【作者】 黄东平；

【导师】 廖芹；

【作者基本信息】 华南理工大学 ， 概率论与数理统计， 2010， 硕士

【摘要】 贝叶斯网络是随机不确定性推理的有效工具,然而在实际应用中,经常遇见模糊随机事物的不确定性推理问题。如预测明天“好天气”的可能性,明天天气“状态”是一随机事件,但“天气是好状态”是模糊事件。如何应用贝叶斯网络解决这一类混合不确定性的知识推理问题已成为研究热点。本文提出了“基于遗传算法的模糊贝叶斯网络建立算法”:首先采用模糊数学的原理,定义了混合事件及其概率,第一次提出了条件模糊概率表的概念,有效解决了同时具备模糊性和随机性的变量的问题表示,通过聚类寻找高斯隶属度的参数,用遗传算法优化结构学习与参数学习,根据推理的分类误差,隶属度误差等反馈寻找最优的网络结构,同时通过修正隶属度函数的参数同步修正网络参数,特别地对模糊概率定义的参数α进行了优化确定,进而建立起模糊贝叶斯网络。针对当前食品安全控制领域出现的难以进行事前风险诊断、安全预警、事后科学界定责任等实际问题,根据食品安全控制领域数据的特点,本文提出了建立基于模糊贝叶斯网络的食品安全风险知识推理模型:通过对广州市质监局的溯源系统的数据研究,提取了与食品安全风险有关的指标,以统计方法定义指标的取值,获取了样本数据,并用“基于遗传算法的模糊贝叶斯网络建立算法”建立起食品安全控制知识的推理与诊断模型。应用结果表明,基于遗传算法的模糊贝叶斯网络虽然因模糊数学的处理增加了计算复杂度和运行时间,但是由于其采用模糊逻辑,能直接反映食品生产过程某一环节“风险出现高”的可能性的模糊随机问题的推理与诊断,且与一般贝叶斯网络比较发现,模糊贝叶斯网络有着更高的推理正确率。更多还原

【Abstract】 Bayesian network is an effective tool for uncertainty reasoning, but in practice, we often meet fuzzy stochastic uncertain reasoning problems. For example we want to forecast the possibility of tomorrow as "good weather" ,future weather "state" is a random event, but "the weather is good state" is a fuzzy event. How to use Bayesian network to solve this kind of mixed uncertainty knowledge inference has become a research hotspot. This paper proposes a novel approach of Fuzzy Bayesian Network Construction which is based on Genetic Algorithm : first of all, according to the principle of fuzzy mathematics, define of a mixed event and its probability, define the conditions fuzzy probability table for the first time, effectively solves the problem of the denote of the random variable, find the parameters of the Gaussian membership by clustering, optimize structure learning and parameter learning by Genetic Algorithm, find the optimal network structure according to the classification error and membership error of reasoning, and modify network parameters by modifying the parameters of membership functions, in particular the parameterαin the definition of the fuzzy probability is optimized to determine, in the end set up a fuzzy Bayesian network.View of the current food safety control is difficult to advance the field of risk diagnosis, early warning, scientific definition of responsibility after the real problems. After research the data of food safety control features in this field, this paper proposes reasoning model of control knowledge of food safety based on Fuzzy Bayesian Network: after the research of the data of Guangzhou Municipal Quality Supervision Bureau of the traceability system, extract the indicators related to food safety risk, define the value of this indicators by statistical method, obtain the sample data, and use the approach of Fuzzy Bayesian Network Construction which is based on Genetic Algorithm to establish reasoning model of control knowledge of food safety. The application results shows that the approach of Fuzzy Bayesian Network Construction which is based on Genetic Algorithm, increases the computational complexity and running time because of its use of fuzzy logic, but just using fuzzy logic can directly reflect the possibility of high food safe risk in the food production process in the reasoning and diagnosis. Compared with the general Bayesian network, fuzzy Bayesian network has a higher accuracy of inference.更多还原