基于DS算法的混胺燃料近红外光谱分析模型传递

对混胺燃料的近红外光谱分析模型的传递方法进行研究。采用 K/S(Kennard/Stone)算法选择转换集样品,采摘用直接校正(Direct Standardization,DS)算法对从仪器采集的光谱进行校正。通过光谱平均差异(ARMS)比较奇异值分解(Singular Value Decomposition,SVD)算法和偏最小二乘法(Partial Least Squares,PLS)对光谱校正的效果。当 PLS算法的最佳主因子数为 3时,DS-PLS算法的光谱校正率可达到 97.5%,优于 DS-SVD算法。混胺样品的分析模型经过 DS-PLS算法传递后,对从仪器的混胺样品各项指标的预测标准偏差(SEP)明显好于传递前,与主仪器预测效果接近,说明采用 K/S算法选择合适的转换集样品后,通过 DS-PLS模型传递算法可有效降低仪器间的光谱差异,实现近红外光谱分析模型在各台光谱仪之间共享。

The Calibration Transfer of Near Ifrared(NIR)Spectral Analysis for Mixed-Amine Fuel Based on DS-PLS Algorithm

The Calibration Transfer of Near Ifrared(NIR)Spectral Analysis for Mixed-Amine Fuel Based on DS-PLS Algo?rithm was studied. Transfer samples were selected with kennard/stone(K/S) algorithm, then direct standardization(DS) wasused to correct near infrared spectra in order to share the model set in one instrument (reference instrument) with the otherone (target instrument). Correction effect of singular value decomposition(SVD) and partial least squares(PLS) was com?pared by average of root mean square (ARMS). When principal factor number of PLS was 3, prediction-corrected of DSPLS algorithm can reach 97.5%, which is better than DS-SVD algorithm. After the analytical model for mixed amine sam?ples was transferred through the DS-PLS algorithm, the standard deviation of prediction set (SEP) for each item of target in?strument was significantly better, which was close to the prediction result of the reference instrument. The results indicatethat the DS-PLS algorithm can effectively reduce the spectral difference between instruments and realize the sharing ofNIR model among different spectrometers after selecting the transfer samples by K/S algorithm.