基于CARS变量筛选的固态发酵pH值近红外检测

NIRS detection of pH value in solid-state fermentation process based on variable selection method of CARS

为了提高近红外光谱技术快速检测固态发酵过程中pH值的精度和稳定性,提出了采用竞争自适应重加权采样(competitive adaptive reweighted sampling,CARS)法筛选出与pH相关的波长变量建立PLS预测模型,对验证集样本进行预测的方法。并与2种常见的变量筛选法GA-PLS和蒙特卡罗无信息变量消除法(MC-UVE)相比较.实验结果表明:CARS方法能有效筛选有用波长26个变量建立PLS模型,其校正集交互验证均方根误差(RMSECV)以及交互验证相关系数(Rc)分别为0.0368和0.9950;验证集的预测均方根误差(RMSEP)以及预测相关系数(Rp)分别为0.0589和0.9895。

In order to improve accuracy and stability of NIRS rapid detection model of pH value in solid-state fermentation process, competitive adaptive reweighted sampling （CARS） method was employed to select characteristic wavelength variables. The key variables for pH were selected by CARS method to establish prediction model by partial least squares （PLS） regression. The prediction model was used to predict pH value in validation set samples. Compared with two common variable selection methods GA-PLS and MC-UVE, CARS method can select characteristic wavelength variables effectively and improve accuracy of prediction model, the result of CARS-PLS model are better than GA-PLS and MC-UVE-PLS model. At last, twenty wavelength variables were selected to establish the PLS model, the result of root mean square error cross-validation （RMSECV） is 0.0368 and the correlation coefficient （Rc） is 0.9950 in the training set, the result of root mean square error prediction （RMSEP） is 0.0589 and the correlation coefficient （Rp） is 0.9895 in the validation set.