Study On Detection Method Of Kerosene In Illegal Additives Of Gasoline Based On Vibration Spectrum

With the increasing demand for gasoline,many unscrupulous traders added more illegal additives to the refining,transportation and sales of gasoline in order to obtain more profits.As one of the illegal additives,kerosene is similar in physical and chemical properties to gasoline,which is difficult to identify.The use of gasoline to add kerosene not only damages the ignition system of gasoline engine,but also causes insufficient combustion which results in insufficient power performance of the vehicle and pollutes the environment.Therefore,there is an urgent need for a technical means to detect gasoline blended with kerosene quickly and accurately.In this paper,the Mid-infrared and Raman spectroscopy techniques were used to detect gasoline blended with different content of kerosene,a qualitative and quantitative analysis model was established.The specific results were as follows:1.Using the mid-infrared spectrometer,gasoline blended with 0.5%to 20%content kerosene and pure 92#gasoline as the research object,there are two types of discriminant models were established,random forest and partial least squares discrimination.The result shows that the random forest discriminant model is better,the correct rate of classification is100%,and the contribution rate of feature band to the model is greater;the gasoline blended with 0.5%to 20%content kerosene is used as the research object,a variety of spectral pretreatment methods were used,three quantitative analysis model of the content of kerosene was established by least squares support vector machine,extreme learning machine and random forest.The result showed that the random forest model established by using the 1~stt derivative spectral data was the best.When the decision tree number is 54,the error rate of OBB is the smallest,which is 3.88%.At the same time,the correlation coefficient R_p of the prediction set was 0.982 and the RMSEP was 0.218.2.Using the Raman spectroscopy,gasoline blended with 0.5%to 20%content kerosene and pure 92#gasoline as the research object,the random forest discriminant model was established,the correct classification rate of the model was 96.2%,the model can be carried out classification;Using gasoline blended with 0.5%to 20%content kerosene as the research object,a variety of spectral pretreatment methods and three band screening methods were used to establish a random forest model for predicting kerosene content.Compared with the model established by using band screening and the only spectral pretreatment to establish the model,the result showed that the random forest model which was established by processing Baseline+MSC+SPA+GA,not only the root mean square error of kerosene content was the lowest 1.464,but also the R_p was the highest 0.971.3.Using the mid-infrared spectrometer,taking gasoline blended with 0.5%to 20%content kerosene was used as the research object.The characteristic peak intensity and peak area of each spectrum at 1000 cm~-11 were calculated,a linear regressions were respectively performed with the kerosene content,a mathematical model for predicting the kerosene content was established,and it was found that the effect of using the peak intensity to predict the kerosene concentration was relatively good.The correlation coefficient reached 0.816 and the predicted root mean square error was 0.793.4.For Raman spectroscopy,the multivariate linear regression model established by chemometrics was compared with the one-variable regression model established by the peak intensity and peak area for predicting kerosene content.The multivariate regression analysis method established the model with higher accuracy.The requirements of the instrument are also high,and the one-variable regression model established using the peak intensity or peak area only analyzes a single peak,which can realize the quick and simple prediction of kerosene.