基于贝叶斯网络的酱卤肉制品安全预警研究

Research on safety early warning of sauced meat products based on Bayesian network

中文关键词:  贝叶斯网络  酱卤肉制品  风险预警

英文关键词:Bayesian network  sauced meat products  risk warning

中文摘要:

      目的 简要介绍贝叶斯网络基本概念和算法, 建立贝叶斯网络预警模型对酱卤肉制品进行安全预警。方法 利用领域专家知识对可能影响酱卤肉食品安全的重金属污染物、兽药残留、食品添加剂、微生物、非食用物质5个方面的因素进行分析, 划分食品安全状况等级与预警指标; 运用最大似然估计算法和贝叶斯网络建立酱卤肉制品安全预警模型结构, 使用VS code软件进行仿真实验, 对酱卤肉制品安全的风险程度进行分类预测。结果 贝叶斯网络模型得到的酱卤肉制品总体情况与实际的数据统计值的误差在0.005~0.006的范围内, 属于合理误差范围。BP神经网络和贝叶斯网络的平均准确率分别为0.85和0.99。在此次实验中, 贝叶斯网络的准确率较高。结论 在小样本情况下, 贝叶斯网络在酱卤肉制品安全风险预警中具有较高的准确率, 是一种能准确、稳定实现酱卤肉制品安全风险预警的算法, 且方法优于BP神经网络。

英文摘要:

      Objective To briefly introduce the basic concept and algorithm of Bayesian network, and establish the early warning model of Bayesian network for the safety early warning of sauced meat products. Methods Based on the knowledge of experts, the five factors that may affect the safety of sauced meat, including heavy metal pollutants, veterinary drug residues, food additives, microorganisms and nonfood substances were analyzed, and classified the food safety status level and early warning indicators, and established the safety early warning model structure of sauced meat products by using maximum likelihood estimation algorithm and Bayesian network. VS code software was used to carry on the simulation experiment to classify and predict the food safety risk degree of the sauce marinated meat. Results The error between the overall situation of the stewed meat products obtained by the Bayesian network model and the actual data statistics was within the range of 0.005 to 0.006, which was a reasonable error range. The average accuracy rates of BP neural network and Bayesian network were 0.85 and 0.99 respectively. In this experiment, the accuracy of the Bayesian network was relatively high. Conclusion In the case of small samples, Bayesian network has a high accuracy rate in the safety risk early warning of sauced meat products. It is an accurate and stable algorithm to realize the safety risk early warning of sauced meat products, and the method is better than BP neural network.