A digital tobacco formulation design method based on detection, proportioning, correction, and adjustment.

Abstract

This invention discloses a digital tobacco formulation design method based on detection, proportioning, correction, and adjustment, comprising the following steps: S1. Sampling and near-infrared spectroscopy detection of various types of purchased tobacco leaves prepared for processing, according to the minimum particle size of the tobacco leaf formulation; S2. Obtaining digital indicators for each grade of tobacco leaves based on near-infrared spectroscopy detection and corresponding indicator prediction models; S3. Generating an empirical formulation sheet by a formulator based on the indicator results and the quantity of tobacco leaves; S4. Generating various digital indicators for the formulation module based on the proportions of the empirical formulation sheet and the indicators of each unit; S5. Correcting the digital indicators of the generated formulation module based on the differences between historical raw tobacco data and actual data, obtaining a correction model; S6. The formulator judges the digital indicators of the formulation module and adjusts the formulation sheet in step S3 based on historical indicator values ​​and the current year's tobacco leaf production situation; S7. Repeating steps S3-S6 to obtain the adjusted formulation sheet.

Description

Technical Field

[0001] This invention belongs to the field of tobacco formulation design, specifically relating to a digital tobacco formulation design method based on detection, proportioning, correction, and adjustment. Background Technology

[0002] Currently, the tobacco industry processes various single-grade tobacco leaves into sheet tobacco modules through leaf threshing and re-drying. After appropriate aging, these sheet tobaccos are then used in cigarette factories for cigarette production. Formula design is required during the processing of various single-grade tobacco leaves into sheet tobacco modules.

[0003] Traditional formulation design relies on the experience of formulators combined with sensory evaluation. This involves sampling all raw tobacco leaves, designing a formulation based on experience, and then sampling and adjusting the formulation samples. Since tobacco leaves are agricultural products with similar maturity times, formulators need to sample and design a large number of samples within a short period, resulting in a high workload. Currently, there is no method other than sensory evaluation to assess the differences between the designed formulation and the actual production samples.

[0004] Currently, publicly available digital tobacco formulation design methods include CN114246356B (Design method, system, medium, and device for cigarette leaf blend formulation), CN115879254A (Design method, device, and storage medium for cigarette formulation based on multiple constraints), and CN116266475A (An auxiliary design method for cigarette leaf blend formulation based on virtual mixed near-infrared spectroscopy). The main idea behind these methods is to use operations research and other methods to plan and solve problems based on the target value of the given formulation and the data values ​​of various raw materials. However, existing methods have not yet solved the following problems: ① The main chemical components of tobacco raw materials change after aging, resulting in a mismatch between the raw material data at the time of storage and the state at the time of use; ② The tobacco processing process has a significant impact on the tobacco raw materials. These problems ultimately lead to a large discrepancy between the digitally designed values ​​and the actual product values, making it difficult to meet quality requirements. Summary of the Invention

[0005] To address the technical problem of significant discrepancies between existing digital design values ​​and actual product values, which makes it difficult to meet quality requirements, this invention provides a digital tobacco formulation design method based on detection, proportioning, correction, and adjustment. Based on years of testing data and actual formulation results, it analyzes various factors from the design end to the production end, forming multiple correction models to solve the problem of mismatch between design values ​​and actual product values.

[0006] The technical solution adopted in this invention is:

[0007] A digital tobacco formulation design method based on detection, proportioning, correction, and adjustment, characterized by comprising the following steps:

[0008] S1. Testing: For all types of tobacco leaves that have been purchased and are ready for processing, samples are taken according to the minimum particle size of the tobacco leaf formula and near-infrared spectroscopy is performed.

[0009] S2. Based on near-infrared spectroscopy detection and corresponding index prediction models, digital indicators for each grade of tobacco leaves are obtained.

[0010] S3. Blending: Based on the indicator results and the quantity of tobacco leaves, the blender generates an empirical blending sheet.

[0011] S4. Based on the proportions of the experience formula sheet and the indicators of each unit, generate various digital indicators for the formula module.

[0012] S5. Correction: Based on the differences between historical raw tobacco data and actual data, the digital indicators of the generated formula module are corrected to obtain the corrected model;

[0013] S6, Adjustment:

[0014] The formulator judges the digital indicators of the formula module, and adjusts the formula sheet in step S3 by combining the indicator values ​​of previous years with the tobacco production situation of the current year.

[0015] S7. Repeat steps S3-S6 to obtain the adjusted formula sheet for subsequent production.

[0016] Furthermore, in step S1, the sampling factors include, but are not limited to, tobacco production area, grade, variety, and batch.

[0017] Furthermore, in step S2, the prediction indicators include conventional chemical components such as total sugar, nicotine, reducing sugar, chlorine, potassium, and total nitrogen, and may also include qualitative and quantitative quality indicators such as part, aroma, and sensory characteristics.

[0018] Furthermore, in step S3, the formula for calculating the numerical value of the empirical formula is as follows:

[0019] For each indicator, the formula module value y is obtained by the following formula (1):

[0020]

[0021] Where n is the number of raw material categories in the formula sheet, y h Let b be the index value of raw material of type h. h This refers to the proportion of this type of raw material in the total ingredients in the formula.

[0022] Furthermore, in step S5, the correction model includes, but is not limited to: direct correction, correction by part, correction by processing batch, and correction by both part and processing batch.

[0023] Furthermore, the method for directly correcting any selected indicator is as follows:

[0024] The error y in estimating the predicted tobacco sheet from raw tobacco data in historical data. e =yy py Where y is the formula value estimated in step S3 based on the raw tobacco and formula ratio, y py These are the actual values ​​measured in tobacco sheets;

[0025] The correction factor is:

[0026] a = mean(y) e (2)

[0027] For formulations requiring design, after obtaining the raw tobacco test data and formulation ratios, the design value is:

[0028]

[0029] Among them, y new The formula value is estimated based on the existing raw tobacco and formula ratio calculated in step S3, and 'a' is a correction coefficient.

[0030] Furthermore, the method for correcting any selected indicator by location is as follows:

[0031] The error y in predicting tobacco leaves from raw tobacco data in historical data. e The calculation is performed separately for the upper, middle, and lower parts of the tobacco leaf, with three correction coefficients calculated for each part. 1, 2, and 3 represent the upper, middle, and lower parts, respectively. The correction coefficients for the formulations of these three parts are:

[0032]

[0033] The predicted formula is substituted into the corresponding correction coefficient according to its part design value; that is, if the processing formula is an upper part formula, then...

[0034]

[0035] That is, if the processing formula is a central formula, then

[0036]

[0037] That is, if the processing formula is the lower part of the formula, then

[0038]

[0039] in, Represents the design value of the formula, y new The formula values ​​are estimated based on the existing raw tobacco and formula ratios calculated in step S3.

[0040] Furthermore, the method for correcting any selected indicator by processing batch is as follows:

[0041] The error y in predicting tobacco leaves from raw tobacco data in historical data. e The cigarettes were separated into three batches: the front, middle, and rear batches, and three correction factors were calculated for each batch. 1, 2, and 3 represent the pre-processing, mid-processing, and post-processing batches, respectively. The correction factors for the three batches are:

[0042]

[0043] The prediction model incorporates its batch design values ​​into the corresponding correction coefficients; that is, if the processing formula is the initial formula, then...

[0044]

[0045] That is, if the processing formula is a mid-term formula, then

[0046]

[0047] That is, if the processing formula is a post-processing formula, then

[0048]

[0049] in, Represents the design value of the formula, y new The formula values ​​are estimated based on the existing raw tobacco and formula ratios calculated in step S3.

[0050] Furthermore, the method for correcting any selected indicator by location and processing batch is as follows:

[0051] The error y in predicting tobacco leaves from raw tobacco data in historical data. e The parts and batches are divided into 9 categories: upper, middle, and lower parts, and front, middle, and rear batches. A correction coefficient is estimated for each category. 1-9 represent the recipes for the early stage of the upper part, the middle stage of the upper part, the late stage of the upper part, the early stage of the middle part, the middle stage of the middle part, the late stage of the middle part, and the early stage of the lower part, respectively.

[0052] Early stage formula, mid-term formula, late stage formula; 9 correction coefficients are:

[0053]

[0054] The prediction model is used to input the corresponding correction coefficients according to the design values ​​of its location and batch.

[0055] Compared with the prior art, the beneficial effects of the present invention are reflected in:

[0056] 1. This invention can establish a correction model based on historically accumulated data, making the digital design value of the formula closer to the actual test value, thereby improving the usability of the model.

[0057] 2. Based on years of testing data and actual formula results, this invention analyzes various factors from the design end to the production end, and forms multiple correction models to solve the problem of mismatch between design values ​​and actual product values. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the process of this invention.

[0059] Figure 2 This is a diagram showing the comparison between the predicted and measured values ​​of total sugar in tobacco leaf module in 2021 (after comparison and correction).

[0060] Figure 3 This is a diagram comparing the predicted and measured values ​​of nicotine in the tobacco leaf module in 2021 (after comparison and correction).

[0061] Figure 4 This is a diagram showing the comparison between the predicted and measured values ​​of total sugar in tobacco leaf modules in 2022 (after comparison and correction).

[0062] Figure 5 This is a diagram comparing the predicted and measured values ​​of nicotine in the tobacco leaf module in 2022 (after comparison and correction). Detailed Implementation

[0063] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0064] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0065] The present invention will now be described in detail with reference to the accompanying drawings and exemplary embodiments.

[0066] Example 1

[0067] Take the tobacco leaf module formula design of a re-drying processing center in a certain area as an example.

[0068] Select all batches of selected raw tobacco from the processing center and sample them according to the number of tobacco leaf batches.

[0069] After sampling, the samples were prepared into powder samples according to the tobacco industry standard "YC / T 31-1996 Preparation of Tobacco and Tobacco Products Samples and Determination of Moisture by Oven Method" (the tobacco leaves were placed in an oven and dried at 40°C for 4 hours, then ground with a cyclone mill (FOSS) and passed through a 40-mesh sieve), and sealed for equilibration.

[0070] The powder sample was collected using a fixed near-infrared instrument at the re-drying processing center, and the spectrum was converted into various digital indicators according to the model, including total sugar, nicotine, reducing sugar, fraction, clarity, and elegance.

[0071] The test results of multiple samples from the same batch are averaged to form a digital indicator for that batch of tobacco leaves.

[0072] The formula is designed by determining the batches used and the proportion of batches used. The digital index of the batches used in the tobacco sheet formula module is the weighted average of the batches, which is the digital index of the tobacco sheet formula module (as shown in Table 1 below).

[0073] Table 1. Digital Indicators of the Tobacco Formula Module

[0074]

[0075]

[0076] After the formula is processed, the actual finished product is subjected to near-infrared spectroscopy, and the digital index value of the finished product is obtained.

[0077] Table 2. Digital Indicator Values ​​of Finished Products

[0078] Total sugar Nicotine reducing sugars Part clear grace YN0122NAC 29.1 2.5 21.0 1.9 1.2 6.4

[0079] Continuing the above process, we modeled the data from 2017 to 2020, calculated various correction coefficients, and used the raw tobacco data, formula ratios, and finished product values ​​from 2021 and 2022 as verification.

[0080] The near-infrared measured values ​​from the tobacco re-drying plant laboratory in 2021 and 2022 were used as the standard values, and the predicted values ​​from the raw tobacco were used as the control. The average absolute errors of the control and the four correction methods were recorded as e0, e1, e2, e3, and e4, respectively.

[0081] Table 3 shows the average absolute error of the four correction methods.

[0082] Predictions for 2021 Mean Absolute Error Total sugar Nicotine reducing sugars Part clear Elegance 2 Comparison e0 2.69 0.08 1.86 0.11 0.33 0.27 direct correction e1 1.06 0.08 1.68 0.08 0.25 0.14 Parts e2 1.02 0.07 1.63 0.07 0.25 0.14 In batches e3 1.07 0.08 1.67 0.08 0.25 0.14 Parts and batches e4 0.99 0.07 1.58 0.07 0.24 0.13

[0083] Table 4 shows the predicted average absolute error for the 2022 comparison and the four correction methods.

[0084] Predictions for 2022 Mean Absolute Error Total sugar Nicotine reducing sugars Part clear Elegance 2 Comparison e0 3.58 0.13 2.49 0.12 0.23 0.38 direct correction e1 1.86 0.12 2.30 0.12 0.16 0.24 Parts e2 1.67 0.11 2.24 0.11 0.17 0.23 In batches e3 1.76 0.12 2.30 0.11 0.16 0.24 Parts and batches e4 1.63 0.10 2.23 0.10 0.16 0.22

[0085] It can be seen that, compared with no correction (control), the error between the formula design value and the actual product value was significantly reduced after model correction. Among them, the correction effect of different parts and batches was the best. The average absolute error of total sugar in the tobacco leaf module decreased from 2.69 to 0.99 in 2021, and the average absolute error of total sugar in the tobacco leaf module decreased from 3.58 to 1.63 in 2022. The errors of other indicators also decreased simultaneously.

[0086] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A digital tobacco formula design method based on detection, matching, correction, adjustment, characterized in that, Includes the following steps: S1. Testing: For all types of tobacco leaves that have been purchased and are ready for processing, samples are taken according to the minimum particle size of the tobacco leaf formula and near-infrared spectroscopy is performed. S2. Based on near-infrared spectroscopy detection and corresponding index prediction models, digital indicators for various types of tobacco leaves are obtained. S3. Blending: Based on the indicator results and the quantity of tobacco leaves, the blender generates an empirical blending sheet. S4. Based on the proportions of the empirical formula sheet and the digital indicators of various types of tobacco leaves, generate the digital indicators of the empirical formula sheet. S5. Correction: Based on the differences between historical raw tobacco data and actual data, the digital indicators of the generated empirical formula sheet are corrected to obtain the correction model. S6, Adjustment: The formulator judges the digital indicators of the experience formula sheet, and adjusts the experience formula sheet in step S3 by combining the indicator values ​​of previous years with the tobacco production situation of the current year. S7. Repeat steps S3-S6 to obtain the adjusted empirical formula sheet for subsequent production. In step S1, the sampling factors include, but are not limited to, tobacco production area, grade, variety, and batch. In step S2, the prediction indicators include conventional chemical composition indicators and quality indicators. The conventional chemical composition indicators include total sugar, nicotine, reducing sugar, chlorine, potassium, and total nitrogen. The quality indicators include part, aroma, and sensory characteristics. In step S4, the formula for calculating the numerical value of the empirical formula is as follows: （1） Where n is the number of raw material categories in the empirical formula sheet. For the index value of raw material category h, This refers to the proportion of this type of raw material among all the raw materials in the formula; In step S5, the model correction includes, but is not limited to: direct correction, correction by part, correction by processing batch, and correction by both part and processing batch. The method for directly correcting any selected indicator is as follows: Estimating the error in predicting tobacco flakes from raw tobacco data in historical data Where y is the value of the empirical formula estimated in step S4 based on the proportion of raw tobacco and empirical formula. These are the actual values ​​measured in tobacco sheets; The correction factor is: （2） For formulations requiring design, after obtaining the raw tobacco test data and the proportions of the empirical formulation sheet, the design value is: （3） in, The empirical formula value is estimated based on the existing raw tobacco and empirical formula ratio calculated in step S4, and 'a' is the correction coefficient. The method for correcting any selected index by location is as follows: Error in predicting tobacco leaves from historical data (using raw tobacco data) The calculations are performed separately for the upper, middle, and lower parts of the tobacco leaf, with three errors calculated for each part. 1, 2, and 3 represent the upper, middle, and lower parts, respectively, and the correction coefficients for the three parts are: ；（4） The predicted formula is substituted into the corresponding correction coefficient according to its part design value; that is, if the processing formula is an upper part formula, then... （5） That is, if the processing formula is a central formula, then （6） That is, if the processing formula is the lower part of the formula, then （7） in, Represents the design value of the formula. The empirical formula value is the estimated value of the existing raw tobacco and empirical formula ratio calculated according to step S4.

2. The digital tobacco formulation design method based on detection, proportioning, correction, and adjustment as described in claim 1, characterized in that, The method for correcting any selected index by processing batch is as follows: Error in predicting tobacco leaves from historical data (using raw tobacco data) The tobacco sheets were separated into three batches: the front, middle, and rear batches, and three errors were calculated for each batch. 1, 2, and 3 represent the pre-processing, mid-processing, and post-processing batches, respectively. The correction factors for the three batches are: （8） The prediction model incorporates its batch design values ​​into the corresponding correction coefficients; that is, if the processing formula is the initial formula, then... （9） That is, if the processing formula is a mid-term formula, then （10） That is, if the processing formula is a post-processing formula, then （11） in, Represents the design value of the formula. The empirical formula value is the estimated value of the existing raw tobacco and empirical formula ratio calculated according to step S4.

3. The digital tobacco formulation design method based on detection, proportioning, correction, and adjustment as described in claim 1, characterized in that, The method for correcting any selected index by location and processing batch is as follows: Error in predicting tobacco leaves from historical data (using raw tobacco data) The parts and batches are divided into nine categories: upper, middle, and lower parts, and front, middle, and rear batches. Nine errors are calculated for each category. 1-9 represent the upper early stage formula, upper middle stage formula, upper late stage formula, middle early stage formula, middle middle stage formula, middle late stage formula, lower early stage formula, lower middle stage formula, and lower late stage formula, respectively; the 9 part-batch correction coefficients are: (12) The prediction model is used to input the corresponding correction coefficients according to the design values ​​of its location and batch.

Images