基于BP神经网络的葡萄气体射流冲击干燥含水率预测

BP Neural Network Modeling to Predict Moisture Content of Grapes after Air Impingement Drying

摘要:

为实现葡萄气体射流冲击干燥过程中含水率的预测，本文探究了不同烫漂前处理时间（30、60、90和120 s）和干燥温度（55、60、65和70 ℃）对葡萄干燥时间和干燥速率的影响，建立了输入层为烫漂时间、干燥温度和干燥时间，隐藏层节点数为7，输出层为葡萄含水率，结构为“3-7-1”的BP神经网络模型。结果表明：烫漂预处理时间和干燥温度均对葡萄干燥速率有影响，增加烫漂时间和提高干燥温度能够有效的缩短葡萄干燥时间，提高干燥效率。采用Levenberg-Marquardt (LM)算法为训练函数，选择tansig-purelin为网络传递函数，经过有限次训练得到的BP神经网络模型，其预测值与实测值之间的决定系数R2为0.9915，均方根误差RMSE为0.03376，预测快速且准确，为葡萄在干燥过程中的含水率在线预测提供理论依据和技术支持。

Abstract:

The effect of different blanching pretreatment times (30, 60, 90 and 120 s) and drying temperatures (55 ℃, 60 ℃, 65 ℃, and 70 ℃) on the drying rate and drying time of grapes was investigated, and a BP neural network model was established. This network used an architecture of ‘3-7-1’, which includes input layers of blanching time, drying temperature and drying time, seven hidden layer variables, and a single output layer of moisture content. The results provided a mechanism to predict the moisture content of grapes under air impingement drying. The results demonstrated that blanching pretreatment time and drying temperature showed a significant impact on drying rate and longer blanching time and higher drying temperature resulted in a higher drying rate. The BP neural network model was configured for finite iteration calculation with Levenberg-Marquardt (LM) algorithm as the training function and tansig-purelin as the network transfer function. The correlation coefficient (R2) and root mean squared error (RMSE) between the predicted and measured values were 0.9915 and 0.03376, respectively.