A powerful tool for near-infrared spectroscopy: Synergy adaptive moving window algorithm based on the immune support vector machine

• Published in: Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy Vol. 282; p. 121631
• Main Authors: Wang, Shenghao, Zhang, Peng, Chang, Jing, Fang, Zeping, Yang, Yi, Lin, Manman, Meng, Yanhong, Lin, Zhixin
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 05.12.2022
• Subjects: Diesel fuel; Gasoline; Immune clone algorithm; Near-infrared spectrum; Optimization calculation; Support vector machine; Support vector machine; Optimization calculation; Diesel fuel; Immune clone algorithm; Gasoline; Near-infrared spectrum
• ISSN: 1386-1425
• DOI: 10.1016/j.saa.2022.121631

[Display omitted] •It proposes a new convenient and efficient method for near-infrared spectroscopy base on AI algorithm.•It solves the problem of optimizing preprocessing methods, wavelength variables, and the hyper-parameters of the calibration model simultaneously.•It can effectively improve the accuracy of the model, select useful wavelength variable, and give reasonable preprocessing method.•It can deal not only with near-infrared spectroscopy but also with other related data.•If the type of calibration model or the optimization strategy were replaced, it can be easily transformed into other related modeling strategy. Traditional trial-and-error methods are time-consuming and inefficient, especially very unfriendly to inexperienced analysts, and are sometimes still used to select preprocessing methods or wavelength variables in near-infrared spectroscopy (NIR). To deal with this problem, a new optimization algorithm called synergy adaptive moving window algorithm based on the immune support vector machine (SA-MW-ISVM) is proposed in this paper. Following the principle of SA-MW-ISVM, the original problem of calibration model optimization is transformed into a mathematical optimization problem that can be processed by the proposed immune support vector machine regression algorithm. The main objective of this optimization problem is the calibration model performance; meanwhile, the constraint conditions include a reasonable spectral data value, spectral data preprocessing method, and calibration model parameters. A unique antibody structure and specific coding and decoding method are used to achieve collaborative optimization in NIR spectroscopy. The tests on four actual near-infrared datasets, including a group of gasoline and three groups of diesel fuels, have shown that the proposed SA-MW-ISVM algorithm can significantly improve the calibration performance and thus achieve accurate prediction results. In the case of gasoline, the SA-MW-ISVM algorithm can decrease the prediction error by 44.09% compared with the common benchmark partial least square (PLS). Meanwhile, in the case of diesel fuels, the SA-MW-ISVM algorithm can decrease the prediction error of cetane number, freezing temperature, and viscosity by 9.99%, 28.69%, and 43.85%, respectively, compared with the PLS. The powerful prediction performance of the SA-MW-ISVM algorithm makes it an ideal tool for modeling near-infrared spectral data or other related application fields.