节点文献

食品安全风险过程控制的贝叶斯统计与知识发现

Bayesian Statistics and Knowledge Discovery Based on the Food Safety Risk’s Process Control

【作者】 颜丽

【导师】 廖芹;

【作者基本信息】 华南理工大学 , 概率论与数理统计, 2011, 硕士

【摘要】 食品安全已成为世界范围内广泛关注的问题,理想的食品质量控制模式是“从农田到餐桌”的全过程质量控制。食品安全追溯系统提供了“从农田到餐桌”的追溯模式,建立了食品安全信息数据库。应用食品安全追溯系统,一旦发现问题,能够通过溯源进行有效的控制和召回,从源头上保障消费者的合法权益。但这依然是一种事后控制,并不能进行食品安全风险预警与控制。本文利用食品安全追溯系统各管理过程的抽样、检验以及有关的监控数据,提取影响食品安全风险因素,获取因素变化取值,建立基于贝叶斯网络的知识表示与知识推理模型,以达到食品安全风险的预测、预警与控制。由于食品链中涉及的食品原料、加工、包装、储藏、运输、销售、消费等环节都会对最终食品的安全性造成影响,本文选取各环节的安全状况作为节点变量,变量的安全状况依赖于不同的危害后果严重程度与可能性。通过由危害后果严重程度与可能性定义的食品安全潜在风险,赋予各环节随机变量的取值,获取相关样本,在此基础上,建立贝叶斯网络结构,并依据网络中各节点条件概率分布的先验信息及获取的样本信息,应用贝叶斯估计方法进行网络参数的估计与更新学习,在食品安全各环节潜在风险程度的影响下,实现原因推理原因、原因推理结果、结果诊断原因的风险知识推理。进一步地,对已建立的食品安全风险推理的贝叶斯网络,针对贝叶斯网络知识推理的计算过程复杂与耗时问题,进行简化知识推理研究,在软件Matlab辅助推理下,论证了具有更高效率的简化知识推理的实现过程。本文建立的基于贝叶斯网络食品安全风险知识推理模型,可以通过食品安全追溯系统各环节的原始数据,预测相应环节的潜在风险,通过环节与环节的潜在风险,实现任意食品生产环节潜在风险与食品安全风险的推理与诊断,当食品安全风险推理结果达到某阈值时,进行预警并对影响环节与原因进行逆推理,达到控制或避免风险的出现。经数据检验表明:模型的正确推理识别率达到93%。

【Abstract】 Food Safety has become a world-wide concerned problem. The ideal mode of food quality control is the ’from farm to fork’ whole process quality control. Food Safety Traceability System provided the ’from farm to fork’ traceability model and established a food safety information database. If we apply the Food Safety Traceability System to food quality control, we can control effectively and recall the trouble as soon as we find it,so as to protect the legitimate rights and interests of consumers from the source. But this mode is still a risk control after we find the trouble,it can not warn and control the risk of food safety.In this paper, we select the factors which affect the risk of food safety and assess which value by extracting relational sampling, testing and monitoring data from the Food Safety Traceability System. In order to achieve the prediction warning and control of the risk of the Food Safety, we establish knowledge representation and reasoning model based on Bayesian network.The sectors involved in food chain of food ingredients, processing, packing, storage, transportation, distribution, consumption and so on which will impact on the final food safety.We select the security situation of all sectors as the node variable, and the variable’s security situation depends on the likelihood of harm and severity of harmful consequences. In order to assess values of variables and obtain the relevant samples, we define potential risk of food safety by likelihood of harm and severity of harmful consequences. After establisted the structure of Bayesian network and got samles for study, we can according conditional probability distribution’s priori information of each node and sample information to update learning network parameters applying Bayesian estimation method.Under the influence of the potential risk level involved in food safety’s all sectors, we achieve the reasoning of risk knowledge which contains cause to reason, cause to cause and result to cause. Further, aiming at reduce the calculation’s complexity and time-consuming of the Bayesian network knowledge reasoning, we research on the simplified knowledge reasoning of Bayesian network. Under the help of Matlab, we demonstrate the implementation process of more efficient simplified knowledge reasoning.We can predict relevant sector’s potential risk through origin data of this sectors in Food Safety Traceability System. Through the potential risk between sector and sector, the model establist in this paper can reason and diagnose any potential risk in food production process and any risk of food safety. Once the result of food safety risk reasoning reaching the threshold, the model will be warning and doing inverse reference to the involved sectors and the cause for controlling or avoiding the risk. The test depending on data show that the correct rate of model’s reasoning reach 93%.

  • 【分类号】O212.8;F203
  • 【被引频次】1
  • 【下载频次】411
  • 攻读期成果
节点文献中: 

本文链接的文献网络图示:

本文的引文网络