因果图的两种不精确推理探索

发布时间：2018-01-07 09:09

本文关键词：因果图的两种不精确推理探索　出处：《重庆师范大学》2017年硕士论文　论文类型：学位论文

【摘要】：不确定性问题作为人工智能最核心的研究任务,将不确定性问题的求解方法大致分为两类:一类是基于概率的方法,一类是基于非概率的方法。因果图推理是一种概率的方法,因果图以图形的方式表达复杂系统的因果关系。由于因果图推理中存在不确定性问题,为了更形象的将系统的不确定性进行表达,本文主要是研究因果图的两种不精确推理:将概率矩阵近似处理转化为精确概率值;将求基本事件精确概率值扩充为求区间概率。主要内容如下:(1)把一个复杂系统用因果图知识表达,进行系统故障诊断时,用节点事件表示故障源,用有向边表示因果关系。由于子变量的赋值状态数不同,将因果图分为单值因果图和多值因果图。将传统因果图推理用于多值因果图中会出现概率不归一的现象,因此提出一种多值因果图的不精确推理。该推理方法是根据因果影响程度找到连接事件概率值,而该概率值是在引入了事件缺省状态,并假设事件各状态之间互斥的情况下求得的。根据概率矩阵中事件各状态发生的概率找到其发生的可能性大小,再进行概率分配,使概率满足归一性,将多值因果图转化为单值因果图。(2)因果图作为一种基于概率的知识表达方法,是对基本事件发生概率已知时进行推导计算,而实际应用中,由于数据的误差、缺失,专家的主观偏见等很难获得精确概率值,针对此情况本文提出将精确值扩充为区间数。根据Dempster-Shafer证据理论(简称D-S理论),将专家知识或者系统数据进行融合,通过计算得到似然函数Pls(Plausibility Function)和信任函数Bel(Belief Function),将其分别作为概率区间的上下界,形象表达系统的模糊性和不确定性,同时还降低了获取精确概率值的难度。
[Abstract]:As the core research task of artificial intelligence, uncertainty problem is divided into two categories: one is probabilistic method. One is based on non-probabilistic method. Causal graph reasoning is a probabilistic method. Causality diagrams express the causality of complex systems graphically. There is uncertainty in causal graph reasoning. In order to express the uncertainty of the system more vividly, this paper mainly studies two kinds of inexact reasoning of causality diagram: the approximate processing of probabilistic matrix is transformed into exact probabilistic value; The main contents are as follows: (1) A complex system is represented by causality diagram knowledge, and the node event is used to represent the fault source in system fault diagnosis. Use directed edges to represent causality. Due to the number of assigned states of subvariables. The causality diagram is divided into single value causality diagram and multivalued causality diagram. The phenomenon of probability disunity will occur when traditional causality diagram reasoning is used in multivalued causality diagram. Therefore, an inexact reasoning method of multi-valued causality graph is proposed. The method is to find the probability value of connected event according to the degree of causality influence, and the probability value is to introduce the default state of event. According to the probability of the occurrence of each state in the probability matrix, the probability of occurrence is found, and then the probability distribution is carried out to make the probability meet the normalization. As a method of knowledge representation based on probability, the multi-valued causality diagram is transformed into a single-valued causality diagram. It is a method to derive and calculate the probability of occurrence of a basic event when the probability is known, but it is applied in practice. It is difficult to obtain the accurate probability value because of the error of the data, the lack of the data, the subjective bias of the expert and so on. According to the Dempster-Shafer evidence theory (D-S theory for short), the expert knowledge or system data are fused. The likelihood function (Pls(Plausibility function) and the trust function (Bel(Belief function) are obtained. It is regarded as the upper and lower bound of the probability interval to express the fuzziness and uncertainty of the system, and at the same time, the difficulty of obtaining the exact probability value is reduced.
【学位授予单位】：重庆师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18;O211

【参考文献】