混合气体红外光谱支持向量机分析的新方法

介绍了一种基于支持向量机的混合气体红外光谱组分浓度和种类分析的新方法。利用核函数将组分气体特征吸收谱线重叠严重的混合气体光谱在高维空间变换后，建立SVM回归校正模型，进行混合气体浓度分析。在利用支持向量机回归校正模型进行混合气体组分浓度分析的同时，证明支持向量机回归校正模型也可用于混合气体组分种类分析。对不同组分和不同组分浓度的混合气体红外光谱数据进行了实验，研究了谱仪扫描间隔、分析特征波长范围、核函数和惩罚因子等因素对分析结果的影响。混合气体组分浓度实验结果的最大平均绝对误差Mean AE为0.132%;混合气体组分种类识别的准确率大于94%。解决了传统的光谱分析方法中光谱特征谱线重叠、光谱数据的维数大、定性和定量分析无法使用同一方法等问题，可用于其他混合气体的红外光谱分析，具有实际应用价值。

New Method of Mixed Gas Infrared Spectrum Analysis Based on SVM

A new method of infrared spectrum analysis based on support vector machine(SVM) for mixture gas was proposed.The kernel function in SVM was used to map the seriously overlapping absorption spectrum into high-dimensional space,and after transformation,the high-dimensional data could be processed in the original space,so the regression calibration model was established,then the regression calibration model with was applied to analyze the concentration of component gas.Meanwhile it was proved that the regression calibration model with SVM also could be used for component recognition of mixture gas.The method was applied to the analysis of different data samples.Some factors such as scan interval,range of the wavelength,kernel function and penalty coefficient C that affect the model were discussed.Experimental results show that the component concentration maximal Mean AE is 0.132%,and the component recognition accuracy is higher than 94%.The problems of overlapping absorption spectrum,using the same method for qualitative and quantitative analysis,and limit number of training sample,were solved.The method could be used in other mixture gas infrared spectrum analyses,promising theoretic and application values.