关键词: 拉曼光谱, 荧光背景, 痕量原油, 特征变量选择, 偏最小二乘回归

Abstract: In the process of heat exchange between crude oil and light oil on the side line of atmospheric distillation column, there is the problem of crude oil leakage in refining and chemical enterprises, which has an adverse effect on the safety production and subsequent processing. Based on the sensitivity of Raman fluorescence background to crude oil, the detection of trace amounts of crude oil in light oil was introduced. Raman spectra of 90 naphtha samples containing trace amounts of crude oil were scanned as experimental data, and the model was established using partial least squares regression (PLSR) combined with genetic algorithm (GA), random frog algorithm (RF) and competitive adaptive reweighting sampling algorithm (CARS). The best model is the combination of CARS and PLSR, in which RMSEP is 1.567 4 μg/g. The results show that the detection method using Raman spectral fluorescence background can be used to accurately detect trace crude oil of 1-100 μg/g after algorithm optimization, which can fully meet the requirements of refining and chemical enterprises.

Key words: Raman spectroscopy, fluorescence background, trace amounts of crude oil, feature variable selection, partial least squares regression