CN110174392B

CN110174392B - Fingerprint spectrum construction and identification method of high-identification-capacity multi-component complex oil product

Info

Publication number: CN110174392B
Application number: CN201910544009.1A
Authority: CN
Inventors: 于迎涛; 孙玉叶; 王季锋
Original assignee: Dalian Maritime University
Current assignee: Dalian Maritime University
Priority date: 2019-06-21
Filing date: 2019-06-21
Publication date: 2021-08-31
Anticipated expiration: 2039-06-21
Also published as: CN110174392A

Abstract

The invention relates to a fingerprint spectrum construction and identification method of a high-identification-capacity multi-component complex oil product based on Raman spectrum. The method comprises the steps of collecting spectrums containing fluorescence and Raman characteristics of a standard sample and a blind sample, calculating two-dimensional synchronous related spectrums of each standard sample and each blind sample, and converting the two-dimensional synchronous related spectrums into a one-dimensional array after resampling the two-dimensional synchronous related spectrum data under the condition of keeping the main characteristics of fingerprint spectrums; eliminating variables which are not important for discrimination by adopting a stepwise discrimination method, and only reserving variable groups with strong discrimination capability; and projecting the intensity data corresponding to the optimized variable group to a new low-dimensional space, and drawing a columnar stack diagram for visual comparison or drawing a three-dimensional space projection diagram for fingerprint feature comparison according to the projection coordinates in the new low-dimensional space. The method has high identification efficiency, and improves the accuracy and persuasion of crude oil identification compared with the existing crude oil identification method.

Description

Fingerprint spectrum construction and identification method of high-identification-capacity multi-component complex oil product

Technical Field

The invention relates to a high-identification-power identification method based on Raman spectrum, in particular to a fingerprint spectrum construction method of a multi-component complex oil product and an oil spill identification method based on the fingerprint spectrum.

Background

In ocean oil transportation and ocean oil and gas development, oil spill accidents happen occasionally. Tracing and identifying according to physical and chemical fingerprint information of the oil spill sample and the suspicious oil spill source can provide reliable scientific basis for responsibility identification and law adjudication of accidents, and is an important premise for taking effective control countermeasures in time.

On the basis of field law enforcement for many years and reference to foreign technologies, the national ocean administration published the industry standard of the sea surface oil spill identification system specification in 1997 (HY 043-. In 2007, the State administration of quality supervision, inspection and quarantine and the State Committee of standardization and administration issue the national recommended Standard of the sea surface oil spill identification System Standard (GB/T21247-: firstly, screening suspected oil spill source samples, and primarily screening by taking a fluorescence spectroscopy method or an infrared spectroscopy method as an optional method prior to a gas chromatography method to remove the suspected oil spill samples which are obviously inconsistent; then, gas chromatography and gas chromatography/mass spectrometry are performed, and if necessary, a monomolecular hydrocarbon stable carbon isotope analysis is performed to perform final discrimination. All samples were run in the same analytical instrument under the same analytical conditions. "

The Raman spectrum has the characteristics of simple and convenient test, small sample consumption, no need of pretreatment, no damage to the sample, no interference from seawater, easy portability and the like, and has wide application prospect in the field of oil sample identification. When raman spectroscopy is used to measure oil samples, the fluorescence transition and raman transition are excited simultaneously. Generally, the fluorescence signal intensity is large, and the spectrum peak is smooth; the raman signal is relatively weak and rich in detail. In the prior art, ZL201310347484.2 takes the original Raman spectrum containing fluorescence and Raman signals as the spectrum fingerprint of crude oil; and (3) converting the extracted spectral fingerprint data into a multi-dimensional space vector for identification without performing any fluorescence background subtraction treatment in fingerprint acquisition. ZL201610647605.9 carries out gradual discriminant analysis on intensity variables in the one-dimensional fingerprint spectrum of the crude oil, so that the identification capability of crude oil identification is improved. The existing method for detecting crude oil by using Raman spectrum is accurate in determination, convenient and fast.

Crude oil identification is a difficult point and a hot point in oil spill identification research, crude oil components are complex and diverse, accuracy and persuasion of oil spill identification are difficult to ensure by using a single detection method, internationally, crude oil identification commonly adopts multiple detection methods to be used together, particularly in judicial identification, different identification methods are required to be respectively adopted for the same oil spill sample, and obtained detection results are mutually verified. Therefore, the research and development of a new high-identification crude oil identification method has important practical value.

Disclosure of Invention

In order to be used in combination with a crude oil identification method in the prior art and improve the identification accuracy and persuasion of an oil sample, the invention provides a method for identifying a multi-component complex sample by combining two-dimensional correlation spectra, data resampling, stepwise discriminant analysis and low-dimensional space projection.

In order to realize the purpose, the invention adopts the following technical scheme:

a method for constructing and identifying a multi-component complex oil sample fingerprint spectrum comprises the following steps,

(1) respectively collecting parallel samples of a standard sample of a known type and a blind sample to be identified; aiming at each oil sample, a Raman spectrometer is adopted to measure a series of spectral data containing fluorescence and Raman information under different excitation conditions;

(2) performing interpolation processing on the spectral data measured under different excitation conditions to enable the abscissa of all the spectral data to correspond one to one, and calculating two-dimensional synchronous related spectra of each standard sample and each blind sample;

(3) under the condition of keeping the main characteristics of the fingerprint spectrum, resampling the two-dimensional synchronous related spectrum data; extracting an upper triangular array or a lower triangular array in the two-dimensional spectrum matrix after resampling, and converting the upper triangular array or the lower triangular array into a one-dimensional array;

(4) based on the one-dimensional array, discriminating the intensity variable by adopting gradual discrimination analysis, removing the unimportant variable for discrimination, and only keeping the variable group with strong discrimination as the preferred variable group; establishing a dictionary according to the principle of 'minimum same-class dispersion and maximum inter-class dispersion' to judge a function group; judging a function group based on the dictionary rules, and projecting the intensity data corresponding to the optimized variable group to a new low-dimensional space;

(5) and drawing a columnar stack diagram for visual comparison or drawing a three-dimensional space projection diagram for fingerprint feature comparison according to the new projection coordinates in the low-dimensional space, so that the blind sample can be identified.

The method for constructing and identifying the multi-component complex oil sample fingerprint spectrum comprises the steps of (5) visually comparing similar samples in a columnar stack diagram, and then drawing a three-dimensional space projection diagram fingerprint characteristic comparison.

According to the method for constructing and identifying the multi-component complex oil sample fingerprint spectrum, the excitation light wavelength used in the step (1) is two or more of 532nm, 514.5nm, 488nm, 457.9nm, 638nm, 632.8nm and 785nm, all the spectral data are not subjected to any fluorescence background subtraction treatment, and all the spectral data are subjected to standardization treatment, so that the maximum values and the minimum values of the intensities of all the spectra are the same.

In the method for constructing and identifying the multi-component complex oil sample fingerprint spectrum, in the step (1), the attenuation sheet is set to be 0.1 percent or more than one of 1 percent, the objective lens is set to be 10 multiplied by 10 or 50 multiplied by 50, and the detection range is 50-6000cm^-1Or 200-^-1。

According to the method for constructing and identifying the multi-component complex oil sample fingerprint spectrum, when the two-dimensional spectrum data is resampled in the step (3), the data interval of the two-dimensional spectrum is 1-80cm^-1。

Preferably, the two-dimensional spectrum has a data interval of 20-40cm^-1。

In the method for constructing and identifying the multicomponent complex oil sample fingerprint spectrum, the stepwise discriminant method in the step (4) is selected from Wilk's lambda, unexplained variance, Mahalanobis distance, minimum F value or Rao's V.

Preferably, the stepwise decision method is Wilk's lambda method, the decision criterion is the value of the statistic F, and when F is greater than F_{Enter into}Then, the variable is retained; when the F value is less than F_DeletingIf so, rejecting the variable; wherein, F_{Enter into}Greater than F_Deleting，F_{Enter into}Is 5.84-1.84, F_Deleting4.71-0.71; more preferably, F_{Enter into}Is 3.84, F_DeletingWas 2.71.

The calculation of the Wilk's lambda and the statistic F is explained as follows:

is the sum of the squares of the total deviations;

is the square sum of the differences between groups;

as the sum of squared deviations within the group;

Wilk’s lambda＝SSE/SST；

MSA ═ SSA/(m-1), the mean squared difference between groups;

MSE is SSA/(n-m), which is the mean intra-group dispersion sum of squares;

F＝MSA/MSE；

wherein n is the total sample size, m is the level number of the control variable, and X_ikFor the k sample value at the i level, n_iTo control the amount of samples that vary the ith level,

is the average at the i-th level,

is the mean of the observations.

The invention has the beneficial effects that: the identification method of the invention adopts the two-position spectrum to obtain the sample characteristic information richer than the one-dimensional spectrum, thereby improving the identification precision; by adopting a resampling technology, the identification analysis based on the two-dimensional spectral information is simpler, more convenient and faster under the condition of keeping the main characteristic information of the sample; the two-dimensional spectral information is concentrated through low-dimensional space projection, so that the fingerprint characteristic difference among different samples can be more visual and obvious, and the identification degree can be improved; the identification method of the invention complements and verifies with the existing crude oil identification method, and improves the accuracy and persuasion of crude oil identification.

Drawings

FIG. 1. two-dimensional simultaneous correlated Raman spectra without resampling; a. standard sample _1#, b blind sample _1 #;

FIG. 2 shows two-dimensional simultaneous correlated Raman spectra after resampling (sampling step length of 5 cm)^-1) (ii) a a. Standard sample _1#, b blind sample _1 #;

FIG. 3 shows two-dimensional simultaneous correlated Raman spectrum after resampling (sampling step length is 10 cm)^-1) (ii) a a. Standard sample _1#, b blind sample _1 #;

FIG. 4 shows two-dimensional simultaneous correlated Raman spectra after resampling (sampling step size is 20 cm)^-1) (ii) a a. Standard sample _1#, b blind sample _1 #;

FIG. 5 shows two-dimensional simultaneous correlated Raman spectra after resampling (sampling step size is 40 cm)^-1) (ii) a a. Standard sample _1#, b blind sample _1 #;

FIG. 6. two-dimensional simultaneous correlated Raman spectra without resampling; a. standard sample _2#, b blind sample _2 #;

FIG. 7 shows two-dimensional simultaneous correlated Raman spectra after resampling (sampling step length of 5 cm)^-1) (ii) a a. Standard sample _2#, b blind sample _2 #;

FIG. 8 shows two-dimensional simultaneous correlated Raman spectra after resampling (sampling step size is 10 cm)^-1) (ii) a a. Standard sample _2#, b blind sample _2 #;

FIG. 9 shows two-dimensional simultaneous correlated Raman spectra after resampling (sampling step size is 20 cm)^-1) (ii) a a. Standard sample _2#, b blind sample _2 #;

FIG. 10 shows two-dimensional simultaneous correlated Raman spectra after resampling (sampling step size of 40 cm)^-1) (ii) a a. Standard sample _2#, b blind sample _2 #;

FIG. 11 is a barycenter stack diagram (F) of the projected coordinates of the fingerprint spectrum of the standard sample and the blind sample in the low-dimensional space_{Upper limit of}＝3.84，F_{Lower limit of}＝2.71)；

FIG. 12 is a projection diagram of the centroid of the fingerprint spectrum of the standard sample and the blind sample in the low-dimensional space coordinate plane (F)_{Upper limit of}＝3.84，F_{Lower limit of}＝2.71)。

Detailed Description

The following non-limiting examples are presented to enable those of ordinary skill in the art to more fully understand the present invention and are not intended to limit the invention in any way.

Example 1

This example uses 15 crude oil standards from the petrochemical sciences research institute of Beijing, petrochemical, China. Crude oil standards and blind samples are numbered sequentially from # 1 to # 15, respectively, as in Table 1.

TABLE 1 crude oil numbering and nomenclature

Serial number	Crude oil name
		1	Horse power cloth
2	Jenno
		3	Shiliya (Chinese character of' Shiliya
4	Kohler et al
		5	Mix in southern Xinjiang
6	Boni light
		7	Yunnan mixture
8	Southern Xinjiang-1
		9	Southern Xinjiang-2
10	Kango brand wine
		11	Lucky rope
12	Aman
		13	Weight of bus
14	Kelaowei (Kelaowei)
		15	Carren

The spectral fingerprint of each oil sample is collected by a Horiba JY XploRA microscopic confocal Raman spectrometer, and more than 4 parallel samples are collected. The detection range is 50-3200cm^-1The attenuator is set to 0.1%, the aperture and slit are 500 μm and 200 μm, respectively, the objective lens is 10 ×, the CCD detector temperature is-70 deg.C, the exposure time is 1s, and each spectrum is accumulated 30 times. The grating is 2400T when the excitation light is 532nm, the grating is 1200T when the excitation light is 785nm, the same test conditions are adopted for all samples, and the spectral data are not subjected to any fluorescence background subtraction treatment. Cubic interpolation processing is carried out on each spectral data, and the step length is 1cm^-1All the spectral data are aligned on the abscissa one by one, and then normalized, with the maximum value set to 100 and the minimum value set to 0. Then, the normalized data is adopted to calculate a two-dimensional correlation spectrum, the two-dimensional correlation spectrum is resampled, and the main characteristics of the fingerprint spectrum are keptUnder the condition (2), the data amount of the two-dimensional spectrum is reduced; and extracting an upper triangular array or a lower triangular array in the two-dimensional spectrum matrix after resampling, and converting the two-dimensional spectrum matrix into a one-dimensional array by using matlab according to the same rule.

FIG. 1 is a two-dimensional simultaneous correlation Raman spectrum of a standard sample _1# and a blind sample _1# without resampling; resampling the two-dimensional synchronous Raman spectrums of the standard sample _1# and the blind sample _1# with the step length of 5cm respectively^-1、10cm^-1、20cm^-1、40cm^-1。

FIG. 2 shows a resampling step size of 5cm^-1The two-dimensional synchronous correlated Raman spectrums of the standard sample 1# and the blind sample 1# are obtained, and it can be seen that compared with the fingerprint characteristic of the oil sample 1# in the figure 1, the fingerprint characteristic is well preserved.

FIG. 3 shows a resampling step size of 10cm^-1The two-dimensional synchronous correlated Raman spectrums of the standard sample 1# and the blind sample 1# are obtained, and it can be seen that compared with the fingerprint characteristic of the oil sample 1# in the figure 1, the fingerprint characteristic is well preserved.

FIG. 4 shows a resampling step size of 20cm^-1The two-dimensional synchronous correlated Raman spectrums of the standard sample 1# and the blind sample 1# are obtained, and it can be seen that compared with the fingerprint characteristic of the oil sample 1# in the figure 1, the fingerprint characteristic is well preserved.

FIG. 5 shows a resampling step size of 40cm^-1Two-dimensional synchronous correlated Raman spectrums of the standard sample 1# and the blind sample 1# are obtained, and it can be seen that compared with the chart 1, the main fingerprint characteristics of the 1# oil sample are well preserved.

FIG. 6 is a two-dimensional simultaneous correlation Raman spectrum of a standard sample _2# and a blind sample _2# without resampling; resampling the two-dimensional synchronous Raman spectrums of the standard sample _2# and the blind sample _2# with the step length of 5cm respectively^-1、10cm^-1、20cm^-1、40cm^-1。

FIG. 7 shows a resampling step size of 5cm^-1The two-dimensional synchronous correlated Raman spectrums of the standard sample 2# and the blind sample 2# are obtained, and it can be seen that compared with the fingerprint characteristic of the 2# oil sample in FIG. 6, the fingerprint characteristic of the 2# oil sample is well preserved.

FIG. 8 shows a resampling step size of 10cm^-1The two-dimensional synchronous correlated Raman spectra of the standard sample 2# and the blind sample 2# obtained in the time are shown to be the same as those in FIG. 6In comparison, the fingerprint characteristics of the No. 2 oil sample are well preserved.

FIG. 9 shows a resampling step size of 20cm^-1The two-dimensional synchronous correlated Raman spectrums of the standard sample 2# and the blind sample 2# are obtained, and it can be seen that compared with the fingerprint characteristic of the 2# oil sample in FIG. 6, the fingerprint characteristic is well preserved.

FIG. 10 shows a resampling step size of 40cm^-1Two-dimensional synchronous correlated Raman spectrums of the standard sample 2# and the blind sample 2# are obtained, and it can be seen that compared with the chart of FIG. 6, the main fingerprint characteristics of the 2# oil sample are well preserved.

Comparing fig. 1 and 6, fig. 2 and 7, fig. 3 and 8, fig. 4 and 9, and fig. 5 and 10, it can be seen that the difference between the 1# oil sample and the 2# oil sample can be still clearly reflected in the comparison of the fingerprint spectra after resampling. On the other hand, in the fingerprint spectrum after resampling, the number of data variables is obviously reduced, and the data processing workload of corresponding identification analysis can be greatly reduced.

With a resampling step length of 20cm^-1The synchronous relevant fingerprint spectrum data is removed through stepwise discrimination analysis, and the optimal variable group with strong discrimination capability is reserved; the discrimination method is Wilk's lambda method, the discrimination criterion is the value of statistic F, when the value of F is larger than F_{Enter into}When 3.84, the variable is retained; when the F value is less than F_DeletingWhen the value is 2.71, the variable is removed; and then using the sps software, constructing a dictionary rule discrimination function group according to the principle of 'minimum homogeneous dispersion and maximum inter-class dispersion', projecting the optimal variable group with strong discrimination capability to a low-dimensional space by adopting the dictionary rule discrimination function, and 'concentrating' the main characteristic difference of the oil sample to a plurality of dimensional variables with the most obvious cumulative variance contribution rate.

In a low-dimensional space, the variance contribution of the dimensional variable with the most significant variance contribution rate is shown in table 2, the cumulative variance contribution rate of the first three dimensional variables with the most significant variance contribution rate reaches 99.8%, and the feature difference between different types of samples can be more comprehensively reflected by using the coordinates of the first three dimensions.

TABLE 2 percent contribution of variance of dimensional variables for low dimensional space

Dimensional variable	Variance contribution%	Cumulative contribution%
			1	95.2	95.2
2	3.5	98.7
			3	1.1	99.8
4	0.2	100.0

The mean of the coordinates of all parallel samples of the standard sample of # 1-15 oil sample in the new three-dimensional space was taken as the barycentric coordinate of this standard sample. Similarly, the mean of the coordinates in the new three-dimensional space of all parallel samples of the blind sample of # 1-15 oil samples was taken as the barycentric coordinate of this type of blind sample. The front three-dimensional barycentric coordinates with the most significant variance contribution are plotted into a column stack chart, and the result is shown in fig. 11. In fig. 11, the characteristic difference between the 15 crude oil standard samples is obvious, and the accurate identification of the 15 blind samples is realized by directly comparing the fingerprint spectrums of the blind samples with the fingerprint spectrums of the standard samples.

For the more similar 7#, 8#, 9#, 12#, 13# in fig. 11, the projections of the barycenter of each of the standard sample and the blind sample on three two-dimensional coordinate planes are plotted. As shown in fig. 12, differences of 7#, 8#, 9#, 12#, and 13# can be clearly identified by visual comparison.

The fingerprint spectrum construction and identification method can be matched with other identification methods, and the identification efficiency, accuracy and persuasion are obviously improved. The novel fingerprint spectrum construction and identification method provided by the invention has wide application prospects in the fields of nondestructive identification of artworks, cultural relics, jewelry and criminal investigation material evidence, identification of producing areas of genuine medicinal materials and marine products, medical disease diagnosis and the like.

It will be apparent to those skilled in the art from this disclosure that many changes and modifications can be made, or equivalents modified, in the embodiments of the invention without departing from the scope of the invention. Therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention shall still fall within the protection scope of the technical solution of the present invention, unless the contents of the technical solution of the present invention are departed.

Claims

1. A multi-component complex oil sample fingerprint spectrum construction and identification method is characterized by comprising the following steps:

(5) according to the new projection coordinates in the low-dimensional space, drawing a columnar stack diagram for visual comparison or drawing a three-dimensional space projection diagram for fingerprint feature comparison, namely identifying a blind sample;

when the two-dimensional spectrum data is resampled in the step (3), the data interval of the two-dimensional spectrum is 1-80cm^-1。

2. The method for constructing and identifying the fingerprint spectrum of the multi-component complex oil sample according to claim 1, wherein the method comprises the following steps: and (5) visually comparing similar samples in the columnar stack diagram, and then drawing a three-dimensional space projection diagram fingerprint characteristic comparison.

3. The method for constructing and identifying the fingerprint spectrum of the multi-component complex oil sample according to claim 1, wherein the method comprises the following steps: the excitation light wavelength used in the step (1) is two or more of 532nm, 514.5nm, 488nm, 457.9nm, 638nm, 632.8nm and 785nm, all the spectral data are not subjected to any fluorescence background subtraction treatment, and all the spectral data are subjected to standardization treatment, so that the maximum values and the minimum values of the intensities of all the spectrograms are the same.

4. The method for constructing and identifying the fingerprint spectrum of the multi-component complex oil sample according to claim 1, wherein the method comprises the following steps: in the step (1), the attenuation sheet is set to be 0.1% or 1% or more, the objective lens is 10 x or 50 x, and the detection range is 50-6000cm^-1Or 200-^-1。

5. The method for constructing and identifying the fingerprint spectrum of the multi-component complex oil sample according to claim 1, wherein the method comprises the following steps: the data interval is 20-40cm^-1。

6. The method for constructing and identifying the fingerprint spectrum of the multi-component complex oil sample according to claim 1, wherein the method comprises the following steps: the stepwise discriminant method in step (4) is selected from Wilk's lambda, unexplained variance, Mahalanobis distance, minimum F-number or Rao's V.

7. The method for constructing and identifying the fingerprint spectrum of the multi-component complex oil sample according to claim 6, wherein the method comprises the following steps: the stepwise discrimination method is Wilk's lambda method, the discrimination criterion is the value of statistic F, when F is larger than F_{Enter into}Then, the variable is retained; when the F value is less than F_DeletingIf so, rejecting the variable; wherein, F_{Enter into}Greater than F_Deleting，F_{Enter into}Is 5.84-1.84, F_DeletingIs 4.71-0.71.

8. The method for constructing and identifying the fingerprint spectrum of the multi-component complex oil sample according to claim 7, wherein the method comprises the following steps: said F_{Enter into}Is 3.84, F_DeletingWas 2.71.