基于BP神经网络的仓储烟草霉变预测

Mildew Prediction of Warehousing Tobacco Based on BP Neural Network

烟草霉变预测尚没有有效的方法。为实时预测仓储烟草的霉变程度,选取仓储环境的温湿度和烟草的自身含水量参数作为神经网络的输入层,烟草霉变度作为输出层,建立BP神经网络烟草霉变预测模型。选取78组实测数据作为训练样本对预测模型进行训练,得出了神经网络的阈值和权值。利用14组预测样本针对该预测模型进行了仿真,并进行了线性回归分析。结果表明,建立的烟草霉变预测模型具有较高的预测精度,预测值和实际值的偏差在[-0.028,0.033]之间,相对误差绝对值的平均值为0.001 9。最后,在基于嵌入式ARM+Linux+Web的某公司烟草仓库智能监测系统中,实现了烟草霉变实时预测功能,取得了较好的效果。

There is still no effective solution to mildew prediction.In order to forecast the moldy degree of warehousing tobacco in real time,a tobacco mildew prediction model is established by BP neural network,in which temperature,humidity,and tobacco moisture are selected as the network input and the mildew degrees of the tobacco are extracted as the network output.Firstly,the measured data of 78 sets are used as training samples to obtain the threshold value and the weight value of BP neural network.The data of 14 samples are simulated with linear regression analysis to validate the proposed model.The results show that the deviation range between the predictive value and the actual value is[-0.028,0.033],and the relative error’s absolute mean is 0.0019.Finally,the tobacco mildew real-time prediction is proved to have higher prediction precision in the tobacco warehouse intelligent monitoring system based on embedded ARM＋Linux＋Web.