call Request Message, Relay Request message and Confirm Message. According to these messages and emotional demand of telecom system, there are three kinds of primitives in communication between multi-agents: Relay Request, Confirm and Call Inform.
The Multi-Agent based emulation system adopts the modified agent structure and message transfer scheme to satisfy the real-time and concurrency character. We tested the Intelligent Route Resource Allocation policy on the emulation system and achieved satisfying results.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2001年第3期266-271,共6页
Pattern Recognition and Artificial Intelligence
基金
上海贝尔技术中心资助
