基于演化博弈观点的多源冲突证据融合研究

Dempster-Shafer evidence theory is widely used in the field of multi-source information fusion because of its relatively strong ability in dealing with uncertain information. However, in evidence theory there are some key open issues remained unsolved. Among them, the fusion of conflicting evidences is a core issue, which has important impact on enhancing the overall performance of information fusion systems. From the perspective of game and evolution, in this project an evolutionary game based solution is proposed for the conflicting evidence fusion. At first, a hierarchical model of conflict representation and measure has been presented by investigating the existing competitive and cooperative relations in the multi-source evidence system. Then on the basis of that, two game models, for evidences and propositions, have been built to implement the game situation analysis of multi-source information from the evidential and propositional levels respectively, and evolutionary dynamics mechanisms adapted to these two game models have been constructed for exploring the dynamical evolution process of evidences and propositions. Subsequently, based on the static game situation analysis and dynamical evolution process analysis, a two-layer evidence fusion algorithm based on evolutionary game is proposed to implement the efficient combination of conflicting evidences. At last, the multi-task management for aircrafts under uncertain environment has been used as the background to verify and apply the proposed algorithms of conflict analysis and evidence fusion to elevate the performance and intelligence of the multi-task management system. This research is of important significance for both the theoretical studies and engineering applications of information fusion.

证据理论以其较强的不确定信息处理能力在多源信息融合领域获得了广泛应用，但目前仍存在一些共性关键问题有待解决。其中，冲突证据融合是证据理论研究的核心问题，对提升信息融合系统的总体性能具有重要影响。本项目从博弈和演化的观点出发，提出基于演化博弈的冲突证据融合方案。首先，通过研究多源证据系统中存在的竞争与合作关系，提出层次化的冲突表示与度量模型。然后以此为基础，建立起证据博弈模型和命题博弈模型，以实现证据和命题两个层面的多源信息博弈态势分析；并构建与之相适应的演化动力学机制，探索证据或命题的动态演化过程。随后，在静态的博弈态势分析和动态的演化过程分析的基础上，提出基于两层博弈的证据演化博弈融合算法，实现冲突证据的有效融合。最后，以不确定环境下的机载多任务管理为背景，验证和应用所提出的冲突分析与证据融合算法，提升系统的性能和智能化程度。本项目的研究对于信息融合的理论探索与工程实践都具有重要意义。

证据理论是一种有效的不确定信息表示与处理理论工具，冲突证据融合是证据理论信息融合研究的核心问题。本项目针对多源证据系统中存在的信息竞争与合作关系，从博弈和演化的思想出发，研究基于演化博弈的冲突证据融合方案。通过三年来的研究，本项目在证据冲突分析、冲突证据博弈融合、证据不确定性度量、不确定信息推理、不确定信息最优决策与信息融合应用等方面取得了大量的研究成果：（1）开展了证据冲突来源研究，揭示了命题排他性和信度值多值离散性对证据冲突的影响。（2）建立了多种冲突证据博弈融合模型。提出了基于单种群演化博弈的冲突信息选择性融合模型和基于多种群进化动力学的冲突证据演化博弈融合方法，构建了图结构上的信度结构连接零和多矩阵博弈模型，提出了排他性冲突条件下的对抗环境博弈融合决策框架。（3）总结和整理了现有的大量证据不确定性度量模型，分析了单调性在证据结构不确定度量中的作用和意义，提出了一种基于信度区间的证据不确定性度量方法，并将该方法推广到了辨识框架元素非互斥的排他性冲突情况。（4）开展了不确定信息推理研究。定义了基于最大不确定性分配的证据理论信度结构逻辑否定计算方法，提出了基于最小承诺原则的证据网络推理模型，建立了面向概率不确定性和模糊不确定性环境的离散Z数推理网络。（5）研究了不确定信息最优决策与信息融合理论应用。建立了基于证据关联系数的最优概率转化模型，提出了面向高质量融合效果的最优数据融合方法，运用证据理论对认知决策中的分离效应给出了新的解释，并开展了不确定信息融合理论在多个领域的应用研究。本项目研究对不确定信息融合的理论研究和工程实践具有一定的参考价值和借鉴意义，相关成果在多传感器信息融合、目标识别等领域具有潜在应用价值。