Smoke index prediction method based on LASSO and partial least squares regression

Abstract

The invention discloses a smoke index prediction method based on LASSO and partial least squares regression. The method comprises the following steps: screening representative samples from all stock tobacco leaf samples; rolling each representative sample to obtain a plurality of rolled single tobacco samples; measuring the smoke index content and the chemical component content of the rolled single-material tobacco sample; an LASSO regression algorithm is adopted to screen important variables from the flue gas index content and the chemical component content; establishing a smoke index content prediction model by using the important variables and adopting a partial least squares regression algorithm; and predicting the flue gas index content according to the flue gas index content prediction model. According to the smoke index prediction method based on LASSO and partial least squares regression, the LASSO regression algorithm is used for screening important variables, the partial least squares regression algorithm is used for establishing the prediction model, an objective and accurate means is provided for prediction of the smoke index content of tobacco leaves, and a scientific basis is provided for cigarette digital formula design.

Description

Technical Field

[0001] The present invention relates to the technical field of tobacco analysis, and in particular to a smoke index prediction method based on LASSO and partial least squares regression. Background Art

[0002] Although numerous studies have attempted to develop predictive models for the content of smoke indicators in tobacco leaf smoke, the accuracy and generalization performance of these models are often limited by factors such as uneven data quality and imperfect feature selection. Furthermore, traditional modeling methods often require extensive manual intervention and prior knowledge, which undoubtedly increases the difficulty and cost of modeling.

[0003] Therefore, a flue gas index prediction method based on LASSO and partial least squares regression is urgently needed. Summary of the Invention

[0004] The purpose of the present invention is to provide a smoke index prediction method based on LASSO and partial least squares regression to solve the problems in the above-mentioned prior art. It can use the LASSO regression algorithm to screen important variables and use the partial least squares regression algorithm to establish a prediction model, providing an objective and accurate means for predicting the smoke index content of tobacco leaves and providing a scientific basis for the digital formula design of cigarettes.

[0005] The present invention provides a flue gas index prediction method based on LASSO and partial least squares regression, which includes:

[0006] Select several representative samples from all tobacco leaf samples in stock;

[0007] Rolling each of the selected representative samples to obtain a number of rolled single-ingredient cigarette samples;

[0008] Determining the smoke index content and chemical component content of the rolled single-ingredient cigarette sample corresponding to each of the representative samples;

[0009] LASSO regression algorithm is used to screen out several important variables from the flue gas index content and chemical component content corresponding to each representative sample;

[0010] Using the selected important variables, a partial least squares regression algorithm is used to establish a flue gas index content prediction model;

[0011] The smoke index content in the tobacco leaves is predicted according to the smoke index content prediction model.

[0012] In the above-mentioned method for predicting smoke indexes based on LASSO and partial least squares regression, it is preferred that a plurality of representative samples are screened out from all the tobacco leaf samples in stock, specifically including:

[0013] Conduct near infrared detection on all tobacco leaves in stock to obtain near infrared spectra of all tobacco leaves in stock;

[0014] Based on the near-infrared spectra of all stock tobacco leaves, the Kennard-Stone algorithm was used to screen out several representative samples from all stock tobacco leaf samples.

[0015] In the above-mentioned method for predicting smoke indexes based on LASSO and partial least squares regression, preferably, the method uses the Kennard-Stone algorithm based on the near-infrared spectra of all tobacco leaves in stock to screen out several representative samples from all tobacco leaves in stock, specifically including:

[0016] All near-infrared spectral samples of tobacco leaves in stock are regarded as candidate samples of the training set, and samples are selected from the candidate samples to enter the training set in turn;

[0017] Among all candidate samples, select the two samples with the farthest Euclidean distance to enter the training set;

[0018] For each remaining sample, calculate the Euclidean distance from the sample to each known sample in the training set;

[0019] Find the two samples closest to the selected samples in the training set, and select the one with the farther distance between them into the training set;

[0020] Repeat the steps of selecting samples to enter the training set until the number of samples in the training set reaches the preset requirement.

[0021] In the above-mentioned method for predicting smoke indexes based on LASSO and partial least squares regression, preferably, the screened representative samples are rolled to obtain a plurality of rolled single-ingredient cigarette samples, specifically comprising:

[0022] Set the rolling processing parameters: drum speed 8r / min-12r / min, hot air temperature 100℃-120℃, drum wall temperature 110℃-130℃, outlet moisture content 12%-13%;

[0023] The representative samples were rolled according to the rolling processing parameters, wherein the control requirements for various indicators of the cigarettes are as follows: the filter rod uses a 24.1mm*100mm*2700Pa ordinary acetate filter rod, the cigarette paper uses a 26.5mm*28.5g / m2A70CU wood pulp cross-grain cigarette paper, the tipping paper uses a 64mm hot stamping tipping paper without perforations, the draw resistance control requirement is 1000Pa-1100Pa, the circumference control requirement is 24.1mm-24.5mm, the hardness control requirement is 58%-78%, the quality control requirement for a single cigarette is 0.82g-0.92g, the quality requirement for 20 cigarettes is 16.4g-18.4g, the length control requirement is 83.5mm-84.5mm, and the mass fraction control requirement for moisture content is 11.00%-13.00%.

[0024] In the above-mentioned method for predicting smoke indexes based on LASSO and partial least squares regression, preferably, the determining of the smoke index content and chemical component content of the rolled single-ingredient cigarette sample corresponding to each representative sample specifically includes:

[0025] The tar content in the smoke index of the rolled single-ingredient cigarette sample corresponding to each representative sample is measured by a smoking machine method;

[0026] Determining the nicotine content in the smoke indicators of the rolled single-ingredient cigarette samples corresponding to each of the representative samples using gas chromatography;

[0027] Determine the carbon monoxide content in the smoke index of the rolled single-ingredient cigarette sample corresponding to each representative sample using a non-scattering infrared method;

[0028] Determine the content of conventional chemical components of the rolled single-ingredient cigarette samples corresponding to each representative sample using a continuous flow method, wherein the conventional chemical components include reducing sugars, total alkaloids, total nitrogen, chlorine, potassium, and starch;

[0029] Determine the pH value of the rolled single-ingredient cigarette sample corresponding to each representative sample using a pH meter method;

[0030] Ion chromatography is used to determine the content of inorganic anions in the rolled single-ingredient cigarette samples corresponding to each representative sample, wherein the inorganic anions include sulfate and phosphate;

[0031] Determining the content of inorganic cations in the rolled single-ingredient cigarette samples corresponding to each representative sample using atomic absorption spectrometry, wherein the inorganic cations include magnesium ions and calcium ions;

[0032] Liquid chromatography was used to determine the content of polyphenols in the rolled single-ingredient cigarette samples corresponding to the representative samples, wherein the polyphenols included neochlorogenic acid, chlorogenic acid, cryptochlorogenic acid, scopoletin, and rutin;

[0033] The GC-MS / MS method is used to determine the content of polybasic acids and higher fatty acids in the rolled single-ingredient cigarette samples corresponding to each representative sample, wherein the polybasic acids include oxalic acid, malonic acid, succinic acid, malic acid, citric acid and vanillic acid, and the higher fatty acids include myristic acid, hexadecanoic acid, linoleic acid, oleic acid + linolenic acid, octadecanoic acid and eicosanoic acid;

[0034] An amino acid analyzer is used to determine the content of free amino acids in the rolled single-ingredient cigarette samples corresponding to each representative sample, wherein the free amino acids include aspartic acid, threonine, serine, asparagine, glutamic acid, glutamine, glycine, alanine, valine, cystine, methionine, isoleucine, leucine, tyrosine, phenylalanine, 4-aminobutyric acid, lysine, histidine, tryptophan, arginine, and proline;

[0035] The content of Amadori compounds in the rolled single-ingredient cigarette samples corresponding to each representative sample was determined by HPLC-MS / MS, wherein the Amadori compounds include Glu-An, Fru-Amb, Fru-His, Fru-Pro, Fru-Val, Fru-Thr, Fru-Gly, Fru-Ala, Fru-Asn, Fru-Asp, Fru-Gln, Fru-Glu, Fru-Ile, Fru-Leu, Fru-Tyr, Fru-Phe and Fru-Trp;

[0036] Determine the content of dichloromethane extract of the rolled single-ingredient cigarette sample corresponding to each representative sample by using a filtration method;

[0037] Determining the solanesol content of the rolled single-ingredient tobacco samples corresponding to each representative sample using high performance liquid chromatography;

[0038] The content of neophytadiene in the rolled single-ingredient cigarette sample corresponding to each representative sample was determined by back-flushing gas chromatography.

[0039] As described above, the smoke index prediction method based on LASSO and partial least squares regression is preferably:

[0040] The LASSO regression algorithm is used to screen out several important variables from the flue gas index content and chemical component content corresponding to each representative sample, specifically including:

[0041] The objective function of the linear regression model for smoke index content is:

[0042] J(w)=∑(y-Vw) 2 +λ||w||1=∑(y-Vw) 2+∑λ|w|=ESS(w)+Al1(w) (1)

[0043] Wherein, y represents the smoke index content, which is the dependent variable. The smoke index content includes tar content, nicotine content or CO content. V=[v1,…,v t ] represents the content of all chemical composition indicators, which is the independent variable, t represents the number of chemical composition indicators, w=[w1,…,w t ] T represents the regression coefficient corresponding to each independent variable, λ||w||1 represents the penalty term of the L1 norm, ESS(w) represents the sum of squared errors of n groups of observation data corresponding to n representative samples, n represents the number of representative samples screened, and λl1(w) represents the penalty term;

[0044] Using the coordinate descent method, control the other regression parameters w unchanged, and for one of the w in the objective function J(w) q Find the partial derivative, g = 1, 2, ..., t, and so on,

[0045] Find the partial derivatives of the remaining regression parameters, and finally make each derivative function equal to 0, so as to obtain the objective function that reaches the global minimum.

[0046] at this time,

[0047] Taking partial derivative of formula (2), we get:

[0048]

[0049] in,

[0050] Since the penalty term is not differentiable, the following derivatives are used:

[0051]

[0052] Let the sum of the two partial derivatives of formula (3) and formula (4) be equal to 0, and we can get:

[0053]

[0054] From this we can get:

[0055]

[0056] All regression coefficients w q Remove the independent variables that are 0 and retain all regression coefficients w q Independent variables that are not 0 are considered as important variables that are screened out.

[0057] The above-mentioned method for predicting flue gas indexes based on LASSO and partial least squares regression, wherein preferably, the LASSO regression algorithm is used to screen out several important variables from the flue gas index content and chemical component content corresponding to each representative sample, further comprising:

[0058] The cross-validation method is used to determine the value of the penalty coefficient λ, which includes:

[0059] The dataset is divided into several parts, one of which is selected as the validation set each time, and the rest are used as the training set. The optimal λ value is selected by adjusting the value of λ and calculating the performance index of the model on the validation set.

[0060] In the above-mentioned method for predicting smoke index based on LASSO and partial least squares regression, preferably, the smoke index content prediction model is established by using the partial least squares regression algorithm using the selected important variables, specifically including:

[0061] From the three groups of independent variables corresponding to the screened tar content, nicotine content and CO content, select the independent variables that appear at least once as the independent variables of the partial least squares regression model;

[0062] Let p dependent variables y1,…,y p With m independent variables x1,…,x m The standardized variables are homogenized, and the n-times standardized observation data matrices corresponding to the n representative samples of the independent variable group and the dependent variable group are recorded as:

[0063]

[0064] Extract the first pair of components of the two variable groups respectively, and maximize the correlation between the first pair of components of the two variable groups. Let the first pair of components of the two groups of variables be u1 and v1 respectively, u1 represents the independent variable set X = [x1, ..., x m ] T The linear combination u1=α 11 x1+...+α 1m x m =ρ (1)T X, v1 represents the dependent variable set Y = [y1,…,y p ] T The linear combination v1=β 11 y1+…+β 1m y m =γ (1)T Y;

[0065] According to the standardized observation data matrices A and B of the two sets of variables, the score vectors of the first pair of components are calculated by the following formula, which are respectively denoted as and

[0066]

[0067]

[0068] The score vector of the first pair of components and The inner product of , calculate the covariance index v(u1, v1) of the first pair of components u1 and v1;

[0069] Calculate the following conditional extreme value problem to maximize the correlation between u1 and v1, and thus enable u1 and v1 to extract as much variation information as possible from their respective variable groups:

[0070]

[0071]

[0072] According to the Lagrange multiplier method, the above conditional extreme value problem is transformed into: solving the unit vector ρ (1) and γ (1) , so that θ1=ρ (1)T A T Bγ (1) reach the maximum;

[0073] By calculating the m×m matrix M=A T BB T The eigenvalues ​​and eigenvectors of A, and the largest eigenvalue of M is The corresponding unit eigenvector is the desired solution ρ (1) , and γ (1) The following formula can be used to calculate ρ (1) Calculation yields:

[0074]

[0075] Establish y1,…,y p The regression model for u1 and x1,…,x m Regression model for v1:

[0076]

[0077] Among them, σ (1) =[σ 11 ,…,σ 1m ] T , τ (1) =[τ 11 ,…,τ 1p ] T They represent the parameter vectors in the many-to-one regression model, and A1 and B1 represent the residual matrices;

[0078] The regression coefficient vector σ is calculated by the following formula (1) and τ (1) The least squares estimation results of :

[0079]

[0080] σ (1) and τ (1) It is called model effect loading;

[0081] Use residual matrices A1 and B1 to replace A and B, repeat the above steps, and record Then the residual matrix

[0082] If the absolute value of the elements in the residual matrix B1 is approximately 0, it is considered that the accuracy of the regression model established with the first component has met the requirements, and the extraction of components can be stopped. Otherwise, replace A and B with residual matrices A1 and B1, and repeat the above steps to obtain

[0083] ρ (2) =[α 21 ,…,α 2m ] T , γ (2) =[β 21 ,…,β 2p ] T (15)

[0084] and is the score vector of the second pair of components, and

[0085] and are the loadings of the second pair of components of X and Y, respectively.

[0086]

[0087] Assume that the rank of matrix A is r≤min(n-1,m), then there are r principal components u1, u2, ..., u r , such that:

[0088]

[0089] will u k =α k1 x1+…+α km x m (k=1,2,…,r)Substitute into Y=u1τ (1) +…+u r τ (r) , that is, the partial least squares regression equation for p dependent variables is:

[0090] y j =c j1 x1+…+c jm x m ,j=1,2,…,p (18).

[0091] In the above-mentioned method for predicting smoke indexes based on LASSO and partial least squares regression, preferably, the smoke index content in tobacco leaves is predicted according to the smoke index content prediction model, specifically comprising:

[0092] Detecting the content of important variables of the rolled single-ingredient cigarette sample corresponding to the sample to be tested;

[0093] The contents of important variables of the sample to be tested are input into the smoke index content prediction model to obtain the smoke index content of the sample to be tested.

[0094] As described above, the smoke index prediction method based on LASSO and partial least squares regression, wherein preferably, the smoke index prediction method based on LASSO and partial least squares regression further includes:

[0095] The prediction effect of the flue gas index content prediction model is evaluated, specifically including:

[0096] Statistical indicators are used to evaluate the prediction effect of the flue gas index content prediction model, wherein the statistical indicators include: at least one of the mean square error, the root mean square error and the determination coefficient. The closer the determination coefficient is to 1, the smaller the mean square error and the root mean square error are, indicating that the prediction accuracy of the flue gas index content prediction model is better.

[0097] The present invention provides a smoke index prediction method based on LASSO and partial least squares regression. Representative tobacco leaf samples are screened out from all stock tobacco leaves, and the screened tobacco leaves are rolled according to process requirements. The smoke index content and chemical composition of the rolled representative samples are then measured. Then, the LASSO regression algorithm is used to screen important variables from training set sample data, and a prediction model is established using the partial least squares regression algorithm. This provides an objective and accurate means for predicting the smoke index content of tobacco leaves and a scientific basis for digital cigarette formula design. The Kennard-Stone algorithm takes into account the diversity and representativeness of the samples, thereby ensuring that the selected samples have a wider quality and chemical composition distribution, thereby improving the efficiency and accuracy of modeling. The LASSO regression algorithm can screen out chemical components that have a greater impact on the smoke index content as important variables, and has the advantages of controlling multicollinearity and reducing data noise, which helps to improve the accuracy and reliability of the prediction model. The prediction model is established using the partial least squares regression algorithm, thereby achieving accurate prediction of the smoke index content. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described below with reference to the accompanying drawings, in which:

[0099] Figure 1 A flowchart of an embodiment of a flue gas index prediction method based on LASSO and partial least squares regression provided by the present invention;

[0100] Figure 2 Schematic diagram of the cross-validated LASSO fitted MSE (tar);

[0101] Figure 3 is the cross-validated LASSO fitted MSE (nicotine);

[0102] Figure 4 MSE(CO) for the cross-validated LASSO fit;

[0103] Figure 5 is the LASSO fitting coefficient trajectory diagram (TAR);

[0104] Figure 6 is the LASSO fitting coefficient trajectory diagram (nicotine);

[0105] Figure 7 is the LASSO fitting coefficient trajectory diagram (CO);

[0106] Figure 8 This is a comparison chart between the actual and predicted values ​​of tar;

[0107] Figure 9 This is a comparison chart of the actual and predicted values ​​of nicotine;

[0108] Figure 10 This is a comparison chart of the actual CO value and the predicted value. DETAILED DESCRIPTION

[0109] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. The description of the exemplary embodiments is merely illustrative and is in no way intended to limit the present disclosure, its application, or use. The present disclosure can be implemented in many different forms and is not limited to the embodiments described herein. These embodiments are provided to make the present disclosure thorough and complete and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that unless otherwise specifically stated, the relative arrangement of parts and steps, the composition of materials, numerical expressions, and numerical values ​​set forth in these embodiments should be interpreted as being merely exemplary and not as limiting.

[0110] The terms "first," "second," and similar terms used in this disclosure do not indicate any order, quantity, or importance, but are simply used to distinguish different parts. Terms such as "include" or "comprising" mean that the elements preceding the term include the elements listed after the term, and do not exclude the possibility of also including other elements. Terms such as "upper," "lower," and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0111] In the present disclosure, when a specific component is described as being located between a first component and a second component, there may or may not be an intervening component between the specific component and the first component or the second component. When a specific component is described as being connected to another component, the specific component may be directly connected to the other component without an intervening component, or may not be directly connected to the other component but have an intervening component.

[0112] All terms (including technical or scientific terms) used in this disclosure have the same meaning as those understood by one of ordinary skill in the art to which this disclosure belongs, unless otherwise specifically defined. It should also be understood that terms defined in, for example, general dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an idealized or highly formal sense, unless explicitly defined herein.

[0113] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0114] like Figure 1 As shown, the smoke index prediction method based on LASSO and partial least squares regression provided in this embodiment includes the following steps in actual implementation:

[0115] Step S1: Select several representative samples from all tobacco leaf samples in stock.

[0116] In the present invention, all tobacco leaves in stock are scanned using near-infrared spectra. Based on the near-infrared spectra of all tobacco leaves in stock, the Kennard-Stone algorithm is used to select several representative samples from all tobacco leaves in stock for constructing a smoke index content prediction model. In one embodiment of the smoke index prediction method based on LASSO and partial least squares regression of the present invention, step S1 may specifically include:

[0117] Step S11: Perform near-infrared detection on all tobacco leaves in stock to obtain near-infrared spectra of all tobacco leaves in stock.

[0118] Step S12: Based on the near-infrared spectra of all the tobacco leaves in stock, a Kennard-Stone algorithm is used to screen out several representative samples from all the tobacco leaves in stock.

[0119] The Kennard-Stone algorithm is an algorithm for selecting samples. It can select a set of the most representative samples from a given data set. Its basic idea is to select samples by minimizing the distance between the selected samples to ensure that the selected samples are as representative of the original data set as possible. In one embodiment of the smoke index prediction method based on LASSO and partial least squares regression of the present invention, step S12 may specifically include:

[0120] Step S121: All near-infrared spectrum samples of tobacco leaves in stock are regarded as candidate samples of the training set, and samples are selected from the candidate samples in turn to enter the training set.

[0121] Step S122: Among all candidate samples, select the two samples with the farthest Euclidean distance to enter the training set.

[0122] Step S123: For each remaining sample, calculate the Euclidean distance between the sample and each known sample in the training set.

[0123] Step S124: Find the two samples closest to the selected samples in the training set, and select the one with the farther distance between the two samples into the training set.

[0124] Step S125: Repeat the step of selecting samples to enter the training set until the number of samples in the training set reaches the preset requirement.

[0125] In one embodiment of the present invention, near-infrared spectra of all stock tobacco leaves are scanned, and the Kennard-Stone algorithm is used to screen out 120 representative samples from all stock tobacco leaf samples, covering multiple production areas at home and abroad from 2013 to 2020. The tar content per unit tobacco is between 12-20 mg / g, the nicotine content is between 0.3-2.5 mg / g, and the CO content is between 11-15 mg / g, and a smoke index prediction model is constructed.

[0126] Step S2: rolling the selected representative samples to obtain a number of rolled single-ingredient cigarette samples.

[0127] In one embodiment of the smoke index prediction method based on LASSO and partial least squares regression of the present invention, step S2 may specifically include:

[0128] Step S21, set the rolling processing parameters: drum speed is 8r / min-12r / min (for example, 10r / min), hot air temperature is 100℃-120℃ (for example, 110℃), drum wall temperature is 110℃-130℃ (for example, 120℃), outlet moisture content is 12%-13% (for example, 12.5%).

[0129] Step S22, rolling each of the representative samples according to the rolling processing parameters, wherein the control requirements for various indicators of the cigarette are as follows: the filter rod uses a 24.1mm*100mm*2700Pa ordinary acetate filter rod, the cigarette paper uses a 26.5mm*28.5g / m2A70CU wood pulp cross-grain cigarette paper, the tipping paper uses a 64mm hot stamping tipping paper without perforations, the draw resistance control requirement is 1000Pa-1100Pa, the circumference control requirement is 24.1mm-24.5mm, the hardness control requirement is 58%-78%, the single cigarette quality control requirement is 0.82g-0.92g, the 20 cigarette quality requirement is 16.4g-18.4g, the length control requirement is 83.5mm-84.5mm, and the moisture content mass fraction control requirement is 11.00%-13.00%.

[0130] Step S3: measuring the smoke index content and chemical component content of the rolled single-ingredient cigarette sample corresponding to each representative sample.

[0131] In one embodiment of the flue gas index prediction method based on LASSO and partial least squares regression of the present invention, step S3 may specifically include:

[0132] Step S31: using a smoking machine method to measure the tar content in the smoke index of the rolled single-ingredient cigarette sample corresponding to each representative sample.

[0133] Specifically, the tar content in the smoke indicators of the rolled single-ingredient cigarettes corresponding to the representative tobacco leaf samples is determined with reference to the relevant national standard "GB / T 19609-2004 Determination of total particulate matter and tar using a conventional analytical smoking machine for cigarettes".

[0134] Step S32: using gas chromatography to determine the nicotine content in the smoke indicators of the rolled single-ingredient cigarette samples corresponding to the representative samples.

[0135] Specifically, the nicotine content in the smoke indicators of the rolled single-ingredient cigarettes corresponding to the representative tobacco leaf samples is determined with reference to "GB / T 23355-2009 Determination of Nicotine in Total Particulate Matter of Cigarettes - Gas Chromatography Method".

[0136] Step S33: using a non-scattering infrared method to measure the carbon monoxide content in the smoke indicators of the rolled single-ingredient cigarette samples corresponding to each of the representative samples.

[0137] Specifically, the carbon monoxide content in the smoke indicators of the rolled single-ingredient cigarettes corresponding to the representative tobacco leaf samples is determined in accordance with "GB / T 23356-2009 Determination of Carbon Monoxide in the Gas Phase of Cigarette Smoke - Non-scattering Infrared Method".

[0138] Step S34: using a continuous flow method to determine the content of conventional chemical components of the rolled single-ingredient cigarette samples corresponding to each of the representative samples, wherein the conventional chemical components include reducing sugars, total alkaloids, total nitrogen, chlorine, potassium and starch.

[0139] Specifically, the determination of conventional chemical components (total alkaloids, reducing sugars, total sugars, total nitrogen, potassium, chlorine, and starch) refers to "YC / T 159-2002 Tobacco and Tobacco Products - Determination of Water-soluble Sugars - Continuous Flow Method", "YC / T 160-2002 Tobacco and Tobacco Products - Determination of Total Alkaloids - Continuous Flow Method", "YC / T 161-2002 Tobacco and Tobacco Products - Determination of Total Nitrogen - Continuous Flow Method", "YC / T 162-2011 Tobacco and Tobacco Products - Determination of Chlorine - Continuous Flow Method", "YC / T 217-2007 Tobacco and Tobacco Products - Determination of Potassium - Continuous Flow Method", and "YC / T 216-2007 Tobacco and Tobacco Products - Determination of Starch - Continuous Flow Method".

[0140] Step S35: using a pH meter method to measure the pH value of the rolled single-ingredient cigarette sample corresponding to each representative sample.

[0141] Specifically, the pH value is determined with reference to "YC / T 222-2007 Tobacco and Tobacco Products - Determination of Tobacco pH Value".

[0142] Step S36: using ion chromatography to determine the content of inorganic anions in the rolled single-ingredient cigarette samples corresponding to the representative samples, wherein the inorganic anions include sulfate and phosphate.

[0143] Specifically, the determination of inorganic anions (sulfate, phosphate) refers to "YC / T 248-2008 Tobacco and tobacco products - Determination of inorganic anions - Ion chromatography method".

[0144] Step S37: using atomic absorption spectrometry to determine the content of inorganic cations in the rolled single-ingredient cigarette samples corresponding to the representative samples, wherein the inorganic cations include magnesium ions and calcium ions.

[0145] Specifically, the determination of inorganic cations (Mg, Ca) refers to "YC / T 174-2003 Determination of Calcium in Tobacco and Tobacco Products - Atomic Absorption Spectrometry" and "YC / T 175-2003 Determination of Magnesium in Tobacco and Tobacco Products - Atomic Absorption Spectrometry".

[0146] Step S38: Using a liquid chromatograph to determine the content of polyphenols in the rolled single-ingredient cigarette samples corresponding to the representative samples, wherein the polyphenols include neochlorogenic acid, chlorogenic acid, cryptochlorogenic acid, scopoletin and rutin.

[0147] Specifically, the determination of polyphenols (neochlorogenic acid, chlorogenic acid, cryptochlorogenic acid, scopoletin, and rutin) refers to "YC / T 202-2006 Determination of polyphenol compounds chlorogenic acid, scopoletin, and rutin in tobacco and tobacco products."

[0148] Step S39: Using GC-MS / MS to determine the content of polyacids and higher fatty acids in the rolled single-ingredient cigarette samples corresponding to each of the representative samples, wherein the polyacids include oxalic acid, malonic acid, succinic acid, malic acid, citric acid and vanillic acid, and the higher fatty acids include tetradecanoic acid, hexadecanoic acid, linoleic acid, oleic acid + linolenic acid, octadecanoic acid and eicosanoic acid.

[0149] Specifically, the determination of polybasic acids (oxalic acid, malonic acid, succinic acid, malic acid, citric acid, vanillic acid) and higher fatty acids (tetradecanoic acid, hexadecanoic acid, linoleic acid, oleic acid + linolenic acid, octadecanoic acid, eicosanoic acid) was carried out with reference to the paper "Simultaneous Analysis of 42 Organic Acids in Tobacco Leaves by GC-MS / MS Method".

[0150] Step S310: Determine the free amino acid content of the rolled single-ingredient cigarette sample corresponding to each representative sample using an amino acid analyzer, wherein the free amino acids include aspartic acid, threonine, serine, asparagine, glutamic acid, glutamine, glycine, alanine, valine, cystine, methionine, isoleucine, leucine, tyrosine, phenylalanine, 4-aminobutyric acid, lysine, histidine, tryptophan, arginine, and proline.

[0151] Specifically, the determination of free amino acids (aspartic acid, threonine, serine, asparagine, glutamic acid, glutamine, glycine, alanine, valine, cystine, methionine, isoleucine, leucine, tyrosine, phenylalanine, 4-aminobutyric acid, lysine, histidine, tryptophan, arginine, and proline) was performed according to the method described in YC / T 282-2009 Determination of Free Amino Acids in Tobacco Leaves - Amino Acid Analyzer Method.

[0152] Step S311: Determine the content of Amadori compounds in the rolled single-ingredient cigarette samples corresponding to each of the representative samples using HPLC-MS / MS, wherein the Amadori compounds include Glu-An, Fru-Amb, Fru-His, Fru-Pro, Fru-Val, Fru-Thr, Fru-Gly, Fru-Ala, Fru-Asn, Fru-Asp, Fru-Gln, Fru-Glu, Fru-Ile, Fru-Leu, Fru-Tyr, Fru-Phe, and Fru-Trp.

[0153] Specifically, the determination of Amadori compounds (Glu-An, Fru-Amb, Fru-His, Fru-Pro, Fru-Val, Fru-Thr, Fru-Gly, Fru-Ala, Fru-Asn, Fru-Asp, Fru-Gln, Fru-Glu, Fru-Ile, Fru-Leu, Fru-Tyr, Fru-Phe, Fru-Trp) was carried out with reference to the paper "Simultaneous Determination of 22 Amadori Compounds in Tobacco by HPLC-MS / MS".

[0154] Step S312: Determine the content of dichloromethane extract of the rolled single-ingredient cigarette sample corresponding to each representative sample by using a filtration method.

[0155] Specifically, the determination of dichloromethane extract refers to the patent "A filtration device for determining the content of dichloromethane extract of tobacco".

[0156] Step S313: using high performance liquid chromatography to determine the solanesol content of the rolled single-ingredient cigarette samples corresponding to the representative samples.

[0157] The determination of solanesol refers to "GB / T 31758-2015 Determination of solanesol in tobacco leaves and tobacco leaf extracts - High performance liquid chromatography method".

[0158] Step S314: using backflushing gas chromatography to determine the content of neophytadiene in the rolled single-ingredient cigarette sample corresponding to each representative sample.

[0159] Specifically, the determination of phytadiene refers to the paper "Detection and Analysis of the Content of Phytadiene in Tobacco Leaves from Different Ecological Zones by Backflush Gas Chromatography".

[0160] Step S4: using the LASSO regression algorithm to screen out several important variables from the flue gas index content and chemical component content corresponding to each representative sample.

[0161] The selected important variables are used in the subsequent process of establishing a smoke index content prediction model. In one embodiment of the smoke index prediction method based on LASSO and partial least squares regression of the present invention, step S4 may specifically include:

[0162] Step S41, the LASSO regression algorithm is used to screen out several important variables from the flue gas index content and chemical component content corresponding to each representative sample, specifically including:

[0163] The objective function of the linear regression model for smoke index content is:

[0164] J(w)=∑(y-Vw) 2 +λ||w||1=∑(y-Vw) 2 +∑λ|w|=ESS(w)+Al1(w) (1)

[0165] Wherein, y represents the smoke index content, which is the dependent variable. The smoke index content includes tar content, nicotine content or CO content. V=[v1,…,v t ] represents the content of all chemical composition indicators, which is the independent variable, t represents the number of chemical composition indicators, w=[w1,…,w t ] T represents the regression coefficient corresponding to each independent variable, λ||w||1 represents the penalty term of the L1 norm, ESS(w) represents the sum of squared errors of n groups of observation data corresponding to n representative samples, n represents the number of representative samples screened out, and λl1(w) represents the penalty term.

[0166] In the present invention, in order to ensure that the regression coefficient w can be calculated, the LASSO regression model adds an L1 norm penalty term to the objective function, which can reduce some unimportant regression coefficients to 0, thereby achieving the effect of eliminating variables.

[0167] Step S42: Using the coordinate descent method, control the other regression parameters w to be constant, and calculate one of the w in the objective function J(w) q Find the partial derivative, q = 1, 2, …, t, and so on.

[0168] Find the partial derivatives of the remaining regression parameters, and finally make each derivative function equal to 0, so as to obtain the objective function that reaches the global minimum.

[0169] at this time,

[0170] Taking partial derivative of formula (2), we get:

[0171]

[0172] in,

[0173] Step S43: Since the penalty term is not differentiable, the following derivatives are used:

[0174]

[0175] Step S44: Let the sum of the two partial derivatives of formula (3) and formula (4) be equal to 0, and we can get:

[0176]

[0177] From this we can get:

[0178]

[0179] Step S45: All regression coefficients w q Remove the independent variables that are 0 and retain all regression coefficients w q Independent variables that are not 0 are considered as important variables that are screened out.

[0180] Furthermore, in one embodiment of the smoke index prediction method based on LASSO and partial least squares regression of the present invention, step S4 may further include:

[0181] Step S46: Determine the value of the penalty coefficient λ using a cross-validation method.

[0182] Specifically, the dataset is divided into several parts, one of which is selected as the validation set each time, and the rest are used as the training set. The optimal λ value is selected by adjusting the value of λ and calculating the performance index of the model on the validation set.

[0183] In one embodiment of the present invention, the LASSO regression algorithm is used to screen variables, and the penalty coefficient λ corresponding to the minimum mean square error (MSE) is obtained by cross-validation, as shown in FIG. Figure 2-Figure 4 As shown. According to the selected optimal penalty coefficient λ, the independent variables whose estimated parameters are compressed to 0 can be eliminated, and the independent variables whose coefficients are not 0 can be retained, as shown in Figure 5-Figure 7 As shown in the figure, from the three groups of independent variables, the independent variables that appeared at least once were selected, and finally 12 independent variables were screened, namely total plant alkaloids, K, chloride, sulfate, cryptochlorogenic acid, scopoletin, rutin, octadecanoic acid, serine, cystine, isoleucine, and tryptophan.

[0184] Step S5: using the selected important variables, a partial least squares regression algorithm is used to establish a flue gas index content prediction model.

[0185] In one embodiment of the flue gas index prediction method based on LASSO and partial least squares regression of the present invention, step S5 may specifically include:

[0186] Step S51 : From the three groups of independent variables corresponding to the screened tar content, nicotine content, and CO content, select the independent variables that appear at least once as the independent variables of the partial least squares regression model.

[0187] Step S52: Set p dependent variables y1, ..., y p With m independent variables x1,…,x m The standardized variables are homogenized, and the n-times standardized observation data matrices corresponding to the n representative samples of the independent variable group and the dependent variable group are recorded as:

[0188]

[0189] Extract the first pair of components of the two variable groups respectively, and maximize the correlation between the first pair of components of the two variable groups. Let the first pair of components of the two groups of variables be u1 and v1 respectively, u1 represents the independent variable set X = [x1, ..., x m ] T The linear combination u1=α 11 x1+…+α 1m x m =ρ (1)T X, v1 represents the dependent variable set Y = [y1,…,y p ] T The linear combination v1=β 11 y1+…+β 1m y m =γ (1)T Y;

[0190] According to the standardized observation data matrices A and B of the two sets of variables, the score vectors of the first pair of components are calculated by the following formula, which are respectively denoted as and

[0191]

[0192]

[0193] The score vector of the first pair of components and The inner product of , calculate the covariance index v(u1, v1) of the first pair of components u1 and v1;

[0194] Step S53: Calculate the following conditional extreme value problem to maximize the correlation between u1 and v1, and thereby enable u1 and v1 to extract as much variation information of their respective variable groups as possible:

[0195]

[0196]

[0197] Step S54: According to the Lagrange multiplier method, the above conditional extreme value problem is transformed into: solving the unit vector ρ (1) and γ (1) , so that θ1=ρ (1)T A T Bγ (1) reach the maximum;

[0198] Step S55, by calculating the m×m matrix M=A T BB T The eigenvalues ​​and eigenvectors of A, and the largest eigenvalue of M is The corresponding unit eigenvector is the desired solution ρ (1) , and γ (1) The following formula can be used to calculate ρ (1) Calculation yields:

[0199]

[0200] Step S56: Establish y1, ..., y p The regression model for u1 and x1,…,x m Regression model for v1:

[0201]

[0202] Among them, σ (1) =[σ 11 ,…,σ 1m ] T , τ (1) =[τ 11 ,…,τ 1p ] T They represent the parameter vectors in the many-to-one regression model, and A1 and B1 represent the residual matrices;

[0203] Step S57: Calculate the regression coefficient vector σ using the following formula: (1) and τ (1) The least squares estimation results of :

[0204]

[0205] σ (1) and τ (1) It is called model effect loading;

[0206] Step S57, replace A and B with residual matrices A1 and B1, repeat the above steps, and record Then the residual matrix

[0207] Step S58: If the absolute value of the elements in the residual matrix B1 is approximately 0, it is considered that the accuracy of the regression model established with the first component has met the requirements, and the extraction of components can be stopped. Otherwise, the residual matrices A1 and B1 are used to replace A and B, and the above steps are repeated to obtain

[0208] ρ (2) =[α 21 ,…,α 2m ] T , γ (2) =[β 21 ,…,β 2p ] T (15)

[0209] and is the score vector of the second pair of components, and and are the loadings of the second pair of components of X and Y, respectively.

[0210]

[0211] Step S59: Assume that the rank of matrix A is r≤min(n-1,m), then there are r principal components u1, u2, ..., u r , such that:

[0212]

[0213] Step S510: k =α k1 x1+…+α kn x m (k=1,2,...,r)Substitute into Y=u1τ (1) +…+u r τ (r) , that is, the partial least squares regression equation for p dependent variables is:

[0214] y j =c j1 x1+…+c jm x m ,j=1,2,…,p (18).

[0215] In one embodiment of the present invention, first define the tar content as the dependent variable y1, the nicotine content as the dependent variable y2, the CO content as the dependent variable y3, the total alkaloid content as the independent variable x1, the K content as the independent variable x2, the chlorine content as the independent variable x3, the sulfate content as x4, the cryptochlorogenic acid content as x5, the scopoletin content as the independent variable x6, the rutin content as the independent variable x7, the octadecanoic acid content as the independent variable x8, the serine content as the independent variable x9, and the cystine content as the independent variable x1. 10 , isoleucine content is the independent variable x 11 , tryptophan content is the independent variable x 12 .

[0216] Then, the partial least squares regression algorithm was used to establish a flue gas index content prediction model, and 6 pairs of principal components were obtained, which are:

[0217] The first pair of principal components:

[0218] v1=-0.0455x1-0.0158x2-0.005x3-0.0086x4+0.0230x5-0.0293x6+0.0059x7+0.0086x8+0.0127x9+0.0009x 10 -0.0021x 11 +0.0012x 12

[0219] u1=1.7848y1-7.7142y2-6.2378y3

[0220] The second pair of principal components:

[0221] v2=-0.0269x1-0.0271x2+0.0205x3+0.0075x4-0.0245x5-0.0070x6-0.0319x7+0.0171x8-0.0159x9-0.0140x 10 +0.0139x 11 -0.0131x 12

[0222] u2=3.8139y1-2.1968y2+2.5001y3

[0223] The third pair of principal components:

[0224] v3=-0.0393x1+0.0554x2-0.0314x3+0.0101x4+0.0259x5+0.0114x6+0.0116x7+0.0157x8-0.0030x9-0.0079x 10 -0.0192x 11-0.0053x 12

[0225] u3=-4.3271y1-1.8663y2-1.2135y3

[0226] The fourth pair of principal components:

[0227] v4=-0.0660x1+0.0733x2+0.0373x3-0.0121x4+0.0344x5+0.0228x6-0.0269x7-0.0301x8-0.0004x9+0.0103x 10 +0.0132x 11 +0.0223x 12

[0228] u4=-1.1291y1+0.0779y2-2.4013y3

[0229] The fifth pair of principal components:

[0230] v5=-0.0713x1+0.0636x2+0.0195x3+0.0022x4-0.0077x5+0.0143x6+0.0045x7-0.0568x8-0.0294x9+0.0462x 10 -0.0687x 11 +0.0041x 12

[0231] u5=-2.8502y1-1.0543y2-0.1338y3

[0232] The sixth pair of principal components:

[0233] v6=-0.0802x1-0.0319x2-0.0292x3+0.0007x4-0.0634x5+0.0736x6+0.0439x7-0.0038x8-0.0135x9-0.0025x 10 +0.0072x 11 +0.0708x 12

[0234] u6=-0.7067y1-1.5309y2+1.3760y3

[0235] Finally, the following regression equation is obtained:

[0236] y1=0.3145x1-0.9395x2+0.3469x3-0.0033x4-0.2915x5-1.3711x6-0.0713x7+4.7915x8+0.0005x9-0.0119x 10 +0.1128x 11 -0.0015x 12 +12.9340

[0237] y2=0.6872x1-0.0950x2-0.0150x3+0.0036x4-0.0298x5-0.2752x6-0.0114x7+0.2771x8+0.0003x9-0.0016x 10 +0.0182x 11 -0.0019x 12 +0.0443

[0238] y3=1.7110x1-1.7325x2-0.4964x3+0.0295x4-1.6468x5+0.9866x6+0.0157x7+6.7266x8-0.0006x9-0.0190x 10 +0.1330x 11 -0.0013x 12 +14.9080.

[0239] Step S6: predicting the smoke index content in the tobacco leaves according to the smoke index content prediction model.

[0240] In one embodiment of the flue gas index prediction method based on LASSO and partial least squares regression of the present invention, step S6 may specifically include:

[0241] Step S61: detecting the contents of important variables of the rolled single-ingredient cigarette sample corresponding to the sample to be tested.

[0242] Step S62: Input the contents of the important variables of the sample to be tested into the smoke index content prediction model to obtain the smoke index content of the sample to be tested.

[0243] Furthermore, in one embodiment of the present invention, the flue gas index prediction method based on LASSO and partial least squares regression further includes:

[0244] Step S7: Evaluate the prediction effect of the flue gas index content prediction model.

[0245] Specifically, statistical indicators are used to evaluate the prediction effect of the flue gas index content prediction model.

[0246] The statistical indicators include mean square error (MSE), root mean square error (RMSE) and coefficient of determination (R 2 ), the closer the determination coefficient is to 1, the smaller the mean square error and root mean square error are, indicating that the prediction accuracy of the flue gas index content prediction model is better.

[0247] The R of the partial least squares regression model of tar obtained in step S5 2 =0.910, MSE=0.2129, RMSE=0.4614, indicating that the overall model fitting effect is good. The actual value of the tar test sample and the predicted value are regressed and the actual value is used as the independent variable x α , the predicted value is the cause vector y α , we get the regression equation y α =0.875x α +2.133, compared with the following figure Figure 8 , indicating that the prediction effect of the model is good.

[0248] Furthermore, in some embodiments of the present invention, the R of the nicotine partial least squares regression model obtained in step S5 is 2 =0.939, MSE=0.0105, RMSE=0.1025, indicating that the overall model fitting effect is good. The actual value of the nicotine test sample and the predicted value are regressed and the actual value is used as the independent variable x β , the predicted value is the cause vector y β , we get the regression equation y β =0.946x β +0.091, compared with the following figure Figure 9 , indicating that the prediction effect of the model is good.

[0249] Furthermore, in some embodiments of the present invention, the R of the CO partial least squares regression model obtained in step S5 is 2 =0.839, MSE=0.1214, RMSE=0.3484, indicating that the overall model fitting effect is good. The actual value of the CO test sample and the predicted value are regressed and the actual value is used as the independent variable x γ , the predicted value is the cause vector y γ , we get the regression equation y γ =1.137x γ -1.603, compared with the following figure Figure 10 , indicating that the prediction effect of the model is good.

[0250] The smoke index prediction method based on LASSO and partial least squares regression provided in an embodiment of the present invention screens out representative tobacco leaf samples from all stock tobacco leaves, and rolls the screened tobacco leaves according to process requirements, then measures the smoke index content and chemical composition of the rolled representative samples, and then uses the LASSO regression algorithm to screen important variables from the training set sample data, and uses the partial least squares regression algorithm to establish a prediction model, which provides an objective and accurate means for predicting the smoke index content of tobacco leaves and provides a scientific basis for the digital formula design of cigarettes; the Kennard-Stone algorithm takes into account the diversity and representativeness of the samples, thereby ensuring that the selected samples have a wider quality and chemical composition distribution, thereby improving the efficiency and accuracy of modeling; the LASSO regression algorithm can screen out chemical components that have a greater impact on the smoke index content as important variables, and has the advantages of controlling multicollinearity and reducing data noise, which helps to improve the accuracy and reliability of the prediction model; by using the partial least squares regression algorithm to establish a prediction model, accurate prediction of the smoke index content is achieved.

[0251] Thus far, various embodiments of the present disclosure have been described in detail. To avoid obscuring the concept of the present disclosure, some details known in the art have not been described. Based on the above description, those skilled in the art can fully understand how to implement the technical solutions disclosed herein.

[0252] Although some specific embodiments of the present disclosure have been described in detail through examples, those skilled in the art will understand that the above examples are for illustration only and are not intended to limit the scope of the present disclosure. Those skilled in the art will understand that the above embodiments may be modified or some technical features may be replaced with equivalents without departing from the scope and spirit of the present disclosure. The scope of the present disclosure is defined by the appended claims.

Claims

1. A flue gas index prediction method based on LASSO and partial least squares regression, characterized in that: include: Select several representative samples from all tobacco leaf samples in stock; Rolling each of the selected representative samples to obtain a number of rolled single-ingredient cigarette samples; Determining the smoke index content and chemical component content of the rolled single-ingredient cigarette sample corresponding to each of the representative samples; LASSO regression algorithm is used to screen out several important variables from the flue gas index content and chemical component content corresponding to each representative sample; Using the selected important variables, a partial least squares regression algorithm is used to establish a flue gas index content prediction model; The smoke index content in the tobacco leaves is predicted according to the smoke index content prediction model.

2. The flue gas index prediction method based on LASSO and partial least squares regression according to claim 1 is characterized in that: The representative samples selected from all tobacco leaf samples in stock include: Conduct near infrared detection on all tobacco leaves in stock to obtain near infrared spectra of all tobacco leaves in stock; Based on the near-infrared spectra of all stock tobacco leaves, the Kennard-Stone algorithm was used to screen out several representative samples from all stock tobacco leaf samples.

3. The flue gas index prediction method based on LASSO and partial least squares regression according to claim 1 is characterized in that: Based on the near-infrared spectra of all tobacco leaves in stock, the Kennard-Stone algorithm was used to screen out several representative samples from all tobacco leaves in stock, including: All near-infrared spectral samples of tobacco leaves in stock are regarded as candidate samples of the training set, and samples are selected from the candidate samples to enter the training set in turn; Among all candidate samples, select the two samples with the farthest Euclidean distance to enter the training set; For each remaining sample, calculate the Euclidean distance from the sample to each known sample in the training set; Find the two samples closest to the selected samples in the training set, and select the one with the farther distance between them into the training set; Repeat the steps of selecting samples to enter the training set until the number of samples in the training set reaches the preset requirement.

4. The flue gas index prediction method based on LASSO and partial least squares regression according to claim 1 is characterized in that: The representative samples screened out are rolled to obtain a number of rolled single-ingredient cigarette samples, specifically comprising: Set the rolling processing parameters: drum speed 8r / min-12r / min, hot air temperature 100℃-120℃, drum wall temperature 110℃-130℃, outlet moisture content 12%-13%; The representative samples were rolled according to the rolling processing parameters, wherein the control requirements for various indicators of the cigarettes are as follows: the filter rod uses a 24.1mm*100mm*2700Pa ordinary acetate filter rod, the cigarette paper uses a 26.5mm*28.5g / m2A70CU wood pulp cross-grain cigarette paper, the tipping paper uses a 64mm hot stamping tipping paper without perforations, the draw resistance control requirement is 1000Pa-1100Pa, the circumference control requirement is 24.1mm-24.5mm, the hardness control requirement is 58%-78%, the quality control requirement for a single cigarette is 0.82g-0.92g, the quality requirement for 20 cigarettes is 16.4g-18.4g, the length control requirement is 83.5mm-84.5mm, and the mass fraction control requirement for moisture content is 11.00%-13.00%.

5. The flue gas index prediction method based on LASSO and partial least squares regression according to claim 1 is characterized in that: The determination of the smoke index content and chemical component content of the rolled single-ingredient cigarette sample corresponding to each representative sample specifically includes: The tar content in the smoke index of the rolled single-ingredient cigarette sample corresponding to each representative sample is measured by a smoking machine method; Determining the nicotine content in the smoke indicators of the rolled single-ingredient cigarette samples corresponding to each of the representative samples using gas chromatography; Determine the carbon monoxide content in the smoke index of the rolled single-ingredient cigarette sample corresponding to each representative sample using a non-scattering infrared method; Determine the content of conventional chemical components of the rolled single-ingredient cigarette samples corresponding to each representative sample using a continuous flow method, wherein the conventional chemical components include reducing sugars, total alkaloids, total nitrogen, chlorine, potassium, and starch; Determine the pH value of the rolled single-ingredient cigarette sample corresponding to each representative sample using a pH meter method; Ion chromatography is used to determine the content of inorganic anions in the rolled single-ingredient cigarette samples corresponding to each representative sample, wherein the inorganic anions include sulfate and phosphate; Determining the content of inorganic cations in the rolled single-ingredient cigarette samples corresponding to each representative sample using atomic absorption spectrometry, wherein the inorganic cations include magnesium ions and calcium ions; Liquid chromatography was used to determine the content of polyphenols in the rolled single-ingredient cigarette samples corresponding to the representative samples, wherein the polyphenols included neochlorogenic acid, chlorogenic acid, cryptochlorogenic acid, scopoletin, and rutin; The GC-MS / MS method is used to determine the content of polybasic acids and higher fatty acids in the rolled single-ingredient cigarette samples corresponding to each representative sample, wherein the polybasic acids include oxalic acid, malonic acid, succinic acid, malic acid, citric acid and vanillic acid, and the higher fatty acids include myristic acid, hexadecanoic acid, linoleic acid, oleic acid + linolenic acid, octadecanoic acid and eicosanoic acid; An amino acid analyzer is used to determine the content of free amino acids in the rolled single-ingredient cigarette samples corresponding to each representative sample, wherein the free amino acids include aspartic acid, threonine, serine, asparagine, glutamic acid, glutamine, glycine, alanine, valine, cystine, methionine, isoleucine, leucine, tyrosine, phenylalanine, 4-aminobutyric acid, lysine, histidine, tryptophan, arginine, and proline; The content of Amadori compounds in the rolled single-ingredient cigarette samples corresponding to each representative sample was determined by HPLC-MS / MS, wherein the Amadori compounds include Glu-An, Fru-Amb, Fru-His, Fru-Pro, Fru-Val, Fru-Thr, Fru-Gly, Fru-Ala, Fru-Asn, Fru-Asp, Fru-Gln, Fru-Glu, Fru-Ile, Fru-Leu, Fru-Tyr, Fru-Phe and Fru-Trp; Determine the content of dichloromethane extract of the rolled single-ingredient cigarette sample corresponding to each representative sample by using a filtration method; Determining the solanesol content of the rolled single-ingredient tobacco samples corresponding to each representative sample using high performance liquid chromatography; The content of neophytadiene in the rolled single-ingredient cigarette sample corresponding to each representative sample was determined by back-flushing gas chromatography.

6. The flue gas index prediction method based on LASSO and partial least squares regression according to claim 1 is characterized in that: The LASSO regression algorithm is used to screen out several important variables from the flue gas index content and chemical component content corresponding to each representative sample, specifically including: The objective function of the linear regression model for smoke index content is: J(w)=∑(y-Vw) 2 +λ||w||1=∑(y-Vw) 2 +∑λ|w|=ESS(w)+λl1(w) (1) Wherein, y represents the smoke index content, which is the dependent variable. The smoke index content includes tar content, nicotine content or CO content. V=[v1,…,v t ] represents the content of all chemical composition indicators, which is the independent variable, t represents the number of chemical composition indicators, w=[w1,…,w t ] T represents the regression coefficient corresponding to each independent variable, λ||w||1 represents the penalty term of the L1 norm, ESS(w) represents the sum of squared errors of n groups of observation data corresponding to n representative samples, n represents the number of representative samples screened, and λl1(w) represents the penalty term; Using the coordinate descent method, control the other regression parameters w unchanged, and for one of the w in the objective function J(w) q Find the partial derivative, q = 1, 2, ..., t, and so on, Find the partial derivatives of the remaining regression parameters, and finally make each derivative function equal to 0, so as to obtain the objective function that reaches the global minimum. at this time, Taking partial derivative of formula (2), we get: in, Since the penalty term is not differentiable, the following derivatives are calculated: Let the sum of the two partial derivatives of formula (3) and formula (4) be equal to 0, and we can get: From this we can get: All regression coefficients w q Remove the independent variables that are 0 and retain all regression coefficients w q Independent variables that are not 0 are considered as important variables that are screened out.

7. The flue gas index prediction method based on LASSO and partial least squares regression according to claim 6 is characterized in that: The LASSO regression algorithm is used to screen out several important variables from the flue gas index content and chemical component content corresponding to each representative sample, and further includes: The cross-validation method is used to determine the value of the penalty coefficient λ, which includes: The dataset is divided into several parts, one of which is selected as the validation set each time, and the rest are used as the training set. The optimal λ value is selected by adjusting the value of λ and calculating the performance index of the model on the validation set.

8. The flue gas index prediction method based on LASSO and partial least squares regression according to claim 1 is characterized in that: The smoke index content prediction model is established by using the selected important variables and adopting the partial least squares regression algorithm, which specifically includes: From the three groups of independent variables corresponding to the screened tar content, nicotine content and CO content, select the independent variables that appear at least once as the independent variables of the partial least squares regression model; Let p dependent variables y1,…,y p With m independent variables x1,…,x m The standardized variables are homogenized, and the n-times standardized observation data matrices corresponding to the n representative samples of the independent variable group and the dependent variable group are recorded as: Extract the first pair of components of the two variable groups respectively, and maximize the correlation between the first pair of components of the two variable groups. Let the first pair of components of the two groups of variables be u1 and v1 respectively, u1 represents the independent variable set X = [x1, ..., x m ] T The linear combination u1=α 11 x1+…+α 1m x m =ρ (1)T X, v1 represents the dependent variable set Y = [y1,…,y p ] T The linear combination v1=β 11 y1+…+β 1m y m =γ (1)T Y; According to the standardized observation data matrices A and B of the two sets of variables, the score vectors of the first pair of components are calculated by the following formula, which are respectively denoted as and The score vector of the first pair of components and The inner product of , calculate the covariance index v(u1, v1) of the first pair of components u1 and v1; In order for u1 and v1 to extract as much variation information of their respective variable groups as possible, the correlation between u1 and v1 is required to be maximized. The following conditional extreme value problem is calculated: According to the Lagrange multiplier method, the above conditional extreme value problem is transformed into: solving the unit vector ρ (1) and γ (1) , so that θ1=ρ (1)T A T Bγ (1) reach the maximum; By calculating the m×m matrix M=A T BB T The eigenvalues ​​and eigenvectors of A, and the largest eigenvalue of M is The corresponding unit eigenvector is the desired solution ρ (1) , and γ (1) The following formula can be used to calculate ρ (1) Calculation yields: Establish y1,…,y p The regression model for u1 and x1,…,x m Regression model for v1: Among them, σ (1) =[σ 11 ,…,σ 1m ] T , τ (1) =[τ 11 ,…,τ 1p ] T They represent the parameter vectors in the many-to-one regression model, and A1 and B1 represent the residual matrices; The regression coefficient vector σ is calculated by the following formula (1) and τ (1) The least squares estimation results of : σ (1) and τ (1) It is called model effect loading; Use residual matrices A1 and B1 to replace A and B, repeat the above steps, and record Then the residual matrix If the absolute value of the elements in the residual matrix B1 is approximately 0, it is considered that the accuracy of the regression model established with the first component has met the requirements, and the extraction of components can be stopped. Otherwise, replace A and B with residual matrices A1 and B1, and repeat the above steps to obtain r (2) =[a 21 ,…,a 2m ] T ,c (2) =[β 21 ,…,b 2p ] T (15) and is the score vector of the second pair of components, and and are the loadings of the second pair of components of X and Y, respectively. Assume that the rank of matrix A is r≤min(n-1,m), then there are r principal components u1, u2, ..., u r , such that: will u k =α k1 x1+…+α km x m (k=1,2,…,r)Substitute into Y=u1τ (1) +...+u r τ (r) , that is, the partial least squares regression equation of p dependent variables is: y j =c j1 x1+…+c jm x m ,j=1,2,…,p (18)。 9. The flue gas index prediction method based on LASSO and partial least squares regression according to claim 8 is characterized in that: The predicting of the smoke index content in the tobacco leaves according to the smoke index content prediction model specifically includes: Detecting the content of important variables of the rolled single-ingredient cigarette sample corresponding to the sample to be tested; The contents of important variables of the sample to be tested are input into the smoke index content prediction model to obtain the smoke index content of the sample to be tested.

10. The flue gas index prediction method based on LASSO and partial least squares regression according to claim 1 is characterized in that: The flue gas index prediction method based on LASSO and partial least squares regression also includes: The prediction effect of the flue gas index content prediction model is evaluated, specifically including: Statistical indicators are used to evaluate the prediction effect of the flue gas index content prediction model, wherein the statistical indicators include: at least one of the mean square error, the root mean square error and the determination coefficient. The closer the determination coefficient is to 1, the smaller the mean square error and the root mean square error are, indicating that the prediction accuracy of the flue gas index content prediction model is better.

Images