摘要
为建立一种能够同时适用于多种新鲜水产品货架期的预测模型,采用反向传播(BP)神经网络模型、遗传算法(GA)优化的BP神经网络模型(GA-BP)、径向基函数(RBF)神经网络模型、极限学习机(ELM)神经网络模型和支持向量回归机(SVR)模型分别对金枪鱼、三文鱼、大菱鲆和鲷鱼的货架期进行预测,寻找最优的模型预测结果。首先通过试验获得4种水产品在0,4,10℃贮藏条件下的感官评分、菌落总数、挥发性盐基氮值、K值、pH值,构建训练样本和测试样本。经相关性分析,选择与水产品货架期相关性较高的感官评分、菌落总数、挥发性盐基氮值、K值作为模型的输入层单元,然后确定各模型的网络拓扑结构以及参数,进行模型的训练,最后使用训练好的5种模型对测试样本的货架期进行预测。结果表明:5种预测模型的预测精度排序为:SVR模型>RBF神经网络模型>GA-BP神经网络模型>ELM神经网络模型>BP神经网络模型,其中BP神经网络模型的预测精度最差,均方误差(MSE)为9.5127×10^(-4),平均绝对误差(MAE)为0.0197,平均绝对百分比误差(MAPE)为0.0825,R^(2)为0.9766;SVR模型的预测精度最优,预测误差均在12%以内,MSE为2.2971×10^(-4),MAE为0.0128,MAPE为0.0631,R^(2)为0.9944,能够很好地同时预测4种水产品在不同储藏温度下的货架期。本研究为水产品的品质控制提供一定的理论基础。
In order to establish a shelf-life prediction model that can be applied to a variety of fresh aquatic products at the same time.Using back propagation(BP)neural network model,genetic algorithm-back propagation(GA-BP)neural network model,radial basis function(RBF)neural network model,extreme learning machine(ELM)neural network model and support vector machine for regression(SVR)model to predict the remaining shelf life of four aquatic products respectively.First measure 4 indicators of tuna,salmon,turbot and sea bream at 0,4℃and 10℃,in order to construct training samples and test samples.Including sensory score,total number of colonies,TVB-N value,K value,p H value.After correlation analysis,the p H indicator with the lowest correlation was eliminated,and other indicators were selected as input indicators for each model.Then,determine the network topology and parameters of each model,and train each model.Finally,this study used each model trained to predict the remaining shelf life of the test sample.The result showed that the order of model prediction accuracy was:SVR prediction model>RBF neural network model>GA-BP neural network model>ELM neural network model>BP neural network model.The prediction accuracy of the BP neural network model was the worst,mean square error(MSE)was 9.5127×10^(-4),mean absolute error(MAE)was 0.0197,mean absolute percentage error(MAPE)was 0.0825,R^(2)was 0.9766.The prediction accuracy of the SVR prediction model was the best,its prediction error was within 12%,MSE=2.2971×10^(-4),MAE=0.0128,MAPE=0.0631,R^(2)=0.9944,it could well predict the remaining shelf life of 4 aquatic products.This study provides a theoretical basis for the quality control of aquatic products.
作者
崔方超
毛智超
李婷婷
励建荣
Cui Fangchao;Mao Zhichao;Li Tingting;Li Jianrong(College of Food Science and Technology,Bohai University,National&Local Joint Engineering Research Center of Storage,Processing and Safety Control Technology for Fresh Agricultural and Aquatic Products,China Light Industry Key Laboratory of Marine Fish Processing,Jinzhou 121013,Liaoning;College of Life Science,Dalian Minzu University,Dalian 116600,Liaoning)
出处
《中国食品学报》
EI
CAS
CSCD
北大核心
2023年第11期254-265,共12页
Journal of Chinese Institute Of Food Science and Technology
基金
国家自然科学基金重点项目(U20A2067)。
