粒子群算法结合支持向量机回归法用于近红外光谱建模

Application of Particle Swarm Optimization-Least Square Support Vector Machine Regression to Modeling of Near Infrared Spectra

研究了最小二乘法支持向量机(LSSVM)应用于烟丝样品和小麦样品的近红外光谱建模,采用粒子群优化算法(PSO)优化LSSVM的参数。通过对烟草样品和小麦样品的近红外光谱建模和预测,并与常规的偏最小二乘法(PLS)比较发现,PSO-LSSVM法具有更好的预测效果和稳健性。

In this paper,the application of least square support vector machine（LSSVM） to the modeling of near infrared（NIR） spectra of tobacco and wheat samples was studied.The modeling parameters for LSSVM were optimized by particle swarm optimization（PSO）.Thus,a quantitative modeling of infrared spectra based on PSO-LSSVM for tobacco and wheat samples was established and was used to predict the unknown samples.The NIR spectra were divided into calibration set and prediction set.The calibration set was mainly used in the modeling,while the prediction set was used for unknown samples to test the validity of the models.In the modeling of NIR spectra of tobacco samples,the reducing sugar,total sugar,total nitrogen and total plant alkaloid were of target analytes.It was found that the root-mean-square error of cross-valication（RMSECV） obtained by PSO-LSSVM was smaller than that by PLS.In the prediction of NIR spectra of tobacco samples and wheat samples,the root mean square error of prediction（RMSEP） obtained by PSO-LSSVM was also smaller than that by PLS.In conclusion,the PSO-LSSVM gave good precision in modeling and good accuracy in prediction.