Method for improving prediction accuracy of gasoline properties

Abstract

The present invention provides a method for improving the prediction accuracy of gasoline properties. According to the method, the de-noising treatment is conducted through the wavelet transform, and a de-noised spectrum is subjected to fast Fourier transform. In this way, the useful information can be effectively extracted from the overlapped absorption peaks of the spectrum. The Euclidean distance is calculated based on first several Fourier transform coefficients with the cumulative proportion thereof to be up to 85%. Meanwhile, a certain number of samples smaller in distance are adopted as correction samples to establish a partial least squares model. Finally, a to-be-tested sample is predicted. Based on the above method, the noise is better removed, and the baseline drift is eliminated. Meanwhile, the feature information of a gasoline spectrum is effectively extracted, and the prediction accuracy of the model is improved. Therefore, the method has important engineering application value.

Description

technical field [0001] The invention relates to a method for improving the prediction accuracy of gasoline properties, and in particular relates to the field of gasoline property detection, especially a method for measuring the octane number of a research method. Background technique [0002] The analysis of properties such as research octane number (RON) and saturated vapor pressure is a must-test item in gasoline production, which is of great significance to the quality control of gasoline products. Traditional detection methods for these properties are complex, labor-intensive and time-consuming. The near-infrared spectroscopy analysis method has the advantages of simplicity, rapidity, and non-destructiveness, so it is more and more widely used in the analysis and detection of gasoline properties. [0003] In actual use, since the near-infrared original spectrum contains not only chemical information related to the composition of the sample, but also interference signals...

AI Technical Summary

Problems solved by technology

[0003] In actual use, since the near-infrared original spectrum contains not only chemical information related to the composition of the sample, but also interference signals caused by the state of the sample, stray light and high-frequency noise of the instrument, these interference signals will affect the spectrum. information, resulting in near-infrared spectrum baseline drift, low signal-to-noise ratio, and adverse effects on detection accuracy, the spectrum must be denoised

[0004] In addition, the near-infrared spectrum is the frequency doubling and combined spectrum band of the molecular vibration spectrum, the absorption is weak, and there is a phenomenon of overlapping absorption peaks, which is not conducive to directly extracting information related to the properties and composition of the sample. Improve the prediction accuracy of the model

Images