Quantitative Analysis of Infrared Spectroscopy of Alkane Gas Based on Random Forest Algorithm

The advantages of infrared spectroscopy analysis method are safe, effective, fast, pollution-free, and low-cost, which perfectly meets the current industrial measurement and measurement needs. In this paper, the infrared spectra of various elemental alkane gases are treated by random forest algorithm combined with multiple pretreatment schemes. Firstly, the number of components and concentration labels of alkane gas in the dataset are analyzed for manual data division, and three pretreatment methods have been selected: principal component analysis, Lowess smoothing method and S-G smoothing method, and two parameters in the random forest are fixed, and the regression prediction of ethane low concentration components is carried out by combining three pretreatments. S-G smoothing method is preferred to be the optimal pretreatment method. Then, S-G smoothing is used as the preprocessing to optimize the two parameters of the random forest. It is concluded that the optimal parameters are optimal when the number of decision trees is 300 and the number of leaves is 4, and the low concentration components of ethane are best predicted, with
$\mathrm{R}^{2}=0.9991$
and
$\text{RMSE}=4.1495$
.