基于特征波段的黄酒近红外光谱检测模型递归更新方法

发布时间：2018-03-31 12:51

本文选题：近红外光谱　切入点：模型更新　出处：《光谱学与光谱分析》2017年11期

【摘要】：近红外光谱是一种快速、无损的定量分析工具。为了提高黄酒关键参数的检测水平,采用近红外光谱法进行定量分析。检测过程中,由于受环境波动、仪器老化、原料变化等因素的影响,基于旧样品所建的模型的精确度逐渐下降。为保持模型的预测精度,引入递归偏最小二乘(recursive partial least square,RPLS)对模型进行更新。以往此方法多使用全谱信息扩充建模集并进行递归计算,光谱的变量多,且包含环境影响等干扰信息,更新计算量大,且精度的提升效果不明显。考虑到黄酒生产过程中特征波段变化小的特性,提出了一种基于特征波段的黄酒近红外光谱检测模型递归更新方法。先采用相关系数法提取特征波段建立低维模型,在采集到新样品理化值后,再利用其特征波段光谱信息,使用递归偏最小二乘对低维模型进行更新。此方法被应用于黄酒总酸的近红外检测模型更新。模型评价使用相关系数r,预测标准偏差RMSEP和预测相对分析误差RPD三个指标。结果表明:采用本方法后,模型稳定性显著优化,计算效率有所提升,模型预测效果良好,三个评价指标分别达到0.965 7,0.184 3和3.736 2,较全谱PRLS时分别提高3%,24%和31%,在实际应用中有一定的参考价值。
[Abstract]:Near-infrared spectroscopy (NIR) is a rapid and non-destructive tool for quantitative analysis. In order to improve the detection level of key parameters of yellow rice wine, the near-infrared spectroscopy is used for quantitative analysis. The accuracy of the model based on the old sample has gradually declined due to the influence of factors such as the change of raw material. In order to maintain the prediction accuracy of the model, The recursive partial least square (RPLS) is introduced to update the model. In the past, the whole spectrum information is used to expand the modeling set and calculate recursively. The spectral variables are many, and the disturbance information such as environmental impact is included. And the effect of raising precision is not obvious. Considering the small change of characteristic band in rice wine production, In this paper, a recursive updating method of near infrared spectrum detection model of yellow rice wine based on feature band is proposed. Firstly, the correlation coefficient method is used to extract the feature band to establish a low dimensional model. After the new physical and chemical values of the sample are collected, the spectral information of the characteristic band is then used. This method is used to update the low-dimensional model by recursive partial least squares. This method has been applied to the near infrared detection model updating of total acid in rice wine. The model evaluation uses correlation coefficient r, prediction standard deviation (RMSEP) and prediction relative analysis error (RPD). The results show that, after using this method, The stability of the model is significantly optimized, the calculation efficiency is improved, and the prediction effect of the model is good. The three evaluation indexes are 0.965 0.1843 and 3.736 2, respectively, which are 34% and 31% higher than those of the full-spectrum PRLS, respectively. It has certain reference value in practical application.
【作者单位】：江南大学轻工过程先进控制教育部重点实验室;
【基金】：国家自然科学基金项目(61573169) 流程工业综合自动化国家重点实验室开放课题基金项目(PAL-N201502) 中央高校基本科研业务费专项资金项目(JUSRP51407B)资助
【分类号】：O657.33;TS262.4

【相似文献】