基于LSTM的粮堆“热皮”厚度预测模型构建  

Construction of“Hot Skin”Thickness Prediction Model of Grain Pile Based on LSTM

作　　者：赵旭 邢思敏 杨晓鑫[2] ZHAO Xu;XING Simin;YANG Xiaoxin(Liaoning Grain Science Research Institute,Shenyang 110032,China;Tieling Grain Science Research Institute,Tieling 112000,China)

机构地区：[1]辽宁省粮食科学研究所,辽宁沈阳110032 [2]铁岭市粮食科学研究所,辽宁铁岭112000

出　　处：《现代食品》2024年第21期6-9,13,共5页Modern Food

摘　　要：本文根据粮仓密集阵列式测温系统所提取的温度数据,构建了一种基于长短期记忆网络(Long Short-Term Memory,LSTM)的粮堆“热皮”厚度预测模型。结果表明,该模型可以较好地预测粮堆“热皮”厚度。东侧列间、西侧列间和层间3种基于LSTM的“热皮”厚度预测模型拟合效果均较好,3种模型R2均高于0.889,预测误差较低。其中层间“热皮”预测模型的预测效果最好,训练集和验证集R2分别为0.936和0.921,平均绝对预测误差分别为0.058 m和0.049 m。表明基于LSTM的粮堆“热皮”厚度预测模型的预测性能具有一定优势,能够实现对储粮过程中粮仓“热皮”厚度的准确预测,为粮堆“热皮”的定量分析提供了新的思路,有助于仓储管理人员实现辅助决策,为保障粮食安全储藏提供理论借鉴。In this paper,a prediction model of“hot skin”thickness of grain pile based on long short-term memory(LSTM)was proposed based on the temperature data extracted from the dense array temperature measurement system of grain silo.The results show that LSTM method can be used to predict the thickness of“hot skin”of grain pile.The three LSTM“hot skin”thickness prediction models proposed in this study,inter row east,inter row west and interlayer,all have good fitting effect,and the R2 of the three models is higher than 0.889,with low prediction error.Among them,the inter-layer“hot skin”prediction model has the best prediction effect,and the training set and validation set R2 are 0.936 and 0.921,respectively.The mean absolute prediction errors were 0.058 m and 0.049 m,respectively.The research shows that the prediction performance of the“hot skin”thickness prediction model based on LSTM has certain advantages,which can realize the accurate prediction of the“hot skin”thickness of grain storage,provide a new idea for the quantitative analysis of“hot skin”of grain storage,help warehouse managers to realize auxiliary decision-making,and provide theoretical reference for ensuring food safety storage.

关 键 词：长短期记忆网络(LSTM) “热皮”厚度 预测模型 

分 类 号：TU249.2[建筑科学—建筑设计及理论]