黄酒总酚含量检测:一种基于GA-LSSVM的近红外光谱波段选择方法

摘要

主要研究了近红外光谱技术对成品黄酒中总酚含量快速检测的可行性.针对近红外光谱样本少、非线性等特点,首次将最小二乘支持向量机(Least squares support vector machines, LSSVM)方法引入到传统遗传算法(genetic algorithms,GA)的波长选择中,提出一种基于GA-LSSVM的近红外光谱波段选择方法.该方法采用LSSVM建立小样本下不同波段的非线性模型,然后通过GA算法进行波长的优化选择.应用中,基于GA-LSSVM模型的总酚预测集相关系数(Rp)为0.9734,预测均方根误差(RMSEP)为5.5596,相比于传统方法,GA-LSSVM算法能够较好地提取非线性信息,预测效果更好.%The objective of the paper is to achieve the rapid detection of the total phenol in the Chinese rice wine by NIRS. In order to develop the model for nonlinear NIRS with small sample, least squares support vector machine (LSSVM) is introduced into the genetic algorithm-based (GA-based) wavelength-selection method, and a GA-LSSVM method is proposed. In the proposed method, each segment of wavelength is modeled by the LSSVM method, and the optimal segments are determined by the GA algorithm. By employing the GA-LSSVM model, the prediction correlation coefficient of the total phenol is 0.9734, and the root mean square error for prediction (RMSEP) is 5.5596. The application results demonstrate that compared with the conventional methods, the proposed method can achieve better extraction of the nonlinear information hiding in the NIRS, and get the better prediction performance.