基于极限学习机的低温液态乳制品中金黄色葡萄球菌污染程度预测方法  

Prediction method of Staphylococcus aureus contamination in low temperature liquid dairy products based on the extreme learning machine

作　　者：王建新[1] 张懿文 连梦雪 王晔茹 WANG Jianxin;ZHANG Yiwen;LIAN Mengxue;WANG Yeru(School of Information,Beijing Forestry University,Beijing 100083,China;China National Center for Food Safety Risk Assessment,Beijing 100021,China)

机构地区：[1]北京林业大学信息学院,北京100083 [2]国家食品安全风险评估中心,北京100021

出　　处：《生物加工过程》2021年第6期693-698,共6页Chinese Journal of Bioprocess Engineering

基　　金：国家重点研发计划(2019YFC1605903);国家食品安全风险评估中心高层次人才队伍建设523项目。

摘　　要：因检测能力和检测成本的限制,金黄色葡萄球菌在液态乳制品生产加工、运输储存以及消费等多个环节的繁殖情况数据一般并不完整,相关的环境条件数据也可能存在一个或多个环节的缺失。极限学习机是神经网络的一种,预测精度通常比较高,研究者可以利用极限学习机建立多原因变量和结果变量之间的定量关联关系,从而预测给定条件下金黄色葡萄球菌的污染程度,并逆向推断一个或多个环境条件变量的值。微生物繁殖一般总体上符合指数增长函数,但极限学习机不能很好地拟合该函数,因此,本文中,笔者将指数增长的繁殖机制和极限学习机相结合,先计算指数增长的值,作为极限学习机的输入之一,再启用极限学习机,从而综合提升预测准确率。以储存环节为例,平均误差从14.77%降至2.85%。通过本算法的初筛,检测人员可以快速定位最有可能发生污染风险的环节,采取重点检测而非全方位检测的策略,从而节约检测成本,提高检测成效;检测人员也可以在严重污染发生后,定位引发污染的环境条件,从而发现问题根源。Due to the limitation of detecting capacity and detecting cost,the data on the reproduction of Staphylococcus aureus in liquid dairy products was generally incomplete during the producing,processing,transportation,storage,consumption,and other links,and the relevant environmental condition data was also missing in one or more links.The extreme learning machine is a kind of neural network with high prediction accuracy,it could be used to establish a quantitative relationship between multiple cause variables and the outcome variable,to predict the pollution degree of S.aureus under given conditions,and also to infer the value of one or more environmental condition variables.Microbial reproduction generally conforms to the exponential growth function,but the extreme learning machine does not fit this function well.Therefore,we combined the exponential growth mechanism with the extreme learning machine,by means that the exponential growth value could be calculated first and then input into the extreme learning machine.Such combination improved the accuracy of prediction,and the average error was reduced from 14.77%to 2.85%during storage.Through the preliminary screening of this algorithm,it is possible to quickly locate the link that is most likely to have the risk of infection,and to adopt a strategy of focused detection rather than comprehensive detection,thereby saving detection costs and improving detection effectiveness.It is also possible to locate the environmental conditions that caused the pollution after severe pollution occurs,so as to find the source of the problem.

关 键 词：低温液态乳制品 食物中毒 金黄色葡萄球菌 食源性致病菌 极限学习机 冷链运输 风险评估 食品安全 

分 类 号：TS201.6[轻工技术与工程—食品科学] TP399[轻工技术与工程—食品科学与工程]