基于BP神经网络的啤酒酒精度近红外光谱快速检测

Study on Content of Beer Alcohol with Near Infrared Spectroscopy Based on Back-Propagation Neural Network

以啤酒酒精度的快速测定为研究对象,采用误差反向传播算法(Back-Propagation,BP),结合主成分分析(PCA),构造了三层的神经网络结构,建立了PCA-BP神经网络模型,达到满意的预测精度。结果表明:使用BP神经网络方法后,验证集预测均方差、平均相对误差和回收率范围分别为0.114、1.131%和97.91%～104.59%,其效果优于PLS模型。

In order to fast determinate the beer alcohol content with NIR spectroscope, the Back-Propagation（BP） algorithm connected with Prince Component Analysis（PCA） was studied. The PCA-BP calibration model with three-lever Artificial Neural Network（ANN） structure was built to estimate the content of beer alcohol in predicted set. The results showed that the root mean square errors of prediction（RMSEP）, the mean relative errors of prediction（MREP）and the recovery rate were 0.114, 1.131% and 97.91%--104.59% respectively in PCA-BP model. Compared with the result of PLS model, the algorithm of BP was more effective and applicable, which could improve the ability of prediction and enforced the rebustness of calibration model.