论文部分内容阅读
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 the accuracy and stability of pH value in solid-state fermentation by near-infrared spectroscopy (NIRS), a method of competitive adaptive weighted reweighted sampling (CARS) was proposed to select pH-dependent wavelength variables to establish PLS prediction model , A method of predicting samples of the validation set. And compared with the two common variable screening GA-PLS and Monte Carlo non-information variable elimination method (MC-UVE) .The experimental results show that: CARS method can effectively filter the useful wavelength 26 variables to establish PLS model, the calibration set RMSECV and Rc were 0.0368 and 0.9950 respectively. The root mean square error of prediction (RMSEP) and prediction correlation coefficient (Rp) of validation set were 0.0589 and 0.9895, respectively.