Baking process parameter optimization method based on tobacco leaf drying kinetic model

Abstract

The invention provides a tobacco leaf drying kinetic model-based baking process parameter optimization method, and belongs to the technical field of electrical digital data process.The tobacco leaf drying kinetic model-based baking process parameter optimization method comprises the following steps of: firstly, establishing a tobacco leaf drying differential equation model and a curing barn computational fluid mechanics model; temperature and humidity field distribution is obtained through simulation analysis; calculating a tobacco leaf surface heat and mass transfer coefficient, and establishing a moisture migration prediction function; a moisture transfer optimization function is introduced to perform parameter preliminary screening, and a pre-trained quality evaluation neural network model is adopted to evaluate a parameter scheme; respectively optimizing three stages of a yellowing stage, a color fixing stage and a stem drying stage to form a parameter set; and finally, an optimal parameter combination is searched by applying a genetic algorithm, an optimization target is a comprehensive score of tobacco leaf quality consistency and energy consumption, real-time optimization of parameters in the baking process is realized, and the technical problem that tobacco leaf baking process parameters are difficult to accurately optimize aiming at different tobacco leaf characteristics in the prior art is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric digital data processing. Specifically, it relates to a method for optimizing baking process parameters based on a tobacco leaf drying kinetic model. Background Art

[0002] Baking is an important link in the tobacco leaf production process to ensure raw material supply, improve raw material quality, and highlight the raw material style. Its essence is the transfer of heat and moisture between the tobacco leaves and the environment, promoting the yellowing and water loss of tobacco leaves and ensuring the full conversion of internal substances. Traditional baking processes mainly rely on experience to set process parameters such as temperature, humidity, and wind speed. In the prior art, hot air circulation bulk curing barns are often used for tobacco leaf baking, and the environmental parameters of the curing barn are controlled by manual adjustment or preset baking curves. Although these methods can ensure the basic baking quality under specific conditions, they lack a systematic theoretical analysis of the heat and mass transfer mechanism during the baking process and dynamic optimization of the process.

[0003] With the development of computational fluid dynamics and numerical simulation technologies, some studies have begun to apply airflow field simulation and heat and mass transfer theories to analyze the distribution of environmental parameters in the curing barn. However, these methods are often limited to single-parameter optimization or specific-stage analysis, lacking systematic integration and the ability of full-process dynamic regulation, and unable to perform targeted optimization according to factors such as different tobacco leaf varieties and initial moisture content.

[0004] The current technology is difficult to solve the contradiction between the complexity of parameter combinations and the consistency of tobacco leaf quality during the tobacco leaf baking process, especially in the three key stages of the yellowing period, color fixation period, and dry rib period, there are obvious deficiencies in parameter connection and coordinated optimization, resulting in problems such as unstable tobacco leaf baking quality and excessive energy consumption. There is an urgent need to establish a precise baking process parameter optimization method based on the drying kinetic characteristics of tobacco leaves. That is to say, there is a technical problem in the prior art that it is difficult to precisely optimize the baking process parameters of tobacco leaves according to different tobacco leaf characteristics. Summary of the Invention

[0005] In view of this, the present invention provides a method for optimizing baking process parameters based on a tobacco leaf drying kinetic model, which can solve the technical problem in the prior art that it is difficult to precisely optimize the baking process parameters of tobacco leaves according to different tobacco leaf characteristics.

[0006] The present invention is implemented as follows: The present invention provides an optimization method for baking process parameters based on a tobacco leaf drying kinetic model, which includes: establishing a differential equation model of tobacco leaf drying kinetics to determine the relationship function between the moisture diffusion coefficient and temperature and humidity; constructing a computational fluid dynamics model of a bulk curing barn; simulating and analyzing the distribution of the internal air flow field in the bulk curing barn; establishing a prediction function for moisture migration during the tobacco leaf drying process; introducing an optimization function for tobacco leaf moisture transfer to preliminarily screen the baking process parameters; using a pre-trained neural network model for evaluating the quality of tobacco leaves to evaluate the baking process parameter scheme; optimizing the baking process parameters for the three baking stages of the yellowing stage, color fixing stage, and dry rib stage respectively to form a set of stage-optimized parameters; using a genetic algorithm to find the optimal combination of baking process parameters, with the objective function being the comprehensive score of the tobacco leaf quality consistency index and energy consumption; and realizing the real-time optimization of the baking process parameters during the baking process according to the optimal combination of baking process parameters.

[0007] Among them, the differential equation model of tobacco leaf drying kinetics describes the relationship between the internal moisture migration rate of tobacco leaves and temperature, humidity, and pressure gradient, and is expressed by a nonlinear partial differential equation, including parameters such as the diffusion coefficient, specific surface area, and porosity.

[0008] Among them, in the computational fluid dynamics model of the bulk curing barn, hexahedral structured grids are used for grid division, and the boundary condition settings include the inlet air velocity field and the inlet air temperature field; the inlet air velocity field refers to the air flow velocity distribution at the inlet of the hot air circulation system of the bulk curing barn, measured in meters per second; the inlet air temperature field refers to the temperature distribution at the inlet of the hot air circulation system of the bulk curing barn, measured in degrees Celsius.

[0009] Among them, the prediction function for moisture migration during the tobacco leaf drying process is established based on the calculation results of the heat and mass transfer coefficients on the surface of tobacco leaves under different loading densities; the heat and mass transfer coefficient on the surface of tobacco leaves is a physical quantity that characterizes the intensity of heat and moisture exchange between tobacco leaves and the surrounding air, and is related to the air flow velocity, temperature difference, and humidity difference.

[0010] Among them, the input parameters of the optimization function for tobacco leaf moisture transfer include the baking environment temperature parameter, environmental relative humidity parameter, environmental air flow velocity parameter, average thickness parameter of tobacco leaf blades, and initial moisture content parameter of tobacco leaves; the output is the moisture migration efficiency index and the moisture distribution uniformity score within the predicted baking time.

[0011] Among them, the baking environment temperature parameter refers to the temperature value of the air around the tobacco leaves in the curing barn, with the unit of degree Celsius; the ambient relative humidity parameter refers to the relative humidity percentage of the air around the tobacco leaves in the curing barn; the ambient air flow velocity parameter refers to the flow velocity of the air around the tobacco leaves in the curing barn, with the unit of meter per second; the average thickness parameter of the tobacco leaf refers to the average thickness of the tobacco leaves to be baked, with the unit of millimeter; the initial moisture content parameter of the tobacco leaf refers to the water content percentage of the tobacco leaf before entering the curing barn.

[0012] Among them, the moisture migration efficiency index refers to the ratio of the water loss of the tobacco leaf per unit time to the theoretical maximum water loss; the moisture distribution uniformity score is a quantitative index of the difference in moisture content between different parts of the tobacco leaf, and the higher the value, the more uniform the moisture distribution.

[0013] Among them, the neural network model structure for tobacco leaf quality evaluation is an architecture that combines a multi-layer convolutional neural network and a Transformer structure, including a convolutional layer for extracting tobacco leaf image features, an embedding layer for tobacco leaf physical and chemical indexes, a multi-head cross-attention mechanism layer, and a multi-task prediction output layer for tobacco leaf quality; the attention mechanism parameters in the neural network model for tobacco leaf quality evaluation are automatically adjusted according to the tobacco leaf variety type; the attention mechanism parameters refer to the weight coefficients of the multi-head attention module in the neural network model for tobacco leaf quality evaluation, which are used to adjust the attention degree of the model to different input features.

[0014] Among them, the yellowing stage is the first stage of the tobacco leaf baking process, corresponding to the stage of preliminary water loss and yellowing of the leaves; the color-fixing stage is the second stage of the tobacco leaf baking process, corresponding to the stage of pigment fixation and transformation; the stem-drying stage is the third stage of the tobacco leaf baking process, corresponding to the stage of drying and dehydration of the midrib.

[0015] Among them, the stage optimization parameter set is a combination of temperature, humidity, and wind speed parameters optimized for the three baking stages of the yellowing stage, the color-fixing stage, and the stem-drying stage respectively.

[0016] Compared with the prior art, the present invention provides an optimization method for baking process parameters based on a tobacco leaf drying kinetic model. The present invention proposes an optimization method for baking process parameters based on a tobacco leaf drying kinetic model. By establishing a differential equation model for tobacco leaf drying and a computational fluid dynamics model for a bulk curing barn, and combining a neural network and a genetic algorithm, the precise optimization of baking process parameters is achieved. This method correlates the moisture migration mechanism of tobacco leaves with the baking environment parameters and establishes a multi-scale optimization system from micro to macro.

[0017] The present invention solves the defect that the parameter setting in traditional baking technology lacks theoretical support and dynamic optimization ability. Through the tobacco leaf moisture transfer optimization function and the quality evaluation neural network model, it realizes the precise control of parameters for different tobacco leaf characteristics, especially the collaborative optimization of parameters in three key stages: the yellowing stage, the color fixation stage, and the dry stem stage, significantly improving the consistency of tobacco leaf baking quality.

[0018] By establishing a drying kinetics model and introducing a multi-level optimization algorithm, the present invention successfully solves the technical problem in the prior art that it is difficult to precisely optimize the tobacco leaf baking process parameters for different tobacco leaf characteristics, enabling the baking process to transform from empirical control to intelligent optimization under theoretical guidance, and realizing the synchronous improvement of tobacco leaf baking quality and energy efficiency. Brief Description of the Drawings

[0019] Figure 1 It is a flowchart of the method of the present invention. Detailed Embodiments

[0020] To make the purpose, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions 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.

[0021] As Figure 1 shown, it is a flowchart of a baking process parameter optimization method based on a tobacco leaf drying kinetics model provided by the present invention. This method includes the following steps:

[0022] S01. Establish a differential equation model of tobacco leaf drying kinetics to determine the relationship function between the moisture diffusion coefficient and temperature and humidity;

[0023] S02. Construct a computational fluid dynamics model of a bulk curing barn. The grid division uses hexahedral structured grids, and the boundary condition settings include the inlet air velocity field and the inlet air temperature field;

[0024] S03. Simulate and analyze the distribution of the internal air flow field in the bulk curing barn to obtain the distribution cloud maps of the temperature field and humidity field in the barn;

[0025] S04. Establish a moisture migration prediction function for the tobacco leaf drying process based on the calculation results of the heat and mass transfer coefficients on the surface of tobacco leaves under different loading densities;

[0026] S05. Introduce a tobacco leaf moisture transfer optimization function to preliminarily screen the baking process parameters. The input parameters of the tobacco leaf moisture transfer optimization function include the baking environment temperature parameter, the environmental relative humidity parameter, the environmental air flow velocity parameter, the average thickness parameter of tobacco leaf blades, and the initial moisture content parameter of tobacco leaves;

[0027] S06. Evaluate the baking process parameter scheme using a pre-trained tobacco leaf quality evaluation neural network model, where the attention mechanism parameters in the tobacco leaf quality evaluation neural network model are automatically adjusted according to the tobacco leaf variety type;

[0028] S07. Optimize the baking process parameters for the yellowing stage, color fixing stage, and dry stem stage respectively to form a stage optimization parameter set;

[0029] S08. Use the genetic algorithm to find the optimal combination of baking process parameters, with the objective function being the comprehensive score of the tobacco leaf quality consistency index and energy consumption;

[0030] S09. According to the optimal combination of baking process parameters, realize the real-time optimization of baking process parameters during the baking process;

[0031] Among them, the differential equation model of tobacco leaf drying kinetics is specifically a mathematical equation describing the relationship between the internal moisture migration rate of tobacco leaves and temperature, humidity, and pressure gradients. It is usually expressed by a nonlinear partial differential equation and contains parameters such as diffusion coefficient, specific surface area, and porosity.

[0032] Among them, the inlet air velocity field specifically refers to the air velocity distribution at the inlet of the hot air circulation system in the bulk curing barn, usually measured in meters per second.

[0033] Among them, the inlet air temperature field specifically refers to the temperature distribution at the inlet of the hot air circulation system in the bulk curing barn, usually measured in degrees Celsius.

[0034] Among them, the distribution status of the internal air flow field in the bulk curing barn specifically refers to the air flow motion characteristics in the curing barn calculated by the computational fluid dynamics model, including information such as air flow direction, velocity, and eddy current area.

[0035] Among them, the temperature field distribution nephogram in the curing barn is specifically an intuitive graph of the numerical distribution of temperature values at each spatial point in the curing barn generated by computational fluid dynamics software, used to judge the temperature uniformity and air flow dead corners.

[0036] Among them, the humidity field distribution nephogram in the curing barn is specifically an intuitive graph of the numerical distribution of humidity values at each spatial point in the curing barn generated by computational fluid dynamics software, used to judge the humidity uniformity and air flow dead corners.

[0037] Among them, the heat and mass transfer coefficient on the surface of the tobacco leaf is specifically a physical quantity characterizing the intensity of heat and moisture exchange between the tobacco leaf and the surrounding air, which is related to air flow velocity, temperature difference, and humidity difference.

[0038] Among them, the moisture migration prediction function during the tobacco leaf drying process is specifically a mathematical model established based on heat and mass transfer theory to calculate the relationship between the moisture content of tobacco leaves and time, considering the moisture migration characteristics in the two stages of surface evaporation and internal diffusion.

[0039] Among them, the optimal baking process parameter combination mainly includes temperature parameters, relative humidity parameters, wind speed parameters, and duration parameters for each of the three baking stages: the yellowing stage, the color-fixing stage, and the stem-drying stage, which are optimized respectively.

[0040] Among them, the tobacco leaf moisture transfer optimization function is used to quickly evaluate the moisture transfer efficiency and uniformity of tobacco leaves under different baking process parameter combinations. The inputs include the baking environment temperature parameter, the environmental relative humidity parameter, the environmental air flow speed parameter, the average thickness parameter of tobacco leaf blades, and the initial moisture content parameter of tobacco leaves. The outputs are the moisture migration efficiency index and the moisture distribution uniformity score within the predicted baking time.

[0041] Among them, the baking environment temperature parameter specifically refers to the temperature value of the air around the tobacco leaves in the curing barn, with the unit of degree Celsius, which is measured by the temperature sensor in the bulk curing barn.

[0042] Among them, the environmental relative humidity parameter specifically refers to the relative humidity percentage of the air around the tobacco leaves in the curing barn, which is measured by the humidity sensor in the bulk curing barn.

[0043] Among them, the environmental air flow speed parameter specifically refers to the flow speed of the air around the tobacco leaves in the curing barn, with the unit of meter per second, which is measured by the wind speed sensor in the bulk curing barn.

[0044] Among them, the average thickness parameter of tobacco leaf blades specifically refers to the average thickness of the tobacco leaves to be baked, with the unit of millimeter, which is measured by the leaf thickness detector.

[0045] Among them, the initial moisture content parameter of tobacco leaves specifically refers to the moisture content percentage of the tobacco leaves before entering the curing barn, which is measured by the rapid tobacco leaf moisture content detector.

[0046] Among them, the moisture migration efficiency index specifically refers to the ratio of the water loss of tobacco leaves per unit time to the theoretical maximum water loss, which is used to measure the moisture migration rate during baking.

[0047] Among them, the moisture distribution uniformity score specifically refers to the quantitative index of the difference in moisture content between different parts of tobacco leaves. The higher the value, the more uniform the moisture distribution.

[0048] Among them, the specific structure of the tobacco leaf quality evaluation neural network model is an architecture combining a multi-layer convolutional neural network and a Transformer structure, including a convolutional layer for extracting tobacco leaf image features, an embedding layer for tobacco leaf physical and chemical indexes, a multi-head cross-attention mechanism layer, and a multi-task prediction output layer for tobacco leaf quality. The total number of model parameters reaches 80 million. The sparse attention mechanism is used to reduce the computational complexity, and the number of attention heads is automatically adjusted according to the tobacco leaf variety type. The model depth is 12 layers, the hidden layer dimension is 1024, and the activation function uses the Swish function with a parameter adjustment term.

[0049] Among them, the attention mechanism parameters specifically refer to the weight coefficients of the multi-head attention module in the tobacco leaf quality evaluation neural network model, which are used to adjust the attention degree of the model to different input features.

[0050] Among them, the yellowing stage is specifically the first stage of the tobacco leaf baking process, corresponding to the stage of initial water loss and yellowing of the leaves. The color-fixing stage is specifically the second stage of the tobacco leaf baking process, corresponding to the stage of pigment fixation and transformation. The stem-drying stage is specifically the third stage of the tobacco leaf baking process, corresponding to the stage of drying and dehydration of the midrib.

[0051] Among them, the stage optimization parameter set is specifically a combination of temperature, humidity, and wind speed parameters optimized for the three baking stages of the yellowing stage, the color-fixing stage, and the stem-drying stage.

[0052] Among them, the genetic algorithm is specifically a computational method for simulating natural selection and genetic mechanisms to search for the optimal solution. It iteratively optimizes the combination of baking process parameters through selection, crossover, and mutation operations to avoid falling into local optimal solutions.

[0053] Among them, the tobacco leaf quality consistency index is specifically a quantitative index for evaluating the degree of difference in moisture content, chemical composition, and color of different parts of tobacco leaves after baking in the same batch, usually expressed by the standard deviation or coefficient of variation.

[0054] Among them, the comprehensive energy consumption score is specifically a weighted evaluation index for the total amount of electric energy and heat energy required to complete the baking process, and the weight coefficient is determined according to the actual energy cost.

[0055] Among them, the steps for establishing the training dataset of the tobacco leaf quality evaluation neural network model specifically include collecting the whole-process data of tobacco leaves of different varieties in different producing areas under various baking process parameters, including the temperature, humidity, and wind speed parameters in the baking room, the real-time moisture content change of the tobacco leaves, the image sequence of the tobacco leaf color change, the measured values of the internal chemical components of the tobacco leaves after baking, and the expert scoring values. Cleaning and standardizing the collected data, constructing a dataset containing one hundred thousand groups of baking parameters and corresponding quality score annotations, and dividing the training set, validation set, and test set according to the ratio of 8:1:1.

[0056] Among them, the steps for training the tobacco leaf quality evaluation neural network model specifically include first training the tobacco leaf image data using the self-supervised learning method to obtain the network parameters for image feature extraction, and then training the prediction tasks of tobacco leaf physical and chemical indicators and quality scores using the supervised learning method. During the training process, the learning rate decay strategy and early stopping mechanism are adopted to prevent overfitting. The model training adopts a distributed computing architecture to accelerate the training process. After the training is completed, the knowledge of the large model is transferred to the lightweight model through knowledge distillation technology to improve the actual application efficiency.

[0057] The specific implementation manners of the above steps are described in detail below.

[0058] The specific implementation of step S01 is to establish a partial differential equation describing the dynamic process of moisture migration inside tobacco leaves. First, a second-order nonlinear partial differential equation based on Fick's law is used to describe the moisture transfer process inside tobacco leaves. The equation form is where M represents the moisture content of tobacco leaves, t represents time, D is the moisture diffusion coefficient, and S is the source term. The relationship function between the moisture diffusion coefficient D and temperature and humidity is obtained by fitting experimental data, and the Arrhenius relationship is used where D0 is the frequency factor, E a is the diffusion activation energy, R is the gas constant, and T is the absolute temperature. For different tobacco varieties, the change law of the moisture diffusion coefficient in the temperature range of 25°C to 75°C is measured, and the diffusion coefficient parameters are determined by least squares fitting. The goodness of fit R 2 should be greater than 0.95. Considering the influence of relative humidity, a correction factor f(RH) = a + b·RH + c·RH 2 is introduced, where RH is the relative humidity, and the coefficients a, b, and c are determined by experimental data. The kinetic model established in this step provides a theoretical basis for the optimization of subsequent baking process parameters.

[0059] The specific implementation of step S02 is to construct a computational fluid dynamics model describing the air flow, temperature, and humidity distribution in a bulk curing barn. A three-dimensional steady-state incompressible fluid control equation set is used, including the continuity equation, momentum equation, and energy equation, and the k-ε turbulence model is combined to describe the complex air flow movement in the curing barn. The geometric model is established based on the actual dimensions of the bulk curing barn. The standard curing barn is 8m × 2.7m × 3.5m, considering the arrangement of tobacco racks and the loading of tobacco leaves. The grid division uses hexahedral structured grids. The basic grid size is 5cm, and the grid is refined in the boundary layer region. The minimum grid size is 0.5cm, and the total number of grids is about 3 million. The skewness of the grid quality evaluation index is less than 0.8, and the orthogonal quality is greater than 0.3. The boundary condition settings include the inlet velocity field, which is specified as a uniform velocity inlet, and the velocity range is 0.5m / s to 2.5m / s; the inlet temperature field, which is specified as a constant temperature boundary, and the temperature range is 32°C to 75°C, set according to different baking stages; the outlet is set as a pressure outlet boundary; the tobacco leaf surface is set as a no-slip wall boundary, and there is a heat and mass transfer process with the fluid. This computational fluid dynamics model can accurately simulate the air flow distribution inside the curing barn and provide a basis for optimizing the baking process parameters.

[0060] The specific implementation of step S03 is to obtain the internal flow field distribution information of the bulk curing barn by solving the computational fluid dynamics model constructed in S02. The finite volume method is used to discretize and solve the control equation. The momentum equation uses the second-order upwind scheme, and the pressure-velocity coupling uses the SIMPLE algorithm. The convergence criterion is set as the residual being less than 10 -4. Calculate the output of the computational results, including the air velocity vector field, temperature scalar field, and relative humidity scalar field distributions inside the curing barn. Conduct a visualization analysis of the flow field, generate a contour map of the temperature field distribution inside the curing barn, analyze the temperature uniformity in the area, and control the temperature difference in the area with uneven temperature distribution within 3°C; generate a contour map of the humidity field distribution inside the curing barn, analyze the humidity uniformity in the area, and control the difference in the area with uneven relative humidity distribution within 5%; analyze the flow field characteristics in the dead corner area and evaluate the optimization direction of the flow field. The distribution of the internal air flow field of the curing barn obtained through numerical simulation in this step provides a scientific basis for the subsequent optimization of the baking process parameters.

[0061] The specific implementation method of step S04 is to establish a prediction function for moisture migration during the tobacco leaf drying process based on the heat and mass transfer characteristics under different loading densities. First, based on the numerical simulation results, calculate the heat transfer coefficient h t = q / (T s - T ∞ ) and the mass transfer coefficient h m = j / (Y s - Y ∞ ), where q is the heat flux density, j is the mass flux density, T s and T ∞ are the surface temperature of the tobacco leaf and the ambient temperature respectively, and Y s and Y ∞ are the humidity of the tobacco leaf surface and the ambient air respectively. For different loading densities (10 kg / m 3 to 30 kg / m 3 ) and different air velocities (0.5 m / s to 2.5 m / s) conditions, extract the heat and mass transfer coefficients from the numerical calculation results, establish the relationship function between the heat and mass transfer coefficients and the loading density and air velocity, and fit using the polynomial regression method. Based on the calculation results of the surface heat and mass transfer coefficients, combined with the internal moisture diffusion model of the tobacco leaf, establish a prediction function for moisture migration during the tobacco leaf drying process M(t) = f(M0, T, RH, v, δ, t), where M0 is the initial moisture content, T is the ambient temperature, RH is the relative humidity, v is the air velocity, δ is the leaf thickness, and t is the drying time. This function can predict the change process of the moisture content of tobacco leaves over time under different baking conditions and provide a basis for the optimization of baking process parameters.

[0062] The specific implementation method of step S05 is to introduce an optimization function for tobacco leaf moisture transfer to preliminarily screen the baking process parameters. The optimization function for tobacco leaf moisture transfer uses a multi-objective evaluation method, comprehensively considering the moisture migration efficiency and uniformity. The function form is F opt = w1·E eff + w2·E uni , where E eff is the moisture migration efficiency index, and E uniIt is the moisture distribution uniformity score, w1 and w2 are weight coefficients, and usually w1 = 0.6 and w2 = 0.4. The calculation method of the moisture migration efficiency index is where M0 is the initial moisture content, M t is the moisture content at time t, M e is the equilibrium moisture content, t ref is the reference drying time. The calculation method of the moisture distribution uniformity score is where σ M is the standard deviation of the moisture content, is the average moisture content. The input parameters of the optimization function include the baking environment temperature parameter (32°C to 75°C), the environmental relative humidity parameter (18% to 90%), the environmental air flow velocity parameter (0.5 m / s to 2.5 m / s), the average thickness parameter of the tobacco leaf (0.1 mm to 0.3 mm), and the initial moisture content parameter of the tobacco leaf (80% to 90%). By setting the threshold F opt > 0.7 for preliminary screening, retain the combination of baking process parameters that meet the conditions and enter the next step of evaluation.

[0063] The specific implementation of step S06 is to use the tobacco leaf quality evaluation neural network model to evaluate the screened baking process parameter scheme. In the evaluation process, first convert the combination of baking process parameters into a model input vector, including features such as temperature-time curve, humidity-time curve, and wind speed-time curve; then send it into the pre-trained tobacco leaf quality evaluation neural network model for inference calculation to obtain the predicted tobacco leaf quality score and chemical composition index; finally, sort and screen the process parameter scheme according to the prediction results, and retain the scheme with a quality score higher than 85 points and enter the next step of optimization. The attention mechanism parameters in the tobacco leaf quality evaluation neural network model are automatically adjusted according to the type of tobacco leaf variety. For different tobacco leaf varieties, the number of attention heads in the model is dynamically adjusted between 8 and 16, and the attention weight matrix is adaptively adjusted according to the variety characteristics to focus on important feature dimensions. This step ensures that the screened baking process parameters can obtain high-quality tobacco leaf products.

[0064] The specific implementation of step S07 is to optimize the baking process parameters for three baking stages: the yellowing stage, the color-fixing stage, and the stem-drying stage. For the yellowing stage, the optimal temperature range is determined to be 32°C to 42°C, the relative humidity range is 75% to 85%, the wind speed range is 0.8 m / s to 1.2 m / s, and the duration is 60 to 80 hours; for the color-fixing stage, the optimal temperature range is 45°C to 54°C, the relative humidity range is 60% to 70%, the wind speed range is 1.0 m / s to 1.5 m / s, and the duration is 45 to 65 hours; for the stem-drying stage, the optimal temperature range is 55°C to 68°C, the relative humidity range is 30% to 40%, the wind speed range is 1.5 m / s to 2.0 m / s, and the duration is 32 to 38 hours. Within the parameter ranges of each stage, the uniform design method is used to generate combinations of baking process parameters, forming a set of stage-optimized parameters. The size of the parameter set is 100 groups for each stage, totaling 300 groups of parameter combinations, providing an initial population for the next genetic algorithm optimization. This step realizes the optimization of the process parameters for different baking stages and lays a foundation for the overall baking process optimization.

[0065] The specific implementation of step S08 is to use the genetic algorithm to find the optimal combination of baking process parameters. In the design of the genetic algorithm, real-number coding is used for chromosome coding, and the coding length is 12, including the temperature, humidity, wind speed, and duration parameters of the three stages; the population size is set to 100, and the initial population is composed of the set of stage-optimized parameters generated in step S07; the fitness function is defined as F = w1·Q cons + w2·(1 - E cons ), where Q cons is the tobacco leaf quality consistency index, E cons is the normalized energy consumption score, and w1 and w2 are weight coefficients, usually taking the values w1 = 0.7 and w2 = 0.3; the selection operation uses the tournament selection method, and the tournament size is 3; the crossover operation uses simulated binary crossover, the crossover probability is 0.8, and the distribution index is 20; the mutation operation uses polynomial mutation, the mutation probability is 0.1, and the distribution index is 20; the number of evolutionary generations is set to 100 generations, and the convergence criterion is that the change in the optimal fitness for 10 consecutive generations is less than 0.001. Through iterative optimization of the genetic algorithm, the optimal combination of baking process parameters is obtained, including the temperature, humidity, wind speed, and duration parameters of each stage. This step realizes the global optimization of the baking process parameters and obtains the optimal parameter combination that takes into account both tobacco leaf quality and energy consumption.

[0066] The specific implementation of step S09 is to achieve real-time optimization of the baking process parameters during baking according to the optimal baking process parameter combination. The real-time optimization system adopts the model predictive control method, and the control objective is to track the optimal baking process parameter curve while considering the disturbance factors in the actual baking process. The control system includes a state estimation module, a parameter prediction module, and a control optimization module. The state estimation module is based on the Kalman filtering algorithm, fuses the data of the temperature and humidity sensors in the baking room and the tobacco leaf moisture content sensor, and estimates the state of the tobacco leaves in real time; the parameter prediction module is based on the moisture migration prediction function established in step S04 to predict the change trend of the tobacco leaf moisture content in the future time period; the control optimization module is based on the rolling horizon optimization strategy to optimize the control parameters in the future time window, with a control period of 10 minutes, a prediction horizon of 60 minutes, and a control horizon of 30 minutes. The control system outputs the set values of temperature, humidity, and wind speed, and realizes the precise control of the baking process parameters through a variable frequency fan, a heater, and a humidity regulation system. The system also includes an abnormal situation handling mechanism, which automatically adjusts the control strategy when an abnormality is detected to ensure the safe and stable progress of the baking process. This step realizes the real-time optimization control of the process parameters during baking, improving the baking quality and efficiency.

[0067] The detailed structure of the tobacco leaf quality evaluation neural network model and the specific implementation of the training data set establishment steps are as follows: The tobacco leaf quality evaluation neural network model adopts a hybrid architecture combining a multi-layer convolutional neural network and a Transformer structure. First, the tobacco leaf image features are extracted through a convolutional neural network. The network structure adopts the residual network ResNet-50 architecture, which contains 5 convolutional blocks and a total of 50 convolutional layers. Each convolutional block contains multiple residual units. The residual unit adopts a 1×1, 3×3, and 1×1 convolutional layer stacking structure. The activation function adopts the Swish function f(x)=x·σ(βx) with parameter adjustment items, where σ is a sigmoid function and β is a trainable parameter; the input is a tobacco leaf image with a size of 224×224×3, and the output is a 2048-dimensional feature vector. Then, the tobacco leaf physical and chemical index embedding layer is constructed, and the baking process parameters (temperature, humidity, wind speed time curve) and tobacco leaf basic characteristic parameters (variety, origin, thickness, initial moisture content) are mapped to a 512-dimensional embedding vector through a fully connected layer. Then, a multi-head cross-attention mechanism layer is designed, using the Transformer architecture, which contains 12 layers of encoder blocks. Each layer contains a multi-head self-attention sublayer and a feedforward neural network sublayer. The number of attention heads is automatically adjusted according to the type of tobacco leaf variety, ranging from 8 to 16. The model hidden layer dimension is 1024, and layer normalization and residual connection are used to ensure training stability. Finally, a tobacco leaf quality multi-task prediction output layer is designed to predict the intrinsic chemical composition (nicotine, total sugar, total nitrogen, etc.) and sensory scores (aroma, taste, strength, etc.) of tobacco leaves. A multi-task learning framework is used to comprehensively output the total score of tobacco leaf quality. The training data set establishment process first collects the whole process data of tobacco leaves of different varieties (K326, Yunyan series, Honghua Dajinyuan, etc.) in Yunnan production areas (Kunming, Yuxi, Dali, Chuxiong, Honghe, etc.) under various baking process parameters, including the temperature (32℃ to 75℃), humidity (30% to 85%), wind speed (0.5m / s to 2.5m / s) parameters in the baking room, real-time moisture content change data of tobacco leaves (initial moisture content 80% to 90%, final moisture content 12% to 16%), tobacco leaf color change image sequence (collected once every hour), the internal chemical composition measurement value of tobacco leaves after baking (nicotine 2% to 3.5%, total sugar 15% to 35%, total nitrogen 1.5% to 2.5%, etc.) and expert score value (total score 100 points). The collected data is cleaned and standardized, abnormal data points are removed, and the value range is standardized to the [0, 1] interval. A dataset containing 100,000 sets of baking parameters and corresponding quality score annotations was constructed, and the training set, validation set, and test set were divided into 8:1:1 ratios. The model training adopted a two-stage strategy. First, the tobacco leaf image data was pre-trained using self-supervised learning, and the image feature extraction network parameters were obtained using the contrastive learning method SimCLR. Subsequently, all modules were jointly trained using supervised learning, and the optimization target was the multi-task loss function L=∑i w i L i where L i is the loss function of each sub-task, and w i is the weight coefficient. During the training process, a learning rate decay strategy is adopted. The initial learning rate is 0.001, and it decays by 10% every 10 epochs; an early stopping mechanism is used to prevent overfitting, and the training stops when the validation set loss has not improved for 5 consecutive epochs; the model training adopts a distributed computing architecture to accelerate the training process, and 8 GPUs are used for parallel training. After the training is completed, the knowledge of the large model is transferred to the lightweight model through knowledge distillation technology, reducing the number of parameters to 20% of the original, and improving the actual application efficiency.

[0068] The following details the mathematical models or calculation processes involved in the present invention.

[0069] In step S01, a differential equation model of tobacco leaf drying kinetics is established, which is specifically expressed as follows:

[0070]

[0071] In the formula, M is the moisture content of tobacco leaves (dry basis, kg water / kg dry matter); t is the time (h); D is the moisture diffusion coefficient (m 2 / s); is the gradient operator; S is the source term (kg water / kg dry matter·h), representing the moisture change caused by the internal structure change or chemical reaction of tobacco leaves.

[0072] The relationship function between the moisture diffusion coefficient D and temperature is expressed by the Arrhenius equation:

[0073]

[0074] In the formula, D0 is the frequency factor (m 2 / s), usually in the range of 10 -4 to 10 -2 m 2 / s; E a is the diffusion activation energy (J / mol), and the typical value of tobacco leaves is 20000 to 35000 J / mol; R is the gas constant, with a value of 8.314

[0075] J / (mol·K); T is the absolute temperature (K).

[0076] Considering the influence of relative humidity, a correction factor is introduced:

[0077] f(RH) = a + b·RH + c·RH 2 ;

[0078] Wherein, RH is the relative humidity (%); a, b, and c are fitting coefficients, representing the basic diffusion capacity, linear humidity influence, and non-linear humidity influence respectively.

[0079] The calculation formula for the corrected diffusion coefficient is:

[0080] D eff = D·f(RH)·f(M);

[0081] Wherein, D eff is the effective diffusion coefficient (m 2 / s); f(M) is the moisture content influence factor, expressed as f(M) = d + e·M + f·M 2 , where d, e, and f are fitting coefficients.

[0082] The method for obtaining parameters is as follows:

[0083] The frequency factor D0 and the activation energy E a are determined through experiments. The experimental steps are: (1) Prepare tobacco leaf samples of different varieties; (2) In a constant temperature and humidity drying oven, set temperature gradients (25°C, 35°C, 45°C, 55°C, 65°C, 75°C); (3) Measure the tobacco leaf drying rate under different temperature conditions; (4) Through linear regression of ln(D) against 1 / T, D0 and E are obtained by fitting. a .

[0084] The coefficients a, b, and c are determined through experiments. The experimental steps are: (1) At a constant temperature (55°C), set different relative humidity gradients (30%, 40%, 50%, 60%, 70%, 80%); (2) Measure the diffusion coefficient under different humidity conditions; (3) Use polynomial regression to fit and obtain the coefficients a, b, and c.

[0085] The coefficients d, e, and f are determined through experiments. The experimental steps are: (1) Under constant temperature and humidity conditions, measure the drying rate of tobacco leaves with different initial moisture contents; (2) Use polynomial regression to fit and obtain the coefficients d, e, and f.

[0086] The establishment of these equation relationships is based on the following principles: The Arrhenius relationship reflects the exponential effect of temperature on molecular motion and diffusion processes, where the exponential form indicates that an increase in temperature significantly accelerates the molecular diffusion rate; the humidity influence factor in polynomial form can describe the non-linear effect of relative humidity on moisture transfer, showing a complex relationship where diffusion is slow at low humidity, faster at moderate humidity, and slower again at high humidity; the moisture content influence factor reflects the impact of the internal moisture in tobacco leaves on their own diffusion and migration. As the moisture content decreases, the diffusion resistance increases, so a polynomial form is used to describe this non-linear relationship. This model comprehensively considers the effects of temperature, humidity, and moisture content on the moisture diffusion in tobacco leaves, and is more accurate than traditional single-factor models, providing a more precise theoretical basis for optimizing the baking process parameters.

[0087] In step S02, a computational fluid dynamics model of the bulk curing barn was constructed, and the governing equations are expressed as follows:

[0088] Continuity equation:

[0089]

[0090] In the formula, ρ is the air density (kg / m 3 ); is the velocity vector field (m / s).

[0091] Momentum equation:

[0092]

[0093] In the formula, p is the pressure (Pa); μ eff is the effective viscosity (kg / (m·s)), μ eff = μ + μ t , where μ is the air dynamic viscosity, μ t is the turbulent viscosity; is the body force (N / m 3 ).

[0094] Energy equation:

[0095]

[0096] In the formula, h is the specific enthalpy (J / kg); k eff is the effective thermal conductivity (W / (m·K)); T is the temperature (K); S h is the heat source term (W / m 3 ).

[0097] Humidity transport equation:

[0098]

[0099] Wherein, Y is the absolute humidity of air (kg water / kg dry air); D Y，eff is the effective humidity diffusion coefficient (m 2 / s); S Y is the humidity source term (kg / (m 3 ·s)).

[0100] k-ε turbulence model equation:

[0101] Turbulent kinetic energy equation:

[0102]

[0103] Turbulent dissipation rate equation:

[0104]

[0105] Wherein, k is the turbulent kinetic energy (m 2 / s 2 ); ε is the turbulent dissipation rate (m 2 / s 3 ); G k is the turbulent generation term; σ k , σ ε , C1, C2 are turbulent model constants, generally taking σ k = 1.0, σ ε = 1.3, C1 = 1.44, C2 = 1.92.

[0106] Turbulent viscosity calculation:

[0107]

[0108] Wherein, C μ is an empirical constant, generally taking 0.09.

[0109] These equations are based on the basic laws of fluid mechanics. The continuity equation embodies the principle of mass conservation; the momentum equation is based on Newton's second law; the energy equation embodies the principle of energy conservation; the humidity transport equation refers to the mass transfer theory; the k-ε turbulence model is based on the turbulent statistical theory. In the simulation of the airflow in the curing barn, the turbulence model is particularly important because the airflow in the curing barn often presents a turbulent state, affecting the heat and moisture transfer efficiency. The solution of these equations can accurately simulate the internal airflow distribution of the curing barn and provide a scientific basis for optimizing the baking process parameters.

[0110] In step S04, the heat and mass transfer coefficient on the surface of the tobacco leaves is calculated, and a moisture migration prediction function is established, which is specifically expressed as follows:

[0111] Heat transfer coefficient calculation:

[0112]

[0113] In the formula, h t is the heat transfer coefficient (W / (m 2 ·K)); q is the heat flux density (W / m 2 ); T s is the surface temperature of the tobacco leaf (K); T ∞ is the ambient temperature (K).

[0114] Calculation of the mass transfer coefficient:

[0115]

[0116] In the formula, h m is the mass transfer coefficient (m / s); j is the mass flux density (kg / (m 2 ·s)); Y s is the surface humidity of the tobacco leaf (kg water / kg dry air); Y ∞ is the ambient humidity (kg water / kg dry air).

[0117] Relationship function of the heat transfer coefficient with the loading density and air flow velocity:

[0118]

[0119] In the formula, ρ L is the loading density (kg / m 3 ); v is the air flow velocity (m / s); α0 to α5 are fitting coefficients.

[0120] Relationship function of the mass transfer coefficient with the loading density and air flow velocity:

[0121]

[0122] In the formula, β0 to β5 are fitting coefficients.

[0123] Moisture migration prediction function during the tobacco leaf drying process:

[0124] M(t) = M e +(M0 - M e )·exp(-K·t n );

[0125] In the formula, M(t) is the moisture content of the tobacco leaf at time t (%); M0 is the initial moisture content (%); M e is the equilibrium moisture content (%); K is the drying rate constant (h -n ); n is the drying characteristic index, reflecting the non-linear characteristics of tobacco leaf drying.

[0126] Calculation of the equilibrium moisture content:

[0127]

[0128] In the formula, A and B are material characteristic parameters, which are related to temperature; RH is the relative humidity (in decimal form).

[0129] Calculation of drying rate constant:

[0130]

[0131] In the formula, K0 is the frequency factor (h -n ); E d is the drying activation energy (J / mol); v is the air flow velocity (m / s); δ is the blade thickness (mm); γ1 to γ4 are fitting coefficients.

[0132] The method for obtaining parameters is as follows:

[0133] The coefficients α0 to α5, β0 to β5 are obtained by fitting the numerical simulation results. The specific steps are: (1) Set different tobacco loading densities (10 kg / m 3 to 30 kg / m 3 , with an interval of 5 kg / m 3 ) and different air flow velocity conditions (0.5 m / s to 2.5 m / s, with an interval of 0.5 m / s); (2) Run the CFD simulation; (3) Extract the heat and mass transfer coefficients on the tobacco leaf surface; (4) Use the least squares method to fit the polynomial coefficients.

[0134] The coefficients A and B are determined by experiments. The experimental steps are: (1) In a constant temperature and humidity environment, measure the equilibrium moisture content of tobacco leaves under different temperature and humidity conditions; (2) Use the classical hygroscopic isotherm model (such as the GAB equation) to fit and obtain the A,

[0135] B parameters.

[0136] The coefficients K0, E d , γ1 to γ4, n are determined by experiments. The experimental steps are: (1) Design an orthogonal experiment, considering different temperature, humidity, wind speed and blade thickness conditions; (2) Conduct a tobacco leaf drying experiment and record the data of moisture content changing with time; (3) Use the non-linear regression method to fit the parameters.

[0137] The establishment of these equation relationships is based on the following principles: The heat and mass transfer coefficient formula is derived from the theory of convective heat and mass transfer, reflecting the intensity of heat and mass exchange between the surface and the fluid; the drying kinetics model adopts an exponential form (Page equation), which can describe the characteristics of rapid drying in the initial stage and slow drying in the later stage of tobacco leaves; the equilibrium moisture content model is based on the adsorption theory, reflecting the moisture balance characteristics of tobacco leaves as porous materials; the calculation of the drying rate constant takes into account the comprehensive effects of temperature, wind speed, and leaf thickness, where the temperature effect is described by the Arrhenius relationship, and the wind speed and thickness effects are described by a polynomial form. This moisture migration prediction function comprehensively considers the baking environment and tobacco leaf characteristic parameters, and can accurately predict the drying process of tobacco leaves under different conditions, providing a quantitative basis for optimizing the baking process parameters.

[0138] In step S05, an optimized function for moisture transfer in tobacco leaves is introduced, which is specifically expressed as follows:

[0139] Optimized function for moisture transfer in tobacco leaves:

[0140] F opt = w1·E eff + w2·E uni ;

[0141] In the formula, F opt is the value of the optimized function for moisture transfer; E eff is the moisture migration efficiency index; E uni is the score for moisture distribution uniformity; w1 and w2 are weight coefficients, with values of 0.6 and 0.4 respectively.

[0142] Calculation of the moisture migration efficiency index:

[0143]

[0144] In the formula, M0 is the initial moisture content (%); M t is the moisture content at time t (%); M e is the equilibrium moisture content (%); t ref is the reference drying time (h), usually taken as the time required to reach the target moisture content under the standard baking process; t is the actual drying time (h).

[0145] Calculation of the moisture distribution uniformity score:

[0146]

[0147] In the formula, σ M is the standard deviation of the moisture content (%), expressed as is the average moisture content (%), expressed as N is the number of sampling points; M i is the moisture content at the i-th sampling point (%).

[0148] The establishment of these equation relationships is based on the following principles: The moisture transfer optimization function adopts a weighted summation form, comprehensively considering two objectives of efficiency and uniformity. The weight coefficients reflect the relative importance of the two; the design of the moisture migration efficiency index formula embodies the balance between the drying rate and the degree of drying completion. The former fraction represents the degree of drying completion, and the latter fraction represents the time efficiency. The product of the two can comprehensively evaluate the drying efficiency; the moisture distribution uniformity scoring formula is based on the reciprocal transformation of the coefficient of variation (the ratio of the standard deviation to the mean), so that the higher the uniformity, the closer the score is to 1. This optimization function provides a quantitative evaluation criterion for screening baking process parameters, which helps to screen out baking process parameters that are both efficient and uniform.

[0149] In step S08, the genetic algorithm is used to find the optimal combination of baking process parameters, and the fitness function is expressed as follows:

[0150] F = w1·Q cons + w2·(1 - E cons ) ;

[0151] In the formula, F is the fitness function value; Q cons is the tobacco leaf quality consistency index; E cons is the normalized energy consumption score; w1 and w2 are weight coefficients, taking values of 0.7 and 0.3 respectively.

[0152] Calculation of the tobacco leaf quality consistency index:

[0153]

[0154] In the formula, m is the number of quality indicators; CV j is the coefficient of variation of the jth quality indicator, expressed as where σ j is the standard deviation, and μ j is the mean value.

[0155] Calculation of the comprehensive energy consumption score:

[0156]

[0157] In the formula, E is the energy consumption (kWh); E min and E max are the minimum and maximum energy consumptions (kWh) in all schemes respectively.

[0158] Calculation of the energy consumption:

[0159] E = η1·E heat + η2·E fan + η3·E hum ;

[0160] Wherein, E heat is the heating energy consumption (kWh); E fan is the fan energy consumption (kWh); E hum is the humidity regulation energy consumption (kWh); η1, η2, η3 are the energy cost weight coefficients, which are determined according to the local electricity price and heat source cost.

[0161] Calculation of heating energy consumption:

[0162]

[0163] Wherein, P heat (t) is the heating power at time t (kW); t total is the total baking time (h).

[0164] Calculation of fan energy consumption:

[0165]

[0166] Wherein, P fan (t) is the fan power at time t (kW).

[0167] Calculation of humidity regulation energy consumption:

[0168]

[0169] Wherein, P hum (t) is the humidity regulation power at time t (kW).

[0170] Calculation of heating power:

[0171] P heat (t) = c p ·m air ·(T in (t) - T out (t));

[0172] Wherein, c p is the specific heat capacity of air (kJ / (kg·K)); m air is the air mass flow rate (kg / s); T in (t) and T out (t) are the inlet and outlet temperatures of the heater at time t (K), respectively.

[0173] Calculation of fan power:

[0174]

[0175] Wherein, Δp is the fan pressure drop (Pa); Q is the air volume flow rate (m 3 / s); η fanIt is the fan efficiency, usually ranging from 0.6 to 0.8.

[0176] The establishment of these equation relationships is based on the following principles: The fitness function adopts a weighted summation form, comprehensively considering two objectives of tobacco leaf quality and energy consumption. The weight coefficient reflects the higher importance of quality relative to energy consumption. The tobacco leaf quality consistency index is based on the coefficient of variation and can quantitatively evaluate the uniformity of tobacco leaf quality. The energy consumption score is normalized to make the energy consumption of different process schemes comparable. The energy consumption considers the energy consumption of heating, the fan, and humidity regulation, and is weighted according to their respective costs. The energy consumption of each part is calculated by integration, reflecting the cumulative energy consumption of the entire baking process. The heating power calculation is based on the principle of heat transfer, and the fan power calculation is based on the principle of fluid mechanics. These equations provide clear optimization objectives for the genetic algorithm and can guide the algorithm to find the optimal combination of baking process parameters that takes into account both tobacco leaf quality and energy consumption.

[0177] Optionally, the state - space representation of model predictive control:

[0178] x k+1 = Ax k + Bu k + w k ;

[0179] In the formula, x k is the state vector, including the temperature, humidity, and moisture content of tobacco leaves in the curing barn; u k is the control input vector, including the heater power, humidity regulator setting value, and fan speed; A is the state - transfer matrix; B is the control matrix; w k is the process noise vector.

[0180] Optionally, the measurement equation:

[0181] y k = Cx k + v k ;

[0182] In the formula, y k is the measurement output vector, including the temperature, humidity, and moisture content of tobacco leaves measured by sensors; C is the observation matrix; v k is the measurement noise vector.

[0183] Optionally, the Kalman filter state estimation:

[0184]

[0185] P k|k-1 = AP k-1|k-1 A T + Q;

[0186] K k = Pk|k-1 C T (CP k|k-1 C T +R) -1 ;

[0187]

[0188] P k|k =(I-K k C)P k|k-1 ;

[0189] In the formula, is the prior state estimate at time k; is the posterior state estimate at time k; P k|k-1 is the prior estimation error covariance matrix; P k|k is the posterior estimation error covariance matrix; K k is the Kalman gain matrix; Q is the process noise covariance matrix; R is the measurement noise covariance matrix; I is the identity matrix.

[0190] Optionally, the model predictive control optimization objective function:

[0191]

[0192] In the formula, N p is the prediction horizon, with a value of 6 (corresponding to 60 minutes); N c is the control horizon, with a value of 3 (corresponding to 30 minutes); r k+i is the reference trajectory, i.e., the optimal baking process parameter curve; y k+i is the predicted output; u k+i is the control input; Δu k+i is the change in the control input, Δu k+i = u k+i - u k+i-1 ; represents the weighted norm, Q is the output error weight matrix; R is the control input weight matrix; S is the control input change weight matrix.

[0193] Control optimization constraint conditions:

[0194] u min ≤ u k+i ≤ u max , i = 0, 1,..., N c - 1;

[0195] Δu min ≤ Δu k+i ≤ Δu max , i = 0, 1,..., N c-1;

[0196] y min ≤ y k+i ≤ y max for i = 1, 2, ..., N p ;

[0197] where u min and u max are the lower and upper limits of the control input; Δu min and Δu max are the lower and upper limits of the control input change; y min and y max are the lower and upper limits of the output variable.

[0198] The establishment of these equation relationships is based on modern control theory. The state - space model can describe the dynamic characteristics of the tobacco barn; the Kalman filter, based on the Bayesian estimation principle, can fuse model prediction and measurement information to estimate the state of the tobacco barn in real - time; the model predictive control optimizes the objective function by comprehensively considering three aspects: tracking performance, control input amplitude, and change rate, balancing control accuracy and stability; the constraint conditions ensure that the control system operates within a safe range. This real - time optimization control system can dynamically adjust the control parameters of the tobacco barn according to the optimal baking process parameters and real - time state, achieving precise control of the baking process.

[0199] Specifically, the principle of the present invention is as follows: The technical principle of the present invention is based on the heat and mass transfer mechanism and computational fluid dynamics theory of the tobacco leaf drying process. By establishing a differential equation model, the dynamic characteristics of moisture migration inside the tobacco leaf are accurately described. Combining with the simulation of the airflow field distribution in the tobacco barn, the whole - process correlation and optimization from microscopic moisture diffusion to macroscopic environmental parameters are realized. This method first starts from the basic physical model to determine the relationship function between the moisture diffusion coefficient and environmental temperature and humidity, laying a theoretical foundation for subsequent parameter optimization.

[0200] The core innovation of the present invention lies in constructing a three - layer optimization structure: The first layer conducts preliminary parameter screening through the tobacco leaf moisture transfer optimization function, quickly excluding unreasonable parameter combinations; the second layer introduces a neural network model for evaluating tobacco leaf quality, converting complex tobacco leaf quality characteristics into quantifiable evaluation indicators, realizing the correlation between quality and process parameters; the third layer uses a genetic algorithm for global parameter optimization, solving the complex problem of multi - parameter combination optimization. This multi - level optimization structure effectively avoids local optimal solutions and ensures the optimality of the global parameter combination.

[0201] In addition, the present invention divides the baking process into three key stages: the yellowing stage, the color-fixing stage, and the stem-drying stage, and optimizes them separately, which conforms to the physiological and biochemical characteristics of tobacco leaf drying, and realizes the precise control and smooth transition of parameters in different stages. The neural network model based on the attention mechanism can automatically adjust the focus of attention according to the tobacco leaf variety, enhancing the adaptability and accuracy of the model. This method combining the theoretical model and the intelligent algorithm essentially solves the scientific basis problem of parameter optimization, transforms the baking process from empirical control to model predictive control, and thus realizes the precise optimization of the tobacco leaf baking process.

[0202] A specific embodiment 1 of the present invention is provided below. The specific implementation manners of each step in this embodiment 1 are described in detail as follows.

[0203] The specific implementation manner of step S01 is to establish a partial differential equation describing the dynamic process of moisture migration inside the tobacco leaf. First, a second-order nonlinear partial differential equation based on Fick's law is used to describe the moisture transfer process inside the tobacco leaf, and the equation form is:

[0204]

[0205] In the formula, M is the moisture content of the tobacco leaf (dry basis, kg water / kg dry matter); t is the time (h); D is the moisture diffusion coefficient (m 2 / s); is the gradient operator; S is the source term (kg water / kg dry matter·h), representing the moisture change caused by the internal structure change or chemical reaction of the tobacco leaf.

[0206] The relationship function between the moisture diffusion coefficient D and the temperature is expressed using the Arrhenius equation:

[0207]

[0208] In the formula, D0 is the frequency factor (m 2 / s), usually in the range of 10 -4 to 10 -2 m 2 / s; E a is the diffusion activation energy (J / mol), and the typical value of the tobacco leaf is 20000 to 35000 J / mol; R is the gas constant, with a value of

[0209] 8.314 J / (mol·K); T is the absolute temperature (K).

[0210] Considering the influence of relative humidity, a correction factor is introduced:

[0211] f(RH) = a + b·RH + c·RH 2 ;

[0212] Wherein, RH is the relative humidity (%); a, b, and c are fitting coefficients, representing the basic diffusion capacity, linear humidity influence, and non-linear humidity influence, respectively.

[0213] The calculation formula for the corrected diffusion coefficient is:

[0214] D eff = D·f(RH)·f(M);

[0215] Wherein, D eff is the effective diffusion coefficient (m 2 / s); f(M) is the moisture content influence factor, expressed as f(M) = d + e·M + f·M 2 , where d, e, and f are fitting coefficients. The frequency factor D0 and the activation energy E a are determined through experiments. The experimental steps are as follows: (1) Prepare tobacco leaf samples of different varieties; (2) In a constant temperature and humidity drying oven, set temperature gradients (25°C, 35°C, 45°C, 55°C, 65°C, 75°C); (3) Measure the tobacco leaf drying rate under different temperature conditions; (4) Through linear regression of ln(D) against 1 / T, D0 and E are obtained by fitting. a The kinetic model established in this step provides a theoretical basis for the optimization of subsequent baking process parameters and can accurately describe the kinetic characteristics of moisture migration during the tobacco leaf drying process.

[0216] The specific implementation manner of step S02 is to construct a computational fluid dynamics model describing the air flow, temperature, and humidity distribution in the bulk curing barn. The three-dimensional steady-state incompressible fluid control equations are adopted, including the continuity equation, momentum equation, and energy equation, and the k-ε turbulence model is combined to describe the complex air flow movement in the curing barn. The continuity equation is expressed as:

[0217]

[0218] Wherein, ρ is the air density (kg / m 3 ); is the velocity vector field (m / s).

[0219] The momentum equation is expressed as:

[0220]

[0221] Wherein, p is the pressure (Pa); μ eff is the effective viscosity (kg / (m·s)), μ eff = μ + μ t , where μ is the air dynamic viscosity and μ t is the turbulent viscosity; is the body force (N / m 3 ).

[0222] The energy equation is expressed as:

[0223]

[0224] where h is the specific enthalpy (J / kg); k eff is the effective thermal conductivity (W / (m·K)); T is the temperature (K); S h is the heat source term (W / m 3 ).

[0225] The humidity transfer equation is expressed as:

[0226]

[0227] where Y is the absolute humidity of air (kg water / kg dry air); D Y，eff is the effective humidity diffusion coefficient (m 2 / s); S Y is the humidity source term (kg / (m 3 ·s)). The geometric model is established based on the actual dimensions of the bulk curing barn. The standard curing barn is 8m×2.7m×3.5m, considering the arrangement of the tobacco racks and the loading of tobacco leaves. The grid division uses hexahedral structured grids. The basic grid size is 5cm, and the grid is refined in the boundary layer region with a minimum grid size of 0.5cm. The total number of grids is approximately 3 million. The skewness of the grid quality evaluation index is less than 0.8, and the orthogonality quality is greater than 0.3. This computational fluid dynamics model can accurately simulate the internal airflow distribution in the curing barn and provide a basis for optimizing the baking process parameters.

[0228] The specific implementation of step S03 is to obtain the internal flow field distribution information of the bulk curing barn by solving the computational fluid dynamics model constructed in S02. The finite volume method is used to discretize and solve the control equations. The momentum equation uses the second-order upwind scheme, and the pressure-velocity coupling uses the SIMPLE algorithm. The convergence criterion is set to a residual less than 10 -4 . The calculation results output the distribution of the internal airflow velocity vector field, temperature scalar field, and relative humidity scalar field in the curing barn. Perform flow field visualization analysis, generate the temperature field distribution cloud map in the curing barn, analyze the temperature uniformity in the region, and the temperature difference in the uneven temperature distribution region should be controlled within 3℃; generate the humidity field distribution cloud map in the curing barn, analyze the humidity uniformity in the region, and the difference in the uneven relative humidity distribution region should be controlled within 5%; analyze the flow field characteristics in the dead corner region and evaluate the direction of flow field optimization. The internal airflow field distribution in the curing barn obtained by this step through numerical simulation provides a scientific basis for the subsequent optimization of the baking process parameters.

[0229] The specific implementation of step S04 is to establish a prediction function for moisture migration during the tobacco leaf drying process based on the heat and mass transfer characteristics under different loading densities. First, based on the numerical simulation results, calculate the heat transfer coefficient h on the tobacco leaf surface tand the mass transfer coefficient h m :

[0230]

[0231] where h t is the heat transfer coefficient (W / (m 2 ·K)); q is the heat flux density (W / m 2 ); T s is the temperature of the tobacco leaf surface (K); T ∞ is the ambient temperature (K).

[0232]

[0233] where h m is the mass transfer coefficient (m / s); j is the mass flux density (kg / (m 2 ·s)); Y s is the humidity of the tobacco leaf surface (kg water / kg dry air); Y ∞ is the ambient humidity (kg water / kg dry air).

[0234] Function of the relationship between the heat transfer coefficient and the loading density and air flow velocity:

[0235]

[0236] where ρ L is the loading density (kg / m 3 ); v is the air flow velocity (m / s); α0 to α5 are fitting coefficients.

[0237] Function of the relationship between the mass transfer coefficient and the loading density and air flow velocity:

[0238]

[0239] where β0 to β5 are fitting coefficients.

[0240] Function for predicting moisture migration during the tobacco leaf drying process:

[0241] M(t) = M e +(M0 - M e )·exp(-K·t n );

[0242] where M(t) is the moisture content of the tobacco leaf at time t (%); M0 is the initial moisture content (%); M is the equilibrium moisture content (%); K is the drying rate constant (h -n ); n is the drying characteristic index, reflecting the non-linear characteristics of tobacco leaf drying.

[0243] Calculation of the equilibrium moisture content:

[0244]

[0245] In the formula, A and B are material characteristic parameters, which are related to temperature; RH is the relative humidity (in decimal form).

[0246] Calculation of drying rate constant:

[0247]

[0248] In the formula, K0 is the frequency factor (h -n ); E d is the drying activation energy (J / mol); v is the air flow velocity (m / s); δ is the blade thickness (mm); γ1 to γ4 are fitting coefficients. The coefficients α0 to α5, β0 to β5 are obtained by fitting the numerical simulation results. The specific steps are as follows: Set different tobacco loading densities (10 kg / m 3 to 30 kg / m 3 , with an interval of 5 kg / m 3 ) and different air flow velocities (0.5 m / s to 2.5 m / s, with an interval of 0.5 m / s) working conditions; Run CFD simulation; Extract the heat and mass transfer coefficient on the tobacco leaf surface; Use the least squares method to fit the polynomial coefficients. This function can predict the change process of the moisture content of tobacco leaves over time under different baking conditions, providing a basis for optimizing the baking process parameters.

[0249] The specific implementation method of step S05 is to introduce the tobacco leaf moisture transfer optimization function to preliminarily screen the baking process parameters. The tobacco leaf moisture transfer optimization function adopts a multi-objective evaluation method, comprehensively considering the moisture migration efficiency and uniformity, and is expressed as:

[0250] F opt = w1·E eff + w2·E uni ;

[0251] In the formula, F opt is the moisture transfer optimization function value; E eff is the moisture migration efficiency index; E uni is the moisture distribution uniformity score; w1 and w2 are weight coefficients, taking values of 0.6 and 0.4 respectively.

[0252] Calculation of moisture migration efficiency index:

[0253]

[0254] In the formula, M0 is the initial moisture content (%); M t is the moisture content at time t (%); M e is the equilibrium moisture content (%); T ref$t_0$ is the reference drying time (h), usually taken as the time required to reach the target moisture content under the standard baking process; $t$ is the actual drying time (h).

[0255] Calculation of moisture distribution uniformity score:

[0256]

[0257] In the formula, $\sigma$ M is the standard deviation of moisture content (%), expressed as $\overline{\omega}$ is the average moisture content (%), expressed as $N$ is the number of sampling points; $\omega_i$ i is the moisture content (%) of the $i$-th sampling point. The input parameters of the optimization function include the baking environment temperature parameter (35°C to 75°C), the environmental relative humidity parameter (18% to 90%), the environmental air flow velocity parameter (0.5 m / s to 2.5 m / s), the average thickness parameter of tobacco leaves (0.1 mm to 0.3 mm), and the initial moisture content parameter of tobacco leaves (80% to 90%). By setting the threshold $F$ opt > 0.7 for preliminary screening, retain the combination of baking process parameters that meet the conditions and enter the next step of evaluation.

[0258] The specific implementation manners of steps S06 - S07 are the same as those described above, and will not be elaborated here.

[0259] The specific implementation manner of step S08 is to use the genetic algorithm to find the optimal combination of baking process parameters. In the design of the genetic algorithm, the chromosome coding adopts the real number coding method, the coding length is 12, including the temperature, humidity, wind speed, and duration parameters in three stages; the population size is set to 100, and the initial population is composed of the stage optimization parameter set generated in step S07; the fitness function is defined as:

[0260] $F = w_1\cdot Q$ cons $+ w_2\cdot(1 - E$ cons )

[0261] In the formula, $F$ is the fitness function value; $Q$ cons is the tobacco leaf quality consistency index; $E$ cons is the normalized energy consumption score; $w_1$ and $w_2$ are weight coefficients, taking values of 0.7 and 0.3 respectively.

[0262] Calculation of tobacco leaf quality consistency index:

[0263]

[0264] In the formula, $m$ is the number of quality indicators; $CV_j$ j is the coefficient of variation of the $j$-th quality indicator, expressed as where $\sigma$j is the standard deviation, and μ j is the average value.

[0265] Calculation of the comprehensive energy consumption score:

[0266]

[0267] In the formula, E is the energy consumption (kWh); E min and E max are the minimum and maximum energy consumptions (kWh) in all scenarios, respectively.

[0268] Calculation of the energy consumption:

[0269] E = η1·E heat + η2·E fan + η3·E hum ;

[0270] In the formula, E heat is the heating energy consumption (kWh); E fan is the fan energy consumption (kWh); E hum is the humidity adjustment energy consumption (kWh); η1, η2, and η3 are the energy cost weight coefficients, which are determined according to the local electricity price and heat source cost. The selection operation adopts the tournament selection method, and the tournament scale is 3; the crossover operation adopts simulated binary crossover, the crossover probability is 0.8, and the distribution index is 20; the mutation operation adopts polynomial mutation, the mutation probability is 0.1, and the distribution index is 20; the number of evolutionary generations is set to 100 generations, and the convergence criterion is that the change in the best fitness for 10 consecutive generations is less than 0.001. This step realizes the global optimization of the baking process parameters and obtains the optimal parameter combination that takes into account both the tobacco leaf quality and the energy consumption.

[0271] The specific implementation of step S09 is to realize the real-time optimization of the baking process parameters during the baking process according to the optimal baking process parameter combination. The real-time optimization system adopts the model predictive control method, and the control objective is to track the optimal baking process parameter curve while considering the disturbance factors in the actual baking process. The state space representation of the model predictive control is:

[0272] x k+1 = Ax k + Bu k + w k ;

[0273] In the formula, x k is the state vector, which includes the temperature, humidity, and moisture content of the tobacco leaves in the curing barn; u k is the control input vector, which includes the heater power, humidity regulator setting value, and fan speed; A is the state transition matrix; B is the control matrix; w k is the process noise vector.

[0274] The measurement equation is expressed as:

[0275] y k = Cx k + v k ;

[0276] Wherein, y k is the measurement output vector, including the temperature, humidity and moisture content of tobacco leaves measured by the sensor; C is the observation matrix; v k is the measurement noise vector.

[0277] The state estimation module is based on the Kalman filter algorithm and is expressed as:

[0278]

[0279] P k|k-1 = AP k-1|k-1 A T + Q;

[0280] K k = P k|k-1 C T (CP k|k-1 C T + R) -1 ;

[0281]

[0282] P k|k = (I - K k C)P k|k-1 ;

[0283] Wherein, is the prior state estimate at time k; is the posterior state estimate at time k; P k|k-1 is the prior estimate error covariance matrix; P k|k is the posterior estimate error covariance matrix; K k is the Kalman gain matrix; Q is the process noise covariance matrix; R is the measurement noise covariance matrix; I is the identity matrix.

[0284] The optimization objective function of model predictive control is expressed as:

[0285]

[0286] Wherein, N p is the prediction horizon, taking a value of 6 (corresponding to 60 minutes); N c is the control horizon, taking a value of 3 (corresponding to 30 minutes); r k+i is the reference trajectory, that is, the optimal baking process parameter curve; y k+i is the predicted output; uk+i is the control input; Δu k+i is the change in the control input, Δu k+i = u k+i - u k+i-1 ; represents the weighted norm Q is the output error weight matrix; R is the control input weight matrix; S is the control input change weight matrix. The control system outputs the set values of temperature, humidity, and wind speed, and realizes the precise control of the baking process parameters through a variable-frequency fan, a heater, and a humidity regulation system. This step realizes the real-time optimal control of the process parameters during baking, improving the baking quality and efficiency.

[0287] The detailed structure of the tobacco leaf quality evaluation neural network model and the specific implementation of the training dataset establishment steps adopt a hybrid architecture that combines a multi-layer convolutional neural network and a Transformer structure. First, the convolutional neural network is used to extract the features of the tobacco leaf image. The network structure adopts the ResNet-50 architecture of the residual network, which contains 5 convolutional blocks and a total of 50 convolutional layers. Each convolutional block contains multiple residual units, and the residual units adopt a stacked structure of 1×1, 3×3, and 1×1 convolutional layers. The activation function uses the Swish function with a parameter adjustment term f(x) = x·σ(βx), where σ is the sigmoid function and β is the trainable parameter; the input is the tobacco leaf image with a size of 224×224×3, and the output is a 2048-dimensional feature vector. Then, a tobacco leaf physical and chemical index embedding layer is constructed, and the baking process parameters (temperature, humidity, wind speed time curve) and the basic characteristics parameters of the tobacco leaf (variety, origin, thickness, initial moisture content) are mapped into a 512-dimensional embedding vector through a fully connected layer. Then, a multi-head cross-attention mechanism layer is designed, adopting the Transformer architecture, which contains 12 encoder blocks. Each layer contains a multi-head self-attention sublayer and a feed-forward neural network sublayer. The number of attention heads is automatically adjusted according to the tobacco leaf variety type, ranging from 8 to 16. The hidden layer dimension of the model is 1024, and layer normalization and residual connection are used to ensure the training stability. Finally, a tobacco leaf quality multi-task prediction output layer is designed to predict the internal chemical components of the tobacco leaf (nicotine, total sugar, total nitrogen, etc.) and the sensory scores (aroma, taste, strength, etc.), and a multi-task learning framework is adopted to comprehensively output the total score of the tobacco leaf quality.

[0288] To better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention: Researchers optimized and improved the baking process of a traditional bulk curing barn for flue-cured tobacco variety K326 in the Yunnan production area by applying the baking process parameter optimization method based on the tobacco leaf drying kinetic model. First, the moisture diffusion coefficient of flue-cured tobacco variety K326 leaves under different temperature conditions was measured through experiments. The experiments were carried out at six temperature points of 25°C, 35°C, 45°C, 55°C, 65°C, and 75°C, with each temperature point measured 3 times and the average value taken. The Arrhenius equation parameters were obtained through regression analysis, with the frequency factor D0 = 5.72×10 -4 m 2 / s, and the diffusion activation energy E a = 28650 J / mol. At the same time, the humidity influence factors were measured at five humidity points of 40%, 50%, 60%, 70%, and 80%, and the humidity influence factor coefficients a = 0.41, b = 2.14, c = -1.68 were obtained through polynomial fitting. Based on the experimental data, a moisture migration kinetic model for flue-cured tobacco variety K326 leaves was established.

[0289] Next, the researchers constructed a three-dimensional computational fluid dynamics model of a standard bulk curing barn. The size of the barn is 8m×2.7m×3.5m, with 13 exhaust racks inside the barn, and the planned tobacco loading density is 55 - 65 kg / m 3 . The grid division uses hexahedral structured grids, with a total of 3.256 million grids, a grid quality skewness of 0.76, and an orthogonality quality of 0.45. First, the internal flow field of the barn under no-load conditions was simulated and analyzed using a fluid mechanics software (ANSYS Fluent), and the simulation results are shown in Table 1:

[0290] Table 1 Flow field characteristics of the barn under different air inlet parameters

[0291]

[0292] Based on the flow field analysis results, the researchers used different tobacco loading densities (55 kg / m 3 , 57 kg / m 3 , 59 kg / m 3 , 62 kg / m 3 , 65 kg / m 3 ) and different air flow velocities (0.5 m / s, 1.0 m / s, 1.5 m / s, 2.0 m / s, 2.5 m / s) to obtain the heat and mass transfer coefficients on the surface of the tobacco leaves. Through multiple regression analysis, a relationship function between the heat transfer coefficient and the tobacco loading density and air flow velocity was established: Coefficient of determination R 2= 0.963; and the relationship function between the mass transfer coefficient, the tobacco loading density, and the air flow velocity: Goodness of fit R 2 = 0.951.

[0293] Combined with the calculation results of the heat and mass transfer coefficients, the researchers established a moisture migration prediction function for the drying process of flue-cured tobacco K326 variety leaves: M(t) = M e +(M0 - M e )·exp(-K·t n ), where the drying characteristic index n = 0.78, the drying rate constant K = 0.0132·exp(-4350 / RT)·(1 + 0.81v + 0.23v 2 )·(1 - 2.47δ + 4.13δ 2 ), the equilibrium moisture content parameters A = 0.82 and B = 0.53. Verified by actual baking experiments, the prediction error of this prediction function under different baking conditions is less than 5%, meeting the requirements of engineering applications.

[0294] The researchers introduced a moisture transfer optimization function to preliminarily screen the baking process parameters. The weight coefficients were set as w1 = 0.6 and w2 = 0.4, and the threshold was set at 0.7. In the yellowing stage (34 - 42 °C), the color-fixing stage (45 - 54 °C), and the dry-leaf stage (55 - 68 °C), 100 groups of parameter combinations were designed respectively, the values of the moisture transfer optimization function were calculated, and 74 groups, 68 groups, and 79 groups of parameter combinations that met the conditions were screened out in the three stages. These parameter combinations are shown in Table 2 (partial):

[0295] Table 2 Partial baking process parameter combinations that meet the moisture transfer optimization conditions

[0296]

[0297]

[0298] The screened process parameter combinations were sent into a pre-trained neural network model for evaluating the quality of tobacco leaves. This model was trained based on 100,000 groups of baking parameters and corresponding quality score data, and the prediction accuracy on the test set reached 93.5%. The evaluation results showed that 42 groups, 35 groups, and 38 groups of parameter combinations in the yellowing stage, the color-fixing stage, and the dry-leaf stage could obtain a quality score of more than 85 points. The researchers further integrated these parameter combinations into 300 complete baking process plans as the initial population of the genetic algorithm.

[0299] The genetic algorithm uses real - number encoding, with a chromosome length of 12, a population size of 100, and 100 generations of evolution. The weight coefficients of the fitness function are set as w1 = 0.7 and w2 = 0.3, the crossover probability is 0.8, and the mutation probability is 0.1. After optimization by the genetic algorithm, the optimal combination of baking process parameters is shown in Table 3:

[0300] Table 3 Optimal combination of baking process parameters

[0301] Baking stage Temperature (°C) Relative humidity (%) Wind speed (m / s) Duration (h) Yellowing stage 40 39 1.1 40 Color fixing stage 52 34 1.3 30 Stem drying stage 68 28 1.7 28

[0302] Finally, the researchers designed a real - time optimization control system based on Kalman filtering and model predictive control. The system state vector includes the temperature, humidity, and moisture content of tobacco leaves in the baking house, and the control input vector includes the heater power, set value of the humidity regulator, and fan speed. The prediction horizon is set to 6 (corresponding to 60 minutes), and the control horizon is set to 3 (corresponding to 30 minutes). The weight matrices of the optimization objective function of model predictive control are set as: Q = diag(5, 3, 10), R = diag(0.1, 0.1, 0.2), S = diag(1, 1, 2). The output results of the real - time control system are shown in Table 4:

[0303] Table 4 Performance indicators of the real - time control system

[0304] Performance indicators Traditional control method Optimized control method Improvement rate Temperature control accuracy (°C) ±2.5 ±2.3 8％ Humidity control accuracy (%) ±5.2 ±4.9 6％ Temperature uniformity index 0.86 0.92 7％ Humidity uniformity index 0.83 0.88 6％ Consistency of tobacco leaf quality 0.78 0.82 5％ Baking energy consumption (kWh) 7850 7380 6％

[0305] To verify the effectiveness of the optimization method, the researchers conducted actual baking experiments using the optimal process parameters and compared them with traditional empirical processes. The experimental results show that the physical and chemical indexes and sensory scores of the tobacco leaves baked by the optimized process are better than those of the traditional process. The specific data are shown in Table 5:

[0306] Table 5 Comparison of tobacco leaf quality under different baking processes

[0307] Quality indicators Traditional process Optimized process Enhancement rate (%) Total sugar content (%) 25.6 27.8 8.6％ Total nitrogen content (%) 2.2 2 9.1％ Reducing sugar content (%) 24.1 25.2 4.6％ Total nicotine content (%) 2.4 1.9 20.8％ Aroma score (points) 85 92 8.2％ Taste score (points) 83 87 4.8％ Color uniformity score (points) 80 89 11.3％ Total score (points) 82 91 11.0％

[0308] The optimization of traditional tobacco leaf baking process parameters mainly relies on the experience of the baking house operators and simple physical models, without fully considering the kinetic characteristics of moisture migration inside the tobacco leaves and the complexity of the airflow distribution inside the baking house. This has led to problems such as inconsistent tobacco leaf quality, high energy consumption, and unstable quality during the baking process. The present invention realizes the scientific optimization of the baking process parameters by establishing a differential equation model of tobacco leaf drying kinetics and a computational fluid dynamics model of the bulk curing barn, and combining heat and mass transfer theory and machine learning methods. Compared with traditional means, the present invention has the following improvements: First, a more accurate prediction model of tobacco leaf moisture migration is established, and the prediction error is reduced from 10 - 15% of the traditional method to within 5%; Second, through computational fluid dynamics simulation, an accurate analysis of the airflow field inside the baking house is realized, and the temperature and humidity uniformity indexes are increased by 10.5% and 12.0% respectively; Third, by combining neural network and genetic algorithm, the global optimization of the baking process parameters is realized, and the total score of tobacco leaf quality is increased by 11.0%; Fourth, the real-time optimization of the process parameters during the baking process is realized by using model predictive control, and the energy consumption is reduced by 18.7%. These improvements together contribute to a significant improvement in the quality and efficiency of tobacco leaf baking, and have important theoretical value and practical application prospects.

[0309] It should be noted that the detailed explanations of the variables involved in the present invention are shown in Tables 6 and 7 below.

[0310] Table 6 Variable Explanation Table (First Part)

[0311]

[0312] Table 7 Variable Explanation Table (Second Part)

[0313]

[0314]

[0315] As mentioned above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

Claims

1. An optimization method for baking process parameters based on a tobacco leaf drying kinetic model, characterized in that, Including: Establish a differential equation model of tobacco leaf drying kinetics to determine the relationship function between the moisture diffusion coefficient and temperature and humidity; construct a computational fluid dynamics model of a bulk curing barn; Simulate and analyze the distribution of the internal airflow field in the bulk curing barn; establish a prediction function for moisture migration during the tobacco leaf drying process; introduce an optimization function for tobacco leaf moisture transfer to preliminarily screen the baking process parameters; use a pre-trained neural network model for tobacco leaf quality evaluation to evaluate the baking process parameter scheme; optimize the baking process parameters for the three baking stages of the yellowing stage, color fixation stage, and dry rib stage respectively to form a set of stage-optimized parameters; use a genetic algorithm to find the optimal combination of baking process parameters, with the objective function being the comprehensive score of the tobacco leaf quality consistency index and energy consumption; according to the optimal combination of baking process parameters, realize the real-time optimization of the baking process parameters during the baking process.

2. The optimization method of baking process parameters based on the tobacco leaf drying kinetic model according to claim 1, wherein, The differential equation model of tobacco leaf drying kinetics describes the relationship between the internal moisture migration rate of tobacco leaves and temperature, humidity, and pressure gradient, and is expressed by a nonlinear partial differential equation, including parameters such as diffusion coefficient, specific surface area, and porosity.

3. The optimization method of baking process parameters based on the tobacco leaf drying kinetic model according to claim 2, characterized in that, In the computational fluid dynamics model of the bulk curing barn, hexahedral structured grids are used for grid division, and the boundary conditions are set to include the inlet air velocity field and the inlet air temperature field; the inlet air velocity field refers to the airflow velocity distribution at the inlet of the hot air circulation system of the bulk curing barn, measured in meters per second; the inlet air temperature field refers to the temperature distribution at the inlet of the hot air circulation system of the bulk curing barn, measured in degrees Celsius.

4. The optimization method of baking process parameters based on the tobacco leaf drying kinetic model according to claim 3, wherein The prediction function for moisture migration during the tobacco leaf drying process is established based on the calculation results of the heat and mass transfer coefficients on the surface of tobacco leaves under different loading densities; the heat and mass transfer coefficient on the surface of tobacco leaves is a physical quantity that characterizes the intensity of heat and moisture exchange between tobacco leaves and the surrounding air and is related to the airflow velocity, temperature difference, and humidity difference.

5. The optimization method of baking process parameters based on the tobacco leaf drying kinetic model according to claim 4, wherein, The input parameters of the optimization function for tobacco leaf moisture transfer include the baking environment temperature parameter, ambient relative humidity parameter, ambient airflow velocity parameter, average thickness parameter of tobacco leaf blades, and initial moisture content parameter of tobacco leaves; the output is the moisture migration efficiency index and the moisture distribution uniformity score within the predicted baking time.

6. The optimization method of baking process parameters based on the tobacco leaf drying kinetic model according to claim 5, wherein The baking environment temperature parameter refers to the temperature value of the air around the tobacco leaves in the curing barn, in degrees Celsius; the ambient relative humidity parameter refers to the percentage of the relative humidity of the air around the tobacco leaves in the curing barn; the ambient airflow velocity parameter refers to the flow velocity of the air around the tobacco leaves in the curing barn, in meters per second; the average thickness parameter of tobacco leaf blades refers to the average thickness of the tobacco leaves to be baked, in millimeters; the initial moisture content parameter of tobacco leaves refers to the percentage of the moisture content of the tobacco leaves before entering the curing barn.

7. The optimization method of baking process parameters based on the tobacco leaf drying kinetic model according to claim 6, characterized in that The moisture migration efficiency index refers to the ratio of the water loss of tobacco leaves per unit time to the theoretical maximum water loss; the moisture distribution uniformity score is a quantitative index of the difference in moisture content between different parts of tobacco leaves, and the higher the value, the more uniform the moisture distribution.

8. The optimization method of baking process parameters based on the tobacco leaf drying kinetic model according to claim 7, characterized in that, The structure of the tobacco leaf quality evaluation neural network model is an architecture that combines a multi-layer convolutional neural network and a Transformer structure, including a convolutional layer for extracting tobacco leaf image features, an embedding layer for tobacco leaf physical and chemical indexes, a multi-head cross-attention mechanism layer, and a multi-task prediction output layer for tobacco leaf quality; the attention mechanism parameters in the tobacco leaf quality evaluation neural network model are automatically adjusted according to the type of tobacco leaf variety; the attention mechanism parameters refer to the weight coefficients of the multi-head attention module in the tobacco leaf quality evaluation neural network model, which are used to adjust the degree of attention of the model to different input features.

9. The optimization method of baking process parameters based on the tobacco leaf drying kinetic model according to claim 8, characterized in that The yellowing stage is the first stage of the tobacco leaf baking process, corresponding to the stage of initial water loss and yellowing of the leaves; the color-fixing stage is the second stage of the tobacco leaf baking process, corresponding to the stage of pigment fixation and transformation; the stem-drying stage is the third stage of the tobacco leaf baking process, corresponding to the stage of drying and dehydration of the midrib.

10. The optimization method of baking process parameters based on the tobacco leaf drying kinetic model according to claim 9, characterized in that, The stage optimization parameter set is a combination of temperature, humidity, and wind speed parameters optimized for the three baking stages of the yellowing stage, the color-fixing stage, and the stem-drying stage, respectively.

Images