【作者】 侯俊；

【导师】 潘泉； 程咏梅；

【作者基本信息】 西北工业大学 ， 控制理论与控制工程， 2006， 博士

【摘要】 与其他推理方法相比，证据推理具有符合人推理决策过程，可以对其进行合理的信息论解释，能够区分确定与不确定区间，且不需要先验信息，能够处理随机性和模糊性导致的不确定，而且计算直观，使用方便等优点。在对不确定信息处理与综合中得到了广泛的应用。本论文在对前人的工作整理完善的基础上，开展D-S理论组合方法、评价体系以及典型应用等方面的研究，具体贡献如下：1．在分析D-S自身特点以及与其他方法相对比的基础上，给出了解决冲突的两种新的组合方法——吸收法和加权分配冲突法，以及最新发展的DSmT方法的特点，并通过算例分析了上述方法解决冲突的不同思路和性能；2．在借鉴、整理和发展前人工作的基础上，给出了证据推理的性能评价体系，包括基本性质、特有性质和工程可用性等三大类9条准则。提出了抗高冲突性、便利性、计算复杂性、无条件极化性和条件极化性，完善了同一性的定义，聚焦性的分析。并依次对9条准则进行了定义和相应的证明，给出了12种典型方法基于9条准则的基本性能比较；3．设计的大量的算例，对D-S、DSmT、Smets、孙全、向阳、吸收法、邢清华、Murphy、邓勇、Yager、Dubois和Toshiyuki等12种典型的方法，以9条准则的依据，逐一进行了计算、分析、对比，分析结果对这些方法的实际应用给出了有价值的指导建议；4．结合图像目标识别典型应用，研究了证据推理在工程应用中的一般流程，分析比较了三种基本置信指派函数的构造方法，最后分别采用指数函数法和神经网络法给出了证据推理的序列图像目标识别算法，并对其中不同组合和决策方法的识别结果进行了仿真和分析，效果良好。更多还原

【Abstract】 Compared with other reasoning theories, evidence reasoning theory makesdecisions according to human’s way, and can be explained reasonably by usinginfonnation theory. Certain and uncertain zones can be divided with no priorinformation and randomness, and uncertainty from imprecise sources can be managedintuitively. Evidence reasoning is widely used in dealing with the uncertaininformation. In this dissertation, the combination rules, performance indexes andsome applications are discussed in detail. The main contributions are as follows:1. After analyzing the D-S theory and the DSmT developed recently, two newcombination rules, "weigh distribution" and "absorptive method", are given tosolve the conflict problem. Substantive simulations are given to analyze thedifferent ideas and performances of these two rules.2. A performance index scheme of evidence reasoning rules is put forward, including3 parts and 9 items of evaluation criteria. We put forward Anti-high-conflict,complexity, convenience, polarization and conditional polarization, and theevaluation criteria of identity and focusing are also perfected. The performancesof the 12 combination rules are analyzed with the 9 evaluation criteria.3. An amount of simulations are given to evaluate the 12 rules including D-S, DSmT,Smets, Sunquan, Xiangyang, Absorptive method, Xingqinghua, Murphy,Dengyong, Yager, Dubois and Toshiyuki. Using the 9 criteria, we analyze theperformance of the 12 methods in detail. Some suggestions are given from thesimulation results.4. With the application in the recognition of sequential images, the normal process ofevidence reasoning’s application in engineering are studied. And the 3 methods ofconstructing basic belief assignment are compared, the recognition of sequentialimages is given based on BP neural network and exponential function. Therecognition results of different methods are then analyzed更多还原