大曲酸度值的快速预测模型及方法研究

该研究利用在大曲发酵周期（1～28 d）内采集的大曲内部温度和水分数据，并结合电位滴定法测定的大曲酸度值数据，建立发酵过程中大曲酸度值快速检测的数学模型。首先对原始数据进行异常样本剔除，划分样本集，再分别运用偏最小二乘回归（PLSR）、支持向量回归机（SVR）和反向传播神经网络（BPNN）建立大曲内部温度、水分与酸度值之间相关性预测模型，最后运用决定系数（R2）与均方根误差（RMSE）对训练集、测试集进行效果评价，探索最佳预测方法。结果表明，支持向量回归机（SVR）建立的酸度值预测模型最好，测试集上的R2为0.874 5，RMSE为0.104 8。经外部验证后，该模型酸度的预测值与实际值的相对误差为1.6%～11.1%，可以用于实际大曲酸度值预测，为智能调控大曲发酵温度、湿度等环境参数提供理论支撑和依据。

Rapid prediction model and method of Daqu acidity value

In this study, a mathematical model for the rapid detection of Daqu acidity value during fermentation was established by using Daqu internal temperature and moisture data collected during the fermentation period (1-28 d), combined with Daqu acidity data determined by potentiometric titration. Firstly, the abnormal samples of original data were removed, the sample sets were divided, and then the prediction model of the correlation between internal temperature, moisture and acidity in Daqu was established by partial least square regression (PLSR), support vector regression machine (SVR) and Back propagation neural network (BPNN). Finally, the effect of the training set and test set was evaluated using the determination coefficient (R2) and root mean square error (RMSE), to explore the optimal prediction method. The results showed that the optimal acidity value prediction model was established by support vector regression machine (SVR), the R2 on the test set was 0.874 5, and the RMSE was 0.104 8. After external verification, the relative error between the predicted and actual values of the model was 1.6%-11.1%, which could be used to predict the actual Daqu acidity value, and provided theoretical support and basis for intelligent adjustment of temperature, humidity and other environmental parameters of Daqu fermentation.