Method for rapid determination of naphtha monomer hydrocarbon distribution using near infrared spectroscopy

Through the combination of near-infrared spectroscopy and gas chromatography, a ratio library of hydrocarbon distribution of naphtha monomer was constructed. The problem of insufficient speed and detail of hydrocarbon distribution of naphtha monomer was solved by using virtual samples and correction models, and the problem of insufficient measurement speed and detail of naphtha monomer hydrocarbon distribution was achieved. Fast and accurate analysis of hydrocarbon distribution of naphtha monomers.

CN114858747BActive Publication Date: 2024-12-06CHINA PETROLEUM & CHEMICAL CORP +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202110155970.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-02-04
Publication Date
2024-12-06
Estimated Expiration
2041-02-04

AI Technical Summary

Technical Problem

Existing methods are difficult to quickly and accurately determine the monomer hydrocarbon distribution of naphtha, especially when the analysis speed and detail requirements are high during the refining process.

Method used

Near infrared spectroscopy combined with gas chromatography was used to construct a proportion library of monomer hydrocarbon distribution of naphtha, and virtual samples were generated, and a correction model was established using sample enhancement methods and multiple regression analysis to quickly determine the monomer hydrocarbon distribution of the sample to be tested.

Benefits of technology

It realizes the rapid and accurate determination of the monomer hydrocarbon distribution of naphtha, meets the needs of industrial online analysis, improves the determination speed and detail, and is suitable for the determination of almost all monomer hydrocarbon composition in naphtha samples.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114858747B_ABST
    Figure CN114858747B_ABST
Patent Text Reader

Abstract

The present invention proposes a method for rapidly determining the distribution of naphtha monomer hydrocarbons using near-infrared spectroscopy. The method is based on the distribution of group components and the content of monomer hydrocarbons of naphtha determined by a standard gas chromatography method, a monomer hydrocarbon distribution ratio library of naphtha is established, a large number of virtual samples are generated by a sample enhancement method, and a prediction model for the distribution of naphtha group components is constructed by using near-infrared spectroscopy as an input variable combined with a multivariate regression analysis method; the absorbance of the near-infrared spectrum of the sample to be tested in the characteristic spectral region is substituted into the prediction model for the distribution of group components to predict the content of each group component of the sample to be tested, and then the near-infrared spectrum of the sample to be tested is used to find similar samples in the mixed sample to calculate the distribution ratio of monomer hydrocarbons of the sample to be tested, and the content of each group component is multiplied by the distribution ratio of the corresponding monomer hydrocarbon to obtain the monomer hydrocarbon distribution result of the sample to be tested. The method of the present invention has a fast determination speed and can simultaneously determine almost all detailed monomer hydrocarbon compositions in the naphtha sample with high accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a method for rapidly measuring the distribution of naphtha monomer hydrocarbons by using near infrared spectroscopy. Background Art

[0002] In recent years, optimizing the refining process has become an important task in the oil processing industry. The main reasons are: (1) Crude oil is becoming increasingly heavy, while the market demand for light oil is increasing, and the conversion efficiency of heavy oil to light oil needs to be improved; (2) The country pays more and more attention to environmental protection and has higher and higher requirements, and needs to optimize the production process to save energy and reduce pollution emissions. How to optimize the refining process is closely related to the properties of its raw materials, and the physical and chemical properties of oil are determined by its molecular composition. Understanding the composition and properties of oil at the molecular level and in-depth research on the chemical reactions of each molecule in the refining process can more accurately understand the processing properties of oil and its various fractions and determine a more efficient processing route.

[0003] Naphtha is one of the main raw materials for ethylene cracking and catalytic reforming in my country. The most direct and accurate method to characterize its molecular composition is chromatographic analysis, which can qualitatively and quantitatively identify hundreds of monomer hydrocarbon molecules. However, the experimental cycle of chromatographic analysis is long, and it also requires some professionals to operate, which makes it difficult to meet the requirements of refineries for analysis speed in terms of optimization control. Near-infrared spectroscopy (NIR) has received more and more attention due to its fast measurement speed and ability to measure multiple components at the same time. In the 1990s, the method of combining near-infrared spectroscopy (NIR) with chemometrics has been proven to be able to determine the composition of naphtha well. However, so far, the composition predicted by the established model is not detailed enough, mostly group composition and several specific components, and has not yet been accurate to all the molecular compositions of naphtha. Establishing a method that can quickly determine the composition of naphtha monomer hydrocarbons has become a problem to be solved. Summary of the invention

[0004] The purpose of the present invention is to provide a method for rapidly determining the distribution of naphtha monomer hydrocarbons by using near infrared spectroscopy. The method of the present invention is simple to operate, fast in analysis speed, accurate in prediction, and suitable for industrial online analysis.

[0005] The method of the present invention for rapidly measuring the distribution of naphtha monomer hydrocarbons by near infrared spectroscopy comprises the following steps:

[0006] (1) collecting n actual naphtha samples, measuring the near infrared spectrum of each actual sample, and measuring the group component distribution and monomer hydrocarbon content of each actual sample by gas chromatography, where n is a positive integer;

[0007] (2) arranging the near infrared spectra and monomer hydrocarbon contents of the actual samples in step (1) into a matrix according to the wave number and the molecular type, respectively, calculating the distribution ratio of each monomer hydrocarbon in each actual sample, and constructing a naphtha monomer hydrocarbon distribution ratio library;

[0008] (3) according to the near infrared spectrum and group component distribution of the actual sample measured in step (1) and the monomer hydrocarbon distribution ratio calculated in step (2), a virtual sample is generated by using a sample enhancement method, the virtual sample including the near infrared spectrum, group component distribution and monomer hydrocarbon distribution ratio of the virtual sample and forming a mixed sample together with the actual sample collected in step (1);

[0009] (4) performing noise reduction processing on the spectrum of the characteristic spectrum region of the near infrared spectrum of the mixed sample described in step (3);

[0010] (5) Correlating the absorbance of the spectrum of the mixed sample after the noise reduction process in step (4) with the content of each group component in the group component distribution results of each sample, respectively, performing regression analysis, and establishing a calibration model between the near-infrared spectrum and the content of each group component;

[0011] (6) using the same method as step (1) to measure the near infrared spectrum of the sample to be tested and perform noise reduction on the spectrum, substituting the absorbance of the characteristic spectrum area into the calibration model of the content of each group component established in step (5) to obtain the content of each group component of the sample to be tested;

[0012] (7) Using the near infrared spectrum of the sample to be tested measured in step (6), find a similar sample in the mixed sample constructed in step (2), use the similar sample to calculate the monomer hydrocarbon distribution ratio of the sample to be tested, and multiply it by the group component content of each monomer hydrocarbon molecule obtained in step (6) to obtain the monomer hydrocarbon distribution of the sample to be tested.

[0013] According to the present invention, preferably, in step (1), the near infrared spectrum is measured at a temperature of 25 to 40°C and the spectrum collection range is 3500 to 10000 cm -1 .

[0014] According to the present invention, preferably, in step (1), the group component distribution and monomer hydrocarbon content of the actual sample are determined according to the national standard method for determining monomer hydrocarbon composition of naphtha (SH / T 0714).

[0015] According to the present invention, in step (1), n ​​is the number of actual samples. n is preferably a positive integer greater than or equal to 5, for example, a positive integer between 10 and 1000. When selecting actual samples, it is preferred to select actual naphtha samples with large sample differences, such as actual samples from different origins.

[0016] According to the present invention, in step (1), the group component distribution refers to the content of normal alkanes, isoalkanes, cycloalkanes and aromatic hydrocarbons classified according to carbon number. The normal alkanes are preferably normal alkanes with carbon numbers of 3 to 12, the isoalkanes are preferably isoalkanes with carbon numbers of 4 to 12, the cycloalkanes are preferably cycloalkanes with carbon numbers of 5 to 12, and the aromatic hydrocarbons are preferably aromatic hydrocarbons with carbon numbers of 6 to 11.

[0017] According to the present invention, in step (1), the monomer hydrocarbon content refers to the type and content of monomer hydrocarbon molecules contained in normal alkanes, isoalkanes, cycloalkanes and aromatic hydrocarbons, wherein the normal alkanes are preferably normal alkanes with a carbon number of 3 to 12, the isoalkanes are preferably isoalkanes with a carbon number of 4 to 12, the cycloalkanes are preferably cycloalkanes with a carbon number of 5 to 12, and the aromatic hydrocarbons are preferably aromatic hydrocarbons with a carbon number of 6 to 11.

[0018] According to the present invention, in step (2), the distribution ratio of each monomer hydrocarbon is obtained by dividing the content of each monomer hydrocarbon molecule by the group composition content of the corresponding carbon number in the group to which it belongs.

[0019] According to the present invention, optionally, in step (3), the enhancement method comprises: selecting m samples from the actual sample by a preset selection rule, combining the m samples by a preset combination rule to obtain the virtual sample, and the near-infrared spectrum, group component distribution and monomer hydrocarbon distribution ratio of each sample in the virtual sample are obtained by linearly adding the near-infrared spectrum, group component distribution and monomer hydrocarbon distribution ratio of the actual samples constituting the sample according to their combination ratio, wherein m is a positive integer greater than or equal to 2 and less than or equal to n, for example, it can be a positive integer between 2 and 1000.

[0020] According to the present invention, optionally, in step (3), the preset selection rule includes a random selection rule, a fixed order selection rule and a near-infrared spectrum similarity selection rule.

[0021] According to the present invention, optionally, in step (3), the near infrared spectrum similarity selection rule includes: by respectively calculating the similarity between the near infrared spectrum of the sample to be tested and the infrared spectrum of n actual samples, m actual samples with the greatest similarity are selected therefrom. The greater the similarity, the closer the infrared spectrum is to the actual sample.

[0022] According to the present invention, optionally, in step (3), the preset combination rule includes a fixed ratio sum rule and / or a random ratio sum rule.

[0023] According to the present invention, optionally, in step (3), the fixed ratio addition is to obtain a virtual sample by adding m samples in the actual sample at a fixed ratio, and the near-infrared spectrum of the obtained virtual sample is obtained by adding the near-infrared spectra of the actual samples constituting the sample at a fixed ratio and optionally adding noise, and the corresponding group component distribution and monomer hydrocarbon distribution ratio are obtained by adding the group component distribution and monomer hydrocarbon distribution ratio of the actual samples constituting the sample at the same fixed ratio. For example, a virtual sample is obtained at a ratio of 1:1 for every two actual samples, and the near-infrared spectrum of the virtual sample is obtained by adding the near-infrared spectra of the two actual samples constituting the sample in a ratio of 1:1 and optionally adding noise. The group component distribution and monomer hydrocarbon distribution ratio of the virtual sample are obtained by adding the group component distribution and monomer hydrocarbon distribution ratio of the two actual samples constituting the sample in a ratio of 1:1. For another example, a virtual sample is obtained at a ratio of 1:1:1 for every three actual samples, and the near-infrared spectrum of the virtual sample is obtained by adding the near-infrared spectra of the three actual samples constituting the sample in a ratio of 1:1:1 and optionally adding noise. The group component distribution and monomer hydrocarbon distribution ratio of the virtual sample are obtained by adding the group component distribution and monomer hydrocarbon distribution ratio of the three actual samples constituting the sample in a ratio of 1:1:1.

[0024] According to the present invention, optionally, in step (3), the random proportion addition is a virtual sample obtained by adding m samples in the actual sample according to the random proportion addition rule, and the near-infrared spectrum of the obtained virtual sample is obtained by adding the near-infrared spectra of the actual samples constituting the sample according to their combined proportions and optionally adding noise, and the corresponding family component distribution and monomer hydrocarbon distribution ratio are obtained by adding the family component distribution and monomer hydrocarbon distribution ratio of the actual samples constituting the sample according to the same proportions. For example, a virtual sample is obtained by adding the near-infrared spectrum of the virtual sample according to the random proportions of the two actual samples constituting the sample according to their combined proportions and optionally adding noise, and the family component distribution and monomer hydrocarbon distribution ratio of the virtual sample are obtained by adding the family component distribution and monomer hydrocarbon distribution ratio of the two actual samples constituting the sample according to their combined proportions.

[0025] According to the present invention, optionally, in step (3), the noise includes difference spectrum and / or Gaussian noise.

[0026] According to the present invention, optionally, in step (3), the difference spectrum refers to the difference spectrum of the near-infrared spectra of the same actual sample measured twice. For example, the near-infrared spectra of the same sample measured twice are xk1 and xk2 respectively, then R = xk1-xk2 or R = xk2-xk1 is the difference spectrum. When adding the difference spectrum, m' actual samples can be randomly selected from the m actual samples in step (2), and the m' actual samples are tested twice to obtain m' difference spectra. In the process of generating each virtual sample, one difference spectrum can be randomly selected for summation.

[0027] According to the present invention, optionally, in step (3), the Gaussian noise refers to a type of noise whose probability density function obeys a normal distribution. The calculation formula of the normal distribution function is as follows:

[0028]

[0029] Where Z is the independent variable of the distribution, μ is the expectation of the distribution, and σ^2 is the variance of the distribution.

[0030] According to the present invention, optionally, the expectation of the Gaussian noise is 0, and the variance range is 0.0010 to 0.0020, preferably 0.0015.

[0031] According to the present invention, optionally, in step (3), the process of using the difference spectrum as noise to generate the spectrum of the virtual sample according to the preset combination rule for m actual samples is as follows:

[0032] The spectra of the m actual samples that make up the virtual sample are x1, x2, …, xm, and the difference spectra are R1, R2, …, Rm. Then the spectrum of the generated virtual sample is X = a1*x1+a2*x2+ …+am*xm+R, where a1, a2, …, am are the combination coefficients of each sample in the spectrum of the generated virtual sample, and R is a difference spectrum randomly selected from R1, R2, …, Rm. The R in the spectra of different virtual samples can be the same or different.

[0033] According to the present invention, optionally, in step (3), the steps of generating virtual samples by using a preset random selection rule and a preset combination rule are as follows:

[0034] The near-infrared spectra of n actual samples are x1, x2, x3, …, xn, the corresponding group compositions are y1, y2, y3, …, yn, and the noise is z, which can be the difference spectrum R or Gaussian noise G. R is a difference spectrum randomly selected from the difference spectra of the near-infrared spectra of n actual samples. The R in the spectra of different virtual samples can be the same or different; randomly select m samples from the above n actual samples and combine them according to a preset combination rule to obtain N virtual samples. N is a positive integer, generally a positive integer greater than or equal to 100, such as a positive integer between 100 and 5000; the near-infrared spectrum X of the virtual sample = a1*x1 + a2*x2 + … + am*xm + z, where z is the noise, and the corresponding group composition Y = a1*y1 + a2*y2 + … + am*ym, where a1, a2, a3, …, am (a1 + a2 + a3 + … + am = 1, m < n) are the combination coefficients of each sample in the virtual samples generated according to the preset combination rule.

[0035] According to the present invention, optionally, in step (3), virtual samples are generated by a preset near-infrared spectrum similarity selection rule. The similarity selection rule is to calculate the similarity between the spectrum of the sample to be measured and the spectrum of each of the n actual samples. The m samples with the largest similarity can be selected as similar samples, and the remaining n - m samples are non-similar samples. m1 (2 ≤ m1 ≤ m, and m1 is a positive integer) samples can be randomly selected from the m samples and n1 virtual samples are generated according to a preset combination rule; or m2 (2 ≤ m2 ≤ m, and m2 is a positive integer) similar samples can be selected from the m similar samples and m3 (2 ≤ m3 ≤ m, and m3 is a positive integer) non-similar samples can be selected from the n - m non-similar samples and n2 virtual samples are generated according to a preset combination rule; or the above n1 + n2 virtual samples can be selected together. n1 and n2 are both positive integers, for example, they can each independently be a positive integer between 100 and 5000.

[0036] According to the present invention, optionally, in step (3), similar samples are found by calculating the similarity between the sample to be measured and the actual samples. The calculation method of the similarity is realized by calculating one or more of the Euclidean distance, Chebyshev distance, Manhattan distance, Mahalanobis distance, and cosine of the included angle. The method of calculating the Euclidean distance is preferably used. The smaller the value of the Euclidean distance, Chebyshev distance, Manhattan distance, Mahalanobis distance, or cosine of the included angle, the greater the similarity, that is, the sample to be measured is most similar to some of the actual samples or virtual samples.

[0037] According to the present invention, optionally, in step (3), the process of selecting similar samples by using the near-infrared spectrum similarity selection rule is as follows:

[0038] (a) Determine the near infrared spectrum of the sample to be tested, and determine the similarity between the near infrared spectrum of the sample to be tested and the near infrared spectrum of each sample in the actual sample by calculating the Euclidean distance.

[0039] The calculation formula of the Euclidean distance is as follows:

[0040]

[0041] where (x1 x2 … x n )、(y1 y2 … y n ) are two points in n-dimensional space.

[0042] (b) Arrange the samples in ascending order according to the size of the Euclidean distance and select the first m samples with the smallest Euclidean distance, which are the similar samples.

[0043] According to the present invention, optionally, in step (3), the steps of generating virtual samples by using the preset near-infrared spectrum similarity selection rule and the random proportion addition combination rule are as follows:

[0044] Suppose the near infrared spectrum of the sample to be tested is xi, and m similar samples are found among n actual samples, then the number of non-similar samples is h = nm, and N virtual samples are obtained by random proportion addition. The spectra of similar samples are xa1, xa2, ..., xam, and the spectra of non-similar samples are xb1, xb2, ..., xbh. The corresponding family composition of similar samples is ya1, ya2, ... yam, and the corresponding family composition of non-similar samples is yb1, yb2, ..., ybh;

[0045] Then the spectrum of the virtual sample is:

[0046] X=a1*xa1+a2*xa2+…+am2*xam2+b1*xb1+b2*xb2+…+bm3*xbm3+z, where z is noise.

[0047] The corresponding groups are:

[0048] Y=a1*ya1+a2*ya2+…+am2*yam2+b1*yb1+b2*yb2+…+bm3*ybm3;

[0049] Among them, a1, a2, …, am2, b1, b2, …, bm3 (a1+a2+ …+am2+b1+b2+ …+bm3=1) are the combination coefficients of each sample in the virtual sample generated by the random proportion addition combination rule, and are randomly generated.

[0050] According to the present invention, in step (4), the characteristic spectrum region is preferably 4000 to 4800 cm -1and / or 5200~6250cm -1 Spectrum area.

[0051] According to the present invention, in step (4), in order to reduce or eliminate the influence of information and noise irrelevant to the spectral data, the spectrum is subjected to denoising processing, and the denoising processing method includes second-order differential processing and / or mean centering processing. Preferably, the spectrum is subjected to second-order differential processing and then mean centering processing.

[0052] According to the present invention, in step (5), the absorbance of the spectrum after noise reduction is correlated with the content of each group component in the sample group component distribution result determined by gas chromatography, and regression analysis is performed to establish a calibration model between the near infrared spectrum and the content of each group component. The calibration model established is a calibration model for each group component content in the group component distribution result, and each calibration model can predict the corresponding group component content.

[0053] According to the present invention, in step (5), the regression analysis used to establish the correction model includes one or more of partial least squares (PLS), least squares (CLS), inverse least squares (ILS), principal component regression (PCR) and robust partial least squares (RPLS), preferably partial least squares (PLS).

[0054] According to the present invention, the process of establishing the correction model using the partial least squares method is as follows:

[0055] When the PLS method is used to establish a correction model, the modeling is based on the Langbai-Beer law:

[0056] Y=XB+E,

[0057] In the formula,

[0058] Y—is an absorbance matrix (m×n") consisting of m samples and n” wavelength points;

[0059] X—is the (m×1) concentration vector consisting of m samples and 1 component;

[0060] B—is a sensitivity vector consisting of 1 component and n” wavelength points (1×n”);

[0061] E—is the m×n” residual matrix of absorbance.

[0062] The number of wavelength points n" is the number of information recorded by the spectrometer within the characteristic spectrum interval.

[0063] According to the present invention, optionally, the algorithm for establishing the correction model is as follows:

[0064] (a) The absorbance matrix Y (obtained by second-order differential processing of the absorbance in the characteristic spectral region) and each element of the concentration vector X are mean-centered, that is, all data are subtracted from the average value of the corresponding data group (column vector).

[0065] (b) The processed absorbance matrix Y and concentration vector X are decomposed into principal components:

[0066]

[0067]

[0068] t k (m×1)—factor score of absorbance matrix, y-score;

[0069] v k (1×n”)—factor loading of absorbance matrix, y-loading;

[0070] r k (m×1)—factor score of concentration vector, x-score;

[0071] q k (1×1)—factor loading of concentration vector, x-loading;

[0072] f—number of principal factors;

[0073] E Y — is the absorbance matrix residual;

[0074] E X — is the concentration vector residual;

[0075] In order to ensure that T obtained from Y can have a good linear relationship with R obtained from X, we can introduce information about R when Y is decomposed into T, or introduce information about T when X is decomposed into R. This can be achieved by exchanging iteration variables during iteration, that is, combining the above two decomposition processes into one, that is,

[0076] r k =b k t k (3)

[0077] b k (1×1)—r k and t k The regression coefficient of .

[0078] (c) Solve for the eigenvector and the number of principal factors f

[0079] Ignoring the residual matrix E, according to equations (1) and (2), when f = 1, we have:

[0080] Y=tv t

[0081] Multiply t on the left t Get: v t =t t Y / t t t

[0082] Multiply the right side by v and we get: t = Yv / v t v

[0083] X=rq

[0084] Multiply left by r t So: q = r t X / r t r, divide both sides by q to get: r = X / q

[0085] The detailed process of calculating all the main factors of the model and determining the number of main factors to obtain the calibration model includes the following steps:

[0086] [1] Find the weight vector w of the absorbance matrix;

[0087] Take a column of the concentration matrix X as the starting iteration value of r, replace t with r, and calculate w;

[0088] The calculation equation is:

[0089] Y=rw t The solution is: w T =r T Y / r t r

[0090] [2] Normalize the weight vector w;

[0091] w T =w T / ||w T ||

[0092] [3] Find the factor score t of the absorbance matrix and calculate t from the normalized w;

[0093] The calculation equation is:

[0094] Y=tw t The solution is: t = Yw / w t w

[0095] [4] Find the load u value of the concentration vector and use t instead of r to calculate u;

[0096] The calculation equation is:

[0097] X=tu T The solution is: T =t t X / t t t

[0098] [5] Normalize the load u;

[0099] u T =u T / ||u T ||

[0100] [6] Find the factor score r of the concentration vector, given by u T Calculate r;

[0101] The calculation equation is:

[0102] X=ru T The solution is: r = Xu / u T u

[0103] Then replace t with this r and return to step [1] to calculate w T , by w t Calculate t 新 , and repeat the iterations until t converges (‖t 新 -t 旧 ‖≤10 -6 ‖t 新 ‖), continue with the next operation, otherwise return to step [1].

[0104] [7] Calculate the load vector v of the absorbance matrix from the converged t;

[0105] The calculation equation is:

[0106] Y=tv t The solution is: v t =t t Y / t t t

[0107] [8] Calculate the load q value of the concentration vector from r;

[0108] The calculation equation is:

[0109] The solution of X = rq is: q = r t X / r t r

[0110] From this, we can find r1,q1,t1,v1 corresponding to the first principal factor.

[0111] [9] Substitute into formula (3) to obtain b1: b1 = r1t1 / t1t1

[0112]

[10] Calculate the residual matrix E by formulas (1) and (2);

[0113] E X,1 =X-r1q1=X-b1t1q1 (4)

[0114] E Y,1 =Y-t1v1 (5)

[0115] E X,1 Instead of X, E Y,1 Substitute Y and return to step [1] to calculate r2, q2, t2, v2, b2 of the next component, and calculate E according to equations (4) and (5): X,2 、E Y,2 , and so on, find all the principal factors of X and Y.

[0116]

[11] The number of principal factors f was determined by cross-validation method, and the correction model of absorbance and concentration matrix was obtained. The steps of cross-validation are as follows:

[0117] Under a certain principal factor f, among the n samples in the calibration set, select n' samples as the validation set of the model, and the remaining n-n' samples as the training set to establish the calibration model. The samples in the validation set predict the established model, and the predicted values ​​of these samples are calculated and recorded. This process continues until all samples have been predicted once and only once, and the calibration standard deviation SEC of the model under the principal factor number f is calculated. The calculation formula for the calibration standard deviation is as follows:

[0118]

[0119] y i,actual is the measured value of the i-th sample, y i,predicted is the predicted value of the ith sample calibration model, and n is the number of samples in the calibration sample set. Check the calibration standard deviation value under each principal component, and select the principal component number with a small calibration standard deviation value, or the principal component number f when the calibration standard deviation value no longer decreases.

[0120] According to the present invention, in step (6), the near-infrared spectrum of the sample to be tested is measured under the same conditions as in step (1), and the spectrum of its characteristic spectrum region is subjected to noise reduction processing according to the method in step (4) to obtain the absorbance of the sample to be tested, and the absorbance of the sample to be tested is substituted into the calibration model established in step (5) to predict the content of each group component of the sample to be tested, thereby obtaining the content of each group component constituting the sample to be tested.

[0121] According to the present invention, in step (6), the characteristic spectrum region is preferably 4000-4800 cm -1 and / or 5200-6250cm -1 Spectrum area.

[0122] According to the present invention, in step (6), the noise reduction processing method includes second-order differential processing and / or mean centering processing, and preferably, the spectrum is subjected to second-order differential processing and then to mean centering processing.

[0123] According to the present invention, in step (6), the near infrared spectrum of the sample to be tested is measured under the same conditions as in step (1), and the spectrum of the characteristic spectrum region is subjected to noise reduction processing according to the method in step (4) to obtain the absorbance y of the sample to be tested. 待测 , according to the calibration model established in step (5), predict the content x of a certain group component in the sample to be tested 待测 , the calculation process is as follows:

[0124] (a) By y 待测 And the v stored during the calibration process k Calculate t k(待测) ;

[0125] t k(待测) =y 待测 v k / v k v k

[0126] (b) From the obtained t k(待测) and b stored during the calibration process k Calculate r k(待测) ;

[0127] r k(待测) =b k t k(待测)

[0128] (c) From the obtained r k(待测) and q stored during the calibration process k , use formula (2) to calculate x 待测 .

[0129] For a single sample, after all the group component contents are predicted, the group component distribution of the sample is obtained by summarizing the results of the group component contents.

[0130] According to the present invention, in step (7), the near-infrared spectrum of the sample to be tested is used to calculate the similarity of the near-infrared spectrum of each sample in the mixed sample constructed in step (2) to obtain a similar sample, and the similar sample is used to calculate the monomer hydrocarbon distribution ratio of the sample to be tested, and multiplied by the group component content of the corresponding carbon number in the group of each monomer hydrocarbon molecule obtained in step (6) to obtain the monomer hydrocarbon distribution of the sample to be tested.

[0131] According to the present invention, in step (7), the similar samples are found by calculating the similarity between the spectrum of the sample to be tested and the spectrum of each sample in the mixed sample, and selecting the sample with the greatest similarity. The similarity can be one or more of Euclidean distance, Chebyshev distance, Manhattan distance, Mahalanobis distance and angle cosine.

[0132] According to the present invention, optionally, in step (7), the method of using similar samples to calculate the monomer hydrocarbon distribution ratio of the sample to be tested includes: using the near-infrared spectrum of the similar sample to regress the near-infrared spectrum of the sample to be tested to obtain a regression coefficient, and then multiplying the corresponding monomer hydrocarbon distribution ratio of the similar sample by the corresponding regression coefficient to calculate the monomer hydrocarbon distribution ratio of the sample to be tested.

[0133] According to the present invention, optionally, in step (7), the method of finding similar samples by calculating the Euclidean distance and calculating the monomer hydrocarbon distribution of the sample to be tested includes:

[0134] (a) calculating the Euclidean distance between the near infrared spectrum of the sample to be tested and the near infrared spectrum of each sample in the mixed sample to determine the similarity between the sample to be tested and the mixed sample, and arranging them in ascending order according to the size of the Euclidean distance;

[0135] The calculation formula of the Euclidean distance is as follows:

[0136]

[0137] where (x1 x2 … x n )、(y1 y2 … y n ) are two points in n-dimensional space.

[0138] (b) Select the first k samples with the smallest Euclidean distance, and use the near-infrared spectra of these k samples to regress the near-infrared spectrum of the sample to be tested; the calculation formula is as follows:

[0139] Y 待测 =a1Y1+a2Y2+…+a k Y k ; a1, a2, …, a k ≥0; a1+a2+…+a k =1

[0140] where Y 待测 is the near infrared spectrum of the sample to be tested, Y1, Y2, …, Y k are the near-infrared spectra of k selected samples with small Euclidean distances, a1, a2, …, a k is the coefficient in the regression equation.

[0141] The near-infrared spectra of k similar samples constitute the independent variable Y, as shown below:

[0142] Y=(Y1 Y2 Y3 .... Y k )

[0143] a1,a2,…,a k The dependent variable a is composed as follows:

[0144] a=(a1 a2 a3 .... a k )

[0145] (c) Regression coefficients a1, a2, …, a k The calculation formula is as follows:

[0146] a=Y T Y 待测 / (Y T Y)

[0147] (d) The monomer hydrocarbon distribution ratio of the k samples is multiplied by the same coefficient in the above regression equation to regress the monomer hydrocarbon distribution ratio of the sample to be tested. The calculation formula is as follows:

[0148] X 待测 =a1X1+a2X2+…+a k X k ; a1, a2, …, a k ≥0

[0149] Where X 待测 is the monomer hydrocarbon distribution ratio of the sample to be tested, X k is the monomer hydrocarbon distribution ratio of the selected k samples,

[0150] The independent variable X composed of the distribution ratio of monomer hydrocarbons in similar samples is:

[0151] X=(X1 X2 X3 … X k )

[0152] (e) Calculate the monomer hydrocarbon distribution ratio X of the sample to be tested 待测

[0153] X 待测 =aX

[0154] (f) Multiplying each monomer hydrocarbon distribution ratio of the sample to be tested by its corresponding group component content to obtain the monomer hydrocarbon distribution of the sample to be tested; the calculation formula is as follows:

[0155] A 待测 =Xa 待测 *xb 待测

[0156] Among them A 待测 is a monomer hydrocarbon molecule of the sample to be tested, Xa 待测 is the distribution ratio of the monomer hydrocarbon molecule in the sample to be tested, xb 待测 It is the group composition content of the carbon number contained in the monomer hydrocarbon molecule of the sample to be tested.

[0157] The method of the present invention is based on the group component distribution and monomer hydrocarbon content of naphtha determined by the standard method of gas chromatography, and establishes a monomer hydrocarbon distribution ratio library of naphtha; a large number of virtual samples are generated by using the near infrared spectrum, group composition distribution and monomer hydrocarbon distribution ratio of the actual sample using the sample enhancement method, and the actual sample and the virtual sample constitute a mixed sample; the absorbance of the near infrared spectrum of the mixed sample in the characteristic spectrum region is used as the input variable, and the group composition distribution of the mixed sample is used as the output variable, and a prediction model of the group component distribution of naphtha is constructed in combination with a multivariate regression analysis method; the absorbance of the near infrared spectrum of the sample to be tested in the characteristic spectrum region is substituted into the prediction model of the group component distribution to predict the content of each group component of the sample to be tested, and then the near infrared spectrum of the sample to be tested is used to find similar samples in the mixed sample to calculate the monomer hydrocarbon distribution ratio of the sample to be tested, and the content of each group component is multiplied by the distribution ratio of the corresponding monomer hydrocarbon to obtain the monomer hydrocarbon distribution result of the sample to be tested. Compared with the traditional method for determining the monomer hydrocarbon composition of naphtha, the method of the present invention has a fast determination speed, and can simultaneously determine almost all detailed monomer hydrocarbon compositions in the naphtha sample, with high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0158] Figure 1 NIR spectrum of an actual naphtha sample collected.

[0159] Figure 2 This is a comparison chart of the predicted values ​​of the PLS model established by selecting all actual samples and generating virtual samples according to the random proportion combination rule in Example 1 and the gas chromatography analysis method (Figure A is a comparison chart of C7 normal alkanes, Figure B is a comparison chart of C9 isoalkanes, Figure C is a comparison chart of C7 cycloalkanes, and Figure D is a comparison chart of C8 aromatics).

[0160] Figure 3 The following is a comparison chart of the predicted values ​​of the PLS model established by selecting 2 actual samples and generating virtual samples according to the random proportion combination rule in Example 1 and the gas chromatography analysis method (Figure A is a comparison chart of C7 normal alkanes, Figure B is a comparison chart of C9 isoalkanes, Figure C is a comparison chart of C7 cycloalkanes, and Figure D is a comparison chart of C8 aromatics).

[0161] Figure 4 This is a comparison chart of the predicted values ​​of the PLS model established by selecting 2 actual samples and generating virtual samples according to the fixed ratio combination rule in Example 1 and the gas chromatography analysis method (Figure A is a comparison chart of C7 normal alkanes, Figure B is a comparison chart of C9 isoalkanes, Figure C is a comparison chart of C7 cycloalkanes, and Figure D is a comparison chart of C8 aromatics).

[0162] Figure 5For the example in Example 1, two samples are selected using the near-infrared spectral similarity selection rule, and the predicted values ​​of the PLS model established by the random proportion combination rule to generate virtual samples are compared with the gas chromatography analysis method (Figure A is a comparison chart of C7 normal alkanes, Figure B is a comparison chart of C9 isoalkanes, Figure C is a comparison chart of C7 cycloalkanes, and Figure D is a comparison chart of C8 aromatics). DETAILED DESCRIPTION

[0163] The present invention is further described in detail below by examples, but the present invention is not limited thereto.

[0164] The method of the present invention collects actual naphtha samples, measures the near-infrared spectrum of the actual samples, and measures the group component distribution and monomer hydrocarbon content of the actual samples by gas chromatography; the near-infrared spectrum and monomer hydrocarbon content of the measured actual samples are arranged into a matrix according to wave number and molecular type, the distribution ratio of each monomer hydrocarbon is calculated, and a naphtha monomer hydrocarbon distribution ratio library is constructed; a large number of virtual samples are generated by using a sample enhancement method based on the near-infrared spectrum, group component distribution and monomer hydrocarbon distribution ratio of the measured actual samples, and a mixed sample is formed together with the collected actual samples; the near-infrared spectrum of the mixed sample is taken at 4000-4800cm -1 and 5200-6250cm -1 The spectral interval is the characteristic spectral region, and the spectrum of the characteristic spectral region is subjected to second-order differential processing. The absorbance of the processed spectrum is respectively associated with the content of each group component in the group component distribution result of the mixed sample, and regression analysis is performed respectively to establish a correction model for the content of each group component; the near-infrared spectrum of the naphtha sample to be tested is measured under the same conditions as those for establishing the correction model, and the same spectrum processing is performed. The content of each group component of the sample to be tested is predicted by each correction model based on the absorbance of the characteristic spectral region; the near-infrared spectrum of the sample to be tested is used to calculate the similarity with each sample in the mixed sample, the monomer hydrocarbon distribution ratio of the similar sample is selected to calculate the monomer hydrocarbon distribution ratio of the sample to be tested, and it is multiplied by the corresponding group component content obtained by each correction model to obtain the monomer hydrocarbon distribution result of the sample to be tested.

[0165] Example 1

[0166] (1) 41 actual naphtha samples were collected, their near-infrared spectra were measured, and the group component distribution and monomer hydrocarbon content of the samples were determined by gas chromatography.

[0167] The instrument used to measure the near-infrared spectrum of the sample is an Antaris II near-infrared spectrometer (Thermo Fisher) equipped with a temperature control module. The specific operation is as follows: fill the cuvette to about two-thirds of the naphtha sample and seal it with a sealing film. Place the sealed cuvette into a temperature-controlled sample pool rack for transmission spectrum collection. The spectrum collection temperature is 25°C and the collection range is 3500-10000cm -1 , repeated scanning 128 times, with a resolution of 8cm -1 A total of 41 samples were collected. The near infrared spectrum of an actual naphtha sample collected is shown in Figure 1 shown.

[0168] Gas chromatography is a method for detecting the composition of naphtha monomer hydrocarbons by capillary gas chromatography in accordance with the national standard ST / H0714. Gas chromatography is used to obtain C3~C 12 The distribution of various group components and the content of each monomer hydrocarbon.

[0169] (2) Construction of Monomer Hydrocarbon Distribution Ratio Library

[0170] The near-infrared spectra and monomer hydrocarbon contents of the 41 naphtha samples collected were organized into a matrix according to the wavenumber and molecular type, which included a total of 233 monomer hydrocarbon molecules. The monomer hydrocarbon distribution ratio in each sample was calculated, that is, the single monomer hydrocarbon content / corresponding group component content, and a naphtha monomer hydrocarbon distribution ratio library was constructed.

[0171] (3) Sample enhancement method generates virtual samples

[0172] The spectral data was processed and the naphtha samples were divided into a calibration sample set (31 samples) and a prediction sample set (10 samples). The calibration sample set was used for virtual sample generation and model building, and the prediction sample set was not used for model building and was used to test the accuracy of the model. In the process of generating virtual samples, noise was added when calculating the near-infrared spectrum of the virtual samples. The noise was a difference spectrum, which was randomly selected from 3 samples in the calibration sample set. Each sample was measured twice to obtain 3 difference spectra. One of them was randomly selected for summation in the process of generating each virtual sample.

[0173] The methods for generating virtual samples include the following four sample enhancement methods:

[0174] Method 1: All calibration samples are combined in random proportions to generate 930 virtual samples.

[0175] Method 2: Randomly select 2 actual samples from the calibration sample set and combine them in random proportions to generate 930 virtual samples;

[0176] Method 3: Randomly select 2 actual samples from the calibration sample set and combine them in a ratio of 1:2 and 2:1 to generate 930 virtual samples;

[0177] Method 4: Determine the near-infrared spectrum of the sample to be tested, calculate the Euclidean distance between the near-infrared spectrum of the sample to be tested and the spectrum of each sample in the calibration sample set, find 5 similar samples, and then randomly select 2 samples from these 5 similar samples and combine them in random proportions to generate 465 virtual samples. At the same time, randomly select 1 sample from these 5 similar samples and 1 sample from the other 26 non-similar samples and combine them in random proportions to generate 465 virtual samples, generating a total of 930 virtual samples.

[0178] The reason for selecting the difference spectrum as noise is that the noise of the actual instrument is irregular, and the spectrum of the virtual sample generated by using the difference spectrum is closer to the actual situation.

[0179] (4) Establishment of partial least squares model

[0180] The actual samples and 961 mixed samples generated by the above methods 1, 2, 3, and 4 were respectively analyzed in the characteristic interval 4000-4800cm -1 and 5200-6250cm -1 The near-infrared spectrum is subjected to second-order differential processing, and then mean-centered processing is performed to obtain the absorbance matrix Y. The content of a certain group component (such as the content of C6 normal alkane component) corresponding to each sample spectrum in the Y matrix is ​​composed of a concentration vector X; the absorbance matrix Y and the concentration vector X are associated with PLS to establish a correction model to obtain the correction model of the group component content (the content of C6 normal alkane component). According to the above method, the content of different group components corresponding to each sample spectrum is composed of different concentration vectors X, and the correction model of the content of each group component can be obtained.

[0181] (5) Evaluation of partial least squares model

[0182] The 10 actual samples in the prediction sample set are used to predict the distribution of family components. The 10 samples in the prediction sample set are in the characteristic interval of 4000-4800cm -1 and 5200~6250cm -1 The near-infrared spectrum is subjected to second-order differential processing and then mean-centered processing to obtain the absorbance y of each sample to be measured. The absorbance is substituted into the calibration model to obtain the predicted value of the content of each group component of the sample, that is, the group component distribution of the sample.

[0183] The model is evaluated by calculating the prediction standard deviation (SEP) of the samples in the prediction sample set and the coefficient correlation coefficient (R) between the measured value and the predicted value. The smaller the prediction standard deviation (SEP) and the closer the correlation coefficient (R) is to 1, the better the prediction effect of the model. The calculation formula of the prediction standard deviation (SEP) is as follows:

[0184]

[0185] where y i,actual is the gas chromatography measured value of a certain group of components of the i-th sample, y i,predicted is the predicted value of the correction model for the corresponding family component of the i-th sample, and m is the number of samples in the prediction set.

[0186] The calculation formula of correlation coefficient (R) is as follows:

[0187]

[0188] where y i,actual is the gas chromatography measured value of a certain group of components of the i-th sample, is the average value of the gas chromatography measured value of a certain family component of all samples in the prediction set, y i,predicted is the predicted value of the correction model for the corresponding family component of the i-th sample, and m is the number of samples in the prediction set.

[0189] Table 1 is a comparison of the prediction results of the PLS models built before and after sample enhancement. The PLS model built before sample enhancement is a PLS model obtained without sample enhancement, which can be used as a comparative example of the method of the present invention; the PLS model built after sample enhancement is a PLS model obtained after sample enhancement, which can be used as an embodiment of the method of the present invention. Table 1 lists the prediction standard deviation (SEP) when each model is established for prediction and the correlation coefficient (R) between the measured value and the predicted value of the prediction set. It can be observed from Table 1 that the prediction standard deviation (SEP) of the model built after sample enhancement is smaller than the prediction standard deviation (SEP) of the PLS model built before sample enhancement. At the same time, it can be seen from Table 1 that the correlation coefficient R value of the model prediction before and after sample enhancement has increased. This shows that sample enhancement can improve the technical effect of the constructed PLS model. Figure 2-Figure 5 Shown are comparison graphs of the predicted values ​​of the PLS model established using different sample enhancement methods and the GC analysis method (Figure A is a comparison graph of C7 normal alkanes, Figure B is a comparison graph of C9 isoalkanes, Figure C is a comparison graph of C7 cycloalkanes, and Figure D is a comparison graph of C8 aromatics). These graphs show that the method of using the sample enhancement method to generate virtual samples to participate in the establishment of the PLS model is accurate and reliable.

[0190] (6) Prediction of Monomer Hydrocarbon Distribution Ratio

[0191] When predicting the sample to be tested, the near infrared spectrum of the sample to be tested is used to calculate the Euclidean distance between the near infrared spectrum of the sample to be tested and the near infrared spectrum of each sample in the 961 mixed samples, and the first k samples with the smallest Euclidean distance are selected, and the spectra of the k similar samples are used to regress the spectrum of the sample to be tested; the obtained regression coefficient is used as a coefficient to multiply the monomer hydrocarbon distribution ratio of the corresponding similar sample, that is, the monomer hydrocarbon distribution ratio of the sample to be tested is obtained. In this embodiment, one similar sample is selected to calculate the spectrum of the sample to be tested.

[0192] Table 2 shows the prediction standard deviation (SEP) and correlation coefficient (R) of the monomer hydrocarbon molecule ratio of some prediction set samples before and after sample enhancement. The predicted results before sample enhancement are the prediction results of the monomer hydrocarbon distribution ratio obtained without sample enhancement, which can be used as a comparative example of the method of the present invention; the predicted results after sample enhancement are the prediction results of the monomer hydrocarbon distribution ratio obtained after sample enhancement, which can be used as an embodiment of the method of the present invention. Table 3 shows the prediction results of the monomer hydrocarbon distribution of naphtha. Since there are 233 kinds of monomer hydrocarbon molecules contained in each sample, it is impossible to list them all in the table. Therefore, the prediction results of the monomer hydrocarbon distribution ratio of some prediction set samples greater than 30% and less than 100% are listed in Table 3. It can be seen from Tables 2 and 3 that the prediction standard deviation (SEP) using the sample enhancement method is smaller than the prediction standard deviation (SEP) without sample enhancement, and the predicted correlation coefficient R value is improved, indicating that the accuracy of the method of the present invention is relatively high.

[0193] In reality, collecting naphtha with chromatographic data is very time-consuming and labor-intensive. The four sample enhancement methods in this embodiment save the cost of collecting samples in reality and increase the density of samples. It can be seen from Tables 1, 2 and 3 that the four sample enhancement methods improve the prediction effect of naphtha group composition distribution and monomer hydrocarbon distribution ratio, and ultimately improve the accuracy of naphtha monomer hydrocarbon distribution prediction results.

[0194] Table 1

[0195]

[0196] Table 2

[0197]

[0198] Table 3 (values ​​are percentages)

[0199]

Claims

1. A method for rapidly determining the distribution of naphtha monomer hydrocarbons using near infrared spectroscopy, comprising the following steps: (1) collecting n actual naphtha samples, measuring the near infrared spectrum of each actual sample, and measuring the group component distribution and monomer hydrocarbon content of each actual sample by gas chromatography, where n is a positive integer; (2) The near infrared spectra and monomer hydrocarbon contents of the actual samples in step (1) are arranged into a matrix according to the wave number and the molecular type, and the distribution ratio of each monomer hydrocarbon in each actual sample is calculated to construct a naphtha monomer hydrocarbon distribution ratio library; (3) generating a virtual sample using a sample enhancement method according to the near infrared spectrum, group component distribution and monomer hydrocarbon distribution ratio of the actual sample measured in step (1) and the monomer hydrocarbon distribution ratio calculated in step (2), including the near infrared spectrum, group component distribution and monomer hydrocarbon distribution ratio of the virtual sample, and forming a mixed sample together with the actual sample collected in step (1); (4) performing noise reduction processing on the spectrum of the characteristic spectrum region of the near infrared spectrum of the mixed sample described in step (3); (5) Correlating the absorbance of the spectrum of the mixed sample after the noise reduction process in step (4) with the content of each group component in the group component distribution results of each sample, respectively, performing regression analysis, and establishing a calibration model between the near-infrared spectrum and the content of each group component; (6) Using the same method as step (1), measure the near infrared spectrum of the sample to be tested and perform noise reduction on the spectrum, substitute the absorbance of the characteristic spectrum area into the calibration model of the content of each group component established in step (5), and obtain the content of each group component of the sample to be tested; (7) Using the near infrared spectrum of the sample to be tested measured in step (6), find a similar sample in the mixed sample constructed in step (2), use the similar sample to calculate the monomer hydrocarbon distribution ratio of the sample to be tested, and multiply it by the group component content of each monomer hydrocarbon molecule obtained in step (6) to obtain the monomer hydrocarbon distribution of the sample to be tested.

2. The method according to claim 1, characterized in that In step (1), the near infrared spectrum is measured at a temperature of 25 to 40°C and the spectrum acquisition range is 3500 to 10000 cm -1 .

3. The method according to claim 1, characterized in that In step (1), the group component distribution and monomer hydrocarbon content of the actual sample are determined according to the national standard SH / T 0714 naphtha monomer hydrocarbon composition determination method.

4. The method according to claim 1, characterized in that In step (3), the enhancement method comprises: selecting m samples from the actual sample according to a preset selection rule, combining the m samples according to a preset combination rule to obtain the virtual sample, and obtaining the near-infrared spectrum, group component distribution and monomer hydrocarbon distribution ratio of each sample in the virtual sample by linearly adding the near-infrared spectrum, group component distribution and monomer hydrocarbon distribution ratio of the actual samples constituting the sample according to their combination ratio, wherein m is a positive integer greater than or equal to 2 and less than or equal to n.

5. The method according to claim 4, characterized in that In step (3), the preset selection rules include a random selection rule, a fixed order selection rule and a near infrared spectrum similarity selection rule.

6. The method according to claim 5, characterized in that In step (3), the near-infrared spectrum similarity selection rule includes: calculating the similarity between the near-infrared spectrum of the sample to be tested and the near-infrared spectra of n actual samples respectively, and selecting m actual samples with the greatest similarity.

7. The method according to claim 4, characterized in that In step (3), the preset combination rule includes a fixed ratio addition rule and / or a random ratio addition rule.

8. The method according to claim 7, characterized in that In step (3), the fixed ratio addition is to obtain a virtual sample by adding m samples in the actual sample at a fixed ratio, and the near-infrared spectrum of the obtained virtual sample is obtained by adding the near-infrared spectra of the actual samples constituting the sample at a fixed ratio and optionally adding noise, and the corresponding group component distribution and monomer hydrocarbon distribution ratio are obtained by adding the group component distribution and monomer hydrocarbon distribution ratio of the actual samples constituting the sample at the same fixed ratio; the random ratio addition is to obtain a virtual sample by adding m samples in the actual sample at a random ratio, and the near-infrared spectrum of the obtained virtual sample is obtained by adding the near-infrared spectra of the actual samples constituting the sample at their combined ratio and optionally adding noise, and the corresponding group component distribution and monomer hydrocarbon distribution ratio are obtained by adding the group component distribution and monomer hydrocarbon distribution ratio of the actual samples constituting the sample at the same ratio.

9. The method according to claim 8, characterized in that In step (3), the noise includes difference spectrum and / or Gaussian noise.

10. The method according to claim 5, characterized in that In step (3), a virtual sample is generated by a preset near-infrared spectrum similarity selection rule, wherein the similarity selection rule is to calculate the similarity between the spectrum of the sample to be tested and the spectrum of each sample in the n actual samples, and select the m samples with the largest similarity as similar samples, and the remaining nm samples as non-similar samples; Randomly select m1 samples from m samples and generate n1 virtual samples through a preset combination rule, where 2≤m1≤m, and m1 is a positive integer; Alternatively, m2 similar samples are selected from the m similar samples and m3 non-similar samples are selected from the nm non-similar samples to generate n2 virtual samples through a preset combination rule, where 2≤m2≤m, and m2 is a positive integer, and 2≤m3≤m, and m3 is a positive integer; Alternatively, the n1+n2 virtual samples are selected together; Both n1 and n2 are positive integers.

11. The method according to claim 10, characterized in that In step (3), similar samples are found by respectively calculating the similarity between the sample to be tested and the actual sample. The similarity is calculated by calculating one or more of the Euclidean distance, Chebyshev distance, Manhattan distance, Mahalanobis distance and angle cosine method.

12. The method according to claim 1, characterized in that In step (4), the characteristic spectrum range is 4000-4800 cm -1 and / or 5200~6250cm -1 Spectrum area.

13. The method according to claim 1, characterized in that In step (4), the spectrum is subjected to denoising processing, wherein the denoising processing method includes second-order differential processing and / or mean centering processing.

14. The method according to claim 1, characterized in that In step (5), the absorbance of the spectrum after noise reduction is correlated with the content of each group of components in the sample group component distribution results determined by gas chromatography, and regression analysis is performed to establish a calibration model for the near-infrared spectrum and the content of each group of components.

15. The method according to claim 1, characterized in that In step (5), the regression analysis used to establish the correction model includes one or more of partial least squares method, least squares method, inverse least squares method, principal component regression and robust partial least squares method.

16. The method according to claim 1, characterized in that In step (6), the near infrared spectrum of the sample to be tested is measured under the same conditions as in step (1), and the spectrum of its characteristic spectrum region is subjected to noise reduction processing according to the method in step (4) to obtain the absorbance of the sample to be tested. The absorbance of the sample to be tested is substituted into the calibration model established in step (5) to predict the content of each group component of the sample to be tested, thereby obtaining the content of each group component constituting the sample to be tested.

17. The method according to claim 1, characterized in that In step (7), the near-infrared spectrum of the sample to be tested is used to calculate the similarity of the near-infrared spectrum of each sample in the mixed sample constructed in step (2) to obtain a similar sample. The similar sample is used to calculate the monomer hydrocarbon distribution ratio of the sample to be tested, and multiplied by the group component content of the corresponding carbon number in the group of each monomer hydrocarbon molecule obtained in step (6) to obtain the monomer hydrocarbon distribution of the sample to be tested.

18. The method according to claim 1, characterized in that In step (7), the similar samples are found by calculating the similarity between the spectrum of the sample to be tested and the spectrum of each sample in the mixed sample, and selecting the sample with the greatest similarity.

19. The method according to claim 1, characterized in that In step (7), the method of using similar samples to calculate the monomer hydrocarbon distribution ratio of the sample to be tested includes: using the near-infrared spectrum of the similar sample to regress the near-infrared spectrum of the sample to be tested to obtain a regression coefficient, and then multiplying the corresponding monomer hydrocarbon distribution ratio of the similar sample by the corresponding regression coefficient to calculate the monomer hydrocarbon distribution ratio of the sample to be tested.