BP神经网络在粮食霉变预测中的应用研究

霉变是造成粮食损失的重要原因,为了降低损失,将危害控制在萌芽状态,提前预测预警意义重大。本研究利用MATLAB的神经网络工具箱建立了预测粮食霉变的BP神经网络,给出了稻谷在给定含水率、温度、储藏时间的条件下是否会发生霉变的预测模型。同时,通过合理选择训练样本的数目,探究训练样本数量对网络精度的影响,并通过华北地区实仓数据验证由实验数据得到的BP神经网络在实际应用中所能达到的准确程度。经过验证,对于实验数据,训练样本数目大于400时,神经网络预测正确率可以达到94.3%;样本数越大,正确率越高。随机选择2 500个实验室样本数据进行训练得到的神经网路预测模型,对剩余样本预测准确率达到98%,对于实仓检测数据,正确率可以达到82.1%。

Application of BP Neural Network in Prediction of Grain Mildew

Mildew is a main cause of grain loss during storage. In order to reduce loss and control the damage in its infancy stage, it is of great significance to predict risks in advance .The BP neural network was used to predict grain mildew via the neural network toolbox of MATLAB. Taking the experimental data of paddy as an example, a BP neural network forecast model was built to predict whether the stored paddy would be mildewed under the given condition of water content, temperature and storage time. At the same time, by selecting the quantity of training samples reasonably, the influence of training sample size on the network accuracy was explored. In addition, when applied in practical, the accuracy of the BP neural network obtained from the experimental data was tested by the actual storage data from one granary in North China. As for experimental data, the prediction accuracy rate of the BP neural network will increase as the augment of sample size ,which can reach up to 94.3 % when the training samples is more than 400. As for the actual storage data, the neural network prediction accuracy rate can achieve to 82.1 %.