Variational mode decomposition unfolded partial least squares regression for ultraviolet-visible spectral analysis of edible oil blend, fuel oil and aqueous samples
作者：Deyun Wu, Joel B. Johnson, Kui Zhang, Yugao Guo, Dan Liu, Zhigang Wang, Xihui Bian*
關鍵字：Variational mode decomposition; partial least squares regression; quantitative analysis; chemometrics
論文來源：期刊
具體來源：Microchemical Journal, 2024, 196, 109587
發表時間：2024年
To exploit the abundant information embedded amongst spectra, a novel regression model, named variational mode decomposition unfolded partial least squares regression (VMD-UPLSR), was developed for ultraviolet-visible (UV-Vis) spectral analysis of complex samples. In the method, variational mode decomposition (VMD) is firstly used to decompose each spectrum into K mode components (uk) with different frequencies. Then the mode components are unfolded to an extended matrix in variable dimensions. Finally, partial least squares regression (PLSR) is used to build a quantitative model between the extended matrix and target values. The performance of VMD-UPLSR is verified by three UV-Vis spectral datasets of edible oil blend, fuel oil and aqueous samples for quantification of sunflower oil, polyaromatics and chromium ion, respectively. The contents of the three analytes are in the ranges of 0-98.99 %(m/m), 0-0.9 %(V/V) and 5.01-73.12 mg/L, respectively. Root mean square error of prediction (RMSEP) of VMD-UPLSR is 6.96, 0.057 and 4.19 for edible oil blend, fuel oil and aqueous solution datasets, respectively. Compared with single PLSR and uk-PLSR, results show that the proposed method displays the best prediction accuracy in the three datasets. Therefore, VMD-UPLSR can be a valuable alternative method for the quantitative analysis of complex samples.