基于长短期记忆神经网络的重大活动中鲜(冻)肉制品铅含量风险预测

目的 针对未来时间内重大活动举办地鲜(冻)肉制品铅含量风险预测的问题,建立基于LSTM的时间序列预测模型,对当地鲜(冻)肉制品铅含量进行风险评估与预测预警。方法 通过收集2011-2020年国家市场监督管理总局日常食品监督管理抽检数据,筛选出北京的鲜(冻)肉制品的抽检数据,构建数据集并进行预处理,按照7:2:1的比例划分训练集、验证集和测试集,基于Tensorflow平台构建4层LSTM模型并进行训练,基于重大活动举办前10天的鲜(冻)制品铅含量数据,对未来1天的鲜(冻)制品铅含量风险进行预测。结果 实验表明,经过50轮模型迭代训练,训练集和测试集Loss指标收敛至0.084,经过5次训练后的模型评估参数RMSE为0.192,R2_score为0.916,模型误差较小、准确度较高。结论 基于LSTM的鲜(冻)肉制品铅含量风险预测模型整体性能较好,可应用于重大活动举办地的食品风险预测,并精准指导监督抽检。     

基于监督抽检数据的肉类食品安全风险分析及预测

为通过数据挖掘预测食品安全风险、隐患和趋势,进行预警和快速反应,基于2015—2017年国家肉类食品监督抽检的18 378 批次样品数据,分析我国肉与肉制品主要安全现状与风险种类,并基于检测指标及属性运用BP（back propagation）神经网络方法构建以抽样省份、产品类型、产地、生产日期、年份、是否大型企业6大属性指标为输入层、包含2 个隐藏层、以是否合格为输出层的肉类食品安全神经网络预测模型。结果表明：经数据准备、模型生成、数据训练和验证及参数优化,得到的3 层BP神经网络预警模型总体百分比矫正为96.2%;对于合格样本,判定正确的概率为96.5%,错判概率为3.5%,预测模型具有较好的参考和应用价值。基于BP神经网络的食品安全预警方法能够对输入样本进行有效预测,为食品安全风险研判和风险预警提供技术支撑。     

长短期记忆神经网络在肉制品中铅含量风险预警的应用

本文基于2015～2018年国家肉制品监督抽检的5万批次检测数据,使用长短期记忆神经网络模型建立肉制品中铅含量的风险预警模型。参照国家食品检验标准并结合专家打分,构建肉制品中铅含量的6个食品安全风险等级;运用Softmax和汉宁窗对风险等级数据进行预处理,使用Tensorflow建立三层长短期记忆神经网络的时间序列风险预警模型,通过500轮模型训练实现对肉制品中铅含量的风险趋势预测。结果表明,长短期记忆神经网络对肉制品中铅含量的预测有较高的准确率,31个省份的平均误差为0.27,同实际检测风险基本匹配;模型稳定重现性较好,运行十次的平均误差为0.27。此模型可以实现对全国不同地域铅含量的趋势预警,为日常监督抽检和食品安全风险预警提供技术支撑。     

基于集成学习优化的肉制品安全风险等级预警分析

该研究依据2013~2017年肉制品抽检数据构造了5个安全风险等级,使用特征构造及独热编码进一步关联与肉制品安全相关的影响因素,构建极端梯度提升树算法（Extreme Gradient Boosting,XGBOOST）研究食品生产过程各类因素对于食品安全风险等级的影响程度,并使用多个指标评价模型。此外通过上采样解决样本不平衡问题、贝叶斯优化调节超参数,来提高模型性能及分类效果。相较于模型决策树（Decision Tree,DT）和随机森林（Random Forest,RF）,XGBOOST模型在肉制品安全风险等级分类中的表现效果最佳。研究结果表明,食品生产过程环节错综复杂,使用one-hot encoding处理后的模型能够有效判断出各类因素对于食品安全风险等级的影响程度,集成模型中RF的学习效果比较稳定,XGBOOST经过参数调节后准确率等指标得到有效的提升且优于RF。不同采样下XGBOOST的平均精确率均能达到89.14%,平均F1值为88.59%,说明XGBOOST在肉制品安全风险等级预警中适用性,为日常抽检提供技术指导。     

基于Stacking集成学习的服装网络直播销量预测

为了解决单一模型对直播销量预测效果不佳的情况，提出利用Stacking集成学习模型对4种单一机器学习模型进行融合。利用Spearman相关性分析和3种树模型的特征贡献度来进行特征选择，选用网格搜索以及贝叶斯优化算法进行模型参数选择。利用抖音直播李维斯品牌牛仔裤品类数据对算法进行实例验证。对比不同组合模型的MAE、MSE、RMSE和SMAPE值，实验证明：选用随机森林、支持向量回归、Xgboost为基学习器，线性回归为元学习器的两层Stacking集成学习模型对服装网络直播销量的预测效果优于单一机器学习模型以及其他组合模型，SMAPE的误差较单一模型最高下降6.97%，最低下降2.53%。     

基于食品抽检数据的风险预警智能研究模型构建分析

近年来,国内外由食品安全问题导致的公共健康事件屡见不鲜。食品安全问题涉及食品产业链的各个环节,收集和识别食品安全风险因子,挖掘造成食品安全的重要因素,是食品安全风险预警的重要途径,也是制定管理措施的前提条件。有效利用和分析食品安全信息,开展有效预警,已经成为食品安全监管面临的重要问题。本文首先分析了食品抽检数据特征和食品安全预警模型原则,在此基础上从风险分析、趋势分析、关联挖掘、全息画像和效能评价等视角构建了基于食品抽检数据的风险预警智能分析模型。该模型旨在为食品安全的监管、消费、生产等各个环节提供有力支撑,为推进基于食品抽检数据的风险预警智能研究提供参考。     

上海公布2014年第一季度肉制品 粮油 蔬果监督抽检结果

正2014年第一季度,上海市食品药品监督管理局对上海市食品生产经营企业开展肉、粮油、蔬果及其制品监督抽检。肉及肉制品共抽样872批次,共有38批次安全指标不合格,合格率为95.6%;粮食及其制品共抽样349批次,食用油及其制品117批次,     

2018年全国食品安全监督抽检情况分析及监管建议

目的 分析2018年全国食品安全总体情况。方法 汇总国家食品安全监督抽检结果, 按照不同维度进行分析。结果 2018年国家食品安全监督抽检结果累计公布样品32461批次, 其中合格样品31898批次, 合格率为98.27%; 不合格样品563批次, 约占1.73%。抽检发现的主要问题有违规使用农药兽药、微生物污染、质量指标不合格、食品添加剂使用不规范等。结论 2018年我国食品安全质量形势良好, 但仍存在食用农产品违规使用农药兽药、特殊膳食食品质量指标不合格、方便食品和肉制品微生物超标等食品安全风险, 监管部门、检验机构、生产经营企业应各尽其责、合力共治。     

基于深度置信网络-多类模糊支持向量机的粮食供应链危害物风险预警

近些年粮食供应链安全问题频发,为减少食源性风险威胁,风险预警正逐渐成为强化粮食食品安全体系的有力保障。但现有风险预警方法在面对多源异构非结构化食品数据时,存在预警准确率低、人工成本高等局限问题。本文在分析全国26个省份的大量抽检数据及关联信息基础上,建立了基于深度置信网络(deep belief network,DBN)-多类模糊支持向量机(multiclass fuzzy support vector machine,MFSVM)的风险分级预警模型,先对海量粮食供应链抽检数据进行嵌入编码和归一化处理,获得结构化食品数据;将其输入到DBN模型进行高维度特征提取,自适应地挖掘供应链中各危害因素间风险变化及内在关联概率,最后将高维特征输入到优化的MFSVM进行训练,实现供应链中各主要危害物风险分级预警。对比实验结果表明,DBN-MFSVM模型在粮食抽检数据上具有更好鲁棒性和泛化性,其准确率达到98.44%,运行时间85 s,可快速识别出粮食供应链中危害物风险程度和优先次序,为监管部门制定有针对性的抽检策略、确立优先监管领域和分配风险监管资源提供科学依据。     

基于集成学习模型的交通事故严重程度时空预测

为探究区域交通事故时空特征,精准预测事故严重程度,给交通运输主管部门提供决策支持,以英国交通事故统计数据作为研究基础,首先,将交通事故时空特征数据转化为网格化数据,并对空间特征进行二维卷积,利用时间特征合并二维卷积为三维卷积,解决网格冲突问题;其次,利用卷积神经网络（CNN）、长短期记忆人工神经网络（LSTM）模型的工作原理建立Stacking模型的基学习器和元学习器;最后,输出结果传入分类与回归树（CART）,构建完整的事故严重程度预测集成学习模型。研究结果表明,集成学习模型较单一模型对预测效果更优,其AUC比CNN、LSTM和Conv-LSTM单一模型预测分别提升0.02、0.04和0.01;最终决策树选择中,CART决策树比随机森林（RF）和梯度提升决策树（GBDT）预测效果更优;预测结果在时间纬度上,“严重事故”事件占比较实际低3.95%,在空间纬度上,预测热力区域在0.5～1区间范围内与实际接近。     

Authentication of meat and meat products

In recent years, interest in meat authenticity has increased. Many consumers are concerned about the meat they eat and accurate labelling is important to inform consumer choice. Authentication methods can be categorised into the areas where fraud is most likely to occur: meat origin, meat substitution, meat processing treatment and non-meat ingredient addition. Within each area the possibilities for fraud can be subcategorised as follows: meat origin—sex, meat cuts, breed, feed intake, slaughter age, wild versus farmed meat, organic versus conventional meat, and geographic origin; meat substitution—meat species, fat, and protein; meat processing treatment—irradiation, fresh versus thawed meat and meat preparation; non-meat ingredient addition—additives and water. Analytical methods used in authentication are as diverse as the authentication problems, and include a diverse range of equipment and techniques. This review is intended to provide an overview of the possible analytical methods available for meat and meat products authentication. In areas where no authentication methods have been published, possible strategies are suggested.     

Chemical safety of meat and meat products

Since the Second World War the consumer behaviour in developed countries changed drastically. Primarily there existed the demand for sufficient food after a period of starvation, afterwards the desire for higher quality was arising, whereas today most people ask for safe and healthy food with high quality. Therefore a united approach comprising consistent standards, sound science and robust controls is required to ensure consumers' health and to maintain consumers' confidence and satisfaction. Chemical analysis along the whole food chain downstream (tracking) from primary production to the consumer and upstream (tracing) from the consumer to primary production is an important prerequisite to ensure food safety and quality. In this frame the focus of the following paper is the “chemical safety of meat and meat products” taking into account inorganic as well as organic residues and contaminants, the use of nitrite in meat products, the incidence of veterinary drugs, as well as a Failure Mode and Effect Analysis (FMEA) system assessing (prioritizing) vulnerable food chain steps to decrease or eliminate vulnerability.     

Biogenic amines in meat and meat products

It has been recognized for some time that biogenic amines occur in a wide range of foods, among them meat and meat products. Meat is an important component of the diet in developed countries. The presence of these amines in food is of interest for two reasons: firstly, for toxicological reasons, in the sense that high levels of dietary biogenic amines can be toxic for certain consumers, and secondly, for their role as possible quality indicators. Based on these two premises, the present article offers a new analysis on aspects of toxicology and on the use of biogenic amines as a quality control method, as well as on their presence in meat and meat products. The article focuses particularly on factors affecting the production of biogenic amines, with reference to various parameters relating to microorganisms, meat raw materials, and processing conditions. A better understanding of the factors determining their formation (i.e., microorganisms, raw materials used, and technological processes applied) and their effects could lead to ways of controlling their production, limiting their presence in the end product, and hence, rendering them less toxic.     

肉与肉制品卫生监督管理

为切实把好肉与肉制品的屠宰、加工、贮存、运输、销售等环节的卫生关,确保肉食安全,必须要认真做好食品卫生监督管理工作,包括商业、农业、工商、卫生、外贸等各有关部门的各环节要共同努力,采取综合措施.     

素肉与素肉干产品分析

主要对素肉及素肉干的生产原理、产品优势进行了说明;并介绍了即食全素肉干、风味素肉块、素肉粒(酱类制品专用)、风味素肉类制品四种新型素肉类休闲食品.     

肉及肉制品中彩虹色斑点现象的调查分析

彩虹色斑点是近年来发现存在于肉及肉制品中的一种非正常颜色。对我国几家大型屠宰场的猪、牛分割肉及南京各大超市销售的部分肉制品进行了调查,结果表明:猪肉的各部位均无彩虹色斑点现象;牛肉半腱肌中彩虹色斑点的发生率为83%,其他部位从高到低分别为背最长肌(21%),腰大肌(7%),股二头肌(5%)。牛肉制品中帕斯雀牛肉、烤牛肉和酱牛肉彩虹色斑点最明显,猪肉制品中西式火腿片和咸肉最明显。     

N-nitrosamines and nitrites in meat and meat products]

Content of N-nitrosamines and nitrites in meat and meat products of Russian manufacture made on traditional and recently developed techniques was measured. The influence of conditions and duration of storage of meat products on changes in N-nitrosamines and nitrites was investigated. It is established that in observance of instruction of entering nitrites during manufacture their residual amounts in a ready product do not exceed hygienic levels accepted in Russian Federation. N-nitrosamines (N-nitrosodimethylamine and N-nitrosodiethylamine) were found in amounts from 0.1 up to 30 ppb. The least amount of N-nitrosamines was in canned meat (up to 0.7 ppb) and fresh meat (up to 0.4 ppb), greatest--in smoked products from pork. 8.6% tested samples of meat products contained N-nitrosamines in amounts exceeding the hygienic levels. The reasons of high contents of N-nitrosamines was infringement of conditions and terms of storage of ready meat products.