多值因果图的推理算法研究

针对多值因果图存在的两个困难：（1）不严格满足概率论；（2）将其用于实际问题时，推理结果可能出现错误，提出了一种基于因果影响可能性分配的推理算法，该算法将多值因果图的推量分成3个阶段，首先对多值因果图进行补充定义，使多值因果图能够兼容单值因果图；接着将多值因果图转化为单值因果图进行概率计算，最后对多值因果图进行可能性计算，将单值因果图计算得到的概率按多值因果图计算得到的可能性进行分配，以核电站二回路系统中蒸汽发生器故障诊断因果图为例，展示了该算法推理计算的全过程，实例表明，该算法能够有效地克服多值因果图存在的困难，其推理过程严谨，计算结果符合实际情况，而前面提出的推理算法基础上，针对其不能处理模型情况的局限性，提出了一种模糊推理算法，该算法对多值因果图进行了模糊扩展定义，在读数变量和事件变量之间建立了用于表达模糊知识的模糊对应关系，在事件变量上定义了一个等价的虚拟模糊状态，使读数变量取值对应一个模糊状态，把读数和模糊推理转化为对应模糊状态的非模糊推理，通过本文的工作，目前因果图已发展成了一个能够处理离散变量和连续变量的混合因果图模型。

Reasoning Algorithm in Multi-Value Causality Diagram

The multi value causality diagram developed on the belief network does not satisfy probability theory rigorous, and the inference result may be error when it is used in practice. In order to overcome these difficulties, this paper presents a reasoning algorithm based on possibility allocation. The reasoning process is separates into 3 stages. Firstly, the multi value causality diagram is supplementally defined. It is compatible with a single value causality diagram. Secondly, it transforms a multi value causality diagram to a single value causality diagram that is used to compute the probability; Thirdly, it allocate the probability to every state according to its possibility value that is computed in multi value causality diagram. An example about fault diagnosis of a steam generator in the nuclear power plant demonstrates that this algorithm could overcome efficiently the difficulties in multi value diagram, the reasoning process is rigorous, and the result coincides with the reality. Aimed to the former algorithm's deficiency, e.g. it cannot deal with the fuzzy case, a fuzzy reasoning algorithm is presented in this paper, which extended the definition of the multi value causality diagram with fuzzy, builts the fuzzy mapping relation between the event variable and the reader variable that to represent the fuzzy knowledge, and defines a suppositional equivalent fuzzy state of event variable that maps the reader variable to a fuzzy state and transforms the fuzzy reasoning of a reader variable to the non fuzzy reasoning of a fuzzy state. Now, the causality diagram has become a hybrid probability knowledge representation and reasoning model, which can deal with discrete and continuous variables and represent the fuzzy knowledge under uncertainty.