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.
Multi-Agent dynamic influence diagrams (MADIDs) are presented by extending Multi-Agent Influence Diagrams (MAIDs) over time. Thus the structural relationships of coordination can be represented in dynamic environment. With the guidance of the strategic relevance, a decomposition approximation method of probability distribution and the approximation of probability distribution in inference are discussed t,o compute the probability distribution of MADIDs efficiently. The complexity, inducing error and error propagation over time are analyzed. Furthermore, based on the KL-divergence, a function is introduced to establish equilibrium between the precision and the complexity of approximate distribution. Finally, the experimental results on a dynamic coordination model show the validity of the probability distribution approximation method.