基于multi-agent的分布式专家系统原理及应用

来源期刊：中南大学学报(自然科学版)2001年第1期

论文作者：李松仁 景广军 陈松乔 梁雪梅

文章页码：20 - 24

关键词：智能体; multi-agent;选矿;分布式专家系统

Key words：agent; multi-agent; mineral processing; distributed expert system

摘    要：基于Internet/Intranet松耦合环境设计分布式选矿专家系统,采用面向agent(智能体)的系统开发方法和multi-agent分布协作求解模式,采用异构异质agent,即agent可建构于广域网、局域网和单处理器紧耦合模式,且求解a-gent可以是基于符号系统、神经网络、遗传算法等agent的协作.此外,建立了统一的知识库标准,基于关系数据模式表征了传统的规则、框架、语义网络及谓词逻辑等知识表示方法.其中,基于数据库引导及搜索技术实现了正向推理机制及反向推理机制,易于agent之间知识的传输,大大提高了通信效率;基于DCOM技术,以C++构建框架类加速了agent的实现.以矿石可选性预测求解为例,实现了多agent协作求解的仿真系统,较好地验证了所提出的理论模型及框架.

Abstract: The design of distributed Mineral Processing Expert System based on Internet/Intranethas been explored under slack coupling with agent-oriented programming (AOP) method and multi-agent distributed cooperating solution pattern.Isomery and alloplasm agent is proposed, that is, agent may be built on WAN, LAN and single CPU by tense coupling mode, and solution agent can work among agents based on symbol system, neural network, genetic algorithm. Unified knowledge base standard of mineral processing is built. Based on relational data pattern and search engine, traditional rule, frame, semantic model and predicate logic knowledge representation methods, and forward inference engine, backward reasoning mechanism are realized. This makes it easy for knowledge to transfer among agents, and improves communicating efficiency greatly. Based on DCOM (distributed component object model) technology, building frame class by C++, it accelerates the execution of agent. Take ore beneficiability prediction problems for example. The simulating system of multi-agent cooperating solution has been carried out, which can preferably verify the theoretic models and frames in this thesis.