基于算术优化算法的复杂样品光谱变量选择方法研究OA

Accurate quantification of components in complex samples is challenging in analytical chemistry,yet modern spectroscopy paired with chemometrics performs well for such quantitative analysis.However,spectral data usually contain thousands of variables,many of which are irrelevant or redundant,leading to poor prediction accuracy.Therefore,effective variable selection has become a critical step in model development.In this study,arithmetic optimization algorithm(AOA)was discretized and combined with partial least squares(PLS)for spectral quantitative analysis for the first time.The performance of the proposed method was evaluated using a ultraviolet-visible-near-infrared spectral dataset of pear and a Raman spectral dataset of metabolite samples.Three key parameters of AOA-PLS,including the latent variables,maximum iterations,and transfer function,were optimized.The proposed AOA-PLS method was compared with full-spectrum PLS,uninformative variable elimination-PLS(UVE-PLS),Monte Carlo uninformative variable elimination-PLS(MC-UVE-PLS),randomization test-PLS(RT-PLS),particle swarm optimization-PLS(PSO-PLS)and whale optimization algorithm-PLS(WOA-PLS).The results showed that AOA-PLS retained only 42 variables,corresponding to a 86.27%reduction from the original 306 full-spectrum variables,reducing the root mean square error of prediction(RMSEP)from 1.1644 to 0.5530 and increasing the correlation coefficient(R)from 0.7650 to 0.9520 for pear dataset.For metabolite dataset,AOA-PLS selected only 42 variables,achieving a 90.05%reduction from the original 422 variables,while the RMSEP decreased from 0.0918 to 0.0571 and Rincreased from 0.9667 to 0.9872.Thus,AOA-PLS can substantially reduce the number of variables while notably improves prediction accuracy.