As a kind of molecular fingerprint spectrum, Raman spectroscopy can perform qualitative analysis of substances based on the vibration between the molecules of substances, and is widely used in many fields. Raman spectroscopy can be applied to portable monitoring systems, but the amount of data is too large. If the data is not processed, it will greatly increase the subsequent analysis time and affect the speed of automatic identification. Nine common flammable liquids such as 90#, 93# and 97# gasoline, acetone, xylene, methanol, ethanol, ethylene glycol and tert-butanol are selected, data preprocessing analysis is performed on them, and the characteristic peak extraction effect is significantly, the data is compressed by 512 points, and then the support vector machine (SVM) and the random forest (RF) classification algorithms are selected for model training. Research results show that the cross-validation accuracy of the RF algorithm for identifying combustible liquid samples is higher than that of the SVM algorithm, and the mean square error results of the RF algorithm are also better than that of the SVM algorithm. The use of Raman spectroscopy technology can clearly detect the peaks of combustible liquid samples, and compress the data to increase the analysis speed, which can provide technical references for subsequent instrument miniaturization.