拉曼光谱结合PSO-LSSVM算法检测三组分食用调和油含量

发布时间：2018-03-13 08:09

本文选题：拉曼光谱　切入点：粒子群优化　出处：《光谱学与光谱分析》2017年08期 　论文类型：期刊论文

【摘要】：提出了一种将拉曼光谱和基于粒子群的最小二乘支持向量机(PSO-LSSVM)算法相结合快速定量检测三组分食用调和油含量的方法。以三组分的食用调和油为研究对象,对拉曼光谱分四步进行了预处理,进而准确提取拉曼光谱的特征峰强度。以训练集样本的特征峰强度和调和油样品的百分比含量作为回归预测模型的输入值和输出值,建立LSSVM和PSO-LSSVM数学模型,通过测试集样本的相关系数和均方误差对模型的预测能力进行分析。非线性建模的最小二乘支持向量机(LSSVM)算法的核函数参数σ和正则化参数γ对模型的学习和泛化能力影响很大,导致模型的预测精度和泛化能力过度依赖于参数——在优化步长过小时耗时较长,过大时又无法得到全局最优值。提出的PSO-LSSVM算法,利用粒子群全局优化能力和收敛速度快的特点对LSSVM的模型参数σ和γ进行优化,从而克服LSSVM算法中耗时与盲目性的问题。分析结果表明,PSO-LSSVM算法对三组分食用调和油中大豆油、花生油和葵花仁油定量预测模型的测试集相关系数分别为0.967 7,0.997 2,0.995 3;均方误差分别为0.054 9,0.009 2,0.047 1。与LSSVM算法相比,PSO-LSSVM模型的预测精度更高。因此,该方法可以快速、准确地检测三组分食用调和油的含量。
[Abstract]:In this paper, a fast and quantitative method for detecting the content of three components of edible blending oil by combining Raman spectrum with PSO-LSSVM-based least squares support vector machine (LSSVM) algorithm is proposed. The three-component edible blending oil is taken as the object of study. The Raman spectrum was pretreated in four steps, and the characteristic peak intensity of the Raman spectrum was extracted accurately. The input and output values of the regression prediction model were determined by using the characteristic peak strength of the training set sample and the percentage content of the blending oil sample as the input value and output value of the regression prediction model. Establish mathematical models of LSSVM and PSO-LSSVM, The prediction ability of the model is analyzed by correlation coefficient and mean square error of test set. Learning and generalization of kernel function parameter 蟽 and regularization parameter 纬 of the least squares support vector machine (LSSVM) algorithm for nonlinear modeling. The influence of chemical ability is very great. The prediction accuracy and generalization ability of the model are too dependent on the parameters. The proposed PSO-LSSVM algorithm can not get the global optimal value when the optimization step is too long. The model parameters 蟽 and 纬 of LSSVM are optimized by using the global optimization ability and fast convergence speed of PSO, which can overcome the problems of time consuming and blindness in LSSVM algorithm. The analysis results show that PSO-LSSVM algorithm can be used to solve the problem of soybean oil in three-component edible blending oil. The correlation coefficient of the quantitative prediction model for peanut oil and sunflower kernel oil is 0.967 7 / 0.997 / 20.995 3 and the mean square error is 0.054 9 / 0.009 / 2 / 0.0471.Compared with the LSSVM algorithm, the prediction accuracy of the PSO-LSSVM model is higher than that of the LSSVM algorithm. The content of edible blending oil was detected accurately.
【作者单位】：燕山大学信息科学与工程学院河北省特种光纤与光纤传感重点实验室;
【基金】：国家自然科学基金项目(61205068) 中国博士后科学基金项目(2013M541200) 河北省自然科学基金项目(F2014203125) 燕山大学“新锐工程”人才支持计划项目资助
【分类号】：O433;TP18;TS227

【相似文献】