论文部分内容阅读
将多 Agent 影响图(MAIDs)在时间上进行扩展,提出一种决策模型:多 Agent 动态影响图(MADIDs),用于表示动态环境中多 Agent 协作的结构关系.为了有效计算 MADIDs 的概率分布,以 Agents 之间的策略偏序关系为指导,给出概率分布的一种分解近似方法,进而讨论概率分布在推理中的近似.对 MADIDs 概率分布计算的复杂性、误差以及误差在时间上的传播进行分析,进而基于 KL 差分,给出一个可对近似分布的精度和复杂性进行均衡的函数.最后,针对一个表示协作关系的 MADID 模型,进行实验和算法比较,实验结果显示该概率分布近似方法的有效性.
In this paper, MAIDs are extended in time and a decision model is proposed: Multi-Agent Dynamic Effect Maps (MADIDs) are used to represent the structural relationships of multi-agent collaboration in a dynamic environment.In order to effectively calculate the probability distribution of MADIDs, Guided by the strategy partial order relationship among Agents, a decomposition approximation method of probability distribution is given, and then the approximation of probability distribution in inference is discussed. The complexity, error and error propagation in time of MADIDs probability distribution And then based on the KL difference, a function that can balance the precision and complexity of the approximate distribution is given.Finally, a comparison of the experiment and the algorithm for a MADID model representing the collaboration relationship shows that the probability distribution approximation method Effectiveness.