Shinzawa Hideyuki, Jiang Jian-Hui, Iwahashi Makio, Noda Isao, Ozaki Yukihiro
Department of Chemistry, School of Science and Technology, and Research Center for Near-Infrared Spectroscopy, Kwansei-Gakuin University, Sanda, Hyogo 669-1337, Japan.
Anal Chim Acta. 2007 Jul 9;595(1-2):275-81. doi: 10.1016/j.aca.2006.12.004. Epub 2006 Dec 9.
Particle swarm optimization (PSO) combined with alternating least squares (ALS) is introduced to self-modeling curve resolution (SMCR) in this study for effective initial estimate. The proposed method aims to search concentration profiles or pure spectra which give the best resolution result by PSO. SMCR sometimes yields insufficient resolution results by getting trapped in a local minimum with poor initial estimates. The proposed method enables to reduce an undesirable effect of the local minimum in SMCR due to the advantages of PSO. Moreover, a new criterion based on global phase angle is also proposed for more effective performance of SMCR. It takes full advantage of data structure, that is to say, a sequential change with respect to a perturbation can be considered in SMCR with the criterion. To demonstrate its potential, SMCR by PSO is applied to concentration-dependent near-infrared (NIR) spectra of mixture solutions of oleic acid (OA) and ethanol. Its curve resolution performances are compared with SMCR with evolving factor analysis (EFA). The results show that SMCR by PSO yields significantly better curve resolution performances than those by EFA. It is revealed that SMCR by PSO is less sensitive to a local minimum in SMCR and it can be a new effective tool for curve resolution analysis.
本研究将粒子群优化算法（PSO）与交替最小二乘法（ALS）相结合引入自建模曲线分辨法（SMCR），以进行有效的初始估计。所提出的方法旨在通过PSO搜索能给出最佳分辨结果的浓度分布或纯光谱。由于初始估计不佳，SMCR有时会陷入局部最小值，从而产生分辨率不足的结果。由于PSO的优势，所提出的方法能够减少SMCR中局部最小值的不良影响。此外，还提出了一种基于全局相角的新准则，以提高SMCR的性能。它充分利用了数据结构，也就是说，在具有该准则的SMCR中可以考虑相对于扰动的顺序变化。为了证明其潜力，将基于PSO的SMCR应用于油酸（OA）和乙醇混合溶液的浓度相关近红外（NIR）光谱。将其曲线分辨性能与采用渐进因子分析（EFA）的SMCR进行比较。结果表明，基于PSO的SMCR产生的曲线分辨性能明显优于基于EFA的SMCR。结果表明，基于PSO的SMCR对SMCR中的局部最小值不太敏感，它可以成为曲线分辨分析的一种新的有效工具。
