Method for predicting cut tobacco quality parameters, electronic device and storage medium

Abstract

This invention provides a method, electronic device, and storage medium for predicting cigarette drying quality parameters. The method includes: acquiring real-time collected drying process data; inputting the drying process data into a target mechanism model to obtain preliminary prediction results of the drying process data and drying quality parameters; constructing drying process features based on the drying process data and the drying process data; projecting the drying process features onto the orthogonal space of the preliminary prediction results to obtain orthogonal features of the drying process features; inputting the orthogonal features into a residual prediction model to obtain prediction residuals; and obtaining the final prediction results of the drying quality parameters based on the preliminary prediction results and the prediction residuals. This invention can retain the interpretability of the drying process while improving the accuracy and generalization ability of predicting cigarette drying quality parameters.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of cigarette production and processing technology, and in particular to a method for predicting the quality parameters of tobacco drying, an electronic device, and a storage medium. Background Technology

[0002] Airflow drying is a core process in cigarette manufacturing, primarily utilizing a high-speed hot airflow to rapidly transfer heat and mass in a suspended state, achieving the required drying quality standards. With the increasing demands for product homogenization in the cigarette industry, there is a pressing need to construct a high-precision, robust, and physically interpretable airflow drying process model to ensure controllable tobacco drying quality.

[0003] However, the relationship between process parameters and drying quality parameters in the airflow drying process exhibits multivariate coupling and strong nonlinear correlation. Airflow drying process models built based on traditional mechanistic models suffer from low model parameter-data fit and insufficient prediction accuracy; while airflow drying models built based on deep learning models suffer from poor interpretability and insufficient generalization ability. Summary of the Invention

[0004] This invention provides a method, electronic device, and storage medium for predicting the quality parameters of cigarette drying, which can retain the interpretability of the drying process while improving the accuracy and generalization ability of predicting the quality parameters of cigarette drying.

[0005] In a first aspect, the method for predicting the quality parameters of the dried wire provided in the embodiments of the present invention includes: Acquire real-time collected wire drying process data; input the wire drying process data into the target mechanism model to obtain preliminary prediction results of wire drying process data and wire drying quality parameters; construct wire drying process features based on the wire drying process data and the wire drying process data; project the wire drying process features onto the orthogonal space of the preliminary prediction results to obtain orthogonal features of the wire drying process features; input the orthogonal features into the residual prediction model to obtain prediction residuals; obtain the final prediction results of wire drying quality parameters based on the preliminary prediction results and the prediction residuals.

[0006] Secondly, the device for predicting the quality parameters of the dried wire provided in the embodiments of the present invention includes: The acquisition module is used to acquire real-time collected wire drying process data; The preliminary prediction module is used to input the drying process data into the target mechanism model to obtain preliminary prediction results of the drying process data and drying quality parameters; The feature construction module is used to construct features of the drying process based on drying process data and drying process data; The feature projection module is used to project the features of the wire drying process onto the orthogonal space of the preliminary prediction results to obtain the orthogonal features of the wire drying process. The residual prediction module is used to input orthogonal features into the residual prediction model to obtain the predicted residuals; The final prediction module is used to obtain the final prediction results of the drying quality parameters based on the preliminary prediction results and the prediction residuals.

[0007] Thirdly, the electronic device provided in the embodiments of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the method for predicting the quality parameters of the drying wire as in any embodiment of the present invention.

[0008] Fourthly, the computer-readable storage medium provided in the embodiments of the present invention stores computer instructions thereon, the computer instructions being used to cause a processor to execute a method for predicting the quality parameters of the filament drying process as described in any embodiment of the present invention.

[0009] Fifthly, the computer program product provided in the embodiments of the present invention includes a computer program that, when executed by a processor, implements the method for predicting the quality parameters of the drying wire as described in any embodiment of the present invention.

[0010] In this embodiment of the invention, real-time collected drying process data is acquired; the drying process data is input into the target mechanism model to obtain preliminary prediction results of drying process data and drying quality parameters. The target mechanism model has clear physical meaning and can reflect the inherent laws of the airflow drying process, thereby ensuring the interpretability and generalization ability of the drying process; drying process features are constructed based on the drying process data and drying process data, taking into account intermediate variables reflecting the current drying environment state, which can improve the characterization ability of the drying process features, so that the subsequent residual prediction model can capture the unmodeled features of the target mechanism model; the drying process features are projected onto the orthogonal space of the preliminary prediction results to obtain the drying process... The orthogonal features of the process characteristics are input into the residual prediction model to obtain the prediction residuals. This forces the residual prediction model to learn only the orthogonal complementary components of the drying process characteristics in the output subspace of the target mechanism model, thereby achieving functional decoupling between the target mechanism model and the data-driven residual prediction model. This allows the target mechanism model to focus on modeling the common macroscopic physical laws of the drying process, while the residual prediction model is used to model the small perturbations that the mechanism model cannot explain, thus improving the generalization ability and robustness of the prediction method. Based on the preliminary prediction results and the prediction residuals, the final prediction results of the drying quality parameters are obtained, which can improve the prediction accuracy while maximizing the preservation of the physical interpretability of the prediction results. Attached Figure Description

[0011] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating a method for predicting the quality parameters of wire drying provided in an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the principle of a method for predicting the quality parameters of wire drying provided in an embodiment of the present invention; Figure 3 This is a flowchart illustrating a method for obtaining a target mechanism model provided in an embodiment of the present invention; Figure 4 This is another flowchart illustrating the method for predicting the quality parameters of the filament drying process provided in this embodiment of the invention. Figure 5 This is a flowchart illustrating a method for obtaining a residual prediction model provided in an embodiment of the present invention; Figure 6 This is an algorithm flowchart of the method for predicting the quality parameters of the filament drying process provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of a device for predicting the quality parameters of filament drying provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0013] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0014] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0015] Figure 1 This is a flowchart illustrating a method for predicting the quality parameters of tobacco drying shreds provided in an embodiment of the present invention. This method is applicable to scenarios where the quality parameters of the processed tobacco shreds are predicted based on process parameters during the airflow drying process. This method can be executed by a device for predicting the quality parameters of tobacco drying shreds provided in this embodiment of the invention, which can be implemented using software and / or hardware. In a specific embodiment, this device can be integrated into an electronic device, such as a computer or server. The following embodiment illustrates this by integrating the device for predicting the quality parameters of tobacco drying shreds into an electronic device. (See also...) Figure 1 The method for predicting the quality parameters of the dried wire in this embodiment may include the following steps: Step 101: Obtain real-time collected wire drying process data.

[0016] The tobacco drying process data is a time series of parameters describing the initial state of the tobacco shreds before they enter the gas dryer, as well as the process operating conditions. Types of drying process data may include the tobacco rehydration process flow rate, the moisture content of the tobacco shreds at the gas dryer inlet, the tobacco shred temperature at the gas dryer inlet, and the process gas temperature. The tobacco rehydration process flow rate is the mass flow rate of tobacco shreds entering the gas dryer per unit time, which can be expressed in kg / h; the tobacco shred moisture content at the gas dryer inlet is the percentage of the mass of water contained in the tobacco shreds to their dry weight; the tobacco shred temperature at the gas dryer inlet is the initial temperature of the tobacco shreds before entering the dryer; and the process gas temperature is the temperature of the hot air used to dry the tobacco shreds at the gas dryer inlet.

[0017] Specifically, during the airflow drying process, the gas dryer utilizes process gas to construct a high-speed hot airflow field, allowing the wet tobacco entering the dryer to exchange moisture and heat with the process gas in a suspended state, thereby achieving the required moisture content and temperature standards. Industrial sensors deployed on the tobacco production line can collect drying process data in real time at a fixed frequency, such as 1 Hz, and the data collected at various times can be arranged chronologically to obtain a time series of drying process data. For example, at each moment... k The data for the wire drying process can be expressed as follows: ,in They represent k The flow rate of the tobacco rehydration process, the moisture content of the tobacco at the inlet of the gas dryer, the temperature of the tobacco at the inlet of the gas dryer, and the temperature of the process gas are measured at specific times.

[0018] Step 102: Input the drying process data into the target mechanism model to obtain preliminary prediction results of the drying process data and drying quality parameters.

[0019] The quality parameters of tobacco drying are indicators that quantify the product quality of tobacco shreds obtained through airflow drying. They are used to evaluate whether the processed tobacco shreds meet the quality standards required by the process. These parameters may include the moisture content and temperature of the tobacco shreds at the dryer outlet.

[0020] The target mechanism model is a mathematical model with clear physical meaning used to simulate the changes in the state of tobacco shreds during airflow drying. The target mechanism model has a clear physical meaning, can model the general laws of thermodynamic behavior within the dryer during airflow drying, and has interpretability and generalization ability. The target mechanism model is used to learn low-frequency principal trends, which reflect the common physical laws of the tobacco drying process under different batches and operating conditions, such as the macroscopic dynamics of heat transfer and moisture evaporation.

[0021] Specifically, the target mechanism model can be a system of differential equations based on the conservation of mass and energy, capable of modeling the changes in tobacco shreds and their mass and temperature during the airflow drying process. Real-time collected drying process data can be used as initial conditions, substituted into the system of differential equations based on the conservation of mass and energy. Then, at each prediction time, the drying mass parameter value from the previous time step is substituted into the target mechanism model to obtain the rate of change of the drying mass parameter value at the current time step. The drying mass parameter value at the current time step is then solved using numerical integration. This process is repeated to obtain the drying mass parameter values ​​at multiple consecutive prediction times within the target time period.

[0022] Step 103: Construct the characteristics of the drying process based on the drying process data and drying process data.

[0023] The drying process data can be understood as intermediate variables in the target mechanism model that reflect the current drying environment and are strongly correlated with the drying quality parameters. Specifically, the drying process data can be water activity and effective diffusion coefficient. Water activity is related to the moisture content and temperature of the tobacco, reflecting the relative magnitude of water binding energy; the effective diffusion coefficient is related to the temperature of the tobacco, reflecting the relative magnitude of the resistance to moisture diffusion within the tobacco. The drying process characteristics can be obtained by splicing the drying technology data and the drying process data according to time steps.

[0024] Step 104: Project the features of the drying process onto the orthogonal space of the preliminary prediction results to obtain the orthogonal features of the drying process.

[0025] The orthogonal space of the preliminary prediction results can be understood as the orthogonal space of the linear subspace containing the preliminary prediction results output by the target mechanism model. The orthogonal features of the drying process can be understood as the vertical components obtained by removing the parallel components of the characteristic vector of the drying process from the trend direction represented by the preliminary prediction results.

[0026] For example, if the characteristics of the wire drying process are: The preliminary prediction results are The low-frequency trend represented by the preliminary prediction results can be used to predict the trend vector using a mechanism. This indicates the characteristics of the wire drying process. exist The projection in the direction is Then the orthogonal features of the wire drying process can be: .

[0027] Figure 2 This is a schematic diagram illustrating the principle of the method for predicting the quality parameters of the dried wire provided in this embodiment of the invention. (See attached diagram.) Figure 2 Where E is the original residual, which can be determined based on the measured values ​​of the wire drying quality parameters. and initial prediction results The difference is calculated. The mechanism prediction trend vector is the trend vector of the initial prediction result obtained by the target mechanism model, representing the output subspace that the target mechanism model can explain; For the original residual in The parallel component in the direction indicates the distribution trend of the drying quality parameters that have been explained by the target mechanism model, and needs to be removed when solving the residual prediction network parameters; Orthogonal residuals can be understood as the trend direction of the original residuals in the initial prediction results. The vertical component of the orthogonal space indicates high-frequency perturbations that the target mechanism model fails to model, and these need to be preserved when solving for the parameters of the residual prediction network. in, This is an orthogonal projection operator used to project E onto... The orthogonal space.

[0028] Step 105: Input the orthogonal features into the residual prediction model to obtain the predicted residuals.

[0029] The residual prediction model is a data-driven computational model used to fit the deviation between the preliminary prediction results output by the target mechanism model and the measured values ​​of the drying quality parameters. Specifically, the residual prediction model is used to fit high-frequency disturbances that the mechanism model fails to explain. These high-frequency disturbances reflect local deviations in the prediction results caused by batch differences in raw materials, environmental fluctuations, sensor noise, or other influencing factors that the target mechanism model fails to model. Projecting the drying process characteristics onto the orthogonal space of the preliminary prediction results, and then predicting the residuals based on the drying process characteristics, allows for functional decoupling between the target mechanism model and the residual prediction model. On the one hand, this avoids the overfitting of the residual prediction model masking the generalization advantage that the mechanism model should have, thus reducing the generalization ability of the drying quality parameter prediction method. On the other hand, when aging production line equipment leads to increased prediction deviations, maintenance personnel can clearly distinguish the source of error: if it is a trend deviation, only the parameters of the target mechanism model need to be corrected; if it is an increase in random fluctuations, the residual prediction model needs to be updated. This clear fault diagnosis and maintenance logic is not available in traditional coupled hybrid models, greatly reducing the complexity of operation and maintenance in industrial production sites.

[0030] For example, the specific choice of residual prediction model can be varied. It can be a traditional machine learning model, such as a random forest, or a deep learning model, such as a multilayer perceptron. It can also be a linear computation model built based on the principle of extended dynamic mode decomposition, in which case the parameters of the residual prediction model are the Koopman operator matrix.

[0031] Step 106: Based on the preliminary prediction results and prediction residuals, obtain the final prediction results of the drying quality parameters.

[0032] Specifically, since the residual prediction model only learns the orthogonal complementary components of the drying process features in the output subspace of the mechanism model, the correlation between the initial prediction result and the prediction residual will be significantly reduced. The initial prediction result and the prediction residual can be directly added or weighted to obtain the final prediction result.

[0033] Continuing with the previous example, the final prediction result can be .

[0034] In this embodiment, real-time collected drying process data is acquired; the drying process data is input into the target mechanism model to obtain preliminary prediction results of drying process data and drying quality parameters. The target mechanism model has clear physical meaning and can reflect the inherent laws of the airflow drying process, thus ensuring the interpretability and generalization ability of the drying process. Drying process features are constructed based on the drying process data and drying process data, considering intermediate variables reflecting the current drying environment state, which improves the characterization ability of the drying process features, enabling the subsequent residual prediction model to capture the unmodeled features of the target mechanism model. The drying process features are projected onto the orthogonal space of the preliminary prediction results to obtain the drying process... Orthogonal features are used to input the residual prediction model and obtain the prediction residuals. This forces the residual prediction model to learn only the orthogonal complementary components of the drying process features in the output subspace of the target mechanism model. This achieves functional decoupling between the target mechanism model and the data-driven residual prediction model, allowing the target mechanism model to focus on modeling the common macroscopic physical laws of the drying process, while the residual prediction model is used to model the small perturbations that the mechanism model cannot explain. This improves the generalization ability and robustness of the prediction method. Based on the preliminary prediction results and the prediction residuals, the final prediction results of the drying quality parameters are obtained, which can improve the prediction accuracy while maximizing the preservation of the physical interpretability of the prediction results.

[0035] The method for obtaining the target mechanism model in step 102 is described in detail below. (See also...) Figure 3 The method for obtaining the target mechanism model in this embodiment may include the following steps: Step 201: Obtain historical operating data of the cigarette manufacturing production line. After filtering the historical operating data, construct a sample set containing multiple drying time sequence samples. The drying time sequence samples include historical drying process data and measured values ​​of historical drying quality parameters at multiple consecutive times.

[0036] The data on the tobacco drying process includes the flow rate of the tobacco rehydration process, the moisture content of the tobacco at the inlet of the gas dryer, the temperature of the tobacco at the inlet of the gas dryer, and the temperature of the process gas. The historical tobacco drying quality parameters include the moisture content and temperature of the tobacco at the outlet of the gas dryer.

[0037] The tobacco drying time-series sample is a training sample composed of time-series data generated during the operation of a cigarette manufacturing production line. Specifically, the drying time-series sample consists of drying process data at multiple consecutive moments and measured values ​​of historical drying quality parameters. Some parts of the drying process data are different at different times, such as the process gas temperature, while others are relatively fixed, such as the tobacco rehydration process flow rate, the moisture content of the tobacco at the gas dryer inlet, and the tobacco temperature at the gas dryer inlet. For example, a single tobacco drying time-series sample... ,in ,in These represent time series segments representing the flow rate of the tobacco rehydration process, the moisture content of the tobacco at the inlet of the gas dryer, the temperature of the tobacco at the inlet of the gas dryer, and the temperature of the process gas, respectively. ,in These represent time series segments representing the moisture content and temperature of the tobacco shreds at the outlet of the gas dryer, respectively.

[0038] Specifically, sensors deployed on the cigarette manufacturing production line can be used to collect historical operational data of the airflow drying process at a fixed frequency. Before constructing the drying sequence sample, the collected raw historical operational data can be screened to remove invalid data generated by equipment start-ups, shutdowns, or malfunctions on the production line.

[0039] Optionally, filtering historical operational data may include at least one of the following steps: (1) Remove data from historical operation data where the flow rate of the tobacco re-moistening process is less than or equal to the preset steady-state threshold.

[0040] The preset steady-state threshold is the minimum flow rate of the tobacco re-moistening process gas required for the airflow drying process to enter a steady state. For example, historical operating data with a tobacco re-moistening process flow rate of less than 5800 kg / h can be excluded to retain data from the stable operation phase of the production line.

[0041] (2) Remove data from historical operation data where the moisture content of tobacco at the outlet is less than the preset moisture content threshold.

[0042] The preset moisture content threshold is the minimum tobacco moisture content required to achieve the airflow drying process. For example, historical operating data with a tobacco moisture content of less than 11% at the outlet can be discarded to retain data from the effective drying stage.

[0043] (3) Remove data from historical operation data where the temperature of the tobacco at the outlet is less than the preset temperature threshold.

[0044] The preset temperature threshold is the minimum tobacco temperature required to achieve the airflow drying process. For example, historical operating data with a tobacco moisture content of less than 55°C at the outlet can be discarded to retain data from the effective drying stage.

[0045] In this embodiment, historical operating data is filtered out to eliminate data from non-steady-state conditions. This ensures that the samples used to fit the parameters to be calibrated can truly reflect the essential characteristics of the airflow drying process under steady-state conditions, avoid interference from abnormal and non-steady-state data, and improve the reliability of the parameters to be calibrated.

[0046] Step 202: Construct an initial mechanism model; the initial mechanism model includes multiple parameters to be calibrated.

[0047] The parameters to be calibrated include at least one of the mass transfer coefficient and the heat transfer coefficient.

[0048] The initial mechanism model is established based on the lumped parameter method and the differential equations based on the conservation of mass and energy. Specifically, the initial mechanism model models the general physical laws of heat transfer, moisture evaporation and internal moisture diffusion between the process gas and the tobacco during the airflow drying process.

[0049] The parameters to be calibrated are those in the differential equation system with clear physical meanings, but whose specific values ​​depend on actual production conditions and tobacco characteristics. These parameters are typically assigned empirical initial values ​​when constructing the initial mechanistic model. However, due to factors such as sensor drift and equipment wear and aging in the tobacco processing production line, the parameters may deviate from their empirical initial values ​​in an unpredictable manner. Therefore, subsequent calibration is performed using historical operating data collected during actual production to adapt to specific production line conditions.

[0050] The mass transfer coefficient is a parameter characterizing the ease with which moisture migrates from the interior of tobacco to the surface and enters the process gas. The heat transfer coefficient is a parameter characterizing the rate of heat exchange per unit temperature difference and unit area between the process gas and the tobacco. Both the mass transfer coefficient and the heat transfer coefficient are closely related to the properties of tobacco.

[0051] Specifically, the entire airflow drying process can be divided into three stages. The first stage is the high-temperature drying section, where the process gas exchanges moisture and heat with the tobacco to be processed, so that the processed tobacco reaches the initial temperature requirement. The second stage is the conveyor belt section, where the tobacco leaves the drying tower and cools naturally, and moisture diffuses within it. The effective diffusion coefficient is an important variable affecting the change in the moisture content of the tobacco in this stage. The third stage is the air separation section, where the tobacco is further loosened and separated under the action of airflow. At the same time, due to the sudden pressure change, the water attached to the surface flashes rapidly, and the air cooling makes the temperature of the tobacco quickly approach the ambient temperature, thus completing the final moisture and temperature regulation.

[0052] In one feasible implementation, constructing the initial mechanistic model includes: (1) A mass conservation model is constructed based on the flow rate of the tobacco rehydration process, the moisture content of the tobacco at the inlet of the gas dryer, the moisture content of the tobacco at the outlet of the input, and the evaporation rate of the tobacco moisture.

[0053] The mass conservation model is an ordinary differential equation that models the change in moisture content of tobacco over time. For example, the mass conservation model can be... ,in, This represents the total weight of the tobacco. For the tobacco re-moistening process flow rate, This refers to the moisture content of the tobacco shreds at the inlet of the gas dryer. Let X be the evaporation rate of the tobacco shreds, and let X be the moisture content of the tobacco shreds at the outlet, i.e., the moisture content of the tobacco shreds at the outlet at the previous moment. The derivative of the output tobacco moisture content at the outlet describes the rate of change of the tobacco moisture content at the current moment.

[0054] The evaporation rate of tobacco moisture is calculated based on the mass transfer coefficient, surface area of ​​tobacco, surface water vapor pressure of tobacco, water activity, and process gas water vapor pressure; the surface water vapor pressure of tobacco is calculated based on the temperature of the tobacco at the input outlet; and the water activity is calculated based on the temperature of the tobacco at the input outlet and the moisture content of the tobacco.

[0055] The evaporation rate of tobacco shreds is the mass of water vapor lost from the surface of tobacco shreds per unit time. The surface area of ​​tobacco shreds is the effective outer surface area of ​​the tobacco shreds that participates in heat and mass transfer. The surface water vapor pressure of tobacco shreds is the effective vapor pressure obtained by multiplying the saturated water vapor pressure by the water activity at a certain tobacco temperature, assuming the surface moisture is in local equilibrium; it represents the driving force for moisture loss from the tobacco shreds surface. The process gas water vapor pressure is the actual partial pressure of water vapor present in the process gas, reflecting the process gas's ability to absorb moisture lost from the tobacco shreds.

[0056] For example, the moisture evaporation rate of tobacco shreds The calculation formula can be ,in Let A be the mass transfer coefficient and A be the surface area of ​​the tobacco. The water vapor pressure on the surface of the tobacco shreds. For water activity, This refers to the vapor pressure of the process gas, water vapor. The calculation formula can be ,in and These represent functions of the current temperature and moisture content of the tobacco at the outlet, respectively.

[0057] (2) An energy conservation model is constructed based on the total weight of tobacco, specific heat capacity of tobacco, heat transfer coefficient, surface area of ​​tobacco, temperature of process gas, temperature of tobacco at the outlet, moisture evaporation rate of tobacco and latent heat of vaporization of water vapor.

[0058] The total weight of tobacco shreds is the total weight of the dry weight of the tobacco shreds and the weight of the water contained in the tobacco shreds. The specific heat capacity of tobacco shreds is the amount of heat required to raise the temperature of a unit mass of water-containing tobacco shreds by a unit temperature. The latent heat of vaporization of water vapor is the amount of heat absorbed by a unit mass of water when it changes from a liquid state to a gaseous state.

[0059] The energy conservation model is an ordinary differential equation that models the temperature change of tobacco shreds over time, reflecting the law of energy conversion between the tobacco shreds and the process gas during the drying process. For example, the energy conservation model can be... ,in, This represents the total weight of the tobacco. The specific heat capacity of tobacco shreds, The heat transfer coefficient is... For process gas temperature, The latent heat of vaporization of water vapor is T, and the temperature of the tobacco at the outlet is the temperature of the tobacco at the previous moment. The derivative of the output tobacco temperature at the outlet describes the rate of change of the tobacco temperature at the current moment.

[0060] Step 203: Select a preset number of mechanism fitting samples from the sample set. For each mechanism fitting sample, input the historical wire drying process data into the initial mechanism model so that the initial mechanism model outputs the predicted values ​​of historical wire drying quality parameters for multiple consecutive time periods.

[0061] The mechanism fitting sample is a set of drying time-series samples used to fit the parameters to be calibrated in the initial mechanism model.

[0062] Specifically, at each target time, firstly, based on the predicted moisture content of the tobacco shreds at the outlet at the previous time and the rate of change of the moisture content of the tobacco shreds at the previous time from the historical drying process data, the predicted moisture content of the tobacco shreds at the outlet at the current time is calculated; based on the predicted temperature of the tobacco shreds at the outlet at the previous time and the rate of change of the temperature of the tobacco shreds at the previous time, the predicted temperature of the tobacco shreds at the outlet at the current time is calculated; then, the tobacco shreds rehydration process flow rate, the moisture content of the tobacco shreds at the gas dryer inlet, the moisture content of the tobacco shreds at the current time at the outlet, and the tobacco shreds moisture evaporation rate are input into the mass conservation model to obtain the rate of change of the moisture content of the tobacco shreds at the outlet at the current time; the process gas temperature, the tobacco shreds temperature at the current time at the outlet, and the tobacco shreds moisture evaporation rate at the current time are input into the energy conservation model to obtain the rate of change of the temperature of the tobacco shreds at the outlet at the current time; repeating the above process, the predicted values ​​of historical drying quality parameters for multiple consecutive time periods are obtained.

[0063] Step 204: Calculate the optimization objective function value based on the difference between the predicted and measured values ​​of the historical wire drying quality parameters of all mechanism fitting samples. Calibrate the parameters to be calibrated by minimizing the optimization objective function value to obtain the target mechanism model.

[0064] The objective function is optimized to quantify the deviation between the predicted and measured values ​​of historical wire-drying quality parameters output by the initial mechanistic model. Specifically, least squares algorithms, Bayesian optimization methods, etc., can be used to continuously update the parameters to be calibrated and calculate the objective function value until the objective function value converges or reaches the maximum number of iterations. The optimal combination of parameters to be calibrated obtained at convergence is then used as the parameter combination adopted by the target mechanistic model.

[0065] For example, if the set of parameters to be calibrated is ,in The mass transfer coefficient is . The heat transfer coefficient is given. The sample set contains N drying time series samples, and samples can be randomly selected from these N samples. A sample was used as the mechanism fitting sample, and the parameters to be calibrated in the initial mechanism model were calibrated using the least squares method. The optimization objective function can be defined as minimizing the weighted sum of squared residuals of the differences between the predicted and measured values ​​of historical wire drying quality parameters: ,in The measured value of the moisture content of the tobacco shreds at the outlet in the k-th mechanism fitting sample is given. Let be the predicted value of the moisture content of the tobacco shreds at the outlet in the k-th mechanism fitting sample. This represents the measured value of the tobacco temperature at the outlet in the k-th mechanism fitting sample. Given the predicted tobacco temperature at the outlet in the kth mechanism fitting sample, the calibrated parameters can be obtained by solving the above optimization problem. .

[0066] In this embodiment, an initial mechanism model is constructed; a preset number of mechanism fitting samples are selected from the sample set; for each mechanism fitting sample, historical drying process data is input into the initial mechanism model so that the initial mechanism model outputs predicted values ​​of historical drying quality parameters at multiple consecutive times; the objective function value is calculated based on the difference between the predicted and measured values ​​of historical drying quality parameters of all mechanism fitting samples; the parameter to be calibrated is calibrated by minimizing the objective function value to obtain the target mechanism model, which can achieve accurate matching between the mechanism model and the cigarette manufacturing production line and improve the prediction accuracy of the target mechanism model.

[0067] Figure 4 This is a schematic flowchart of a method for predicting the quality parameters of wire drying provided in an embodiment of the present invention. (See attached diagram.) Figure 4 The method for predicting the quality parameters of the dried wire in this embodiment may include the following steps: Step 301: Obtain real-time collected tobacco drying process data; the tobacco drying process data includes the tobacco re-moistening process flow rate, the tobacco moisture content at the gas dryer inlet, the tobacco temperature at the gas dryer inlet, and the process gas temperature.

[0068] Step 302: At each target time, calculate the predicted value of the moisture content of the tobacco at the outlet at the current time based on the predicted value of the moisture content of the tobacco at the outlet at the previous time and the rate of change of the moisture content of the tobacco at the previous time; calculate the predicted value of the temperature of the tobacco at the outlet at the current time based on the predicted value of the temperature of the tobacco at the outlet at the previous time and the rate of change of the temperature of the tobacco at the previous time.

[0069] The predicted initial tobacco moisture content is the tobacco moisture content at the gas dryer inlet, and the predicted initial tobacco temperature is the tobacco temperature at the gas dryer inlet.

[0070] Specifically, given the predicted values ​​of tobacco temperature and moisture content at the previous moment, and the rate of change of these values ​​at the previous moment, the predicted values ​​of tobacco temperature and moisture content at the next moment are calculated by numerical integration. By iterating in this manner, a predicted sequence of changes in the moisture content and temperature of the exported tobacco over time can be obtained.

[0071] For example, suppose the current time is k and the time step is k. The temperature of the tobacco at the current moment is The current moisture content of the tobacco is ,in, and These represent the tobacco temperature and moisture content at the previous moment, respectively. and These represent the rates of change of the tobacco temperature and moisture content at the previous moment, respectively. At the initial moment, i.e., when k=0, , , These represent the moisture content of the tobacco at the inlet of the gas dryer and the temperature of the tobacco at the inlet of the gas dryer, respectively.

[0072] Step 303: Input the tobacco rehydration process flow rate, the tobacco moisture content at the gas dryer inlet, the predicted tobacco moisture content at the outlet at the current moment, and the tobacco moisture evaporation rate into the mass conservation model to obtain the change rate of tobacco moisture content at the outlet at the current moment.

[0073] The current tobacco moisture evaporation rate is calculated based on the mass transfer coefficient, tobacco surface area, tobacco surface water vapor pressure, current water activity, and process gas water vapor pressure. The current water activity is calculated based on the predicted tobacco temperature and moisture content at the outlet. The tobacco surface water vapor pressure is calculated based on the predicted tobacco temperature at the outlet.

[0074] Specifically, the mass conservation model is constructed based on the tobacco rehydration process flow rate, the tobacco moisture content at the gas dryer inlet, the tobacco moisture content at the input outlet, and the tobacco moisture evaporation rate. The left-hand side of the equation includes the derivative of the rate of change in tobacco moisture content. At the current moment, the surface water vapor pressure of the tobacco can be calculated first based on the predicted temperature of the tobacco at the outlet. Then, the water activity at the current moment can be calculated based on the predicted temperature and moisture content of the tobacco at the outlet. The current tobacco moisture evaporation rate can be calculated based on the mass transfer coefficient, tobacco surface area, surface water vapor pressure, current water activity, and process gas water vapor pressure. The mass transfer coefficient and tobacco surface area have been calibrated and fixed during the construction phase of the target mechanism model. Finally, by substituting the tobacco rehydration process flow rate, the tobacco moisture content at the gas dryer inlet, the predicted moisture content at the outlet, and the tobacco moisture evaporation rate into the constructed mass conservation model, the rate of change in tobacco moisture content at the current moment can be calculated.

[0075] Step 304: Input the process gas temperature, the predicted value of the tobacco temperature at the outlet at the current moment, and the current tobacco moisture evaporation rate into the energy conservation model to obtain the rate of change of the tobacco temperature at the outlet at the current moment.

[0076] Specifically, the energy conservation model is constructed based on the total weight of the tobacco shreds, specific heat capacity of the tobacco shreds, heat transfer coefficient, surface area of ​​the tobacco shreds, process gas temperature, input outlet tobacco shred temperature, tobacco moisture evaporation rate, and latent heat of vaporization of water vapor. The left-hand side of the equation includes the differential of the tobacco shred temperature. The total weight of the tobacco shreds, specific heat capacity, heat transfer coefficient, surface area of ​​the tobacco shreds, and latent heat of vaporization of water vapor have already been calibrated and fixed during the construction phase of the target mechanism model. By substituting the process gas temperature, the current outlet tobacco shred temperature, and the current tobacco moisture evaporation rate into the constructed energy conservation model, the rate of change of the tobacco shred temperature at the current moment can be calculated.

[0077] Step 305: Calculate the effective diffusion coefficient at the current time based on the current tobacco temperature at the outlet.

[0078] Specifically, the effective diffusion coefficient is related to the temperature of the tobacco shreds and reflects the relative magnitude of the resistance to moisture diffusion inside the tobacco shreds.

[0079] For example, the formula for calculating the effective diffusion coefficient can be: ,in, This is the threshold value for the diffusion coefficient. Let R be the activation energy coefficient, T be the ideal gas coefficient, and T be the current temperature of the tobacco at the outlet. The diffusion coefficient threshold is the maximum value of the effective diffusion coefficient, and the activation energy coefficient is the energy barrier that needs to be overcome for moisture diffusion inside the tobacco.

[0080] Step 306: Return to steps 302 to 305 until the predicted values ​​of tobacco moisture content, tobacco temperature, water activity, and effective diffusion coefficient at the outlet for all target times are generated. Arrange the water activity and effective diffusion coefficient of all target times in chronological order to form the drying process data. Arrange the predicted values ​​of tobacco moisture content and tobacco temperature of all target times in chronological order to form the preliminary prediction results of the drying quality parameters.

[0081] Step 307: Construct the characteristics of the drying process based on the drying process data and drying process data.

[0082] Specifically, the drying process data and the yarn drying process data can be spliced ​​together according to time steps to obtain the yarn drying process characteristics.

[0083] Step 308: Based on the preliminary prediction results, use computer algorithms to predict the trend vector.

[0084] The mechanism-predicted trend vector is a directional vector that characterizes the trend represented by the preliminary prediction results.

[0085] For example, the preliminary prediction results include a time series of tobacco temperature at the outlet. Time series of tobacco moisture content at the outlet The two together form a preliminary prediction result matrix. The centered matrix can be obtained by subtracting the average value of the data in each column of the preliminary prediction matrix. And then The decomposition is performed, and the first column of the left singular matrix is ​​taken as the mechanism prediction trend vector.

[0086] Step 309: Calculate the orthogonal projection operator based on the predicted trend vector according to the mechanism.

[0087] The orthogonal projection operator is used to project the features of the drying process onto the orthogonal space of the mechanism prediction trend vector. Specifically, the orthogonal projection operator makes the orthogonal features of the drying process features orthogonal to the direction of the mechanism prediction trend, so that the subsequent residual prediction model can obtain the prediction residual by using only the orthogonal components in the drying process features that cannot be explained by the target mechanism model.

[0088] Step 310: Project the features of the drying process onto the orthogonal space of the mechanism prediction trend vector using the orthogonal projection operator to obtain the orthogonal features of the drying process.

[0089] Specifically, the features of the drying process can be forcibly projected onto the orthogonal space of the preliminary prediction results by calculating the inner product of the orthogonal projection operator and the features of the drying process.

[0090] For example, Orthogonal projection operator for predicting trend vectors based on mechanisms The calculation formula can be , where I is the identity matrix. After The transformed arbitrary vector and The linear correlation is theoretically zero (i.e., the inner product is 0). If the characteristics of the wire drying process are... Then the orthogonal features of the wire drying process can be: .

[0091] Step 311: Input the orthogonal features into the residual prediction model to obtain the prediction residuals.

[0092] Step 312: Based on the preliminary prediction results and prediction residuals, obtain the final prediction results of the drying quality parameters.

[0093] In this embodiment, at each target time, the predicted moisture content of the tobacco shreds at the outlet at the current time is calculated based on the predicted moisture content of the tobacco shreds at the outlet at the previous time and the rate of change of the moisture content of the tobacco shreds at the previous time; the predicted temperature of the tobacco shreds at the outlet at the current time is calculated based on the predicted temperature of the tobacco shreds at the outlet at the previous time and the rate of change of the temperature of the tobacco shreds at the previous time; the tobacco shreds rehydration process flow rate, the moisture content of the tobacco shreds at the gas dryer inlet, the predicted moisture content of the tobacco shreds at the outlet at the current time, and the tobacco shreds moisture evaporation rate are input into the mass conservation model to obtain the rate of change of the moisture content of the tobacco shreds at the outlet at the current time; the process gas temperature, the predicted temperature of the tobacco shreds at the outlet at the current time, and the tobacco shreds moisture evaporation rate at the current time are input into the energy conservation model to obtain the rate of change of the temperature of the tobacco shreds at the outlet at the current time; the effective diffusion coefficient is calculated based on the temperature of the tobacco shreds at the outlet at the current time; the water activity and effective diffusion coefficient of all target times are arranged in chronological order to form the number of drying processes. The method involves arranging the predicted values ​​of tobacco moisture content and temperature at all target times in chronological order to form preliminary prediction results for the tobacco drying quality parameters. This method can fully reproduce the physical mechanisms of heat exchange, moisture evaporation, and internal diffusion during the airflow drying process, giving the mechanistic model high interpretability and cross-condition generalization ability. The method constructs drying process features based on drying process data and drying process data. It then calculates the mechanistic prediction trend vector based on the preliminary prediction results, calculates the orthogonal projection operator based on the mechanistic prediction trend vector, and projects the drying process features onto the orthogonal space of the mechanistic prediction trend vector using the orthogonal projection operator. This yields orthogonal features of the drying process features, forcing the residual prediction model to learn only the information in the orthogonal space of the mechanistic model's prediction results. This avoids the data-driven residual prediction model repeatedly fitting the main trend components already captured by the target mechanistic model, reducing the risk of overfitting in the prediction method while retaining the generalization advantage of the target mechanistic model.

[0094] The method for obtaining the residual prediction model in step 311 is described in detail below. (See also...) Figure 6The method for obtaining the residual prediction model in this embodiment may include the following steps: Step 401: For each sample in the sample set, input the historical wire drying process data into the target mechanism model so that the target mechanism prediction model outputs historical wire drying process data and predicted values ​​of historical wire drying quality parameters for multiple consecutive time periods.

[0095] Specifically, for each sample in the sample set, at each target time, firstly, based on the predicted moisture content of the tobacco shreds at the outlet at the previous time and the rate of change of the moisture content of the tobacco shreds at the previous time from the historical drying process data, the predicted moisture content of the tobacco shreds at the outlet at the current time is calculated; based on the predicted temperature of the tobacco shreds at the outlet at the previous time and the rate of change of the temperature of the tobacco shreds at the previous time, the predicted temperature of the tobacco shreds at the outlet at the current time is calculated; then, the tobacco shreds rehydration process flow rate, the moisture content of the tobacco shreds at the gas dryer inlet, the predicted moisture content of the tobacco shreds at the outlet at the current time, and the tobacco shreds moisture evaporation rate are input into the mass conservation model to obtain the rate of change of the moisture content of the tobacco shreds at the outlet at the current time; the process gas temperature is then... The predicted values ​​of the tobacco temperature at the outlet and the current moisture evaporation rate of the tobacco at the current moment are input into the energy conservation model to obtain the rate of change of the tobacco temperature at the outlet at the current moment. The effective diffusion coefficient at the current moment is calculated based on the predicted value of the tobacco temperature at the outlet at the current moment. The above process is repeated until the predicted values ​​of the moisture content, temperature, water activity, and effective diffusion coefficient of the tobacco at the outlet at all target moments are generated. The water activity and effective diffusion coefficient of all target moments are arranged in time sequence to form historical drying process data. The predicted values ​​of the moisture content and temperature of the tobacco at all target moments are arranged in time sequence to form the preliminary prediction results of historical drying quality parameters.

[0096] Step 402: Construct historical drying process features based on historical drying process data and historical drying process data.

[0097] Specifically, the historical wire drying process data and historical wire drying technology data output by the target mechanism prediction model at multiple consecutive time points can be spliced ​​together according to time steps to obtain the historical wire drying process characteristics.

[0098] For example, if the wire drying process data is time series data The data for the drying process are as follows: ,in Given the water activity sequence and effective diffusion coefficient sequence at multiple consecutive time points corresponding to the wire drying process data, the characteristics of the wire drying process can be described as follows: .

[0099] Step 403: Project the historical drying process features onto the orthogonal space of the predicted historical drying quality parameters to obtain the orthogonal features of the historical drying process features.

[0100] Specifically, a mechanism prediction trend vector can be calculated based on the predicted values ​​of historical drying quality parameters at multiple consecutive times output by the target mechanism prediction model. Then, an orthogonal projection operator is calculated based on the mechanism prediction trend vector. The orthogonal projection operator can project any vector onto the orthogonal space of the mechanism prediction trend vector. Finally, by calculating the inner product of the orthogonal projection operator and the historical drying process features, the drying process features are constrained to obtain the orthogonal features of the historical drying process features.

[0101] Step 404: Calculate the residual based on the predicted and measured values ​​of historical wire drying quality parameters.

[0102] Mechanism prediction residuals are the deviations between the historical drying quality parameters predicted by the target mechanism model and the historical drying quality parameters in the drying time series samples. They represent high-frequency disturbances, unmodeled dynamics, and measurement noise that the target mechanism model failed to capture, and are the targets that the residual prediction model needs to fit.

[0103] For example, the mechanism prediction residual of drying time series sample k ,in, This represents a sequence of measured values ​​for historical wire drying quality parameters at multiple consecutive points in time. This represents the predicted historical wire drying quality parameters output by the target mechanism model at multiple consecutive time points.

[0104] Step 405: Calculate the historical prediction residuals based on the orthogonal features of the historical drying process characteristics and the initial weights of the residual prediction model.

[0105] The initial weights of the residual prediction model are the initial values ​​assigned to the learnable parameters in the residual prediction model. The historical prediction residuals are the residual prediction values ​​calculated by inputting the orthogonal features of the historical silk drying process into the unoptimized residual prediction model.

[0106] Specifically, for each historical sample, the orthogonal features of the drying process obtained in step 403 are input into the unoptimized residual prediction model to obtain the historical prediction residual for that sample. This historical prediction residual will be compared with the mechanism prediction residual in subsequent steps to calculate the loss function and update the weights of the residual prediction model.

[0107] For example, if the residual prediction model is a multilayer perceptron Its input is the orthogonal features of the historical drying process characteristics of sample k. When the output is the historical prediction residual of sample k, the output is... The initial weights are The weight matrix before training.

[0108] Step 406: Calculate the loss function based on the mechanism prediction residuals and historical prediction residuals of all drying time series samples in the sample set.

[0109] In one feasible implementation, a loss function is calculated based on the mechanism-predicted residuals and historical prediction residuals, including: (1) The deviation between the computer-predicted residual and the historical prediction residual.

[0110] Specifically, the deviation between the mechanism prediction residual and the historical prediction residual can be represented by the mean square error of the two.

[0111] (2) Calculate the correlation between the historical prediction residual and the historical mechanism prediction trend vector.

[0112] Specifically, the correlation between the historical prediction residuals and the historical mechanism prediction trend vector can be represented by their inner product. When the inner product is not zero, it indicates that there is a correlation between the trend of the historical prediction residuals output by the residual prediction model and the trend of the preliminary prediction results output by the target mechanism model. The advantage of calculating the correlation is that the update direction of the residual model weights is constrained by the trend of the preliminary prediction results output by the target mechanism model, which ensures that the prediction residuals generated by the final residual prediction model are as orthogonal as possible to the historical mechanism prediction trend vector, thus achieving functional decoupling between the target mechanism model and the residual prediction model.

[0113] (3) Calculate the loss function based on the deviation between the mechanism prediction residual and the historical prediction residual, the correlation between the historical prediction residual and the historical mechanism prediction trend vector, and the preset penalty weight.

[0114] Specifically, the loss function value can be obtained by weighted summing the deviations between the mechanistic prediction residuals and historical prediction residuals for all drying time-series samples in the sample set, as well as the correlation between the historical prediction residuals and historical mechanistic prediction trend vectors. The preset penalty weight is the weighting coefficient of the correlation between the historical prediction residuals and historical mechanistic prediction trend vectors in the loss function calculation.

[0115] Continuing with the previous example, the formula for calculating the loss function is: ,in Mechanistic prediction residuals for sample k Residuals compared to historical forecasts deviation, For sample k, the historical prediction residuals and historical mechanism prediction trend vectors are given. Relevance The preset penalty weight can be set to 0.5.

[0116] Step 407: Optimize the initial weights of the residual prediction model by minimizing the loss function to obtain the residual prediction model.

[0117] Continuing with the previous example, the loss function can be backpropagated using the gradient descent algorithm. The gradient is used to optimize the initial weights of the residual prediction model until the loss function converges, resulting in a trained residual prediction model.

[0118] In another feasible implementation, the weights of the residual prediction model can be solved based on the orthogonal characteristics of the mechanism prediction residuals and historical drying process characteristics of all drying time series samples in the sample set, thus obtaining the residual prediction model.

[0119] Specifically, when the data scale is small and the dimensionality of the features in the drying process is not high, it is not necessary to update the initial weights of the residual prediction model through iterative optimization of each sample. Instead, the residual prediction model can be modeled as a linear model, and the weights of the residual model can be solved by matrix inversion. This eliminates the need for multiple rounds of iterative optimization and results in a faster solution speed.

[0120] For example, if The orthogonal feature matrix of the historical drying features of all samples in the sample set is given, and the residual prediction model is a linear equation. ,in Let W represent the historical prediction residuals, and let W represent the parameters of the residual prediction model. In solving for W, if the mechanism prediction residual is E, the objective of the residual prediction model is to minimize the objective function. We need to minimize E and When, the solution for W is .

[0121] In this embodiment, the mechanistic prediction residual is calculated based on the predicted and measured values ​​of historical drying quality parameters; the historical prediction residual is calculated based on the orthogonal characteristics of historical drying process features and the initial weights of the residual prediction model; and the loss function is calculated based on the mechanistic prediction residuals and historical prediction residuals of all drying time-series samples in the sample set. The loss function is calculated based on the deviation between the mechanistic prediction residual and the historical prediction residual and the correlation between the historical prediction residual and the historical mechanistic prediction trend vector. This ensures that the update direction of the residual model weights is constrained by the trend of the initial prediction results output by the target mechanistic model. Finally, the prediction residual generated by the residual prediction model is decoupled from the historical mechanistic prediction trend, avoiding the residual prediction model learning the trend information already contained in the target mechanistic model and causing parameter compensation. This is more conducive to achieving functional decoupling between the target mechanistic model and the residual prediction model.

[0122] Figure 6 This is an algorithm flowchart of the method for predicting the quality parameters of the dried wire provided in an embodiment of the present invention. (See also...) Figure 6 The method for predicting the quality parameters of the dried wire provided in this embodiment includes the following six steps: Step S1: Data acquisition and preprocessing. The historical operating data of the cigarette processing production line were screened for operating conditions, retaining data where the tobacco rehydration process flow rate F is greater than or equal to 5800 kg / h and the moisture content of the tobacco at the outlet is [not specified]. More than 11% of steady-state data were used to construct the drying time series sample. ,in This indicates the data for the wire drying process. This represents the measured value of the wire drying quality parameter.

[0123] Step S2: Construct a transient mechanism model and calibrate parameters. Using the lumped parameter method, construct the energy conservation equation and mass conservation equation for the airflow drying process based on convective heat transfer and evaporation kinetics. Solve for the parameters to be calibrated in the energy conservation equation and mass conservation equation based on the mechanism fitting sample. The mechanistic prediction model after parameter calibration can be based on... Obtain mechanism prediction value During the intermediate calculations of the transient mechanism model, data on the wire drying process will be generated. ,in , Indicates water activity. This represents the effective diffusion coefficient.

[0124] Step S3: Construct the physical feature space and the original residual, based on the wire drying process data. Data on the drying process Constructing the extended feature matrix Simultaneously, based on the mechanism, the predicted value Measured values ​​of drying quality parameters in the drying time sequence sample Calculate the original residual E, which includes high-frequency disturbances not captured by the transient mechanism model, unmodeled dynamics, and measurement noise.

[0125] Step S4: Orthogonal projection and decoupled training. In order to solve the "parameter compensation" problem, that is, to prevent the residual prediction model from fitting the low-frequency trends explained by the transient mechanism model, this step S4 requires orthogonal projection constraints to be applied to the extended feature matrix.

[0126] (1) Construction of projection operator: First, the output of the mechanism model is extracted as the basis vector for constructing the orthogonal projection operator. . This represents the physical principal subspace that the transient mechanism model can explain. Then, a subspace is constructed on the sample space that can project any vector onto it. Matrix operators in orthogonal spaces .

[0127] (2) Feature space projection: Calculate the inner product of the extended feature matrix and the matrix operator to obtain the orthogonal feature matrix. This forces a projection constraint on the extended feature matrix, which removes components in the extended feature matrix that are related to the output trend of the transient mechanism model.

[0128] (3) Training the orthogonal residual prediction model: Establish the residual prediction model. The residual prediction model can be a multilayer perceptron or kernel dictionary regression, and its input is the orthogonal feature matrix. The output is the predicted residual. ,in The fitting objective is to minimize the deviation between the residual E and the output. By solving this optimization problem, the weights W of the residual prediction model are obtained.

[0129] Step S5: Hybrid Prediction and Validation: The transient mechanism model and the residual prediction model constitute a hybrid model. The final output of the hybrid model is... ,in Provide a steady-state reference. Provides dynamic compensation.

[0130] The prediction method provided in this embodiment was applied to a cigarette manufacturing production line. The process is as follows: First, historical operating data from 20 batches of a cigarette factory were collected to construct a sample set. This sample set was then divided into a training set and a test set at an 8:2 ratio. The training set was used to train the transient mechanism model and the residual prediction model. The effectiveness of the prediction method was verified on the test set. Experimental results show that after adopting this method, the Pearson correlation coefficient between the output of the residual prediction model and the output of the transient mechanism model is 0.044 (close to 0), proving that the two models are functionally strictly decoupled. The transient mechanism model is responsible for macroscopic trends, while the residual prediction model is responsible for microscopic perturbations. Simultaneously, the residuals are predicted. The significantly increased autocorrelation indicates that the prediction residual is closer to white noise, suggesting that the residual prediction model has fully extracted the effective information from the data and has not destroyed the physical structure of the transient mechanism model.

[0131] Table 1 shows the performance metrics of the prediction method provided in this embodiment on the test set: Mean Absolute Error (MAE) and Coefficient of Determination. Analysis of Table 1 shows that the mechanistic model is limited by parameter simplification and model mismatch, resulting in insufficient prediction accuracy; the two hybrid models significantly improve prediction accuracy.

[0132] Table 1 Figure 7 This is a schematic diagram of a device for predicting the quality parameters of the filament drying process provided in an embodiment of the present invention, as shown below. Figure 7 As shown, the device includes: The acquisition module 701 is used to acquire real-time collected wire drying process data; The preliminary prediction module 702 is used to input the drying process data into the target mechanism model to obtain preliminary prediction results of the drying process data and drying quality parameters. Feature construction module 703 is used to construct features of the drying process based on drying process data and drying process data; The feature projection module 704 is used to project the features of the drying process onto the orthogonal space of the preliminary prediction results to obtain the orthogonal features of the drying process. The residual prediction module 705 is used to input orthogonal features into the residual prediction model to obtain the predicted residuals. The final prediction module 706 is used to obtain the final prediction results of the drying quality parameters based on the preliminary prediction results and the prediction residuals.

[0133] In one embodiment, the device further includes a first model building module, used for: Historical operating data of the cigarette manufacturing production line was obtained. After filtering the historical operating data, a sample set containing multiple drying time series samples was constructed. The drying time series samples include historical drying process data and measured values ​​of historical drying quality parameters at multiple consecutive time points. Construct an initial mechanistic model; the initial mechanistic model includes multiple parameters to be calibrated. A preset number of mechanism fitting samples are selected from the sample set. For each mechanism fitting sample, historical wire drying process data is input into the initial mechanism model so that the initial mechanism model outputs predicted values ​​of historical wire drying quality parameters at multiple consecutive times. The objective function is calculated by the difference between the predicted and measured values ​​of the historical wire drying quality parameters of all mechanism fitting samples. The parameters to be calibrated are calibrated by minimizing the objective function to obtain the target mechanism model.

[0134] In one embodiment, the parameter to be calibrated includes at least one of mass transfer coefficient and heat transfer coefficient; the mechanism model training module constructs an initial mechanism model including: A mass conservation model is constructed based on the process flow rate of tobacco rehydration, the moisture content of tobacco at the inlet of the gas dryer, the moisture content of tobacco at the outlet of the input gas dryer, and the evaporation rate of tobacco moisture. The evaporation rate of tobacco moisture is calculated based on the mass transfer coefficient, the surface area of ​​the tobacco, the surface water vapor pressure of the tobacco, the water activity, and the water vapor pressure of the process gas. The surface water vapor pressure of the tobacco is calculated based on the temperature of the tobacco at the outlet of the input gas. The water activity is calculated based on the temperature of the tobacco at the outlet of the input gas and the moisture content of the tobacco. An energy conservation model is constructed based on the total weight of tobacco, specific heat capacity of tobacco, heat transfer coefficient, surface area of ​​tobacco, temperature of process gas, temperature of tobacco at the inlet and outlet, moisture evaporation rate of tobacco, and latent heat of vaporization of water vapor.

[0135] In one embodiment, the tobacco drying process data includes the tobacco rehydration process flow rate, the tobacco moisture content at the gas dryer inlet, the tobacco temperature at the gas dryer inlet, and the process gas temperature; the drying process data includes water activity and effective diffusion coefficient; and the drying quality parameters include the tobacco moisture content and tobacco temperature at the gas dryer outlet. The preliminary prediction module 702 is specifically used for: At each target time, the predicted value of the tobacco moisture content at the outlet at the current time is calculated based on the predicted value of the tobacco moisture content at the outlet at the previous time and the rate of change of the tobacco moisture content at the previous time; the predicted value of the tobacco temperature at the outlet at the current time is calculated based on the predicted value of the tobacco temperature at the outlet at the previous time and the rate of change of the tobacco temperature at the previous time; wherein, the predicted value of the tobacco moisture content at the initial time is the tobacco moisture content at the gas dryer inlet, and the predicted value of the tobacco temperature at the initial time is the tobacco temperature at the gas dryer inlet; The process flow rate of the tobacco rehydration process, the moisture content of the tobacco at the inlet of the gas dryer, the predicted moisture content of the tobacco at the outlet at the current moment, and the evaporation rate of the tobacco moisture are input into the mass conservation model to obtain the rate of change of the moisture content of the tobacco at the outlet at the current moment. The evaporation rate of the tobacco moisture at the current moment is calculated based on the mass transfer coefficient, the surface area of ​​the tobacco, the water vapor pressure on the surface of the tobacco, the water activity at the current moment, and the water vapor pressure of the process gas. The water activity at the current moment is calculated based on the predicted temperature and moisture content of the tobacco at the outlet at the current moment. The water vapor pressure on the surface of the tobacco is calculated based on the predicted temperature of the tobacco at the outlet at the current moment. By inputting the process gas temperature, the predicted value of the tobacco temperature at the outlet at the current moment, and the current moisture evaporation rate of the tobacco at the current moment into the energy conservation model, the rate of change of the tobacco temperature at the outlet at the current moment can be obtained. Calculate the effective diffusion coefficient at the current moment based on the predicted tobacco temperature at the outlet. Repeat the above process until the predicted values ​​of tobacco moisture content, tobacco temperature, water activity, and effective diffusion coefficient at the outlet are generated for all target times. Arrange the water activity and effective diffusion coefficient at all target times in time sequence to form the drying process data. Arrange the predicted values ​​of tobacco moisture content and tobacco temperature at all target times in time sequence to form the preliminary prediction results of the drying quality parameters.

[0136] In one embodiment, the feature projection module 704 is specifically used for: Based on preliminary prediction results, the computer will predict the trend vector. Calculate the orthogonal projection operator based on the predicted trend vector according to the mechanism; The orthogonal projection operator is used to project the characteristics of the drying process onto the orthogonal space of the mechanism prediction trend vector, thus obtaining the orthogonal characteristics of the drying process.

[0137] In one embodiment, the apparatus further includes: a second model building module, configured to: For each sample in the sample set, historical wire drying process data is input into the target mechanism model, so that the target mechanism prediction model outputs historical wire drying process data and predicted values ​​of historical wire drying quality parameters at multiple consecutive time points. Historical wire drying process characteristics were constructed based on historical wire drying process data and historical wire drying process data. By projecting the historical wire drying process characteristics onto the orthogonal space of the predicted historical wire drying quality parameters, the orthogonal characteristics of the historical wire drying process characteristics are obtained. The residual is theoretically predicted by computer based on the predicted and measured values ​​of historical wire drying quality parameters. The historical prediction residuals are calculated based on the orthogonal features of the historical drying process characteristics and the initial weights of the residual prediction model; the loss function is calculated based on the mechanism prediction residuals and historical prediction residuals of all drying time-series samples in the sample set; the initial weights of the residual prediction model are optimized by minimizing the loss function to obtain the residual prediction model; or: The weights of the residual prediction model are solved by considering the orthogonal characteristics of the mechanism prediction residuals and historical drying process features of all drying time series samples in the sample set, thus obtaining the residual prediction model.

[0138] In one embodiment, the loss function is calculated based on the mechanism-predicted residuals and historical prediction residuals, including: The deviation between the computer-generated prediction residuals and the historical prediction residuals; Calculate the correlation between historical forecast residuals and historical mechanism forecast trend vectors; The loss function is calculated based on the deviation between the mechanism prediction residual and the historical prediction residual, the correlation between the historical prediction residual and the historical mechanism prediction trend vector, and the preset penalty weight.

[0139] In one embodiment, filtering historical operational data includes at least one of the following steps: Data from historical operating data that shows the tobacco re-moistening process flow rate is less than or equal to the preset steady-state threshold value; Data from historical operation data in which the moisture content of tobacco at the outlet is less than the preset moisture content threshold is removed; Remove data from historical operation data where the temperature of the tobacco at the outlet is lower than the preset temperature threshold.

[0140] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is merely an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the functional modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0141] The apparatus of this invention acquires real-time collected drying process data; inputs the drying process data into a target mechanism model to obtain preliminary prediction results of drying process data and drying quality parameters. The target mechanism model has clear physical meaning and can reflect the inherent laws of the airflow drying process, thereby ensuring the interpretability and generalization ability of the drying process; constructs drying process features based on the drying process data and drying process data, considering intermediate variables reflecting the current drying environment state, which can improve the characterization ability of the drying process features, enabling the subsequent residual prediction model to capture the unmodeled features of the target mechanism model; projects the drying process features onto the orthogonal space of the preliminary prediction results to obtain the drying process data. Orthogonal features of process characteristics; inputting orthogonal features into the residual prediction model to obtain prediction residuals can force the residual prediction model to learn only the orthogonal complementary components of the drying process characteristics in the output subspace of the target mechanism model, realizing the functional decoupling of the target mechanism model and the data-driven residual prediction model. This allows the target mechanism model to focus on modeling the common macroscopic physical laws of the drying process, while the residual prediction model is used to model the small perturbations that the mechanism model cannot explain, improving the generalization ability and robustness of the prediction method. Based on the preliminary prediction results and prediction residuals, the final prediction results of the drying quality parameters are obtained, which can improve the prediction accuracy while maximizing the preservation of the physical interpretability of the prediction results.

[0142] The following is for reference. Figure 8 It shows a schematic diagram of the structure of a computer system 800 suitable for implementing an electronic device according to embodiments of the present invention. Figure 8 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0143] like Figure 8 As shown, the computer system 800 includes a Central Processing Unit (CPU) 801, which can perform various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) 802 or programs loaded from storage section 808 into Random Access Memory (RAM) 803. The RAM 803 also stores various programs and data required for the operation of the computer system 800. The CPU 801, ROM 802, and RAM 803 are interconnected via a bus 804. An Input / Output (I / O) interface 805 is also connected to the bus 804.

[0144] The following components are connected to I / O interface 805: an input section 806 including a keyboard, mouse, etc.; an output section 807 including a cathode ray tube, liquid crystal display, etc., and speakers, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card, such as a modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to I / O interface 805 as needed. A removable medium 811, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 810 as needed so that computer programs read from it can be installed into storage section 808 as needed.

[0145] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 809, and / or installed from removable medium 811. When the computer program is executed by central processing unit (CPU) 801, it performs the functions defined above in the system of this invention.

[0146] It should be noted that the computer-readable medium shown in this invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical fiber, a portable compact disk read-only memory, an optical storage device, a magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, etc., or any suitable combination thereof.

[0147] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0148] The modules and / or units described in the embodiments of the present invention can be implemented in software or hardware. The described modules and / or units can also be housed in a processor; for example, a processor can be described as including an acquisition module, a preliminary prediction module, a feature construction module, a residual prediction module, and a final prediction module. The names of these modules do not necessarily constitute a limitation on the module itself.

[0149] In another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs, which, when executed by the device, cause the device to include: Acquire real-time data on the drying process; input the drying process data into the target mechanism model to obtain preliminary prediction results of the drying process data and drying quality parameters; construct drying process features based on the drying process data and drying process data; project the drying process features onto the orthogonal space of the preliminary prediction results to obtain orthogonal features of the drying process features; input the orthogonal features into the residual prediction model to obtain the prediction residuals; obtain the final prediction results of the drying quality parameters based on the preliminary prediction results and prediction residuals.

[0150] The technical solution of this invention involves acquiring real-time collected drying process data; inputting the drying process data into a target mechanism model to obtain preliminary prediction results of drying process data and drying quality parameters. The target mechanism model has clear physical meaning and can reflect the inherent laws of the airflow drying process, thereby ensuring the interpretability and generalization ability of the drying process; constructing drying process features based on the drying process data and drying process data, considering intermediate variables reflecting the current drying environment state, which can improve the characterization ability of the drying process features, enabling the subsequent residual prediction model to capture the unmodeled features of the target mechanism model; projecting the drying process features onto the orthogonal space of the preliminary prediction results to obtain the drying process quality parameters. The orthogonal features of the drying process characteristics are input into the residual prediction model to obtain the prediction residuals. This forces the residual prediction model to learn only the orthogonal complementary components of the drying process characteristics in the output subspace of the target mechanism model, thereby achieving functional decoupling between the target mechanism model and the data-driven residual prediction model. This allows the target mechanism model to focus on modeling the common macroscopic physical laws of the drying process, while the residual prediction model is used to model the small perturbations that the mechanism model cannot explain, thus improving the generalization ability and robustness of the prediction method. Based on the preliminary prediction results and the prediction residuals, the final prediction results of the drying quality parameters are obtained, which can improve the prediction accuracy while maximizing the preservation of the physical interpretability of the prediction results.

[0151] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements a method for predicting the quality parameters of the dried wire as provided in any embodiment of this invention.

[0152] In the implementation of a computer program product, computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​as well as conventional procedural programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including local area networks (LANs) or wide area networks (WANs), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0153] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0154] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for predicting the quality parameters of dried wire, characterized in that, include: Acquire real-time data on the wire drying process; The drying process data is input into the target mechanism model to obtain preliminary prediction results of the drying process data and drying quality parameters; Based on the drying process data and the drying process data, the characteristics of the drying process are constructed; The features of the drying process are projected onto the orthogonal space of the preliminary prediction results to obtain the orthogonal features of the drying process. The orthogonal features are input into the residual prediction model to obtain the predicted residuals; The final prediction results of the wire drying quality parameters are obtained based on the preliminary prediction results and the prediction residuals.

2. The method according to claim 1, characterized in that, The target mechanism model was obtained through the following method: Historical operating data of the cigarette manufacturing production line is obtained. After filtering the historical operating data, a sample set containing multiple drying time series samples is constructed. The drying time series samples include historical drying process data and measured values ​​of historical drying quality parameters at multiple consecutive times. An initial mechanistic model is constructed; the initial mechanistic model includes multiple parameters to be calibrated. A preset number of mechanism fitting samples are selected from the sample set. For each mechanism fitting sample, the historical wire drying process data is input into the initial mechanism model so that the initial mechanism model outputs predicted values ​​of historical wire drying quality parameters for multiple consecutive time periods. The objective function value is calculated based on the difference between the predicted and measured values ​​of the historical wire drying quality parameters of all mechanism fitting samples. The parameter to be calibrated is then calibrated by minimizing the objective function value to obtain the target mechanism model.

3. The method according to claim 2, characterized in that, The parameters to be calibrated include at least one of mass transfer coefficient and heat transfer coefficient, and the construction of the initial mechanism model includes: A mass conservation model is constructed based on the tobacco rehydration process flow rate, the tobacco moisture content at the gas dryer inlet, the tobacco moisture content at the input outlet, and the tobacco moisture evaporation rate. The tobacco moisture evaporation rate is calculated based on the mass transfer coefficient, tobacco surface area, tobacco surface water vapor pressure, water activity, and process gas water vapor pressure. The tobacco surface water vapor pressure is calculated based on the tobacco temperature at the input outlet. The water activity is calculated based on the tobacco temperature at the input outlet and the tobacco moisture content. An energy conservation model is constructed based on the total weight of the tobacco, the specific heat capacity of the tobacco, the heat transfer coefficient, the surface area of ​​the tobacco, the temperature of the process gas, the temperature of the tobacco at the outlet, the moisture evaporation rate of the tobacco, and the latent heat of vaporization of water vapor.

4. The method according to claim 3, characterized in that, The tobacco drying process data includes the tobacco rehydration process flow rate, tobacco moisture content at the gas dryer inlet, tobacco temperature at the gas dryer inlet, and process gas temperature. The drying process data includes water activity and effective diffusion coefficient. The drying quality parameters include tobacco moisture content and tobacco temperature at the gas dryer outlet. Inputting the drying process data into the target mechanism model yields preliminary prediction results for the drying process data and drying quality parameters, including: At each target time, the predicted value of the tobacco moisture content at the outlet at the current time is calculated based on the predicted value of the tobacco moisture content at the outlet at the previous time and the rate of change of the tobacco moisture content at the previous time; the predicted value of the tobacco temperature at the outlet at the current time is calculated based on the predicted value of the tobacco temperature at the outlet at the previous time and the rate of change of the tobacco temperature at the previous time; wherein, the predicted value of the tobacco moisture content at the initial time is the tobacco moisture content at the gas dryer inlet, and the predicted value of the tobacco temperature at the initial time is the tobacco temperature at the gas dryer inlet; The process flow rate of the tobacco rehydration process, the moisture content of the tobacco at the inlet of the gas dryer, the predicted moisture content of the tobacco at the outlet at the current moment, and the tobacco moisture evaporation rate are input into the mass conservation model to obtain the rate of change of the tobacco moisture content at the outlet at the current moment. The tobacco moisture evaporation rate at the current moment is calculated based on the mass transfer coefficient, the surface area of ​​the tobacco, the water vapor pressure on the surface of the tobacco, the water activity at the current moment, and the water vapor pressure of the process gas. The water activity at the current moment is calculated based on the predicted temperature and moisture content of the tobacco at the outlet at the current moment. The water vapor pressure on the surface of the tobacco is calculated based on the predicted temperature of the tobacco at the outlet at the current moment. The process gas temperature, the predicted temperature of the tobacco shreds at the outlet at the current moment, and the moisture evaporation rate of the tobacco shreds at the current moment are input into the energy conservation model to obtain the rate of change of the temperature of the tobacco shreds at the outlet at the current moment. The effective diffusion coefficient at the current moment is calculated based on the predicted tobacco temperature at the outlet. Repeat the above process until the predicted values ​​of tobacco moisture content, tobacco temperature, water activity, and effective diffusion coefficient at the outlet are generated for all target times. Arrange the water activity and effective diffusion coefficient at all target times in chronological order to form the drying process data. Arrange the predicted values ​​of tobacco moisture content and tobacco temperature at all target times in chronological order to form the preliminary prediction results of the drying quality parameters.

5. The method according to claim 2, characterized in that, Projecting the features of the wire drying process onto the orthogonal space of the preliminary prediction results yields orthogonal features of the wire drying process features, including: Based on the preliminary prediction results, a trend vector is theoretically predicted by computer. Calculate the orthogonal projection operator based on the predicted trend vector according to the aforementioned mechanism; The orthogonal projection operator projects the features of the drying process onto the orthogonal space of the mechanism prediction trend vector to obtain the orthogonal features of the drying process.

6. The method according to claim 5, characterized in that, The residual prediction model is obtained through the following method: For each sample in the sample set, the historical wire drying process data is input into the target mechanism model, so that the target mechanism prediction model outputs historical wire drying process data and predicted values ​​of historical wire drying quality parameters for multiple consecutive time periods. Based on the historical wire drying process data and the historical wire drying process data, a historical wire drying process feature is constructed; The historical drying process features are projected onto the orthogonal space of the predicted historical drying quality parameters to obtain the orthogonal features of the historical drying process features; The residual is theoretically predicted by computer based on the predicted and measured values ​​of historical wire drying quality parameters. The historical prediction residuals are calculated based on the orthogonal features of the historical drying process characteristics and the initial weights of the residual prediction model; a loss function is calculated based on the mechanism prediction residuals of all drying time-series samples in the sample set and the historical prediction residuals; the initial weights of the residual prediction model are optimized by minimizing the loss function to obtain the residual prediction model; or: The weights of the residual prediction model are calculated by solving the orthogonal features of the mechanism prediction residuals and the historical drying process features of all drying time series samples in the sample set, and the residual prediction model is obtained.

7. The method according to claim 6, wherein a loss function is calculated based on the predicted residuals according to the mechanism and the historical predicted residuals, comprising: The deviation between the computer-generated prediction residuals and the historical prediction residuals; Calculate the correlation between historical forecast residuals and historical mechanism forecast trend vectors; The loss function is calculated based on the deviation between the predicted residual and the historical predicted residual, the correlation between the historical predicted residual and the historical mechanism predicted trend vector, and the preset penalty weight.

8. The method according to claim 4, characterized in that, Filtering the historical operational data includes at least one of the following steps: Data in the historical operating data that shows the tobacco re-moistening process flow rate is less than or equal to a preset steady-state threshold value will be removed. Data from the historical operating data in which the moisture content of the tobacco at the outlet is less than a preset moisture content threshold is removed; Data from the historical operating data that shows the tobacco temperature at the outlet is lower than a preset temperature threshold will be removed.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for predicting the quality parameters of the filament drying process as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method for predicting the drying quality parameters as described in any one of claims 1 to 8.

Images