BP 神经网络在大豆油酸价近红外光谱检测中的应用

对于大豆四级油，采用BP 神经网络对其近红外光谱数据建模，对系统的结构及参数选取进行了分析，对样本训练集的设计和网络输入端的主因子方面进行了处理。对于其他的多变量建模方法，分析了其对近红外光谱有用信息的提取作用。结果显示：多元线性回归、主成分回归和偏最小二乘法对大豆四级油酸价预测的标准偏差分别为0.1472%、0.1801% 和0.1576%，BP 神经网络的预测标准偏差为0.1387%。

Application of BP Neural Network in Detecting Acid Value of Oil Using Near Infrared Spectrum

For the fourth degree of soybean oil, a BP neural network was introduced to construct model for its near-infrared spectral data. The structure and parameters of the neural network were analyzed, and the structure design of samples training aggregation and primary factor of network input were thoroughly investigated. With regard to other multivariable modeling methods, extraction effects of useful information about near infrared spectrum were analyzed. For acid value prediction standard deviation of the fourth degree of soybean oil, the results of multivariable linear regression, primary components regression and partial least square are 0.1472%, 0.1801% and 0.1576% respectively, whereas the result of BP neural network is 0.1387%.