A method for optimizing the formulation of tobacco bud inhibitor

Abstract

The present invention provides a method for optimizing the formula of a tobacco bud inhibitor, which belongs to the technical field of tobacco bud inhibitors, and comprises: firstly obtaining the utility data of a basic formula and an alternative formula, and using a Bayesian network model to analyze the relationship between each component and the utility, and establishing a formula utility equation. On this basis, a sequence ratio is set to calculate theoretical utility data, and multiple groups of preferred formulas are obtained through screening and clustering. The researchers also constructed a multi-branch bidirectional neural network model, and obtained an optimized formula model after training and verification. Finally, Latin hypercube sampling is used to generate preferred utility data, which is input into the model to obtain a preferred formula set, and the optimal formula combination is determined through orthogonal experiments to complete experimental verification. The present invention makes full use of the method combining experimental data and modeling analysis, and cyclically optimizes to finally meet the expected optimized formula, thereby solving the technical problem that most existing methods use a large number of experiments for optimization, and it is difficult to obtain accurate formula optimization results.

Description

Technical Field

[0001] The invention belongs to the technical field of tobacco sprout inhibitors, and in particular relates to a method for optimizing the formula of a tobacco sprout inhibitor. Background Art

[0002] As an important economic crop, the cultivation and processing of tobacco has always received widespread attention. After topping, the apical dominance of tobacco plants disappears, and axillary buds grow wildly, which will reduce the quality and economic value of tobacco leaves. In order to inhibit the germination of tobacco, researchers have developed various tobacco bud inhibitors. These tobacco bud inhibitors interfere with the balance of plant growth hormones and inhibit the development of axillary buds, thereby achieving the purpose of improving tobacco leaf varieties.

[0003] At present, the common tobacco bud inhibitor formulas on the market mainly include chemical ingredients such as dimethoate and butachlor. These bud inhibitors can effectively inhibit the development of tobacco axillary buds and improve the quality of tobacco leaves. However, the existing tobacco bud inhibitor formulas also have some problems:

[0004] 1. The sprout inhibition effect is not ideal. The existing sprout inhibitors have certain differences in the sprout inhibition effect on tobacco varieties or growing environments, making it difficult to achieve comprehensive and effective sprout inhibition.

[0005] 2. The duration of effect is short. Most sprout inhibitors can only play a sprout inhibitory role in the short term, and tobacco is still prone to axillary buds during the growth period.

[0006] 3. Impact on tobacco leaf quality. Some chemical components may have an adverse effect on tobacco leaf quality indicators such as color and smell.

[0007] 4. The formulation optimization lacks systematicity. The existing formulation optimization of sprout inhibitors mostly adopts empirical methods, which makes it difficult to fully consider the impact of various factors on the sprout inhibition effect.

[0008] Most of the existing formulation optimization methods use a large number of experiments for optimization, which makes it difficult to obtain accurate formulation optimization results. Summary of the invention

[0009] In view of this, the present invention provides a method for optimizing the formula of a tobacco bud inhibitor, which can solve the technical problem that most existing formula optimization methods use a large number of experiments for optimization and it is difficult to obtain accurate formula optimization results.

[0010] The present invention is achieved in that:

[0011] The present invention provides a method for optimizing the formulation of a tobacco sprout inhibitor, comprising the following steps:

[0012] S01. Obtain basic formula data of various proportions of tobacco bud inhibitors, including various components and their proportions; and obtain efficacy data of the tobacco bud inhibitors by experimental means, including bud inhibitory effect, duration of effectiveness, and effect on tobacco leaf quality;

[0013] S02, obtaining alternative formula data of various ratios of the tobacco bud inhibitor, including alternative components and their ratios of each component; and obtaining the effectiveness data of the tobacco inhibitor obtained by replacing the corresponding components with the alternative components and their ratios by experimental means;

[0014] S03. Using the Bayesian network model, the relationship between each component and utility is analyzed through conditional probability, and the association mapping between each component, alternative component and each utility data is obtained to represent the probability dependency relationship between each variable;

[0015] S04. Calculate the association matrix of each component, alternative component and all utility data according to the association mapping;

[0016] S05. According to the correlation matrix, power or interaction terms and nonlinear terms are set to establish a group of tobacco bud inhibitor effectiveness equations, including a bud inhibitor effect equation, a lasting effect equation and a quality influence equation;

[0017] S06, fitting the tobacco bud inhibitor effectiveness equation group using basic formula data of various ratios and alternative formula data to obtain a fitting equation group;

[0018] S07, setting a sequence ratio for each component and alternative component, obtaining group ratio data of the sequence ratio, inputting the data into the fitting equation group for calculation, obtaining corresponding theoretical utility data, and generating a formula theoretical utility matrix according to the calculation results of the group ratio data of the sequence ratio and the corresponding theoretical utility data;

[0019] S08, screening the formula theoretical utility matrix, deleting elements that do not meet the preset basic utility data range, and obtaining a screening matrix; the basic utility data range is an array composed of the basic range of each utility in the utility data;

[0020] S09, clustering the screening matrix to obtain multiple groups of cluster centers, and conducting experiments with the recipe data corresponding to the cluster centers to obtain experimental utility data;

[0021] S10, establishing a training set according to the screening matrix, taking the group ratio data of the sequence ratio as the training input, and taking the corresponding theoretical utility data as the training output; and establishing a verification set, specifically, combining the basic formula data of the multiple ratios obtained, the alternative formula data of the multiple ratios obtained, and the formula data corresponding to the multiple cluster centers as the verification input, and taking the utility data obtained from each formula data experiment as the verification output; training a bidirectional neural network model with a multi-branch structure to obtain a tobacco bud inhibitor formula model;

[0022] S11, generating multiple groups of preferred utility data from a preset preferred utility data range, inputting each group of generated preferred utility data into the tobacco bud inhibitor formula model to obtain corresponding preferred formula data, and merging the preferred formula data corresponding to each group of preferred utility data into a preferred formula set;

[0023] S12. Establish an orthogonal experiment based on the preferred formula set, determine the experimental factors and levels through the orthogonal design table, conduct experiments and collect data, determine the significance of each factor through variance analysis, and finally obtain the optimal formula combination

[0024] S13, experimentally verify the optimal formula combination to obtain actual utility data; if the actual utility data meets the preset preferred utility data range, it is determined as the final optimized formula; if not, return to step S10, adjust the neural network model parameters or structure, retrain the model and iterate the optimization process until the final optimized formula that meets the requirements is obtained.

[0025] The various components of the basic formula data are specifically choline chloride, uniconazole, water, emulsifier, preservative, pH regulator, viscosity regulator, stabilizer, and colorant.

[0026] Furthermore, alternative components of the choline chloride include: ethephon or morpholine dehydrogenase inhibitors; alternative components of the uniconazole include: nicotine, methylcyclopropene or 1-methylcyclopropene (1-MCP); alternative components of the emulsifier include: Tween series, Span series or lecithin; alternative components of the preservative include: potassium sorbate, sodium benzoate or natamycin; alternative components of the pH regulator include: citric acid, phosphoric acid or sodium hydroxide; alternative components of the viscosity regulator include: sodium carboxymethyl cellulose (CMC-Na), xanthan gum or guar gum; alternative components of the stabilizer include: hydroxypropyl methylcellulose (HPMC), polyvinyl pyrrolidone (PVP) or gum arabic; alternative components of the colorant include: food coloring, plant extract coloring or mineral pigment.

[0027] The following is the formula expression of each equation in the tobacco bud inhibitor effectiveness equation:

[0028] 1. Bud inhibition effect equation:

[0029] ;

[0030] In the formula, For the sprout inhibition effect; For the The concentration of the components; is the total number of components; For time; is the coefficient to be determined; is a function of concentration changing with time, which can be obtained by fitting experimental data;

[0031] 2. Duration equation:

[0032] ;

[0033] In the formula, The duration of effect; For the The concentration of the components; is the total number of components; The number of key components ; is a complex variable; is a characteristic point on the complex plane; is a complex function to be determined, which can be obtained by fitting experimental data; is the integral path; is the coefficient to be determined; is the Laplace operator;

[0034] 3. Quality influence equation:

[0035] ;

[0036] In the formula, is the quality impact index; For the The concentration of the components; is the total number of components; The maximum number of combinations considered ; is the Heaviside step function; is the coefficient to be determined;

[0037] How to obtain variables and parameters:

[0038] 1. : Obtained by precise weighing and preparation;

[0039] 2. : measured using precise timing equipment;

[0040] 3. : Obtained by fitting experimental data using the least squares method or other optimization algorithms;

[0041] 4. : Obtained by fitting experimental data, polynomial fitting or other nonlinear regression methods can be used;

[0042] 5. : Determined by analyzing the distribution characteristics of experimental data on the complex plane;

[0043] 6. : Heaviside step function, defined as:

[0044] ;

[0045] 7. : The Laplace operator is defined in a three-dimensional rectangular coordinate system as:

[0046]

[0047] These equations take into account the interaction between components, time effects, nonlinear relationships, and complex mathematical operations, including integration, differentiation, and complex number operations. Each equation contains multiple parameters, which makes the model highly flexible and adaptable. By fitting and optimizing experimental data, the specific values ​​of these parameters can be obtained, thereby establishing an accurate prediction model for the effectiveness of tobacco bud inhibitors.

[0048] Among them, in the step of fitting the tobacco bud inhibitor effectiveness equation group using basic formula data of multiple ratios and alternative formula data, the fitting method adopts the least squares method.

[0049] The method for generating multiple groups of preferred utility data is to use Latin hypercube sampling.

[0050] Among them, the bidirectional neural network model with multi-branch structure includes a sprout inhibition effect branch network, a lasting effect branch network, a quality influence branch network and a fusion network.

[0051] Furthermore, the sprout inhibition effect branch network is used to predict the sprout inhibition effect, the input is the data of each component and its ratio, the output is the predicted sprout inhibition effect value, and the structure is a bidirectional LSTM network plus an attention mechanism;

[0052] The effective period branch network is used to predict the effective period, the input is the data of each component and its ratio, the output is the predicted effective period value, and the structure is a bidirectional GRU network plus residual connection;

[0053] The quality impact branch network is used to predict the impact on tobacco leaf quality. The input is the data of each component and its ratio, and the output is the predicted quality impact value. The structure is a bidirectional Transformer network.

[0054] The fusion network is used to integrate the outputs of the three branch networks, the input is the output of the three branch networks, the output is the final utility prediction value, and the structure is a multi-layer perceptron plus a self-attention mechanism.

[0055] The training steps of the tobacco bud inhibitor formula model are specifically:

[0056] Data preprocessing: standardize and normalize the input data.

[0057] Branch network training: train the branch networks for sprout inhibition effect, duration of effectiveness and quality influence separately.

[0058] Fusion network training: Use the outputs of the branch networks to train the fusion network.

[0059] Model integration: Integrate the trained branch network and fusion network into a complete tobacco bud inhibitor formulation model.

[0060] Model Validation: Use validation set data to evaluate model performance and perform fine-tuning if necessary.

[0061] Specifically, the step S01 includes: obtaining basic formula data of tobacco bud inhibitors with various ratios, including the types of various components in the tobacco bud inhibitors and their specific ratios, mainly including choline chloride, uniconazole, water, emulsifiers, preservatives, pH regulators, viscosity regulators, stabilizers and colorants, etc.; and obtaining the efficacy data of the tobacco bud inhibitors under these basic formulas through experiments, including bud inhibition effect, duration of effectiveness and impact on tobacco leaf quality, etc.

[0062] Wherein, the step S02 specifically includes: obtaining alternative formula data of tobacco bud inhibitors with various ratios, including alternative components of each component and their ratios; and also obtaining the effectiveness data of tobacco inhibitors obtained by replacing corresponding components with alternative components and their ratios through experiments.

[0063] Wherein, the step S03 specifically includes: using the Bayesian network model to analyze the relationship between each component and utility through conditional probability, and obtaining an association mapping between each component, alternative component and each utility data, which is used to represent the probabilistic dependency relationship between each variable.

[0064] Wherein, the step S04 specifically includes: according to the association mapping obtained in step S03, calculating the association matrix between each component, alternative component and all utility data, and displaying the correlation strength between each component / alternative component and different utility indicators.

[0065] The step S05 specifically includes: according to the correlation matrix obtained in step S04, power terms, interaction terms and nonlinear terms are set to establish a tobacco bud inhibitor efficacy equation group, including a bud inhibitor effect equation, a lasting effect equation and a quality influence equation. These complex mathematical terms are set to better fit the nonlinearity and interaction between components observed in the experimental data.

[0066] Among them, the step S06 specifically includes: using basic formula data of various ratios and alternative formula data to fit the tobacco bud inhibitor effectiveness equation group established in step S05, and using the least squares method or other optimization algorithms to estimate the parameters of the equation group, so that the equation group can describe the experimentally observed effectiveness data as accurately as possible.

[0067] Among them, the step S07 specifically includes: first, setting a series of different ratio sequences for each component and alternative component, and then inputting the group ratio data of these sequence ratios into the fitting equation group obtained in step S06 for calculation to obtain the corresponding theoretical utility data; next, generating a formula theoretical utility matrix according to the calculation results of the group ratio data of these sequence ratios and their corresponding theoretical utility data.

[0068] The step S08 specifically includes: first setting a basic acceptable range for each utility index, then screening the theoretical utility matrix of the formula obtained in step S07, deleting the matrix elements that do not meet the preset ranges, and obtaining a screened matrix. In this way, those formula solutions that theoretically meet the basic utility requirements can be retained.

[0069] Among them, the step S09 specifically includes: performing cluster analysis on the screening matrix obtained in step S08, using unsupervised learning algorithms such as k-means clustering or Gaussian mixture model to divide these formula schemes into multiple clusters, and then selecting the center point of each cluster as a representative formula, performing experimental verification, and obtaining the actual utility data of these representative formulas.

[0070] Among them, the step S10 specifically includes: first, establishing a training set based on the sequence ratio data generated in step S07, taking these ratio data as input, and taking the corresponding theoretical utility data as training output; at the same time, establishing a verification set, combining the basic formula data obtained in step S01, the alternative formula data obtained in step S02, and the cluster center formula data in step S09 as verification input, and the actual utility data obtained from each formula experiment as verification output; then training a bidirectional neural network model with a multi-branch structure to obtain the final tobacco bud inhibitor formula prediction model.

[0071] Among them, the step S11 specifically includes: first setting an ideal preferred utility data range, and then using the Latin hypercube sampling method to generate multiple groups of target utility data that meet this preferred range; inputting each group of generated target utility data into the tobacco bud inhibitor formula model trained in step S10 to obtain corresponding preferred formula data; integrating these preferred formula data and the corresponding target utility data into a preferred formula set.

[0072] Wherein, the step S12 specifically includes: establishing an orthogonal experimental design according to the preferred formula set obtained in step S11, determining the factors and levels of the experiment, and using the orthogonal experimental design table to determine the specific experimental scheme; conducting experiments and collecting data, and finally determining the significant influence of each factor on the final utility through the variance analysis method, and determining the optimal formula combination based on this.

[0073] Wherein, the step S13 specifically includes: conducting actual experimental verification on the optimal formula combination determined in step S12 to obtain actual utility data; if these actual utility data meet the preferred utility data range preset in step S11, then determining the optimal formula combination as the final optimized formula; if the actual utility data does not meet the requirements, then returning to step S10, adjusting the parameters or structure of the neural network model, retraining the model and iterating the optimization process until the final optimized formula that meets the requirements is obtained.

[0074] Compared with the prior art, the method for optimizing the formulation of a tobacco bud inhibitor provided by the present invention has the following beneficial effects:

[0075] 1. The bud inhibition effect is significantly improved. The method of the present invention can comprehensively analyze the influence of various formula factors on the bud inhibition effect, and establish an accurate bud inhibition effect prediction model based on experimental data. By optimizing the formula, the bud inhibition effect of the bud inhibitor can be significantly improved to meet the needs of tobacco in different varieties and growth environments.

[0076] 2. The effective period is greatly extended. Based on the establishment of the sprout inhibitor effect model, the method of the present invention also establishes a mathematical model for the effective period. By optimizing the ratio of various components, the effective period of the tobacco sprout inhibitor can be significantly extended, and the sprout loss of tobacco during storage and transportation can be reduced.

[0077] 3. No harm to tobacco leaf quality. The method of the present invention takes into account the impact on tobacco leaf quality when optimizing the formula. By establishing a mathematical model of quality impact, a formula combination with less impact on tobacco leaf color, odor and other indicators can be selected to ensure that the tobacco leaf quality is not damaged.

[0078] 4. Systematic optimization method. Different from simple empirical optimization, the present invention adopts advanced data analysis and modeling technologies such as Bayesian network and neural network, comprehensively considers various factors, and establishes a systematic optimization framework. This method can not only obtain the optimal formula, but also the models and data accumulated in the process also provide an important basis for subsequent formula design.

[0079] In summary, the present invention solves the technical problem that most existing formulation optimization methods use a large number of experiments to perform optimization, making it difficult to obtain accurate formulation optimization results. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0081] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0082] like Figure 1 FIG. 1 is a flow chart of a method for optimizing a tobacco bud inhibitor formulation provided by the present invention, and the method comprises the following steps:

[0083] S01. Obtain basic formula data of various proportions of tobacco bud inhibitors, including various components and their proportions; and obtain efficacy data of tobacco bud inhibitors by experimental means, including bud inhibitory effect, duration of effectiveness, and impact on tobacco leaf quality;

[0084] S02. Obtaining alternative formula data of various ratios of tobacco bud inhibitors, including alternative components and their ratios of each component; and obtaining the efficacy data of tobacco inhibitors obtained by replacing corresponding components with alternative components and their ratios by experimental means;

[0085] S03. Using the Bayesian network model, the relationship between each component and utility is analyzed through conditional probability, and the association mapping between each component, alternative component and each utility data is obtained to represent the probability dependency relationship between each variable;

[0086] S04. Calculate the association matrix of each component, alternative component and all utility data according to the association mapping;

[0087] S05. According to the correlation matrix, power or interaction terms and nonlinear terms are set to establish a group of tobacco bud inhibitor effectiveness equations, including a bud inhibitor effect equation, a lasting effect equation and a quality influence equation;

[0088] S06, fitting the tobacco bud inhibitor effectiveness equation group using basic formula data of various ratios and alternative formula data to obtain a fitting equation group;

[0089] S07, setting a sequence ratio for each component and alternative component, obtaining group ratio data of the sequence ratio, inputting the data into the fitting equation group for calculation, obtaining corresponding theoretical utility data, and generating a formula theoretical utility matrix according to the calculation results of the group ratio data of the sequence ratio and its corresponding theoretical utility data;

[0090] S08, screening the recipe theory utility matrix, deleting elements that do not meet the preset basic utility data range, and obtaining a screening matrix; the basic utility data range is an array composed of the basic range of each utility in the utility data;

[0091] S09, clustering the screening matrix to obtain multiple groups of cluster centers, and conducting experiments with the recipe data corresponding to the cluster centers to obtain experimental utility data;

[0092] S10, establishing a training set according to the screening matrix, taking the group ratio data of the sequence ratio as the training input, and taking the corresponding theoretical utility data as the training output; and establishing a verification set, specifically, combining the basic formula data of multiple ratios obtained, the alternative formula data of multiple ratios obtained, and the formula data corresponding to multiple groups of cluster centers as the verification input, and the utility data obtained from each formula data experiment as the verification output; training a bidirectional neural network model with a multi-branch structure to obtain a tobacco bud inhibitor formula model;

[0093] S11, generating multiple groups of preferred utility data from a preset preferred utility data range, inputting each group of generated preferred utility data into a tobacco bud inhibitor formulation model to obtain corresponding preferred formulation data, and merging the preferred formulation data corresponding to each group of preferred utility data into a preferred formulation set;

[0094] S12. Establish an orthogonal experiment based on the optimal formula set, determine the experimental factors and levels through the orthogonal design table, conduct experiments and collect data, determine the significance of each factor through variance analysis, and finally obtain the optimal formula combination

[0095] S13, experimentally verify the optimal formula combination to obtain actual utility data; if the actual utility data meets the preset preferred utility data range, it is determined as the final optimized formula; if not, return to step S10, adjust the neural network model parameters or structure, retrain the model and iterate the optimization process until the final optimized formula that meets the requirements is obtained.

[0096] The specific implementation methods of the above steps are described in detail below:

[0097] The specific implementation method of step S01 is: first, basic formula data of tobacco bud inhibitors with multiple ratios are obtained. These basic formula data include the types of various components in the tobacco bud inhibitors and their specific ratios, mainly including choline chloride, uniconazole, water, emulsifier, preservative, pH regulator, viscosity regulator, stabilizer and colorant, etc. The efficacy data of tobacco bud inhibitors under these basic formulas are obtained by experimental means, including bud inhibitor effect, duration of effect and influence on tobacco leaf quality, etc.

[0098] The specific implementation of step S02 is: obtaining alternative formula data of tobacco bud inhibitors with multiple ratios. These alternative formula data include alternative components and their ratios of each component. Similarly, the effectiveness data of tobacco inhibitors obtained by replacing corresponding components with alternative components and their ratios are obtained by experimental means.

[0099] The specific implementation method of step S03 is: using the Bayesian network model, through conditional probability analysis of the relationship between each component and utility, to obtain the association mapping between each component, alternative component and each utility data. In this way, the probability dependency relationship between each variable can be represented. The Bayesian network model is a probabilistic graph model that can effectively represent the conditional independence relationship and probability dependency relationship between variables. By establishing a Bayesian network model, the correlation between each component and utility can be inferred.

[0100] The specific implementation of step S04 is: according to the association map obtained in step S03, the association matrix between each component, alternative component and all utility data is calculated. This association matrix shows the correlation strength between each component / alternative component and different utility indicators.

[0101] The specific implementation method of step S05 is: according to the correlation matrix obtained in step S04, power terms, interaction terms and nonlinear terms are set to establish a tobacco bud inhibitor utility equation group. This utility equation group includes a bud inhibitor effect equation, a lasting effect equation and a quality influence equation. These complex mathematical terms are set to better fit the nonlinearity observed in the experimental data, the interaction between components and other characteristics.

[0102] The specific implementation of step S06 is: using the basic formula data of multiple ratios and the alternative formula data to fit the tobacco bud inhibitor effectiveness equation group established in step S05 to obtain a fitting equation group. Here, the least square method or other optimization algorithms are used to estimate the parameters of the equation group, so that the equation group can describe the effectiveness data observed in the experiment as accurately as possible.

[0103] The specific implementation of step S07 is: first, a series of different ratio sequences are set for each component and replacement component. Then, the group ratio data of these sequence ratios are input into the fitting equation group obtained in step S06 for calculation to obtain the corresponding theoretical utility data. Next, a formula theoretical utility matrix is ​​generated based on the group ratio data of these sequence ratios and the calculation results of the corresponding theoretical utility data. This matrix shows the utility index values ​​theoretically predicted under different formula combinations.

[0104] The specific implementation of step S08 is: screening the theoretical utility matrix of the formula obtained in step S07. First, the basic acceptable range of each utility index is set, and then the matrix elements that do not meet these preset ranges are deleted to obtain a screened matrix. In this way, those formula solutions that theoretically meet the basic utility requirements can be retained.

[0105] The specific implementation of step S09 is: cluster analysis is performed on the screening matrix obtained in step S08. Unsupervised learning algorithms such as k-means clustering or Gaussian mixture model can be used here to divide these formula schemes into multiple clusters. Then the center point of each cluster is selected as a representative formula, and experimental verification is performed to obtain the actual utility data of these representative formulas.

[0106] The specific implementation method of step S10 is: first, a training set is established based on the sequence ratio data generated in step S07, with these ratio data as input and the corresponding theoretical utility data as training output. At the same time, a verification set is established, and the basic formula data obtained in step S01, the alternative formula data obtained in step S02 and the cluster center formula data in step S09 are combined as verification input, and the actual utility data obtained from each formula experiment is used as verification output. Then a bidirectional neural network model with a multi-branch structure is trained to obtain the final tobacco bud inhibitor formula prediction model. This neural network model includes a bud inhibition effect branch network, a lasting period branch network, a quality influence branch network and a fusion network. The bud inhibition effect branch network adopts a bidirectional LSTM network plus an attention mechanism, the lasting period branch network adopts a bidirectional GRU network plus a residual connection, the quality influence branch network adopts a bidirectional Transformer network, and the fusion network adopts a multi-layer perceptron plus a self-attention mechanism. Through this multi-branch structure, the complex relationship between different utility indicators can be better portrayed.

[0107] The specific implementation method of step S11 is: first set an ideal preferred utility data range. Then use the Latin hypercube sampling method to generate multiple groups of target utility data that meet this preferred range. Input each group of generated target utility data into the tobacco bud inhibitor formula model trained in step S10 to obtain the corresponding preferred formula data. These preferred formula data and the corresponding target utility data are integrated into a preferred formula set. This preferred formula set provides a theoretical basis for subsequent experimental optimization.

[0108] The specific implementation of step S12 is: establish an orthogonal test design according to the preferred formula set obtained in step S11. First, determine the factors and levels of the test, where the factors may include the ratio of each component, etc. Then use the orthogonal test design table to determine the specific test plan. Conduct experiments and collect data, and finally determine the significant impact of each factor on the final effect through the variance analysis method. Based on this, determine the optimal formula combination.

[0109] The specific implementation of step S13 is: the optimal formula combination determined in step S12 is verified by actual experiments to obtain actual utility data. If these actual utility data meet the preferred utility data range preset in step S11, the optimal formula combination is determined as the final optimized formula. If the actual utility data does not meet the requirements, return to step S10, adjust the parameters or structure of the neural network model, retrain the model and iterate the optimization process until the final optimized formula that meets the requirements is obtained.

[0110] Through the above 13 steps, the mathematical model of tobacco bud inhibitor formula was first established, covering the three core indicators of bud inhibitor effect, duration and quality impact. These mathematical models use complex mathematical forms, such as power terms, interaction terms, nonlinear terms, etc., which can better fit the complex relationships observed in the experiment. Then, the Bayesian network and neural network models were used to analyze the correlation between each component and utility, and a formula optimization model integrating a multi-branch structure was established. Finally, the optimal tobacco bud inhibitor formula was determined through orthogonal experimental design and experimental verification.

[0111] The entire optimization process makes full use of statistics, machine learning, numerical analysis and other technical means to achieve the full process design from experimental data to mathematical models and finally to the optimized formula. This systematic optimization method can not only obtain the optimal formula that meets the requirements, but also the models and data accumulated in the process provide important basis and reference for subsequent formula design.

[0112] The core idea of ​​the tobacco sprout inhibitor formula optimization method of the present invention is: first, systematically collect various formula factors and utility data to establish a complete mathematical model; then use advanced data analysis and machine learning technology to dig out the internal laws between various factors and utility; finally, based on the optimization model, explore the best formula combination. This method can comprehensively consider various factors and find the optimal balance point, thereby obtaining a tobacco sprout inhibitor with excellent performance.

[0113] Specifically, the technical principle of the method of the present invention is as follows:

[0114] 1. Data Collection and Modeling

[0115] First, the basic formula data of tobacco bud inhibitors with various ratios were obtained, including various components and their ratios. At the same time, the efficacy data of these formulas were obtained through experimental tests, including bud inhibition effect, duration of effectiveness and impact on tobacco leaf quality.

[0116] Secondly, a mathematical model was established to describe the relationship between these efficacy indicators and formulation factors. , using complex equations containing power terms, interaction terms, and nonlinear terms for modeling:

[0117]

[0118] in For the The concentration of the components, This mathematical form can fit the complex relationship observed in the experiment very well.

[0119] For the duration and quality impact , and also established similar mathematical models involving advanced mathematical tools such as integration, differentiation, and complex functions.

[0120] 2. Association analysis and feature extraction

[0121] With the formula data and utility data, we need to analyze the correlation between them. Here we use the Bayesian network model to analyze the correlation between each component and each utility indicator through conditional probability. The resulting correlation matrix Indicates the impact strength of each factor on the target utility.

[0122] Based on the incidence matrix , and dig out the key characteristic factors that affect each utility index. This provides an important basis for subsequent optimization plans.

[0123] 3. Optimization modeling and solution

[0124] With the above mathematical model and key features, we can carry out the optimization design of the system. First, set a series of ratio sequences for each component and input them into the model to predict the theoretical utility to form a formula theoretical utility matrix. .

[0125] Then, Screening and clustering were performed to find representative formulas that met the preset utility requirements. Next, these representative formulas were used as validation samples to train a multi-branch neural network model to obtain the final tobacco bud inhibitor formula prediction model.

[0126] Finally, a series of target utility data were generated by Latin hypercube sampling method and input into the prediction model to obtain the corresponding optimal formula. Through orthogonal test, further verification and optimization were carried out to finally determine the optimal formula combination that meets various requirements.

[0127] Through the above three steps, the method of the present invention fully explores the laws contained in the experimental data and establishes a system optimization model covering the bud inhibition effect, duration of effectiveness and quality impact. This data-driven, model-oriented method can find the best balance point of various factors and obtain a tobacco bud inhibitor formula with excellent performance.

[0128] In order to better understand and implement the present invention, a specific embodiment 1 of the present invention is provided below, and the steps are specifically described as follows:

[0129] The specific implementation of step S01 is: first, obtain the basic formula data of tobacco sprout inhibitors with multiple ratios. These basic formula data include the types and specific ratios of various components in the tobacco sprout inhibitor, mainly including choline chloride, uniconazole, water, emulsifier, preservative, pH regulator, viscosity regulator, stabilizer and colorant. The concentrations of these components can be To indicate that The total number of components. The efficacy data of tobacco bud inhibitors under these basic formulas were obtained through experiments, including the bud inhibition effect , Duration of effect and its influence on tobacco leaf quality wait.

[0130] The specific implementation of step S02 is: obtaining alternative formula data of tobacco bud inhibitors with multiple ratios. These alternative formula data include alternative components and their ratios for each component. Similarly, the effectiveness data of tobacco inhibitors obtained by replacing corresponding components with alternative components and their ratios are obtained by experimental means. The concentrations of these alternative components can be used To express.

[0131] The specific implementation of step S03 is: using the Bayesian network model, through conditional probability analysis of the relationship between each component and utility, to obtain the association mapping between each component, alternative component and each utility data. The Bayesian network model is a probabilistic graph model that can effectively represent the conditional independence relationship and probabilistic dependence relationship between variables. By establishing a Bayesian network model, the association between each component and utility can be inferred. Here, conditional probability can be used , and To represent this association mapping.

[0132] The specific implementation of step S04 is: according to the association map obtained in step S03, the association matrix of each component, the replacement component and the entire utility data is calculated. This association matrix can be expressed as:

[0133] ;

[0134] in Indicates The components / alternative components and The strength of the correlation between the utility indicators.

[0135] The specific implementation of step S05 is: according to the association matrix obtained in step S04 , set the power term, interaction term and nonlinear term, and establish the utility equation group of tobacco bud inhibitor. This utility equation group can be expressed as:

[0136] Bud inhibition effect equation:

[0137] ;

[0138] Duration equation:

[0139]

[0140] Quality influence equation:

[0141] ;

[0142] in is the coefficient to be determined, A function to be determined. is the Laplace operator, is the Heaviside step function. These complex mathematical forms can better fit the nonlinearity and interaction between components observed in experiments.

[0143] The specific implementation of step S06 is: using the basic formula data of multiple ratios and the alternative formula data to fit the tobacco bud inhibitor effectiveness equations established in step S05 to obtain a fitted equation. The least square method or other optimization algorithms are used to estimate the parameters of the equations so that the equations can describe the effectiveness data observed in the experiment as accurately as possible. The objective function of the least square method can be expressed as:

[0144]

[0145] in is the parameter vector to be optimized, is the utility data observed in the experiment, is the corresponding model prediction value.

[0146] The specific implementation of step S07 is: first, a series of different ratio sequences are set for each component and replacement component. ,in , The group ratio data of these sequence matching are then input into the fitting equation group obtained in step S06 for calculation to obtain the corresponding theoretical utility data. Next, a formula theoretical utility matrix is ​​generated based on the group ratio data of these sequence ratios and their corresponding theoretical utility data calculation results. :

[0147]

[0148] The specific implementation of step S08 is: first set the basic acceptable range of each utility indicator , and Then the recipe theoretical utility matrix obtained in step S07 is Filter and delete matrix elements that do not meet these preset ranges to obtain a filtered matrix .

[0149] The specific implementation of step S09 is: the screening matrix obtained in step S08 Perform cluster analysis. Here, unsupervised learning algorithms such as k-means clustering or Gaussian mixture model can be used to divide these recipes into Clusters Then select the center point of each cluster As representative formulas, experimental verification is carried out to obtain the actual utility data of these representative formulas .

[0150] The specific implementation of step S10 is: first, based on the sequence matching data generated in step S07 Create a training set with these ratio data as input , with the corresponding theoretical utility data As training output. At the same time, a verification set is established, and the basic formula data obtained in step S01, the alternative formula data obtained in step S02, and the cluster center formula data in step S09 are combined as verification input , the actual utility data obtained from each formulation experiment As the verification output. Then a multi-branch bidirectional neural network model is trained to obtain the final tobacco bud inhibitor formula prediction model. This neural network model can be expressed as:

[0151] Bud suppression effect branch network:

[0152]

[0153]

[0154] Duration of validity period:

[0155]

[0156]

[0157] Quality affects branch network:

[0158]

[0159]

[0160] Converged Network:

[0161]

[0162] in and are the parameters of each sub-network. Through this multi-branch structure, the complex relationship between different utility indicators can be better characterized.

[0163] The specific implementation of step S11 is: first set the ideal preferred utility data range Then Latin hypercube sampling is used to generate The target utility data that meets this preferred range is Input each group of generated target utility data into the tobacco bud inhibitor formula model trained in step S10 to obtain the corresponding optimal formula data . Integrate these optimal recipe data and the corresponding target utility data into an optimal recipe set .

[0164] The specific implementation of step S12 is: according to the preferred formula set obtained in step S11 Establish an orthogonal experimental design. First determine the factors of the experiment and level. The factors here can include the ratio of each component, etc. Then use the orthogonal test design table to determine the specific test plan . Conduct experiments and collect data Finally, the variance analysis method was used to determine the significant impact of each factor on the final utility. Based on this, the optimal formula combination is determined .

[0165] The specific implementation of step S13 is: combine the optimal formula determined in step S12 Conduct actual experimental verification to obtain actual utility data If these actual utility data meet the preferred utility data range preset in step S11 , then determine the optimal formula combination is the final optimization formula. If the actual utility data does not meet the requirements, return to step S10 to adjust the parameters or structure of the neural network model, for example:

[0166] 1) Adjust hyperparameters such as the number of network layers and neurons

[0167] 2) Try different activation functions, regularization methods, etc.

[0168] 3) Change the selection and preprocessing of input features

[0169] 4) Adjust the weight of the loss function

[0170] Retrain the model and iterate the optimization process until the final optimization formula that meets the requirements is obtained.

[0171] Through the above 13 steps, the mathematical model of tobacco bud inhibitor formula was first established, covering the three core indicators of bud inhibitor effect, duration and quality impact. These mathematical models use complex mathematical forms, such as power terms, interaction terms, nonlinear terms, etc., which can better fit the complex relationships observed in the experiment. Then, the Bayesian network and neural network models were used to analyze the correlation between each component and utility, and a formula optimization model integrating a multi-branch structure was established. Finally, the optimal tobacco bud inhibitor formula was determined through orthogonal experimental design and experimental verification.

[0172] In order to further better understand and implement the present invention, Example 2 of a specific application scenario of the present invention is provided below: In view of the problems existing in the existing tobacco sprout inhibitor formula, a tobacco research and development institution decided to adopt the optimization method proposed in the present invention to develop a new type of sprout inhibitor with better performance.

[0173] First, the research and development institute collected 10 basic sprout inhibitor formulas with different ratios, as well as formula data for 15 alternative components. Through experimental testing, specific data on these formulas in terms of sprout inhibitor effect, duration of effectiveness, and quality impact were obtained, as shown in Table 1:

[0174] Table 1 Experimental test data table

[0175]

[0176] Based on these experimental data, the research and development institute first used the Bayesian network model to analyze the correlation between each component and the three utility indicators. After calculation, the correlation matrix was obtained.

[0177]

[0178] From the correlation matrix, we can see that choline chloride Ethyleneconazole The biggest influence on the duration of efficacy is water content. This has the greatest impact on quality. This provides an important basis for subsequent optimization design.

[0179] Next, the research and development institute designed a series of different group ratio sequences based on the above correlation analysis results, input them into the established mathematical model for prediction, obtained theoretical utility data, and formed a formula theoretical utility matrix. In order to select the formula that meets the basic requirements, the R&D organization set the following acceptable range of efficacy indicators:

[0180] Bud suppression effect

[0181] Duration sky

[0182] Quality Impact point

[0183] Will The solutions that do not meet the above range are eliminated, and the screened matrix is ​​obtained. .

[0184] Next, the research and development institute K-means cluster analysis was performed to divide the formula into 5 clusters. The center point of each cluster was selected as the representative formula for actual experimental verification. The results are shown in Table 2:

[0185] Table 2 Cluster center formula data table

[0186]

[0187] Based on these experimental data, the research and development institution constructed a training set and a validation set, and trained a multi-branch bidirectional neural network model as the final tobacco bud inhibitor formula prediction model.

[0188] After the model training was completed, the R&D institute further explored a better formulation. First, they set the ideal preferred utility data range:

[0189] Bud suppression effect

[0190] Duration sky

[0191] Quality Impact point

[0192] Then, the Latin hypercube sampling method was used to generate 20 sets of target utility data that met the above requirements. These target utility data were input into the trained prediction model, and the corresponding 20 optimal formulation schemes were obtained, as shown in Table 3:

[0193] Table 3 Preferred formula

[0194]

[0195] In order to further determine the optimal formula, the research and development institute designed an orthogonal experiment based on these 20 preferred solutions. First, five main factors were determined: choline chloride , uniconazole , moisture , emulsifier and pH adjusters Each factor was set at 3 levels, and 9 experimental schemes were designed using the L9 (3^5) orthogonal table, as shown in Table 4:

[0196] Table 4 Experimental plan

[0197]

[0198] Through variance analysis, the R&D institute determined the significant impact of each factor on the final utility:

[0199] Choline chloride :

[0200] Uniconazole :

[0201] Moisture :

[0202] Emulsifier :

[0203] pH Adjusters :

[0204] Based on the above analysis results, the research and development institution finally determined the optimal tobacco bud inhibitor formula as follows:

[0205]

[0206]

[0207]

[0208]

[0209]

[0210] This formula was used for actual verification, and its effectiveness indicators are as follows:

[0211] Bud suppression effect: 92%

[0212] Duration: 46 days

[0213] Quality impact: 97 points

[0214] These results all met the optimization requirements pre-set by the R&D organization, so this formula was determined as the final optimization solution.

[0215] In this embodiment 2, through the optimization design of the above system, the research and development institution successfully developed a new tobacco bud inhibitor with excellent performance. Compared with the prior art, this bud inhibitor not only has a significantly improved bud inhibitory effect, but also greatly extends the effective period and has almost no effect on the quality of tobacco leaves. This is attributed to the optimization method proposed in the present invention, which can comprehensively analyze the influence of various factors on the efficacy and find the best formula combination.

[0216] In addition, this data-driven, model-oriented optimization method is not only applicable to the formulation design of tobacco bud inhibitors, but can also be extended to the formulation optimization of other chemicals or materials. It makes full use of advanced technical means such as statistics and machine learning to systematically solve problems that are difficult to solve with traditional empirical optimization, and ultimately obtain the best formula.

[0217] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for optimizing the formulation of a tobacco bud inhibitor, characterized in that: The following steps are involved: S01. Obtain basic formula data of various proportions of tobacco bud inhibitors, including various components and their proportions; And using experimental methods to obtain the efficacy data of the tobacco bud inhibitor, including bud inhibitor effect, duration of effectiveness, and impact on tobacco leaf quality; S02, obtaining alternative formula data of various ratios of the tobacco bud inhibitor, including alternative components of each component and their ratios; And experimentally obtain the efficacy data of tobacco suppressants obtained by replacing corresponding components with alternative components and their proportions; S03. Using the Bayesian network model, the relationship between each component and utility is analyzed through conditional probability, and the association mapping between each component, alternative component and each utility data is obtained to represent the probability dependency relationship between each variable; S04. Calculate the association matrix of each component, alternative component and all utility data according to the association mapping; S05. According to the correlation matrix, power or interaction terms and nonlinear terms are set to establish a group of tobacco bud inhibitor effectiveness equations, including a bud inhibitor effect equation, a lasting effect equation and a quality influence equation; S06, fitting the tobacco bud inhibitor effectiveness equation group using basic formula data of various ratios and alternative formula data to obtain a fitting equation group; S07, setting a sequence ratio for each component and alternative component, obtaining group ratio data of the sequence ratio, inputting the data into the fitting equation group for calculation, obtaining corresponding theoretical utility data, and generating a formula theoretical utility matrix according to the calculation results of the group ratio data of the sequence ratio and the corresponding theoretical utility data; S08, screening the formula theoretical utility matrix, deleting elements that do not meet the preset basic utility data range, and obtaining a screening matrix; the basic utility data range is an array composed of the basic range of each utility in the utility data; S09, clustering the screening matrix to obtain multiple groups of cluster centers, and conducting experiments with the recipe data corresponding to the cluster centers to obtain experimental utility data; S10, establishing a training set according to the screening matrix, using the group ratio data of the sequence ratio as training input, and using the corresponding theoretical utility data as training output; A validation set is established, specifically, the basic formula data of the multiple ratios obtained, the alternative formula data of the multiple ratios obtained, and the formula data corresponding to the multiple groups of cluster centers are combined as validation input, and the utility data obtained from the experiment of each formula data is used as validation output; A bidirectional neural network model with a multi-branch structure is trained to obtain a tobacco sprout inhibitor formula model; S11, generating multiple groups of preferred utility data from a preset preferred utility data range, inputting each group of generated preferred utility data into the tobacco bud inhibitor formula model to obtain corresponding preferred formula data, and merging the preferred formula data corresponding to each group of preferred utility data into a preferred formula set; S12, establishing an orthogonal experiment according to the preferred formula set, determining the experimental factors and levels through an orthogonal design table, conducting experiments and collecting data, determining the significance of each factor through variance analysis, and finally obtaining the optimal formula combination; S13, experimentally verifying the optimal formula combination to obtain actual utility data; If the actual utility data meets the preset preferred utility data range, it is determined as the final optimization formula; if not, return to step S10, adjust the neural network model parameters or structure, retrain the model and iterate the optimization process until the final optimization formula that meets the requirements is obtained; The sprout inhibition effect equation is specifically expressed as: ； In the formula, For the sprout inhibition effect; For the The concentration of the components; is the total number of components; For time; is the coefficient to be determined; is the function of concentration changing with time; ； In the formula, The duration of effect; For the The concentration of the components; is the total number of components; is the number of key components, and ; is a complex variable; is a characteristic point on the complex plane; is a complex function to be determined; is the integral path; is the coefficient to be determined; is the Laplace operator; The quality influence equation is specifically expressed as: ； In the formula, is the quality impact index; For the The concentration of the components; is the total number of components; is the maximum number of combinations considered and ; is the Heaviside step function; is the coefficient to be determined.

2. A method for optimizing the formulation of a tobacco bud inhibitor according to claim 1, characterized in that: The various components of the basic formula data specifically include choline chloride, uniconazole, water, emulsifier, preservative, pH regulator, viscosity regulator, stabilizer, and colorant.

3. A method for optimizing the formulation of a tobacco bud inhibitor according to claim 2, characterized in that: The alternative components of the choline chloride include: ethephon or morpholine dehydrogenase inhibitor; the alternative components of the uniconazole include: nicotine, methylcyclopropene or 1-methylcyclopropene (1-MCP); the alternative components of the emulsifier include: Tween series, Span series or lecithin; the alternative components of the preservative include: potassium sorbate, sodium benzoate or natamycin; the alternative components of the pH regulator include: citric acid, phosphoric acid or sodium hydroxide; the alternative components of the viscosity regulator include: sodium carboxymethyl cellulose (CMC-Na), xanthan gum or guar gum; the alternative components of the stabilizer include: hydroxypropyl methylcellulose (HPMC), polyvinyl pyrrolidone (PVP) or gum arabic; the alternative components of the colorant include: food coloring, plant extract coloring or mineral pigment.

4. A method for optimizing the formulation of a tobacco sprout suppressant according to claim 1, characterized in that: In the step of fitting the tobacco bud inhibitor effectiveness equation group using basic formula data of various ratios and alternative formula data, the fitting method adopts the least square method.

5. A method for optimizing the formulation of a tobacco sprout suppressant according to claim 1, characterized in that: The method for generating multiple groups of preferred utility data is to adopt Latin hypercube sampling.

6. A method for optimizing the formulation of a tobacco sprout suppressant according to claim 1, characterized in that: The bidirectional neural network model with a multi-branch structure includes a sprout inhibition effect branch network, a lasting effect branch network, a quality influence branch network and a fusion network.

7. A method for optimizing the formulation of a tobacco bud inhibitor according to claim 6, characterized in that: The sprout inhibition effect branch network is used to predict the sprout inhibition effect, the input is the data of each component and its ratio, the output is the predicted sprout inhibition effect value, and the structure is a bidirectional LSTM network plus an attention mechanism; The effective period branch network is used to predict the effective period, the input is the data of each component and its ratio, the output is the predicted effective period value, and the structure is a bidirectional GRU network plus residual connection; The quality impact branch network is used to predict the impact on tobacco leaf quality. The input is the data of each component and its ratio, and the output is the predicted quality impact value. The structure is a bidirectional Transformer network. The fusion network is used to integrate the outputs of the three branch networks, the input is the output of the three branch networks, the output is the final utility prediction value, and the structure is a multi-layer perceptron plus a self-attention mechanism.

Images