基于多Agent动态影响图的协作实现

将MAIDs在时间上进行扩展,提出一种新决策模型——多Agent动态影响图（MADIDs）,对动态环境中的协作关系进行建模;给出MADIDs的一种分层分解的分布近似方法,进而通过将决策结点和效用结点的推理引入到BK算法中,给出MADIDs环境模型的一种扩展BK（EBK）近似推理算法;引入一种BP神经网络学习MADIDs的局部效用函数。最后,针对一个表示协作关系的MADID模型,进行算法比较和仿真实验,实验结果显示了MADIDs模型的有效性。

Coordination Problems Based on Multi-Agent Dynamic Influence diagrams

Multi-Agent dynamic influences(MADIDs)were proposed by extending MAIDs over time,and coordination relationships could be modeled in dynamic environment.Based on the hierarchical decomposition,a method of distribution approximation was discussed;further,an extensional BK(EBK)algorithm was given by adding inference of decision nodes and utility nodes to BK algorithm.Another,a BP Neural Network was given to learn local utility functions of MADIDs.On a dynamic coordination model,the algorithms were compared and the coordination relationships were simulated,and the results of experiments show the validity of MADIDs.