Lyocell fiber spinning pulp porridge process simulation and quality prediction method and system

Abstract

The invention relates to the technical field of fiber production process simulation modeling, in particular to a Lyocell fiber spinning pulp porridge process simulation and quality prediction method and system.The method comprises the following steps that a simulation system is used for simulating the swelling process of cellulose in an NMMO solvent, and multiple sets of input and output data pairs are generated to construct a data set; a physical model of a simulation system is introduced into the gradient boosting tree algorithm to serve as regularization constraint, and an improved gradient boosting tree algorithm is obtained; and training an improved gradient boosting tree algorithm by using the data set to predict the quality of the Lyocell fiber spinning pulp porridge. According to the method, a slurry porridge quality prediction scheme based on simulation data and the improved gradient boosting tree algorithm is put forward for the first time, the slurry porridge research and development time and cost are greatly reduced, the improved gradient boosting tree algorithm integrates a physical model of a simulation system into a model training process by adding a physical constraint regularization item into a loss function, and the development efficiency of the slurry porridge is improved. The scientificity of a prediction result is enhanced, and the reliability of industrial application of the model is improved.

Description

Technical Field

[0001] This invention relates to the field of fiber production process simulation modeling technology, and in particular to a method and system for simulating and predicting the quality of lyocell fiber spinning sizing process. Background Technology

[0002] Lyocell fiber is recognized worldwide as a "green fiber" of the 21st century. Its production process uses pulp as raw material and N-methylmorpholine-N-oxide (NMMO) as a solvent, prepared through wet spinning technology. The core of the Lyocell fiber production process lies in the swelling process of cellulose in the NMMO solution, where swelling, as the initial stage of the entire process, plays a crucial role. Cellulose swelling refers to the process where, under the action of the NMMO solvent, solvent molecules penetrate into the spaces between cellulose molecular chains, disrupting their tight hydrogen bond network. This causes the cellulose structure to loosen and expand, forming a spinning pulp, laying the foundation for subsequent complete dissolution.

[0003] Currently, the production process parameters for spinning pulp mainly rely on laboratory environments and small-scale pilot production lines. Necessary parameters are obtained through a series of experiments, and these parameters are then applied to large-scale actual production. This method can accurately simulate production conditions on a small scale, thus obtaining relatively accurate and stable process parameters. However, the experimental process in laboratories and pilot production lines involves multiple stages and steps, requiring considerable time. In emergency situations requiring rapid adjustments to process parameters, the lengthy experimental cycle may delay the progress of actual production. Furthermore, the construction and operating costs of laboratories and pilot production lines are high due to the consumption of raw materials, the purchase of equipment, and the training of personnel.

[0004] At the same time, no systematic method for predicting the quality of spinning pulp has been proposed yet, and the quality of pulp is affected by a variety of factors, making quality prediction somewhat complex.

[0005] Therefore, in order to improve production efficiency, save production costs, and help production personnel fully understand the operational skills and processes of spinning sizing, it is urgent to digitally present the spinning sizing production process and establish a simulation model of the spinning sizing production process to achieve visualization and effective control of spinning sizing production. At the same time, it is also urgent to explore methods for predicting the quality of spinning sizing to improve the efficiency of sizing quality control, provide real-time guidance for actual production, and further promote the intelligent and efficient development of lyocell fiber production processes. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for simulating and predicting the quality of lyocell fiber spinning sizing process.

[0007] To achieve the above objectives, in a first aspect, this invention provides a simulation and quality prediction method for lyocell fiber spinning sizing process. The method includes the following steps: constructing a sizing production process simulation system based on the swelling mechanism of cellulose in NMMO solution; generating simulation data for lyocell fiber spinning sizing; and obtaining multiple sets of input-output data pairs to construct a dataset; introducing the physical model of the sizing production process simulation system as a regularization constraint into the gradient boosting tree algorithm to obtain an improved gradient boosting tree algorithm; and using the dataset to complete the training and verification of the improved gradient boosting tree algorithm to obtain a quality prediction model for predicting key indicators of lyocell fiber spinning sizing quality. This invention proposes for the first time a sizing quality prediction scheme based on simulation data and an improved gradient boosting tree algorithm, significantly reducing sizing development time and cost. Furthermore, the improved gradient boosting tree algorithm, by adding a physical constraint regularization term to the loss function, integrates the physical model of the simulation system into the model training process, enhancing the scientific validity of the prediction results and improving the reliability of the model for industrial application.

[0008] Optionally, the physical model of the spinning pulp production process simulation system includes a solvent penetration model, a swelling kinetics model, a stirring and uniformity model, and a viscosity calculation model.

[0009] Optionally, the solvent permeation model satisfies the following relationship:

[0010] in, denoted as the solvent concentration of the NMMO solution, t as time, and D as the diffusion coefficient related to the degree of polymerization of cellulose. This is the spatial second derivative of the solvent concentration gradient.

[0011] Optionally, the swelling kinetics model satisfies the following relationship:

[0012] in, The degree of swelling of cellulose in NMMO solvent. The swelling rate constant is The solvent concentration is... To balance the solvent concentration, This represents the maximum degree of swelling.

[0013] Optionally, the stirring action and the uniformity model satisfy the following relationship:

[0014]

[0015] in, The solvent concentration is... The velocity of the flow field caused by stirring characterizes the fluid motion induced by stirring. Let be the spatial first derivative of the solvent concentration gradient, and U be the uniformity of the sizing paste used in Lyocell fiber spinning. V represents the average solvent concentration in the NMMO solution, and V is the volume of the mixture of cellulose and NMMO solution.

[0016] Optionally, the viscosity calculation model satisfies the following relationship:

[0017] in, The viscosity of the slurry for spinning lyocell fibers. Where is the viscosity constant. For cellulose content, The degree of swelling, The solvent concentration is given, and m, n, and p are the weights of the effects of the cellulose content, the degree of swelling, and the solvent concentration on the viscosity of the porridge, respectively. This is the viscosity activation energy.

[0018] Optionally, the input-output data pair includes input data for inputting the quality prediction model and output data corresponding to the input data. The input data includes the degree of polymerization of cellulose, the solvent concentration, the cellulose content, the temperature, and the stirring rate. The output data includes the uniformity and the viscosity of the porridge.

[0019] Optionally, the physical model of the spinning slurry production process simulation system is introduced into the gradient boosting tree algorithm as a regularization constraint, including a uniformity constraint term and a viscosity constraint term. The uniformity constraint term and the viscosity constraint term satisfy the following relationships respectively:

[0020]

[0021] in, For the theoretical uniformity of the j-th sample, The rate at which stirring intensity affects the uniformity of the porridge. Let be the numerical value of the flow field velocity for the j-th sample. This is the proportionality coefficient. Let j be the stirring rate of the j-th sample. Let the theoretical viscosity be the j-th sample. Let m be the viscosity constant, and m, n, and p be the weights of the effects of the cellulose content, the degree of swelling, and the solvent concentration on the viscosity of the porridge, respectively. Let be the cellulose content of the j-th sample. The degree of swelling of the j-th sample, The solvent concentration for the j-th sample is... Where is the viscosity activation energy, and R is the gas constant. Let be the temperature of the j-th sample.

[0022] Optionally, the loss function of the improved gradient boosting tree algorithm satisfies the following relationship:

[0023]

[0024] in, For the total loss, For data-driven items, For physical loss items, This is the viscosity loss term. For uniformity loss term, Here, N represents the regularization weights, and N is the total number of samples. Let be the algorithm's predicted value for the j-th sample. Let be the true value of the j-th sample.

[0025] Secondly, the present invention provides a simulation and quality prediction system for lyocell fiber spinning sizing process. This system includes: a data acquisition device, a data output device, a processor, and a storage device. The storage device includes a computer-readable storage medium storing a computer program. The computer program includes program instructions, which, when executed by the processor, cause the processor to implement the lyocell fiber spinning sizing process simulation and quality prediction method provided by the present invention.

[0026] In summary, the present invention has at least the following beneficial effects: 1. This method proposes for the first time a systematic prediction scheme for the quality of lyocell fiber spinning pulp based on simulation data and an improved gradient boosting tree algorithm, filling a technological gap and providing a complete process from data generation to prediction of lyocell fiber spinning pulp quality, thereby improving the systematicness and scientific nature of pulp quality prediction.

[0027] 2. This method generates a large amount of data through a simulation system, which can obtain enough data to train the prediction model without physical experiments, greatly reducing R&D time and economic costs, and supporting rapid process parameter optimization, especially suitable for emergency adjustment scenarios.

[0028] 3. The improved gradient boosting tree algorithm in this method incorporates a physical constraint regularization term into the loss function, integrating the physical model of the spinning pulp production process simulation system into the model training process. This ensures that the predictions conform to the physical characteristics of cellulose swelling, viscosity, and uniformity, enhancing the scientific rationality of the prediction results, avoiding the physical biases that may occur in traditional data-driven models, and improving the reliability of the model for industrial applications.

[0029] 4. The present invention provides a system adapted to the method, which can improve the practicality of the method. Attached Figure Description

[0030] To more clearly illustrate the technical solutions of the embodiments of this application, 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 this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a flowchart illustrating a method for simulating and predicting the quality of lyocell fiber spinning sizing process according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the framework of a simulation and quality prediction system for lyocell fiber spinning sizing process according to an embodiment of the present invention. Detailed Implementation

[0032] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0033] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0034] It should be noted in advance that, in one alternative embodiment, except for independent descriptions, the same symbols or letters appearing in all formulas have the same meaning and value.

[0035] In one optional embodiment, please refer to Figure 1 This invention provides a method for simulating and predicting the quality of lyocell fiber spinning sizing process, the method comprising the following steps: S1. Based on the swelling mechanism of cellulose in NMMO solution, a simulation system for the production process of spinning pulp is constructed to generate simulation data of lyocell fiber spinning pulp, and then multiple sets of input-output data pairs are obtained to construct a dataset.

[0036] Specifically, in this embodiment, the physical model of the spinning pulp production process simulation system includes a solvent penetration model, a swelling kinetics model, a stirring and uniformity model, and a viscosity calculation model. Furthermore, it should be noted that the cellulose mentioned below refers to lyocell cellulose.

[0037] The solvent permeation model describes the dynamic process by which NMMO solvent molecules diffuse into the porous structure of cellulose. This is the initial step of swelling, and the permeation of solvent molecules provides the driving force for subsequent hydrogen bond breaking and cellulose expansion. The role of the solvent permeation model is to simulate how the solvent permeates into the cellulose microstructure over time, generating a concentration distribution, and providing input for the swelling kinetics model. The output of the solvent permeation model is the distribution of solvent concentration over time and space; the remaining parameters are the model inputs. The solvent permeation model satisfies the following relationship:

[0038] in, is the solvent concentration of the NMMO solution, representing the mass of solvent molecules per unit volume of NMMO solution, which varies with time and spatial location, and its value is usually in the range of 400~700 kg / m³; t is time, representing the evolution of the solvent molecule diffusion process; D is the diffusion coefficient related to the degree of polymerization of cellulose, characterizing the ability of solvent molecules to penetrate cellulose, which is affected by temperature and cellulose structure. Let be the spatial second derivative of the solvent concentration gradient, representing the diffusion direction and intensity of solvent molecules.

[0039] The diffusion coefficient D satisfies the following relationship:

[0040] in, The pre-exponential factor for the diffusion process represents the diffusion capacity under ideal conditions, and its value ranges from 10. -9 ~10 -8 . ΔD is the diffusion activation energy, which characterizes the energy required to overcome the structural resistance of cellulose. It is determined through diffusion experiments at different temperatures and ranges from 20 to 40 kJ / mol. In subsequent simulations, the initial value of the diffusion activation energy can be set to 30 kJ / mol. R is the gas constant; T is the temperature, ranging from 353 to 393 K. Increasing T will accelerate molecular motion, thus increasing D; DP is the degree of polymerization of cellulose, ranging from 400 to 1200. High DP will increase the molecular chain length and hydrogen bond density, thus decreasing D. This is an empirical index characterizing the degree to which DP inhibits solvent molecule diffusion; its value ranges from 0.5 to 1. In subsequent simulations, The initial value can be set to 0.7.

[0041] The swelling kinetics model describes the change in the degree of cellulose swelling over time, characterizing the relaxation and expansion process of cellulose structure. It can simulate the dynamic behavior of increased interchain spacing and hydrogen bond breakage in cellulose molecules after solvent molecule infiltration. The function of the swelling kinetics model is to calculate the degree of cellulose swelling based on solvent concentration, reflecting the pulp formation process. The output of the swelling kinetics model is the change in the degree of swelling over time, and the remaining parameters are the model inputs. The swelling kinetics model satisfies the following relationship:

[0042] in, The degree of swelling of cellulose in NMMO solvent is defined as the ratio of the swollen volume to the initial volume of cellulose. Indicates the unswollen state; The swelling rate constant represents the expansion rate and is affected by temperature. To balance the solvent concentration, characterize the solvent concentration at which swelling saturation is achieved, and correlate it with the cellulose content. Positive correlation, with a reference range of 50~200 kg / m³. In subsequent simulations... The initial value can be set to 100 kg / m³; The maximum swelling degree represents the limit of cellulose swelling, subject to... And DP limitations. The value range is generally 50~150 kg / m³.

[0043] The swelling rate constant and the maximum degree of swelling satisfy the following relationships:

[0044]

[0045] in, is the pre-exponential factor for swelling kinetics, representing the swelling rate under ideal conditions, which will be used in subsequent simulations. The initial value can be set to . If the cellulose swells too slowly, it will increase... Conversely, reduce ; The swelling activation energy is the energy barrier characterizing hydrogen bond breaking. It is determined through swelling experiments at different temperatures, and its value ranges from 20 to 50 kJ / mol. In subsequent simulations, the initial value of the swelling activation energy can be set to 40 kJ / mol. The constant represents the material properties. The value range is generally 5~20 kg / m³. In subsequent simulation processes, The initial value can be set to 8 kg / m³; This is an empirical index representing the suppression effect on DP, with a value ranging from 0.5 to 1. In subsequent simulations... The initial value can be set to 0.7.

[0046] The stirring and homogeneity model comprises a stirring model and a homogeneity model, describing how stirring enhances the mixing of solvent and cellulose through convection, optimizes concentration distribution, and affects slurry homogeneity. The stirring rate modulates solvent distribution through flow field velocity, ultimately influencing slurry quality. The role of the stirring and homogeneity model is to combine diffusion and convection to update the solvent concentration distribution and calculate the homogeneity of lyocell fiber spinning slurry. The output of the stirring and homogeneity model is the updated solvent concentration and homogeneity; the remaining parameters are the model inputs. The stirring model and the homogeneity model satisfy the following relationships:

[0047]

[0048] in, The velocity of the flow field caused by stirring characterizes the fluid motion induced by stirring. The first spatial derivative of the solvent concentration gradient. The convection term represents the spatial transport effect of stirring on concentration; U represents the uniformity of lyocell fiber spinning slurry. The larger the value, the more uniform the distribution of lyocell fiber spinning slurry, which is one of the key indicators of lyocell fiber spinning slurry quality. The average solvent concentration in the NMMO solution is obtained by calculating the spatial average of the solvent concentration; V is the volume of the mixture of cellulose and NMMO solution.

[0049] The flow field velocity satisfies the following relationship:

[0050] in, The proportionality coefficient characterizes the efficiency of the mixing equipment, and its value is 10. −4 m / rpm, determined experimentally or by equipment parameters; This refers to the stirring rate, which typically ranges from 100 to 1000 rpm. Increasing this value... This indicates enhanced convection.

[0051] The viscosity calculation model describes the rheological properties of lyocell fiber spinning slurry, which is influenced by a combination of factors including cellulose content, swelling degree, solvent concentration, and temperature. Its function is to quantify the rheological properties of the slurry and calculate its viscosity. Slurry viscosity describes the hydrodynamic behavior of the slurry and directly affects the feasibility of the spinning process and product quality; therefore, it can serve as a key indicator for quality prediction. The output of the viscosity calculation model is the slurry viscosity, and the other parameters are inputs. The viscosity calculation model satisfies the following relationships:

[0052] in, The viscosity of the slurry for spinning lyocell fibers; is the viscosity constant, characterizing the reference viscosity, and its value is 10. −5 The values ​​were obtained through experimental fitting; m, n, and p were the weights of the effects of cellulose content, swelling degree, and solvent concentration on the viscosity of the porridge, respectively. In this embodiment, the values ​​of m, n, and p were 1, 0.5, and -0.3, respectively. The viscosity activation energy characterizes the temperature effect and is typically in the range of 20–50 kJ / mol. In subsequent simulations, the initial value of the viscosity activation energy can be set to 30 kJ / mol.

[0053] More specifically, the simulation system for the spinning pulp production process was built using Ansys Fluent simulation software. In Ansys Fluent: the solvent penetration model provides the diffusion basis, driving swelling and concentration distribution; the swelling kinetics model connects microstructural changes with macroscopic mass, outputting the degree of swelling; the stirring and uniformity model optimizes solvent distribution, outputting uniformity; and the viscosity calculation model integrates the results of the preceding models, outputting the pulp viscosity.

[0054] Furthermore, since the value ranges and initial values ​​of some parameters in the physical models of the lyocell fiber spinning slurry production simulation system have already been given, these parameters can be adjusted during the simulation process to generate simulation data for the lyocell fiber spinning slurry, thereby generating 10,000 sets of input-output data pairs, each of which can serve as a sample. The input-output data pairs include input data for the quality prediction model and corresponding output data. Input data includes cellulose degree of polymerization, solvent concentration, cellulose content, temperature, and stirring rate; output data includes uniformity and slurry viscosity. This step, through the method of generating a large amount of data using the simulation system, can digitally present the dynamic process of cellulose swelling, obtaining sufficient data for subsequent prediction models without the need for physical experiments.

[0055] Furthermore, the input and output data pairs are normalized using Min-Max normalization, and a dataset is constructed using the 10,000 normalized input and output data pairs.

[0056] S2. The physical model of the spinning pulp production process simulation system is introduced into the gradient boosting tree algorithm as a regularization constraint to obtain an improved gradient boosting tree algorithm.

[0057] Specifically, in this embodiment, the improved gradient boosting tree algorithm can also be called the physically constrained regularized improved gradient boosting tree algorithm. It introduces the physical model of the spinning sizing process simulation system as a regularization constraint on the gradient boosting tree algorithm to optimize the loss function and residual calculation. The physical model of the spinning sizing process simulation system is introduced into the gradient boosting tree algorithm as a regularization constraint, including uniformity constraint terms and viscosity constraint terms, which respectively satisfy the following relationships:

[0058]

[0059] in, For the theoretical uniformity of the j-th sample, The rate at which stirring intensity affects the uniformity of the porridge. Let be the numerical value of the flow field velocity for the j-th sample. Let j be the stirring rate of the j-th sample. Let the theoretical viscosity be the j-th sample. Let be the cellulose content of the j-th sample. Let represent the degree of swelling of the j-th sample. Let J be the solvent concentration of the j-th sample. Let be the temperature of the j-th sample. The value range is generally 0.01 to 0.1, and its initial value can be set to 0.01.

[0060] The homogeneity constraint term adopts the exponential approximation formula for homogeneity calculation in the stirring and homogeneity model, rather than directly using the homogeneity calculation method in the stirring and homogeneity model, mainly for the following considerations: 1. The calculation of uniformity in the stirring and uniformity model involves spatial integration, which is highly complex and difficult to directly embed into the loss function of the machine learning model. Therefore, the uniformity constraint term uses an exponential function to directly link uniformity with the stirring rate, simplifying the calculation and making it suitable for rapid iterative training.

[0061] 2. The simulation system needs to accurately simulate the dynamic distribution of solvent concentration and generate realistic homogeneity data. The homogeneity constraint term only needs to provide a rough but physically reasonable homogeneity estimate to constrain the prediction of the improved gradient boosting tree algorithm, ensuring that the prediction results conform to the trend of stirring, rather than accurately reproducing the simulation.

[0062] 3. Physical constraints need to be calculated quickly in each round of training. Integration will significantly increase the computational cost. The uniformity constraint uses an exponential approximation formula, which is simple and efficient, and retains the positive correlation between stirring intensity and uniformity.

[0063] 4. The simulation data is in steady state or quasi-steady state, and the uniformity is mainly driven by the stirring intensity. The exponential function can effectively capture this relationship, which meets the actual needs of predicting the quality of porridge.

[0064] Furthermore, the loss function of the improved gradient boosting tree algorithm satisfies the following relationship:

[0065]

[0066] in, For total loss, MSE is a data-driven term, and MSE measures... , and , deviation, This is the algorithm's predicted value for the viscosity of the porridge. This is the algorithm's predicted value for uniformity. This represents the true viscosity of the porridge. The true value of uniformity, For physical loss items, measure , and , deviation, For theoretical viscosity, For theoretical uniformity, This is the viscosity loss term. For uniformity loss term, Here, N represents the regularization weights, and N is the total number of samples. Let be the algorithm's predicted value for the j-th sample. This represents the true value for the j-th sample. In this embodiment, the viscosity and uniformity of the porridge output by the simulation system are taken as the true values ​​for the viscosity and uniformity of the porridge.

[0067] The viscosity loss term and the uniformity loss term satisfy the following relationships:

[0068]

[0069] in, Let be the predicted viscosity of the porridge for the j-th sample by the algorithm. Let be the algorithm's predicted value for the uniformity of the j-th sample.

[0070] The residuals of the improved gradient boosting tree algorithm are driven by both data and physical models, balancing data fitting with physical consistency. This ensures that the predictions of the viscosity and uniformity of the porridge are both accurate and conform to the swelling law. The residuals of the improved gradient boosting tree algorithm satisfy the following relationship:

[0071]

[0072] in, Let J be the viscosity residual of the j-th sample. Let j be the uniformity residual of the j-th sample. This represents the true value of the viscosity of the porridge in the j-th sample. Let be the true value of uniformity in the j-th sample. or It represents data error and drives the model to fit the simulation data. or This indicates a physical deviation, constraining the predicted value to be close to the theoretical value. For regularization weights, and These are the task weights, and the initial value for both task weights is 1.

[0073] More specifically, the implementation process of the improved gradient boosting tree algorithm can be divided into 5 stages, as follows: 1. Data preparation stage: Construct a dataset based on the content described in step S1, and divide the dataset into training set, validation set and test set in a ratio of 8:1:1.

[0074] 2. Physical constraint calculation: The theoretical homogeneity and theoretical viscosity of each sample are calculated using the exponential approximation formula of the homogeneity model in the simulation system and the viscosity calculation model as physical constraints. That is, the theoretical homogeneity and theoretical viscosity of each sample are calculated as physical constraints based on the homogeneity constraint term and the viscosity constraint term.

[0075] 3. Model Initialization: First, initialize the predicted values ​​for uniformity and porridge viscosity separately, taking the mean of the target variable, i.e. , ,in, For the initial uniformity prediction value, Initialize the viscosity prediction value of the porridge; then create two empty decision tree sets to store the trees for uniformity prediction value and viscosity prediction value respectively, ensuring that each output has an independent optimization path. The initial tree parameters include a maximum depth of 4~8 and a minimum leaf weight of 1.

[0076] 4. Iterative Training: Calculate the adjusted residual for each sample by combining data error and physical bias. Train a decision tree for both the viscosity and uniformity of the porridge, fit the corresponding residuals, and update the predicted values ​​with a step size of 0.05 to ensure smooth updates. Iterate continuously until the preset number of trees is reached or the validation set error converges, and output the final predicted values ​​for the viscosity and uniformity of the porridge.

[0077] 5. Model Training, Evaluation and Testing: Use the dataset to train, validate and test the improved gradient boosting tree algorithm for predicting key indicators of lyocell fiber spinning sizing quality.

[0078] As discussed above, the improved gradient boosting tree algorithm, by embedding the physical model of the simulation system into the loss function, not only retains the strong fitting ability of gradient boosting trees for nonlinear relationships and feature interactions, but also ensures that the prediction results conform to the physical laws of the swelling process of cellulose in NMMO solvent. This significantly improves physical consistency, prediction accuracy, and the model's adaptability to simulation data, making it particularly suitable for industrial scenarios of predicting the quality of porridge. In contrast, the traditional gradient boosting tree relies solely on data-driven approaches and lacks constraints on physical mechanisms, making it difficult to meet this specific requirement.

[0079] S3. Use the dataset to train and validate the improved gradient boosting tree algorithm to obtain a quality prediction model, which is used to predict key indicators of the quality of lyocell fiber spinning pulp.

[0080] Specifically, in this embodiment, the dataset obtained in step S1 is divided into a training set, a validation set, and a test set in an 8:1:1 ratio to complete the training, validation, and testing of the improved gradient boosting tree algorithm, ultimately obtaining a quality prediction model. Then, by adjusting the input to the quality prediction model, the corresponding key indicators of lyocell fiber spinning quality can be obtained.

[0081] More specifically, multiple quality prediction models can be trained by adjusting parameters such as the number of trees, maximum depth, learning rate, and regularization weights in the improved gradient boosting tree algorithm. Then, each quality prediction model can be tested on a validation set. By calculating the root mean square error and coefficient of determination of each quality prediction model, the root mean square error of the quality prediction model can be ensured to be below 5% and the coefficient of determination can be above 0.9, thereby selecting the best quality prediction model.

[0082] This embodiment utilizes simulation data to train an improved gradient boosting tree algorithm to construct a quality prediction model for predicting key indicators of lyocell fiber spinning sizing quality. This not only significantly reduces R&D time and economic costs, but also ensures that the prediction results conform to physical laws, supporting rapid process parameter optimization, and is especially suitable for emergency adjustment scenarios.

[0083] It should be noted that in some cases, the actions described in the specification can be performed in different orders and still achieve the desired results. In this embodiment, the order of steps is given only to make the embodiment clearer and easier to explain, and not to limit it.

[0084] In one optional embodiment, please refer to Figure 2 To improve the practicality of this method, the present invention also provides a simulation and quality prediction system for lyocell fiber spinning sizing process. The lyocell fiber spinning sizing process simulation and quality prediction system includes: a data acquisition device 1, a data output device 2, a processor 3, and a storage device 4. The storage device 4 includes a computer-readable storage medium storing a computer program. The computer program includes program instructions, which, when executed by the processor 3, cause the processor 3 to perform the contents described in steps S1 to S3.

[0085] In summary, the present invention has at least the following beneficial effects: 1. Existing technologies lack a systematic prediction method for pulp quality, relying solely on indirect derivation from experimental data. This method, for the first time, proposes a systematic prediction scheme for lyocell fiber spinning pulp quality based on simulation data and an improved gradient boosting tree algorithm, filling a technological gap and providing a complete process from data generation to lyocell fiber spinning pulp quality prediction, thus enhancing the systematicness and scientific rigor of pulp quality prediction.

[0086] 2. Existing technologies rely on laboratory and pilot-scale experiments to adjust process parameters to obtain high-quality spinning sizing, which is time-consuming and costly. This method generates a large amount of data through a simulation system, eliminating the need for physical experiments to obtain sufficient data for training the prediction model. This significantly reduces R&D time and economic costs, supports rapid process parameter optimization, and is particularly suitable for urgent adjustment scenarios.

[0087] 3. Existing technologies do not systematically utilize the physical laws governing the swelling process of cellulose in NMMO solvent, and the predicted quality of the pulp may not match the actual mechanism. This improved gradient boosting tree algorithm incorporates a physical constraint regularization term into the loss function, integrating the physical model of the spinning pulp production process simulation system into the model training process. This ensures that the predictions conform to the physical characteristics of cellulose swelling, viscosity, and uniformity, enhancing the scientific rationality of the prediction results, avoiding potential physical biases in traditional data-driven models, and improving the reliability of the model for industrial applications.

[0088] 4. The present invention provides a system adapted to the method, which can improve the practicality of the method.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for simulating and predicting the quality of lyocell fiber spinning sizing process, characterized in that, Includes the following steps: A simulation system for the production process of spinning pulp was constructed based on the swelling mechanism of cellulose in NMMO solution. Simulation data of lyocell fiber spinning pulp was generated, and multiple sets of input-output data pairs were obtained to construct a dataset. An improved gradient boosting tree algorithm is obtained by incorporating the physical model of the spinning pulp production process simulation system as a regularization constraint into the gradient boosting tree algorithm. The improved gradient boosting tree algorithm was trained and validated using the dataset to obtain a quality prediction model, which is used to predict key indicators of the quality of lyocell fiber spinning pulp.

2. The method for simulation and quality prediction of lyocell fiber spinning sizing process according to claim 1, characterized in that: The physical models of the spinning pulp production process simulation system include a solvent penetration model, a swelling kinetics model, a stirring and uniformity model, and a viscosity calculation model.

3. The method for simulation and quality prediction of lyocell fiber spinning sizing process according to claim 2, characterized in that, The solvent permeation model satisfies the following relationship: ， in, denoted as the solvent concentration of the NMMO solution, t as time, and D as the diffusion coefficient related to the degree of polymerization of cellulose. This is the spatial second derivative of the solvent concentration gradient.

4. The method for simulation and quality prediction of lyocell fiber spinning sizing process according to claim 3, characterized in that, The swelling kinetics model satisfies the following relationship: ， in, The degree of swelling of cellulose in NMMO solvent. The swelling rate constant is The solvent concentration is... To balance the solvent concentration, This represents the maximum degree of swelling.

5. The method for simulation and quality prediction of lyocell fiber spinning sizing process according to claim 4, characterized in that, The stirring effect and the homogeneity model satisfy the following relationship: ， ， in, The solvent concentration is... The velocity of the flow field caused by stirring characterizes the fluid motion induced by stirring. Let be the spatial first derivative of the solvent concentration gradient, and U be the uniformity of the sizing paste used in Lyocell fiber spinning. V represents the average solvent concentration in the NMMO solution, and V is the volume of the mixture of cellulose and NMMO solution.

6. The method for simulation and quality prediction of lyocell fiber spinning sizing process according to claim 5, characterized in that, The viscosity calculation model satisfies the following relationship: ， in, The viscosity of the slurry for spinning lyocell fibers. Where is the viscosity constant. The content is cellulose. The degree of swelling, The solvent concentration is given, and m, n, and p are the weights of the effects of the cellulose content, the degree of swelling, and the solvent concentration on the viscosity of the porridge, respectively. This is the viscosity activation energy.

7. The method for simulation and quality prediction of lyocell fiber spinning sizing process according to claim 6, characterized in that: The input-output data pair includes input data for inputting the quality prediction model and output data corresponding to the input data. The input data includes the degree of polymerization of cellulose, the solvent concentration, the cellulose content, the temperature, and the stirring rate. The output data includes the uniformity and the viscosity of the porridge.

8. The method for simulation and quality prediction of lyocell fiber spinning sizing process according to claim 6, characterized in that, The physical model of the spinning slurry production process simulation system is introduced into the gradient boosting tree algorithm as a regularization constraint, including a uniformity constraint term and a viscosity constraint term. The uniformity constraint term and the viscosity constraint term satisfy the following relationships respectively: ， ， in, For the theoretical uniformity of the j-th sample, The rate at which stirring intensity affects the uniformity of the porridge. Let be the numerical value of the flow field velocity for the j-th sample. This is the proportionality coefficient. Let j be the stirring rate of the j-th sample. Let the theoretical viscosity be the j-th sample. Let m be the viscosity constant, and m, n, and p be the weights of the effects of the cellulose content, the degree of swelling, and the solvent concentration on the viscosity of the porridge, respectively. Let be the cellulose content of the j-th sample. The degree of swelling of the j-th sample, The solvent concentration for the j-th sample is... Where is the viscosity activation energy, and R is the gas constant. Let be the temperature of the j-th sample.

9. The method for simulation and quality prediction of lyocell fiber spinning sizing process according to claim 1, characterized in that, The loss function of the improved gradient boosting tree algorithm satisfies the following relationship: ， ， in, For the total loss, For data-driven items, For physical loss items, This is the viscosity loss term. For uniformity loss term, Here, N represents the regularization weights, and N is the total number of samples. Let be the algorithm's predicted value for the j-th sample. Let be the true value of the j-th sample.

10. A simulation and quality prediction system for lyocell fiber spinning sizing process, characterized in that, The lyocell fiber spinning sizing process simulation and quality prediction system includes: a data acquisition device, a data output device, a processor, and a storage device. The storage device includes a computer-readable storage medium storing a computer program. The computer program includes program instructions, which, when executed by the processor, cause the processor to implement the lyocell fiber spinning sizing process simulation and quality prediction method as described in any one of claims 1-9.

Images