Abstract We introduce two methods for the calculation of compositions of liquid mixtures from near infrared (NIR) absorption spectroscopy data. Radial basis function (RBF) and artificial neural network (ANN) approaches are separately applied to establish correlations between spectral data and concentrations of each component. Principal component analysis (PCA) was implemented to establish the correlations and both RBF and ANN methods were trained with the first 5 principal components (PCs) obtained from 200 absorption spectra of liquid mixture samples. The trained systems were tested with 27 laboratory measurements; both results yield very good predictions of component concentrations with root-mean-square errors (RMSE) of 2–3 %. It is shown that RBF and ANN methods yield prediction errors 33 % smaller than with traditional standard multivariate methods.

中文翻译：

---

通过径向基函数和人工神经网络获得的近红外光谱数据的液体混合物成分

摘要 我们介绍了根据近红外 (NIR) 吸收光谱数据计算液体混合物成分的两种方法。径向基函数 (RBF) 和人工神经网络 (ANN) 方法分别用于建立光谱数据和每个成分浓度之间的相关性。实施主成分分析 (PCA) 以建立相关性，并使用从液体混合物样品的 200 个吸收光谱中获得的前 5 个主成分 (PC) 训练 RBF 和 ANN 方法。训练有素的系统通过 27 次实验室测量进行了测试；两种结果都对成分浓度进行了非常好的预测，均方根误差 (RMSE) 为 2–3%。