摘要
在对变量数远大于样本数的红外/近红外光谱数据进行分析时,经常需要对变量进行筛选或降维,为此提出了基于变量重要度指数的离散粒子群优化算法,并应用于五种食用油的偏最小二乘判别分析。变量的重要度指数为偏最小二乘回归系数和光谱纯度的乘积。在粒子群优化算法的初始化阶段引入变量重要度指数,利用轮盘赌算法增大选中重要度大的变量的概率,并且不减少种群的随机性。77个食用油样本的分类实验结果表明,与全变量偏最小二乘和经典粒子群优化算法相比,基于变量重要度指数的离散粒子群优化算法收敛速度较快,并在一定程度上避免了陷入局部最优,利用FT-IR光谱技术并结合化学计量学建立模型是一种有效的食用油分类分析方法。
When analyzing the infrared/near-infrared spectral data with the number of variables much larger than the number of samples, it was often necessary to screen or reduce the dimension of variables. Therefore, a discrete particle swarm optimization algorithm based on the variable importance index was proposed and applied to the partial least squares discriminant analysis of five kinds of edible oil. The variable importance index was the product of partial least squares regression coefficient and spectral purity value. In the initialization stage of particle swarm optimization algorithm, the variable importance index was introduced, and roulette algorithm was used to increase the probability of selecting variables with large importance index without reducing the randomness of the population. The experimental results of 77 edible oil samples showed that compared with all variable partial least squares and classical particle swarm optimization algorithm, the discrete particle swarm optimization algorithm based on variable importance index converged quickly and avoided falling into local optimization to a certain extent. It was an effective classification and analysis method of edible oil to use FT-IR spectroscopy and chemometrics to establish the proposed model.
作者
申琦
李盎
张晓芋
桑泽农
王志莹
SHEN Qi;LI Ang;ZHANG Xiaoyu;SANG Zenong;WANG Zhiying(College of Chemistry,Zhengzhou University,Zhengzhou 450001,China)
出处
《郑州大学学报(理学版)》
CAS
北大核心
2023年第1期84-88,共5页
Journal of Zhengzhou University:Natural Science Edition
基金
国家自然科学基金项目(21575131)
河南省高等学校大学生创新创业训练计划项目(S202110459020)。
关键词
红外光谱
离散粒子群优化算法
偏最小二乘判别分析
食用油
infrared spectrum
discrete particle swarm optimization algorithm
partial least squares discriminant analysis
edible oil