Abstract: To solve multi-level and multi-objective batch delivery collaboration optimization, the modeling tool was analyzed, and the modeling technology by combining Aglet based mobile Agent with Flexsim based simulation was selected. To avoid traditional multi-Agent system's shortcomings, a hybrid control structure was adopted in system. The solving frame of forging multi-level and multi-objective was designed, and multi-Agent collaboration model as well as Unified Modeling Language(UML)based collaborative process were given. An improved Look-ahead Batching Rule(LBCR)strategy based on improved Q-learning algorithm was proposed to serve as multi-level and multi -objective collaborative algorithm. The control experiment was used for production system's parameters which may affect the experiment result, and the tardiness as well as cost of different algorithms in uncertain condition were studied. The results showed that the proposed algorithm was better than the traditional LBCR strategies and the output of production system had strong stability.

Key words: forging, batch, multi-objective optimization, collaborative decision, multi-Agent, reconstruction, game theory

摘要： 为解决多级多目标组批投放决策协同优化算法,对建模工具进行分析，选择基于微软Aglet的移动智能体（Agent）和基于Flexsim的仿真相结合的建模技术，系统控制采用混合控制结构以避免传统多Agent系统的缺陷；设计了锻造多级、多目标组批协同求解框架，给出了多Agent协作模型和基于统一建模语言的协同过程。提出基于强化Q学习的改进型多目标前向投放控制策略作为多级多目标协同算法。对可能影响实验结果的生产系统参数设置作了对照实验，实验研究了不确定条件下不同算法的拖期数和成本表现，结果表明本文算法明显优于传统的前向投放策略,生产系统的输出具有很好的稳定性。

关键词: 锻造, 组批, 多目标优化, 协同决策, 多智能体, 重构, 博弈论