A method for assisting design of tobacco leaf group formula based on virtual mixed near infrared spectrum

Abstract

The application provides a cigarette leaf group formula aided design method based on virtual mixed near-infrared spectroscopy. The method greatly shortens the time for substituting leaf formula of a target flavor cigarette product based on virtual leaf group mixing technology of near-infrared spectroscopy, multivariate correction model and spectral standardization model, reduces the evaluation and smoking workload of a formulator, realizes the intelligentization of a leaf formula technology, and effectively maintains the stability of tobacco product quality.

Description

Technical Field

[0001] This invention relates to a method for designing cigarette leaf blend formulations, specifically a method for auxiliary design of cigarette leaf blend formulations based on virtual hybrid near-infrared spectroscopy, belonging to the field of cigarette product formulation design. Background Technology

[0002] In recent years, with the expansion of cigarette production scale and the development of computer technology, in order to make rational and efficient use of resources, more and more tobacco companies hope to use modern advanced technologies to assist in the design of cigarette formulas. They have gradually begun to pay attention to the research and development and practice of digital technologies for tobacco raw materials, formulas and designs, and are constantly exploring ways to upgrade formula design from experience-based design mode to scientific formula design mode. Digital leaf blend formula design has become a control technology for stabilizing product quality and maintaining the consistency of style and flavor, which is conducive to improving the efficiency of formula research and development and maintenance.

[0003] Near-infrared spectroscopy contains a wealth of information on the components and content of tobacco leaves. If advanced data mining methods can be used to extract as much characteristic information as possible from this data, and then by establishing corresponding mathematical models and using chemometric methods to find and infer suitable substitutes, it will be of great significance for understanding the style characteristics and quality types of tobacco leaves from different producing areas and for guiding formulation design.

[0004] The current research status of applying near-infrared spectroscopy to assist cigarette formulation design and tobacco raw material substitution can be summarized into three aspects: (1) constructing an expert system to assist cigarette formulation based on qualitative and quantitative analysis of tobacco characteristics based on near-infrared spectroscopy [References 1-4]; (2) evaluating the substitution of formulation tobacco based on near-infrared spectroscopy or its characteristic projection similarity measurement [References 5-11]; (3) guiding formulation work through correlation analysis between near-infrared spectroscopy and cigarette quality or sensory evaluation [References 12-14].References: [1] Zhang Jianping, Chen Jianghua, Shu Ruxin, et al. Preliminary exploration of near-infrared information for tobacco leaf style identification and cigarette formulation research [J]. Journal of Tobacco Science, 2007, 13(5): 5-9. [2] Wang Yihui. Preliminary study on near-infrared rapid detection technology to assist cigarette formulation design [D]. Chinese Academy of Agricultural Sciences, 2011. [3] Zhang Yajuan, Ma Xiang, Zhang Yehui, et al. Application of near-infrared diffuse reflectance linear summation spectrum in tobacco leaf re-drying formulation [J]. Spectroscopy and Spectral Analysis, 2011, 31(2): 390-393. [4] Yang Kai, Cai Jiayue, Zhang Chaoping, et al. Analysis of part characteristics of tobacco leaves by near-infrared spectral projection model method [J]. Spectroscopy and Spectral Analysis, 201 4,34(12):2764-2768. [5] Wang Jiajun, Li Juan. SIMCA modeling based on FT-NIR analysis technology and its application in quality monitoring of cigarette formulation process [J]. Tobacco Science and Technology, 2008, 3:6-10. [6] Wen Yadong, Wang Yi, Wang Nengru, et al. Application of projection analysis method of near-infrared spectroscopy in industrial grading and re-drying module formulation [J]. China Tobacco Science, 2009, 15(5):6-10. [7] Tao Shuai, Ma Xiang, Li Junhui, et al. Study on projection method based on principal component data of near-infrared spectroscopy of tobacco leaves and its application in re-drying formulation [J]. Spectroscopy and Spectral Analysis, 2009, 29(11):2970-2974. [8] Qin Yandong, Zhang Yaohua, Xiong Bin, et al. A method for computer-aided design of tobacco leaf blend formulation using near-infrared spectroscopy [P]. Hubei China Tobacco Industry Co., Ltd., 2009. [9] Yu Chunxia, ​​Ma Xiang, Zhang Yehui, et al. Similarity analysis of tobacco leaf parts based on near-infrared spectroscopy and SIMCA algorithm [J]. Spectroscopy and Spectral Analysis, 2011, 31(4):924-927.

[10] Mi Jinrui, Ma Xiang, Zhang Yajuan, et al. Analysis of the upper limit of tobacco leaf formulation ratio based on near-infrared spectral projection and Monte Carlo method [J]. Spectroscopy and Spectral Analysis, 2011, 31(4):915-919.

[11] Zhang Feng, Chen Xiaoming, Chen Qun, et al. A method for computer-aided design of tobacco leaf blend formulation based on near-infrared spectral information. A method for assisting cigarette formulation [P]. Fujian China Tobacco Industry Co., Ltd., 2013.

[12] Li Xueying, Shu Ruxin, Luan Lili, et al. Analysis of the composition ratio of mixed samples by near-infrared spectroscopy combined with non-negative regression coefficient regression method (formulation regression) [J]. Spectroscopy and Spectral Analysis, 2016, 36(4):967-971.

[13] Hao Xianwei, Tie Jinxin, He Wenmiao, et al. Simulation and substitution of Brazilian tobacco style based on near-infrared spectroscopy-sensory evaluation [J]. Tobacco Science and Technology, 2018, 51(10):89-95.

[14] Bi Yiming, Tian Yunong, Hao Xianwei, et al. A method for evaluating the overall sensory quality of tobacco leaves based on near-infrared spectroscopy, Zhejiang China Tobacco Industry Co., Ltd., 2019.

[0005] However, the tobacco formulation design work reported so far requires the preparation of a large number of tobacco samples, the collection of a large amount of spectral information, and also relies on a large number of evaluation experiments. It has the disadvantages of large workload, low efficiency, strong reliance on experience, and difficulty in promotion. Summary of the Invention

[0006] Based on the above background technology, the purpose of this invention is to provide a method for assisting in the design of cigarette leaf blends based on virtual mixed near-infrared spectroscopy. This method fits a large number of spectra with different mixing ratios through spectral standardization, eliminating the process of preparing mixed samples and testing spectra, shortening the time required for tobacco leaf blend development, reducing the workload of formulators in sensory evaluation, and avoiding the subjectivity and instability of the sensory evaluation process. This realizes the intelligentization of tobacco leaf blend technology and effectively maintains the stability of tobacco product quality.

[0007] To achieve the above technical objectives, this invention provides a method for auxiliary design of cigarette leaf blend formulations based on virtual hybrid near-infrared spectroscopy, the method comprising the following steps:

[0008] 1) Establish a calibration model between the near-infrared spectra of tobacco powder samples of different grades and the content of important chemical components contained in the samples; the calibration model y i =f i (x); where x is the near-infrared spectrum of each sample, and y i Let i represent the important chemical components, i = 1, 2, ..., m, where m is the number of important chemical components;

[0009] 2) Measure the near-infrared spectrum of each grade of tobacco powder sample in the inventory, and construct a near-infrared spectral data matrix of the grade of tobacco powder samples; the matrix S = [s1; ...; s i ;…;s n ], where s i , i = 1, 2, ..., n; s i The near-infrared spectra of each grade of tobacco powder sample in the inventory are shown, where n is the number of grades of tobacco leaves in the inventory.

[0010] 3) Prepare mixed sample sets consisting of different grades of tobacco powder mixed at different mass percentages, and measure their near-infrared spectral data matrices; the corresponding grade percentage mixing matrix of the mixed sample sets is C. K×n Where k is the number of mixed sample sets, and n is the number of tobacco powder grades; the near-infrared spectral data matrix X 实测 = [x1; ...; x i ;…;x K ], where x i Here is the near-infrared spectrum of the i-th mixed sample;

[0011] 4) Calculate the theoretical spectral data matrix X of the mixed sample set using the percentage mixing matrix and the near-infrared spectral data matrix S of n single-grade tobacco powders. 计算 ;

[0012] 5) In X 实测 and X 计算 Establish a spectral normalization model x between them 预测 =g(x 计算 );

[0013] 6) In X 实测 X 计算 And existing correction model y i =f i Based on (x), a predictive model y is established to predict the content of important chemical components in actual mixed tobacco samples from theoretically calculated spectral data of mixed tobacco samples. i =h i (x 计算 );

[0014] 7) For any given set of tobacco leaf formulations c = [c1, ..., c2], ... i c n ], where c i To determine the percentage content of the i-th single-grade tobacco leaf in the blended sample, first calculate the theoretical near-infrared spectral data x of this virtual blended tobacco leaf sample. 计算 = c × S, where S is the near-infrared spectrum of a single-grade tobacco powder sample, and then using the x 实测 =g(x 计算 ) and y i =h i (x 计算 The content of important chemical components y in the virtual blended tobacco sample was predicted. 预测 = [y1, ..., y i , ..., y m ], and the corresponding actual near-infrared spectrum x 实际 Finally, calculate y. 预测 and x 实际 The content of key chemical components in the target flavor cigarette product is respectively compared with y 目标 and its near-infrared spectrum x 目标 The correlation coefficient r between them y and r x If both values ​​are large, the established tobacco blend tends to meet the requirements, and then it is verified through a smoking evaluation test; if both values ​​are small, another tobacco blend is established and r is calculated. y and r x value.

[0015] As a preferred embodiment, the single-grade tobacco powder sample undergoes a process before near-infrared spectroscopy measurement. screen and drying process.

[0016] As a preferred embodiment, the mixed sample set is first processed by analyzing the individual grade tobacco powder samples before measuring the near-infrared spectrum. screen After the tobacco powder samples of each grade are mixed, they are dried.

[0017] As a preferred embodiment, the sieving is performed through a 60-mesh sieve; the drying is performed at 40-50°C for 5-7 hours.

[0018] As a preferred option, the correction model is a multivariate linear correction model constructed using the partial least squares method.

[0019] As a preferred embodiment, the theoretical spectral data matrix X of the tobacco powder mixture sample set 计算 From the percentage mixing matrix C K×n Multiplying the near-infrared spectral data matrix S of a single-grade tobacco powder sample yields: X 计算 =C K×n ×S.

[0020] As a preferred option, in X 实测 and X 计算 Establish a spectral normalization model x between them 预测 =g(x 计算 The process is as follows:

[0021] First define X 组合 =[X 实测 X 计算 ], then X can be used 组合 Perform the following singular value decomposition:

[0022] Among them, T s =U s ∑ s ;P s =V s ; 's' and 'n' represent the corresponding factors representing spectral information and noise, respectively; the superscript '′' indicates the matrix transpose operation; arbitrarily assuming that the number of spectrally active chemical components in the tobacco powder mixture sample is r, then T s and P s Both have r columns; P′1 and P′2 are P′ s submatrix P′ s = [P′1, P′2], whose column numbers are respectively equal to X 实测 and X 计算 The number of columns, and X实测 and X 计算 The number of columns is the same, and is equal to the number of wavelength points in the measured near-infrared spectrum; after obtaining P′1 and P′2, the measured near-infrared spectrum X of any tobacco powder mixture sample is... 实测 The corresponding theoretical calculation of the near-infrared spectrum X 计算 Relationship x 预测 =g(x 计算 ) represents the following: x 预测 =x 计算 (P′2) + P′1+x 计算 -x 计算 (P′2) + P′2, where the superscript '+' denotes the Moore-Penrose generalized inverse operation of the matrix.

[0023] As a preferred option, in X 实测 X 计算 And existing correction model y i =f i Based on (x), a predictive model y is established to predict the content of important chemical components in actual mixed tobacco samples from theoretically calculated spectral data of mixed tobacco samples. i =h i (x 计算 The process is as follows: First, for X... 计算 Perform singular value decomposition, X 计算 =T s,计算 P′ s,计算 +E; then the spectral data of any mixed tobacco sample can be calculated theoretically. 计算 The predictive model y predicts the content of important chemical components in the actual blended tobacco sample. i = i (x 计算 This can be specifically represented as follows:

[0024]

[0025] As a preferred approach, a virtual leaf blend formulation screening experiment is conducted using a computer. An arbitrarily set single-leaf tobacco blend formulation c = [c1, ..., c2] is used. i c n Then, the theoretical near-infrared spectral data x of the virtual mixed tobacco sample were calculated. 计算 = c × S.

[0026] As a preferred approach, determining whether a virtual blend of tobacco samples meets the target flavor of the cigarette product is based on x. 预测 =g(x 计算 ) and y i =hi (x 计算 The content of important chemical components in the virtual blended tobacco sample predicted by the two models (y) 预测 = [y1, ..., y i , ..., y m ] and the corresponding near-infrared spectrum x 预测 The content of key chemical components in the target flavor cigarette product is respectively compared with y 目标 and its near-infrared spectrum x 目标 The correlation coefficient r between them y and r x The results are based on the magnitude of the values; if both values ​​are large, the virtual tobacco formulation is likely to meet the requirements and will be verified through a smoking test.

[0027]

[0028]

[0029] The tobacco leaf blend formulation-assisted design method based on virtual hybrid near-infrared spectroscopy provided by this invention includes the following specific steps:

[0030] (1) Near-infrared spectra (x) and some important chemical components (y) of tobacco powder samples of different varieties and grades i A calibration model y is established between the contents of (i = 1, 2, ..., m). i =f i (x), the correction model is a multivariate linear correction model constructed using the partial least squares regression (PLSR) method.

[0031] (2) Every single-grade tobacco powder sample in the inventory must pass through a 60-mesh sieve. screen Furthermore, it needs to be dried in a 45℃ oven for 6 hours and cooled to room temperature before measuring its near-infrared spectrum (S). i (i = 1, 2, ..., n; n is the number of varieties or grades of tobacco leaves in the inventory), construct a near-infrared spectral data matrix S (S = [s1; ...; s2]) for single-grade tobacco leaf powder samples. i ;…;s n ]).

[0032] (3) When preparing mixed samples of tobacco powder of different grades, each single grade of tobacco powder sample needs to be passed through a 60-mesh sieve, then dried in an oven at 45℃ for 6 hours, cooled to room temperature, weighed and mixed in proportion to prepare K mixed sample sets of tobacco powder of different grades mixed in different mass percentages (the corresponding variety or grade percentage mixing matrix is ​​C). K×nThe prepared tobacco powder mixture sample needs to be dried in an oven at 45℃ for 6 hours. After cooling to room temperature, its near-infrared spectrum is measured to obtain the near-infrared spectral data matrix X. 实测 (X 实测 = [x1; ...; x i ;…;x K ], x i (The near-infrared spectrum of the i-th mixed sample).

[0033] (4) Using the percentage mixing matrix C K×n The theoretical spectral data matrix X of the mixed sample set is calculated from the near-infrared spectral data matrix S of n single-grade tobacco powders. 计算 Theoretical spectral data matrix X of the mixed sample set of tobacco powder 计算 Through the percentage mixing matrix C K×n X is obtained by multiplying the near-infrared spectral data matrix S of a single-grade tobacco powder sample. 计算 =C K×n ×S.

[0034] (5) In X 实测 and X 计算 Establish a spectral normalization model x between them 预测 =g(x 计算 First, define X. 组合 =[X 实测 X 计算 ], then X can be used 组合 Perform the following singular value decomposition. Among them, T s =U s ∑ s ;P s =V s ; 's' and 'n' represent the corresponding factors representing spectral information and noise, respectively; the superscript '′' indicates the matrix transpose operation. Let r be the number of spectrally active chemical components in the tobacco powder mixture sample, then T s and P s Both have r columns; P′1 and P′2 are P′ s submatrix (P′) s = [P′1, P′2]), whose column numbers are respectively equal to X 实测 and X 计算 The number of columns (Note: X in this invention) 实测 and X 计算 The number of columns is the same, equal to the number of wavelength points in the measured near-infrared spectrum; after obtaining P′1 and P′2, the measured near-infrared spectrum X of any tobacco powder mixture sample is... 实测 The corresponding theoretical calculation of the near-infrared spectrum X 计算 Relationship x预测 =g(x 计算 This can be specifically represented as follows: x 预测 =x 计算 (P′2) + P′1+x 计算 -x 计算 (P′2) + P′2; where the superscript '+' indicates the Moore-Penrose generalized inverse operation of the matrix.

[0035] (6) In X 实测 X 计算 And existing correction model y i =f i Based on (x), a predictive model y is established to predict the content of important chemical components in actual mixed tobacco samples from theoretically calculated spectral data of mixed tobacco samples. i =h i (x 计算 The specific calculation process is as follows: First, calculate X... 计算 Perform singular value decomposition, X 计算 =T s,计算 P′ s,计算 +E(3); the spectral data of any mixed tobacco sample can be calculated theoretically. 计算 The predictive model y predicts the content of important chemical components in the actual blended tobacco sample. i =h i (x 计算 This can be specifically represented as follows:

[0036] (7) For any given set of tobacco leaf formulations c (c = [c1, ..., c2], ... i c n ];c i Let c be the percentage content of the i-th single-grade tobacco leaf in the mixed sample. A virtual leaf group formulation screening experiment is conducted using a computer. An arbitrarily set single-leaf group tobacco formulation c (c = [c1, ..., c2]) is used. i c n Then, the theoretical near-infrared spectral data x of the virtual mixed tobacco sample were calculated. 计算 (x 计算 =c×S), determining whether a virtual blended tobacco sample meets the requirements of a substitute tobacco formula for a specified flavor brand of cigarettes is based on x 预测 =g(x 计算 ) and y i =h i (x 计算 The two models predict the content of important chemical components in the virtual blended tobacco sample (y).预测 = [y1, ..., y i , ..., y m ]) and the corresponding near-infrared spectrum x 预测 The content of key chemical components in the target flavor brand cigarette products is compared with that of y. 目标 and its near-infrared spectrum x 目标 The correlation coefficient r between them y and r x The magnitude of the values ​​is considered. If both values ​​are large, the virtual tobacco blend has a high probability of meeting the actual blend requirements. It is recommended to formulate the actual leaf blend according to this mixing ratio and conduct a smoking test.

[0037]

[0038] Compared with existing technologies, the beneficial technical effects of the present invention are as follows:

[0039] This invention is the first to propose a virtual leaf blending technology based on near-infrared spectroscopy, multivariate correction models, and spectral standardization models to shorten the time required for tobacco formulation development, reduce the workload of formulators in evaluation, realize the intelligentization of tobacco formulation technology, and effectively maintain the stability of tobacco product quality.

[0040] The present invention first requires the establishment of a multivariate linear calibration model between the near-infrared spectra of tobacco powder samples of different grades and the contents of some important chemical components in the samples; these important chemical components include, but are not limited to, reducing sugars, total sugars, total alkaloids, potassium, starch, total nitrogen, chlorine, sugar-nitrogen ratio, nitrogen-alkaloid ratio, sugar-alkaloid ratio, etc.

[0041] This invention requires the preparation of K mixed sample sets, which are composed of different grades of tobacco powder mixed in different mass percentages; in order to ensure that this invention has good performance and at the same time minimize the amount of experimental work, K is generally no greater than 50.

[0042] In conducting virtual leaf group formulation experiments, this invention allows for the setting of leaf group formulation c (c = [c1, ..., c2]) by incorporating the formulator's experience. i c n This improves the efficiency of virtual recipe experiments.

[0043] The method of this invention is applicable to situations where there is a shortage of raw materials for the leaf blend formula in the production of target flavor brand cigarette products, and it is necessary to use existing grades of tobacco leaves in stock to design a substitute leaf blend formula, which can effectively maintain the stability of tobacco product quality.

[0044] In summary, the technical solution of this invention overcomes the shortcomings of traditional leaf blend formulation methods, which rely entirely on the experience of formulators and require extensive smoking tests. It can assist formulators in quickly designing alternative leaf blend formulations for producing target flavor brand cigarette products using existing varieties or grades of tobacco leaves in stock, thereby maintaining the stability of tobacco product quality. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating the technical solution of the present invention;

[0046] Figure 2 The measured spectrum (a) of the mixed sample of tobacco powder and the theoretical spectrum (b) calculated using the binding percentage of pure single-origin tobacco powder are illustrated.

[0047] Figure 3 The graph shows the correlation coefficients between the measured and theoretical spectra before and after spectral standardization. Detailed Implementation

[0048] The following examples are intended to further illustrate the present invention, rather than to limit the scope of the claims.

[0049] This embodiment uses data accumulated by Hunan Tobacco Industry Co., Ltd. in the field of near-infrared spectroscopy of tobacco leaves in recent years to briefly illustrate the specific implementation process and steps of the present invention in the area of ​​digital formulation of tobacco leaves. After years of accumulation, Hunan Tobacco Industry Co., Ltd. has obtained near-infrared spectral data X of 330 standard tobacco powder samples measured on a Bruker MPA near-infrared spectrometer. 标样 (Wavelength range: 4000nm~10000cm) -1 Wavelength spacing: 4cm -1 ;X 标样 = [x1; x2; ...; x i ;…;x 330 ], where the row vector x i The near-infrared spectrum of the i-th tobacco powder standard sample, and the content data matrix Y of 10 important chemical components in the standard sample set measured by an authoritative institution using the corresponding national standard method. 标样 (Y 标样 =[y1,y2,…,y i ,…,y 10 <; where the column vector y iThe content of the i-th chemical component in the standard sample is given. These 10 important chemical components are: reducing sugar, total sugar, total alkaloids, potassium, starch, total nitrogen, chlorine, sugar-nitrogen ratio, nitrogen-alkaloid ratio, and sugar-alkaloid ratio. Furthermore, the inventors determined the pure near-infrared spectral data matrix S (S = [s1; s2; s3; s4; s5; s6; s7; s8; s9]) of nine grades of tobacco powder samples; where the row vector s... i (Near-infrared spectrum of tobacco powder sample of grade i), and 502 samples of tobacco powder of the nine grades mixed by percentage, matrix C. 502×9 (Note: This matrix can be designed and determined using the simplex mixture experimental design method. The percentage mixing matrix C for a mixture of nine pure grades of tobacco powder, designed using the simplex mixture experimental design method, has 511 rows and 9 columns. After deleting the samples representing the nine pure grades of tobacco powder, 502 mixture samples remain.) Near-infrared spectral data matrix X of the mixed tobacco powder samples. 实测 (X 实测 = [x1; x2; ...; x i ;…;x 511 ]; where the row vector x i (This refers to the near-infrared spectrum of the i-th tobacco powder mixture sample). Based on this, the main implementation process and steps of this experiment are as follows:

[0050] 1) First, the 330 standard samples were divided into a calibration sample set (247 standard samples) and a validation sample set (83 standard samples). Then, partial least squares regression (PLSR) was used to establish 10 calibration models (yi, yi, yi) between the spectral data of the calibration set samples and the 10 chemical components in the calibration set samples. i =f i (x), i = 1, 2, ..., 10). The number of PLS ​​latent variables used in each calibration model is determined by examining the prediction error of the calibration model on the validation set samples.

[0051] 2) Using the percentage mixing matrix C 502×9 The theoretical spectral data matrix X of the mixed sample set was calculated from the near-infrared spectral data matrix S of nine grades of tobacco powder. 计算 ;

[0052] 3) In the case of X 实测 The first 41 rows of the spectral matrix X 实测 (1:41,:) and by X 计算 The first 41 rows of the spectral matrix X 计算 Establish a spectral normalization model x between (1:41,:) 预测 =g(x 计算 Then, the established spectral normalization model is used to determine the spectral normalization of X. 计算The spectral matrix X, consisting of the last 461 rows 计算 (42:502,:) is standardized to X 计算→实测 (42:502,:), calculate X 计算→实测 (42:502,:) and X 实测 (42:502,:) Correlation coefficients between corresponding rows and X 计算 (42:502,:) and X 实测 (42:502,:) The correlation coefficient r between the corresponding rows (r=[r1,r2,…,r 461 ]), and compare r * And the size of the corresponding element in r. If r * If most elements in x are greater than the corresponding elements in r, then the spectral normalization model x is... 预测 =g(x 计算 It has made significant contributions to virtual leaf group formulation technology.

[0053] 4) In X 实测 (1:41,:)X 计算 (1:41,:) and the existing calibration model y i =f i Based on (x), a predictive model y is established that can predict the content of important chemical components in actual mixed tobacco samples from the theoretically calculated spectral data of mixed tobacco samples. i =h i (x 计算 ), using y i =h i (x 计算 ) for X 计算 (42:502,:) was obtained through quantitative analysis. Using the calibration model y i =f i (x) respectively for X 实测 (42:502,:) and X 计算 (42:502,:) Quantitative analysis was performed to obtain Y. 实测→预测 =[y 实1 , ..., y 实i , ..., y 实10 ] and Y 计算→预测 =[y 计1 , ..., y 计i , ..., y 计10 ];Compare and Y 计算→预测 respectively with Y 实测→预测 The degree of conformity, if With Y 实测→预测 The degree of agreement is greater than that of Y 计算→预测 With Y实测→预测 The degree of agreement indicates the predictive accuracy of the chemical composition content prediction model y. i =h i (x 计算 It has made significant contributions to virtual leaf group formulation technology.

[0054] Figure 2 a and 2b are the measured near-infrared spectral data (X) of 502 mixed samples of tobacco powder, respectively. 实测 ) and using the percentage mixing matrix for C 502×9 The theoretical spectral data (X) calculated from the near-infrared spectral data matrix S of nine pure tobacco leaf powders 计算 It is obvious that X 实测 With X 计算 The significant differences demonstrate that virtual leaf group formulations cannot be directly derived using theoretically calculated spectra.

[0055] Figure 3 It is X 计算→实测 (42:502,:) and X 实测 (42:502,:) Correlation coefficient r between corresponding rows * and X 计算 (42:502,:) and X 实测 (42:502,:) represents the correlation coefficient r between the corresponding rows. Clearly, r... * Most elements in x are greater than the corresponding elements in r, indicating that the spectral normalization model x 预测 =g(x 计算 It has made significant contributions to virtual leaf group formulation technology.

[0056] Table 1 lists the usage of y i =h i (x 计算 ) for X 计算 (42:502,:) was obtained through quantitative analysis. Using the calibration model y respectively i =f i (x) for X 实测 (42:502,:) and X 计算 (42:502,:) Y obtained from quantitative analysis 实测→预测 and Y 计算→预测 The correlation coefficient between them. Clearly, With Y 实测→预测 The correlation coefficient between them is higher than that between Y and Y. 计算→预测 With Y 实测→预测 The correlation coefficient between them is significantly larger, indicating that y i =h i (x 计算 It has made significant contributions to virtual leaf group formulation technology.

[0057] Table 1: respectively with Y 实测→预测 and Y 计算→预测 Correlation coefficient between

[0058]

[0059]

[0060] In summary, this invention, by constructing a spectral standardization model, can obtain the spectra of a large number of virtual leaf blend formulation samples without actual mixing and configuration. Simultaneously, based on the established multivariate calibration model, the chemical composition content of the virtual leaf blend formulation samples can be obtained. Then, the correlation coefficients between the spectra and chemical composition content of the virtual leaf blend formulation samples and the spectra and chemical composition content of the target leaf blend are calculated, and the leaf blend with the highest correlation coefficient is selected as the candidate leaf blend. This invention can assist formulation personnel in quickly selecting suitable alternative tobacco leaf formulations for producing a specific flavor brand of cigarettes from a large number of different varieties and grades of tobacco leaves.

[0061] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for assisting in the design of cigarette leaf blend formulations based on virtual hybrid near-infrared spectroscopy, characterized in that, Includes the following steps: 1) Establish a calibration model between the near-infrared spectra of tobacco powder samples of different grades and the content of important chemical components contained in the samples; the calibration model Where x represents the near-infrared spectrum of each sample. y i For each important chemical component, i =1, 2, ..., m m represents the number of important chemical components; 2) Measure the near-infrared spectrum of each grade of tobacco powder sample in the inventory, and construct a near-infrared spectral data matrix for each grade of tobacco powder sample; the matrix S=[s1;…; s i ; …; s n ], where s i , i =1, 2, ..., n ;s i Near-infrared spectra of various grades of tobacco powder samples in the inventory. n This refers to the number of tobacco leaf grades in the inventory. 3) Prepare mixed sample sets consisting of different grades of tobacco powder mixed at different mass percentages, and determine their near-infrared spectral data matrices; the corresponding grade percentage mixing matrix of the mixed sample sets is as follows: Where k is the number of mixed sample sets and n is the number of tobacco powder grades; the near-infrared spectral data matrix =[ x1; …;x i ; …; x K ], where x i For the first i Near-infrared spectra of a mixed sample; 4) Utilizing percentage mixing matrices and n The theoretical spectral data matrix of the mixed sample set was calculated from the near-infrared spectral data matrix S of a single-grade tobacco powder. ; 5) In and Establish a spectral normalization model between them ; 6) In , and existing calibration models Based on this, a predictive model is established that can predict the content of important chemical components in actual mixed tobacco samples from theoretically calculated spectral data of mixed tobacco samples. ; 7) For any given set of tobacco leaf formulations c = [c1, ..., c2], ... i c n ], where c i For the first i The percentage content of each single-grade tobacco leaf in the mixed sample is first calculated to obtain the theoretical near-infrared spectral data of the virtual mixed tobacco leaf sample. =c×S, where S is the near-infrared spectrum of a single-grade tobacco powder sample, and then using the... The content of important chemical components in the virtual blended tobacco sample was predicted. =[ y1, ..., y i , ..., y m ], and the corresponding actual near-infrared spectrum Finally, calculate and The content of key chemical components in the target flavor cigarette products are respectively compared. and its near-infrared spectrum Correlation coefficient between and If both values ​​are large, the established tobacco blend is likely to meet the requirements, and then verified through a smoking evaluation test; if both values ​​are small, a new tobacco blend is established and the results are calculated. and value; Determining whether a virtual blend of tobacco leaves meets the target flavor of a cigarette product is based on The two models predicted the content of key chemical components in the virtual blended tobacco sample. =[ y 1, ..., y i , ..., y m and the corresponding near-infrared spectrum The content of key chemical components in the target flavor cigarette products are respectively compared. and its near-infrared spectrum Correlation coefficient between and The results are based on the magnitude of the values; if both values ​​are large, the virtual tobacco formulation is likely to meet the requirements and will be verified through a smoking test. ； 。 2. The method for auxiliary design of cigarette leaf blend formulation based on virtual hybrid near-infrared spectroscopy according to claim 1, characterized in that: The single-grade tobacco powder samples were sieved and dried before the near-infrared spectroscopy was measured. Before measuring near-infrared spectra, the mixed sample set first undergoes sieving and drying pretreatment of each grade of tobacco powder sample, and then the mixed sample set is dried.

3. The method for auxiliary design of cigarette leaf blend formulation based on virtual hybrid near-infrared spectroscopy according to claim 2, characterized in that: The sieving is performed through a 60-mesh sieve; the drying is performed at 40-50℃ for 5-7 hours.

4. The method for auxiliary design of cigarette leaf blend formulation based on virtual hybrid near-infrared spectroscopy according to claim 1, characterized in that: The correction model is a multivariate linear correction model constructed using the partial least squares method.

5. The method for auxiliary design of cigarette leaf blend formulation based on virtual hybrid near-infrared spectroscopy according to claim 1, characterized in that: The theoretical spectral data matrix of the tobacco powder mixture sample set From the percentage mixing matrix C K×n Multiplying the near-infrared spectral data matrix S of a single-grade tobacco powder sample yields: .

6. The method for auxiliary design of cigarette leaf blend formulation based on virtual hybrid near-infrared spectroscopy according to claim 1, characterized in that: exist and Establish a spectral normalization model between them The process is as follows: First define Then it can be used for Perform the following singular value decomposition: ;in, ; ; ;' s 'and' n ' represents the corresponding factor representing spectral information and noise respectively; the superscript '´' indicates the matrix transpose operation; arbitrarily assume that the number of spectrally active chemical components in the tobacco powder mixture sample is . r ,but and All r List; and yes submatrix Their column numbers are respectively equal to and The number of columns, and and The number of columns is the same, and is equal to the number of wavelength points in the measured near-infrared spectrum; to obtain and Subsequently, the measured near-infrared spectrum of any mixed sample of tobacco powder was obtained. The corresponding theoretical calculations of the near-infrared spectrum Relationship between It is expressed as follows: , where the superscript '+' indicates the Moore-Penrose generalized inverse operation of the matrix.

7. The method for auxiliary design of cigarette leaf blend formulation based on virtual hybrid near-infrared spectroscopy according to claim 1, characterized in that: exist , and existing calibration models Based on this, a predictive model is established that can predict the content of important chemical components in actual mixed tobacco samples from theoretically calculated spectral data of mixed tobacco samples. The process is as follows: First, Perform singular value decomposition. Then, the spectral data of any mixed tobacco sample can be obtained from theoretical calculations. The predictive model that predicts the content of important chemical components in the actual blended tobacco sample. This can be specifically expressed as follows: 。 8. The method for auxiliary design of cigarette leaf blend formulation based on virtual hybrid near-infrared spectroscopy according to claim 1, characterized in that: A virtual leaf blend formulation screening experiment was conducted using a computer. An arbitrarily set single-leaf tobacco blend formulation c=[ c 1, ..., c i , ..., c n Then, the theoretical near-infrared spectral data of the virtual mixed tobacco sample were calculated. =c×S.

Images