汽油多参数拉曼光谱分析仪中的稳健支持向量机方法

Robust SVM in multi-parameter gasoline analyzer using Raman spectroscopy

由于拉曼光谱可以反映汽油各有机物基团丰富的信息,拉曼光谱更加适合于汽油质量指标的快速分析,并可同时测定多种参数。为避免少量异常训练样本对校正模型的影响,本文采用了一种迭代的稳健支持向量机算法。该方法首先求取训练样本的回归残差,然后利用残差的正态分布置信区间来鉴别异常样本并选取正常样本,最后用选出的正常样本作为训练样本并建立最小二乘支持向量机模型,对测试样本进行预测。将本文的算法应用于汽油多参数拉曼光谱快速分析仪中,结果证明:该方法具有很好的稳健性,同时具有很好的预测精度。

Raman spectroscopy is more suitable for analyzing the properties of gasoline, because Raman spectroscopy can reflect the information of the organic compound in gasoline. In order to overcome the influence of several outliers to calibration model, a robust version of support vector machine is introduced to overcome the influence of the outliers. In the proposed approach, regression error of the original training data set is computed, and then the confidence interval of the residuals distribution is applied iteratively to detect those outliers and select normal samples. In fact, a LS-SVM is created from the model being trained with the selected training sub-dataset without outliers to estimate the test samples. Applying this approach in Raman spectrum analysis for gasoline properties, experimental results show its robustness and accuracy.