Training method and prediction method of tobacco product producing area style prediction model

Abstract

The invention relates to a training method and a prediction method of a tobacco product producing area style prediction model, and relates to the technical field of tobacco production. The training method comprises the following steps: determining a plurality of combination formulas of the tobacco product; selecting tobacco product raw material samples of a target producing area and other producing areas according to each combination formula in the plurality of combination formulas to form a tobacco product sample corresponding to each combination formula; determining first characteristic data related to the characteristic temperature of the sample and second characteristic data related to the mass change rate corresponding to the characteristic temperature of the sample from the first thermogravimetric analysis data of the tobacco product sample corresponding to each combination formula; and taking at least one of the first characteristic data and the second characteristic data of the plurality of combination formulas and the first chemical component data as input, taking the production place of the tobacco product raw material sample with the maximum mass ratio in the tobacco product samples as output, and training the production place style prediction model until a training ending condition is met.

Description

Technical Field

[0001] This disclosure relates to the field of tobacco production technology, and in particular to a training method and prediction method for a region-of-origin style prediction model for tobacco products. Background Technology

[0002] Tobacco products are made by combining various tobacco raw materials according to a specific formula. Taking tobacco leaves as the raw material and cigarettes as the finished product, different combinations of tobacco leaves will result in different aromas and tastes in the cigarettes. Origin style is an indicator of the aroma and taste of cigarettes. While the tobacco leaves used as raw materials can have a clearly defined origin, cigarettes made from a combination of tobacco leaves may not have a specific origin. However, if their aroma and taste are similar to those of tobacco leaves from a particular origin, then their origin style can be identified as that specific origin. For example, cigarettes made from a combination of tobacco leaves from Yunnan, Guizhou, and Hunan may, under a specific formula, ultimately have an aroma and taste close to that of Yunnan tobacco leaves, thus their origin style is identified as Yunnan. Summary of the Invention

[0003] According to some embodiments of this disclosure, a training method for a tobacco product origin style prediction model is provided, comprising: determining multiple combination formulations of tobacco products, each combination formulation indicating the mass proportion of tobacco product raw materials from a target origin and tobacco product raw materials from other origins in the tobacco product, wherein the mass proportion of tobacco product raw materials from the target origin is different in each of the multiple combination formulations; selecting tobacco product raw material samples from the target origin and tobacco product raw material samples from other origins according to each of the determined multiple combination formulations to form a tobacco product sample corresponding to each combination formulation; obtaining first thermogravimetric analysis data and first chemical composition data of the tobacco product sample corresponding to each combination formulation, wherein the first thermogravimetric analysis data includes multiple data of the pyrolysis process of the tobacco product sample. A first pyrolysis temperature and a mass change rate corresponding to each of the plurality of first pyrolysis temperatures, wherein the first chemical composition data includes the content of each of the plurality of chemical components in the tobacco product sample; a first characteristic data related to the sample characteristic temperature among the plurality of first pyrolysis temperatures and a second characteristic data related to the mass change rate corresponding to the sample characteristic temperature are determined from the first thermogravimetric analysis data; the origin style prediction model is trained using at least one of the first characteristic data and the second characteristic data of the plurality of combined formulations and the first chemical composition data of the plurality of combined formulations as input and the origin of the tobacco product raw material sample with the largest mass proportion in the tobacco product sample as output, until the training termination condition is met.

[0004] According to some embodiments of this disclosure, a method for predicting the origin style of a tobacco product to be tested is provided, comprising: acquiring third thermogravimetric analysis data and third chemical composition data of the tobacco product to be tested, wherein the tobacco product to be tested is composed of multiple tobacco product raw materials from different origins, the third thermogravimetric analysis data includes multiple third pyrolysis temperatures of the pyrolysis process of the tobacco product to be tested and the mass change rate corresponding to each of the multiple third pyrolysis temperatures, and the third chemical composition data includes the content of each of the multiple chemical components in the tobacco product to be tested; determining third feature data related to a characteristic temperature among the multiple third pyrolysis temperatures and fourth feature data related to the mass change rate corresponding to the characteristic temperature from the third thermogravimetric analysis data; and predicting the origin style of the tobacco product to be tested based on at least one of the third feature data and the fourth feature data and the third chemical composition data, using an origin style prediction model trained according to the training method of any of the above embodiments.

[0005] According to further embodiments of this disclosure, a method for determining a target combination formulation of a tobacco product is provided, comprising: selecting multiple tobacco product raw materials from different origins from a tobacco product raw material library based on each of a plurality of candidate combination formulations to form a tobacco product corresponding to each candidate combination formulation; for each candidate combination formulation, performing the following steps to construct a combination formulation-origin style database: predicting the origin style of the tobacco product corresponding to each candidate combination formulation using an origin style prediction method according to any of the above embodiments; storing each candidate combination formulation and the origin style of the tobacco product corresponding to each candidate combination formulation in the combination formulation-origin style database; and retrieving a combination formulation corresponding to a target origin style from the combination formulation-origin style database to determine a target combination formulation.

[0006] According to further embodiments of this disclosure, a training apparatus for a prediction model of the origin style of tobacco products is provided, comprising: a first determining module configured to determine a plurality of combined formulations of tobacco products, each combined formulation indicating the mass proportion of tobacco product raw materials from a target origin and tobacco product raw materials from other origins in the tobacco product, wherein the mass proportion of the tobacco product raw materials from the target origin is different in each of the plurality of combined formulations; a selecting module configured to select tobacco product raw material samples from the target origin and tobacco product raw material samples from other origins according to each of the determined plurality of combined formulations to form a tobacco product sample corresponding to each combined formulation; and an acquiring module configured to acquire first thermogravimetric analysis data and first chemical composition data of the tobacco product sample composed of the tobacco product raw material samples corresponding to each combined formulation, wherein the first thermogravimetric analysis data... The first chemical composition data includes the content of each of the multiple chemical components in the tobacco product sample, based on a plurality of first pyrolysis temperatures of the pyrolysis process of the tobacco product sample and the mass change rate corresponding to each of the plurality of first pyrolysis temperatures; the second determination module is configured to determine, from the first thermogravimetric analysis data, first feature data related to the sample characteristic temperature among the plurality of first pyrolysis temperatures and second feature data related to the mass change rate corresponding to the sample characteristic temperature; the training module is configured to take at least one of the first feature data and the second feature data and the first chemical composition data as input, and take the predicted origin of the tobacco product raw material sample with the largest mass proportion in the tobacco product sample as output, and train the origin style prediction model until the training termination condition is met.

[0007] According to some embodiments of this disclosure, an origin style prediction device for a tobacco product to be tested is provided, comprising: an acquisition module configured to acquire third thermogravimetric analysis data and third chemical composition data of the tobacco product to be tested, wherein the tobacco product to be tested is composed of multiple tobacco product raw materials from different origins, the third thermogravimetric analysis data including multiple third pyrolysis temperatures of the pyrolysis process of the tobacco product to be tested and a mass change rate corresponding to each of the multiple third pyrolysis temperatures, and the third chemical composition data including the content of each of the multiple chemical components in the tobacco product to be tested; a determination module configured to determine, from the third thermogravimetric analysis data, third feature data related to a characteristic temperature among the multiple third pyrolysis temperatures, and fourth feature data related to the mass change rate corresponding to the characteristic temperature; and a prediction module configured to predict the origin style of the tobacco product to be tested based on at least one of the third feature data and the fourth feature data and the second chemical composition data, using an origin style prediction model trained according to the aforementioned training device.

[0008] According to some embodiments of the present disclosure, an electronic device is provided, including: a memory; and a processor coupled to the memory, the processor being configured to execute, based on instructions stored in the memory device, a method for training a tobacco product origin style prediction model, a method for predicting the origin style of a tobacco product to be tested, or a method for determining a target combination formulation of a tobacco product.

[0009] According to some embodiments of the present disclosure, a computer-readable storage medium is provided, on which computer instructions are stored, which, when executed by a processor, implement the training method for a tobacco product origin style prediction model, the method for predicting the origin style of a tobacco product to be tested, or the method for determining a target combination formulation of a tobacco product.

[0010] According to some embodiments of this disclosure, a computer program product is also provided, including instructions that, when executed by a processor, cause the processor to perform a training method for a tobacco product origin style prediction model, a method for predicting the origin style of a tobacco product to be tested, or a method for determining a target combination formulation of a tobacco product, as described in any of the above embodiments. Attached Figure Description

[0011] The accompanying drawings, which form part of this specification, illustrate embodiments of this disclosure and, together with the specification, serve to explain the principles of this disclosure.

[0012] This disclosure will become clearer with reference to the accompanying drawings and the following detailed description, wherein:

[0013] Figure 1 A flowchart illustrating a training method for a tobacco product origin style prediction model according to some embodiments of the present disclosure;

[0014] Figure 2 A schematic diagram of DTG curves according to some embodiments of the present disclosure is shown;

[0015] Figure 3 This diagram illustrates the structure of a convolutional neural network model according to some embodiments of the present disclosure;

[0016] Figure 4 A flowchart illustrating a specific example of a training method for an origin style prediction model according to some embodiments of the present disclosure;

[0017] Figure 5 A flowchart illustrating a method for predicting the origin style of a tobacco product under test according to some embodiments of the present disclosure is shown.

[0018] Figure 6A flowchart illustrating a method for determining a target combination formulation according to some embodiments of this disclosure;

[0019] Figure 7 A block diagram of a training apparatus for a tobacco product origin style prediction model according to some embodiments of the present disclosure is shown.

[0020] Figure 8 A block diagram of an origin style prediction device for a tobacco product to be tested, according to some embodiments of the present disclosure, is shown.

[0021] Figure 9 A block diagram of an electronic device according to some embodiments of the present disclosure is shown;

[0022] Figure 10 Block diagrams of electronic devices according to other embodiments of the present disclosure are shown. Detailed Implementation

[0023] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present disclosure.

[0024] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0025] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this disclosure or its application or use.

[0026] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0027] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0028] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0029] As mentioned earlier, combining tobacco raw materials from different origins according to a specific formula can create tobacco products that exhibit the style of a particular origin.

[0030] In related technologies, machine learning models are commonly used to predict the regional style of tobacco products. However, during the training of machine learning models, the impact of the quality proportion of tobacco raw materials from different regions in the blending formula on the overall style of the final tobacco product is often overlooked.

[0031] The inventors of this disclosure have discovered that the regional characteristics of tobacco raw materials, which constitute a large proportion of the composition of a blend, have a significant dominant effect on the overall style of the final tobacco product. In other words, the aroma, taste, and other characteristics of tobacco raw materials that make up a large proportion of the blend will largely determine the final style of the tobacco product. For example, if the tobacco leaves from a certain region (such as Yunnan) constitute the largest proportion of the blend, then the regional style of the tobacco product made based on that blend will most likely reflect the regional style of that tobacco leaf (such as the Yunnan style).

[0032] In view of this, this disclosure proposes a training method for a tobacco product origin style prediction model. Based on the research finding that the origin style of tobacco products is dominated by the origin style of the tobacco product raw material (also known as the "main source tobacco product raw material") with the largest mass proportion in the blend, the thermogravimetric analysis data and chemical composition data of tobacco products corresponding to each blend are used as the data basis for model training during the training process. This effectively quantifies the influence of the mass proportion of tobacco product raw materials from different origins in tobacco products on the origin style of tobacco products, thus providing a scientific and accurate tool for predicting the origin style of tobacco products and optimizing the design of blends.

[0033] Figure 1 A flowchart illustrating a training method for a tobacco product origin style prediction model according to some embodiments of the present disclosure is provided.

[0034] like Figure 1 As shown, in step 110, multiple combination formulations of the tobacco product are determined.

[0035] The formulation of a tobacco product may include the mass percentage of multiple tobacco product ingredients in the tobacco product. The mass percentage can be presented as an absolute value or a relative value. In the absolute value form, the formulation may include, for example, the mass of each tobacco product ingredient (e.g., in grams, milligrams, micrograms, etc.), thus indirectly reflecting the mass percentage. In the relative value form, the formulation may include, for example, the ratio of the mass of each tobacco product ingredient to the mass of the tobacco product (e.g., in percentage form), thus directly reflecting the mass percentage.

[0036] Here, each of the multiple blending formulations indicates the mass percentage of tobacco product raw materials from the target origin and tobacco product raw materials from other origins in the tobacco product. The mass percentage of tobacco product raw materials from the target origin varies in the multiple blending formulations.

[0037] In some embodiments, the mass percentage of tobacco product raw materials from the target origin in multiple blended formulations can be distributed in an equally spaced gradient. That is, the mass percentage of tobacco product raw material samples from the target origin in the tobacco product sample in multiple blended formulations can be in the form of an arithmetic sequence. For example, the mass percentage of tobacco product raw material samples from the target origin in the tobacco product sample can vary from 10% to 90% at 10% intervals. The mass percentage of tobacco product raw material samples from the target origin in the tobacco product sample in multiple blended formulations can be 10%, 20%, 30%, 40%...90%, respectively.

[0038] In this approach, by regularly varying the mass proportion of tobacco product raw materials from the target origin in multiple combination formulations, the trained origin style prediction model can comprehensively and systematically learn the quantitative relationship between the mass proportion of tobacco product raw materials and the origin style of tobacco products, thereby improving the scientific rigor and reliability of the model training.

[0039] It should be understood that, in this disclosure, tobacco product samples are products made wholly or partially from tobacco leaves and intended for smoking, chewing, snorting, or other use. Tobacco product samples may include, but are not limited to, cigarettes, cigars, pipe tobacco, or hookah. Cigarettes, distinguished by their tobacco leaf formulation and processing method, may include, but are not limited to, flue-cured tobacco, burley tobacco, aromatic tobacco, or sun-cured tobacco. Tobacco product samples release chemicals such as nicotine (a common term for "nicotine") through heating (e.g., heated non-combustible tobacco products) and / or combustion (e.g., cigarettes, cigars, etc.). Tobacco product samples may or may not have a cigarette wrapping and may or may not have a filter. In this disclosure, for ease of description, cigarettes are sometimes used as examples of tobacco product samples. However, it should be recognized that the various features or limitations described herein regarding cigarettes also apply to other types of tobacco product samples.

[0040] Samples of raw materials for tobacco products may include tobacco leaves in various forms, such as tobacco leaves or tobacco shreds, tobacco sheets, tobacco powder or tobacco blocks processed from tobacco leaves.

[0041] In step 120, tobacco product raw material samples from the target origin and tobacco product raw material samples from other origins are selected according to each of the determined multiple combination formulas to form the tobacco product sample corresponding to each combination formula.

[0042] In some embodiments, tobacco product raw material samples from the target origin can be selected according to the mass percentage of the target origin in the total tobacco product sample in each blend, and tobacco product raw material samples from other origins can be selected according to the mass percentage of the other origins in the total tobacco product sample in each blend, to form the tobacco product sample corresponding to each blend. For example, Monte Carlo sampling can be used to select tobacco product raw material samples from the target origin and tobacco product raw material samples from other origins from a tobacco product raw material library to form the tobacco product sample corresponding to each blend.

[0043] It should be understood that each combination formula corresponds to a tobacco product sample consisting of multiple tobacco product raw material samples from different origins (i.e., the target origin and other origins).

[0044] To improve the generalization ability of the model training, when selecting tobacco product raw material samples, it is advisable to select samples covering multiple different aroma styles. In some embodiments, tobacco product raw material samples from the target origin and / or other origins may cover three aroma styles: light, medium, and strong. For similar purposes, when selecting tobacco product raw material samples, it is also advisable to further consider selecting tobacco product raw material samples covering multiple growth parts for each origin. Here, the growth part of the tobacco product raw material is, for example, the growth part of the tobacco leaf on the tobacco plant, such as the upper, middle, or lower part. In some embodiments, tobacco product raw material samples from the target origin and / or other origins may include upper tobacco leaves, middle tobacco leaves, and lower tobacco leaves.

[0045] In this way, the tobacco product samples corresponding to each combination formula can more comprehensively cover the characteristics of various types of tobacco product raw materials. Subsequently, training the origin style prediction model using tobacco product samples corresponding to each combination formula can help the origin style prediction model learn a wider range of characteristics of tobacco product raw materials from different origins, improving the model's generalization ability and thus contributing to improved prediction accuracy.

[0046] In step 130, the first thermogravimetric analysis data and the first chemical composition data of the tobacco product sample corresponding to each combination formulation are obtained.

[0047] The first thermogravimetric analysis data for each tobacco product sample corresponding to the combined formulation includes multiple first pyrolysis temperatures of the tobacco product sample during the pyrolysis process and the mass change rate corresponding to each of the multiple first pyrolysis temperatures.

[0048] In some embodiments, the first thermogravimetric analysis (TGA) data of the tobacco product sample corresponding to each combined formulation can be obtained from the second TGA data of each of the plurality of tobacco product raw material samples constituting the tobacco product sample. The second TGA data includes a plurality of second pyrolysis temperatures of the pyrolysis process of each tobacco product raw material sample and the mass change rate of each tobacco product raw material sample corresponding to each of the plurality of second pyrolysis temperatures.

[0049] As one implementation method, the first thermogravimetric analysis (TGA) data of a tobacco product sample can be determined based on the average of the second TGA data of multiple tobacco product raw material samples. For example, a tobacco product sample is known to consist of 5 portions of tobacco leaves A from Yunnan, 3 portions of tobacco leaves B from Fujian, and 2 portions of tobacco leaves C from Henan, each portion weighing 100g. The second TGA data for the 5 portions of tobacco leaves A are (S11, S12, S13, S14, S15), the second TGA data for the 3 portions of tobacco leaves B are (S21, S22, S23), and the second TGA data for the 2 portions of tobacco leaves C are (S31, S32). The first thermogravimetric analysis data S of the tobacco product samples can be the average of the second thermogravimetric analysis data of 5 tobacco leaf A samples, 3 tobacco leaf B samples, and 2 tobacco leaf C samples, i.e., S = (S11 + S12 + S13 + S14 + S15 + S21 + S22 + S23 + S31 + S32) / 10.

[0050] It is understood that the thermogravimetric analysis data of tobacco product raw material samples can be obtained by conducting pyrolysis experiments on the samples and recording multiple pyrolysis temperatures and the corresponding rate of mass change at each temperature. The rate of mass change can be determined by calculating the derivative of the mass of the tobacco product raw material with respect to temperature during pyrolysis.

[0051] The following example uses a certain tobacco leaf as a sample of raw materials for tobacco products to illustrate the process of conducting pyrolysis experiments on raw materials for tobacco products.

[0052] First, a tobacco leaf is placed in a constant temperature and humidity chamber for a period of time to equilibrate. Then, the tobacco leaf is ground, and the ground tobacco powder is filtered through a sieve and thoroughly mixed. For example, 20 mg of the tobacco powder is weighed as a sample. Next, the sample is heated to a higher temperature (e.g., 873 Kelvin (K)) at the same rate. The relationship between the sample's mass and temperature during the heating process is recorded, thus obtaining the thermogravimetry (TG) curve. To make the pyrolysis behavior of tobacco more apparent, the TG curve is differentiated to obtain the derivative thermogravimetry (DTG) curve, which reflects the relationship between the rate of mass change and temperature, i.e., the "thermogravimetry data" mentioned in this paper. If the thermogravimetry data is displayed as a graph (i.e., the DTG curve), the horizontal axis represents temperature (i.e., the "pyrolysis temperature" mentioned in this paper), in Kelvin (K), and the vertical axis represents the rate of mass change of the tobacco leaf during the pyrolysis process, in percentage / Kelvin (% / K). This curve is also called a thermal analysis spectrum.

[0053] In some embodiments, the initial thermogravimetric analysis (TGA) data of the tobacco product sample can be obtained by performing a TGA experiment (i.e., a pyrolysis experiment) on the tobacco product sample. Thus, the initial TGA data obtained experimentally can more accurately reflect the thermal decomposition characteristics of the tobacco product sample based on the combined formulation, helping to improve the accuracy of subsequent model training.

[0054] Here, the pyrolysis experiment on tobacco products is similar to the pyrolysis experiment on tobacco product raw materials. For details, please refer to the relevant description of the pyrolysis experiment on tobacco product raw materials above.

[0055] Figure 2 A schematic diagram of DTG curves according to some embodiments of the present disclosure is shown.

[0056] For example, in the example above, 800 data points between 401.138 K and 801.138 K can be taken as the first thermogravimetric analysis data for the tobacco product sample with a step size of 0.5 K, and displayed in the form of a curve to obtain the following results. Figure 2 The DTG curve is shown by the solid black line.

[0057] It can be understood that the pyrolysis temperature in the thermogravimetric analysis (TG) data (i.e., the x-axis of the TG curve) reflects the temperature range of tobacco products during pyrolysis, with different pyrolysis temperatures potentially corresponding to different stages of the pyrolysis reaction. The mass change rate in the TG data (i.e., the y-axis of the TG curve) reflects the intensity of the reaction during pyrolysis, thus revealing the kinetic characteristics of the pyrolysis reaction. In other words, the pyrolysis temperature and mass change rate in the TG data of tobacco products can reflect the pyrolysis characteristics of tobacco products from two different dimensions.

[0058] The first chemical composition data for each tobacco product sample corresponding to each combination formulation includes the content of each chemical component among multiple chemical components in the tobacco product sample.

[0059] In some embodiments, the multiple chemical components may include at least one of total nitrogen, nicotine, total sugar, reducing sugar, potassium, and chlorine. The second chemical component data includes the content of each of the multiple chemical components in each tobacco product raw material sample.

[0060] In some embodiments, the first chemical composition data of a tobacco product sample can be obtained based on the second chemical composition data of each of the plurality of tobacco product raw material samples that make up the tobacco product sample. For example, a continuous flow method can be used to obtain the content of each of the plurality of chemical components in each tobacco product raw material sample.

[0061] As one implementation method, the first chemical composition data of a tobacco product sample can be determined based on the average of the second chemical composition data of multiple tobacco product raw material samples.

[0062] Continuing with the example above, taking total nitrogen as an example of chemical composition, assume that the total nitrogen content in the second chemical composition data of 5 tobacco leaf samples A is (Q11, Q12, Q13, Q14, Q15), the total nitrogen content in the second chemical composition data of 3 tobacco leaf samples B is (Q21, Q22, Q23), and the total nitrogen content in the second chemical composition data of 2 tobacco leaf samples C is (Q31, Q32). The total nitrogen content Q in the first chemical composition data of the tobacco product sample is calculated as follows: Q = (Q11 + Q12 + Q13 + Q14 + Q15 + Q21 + Q22 + Q23 + Q31 + Q32) / 10.

[0063] In some embodiments, the first chemical composition data of a tobacco product sample can be obtained by conducting chemical composition analysis experiments on the tobacco product sample for multiple chemical components. For example, a continuous flow method can be used to obtain the content of each chemical component among multiple chemical components in the tobacco product sample. Thus, the first chemical composition data obtained experimentally can more accurately reflect the chemical composition properties of the tobacco product sample made based on the combined formulation, which helps to improve the accuracy of subsequent model training.

[0064] In step 140, first characteristic data related to the sample characteristic temperature among a plurality of first pyrolysis temperatures and second characteristic data related to the mass change rate corresponding to the sample characteristic temperature are determined from the first thermogravimetric analysis data.

[0065] In some embodiments, a subset of feature points can be selected from the DTG curve corresponding to the first thermogravimetric analysis data. The abscissa value of the subset of feature points in the DTG curve is determined as the feature temperature in the feature data, and the ordinate value of the subset of feature points in the DTG curve is determined as the mass change rate corresponding to the feature temperature in the feature data.

[0066] In step 150, the origin style prediction model is trained by taking at least one of the first and second feature data of multiple combination formulas and the first chemical composition data of multiple combination formulas as inputs and the origin of the tobacco product raw material sample with the largest mass proportion in the tobacco product sample as output, until the training termination condition is met.

[0067] It should be understood that for each tobacco product sample corresponding to a combination formulation, at least one feature data of the tobacco product sample can be extracted as a feature data corresponding to that combination formulation, and the first chemical component data of the tobacco product sample can be extracted as a chemical component data corresponding to that combination formulation. The training data used to train the origin style prediction model includes multiple feature data corresponding to multiple combination formulations, and multiple chemical component data corresponding to multiple combination formulations.

[0068] For example, during training, one set of feature data (i.e., at least one of the first and second feature data of the tobacco product sample corresponding to this combination formula) and one set of chemical composition data (i.e., the first chemical composition data of the tobacco product sample corresponding to this combination formula) are input as training data into the regional style prediction model, until multiple sets of training data corresponding to multiple combination formulas are input. In this way, the regional style prediction model can learn the data characteristics under different combination formulas, thereby better understanding the contribution of tobacco product raw materials from different origins to the regional style of tobacco products.

[0069] In some embodiments, the training termination condition may be, for example, reaching a specified number of training iterations and / or the difference between the predicted origin of the tobacco product raw material sample with the largest mass percentage in the tobacco product sample and the actual origin of the tobacco product raw material sample with the largest mass percentage in the tobacco product sample is less than a threshold. For example, the difference between the predicted origin and the actual origin of the tobacco product raw material sample with the largest mass percentage can be characterized by the loss function value of the origin style prediction model.

[0070] In some embodiments, the first characteristic data, the second characteristic data, and the first chemical composition data of multiple combined formulations can be used together as input to the origin style prediction model.

[0071] It is understood that the types of data input into the origin style prediction model during the subsequent application of the origin style prediction model are the same as the types of data input into the origin style prediction model during the training process.

[0072] In the above embodiments, multiple combination formulas are obtained by designing different proportions of the target origin and other origins. The mass proportion of tobacco product raw materials from the target origin is different in each of the multiple combination formulas. Then, for each combination formula, tobacco product raw material samples from the target origin and other origins are precisely selected according to the designed proportions to form tobacco product samples with specific origin proportion characteristics.

[0073] The tobacco product samples obtained in this way can effectively simulate the situation after the raw materials of tobacco products from different origins are mixed during the tobacco production process, providing a reliable data basis for studying the interaction of raw materials of tobacco products from different origins in the combination formula and their impact on the final product style.

[0074] Thus, by using feature data and chemical composition data extracted from thermogravimetric analysis data of tobacco product samples corresponding to multiple different combination formulations, the regional style prediction model is trained. This allows the trained regional style prediction model to effectively quantify the impact of the mass proportion of tobacco product raw materials from different regions on the regional style of tobacco products, thereby accurately predicting the regional style of tobacco products. This provides a scientific and precise tool for predicting the regional style of tobacco products and optimizing the design of combination formulations.

[0075] In some embodiments, the prediction model may include a convolutional neural network model.

[0076] Figure 3 A schematic diagram of the structure of a convolutional neural network model according to some embodiments of the present disclosure is shown.

[0077] like Figure 3As shown, the convolutional neural network model mainly consists of convolutional layers, pooling layers, and fully connected layers. The input layer receives feature data characterizing tobacco product samples.

[0078] For example, the input feature dimension can be batch size × 1 × 48. Taking the first feature data, second feature data, and first chemical composition data as input, after processing by the first convolutional layer (e.g., a one-dimensional convolutional kernel of size 3), the feature dimension becomes batch size × 16 × 48. After processing by the second convolutional layer (e.g., a one-dimensional convolutional kernel of size 3), the feature dimension remains unchanged. Then, the multi-dimensional feature vector is flattened into a one-dimensional feature vector and input into a fully connected layer. The fully connected layer can output the origin of the tobacco product raw material sample with the largest mass proportion in the tobacco product sample as the predicted origin.

[0079] Here, the batch size can be set according to the hardware resources available for running the convolutional neural network model. For example, if there is ample memory, a larger batch size can be set to improve training speed.

[0080] In some embodiments, tobacco product raw materials from one or more origins other than the target origin are randomly selected as samples of tobacco product raw materials from other origins, according to the mass proportion of tobacco product raw materials from other origins indicated in each of the combined formulations. For example, Monte Carlo sampling can be used to randomly select from tobacco product raw materials from one or more origins other than the target origin.

[0081] In other words, samples of tobacco product raw materials from other origins can be selected from one or more origins other than the target origin.

[0082] For example, if the mass percentage of tobacco product raw material samples from the target origin is 90% and the mass percentage of tobacco product raw material samples from other origins is 10% (hereinafter referred to as "target origin: other origins = 90%: 10%), one or more tobacco product raw material samples from the target origin (e.g., Yunnan) can be selected from the tobacco product sample raw material library, and one or more tobacco product raw material samples from multiple origins other than the target origin (e.g., Fujian, Hunan, Henan) can be selected as tobacco product raw material samples from other origins, thus forming the tobacco product raw material samples corresponding to the "target origin: other origins = 90%: 10%" combination formula.

[0083] In the above embodiments, by randomly selecting tobacco product raw materials from multiple different origins as samples of tobacco product raw materials from other origins, the diversity of the samples can be significantly increased. This allows the training data used for training the origin style prediction model to more comprehensively cover the characteristics of tobacco product raw materials from different origins. This helps the trained origin style prediction model to accurately predict the origin style of complex combinations of tobacco product raw materials in subsequent applications.

[0084] In some embodiments, the tobacco product raw material with the largest weight percentage in each combination formulation originates from only one place (i.e., the tobacco product raw material from a single origin has the largest weight percentage).

[0085] In other words, when designing each blend, it is prohibited for the mass percentage of tobacco product raw materials from multiple different origins to simultaneously reach the maximum value. For example, multiple blends can include a blend such as "Yunnan: Fujian: Henan = 40%: 30%: 30%", but blends such as "Fujian: Yunnan: Guizhou: Shandong = 30%: 30%: 30%: 10%" or "Fujian: Yunnan = 50%: 50%" will not appear.

[0086] It should be noted that the core of designing combination formulations based on these constraints is to ensure that each combination formulation contains a clearly defined "primary source tobacco product ingredient." As mentioned earlier, the regional characteristics of this "primary source tobacco product ingredient" have a clear and dominant influence on the regional characteristics of the final tobacco product.

[0087] Thus, the training data obtained from the combined formulations designed based on these constraints will have strong directionality, enabling the trained regional style prediction model to clearly identify the regional style of the main source tobacco product raw materials that play a dominant role, thereby helping to improve the prediction accuracy and reliability of the regional style prediction model.

[0088] The feature data in step 140 will be illustrated below with reference to some embodiments.

[0089] In some embodiments, the characteristic temperature includes at least one of the starting temperature and the ending temperature among a plurality of first pyrolysis temperatures.

[0090] The first characteristic data determined from the first thermogravimetric analysis data includes at least one of the starting temperature and the ending temperature, and the second characteristic data includes at least one of the mass change rate corresponding to the starting temperature and the mass change rate corresponding to the ending temperature.

[0091] The initial temperature among multiple first pyrolysis temperatures can be used to indicate the start of the pyrolysis process, that is, when the tobacco product sample begins to undergo thermal decomposition. The termination temperature among multiple first pyrolysis temperatures can be used to indicate the termination of the pyrolysis process, that is, when the tobacco product sample has completed thermal decomposition.

[0092] For example, please continue to see Figure 2 As mentioned earlier, using a step size of 0.5 K, a total of 801 data points between 401.138 K and 801.138 K were taken as the thermogravimetric analysis data for this tobacco product sample, in order to obtain the following... Figure 2 The black solid line represents the DTG curve. That is, in Figure 2 In the example shown, the pyrolysis process starts at a temperature of 401.138 K and ends at a temperature of 801.138 K.

[0093] Select the starting point A1 and the ending point A2 as feature points from the DTG curve. The x-axis value of the starting point A1 is the starting temperature TA1, and the y-axis value is the mass change rate WA1 corresponding to the starting temperature. The x-axis value of the ending point A2 is the ending temperature TA2, and the y-axis value is the mass change rate WA2 corresponding to the ending temperature.

[0094] For example, any set of data from (TA1, TA2), (WA1, WA2), (TA1, TA2, WA1, WA2) and the first chemical composition data of tobacco product samples can be used as inputs to train the origin style prediction model.

[0095] In the above embodiments, since the starting temperature of the pyrolysis process marks the beginning of thermal decomposition of the tobacco product sample, and the ending temperature of the pyrolysis process marks the completion of thermal decomposition of the tobacco product sample, including the starting temperature and the corresponding mass change rate, the ending temperature and the corresponding mass change rate in the feature data can enable the trained origin style prediction model to effectively learn the thermal stability characteristics of the tobacco product sample, thereby helping to further improve the prediction accuracy and reliability of the origin style prediction model.

[0096] In some embodiments, the plurality of first pyrolysis temperatures form one or more temperature ranges. The mass change rate corresponding to the characteristic temperature includes at least one extreme value of the mass change rate of the tobacco product sample within one or more temperature ranges.

[0097] The first characteristic data determined from the first thermogravimetric analysis data includes the pyrolysis temperature corresponding to at least one extreme value, and the second characteristic data includes at least one extreme value. Here, at least one extreme value can include a maximum value and / or a minimum value. For example, if the rate of mass change is differentiated with respect to temperature, the extreme value is the rate of mass change corresponding to a derivative of 0. Alternatively, if the pyrolysis temperature and the rate of mass change are plotted as a curve, the extreme value corresponds to a peak or trough on the curve.

[0098] For example, such as Figure 2 As shown, the lower limit of the temperature range formed by the multiple first pyrolysis temperatures is 401.138 K, and the upper limit is 801.138 K. Multiple points (B1~B5) with a slope of 0 are selected as characteristic points on the DTG curve. Among them, B1 and B4 are the troughs on the DTG curve, and B2, B3, and B5 are the peaks on the DTG curve.

[0099] The vertical coordinates of these feature points (WB1~WB5) represent the extreme values ​​of the rate of mass change of the tobacco product sample within the temperature range of 401.138 K to 801.138 K, while the horizontal coordinates of these feature points (TB1~TB5) represent the temperatures corresponding to the extreme values ​​(i.e., the characteristic temperatures).

[0100] For example, any set of data from (TB1~TB5), (WB1~WB5), and ((TB1, WB1)~(TB5, WB5)) along with the first chemical composition data of tobacco products can be used as input to train the origin style prediction model.

[0101] In the above embodiments, since the peaks and troughs in the curves corresponding to the thermogravimetric analysis data can reflect the main thermal decomposition stages of tobacco products, including one or more extreme values ​​of the rate of mass change of tobacco products within the temperature range of the pyrolysis process, along with the corresponding characteristic temperatures, in the feature data allows the trained origin style prediction model to effectively learn the characteristics of tobacco product samples at each thermal decomposition stage. This helps to further improve the prediction accuracy and reliability of the origin style prediction model.

[0102] In some embodiments, a plurality of first pyrolysis temperatures form adjacent first temperature ranges and second temperature ranges.

[0103] The rate of mass change corresponding to the characteristic temperature includes the midpoint between the first extreme value of the rate of mass change of the tobacco product sample within the first temperature range and the second extreme value within the second temperature range. In other words, the rate of mass change corresponding to the characteristic temperature can include the midpoint between the two extreme values ​​of the rate of mass change of the tobacco product sample within two adjacent temperature ranges.

[0104] In some embodiments, the characteristic temperature includes a first temperature corresponding to an intermediate value in the first thermogravimetric analysis data, a temperature corresponding to a first extreme value that is less than the temperature corresponding to a second extreme value, and a first temperature that is greater than the temperature corresponding to the first extreme value and less than the temperature corresponding to the second extreme value. In other words, the characteristic temperature corresponding to an intermediate value in the first thermogravimetric analysis data may include a first temperature between the temperature corresponding to the first extreme value and the temperature corresponding to the second extreme value.

[0105] The first characteristic data determined from the first thermogravimetric analysis data includes the first temperature, and the second characteristic data includes intermediate values.

[0106] Therefore, by further including the median value of the mass change rate of tobacco product samples within two adjacent temperature ranges, as well as the first temperature corresponding to the median value, the trained origin style prediction model can be further trained to learn the characteristics of tobacco products at each stage of thermal decomposition. This will further improve the prediction accuracy and reliability of the origin style prediction model.

[0107] In some embodiments, the characteristic temperature includes a second temperature corresponding to the median value in the first thermogravimetric analysis data. The second temperature is either less than the temperature corresponding to the first extreme value or greater than the temperature corresponding to the second extreme value. In other words, the characteristic temperature corresponding to the median value can include a second temperature that is either less than the temperature corresponding to the first extreme value or greater than the temperature corresponding to the second extreme value.

[0108] The first characteristic data determined from the first thermogravimetric analysis data may also include the second temperature.

[0109] Therefore, considering that this intermediate value may correspond to multiple different temperatures in the thermogravimetric analysis data, further including this intermediate value and its corresponding second temperature (different from the first temperature) in the feature data allows the trained origin style prediction model to learn more about the characteristics of tobacco product raw materials at each stage of thermal decomposition. This further improves the prediction accuracy and reliability of the origin style prediction model.

[0110] The following is combined with Figure 2 Further explanation is provided for the case where the characteristic data includes the median value of the mass change rate of tobacco product samples within two adjacent temperature ranges, as well as the temperature corresponding to the median value.

[0111] like Figure 2 As shown, A1 and A2 are the starting and ending points on the DTG curve, respectively. B1 and B4 are the troughs on the DTG curve, and B2, B3, and B5 are the peaks on the DTG curve.

[0112] For example, B1 and B2 are extreme points within two adjacent temperature ranges. We can use the ordinate of characteristic point B1 as the first extreme value WB1 and the ordinate of characteristic point B2 as the second extreme value WB2. We can then select points C1 and D1 on the DTG curve corresponding to the midpoints of the ordinate values ​​of characteristic points B1 and B2 as characteristic points. The x-coordinate of characteristic point C1 lies between the x-coordinates of characteristic points B1 and B2; that is, the x-coordinate of characteristic point C1 corresponds to the first temperature TC1, which is the midpoint WC1. The x-coordinate of characteristic point D1 is less than the x-coordinate of characteristic point B1; that is, the x-coordinate of characteristic point D1 corresponds to the second temperature TD1, which is the midpoint WD1. And so on.

[0113] It is understood that the midpoint between the ordinate values ​​of feature point B1 and feature point B2, corresponding to point D1' on the DTG curve, can also be selected as the feature point. The x-coordinate value of feature point D1' is greater than the x-coordinate value of feature point B2, meaning the x-coordinate value of feature point D1' corresponds to the second temperature TD1' of the midpoint WD1'.

[0114] Figure 2 The diagram schematically illustrates eight selected feature points (also known as "half-peak points"), namely C1 to C4 and D1 to D4. Any combination of one or two types of data from the temperature and mass change rates of these eight feature points can be used as feature data input into the origin style prediction model to train the model.

[0115] For example, any set of data from (TC1~TC4), (WC1~WC4), and ((TC1, WC1)~(TC4, WC4)) along with the chemical composition data of tobacco products can be used as input to train the origin style prediction model.

[0116] For example, any set of data from (TD1~TD4), (WD1~WD4), and ((TD1, WD1)~(TD4, WD4)) along with the chemical composition data of tobacco products can be used as input to train the origin style prediction model.

[0117] For example, any set of data from (TC1~TC4 and TD1~TD4), (WC1~WC4 and WD1~WD4), ((TC1, WC1)~(TC4, WC4), (TD1, WD1)~(TD4, WD4)) and the chemical composition data of tobacco products can be used as inputs to train the origin style prediction model.

[0118] In some embodiments, the first characteristic data determined from the first thermogravimetric analysis data further includes the difference between the second temperature and the first temperature.

[0119] like Figure 2 As shown, Figure 2 The diagram schematically illustrates eight selected feature points (also known as "half-peak points"), C1 to C4 and D1 to D4. The difference between the first temperature TC1 (the x-coordinate of feature point C1) and the second temperature TD1 (the x-coordinate of feature point D1) is ∆T1 = TC1 - TD1; the difference between the first temperature TC2 (the x-coordinate of feature point C2) and the second temperature TD2 (the x-coordinate of feature point D2) is ∆T2 = TC2 - TD2; the difference between the first temperature TC3 (the x-coordinate of feature point C3) and the second temperature TD3 (the x-coordinate of feature point D3) is ∆T3 = TC3 - TD3; and the difference between the first temperature TC4 (the x-coordinate of feature point C4) and the second temperature TD4 (the x-coordinate of feature point D4) is ∆T4 = TC4 - TD4. Here, the first feature data determined from the first thermogravimetric analysis data may also include ∆T1, ∆T2, ∆T3, and ∆T4.

[0120] In some embodiments, the second characteristic data determined from the first thermogravimetric analysis data includes the difference between at least one extreme value and an intermediate value. In other words, the second characteristic data includes the intermediate value of two extreme values ​​corresponding to the rate of mass change of the tobacco product sample in two adjacent temperature intervals and the difference between either of these extreme values.

[0121] Continue to refer Figure 2 , Figure 2 The diagram schematically illustrates the five selected extreme points B1 to B5. Taking the ordinate of feature point B1 as the first extreme value WB1 and the ordinate of feature point B2 as the second extreme value WB2 as an example, the second feature data can include the difference ∆W1 = WC1 - WB1 between the first extreme value WB1 and the intermediate value WC1. Similarly, the second feature data can also include ∆W2 = WC2 - WB3, ∆W3 = WC3 - WB4, and ∆W4 = WC4 - WB5.

[0122] In the above embodiments, considering the different rates of thermal weight loss at different temperature ranges (i.e., the heterogeneity of thermal weight loss rates), adding the difference between the second temperature and the first temperature and / or the difference between at least one extreme value and an intermediate value as characteristic data of tobacco product samples can more accurately quantify the pyrolysis characteristics of tobacco product samples. Therefore, training based on such characteristic data can further improve the prediction accuracy and reliability of the origin style prediction model.

[0123] The following table, Table 1, schematically illustrates the feature data (also known as "feature parameters") obtained based on these feature points.

[0124] Table 1

[0125] Please continue to refer to this. Figure 2 The curve obtained after curve fitting based on the 15 feature points A1, A2, B1~B5, C1~C4, and D1~D4 is as follows: Figure 3 As shown by the dashed line, the curve fitted based on these 15 feature points is largely consistent with the actual DTG curve. This indicates that the 38 feature data obtained using these 15 feature points can accurately reflect the pyrolysis characteristics of tobacco products during the special decomposition process. Therefore, using this feature data to train the origin style prediction model can help the trained origin style prediction model accurately predict the origin of the tobacco raw material sample that accounts for the largest mass proportion in the tobacco product sample.

[0126] In some embodiments, the second chemical composition data of each of the multiple tobacco product raw material samples constituting each combined formulation may include the ratio of the content of different chemical components among the multiple chemical components of each tobacco product raw material sample. Correspondingly, the first chemical composition data of the tobacco product sample may also include the ratio of the content of different chemical components among the multiple chemical components in the tobacco product.

[0127] For example, the multiple chemical components may include total nitrogen, nicotine, total sugar, reducing sugar, potassium, and chlorine. The ratio of the contents of different chemical components in the multiple chemical components may include at least one of the following: potassium-chlorine ratio (i.e., the ratio of potassium to chlorine content), nitrogen-base ratio (i.e., the ratio of total nitrogen to nicotine content), sugar-base ratio (i.e., the ratio of total sugar to nicotine content), and disaccharide ratio (i.e., the ratio of reducing sugar to total sugar content).

[0128] Here, the calculation method for the first chemical component data of the tobacco product sample can be similar to that in the relevant embodiments above, and will not be repeated here.

[0129] Therefore, considering that the content and ratio of various chemical components in tobacco products can reflect the origin characteristics of tobacco product samples to a certain extent, incorporating this type of data into the origin style prediction model can help further improve the prediction accuracy and reliability of the origin style prediction model.

[0130] It needs to be recognized that, Figure 2 In the example, the number of pyrolysis temperatures and feature points (feature temperatures and their corresponding rates of mass change) is merely illustrative.

[0131] Figure 4 A flowchart illustrating a specific example of a training method for an origin style prediction model according to some embodiments of this disclosure.

[0132] like Figure 4 As shown, for example, the training method for the origin style prediction model may include steps 410 to 480. This method can be used as... Figure 1 This is a specific implementation of the method.

[0133] In step 410, tobacco product raw material samples are selected. For example, tobacco product raw material samples from 8 production areas are selected according to light aroma, medium aroma, and strong aroma, with each production area covering at least 30 different tobacco product raw material samples, and the selected tobacco product raw material samples cover upper tobacco leaves, middle tobacco leaves, and lower tobacco leaves.

[0134] In step 420, chemical composition data for each tobacco product raw material sample is obtained.

[0135] For example, a continuous flow method can be used to obtain chemical composition data for each tobacco product raw material sample. The chemical composition data for each tobacco product raw material sample may include total nitrogen, nicotine, total sugar, reducing sugar, potassium, chlorine, potassium-chlorine ratio, nitrogen-alkali ratio, sugar-alkali ratio, and disaccharide ratio.

[0136] In step 430, a pyrolysis experiment is performed on each tobacco product raw material sample to obtain thermogravimetric analysis data for each tobacco product raw material sample.

[0137] In step 440, multiple combination formulations of tobacco product samples are designed according to the principle of proportional gradient.

[0138] Here, the proportional gradient principle includes the following three constraints.

[0139] (1) Formulas covering combinations of 2 to 8 origins, where the quality of tobacco product raw material samples from each origin is the same. For example, taking a formula with 2 origins as an example, 9 ratios can be designed: 90%:10%, 80%:20%, 70%:30%, 60%:40%, 50%:50%, 40%:60%, 30%:70%, 20%:80%, and 10%:90%, thus forming multiple combination formulas. Formulas with 3 to 8 origins follow the same pattern.

[0140] (2) The mass proportion of tobacco product raw materials from the target origin in multiple combination formulations can be distributed in an equally spaced gradient. For example, the mass proportion of tobacco product raw material samples from the target origin in the tobacco product sample can vary from 10% to 90% in 10% intervals. The mass proportion of tobacco product raw material samples from the target origin in the tobacco product sample in multiple combination formulations can be 10%, 20%, 30%, 40%...90%, respectively.

[0141] (3) The tobacco product raw material with the largest weight in each combination formula comes from only one place (i.e., the tobacco product raw material from a single place has the largest weight).

[0142] In step 450, tobacco product raw material samples from the target origin and tobacco product raw material samples from other origins are selected according to each of the determined multiple combination formulas to form the tobacco product sample corresponding to each combination formula.

[0143] In step 460, the first thermogravimetric analysis data and the first chemical composition data of the tobacco product sample corresponding to each combination formulation are obtained.

[0144] For example, the first thermogravimetric analysis data can be calculated based on the thermogravimetric analysis data of each tobacco product raw material sample obtained in step 430. The first chemical composition data can be calculated based on the chemical composition data of each tobacco product raw material sample obtained in step 420.

[0145] In step 470, first feature data related to the sample characteristic temperature among multiple first pyrolysis temperatures and second feature data related to the mass change rate corresponding to the sample characteristic temperature are determined from the first thermogravimetric analysis data to obtain a training dataset.

[0146] For example, the training dataset includes at least one of the first feature data and second feature data of multiple combination formulations, as well as the first chemical component data of multiple combination formulations. For example, the training dataset can be constructed in the form of sample pairs of "combination formulation ID-feature data-chemical component data-origin".

[0147] In step 480, the convolutional neural network model is trained based on the training dataset to obtain the origin style prediction model.

[0148] For example, the origin style prediction model is trained by taking at least one of the first and second characteristic data of multiple combination formulas and the first chemical composition data of multiple combination formulas as inputs, and taking the origin of the tobacco product raw material sample with the largest mass proportion in the tobacco product sample as output, until the training termination condition is met.

[0149] about Figure 4 For more embodiments of the training method for the origin style prediction model shown, please refer to the previous section. Figure 1 The training method and its implementation examples of the origin style prediction model are described below.

[0150] The previous section introduced the training method and implementation of the origin style prediction model. The following section further describes the prediction method for the origin style of tobacco products using the trained origin style prediction model.

[0151] Figure 5 A flowchart illustrating a method for predicting the origin style of a tobacco product under test, according to some embodiments of this disclosure, is shown. Figure 5 As shown, for example, the origin style prediction method may include steps 510 to 530.

[0152] In step 510, the third thermogravimetric analysis data and the third chemical composition data of the tobacco product to be tested are obtained.

[0153] Here, the tobacco product under test is composed of multiple tobacco product raw materials from different origins. The third thermogravimetric analysis data includes multiple third pyrolysis temperatures of the pyrolysis process of the tobacco product under test and the mass change rate corresponding to each of the multiple third pyrolysis temperatures. The third chemical composition data includes the content of each chemical component in the multiple chemical components in the tobacco product under test.

[0154] It is understood that the third thermogravimetric analysis data of the tobacco product to be tested can be obtained by conducting thermogravimetric analysis experiments on the tobacco product to be tested, or it can be calculated based on the thermogravimetric analysis data of multiple tobacco product raw materials that make up the tobacco product to be tested. The specific implementation is similar to the implementation method of obtaining the first thermogravimetric analysis data of the tobacco product sample mentioned above. For relevant explanations, please refer to the description in the relevant embodiments mentioned above, and will not be repeated here.

[0155] The third chemical component data of the tobacco product to be tested can be obtained by conducting chemical component analysis experiments on the tobacco product to be tested for multiple chemical components, or it can be calculated based on the chemical component data of multiple tobacco product raw materials that make up the tobacco product to be tested. The specific implementation is similar to the method of obtaining the first chemical component data of the tobacco product mentioned above. For relevant explanations, please refer to the description in the relevant embodiments mentioned above, and will not be repeated here.

[0156] In step 520, third characteristic data related to the characteristic temperature among the multiple third pyrolysis temperatures and fourth characteristic data related to the mass change rate corresponding to the characteristic temperature are determined from the third thermogravimetric analysis data.

[0157] The method for selecting feature data of the tobacco products to be tested in the process of predicting the origin style of the tobacco products using the prediction model is similar to the method for selecting feature data of tobacco product samples in the process of training the origin style prediction model. For details, please refer to the relevant embodiments above. It will not be repeated here.

[0158] In step 530, based on at least one of the third feature data and the fourth feature data, as well as the third chemical composition data, the origin style of the tobacco product to be tested is predicted using an origin style prediction model.

[0159] Here, the origin style prediction model is trained based on the training method of the origin style prediction model described above.

[0160] In some embodiments, at least one of the third feature data and the fourth feature data, along with the third chemical composition data, can be used as input to the origin style prediction model to obtain the origin of the tobacco product raw material sample with the largest mass proportion in the tobacco product sample output by the origin style prediction model as the origin style of the tobacco product to be tested.

[0161] For example, the third feature data, the fourth feature data, and the third chemical composition data can be used together as input to the origin style prediction model to obtain the origin of the tobacco product raw material sample with the largest mass proportion in the tobacco product sample (e.g., Yunnan) as the origin style (e.g., Yunnan style) of the tobacco product to be tested.

[0162] In the above embodiments, since the origin style prediction model trained based on the training method of the origin style prediction model of this disclosure can effectively quantify the influence of the quality ratio of tobacco product raw materials from different origins in tobacco products on the origin style of tobacco products, the origin style prediction model can be accurately predicted by using such an origin style prediction model to predict the origin style of the tobacco product to be tested.

[0163] Next, with reference to some embodiments, we will exemplarily illustrate how the technical solution for predicting the origin style of tobacco products to be tested proposed in this disclosure is used to construct a combination formula-origin style database, so that when combination formula design is required, the target combination formula can be quickly determined with the help of the constructed database.

[0164] Figure 6 A flowchart illustrating a method for determining a target combination formulation according to some embodiments of this disclosure is provided. Figure 6 As shown, for example, a method for determining a target combination formulation may include steps 610 to 630.

[0165] In step 610, based on each of the multiple candidate combination formulations, multiple tobacco product raw materials from different origins are selected from the tobacco product raw material library to form the tobacco product corresponding to each candidate combination formulation.

[0166] In step 620, for each candidate combination formulation, steps 621 to 622 are performed to build a combination formulation-origin style database.

[0167] In step 621, the origin style of the tobacco product corresponding to each candidate combination formulation is predicted using a technical solution for predicting the origin style of the tobacco product to be tested. Here, the technical solution for predicting the origin style of the tobacco product to be tested can be as described above. Figure 5 The process shown can be implemented by referring to the previous text for relevant steps. Figure 5 The descriptions in the relevant embodiments shown will not be repeated here.

[0168] In step 622, each candidate combination formula and the origin style of the tobacco product corresponding to each candidate combination formula are stored in the combination formula-origin style database.

[0169] For example, a candidate blend may contain 90% tobacco leaves from Yunnan and 10% from Fujian, and the predicted regional style for this candidate blend is Yunnan. This candidate blend and its corresponding predicted regional style are stored as a single data entry in the blend-regional style database.

[0170] In step 630, the combination formula corresponding to the target origin style is retrieved from the combination formula-origin style database to determine the target combination formula.

[0171] In some embodiments, a target combination formula is determined from the retrieved combination formulas that meet a first criterion. For example, there may be multiple combination formulas retrieved according to the target origin style. The first criterion is to determine the target combination formula by determining whether one or more of the tobacco product raw material inventory or year in the multiple candidate combination formulas meet certain conditions.

[0172] In the above embodiments, for each of the multiple candidate combination formulations based on multiple tobacco product raw materials, the regional style can be predicted using the technical solution for predicting the regional style of tobacco products as described above, and a combination formulation-regional style database can be constructed. When it is necessary to design a combination formulation corresponding to a target regional style, the target combination formulation can be determined by searching the combination formulation-regional style database.

[0173] In this approach, the technology of predicting the regional style of tobacco products can accurately determine the regional style corresponding to each blend formulation. The database built based on this data can provide strong data support for blend formulation design. As a result, the determined target blend formulation can accurately meet the expected standards in practical applications, and the efficiency and flexibility of blend formulation design are also significantly improved.

[0174] The following examples illustrate the predictive performance of the origin style prediction model trained using the training method of this disclosure, and the predictive effect of using the origin style prediction model for origin style prediction.

[0175] First, according to the previous text Figure 4 The specific example shown is used to train the origin style prediction model.

[0176] Then, the origin style prediction model was used to predict the origin style of multiple tobacco product raw material samples (e.g., 434 tobacco leaf samples), and the prediction results are shown in Table 2.

[0177] Table 2

[0178] As shown in Table 2, the average prediction accuracy of the regional style prediction model for 434 tobacco leaf samples was 90.09%, indicating that the regional style prediction model can accurately identify the style characteristics of each region and fully verify the reliability of the regional style prediction model in feature extraction.

[0179] Furthermore, the origin style prediction model was used to predict the origin style of tobacco products corresponding to multiple combination formulas. The prediction results are shown in Table 3.

[0180] Table 3

[0181] As shown in Table 3, the regional style of tobacco products predicted by this regional style prediction model is generally highly consistent with the regional style of the main source tobacco leaves, with an average consistency rate of 87.90%. This result demonstrates the dominant role of the main source tobacco leaves in influencing the regional style of the final product, and the scientific validity and rationality of the training logic of this regional style prediction model. Furthermore, the following key findings are also revealed.

[0182] (1) Threshold effect: When the mass proportion of the main source tobacco leaves is greater than or equal to 40%, the consistency of the origin style of the tobacco product with the origin style of its main source tobacco leaves (hereinafter referred to as "style consistency") is significantly improved, with an average value greater than 87.50%. This indicates that when the main source tobacco leaves account for a large proportion in the formula, its origin style has a more obvious dominant role in the final product.

[0183] (2) Origin specificity: The tobacco leaves from origins B and H have a strong style retention. Even with a quality ratio of 40%, the style consistency remains above 93%, which is significantly better than tobacco leaves from other origins. This indicates that tobacco leaves from certain specific origins have stronger style stability and uniqueness, and can maintain their dominant role in the origin style of the final product even when blended with tobacco leaves from other origins.

[0184] (3) Nonlinear decay: As the mass proportion of the main source tobacco leaves decreased from 90% to 30%, the style consistency nonlinearly decreased from 99.91% to 67.90%. When the mass proportion of the main source tobacco leaves was high, the regional style of the main source tobacco leaves had a significant impact on the regional style of the final product. However, when the mass proportion of the main source tobacco leaves was low, the style characteristics of tobacco leaves from other regions began to emerge, leading to a decrease in style consistency. This indicates that the impact of the mass proportion of the main source tobacco leaves on style consistency is not a linear relationship, and the interaction of tobacco leaves from different regions after blending also has a certain impact on the regional style of the final product.

[0185] These key findings confirm that the regional characteristics of tobacco raw materials, which constitute a significant portion of the blend, play a dominant role in the regional characteristics of the final product. Furthermore, the interactions between tobacco raw materials from different regions after blending in a single formula also significantly impact the regional characteristics of the final product. Therefore, using the thermogravimetric analysis and chemical composition data of tobacco products corresponding to each blend as the data foundation for training machine learning models is both scientifically sound and reasonable.

[0186] The analysis of several combination formulations where the origin style of tobacco products predicted by the origin style prediction model is inconsistent with the origin style of the main source tobacco leaves designed in the combination formulation is shown in Table 4.

[0187] Table 4

[0188] Table 4 shows that among the 36,867 inconsistent style blends, 32,246 blends were due to a difference of less than or equal to 10% between the mass proportions of primary and secondary source tobacco leaves. This factor accounted for 87.5% of the total number of misclassified blends. This indicates that when the mass proportions of primary and secondary source tobacco leaves are relatively close, the regional style of the secondary source tobacco leaves may interfere with the regional style of the primary source tobacco leaves, leading to inconsistencies between the regional style predicted by the regional style prediction model and the designed regional style of the primary source tobacco leaves.

[0189] Although the origin style model can accurately predict the origin style of tobacco products in most cases, there is still a certain misjudgment rate when the quality proportions of primary source tobacco leaves and secondary source tobacco leaves are close.

[0190] Therefore, the design of training data can be optimized to improve the feature extraction capability of the origin style prediction model, enabling it to more accurately distinguish the origin styles of primary and secondary tobacco leaves. For example, for each blend, the difference between the mass percentage of primary and secondary tobacco leaves can be designed to be greater than 10%.

[0191] Figure 7 A block diagram of a training apparatus for a tobacco product origin style prediction model according to some embodiments of the present disclosure is shown.

[0192] like Figure 7 As shown, the training device 700 for the prediction model of the origin style of tobacco products includes a first determination module 701, a selection module 702, an acquisition module 703, a second determination module 704, and a training module 705.

[0193] The first determining module 701 can be configured to determine multiple combined formulations of a tobacco product. Each combined formulation indicates the mass percentage of tobacco product raw materials from a target origin and tobacco product raw materials from other origins in the tobacco product, and the mass percentage of tobacco product raw materials from the target origin varies in the multiple combined formulations.

[0194] The selection module 702 can be configured to select tobacco product raw material samples from the target origin and tobacco product raw material samples from other origins for each of the determined multiple combination formulations, so as to form the tobacco product sample corresponding to each combination formulation.

[0195] The acquisition module 703 can be configured to acquire first thermogravimetric analysis data and first chemical composition data of a tobacco product sample composed of tobacco product raw material samples corresponding to each combined formulation. The first thermogravimetric analysis data includes multiple first pyrolysis temperatures of the pyrolysis process of the tobacco product sample and the mass change rate corresponding to each of the multiple first pyrolysis temperatures. The first chemical composition data includes the content of each chemical component among multiple chemical components in the tobacco product sample.

[0196] The second determining module 704 can be configured to determine, from the first thermogravimetric analysis data, first characteristic data related to the sample characteristic temperature among a plurality of first pyrolysis temperatures, and second characteristic data related to the mass change rate corresponding to the sample characteristic temperature.

[0197] The training module 705 can be configured to take at least one of the first feature data and the second feature data, as well as the first chemical composition data, as input, and the predicted origin of the tobacco product raw material sample with the largest mass proportion in the tobacco product sample as output, to train the origin style prediction model until the training termination condition is met.

[0198] In some embodiments, the training device 700 for the tobacco product origin style prediction model may further include the execution of the foregoing text. Figures 1 to 4 Other modules of other operations in the illustrated embodiments.

[0199] Figure 8 A block diagram of an origin style prediction device for a tobacco product to be tested, according to some embodiments of the present disclosure, is shown.

[0200] like Figure 8 As shown, the origin style prediction device 800 for the tobacco product to be tested includes an acquisition module 801, a determination module 802, and a prediction module 803.

[0201] The acquisition module 801 can be configured to acquire third thermogravimetric analysis (TGA) data and third chemical composition data of the tobacco product under test. The tobacco product under test consists of multiple tobacco raw materials from different origins. The third TGA data includes multiple third pyrolysis temperatures of the tobacco product under test and the corresponding mass change rate for each of these temperatures. The third chemical composition data includes the content of each of the various chemical components in the tobacco product under test.

[0202] The determination module 802 can be configured to determine, from the third thermogravimetric analysis data, third characteristic data related to a characteristic temperature among a plurality of third pyrolysis temperatures, and fourth characteristic data related to the mass change rate corresponding to the characteristic temperature.

[0203] The prediction module 803 can be configured to predict the origin style of the tobacco product under test based on at least one of the third feature data and the fourth feature data, as well as the second chemical composition data, using an origin style prediction model trained by a training device (e.g., training device 700) of the tobacco product origin style prediction model of any of the above embodiments.

[0204] In some embodiments, the origin style prediction device 800 for the tobacco product to be tested may further include the execution of the foregoing description. Figure 5 Other modules of other operations in the illustrated embodiments.

[0205] Figure 9 A block diagram of an electronic device according to some embodiments of the present disclosure is shown.

[0206] like Figure 9As shown, the electronic device 900 of this embodiment includes: a memory 901 and a processor 902 coupled to the memory 901. The processor 902 is configured to execute, based on instructions stored in the memory 901, a training method for a tobacco product origin style prediction model, a method for predicting the origin style of a tobacco product to be tested, or a method for determining a target combination formula for a tobacco product.

[0207] The memory 901 may include, for example, system memory, fixed non-volatile storage media, etc. The system memory may store, for example, an operating system, application programs, a boot loader, a database, and other programs.

[0208] Figure 10 Block diagrams of electronic devices according to other embodiments of the present disclosure are shown.

[0209] like Figure 10 As shown, the electronic device 1000 of this embodiment includes: a memory 1001 and a processor 1002 coupled to the memory 1001. The processor 1002 is configured to execute, based on instructions stored in the memory 1001, a training method for a tobacco product origin style prediction model, a method for predicting the origin style of a tobacco product to be tested, or a method for determining a target combination formula of a tobacco product.

[0210] The memory 1001 may include, for example, system memory, fixed non-volatile storage media, etc. The system memory may store, for example, the operating system, application programs, boot loader, and other programs.

[0211] The electronic device 1000 may also include an input / output interface 1003, a network interface 1004, and a storage interface 1005. These interfaces 1003, 1004, and 1005, as well as the memory 1001 and processor 1002, can be connected, for example, via a bus 1006. The input / output interface 1003 provides a connection interface for input / output devices such as a monitor, mouse, keyboard, touchscreen, microphone, and speakers. The network interface 1004 provides a connection interface for various networked devices. The storage interface 1005 provides a connection interface for external storage devices such as SD cards and USB flash drives.

[0212] This disclosure also provides a computer-readable storage medium including computer program instructions, which, when executed by a processor, constitute a method for training a tobacco product origin style prediction model, a method for predicting the origin style of a tobacco product to be tested, or a method for determining a target combination formulation of a tobacco product, as described in any of the above embodiments.

[0213] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the training method for a tobacco product origin style prediction model, the method for predicting the origin style of a tobacco product to be tested, or the method for determining a target combination formulation of a tobacco product, as described in any of the above embodiments.

[0214] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0215] This concludes the detailed description of the training method for the origin style prediction model of tobacco products according to this disclosure, the method for predicting the origin style of the tobacco product to be tested, and the method for determining the target combination formulation of the tobacco product. To avoid obscuring the concept of this disclosure, some details known in the art have not been described. Those skilled in the art will fully understand how to implement the technical solutions disclosed herein based on the above description.

[0216] The methods and systems of this disclosure may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the methods is for illustrative purposes only, and the steps of the methods of this disclosure are not limited to the specific order described above unless otherwise specifically stated. Furthermore, in some embodiments, this disclosure may also be implemented as a program recorded on a recording medium, the program including machine-readable instructions for implementing the methods according to this disclosure. Thus, this disclosure also covers recording media storing programs for performing the methods according to this disclosure.

[0217] While specific embodiments of this disclosure have been described in detail by way of example, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of this disclosure. Those skilled in the art should understand that modifications can be made to the above embodiments without departing from the scope and spirit of this disclosure. The scope of this disclosure is defined by the appended claims.

Claims

1. A training method for a tobacco product origin style prediction model, comprising: Multiple combination formulations for tobacco products are determined, each combination formulation indicating the mass percentage of tobacco product raw materials from a target origin and tobacco product raw materials from other origins in the tobacco product, and the mass percentage of tobacco product raw materials from the target origin is different in each of the multiple combination formulations; Tobacco product raw material samples from the target origin and tobacco product raw material samples from other origins are selected for each of the determined multiple combination formulas to form the tobacco product sample corresponding to each combination formula; Obtain first thermogravimetric analysis data and first chemical composition data of the tobacco product sample corresponding to each combination formulation. The first thermogravimetric analysis data includes multiple first pyrolysis temperatures of the pyrolysis process of the tobacco product sample and the mass change rate corresponding to each of the multiple first pyrolysis temperatures. The first chemical composition data includes the content of each chemical component among the multiple chemical components in the tobacco product sample. From the first thermogravimetric analysis data, determine first characteristic data related to the sample characteristic temperature among the plurality of first pyrolysis temperatures, and second characteristic data related to the mass change rate corresponding to the sample characteristic temperature; The origin style prediction model is trained by taking at least one of the first feature data and the second feature data of the multiple combined formulations and the first chemical composition data of the multiple combined formulations as inputs and taking the origin of the tobacco product raw material sample with the largest mass proportion in the tobacco product sample as output, until the training termination condition is met.

2. The training method according to claim 1, wherein, The step of selecting tobacco product raw material samples from the target origin and tobacco product raw material samples from other origins according to each of the determined multiple combination formulations includes: According to the mass percentage of tobacco product raw materials from other origins indicated in each of the combined formulations, tobacco product raw materials from one or more origins different from the target origin are randomly selected as samples of tobacco product raw materials from other origins.

3. The training method according to claim 1, wherein, The mass percentages of tobacco product raw materials from the target origin in the multiple combined formulations are distributed in an equally spaced gradient.

4. The training method according to claim 1, wherein, The tobacco product raw material with the largest weight percentage in each of the aforementioned formulations originates from only one place.

5. The training method according to any one of claims 1-4, wherein, The plurality of first pyrolysis temperatures form temperature ranges, and the mass change rate corresponding to the characteristic temperature includes at least one extreme value of the mass change rate of the tobacco product sample within the temperature range. The first feature data includes the pyrolysis temperature corresponding to the at least one extreme value, and the second feature data includes the at least one extreme value.

6. The training method according to any one of claims 1-4, wherein, The plurality of first pyrolysis temperatures form adjacent first temperature ranges and second temperature ranges. The mass change rate corresponding to the characteristic temperature includes the midpoint between the first extreme value of the mass change rate of the tobacco product sample within the first temperature range and the second extreme value of the mass change rate of the tobacco product sample within the second temperature range. The characteristic temperature includes the first temperature corresponding to the intermediate value in the first thermogravimetric analysis data, the temperature corresponding to the first extreme value is less than the temperature corresponding to the second extreme value, and the first temperature is greater than the temperature corresponding to the first extreme value and less than the temperature corresponding to the second extreme value. The first feature data includes the first temperature, and the second feature data includes the intermediate value.

7. The training method according to claim 6, wherein, The characteristic temperature includes the second temperature corresponding to the intermediate value in the first thermogravimetric analysis data, wherein the second temperature is either less than or greater than the temperature corresponding to the first extreme value. The first feature data also includes the second temperature.

8. The training method according to claim 7, wherein, The first feature data also includes the difference between the second temperature and the first temperature.

9. The training method according to claim 6, wherein, The mass change rate corresponding to the characteristic temperature includes at least one extreme value of the mass change rate of the tobacco product sample within the first temperature range and the second temperature range. The second feature data includes the difference between the at least one extreme value and the median value.

10. The training method according to any one of claims 1-4, wherein, The characteristic temperature includes at least one of the starting temperature and the ending temperature among the plurality of first pyrolysis temperatures. The first feature data includes at least one of the starting temperature and the ending temperature. The second feature data includes at least one of the mass change rate corresponding to the starting temperature and the mass change rate corresponding to the ending temperature.

11. The training method according to any one of claims 1-4, wherein, The process of obtaining the first thermogravimetric analysis data and the first chemical composition data of the tobacco product samples corresponding to each of the combined formulations includes: Obtain second thermogravimetric analysis data and second chemical composition data for each of the multiple tobacco product raw material samples that make up the tobacco product sample corresponding to each combined formulation. The second thermogravimetric analysis data includes multiple second pyrolysis temperatures of the pyrolysis process of each tobacco product raw material sample and the mass change rate of each tobacco product raw material sample corresponding to each of the multiple second pyrolysis temperatures. The second chemical composition data includes the content of each chemical component among the multiple chemical components of each tobacco product raw material sample. The first thermogravimetric analysis data is determined based on the average value of the second thermogravimetric analysis data; The first chemical component data is determined based on the average value of the second chemical component data.

12. The training method according to claim 11, wherein, The second chemical composition data includes the ratio of the content of different chemical components in the multiple chemical components of each tobacco product raw material sample. The first chemical composition data also includes a comprehensive ratio determined based on the ratio of the contents of the different chemical components.

13. The method for predicting regional style according to any one of claims 1-4, wherein, The process of obtaining the first thermogravimetric analysis data and the first chemical composition data of the tobacco product samples corresponding to each of the combined formulations includes: Thermogravimetric analysis was performed on the tobacco product sample to obtain the first thermogravimetric analysis data; and Chemical composition analysis experiments were performed on the tobacco product sample to obtain the first chemical composition data.

14. The method for predicting regional style according to any one of claims 1-4, wherein, The various chemical components include at least one of total nitrogen, nicotine, total sugar, reducing sugar, potassium, and chlorine.

15. A method for predicting the regional style of a tobacco product to be tested, comprising: The third thermogravimetric analysis data and the third chemical composition data of the tobacco product to be tested are obtained. The tobacco product to be tested is composed of multiple tobacco product raw materials from different origins. The third thermogravimetric analysis data includes multiple third pyrolysis temperatures of the pyrolysis process of the tobacco product to be tested and the mass change rate corresponding to each of the multiple third pyrolysis temperatures. The third chemical composition data includes the content of each chemical component among the multiple chemical components in the tobacco product to be tested. From the third thermogravimetric analysis data, determine third characteristic data related to the characteristic temperature among the plurality of third pyrolysis temperatures, and fourth characteristic data related to the mass change rate corresponding to the characteristic temperature; Based on at least one of the third feature data and the fourth feature data, as well as the third chemical composition data, the origin style prediction model trained using the training method according to claim 1 is used to predict the origin style of the tobacco product to be tested.

16. A method for determining a target combination formulation of tobacco products, comprising: Based on each of the multiple candidate combination formulations, multiple tobacco product raw materials from different production areas are selected from the tobacco product raw material library to form the tobacco product corresponding to each candidate combination formulation; For each candidate combination recipe, the following steps are performed to build a combination recipe-origin style database: The origin style prediction method according to claim 15 is used to predict the origin style of the tobacco products corresponding to each candidate combination formulation; as well as Each candidate combination formula and the origin style of the tobacco product corresponding to each candidate combination formula are stored in the combination formula-origin style database; The target combination formula is determined by retrieving the combination formula corresponding to the target origin style from the combination formula-origin style database.

17. The method of claim 16, further comprising: The combined formulations that meet the first criterion among the retrieved combined formulations are identified as the target combined formulations.

18. A training device for a tobacco product origin style prediction model, comprising: The first determining module is configured to determine multiple combined formulations of a tobacco product, each combined formulation indicating the mass percentage of tobacco product raw materials from a target origin and tobacco product raw materials from other origins in the tobacco product, wherein the mass percentage of tobacco product raw materials from the target origin is different in each of the multiple combined formulations. The selection module is configured to select tobacco product raw material samples from the target origin and tobacco product raw material samples from other origins for each of the determined multiple combination formulas, so as to form tobacco product samples corresponding to each combination formula; The acquisition module is configured to acquire first thermogravimetric analysis data and first chemical composition data of the tobacco product sample composed of tobacco product raw material samples corresponding to each of the combined formulations. The first thermogravimetric analysis data includes multiple first pyrolysis temperatures of the pyrolysis process of the tobacco product sample and the mass change rate corresponding to each of the multiple first pyrolysis temperatures. The first chemical composition data includes the content of each chemical component among the multiple chemical components in the tobacco product sample. The second determining module is configured to determine, from the first thermogravimetric analysis data, a first characteristic data related to the sample characteristic temperature among the plurality of first pyrolysis temperatures, and a second characteristic data related to the mass change rate corresponding to the sample characteristic temperature; The training module is configured to take at least one of the first feature data and the second feature data, as well as the first chemical composition data, as input, and the predicted origin of the tobacco product raw material sample with the largest mass proportion in the tobacco product sample as output, to train the origin style prediction model until the training termination condition is met.

19. A device for predicting the origin and style of a tobacco product to be tested, comprising: The acquisition module is configured to acquire the third thermogravimetric analysis data and the third chemical composition data of the tobacco product to be tested. The tobacco product to be tested is composed of multiple tobacco product raw materials from different origins. The third thermogravimetric analysis data includes multiple third pyrolysis temperatures of the pyrolysis process of the tobacco product to be tested and the mass change rate corresponding to each of the multiple third pyrolysis temperatures. The third chemical composition data includes the content of each chemical component among the multiple chemical components in the tobacco product to be tested. The determination module is configured to determine, from the third thermogravimetric analysis data, a third feature data related to a feature temperature among the plurality of third pyrolysis temperatures, and a fourth feature data related to the mass change rate corresponding to the feature temperature; The prediction module is configured to predict the origin style of the tobacco product under test based on at least one of the third feature data and the fourth feature data, as well as the second chemical composition data, using an origin style prediction model trained by the training device according to claim 18.

20. An electronic device, comprising: Memory; and A processor coupled to the memory is configured to execute, based on instructions stored in the memory, a training method for a tobacco product origin style prediction model according to any one of claims 1-14, a method for predicting the origin style of a tobacco product to be tested according to any one of claims 15, or a method for determining a target combination formulation of a tobacco product according to any one of claims 16-17.

21. A computer-readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the training method for a tobacco product origin style prediction model according to any one of claims 1-14, the method for predicting the origin style of a tobacco product to be tested according to any one of claims 15, or the method for determining a target combination formulation of a tobacco product according to any one of claims 16-17.

22. A computer program product comprising instructions that, when executed by a processor, cause the processor to perform a training method for a tobacco product origin style prediction model according to any one of claims 1-14, a method for predicting the origin style of a tobacco product to be tested according to any one of claims 15, or a method for determining a target combination formulation of a tobacco product according to any one of claims 16-17.

Images