中文摘要：

本文以有向拓扑结构的传感器网络为背景,通过分析在全局贝叶斯风险最小准则下已建立的集中式和近似分散式两种决策方法各自优势与不足,并考虑到现有的队决策方法求解决策过程中存在着局限性,结合贝叶斯公式和相关图模型理论等,建立了在P-B-P最优准则下针对此类网络的新决策方法.该方法分为在线计算和离线计算两部分,前者主要任务是基于在线测量值获得类条件概率密度;后者主要任务是如何求取用于计算阈值所需的参数,参数的计算过程分别是由正序传递似然函数消息和逆序接收代价函数消息两部分组成.同时,还分析了新方法在调节集中式和近似分散式两种决策方法的计算量和能耗之间矛盾的能力,而其优点也通过计算机仿真结果进行了验证.

英文摘要：

Considered the directed topology structure of the sensor network,on the basis of introducing the optimal centralized strategy and myopic decentralized strategy established by the rule of the minimum global Bayesian risk,and analyzing their merits and demerits,and considering the limitation of the team-theory solution,and combining the belief propagation with some algorithms in graphical models,a new strategy method to this sensor network is proposed by the rule of person-by-person optimality.It consists of online computation and offline computation,the former task is to obtain one class-conditional-probability density based on the online measurements,the later task is get some parameters which are used to calculate one threshold,parameter calculation procedure comprises one forward transmitting process of likelihood function message and another backward receiving process of cost-to-go function message.The new method capability,to trade off the conflict between calculated complexity and energy consumption from the optimal centralized strategy and myopic decentralized strategy,is analyzed,other advantage is also validated by computer simulation.