基于算法集成的调度系统多智能体实现框架

在混合算法研究的基础上,提出了调度系统中算法集成的概念,建立了算法类库。以多智能体系统为实施平台,给出了一种新的具体实现框架。系统中智能体可分为构造型、改进型和决策型三类,且无主控智能体,各智能体间不直接通讯,而是通过共享内存进行合作。描述了单智能体的基本结构。考虑到领域专家知识在调度系统中的重要性,将调度员作为一个特殊的智能体引入调度系统,改善了决策过程。最后分析了系统的特点。

Multi-agent Architecture of Scheduling System Based on Integrated Algorithms

We proposed the concept of integrated algorithms in scheduling system, based on the study of hybrid algorithms. An algorithm class library was founded. We explored an agent-based architecture, in which integrated algorithms were encapsulated intelligently as agents. In our architecture, there are three types of agents including construction , improvement and decision agents. There is no central controller, and agents cooperate by evolving a shared population of solutions instead of inter-agent communication. Single agentg architecture was studied as well. We have enhanced the architecture to enable human scheduler to participate as agents so that the human scheduler can provide deep knowledge of the problem domain not captured by software agents. The human scheduler thus plays a highly interactive role in our system.