人工神经网络法预测干黄酱挥发性成分保留值

Prediction of RI of the volatile flavor compounds in dry yellow soybean sauce by artificial neural network

基于分子连接性及邻接矩阵,计算69种干黄酱挥发性成分的分子连接性指数mχt,借助多元逐步回归法优化筛选了其中的结构参数0χ、5χ、3χc和5χpc,将其作为人工神经网络的输入层神经元,采用4∶8∶1的网络体系结构,以BP算法获得预测保留指数的神经网络模型,其相关系数R和标准偏差S分别为0.985和93.301.结果表明,保留指数与0χ、5χ、3χc、5χpc具有良好的非线性关系,BP神经网络方法预测的结果要优于多元回归方法的结果.

A pattern recognition model for the detection of food management was established by usingback -propagation （BP） algorithm in neural network. Based on molecular connectivity and adjacencymatrix, we calculated 69 volatile flavor compounds in dry yellow soybean sauce in this paper. By usingmultiple stepwise regression method, we screened and optimized the structure parameters xX, 5X, 3Xcand 5 p（p to establish a BP neural network model. The four structural parameters were used as the inputneurons of the artificial neural network, and a 4 ＂ 8 ： 1 network architecture was employed. A neuralnetwork model for predicting retention index （RI） was constructed with the back -propagationalgorithm. The correlation coefficient R and the standard errors was 0.985 and 93. 301, respectively,which showed a significantly nonlinear relationship between RI and the four structural parameters. Itcan be concluded that the prediction results of BP neural network are better than those of multipleregression methods.