基于LSTM网络的粮食干燥机水分预测与优化

Prediction and Optimization of Grain Dryer Outlet Moisture Content Based on LSTM

粮食干燥机的出机粮食水分预测有助于实现干燥机的智能化控制,从而可以减少干燥过程中的粮食损耗,对于粮食产后干燥环节有着重大意义。通过机器学习的方式进行预测,可以规避传统数学模型所存在的一系列缺陷。研究根据连续式谷物干燥机所提取的数据特征,提出了一种基于优化长短期记忆神经网络(LSTM)的稻谷出机水分预测模型。实验结果表明,出机水分与M_(in)、T_(o2)、T_(o3)、T_(d1)、T_(d2)、T_(d3)具有十分明显的相关性,通过设定不同的网络参数,确立了批尺寸50,学习率0.001,迭代次数50,时间步长50,神经元数100×100时效果最佳,此外还发现增加训练数据量,可以有效提高LSTM网络预测性能。将研究建立的LSTM模型与BP、ELMAN、NARX等算法以及普通LSTM网络(无dropout,单隐藏层)进行比较。结果发现,相较于其他网络模型,研究所采用的LSTM模型可以更好地预测稻谷出机水分,其平均绝对误差(MAE)、均方根误差(RMSE)和决定系数(R^(2))分别为0.12%、0.20%和0.94。研究所采用的优化LSTM模型具有较高预测精度,稳定性以及泛化性,可以为粮食干燥机的水分预测控制提供参考。

The outlet moisture content prediction of grain is helpful to realize the intelligent control of the dryer to reduce the grain loss in the drying process,having significance for the post-production drying of grain.Prediction through machine learning can avoid a series of drawbacks of traditional mathematical models.In this paper,in accordance with the data characteristics extracted from continuous grain dryer,a prediction model of paddy outlet moisture content based on optimized long short-term memory neural network(LSTM)was proposed.The results indicated that the outlet moisture content was significantly correlated with M_(in),T_(o2),T_(o3),T d1,T_(d2) and T_(d3).By comparing different network parameters,the optimal batch size of 50,learning rate of 0.001,iteration times of 50,time step of 50 and neural number of 100×100 were established.In addition,increasing the amount of training data could effectively improve the performance of LSTM network.The optimized LSTM prediction model was compared with BP,ELMAN,NARX as well as ordinary LSTM networks(no dropout,single hidden layer).The results indicated that,compared with other network models,the LSTM model used in this paper could better predict the outlet moisture content of rice,and its mean absolute error(MAE),root mean square error(RMSE)and coefficient of determination(R^(2))were 0.12%,0.20% and 0.94,respectively.The optimized LSTM model proposed had high prediction accuracy,stability and generalization performance,providing a reference for the moisture prediction and control of grain dryer.