建模的分布状态对近红外预测小麦蛋白质含量的影响

Modelling Distribution Influence on NIR Prediction of Wheat Protein Content

建模样品水分含量的差异对近红外光谱分析模型的稳健性影响很大。以小麦蛋白质含量为研究对象,定标集的水分含量分布设为3个不同δ值的正态分布和1个均态分布,分别建立预测小麦蛋白质含量的模型。结果表明,当用4个模型预测水分范围在10%～15%,蛋白质含量平均值为14.31%的样品时,3个正态分布预测的结果分别是13.63%、13.83%、14.04%,但均态模型的预测结果是14.24%,明显比正态模型预测效果好。当预测中间水分样品的蛋白含量时,正态模型预测准确性要优于均态模型。应用中,可先用较宽水分含量范围建立的模型对待测样品水分进行粗测,后用与待测样品水分粗测值适配的窄水分含量范围的模型来测定样品的蛋白质含量,可减少由于建模水分的差异所带来的预测误差。

Modelling sample moisture content differences on the robustness of the model of near infrared spec- troscopy is greatly. Research on grain protein content of wheat in this experiment, calibration sets the moisture content distribution to 3 different or values of normal distribution and one uniform distribution. Respectively cre- ation of model for predicting protein content in wheat. Results show that normal distribution forecasting model of RMSE Root Mean Square Error （RMSECV） is less than the uniform distribution model. When using 4 models to prediction moisture ranging from 10%-15%, protein content of an average of 14.31% sample, prediction of 3 normal distribution results respectively is 13.63%, 13.83%, 14.04%, but uniform distribution model forecast the result is 14.24%. Significantly better than the normal model to predict effects. When forecasting protein content of the sample of intermediate moisture , prediction accuracy of normal distribution model better than uniform distribution model. In the application, can be built by using a wide range of moisture content model test samples of water, and test water samples after adaptation narrow range of moisture content of the full scale value model for determination of protein content of samples, can be reduced because of differences in modelling water brought about by the prediction error.