基于LSTM和IGA-BP的酒精度预测模型  

Research on alcohol prediction model based on LSTM and IGA-BP

作　　者：张建华[1] 商建伟 王唱 赵岩[1] 李克祥 李祥利 ZHANG Jian-hua;SHANG Jian-wei;WANG Chang;ZHAO Yan;LI Ke-xiang;LI Xiang-li(School of Mechanical Engineering,Hebei University of Technology,Tianjin 300401,China)

机构地区：[1]河北工业大学机械工程学院,天津300401

出　　处：《食品与机械》2022年第5期71-77,共7页Food and Machinery

基　　金：国家重点研发计划项目(编号:2018YFB1306500);河北省重点研发计划项目(编号:21311801D);天津市智能制造重大专项(编号:19ZXZNGX00100);河北省高层次人才资助项目—博士后科研项目择优资助(编号:B2020003020)。

摘　　要：目的:解决目前分段摘酒过程依赖人工“看花摘酒”,酒精度检测不准确的问题。方法:设计搭建基于酒精度建模的分段摘酒系统,研究采集音叉在不同模态不同浓度酒精溶液内的音叉频率值、音叉内置温度值,酒精溶液温度值和动态条件下泵转速值,基于最小均方算法(LMS)和长短期记忆网络(LSTM)实现音叉频率自适应滤波和动态补偿,基于改进遗传算法优化BP神经网络(IGA-BP)建立酒精度预测模型。结果:模型在迭代次数和预测精度上优于传统遗传算法优化BP神经网络和BP神经网络建立的酒精度预测模型,酒精度平均预测误差为0.381。结论:基于改进遗传算法优化BP神经网络(IGA-BP)建立酒精度数预测模型具有合理性。Objective:In order to solve the problem of inaccurate detection of alcohol accuracy due to manual"flower picking"in the segmented liquor picking process.Methods:Designed and built a segmented liquor taking system based on alcohol accuracy modeling.The research collected tuning fork frequency values,tuning fork built-in temperature values,alcohol solution temperature values and pump speed values under dynamic conditions in different modes of alcohol solutions with different concentrations,implemented adaptive tuning fork frequency filtering and dynamic compensation based on least mean square algorithm(LMS)and long short-term memory network(LSTM),and built a liquor accuracy prediction model based on improved genetic algorithm optimized BP neural network(IGA-BP).Results:The model outperformed the traditional genetic algorithm optimized BP neural network and BP neural network in terms of the number of iterations and prediction accuracy,and the average prediction error of the alcoholic beverages was 0.381.Conclusion:which verifies the reasonableness of the model.In order to solve the limitations of the current manual"flower picking",a method is proposed to improve the accuracy of alcohol detection in the segmented liquor taking process.

关 键 词：摘酒 酒精度 长短期记忆网络 遗传算法 BP神经网络 

分 类 号：TS262.3[轻工技术与工程—发酵工程] TP183[轻工技术与工程—食品科学与工程]