基于集成建模方法的便携式近红外光谱仪酒醅成分检测研究

Study on detection method of fermented grains composition by portable NIR spectrometer based on integrated modeling

为了提升便携式近红外光谱仪酒醅成分检测性能,提出了一种集成建模方法。利用便携式近红外光谱仪对酒醅样本进行光谱数据采集,将采集后的数据按随机及临近方式划分,继而在不同的预处理算法与偏最小二乘(PLS)建模算法组合中筛选基模,并基于其自身的相关系数(R^(2))值进行模型权重计算,最终结合权重值进行模型集成预测。相较于单模型建模预测结果,集成建模方法将酒醅样本各成分预测准确率平均提升了约8.0%。结果表明,基于集成建模的方法在酒醅成分检测应用中,有效提升了模型预测准确率,是一种可以在便携式近红外光谱仪数据建模中推广应用的实用方法。

The collected data of fermented grains samples were divided by random and adjacent modes.The member model was selected among different parameters combinations including preprocessing algorithms and partial least squares(PLS)modeling algorithm. The model weights were calculated based on their own correlation coefficient(R^(2))values,and finally combined with weight values for model integration and prediction.Compared to the prediction results of single model modeling,the integrated modeling method was verified to improve the prediction accuracy of each component of fermented grains samples by about 8. 0% on average.The results show that this method can effectively improve the model prediction accuracy.