A method and device for optimizing the production process of a carbon fiber precursor washing tank
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN UNIV OF TECH
- Filing Date
- 2026-02-07
- Publication Date
- 2026-06-09
Smart Images
Figure CN122174625A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of carbon fiber production technology, and in particular to a method and apparatus for optimizing the production process of carbon fiber precursor washing tank. Background Technology
[0002] In the production of carbon fiber precursor, the water washing tank process is a key step in removing dimethyl sulfoxide (DMSO) from the precursor. The precise control of the DMSO mass fraction directly determines the mechanical properties and product qualification rate of the carbon fiber precursor. Therefore, the carbon fiber precursor water washing tank production system has a high requirement for the accuracy of DMSO mass fraction prediction.
[0003] However, existing DMSO mass fraction prediction methods do not integrate diverse production needs such as prediction accuracy, computational efficiency, and process constraints when designing objective and decision functions, resulting in weak predictive capabilities. Furthermore, the lack of quantitative consideration of process constraints during model optimization leads to insufficient prediction accuracy of DMSO mass fraction, ultimately affecting the optimization effect of the washing process and the stability of carbon fiber precursor production quality. Summary of the Invention
[0004] To address the aforementioned problems, in a first aspect, the present invention provides an optimized method for the production process of a carbon fiber precursor washing tank, comprising: Obtain production demand data and historical production data of the carbon fiber precursor washing tank production system, preprocess the historical production data, and obtain sample process data and label DMSO quality score. Based on production demand data, construct objective and decision functions, and then construct an initial sparse LS-SVM model based on the objective and decision functions. Using sample process data and label DMSO quality scores, the objective function and decision function of the initial sparse LS-SVM model are iteratively optimized using the improved NOA algorithm to obtain the final sparse LS-SVM model. The process data to be tested in the carbon fiber precursor washing tank production system is obtained. The process data to be tested is input into the final sparse LS-SVM model to predict the DMSO mass fraction and obtain the final DMSO mass fraction. Based on the final DMSO mass fraction, the production process of the carbon fiber precursor washing tank is optimized.
[0005] Optionally, the preprocessing of historical production data to obtain sample process data and label DMSO quality scores includes: Obtain historical process data from historical production data, as well as the corresponding historical DMSO quality scores; Historical process data and historical DMSO quality scores were cleaned and normalized to obtain sample process data and labeled DMSO quality scores.
[0006] Optional: The production demand data includes prediction accuracy demand data, computational efficiency demand data, washing process constraint data, local fitting demand data, and global fitting demand data.
[0007] Optionally, the process of constructing the objective function includes: Based on the prediction accuracy requirement data, set the weight vector, error penalty coefficient and regression error term; based on the computational efficiency requirement data, set the L1 sparsity penalty coefficient and L2 regularization penalty coefficient to obtain the importance coefficient of the sample process data. The first target value is obtained by calculating the weight vector, the second target value is obtained by calculating the error penalty coefficient and the regression error term, the third target value is obtained by calculating the L1 sparsity penalty coefficient and the importance coefficient, and the fourth target value is obtained by calculating the L2 regularization penalty coefficient and the importance coefficient. Add the first target value, the second target value, the third target value, and the fourth target value to obtain the total target value; Based on the constraints of the water washing process, set the constraints, and under the constraints, minimize the total objective value as the objective function.
[0008] Optionally, the process of constructing the decision function includes: Set the RBF kernel width based on the local fitting requirements data, and set the polynomial kernel offset, polynomial order, and kernel function weighting coefficients based on the global fitting requirements data. Obtain key support vectors from the sample process data, construct an RBF kernel function based on the sample process data, key support vectors, and RBF kernel width, and construct a polynomial kernel function based on the sample process data, key support vectors, polynomial kernel offset, and polynomial order. Decision functions are constructed based on the kernel function weighting coefficients, RBF kernel function, and polynomial kernel function.
[0009] Optionally, the step of iteratively optimizing the objective function and decision function of the initial sparse LS-SVM model using sample process data and label DMSO quality scores, and obtaining the final sparse LS-SVM model, includes: By improving the NOA algorithm, the error penalty coefficient, L1 sparsity penalty coefficient, and L2 regularization penalty coefficient of the objective function of the initial sparse LS-SVM model, as well as the RBF kernel width, polynomial kernel offset, polynomial order, and kernel function weighting coefficient of the decision function, are iteratively optimized to obtain an optimized sparse LS-SVM model. The sample process data is input into the optimized sparse LS-SVM model to predict the DMSO quality score, and the predicted DMSO quality score is obtained. Obtain the number of key support vectors for predicting DMSO quality scores, as well as the total number of samples in the sample process data; The optimal fitness value is calculated based on the predicted DMSO quality score, the labeled DMSO quality score, the number of key support vectors, and the total number of samples. The initial sparse LS-SVM model was optimized multiple times, and the optimized sparse LS-SVM model with the smallest fitness value was taken as the final sparse LS-SVM model.
[0010] Optionally, the improved NOA algorithm is used to iteratively optimize the error penalty coefficient, L1 sparsity penalty coefficient, and L2 regularization penalty coefficient of the objective function of the initial sparse LS-SVM model, as well as the RBF kernel width, polynomial kernel offset, polynomial order, and kernel function weighting coefficients of the decision function, to obtain an optimized sparse LS-SVM model, including: A set of error penalty coefficients, L1 sparsity penalty coefficients, L2 regularization penalty coefficients, RBF kernel width, polynomial kernel offset, polynomial order, and kernel function weighting coefficients are used as the positions of individual star crows to construct a star crow population. Randomly perturb the Star Crow population and update the position of each Star Crow individual; The root mean square error, constraint violation degree, and dynamic adjustment factor of the individual Star Raven are calculated based on the updated position. The comprehensive fitness value is then calculated based on the root mean square error, constraint violation degree, and dynamic adjustment factor. The population of ravens was subjected to multiple random perturbations. The position with the lowest comprehensive fitness value was taken as the optimal position. The parameters of the initial sparse LS-SVM model were configured according to the optimal position to obtain the optimized sparse LS-SVM model.
[0011] Secondly, the present invention provides a device for optimizing the production process of a carbon fiber precursor washing tank, used to implement the aforementioned method for optimizing the production process of a carbon fiber precursor washing tank, the device comprising: The data acquisition module is used to acquire production demand data and historical production data of the carbon fiber precursor washing tank production system, preprocess the historical production data, and obtain sample process data and label DMSO quality score. The sparse LS-SVM model building module is used to construct the objective function and decision function based on production demand data, and to construct the initial sparse LS-SVM model based on the objective function and decision function. The sparse LS-SVM model training module is used to iteratively optimize the objective function and decision function of the initial sparse LS-SVM model using sample process data and label DMSO quality scores, and obtain the final sparse LS-SVM model. The parameter optimization module is used to acquire the process data to be tested in the carbon fiber precursor washing tank production system. The process data to be tested is input into the final sparse LS-SVM model to predict the DMSO mass fraction, obtain the final DMSO mass fraction, and optimize the production process of the carbon fiber precursor washing tank based on the final DMSO mass fraction.
[0012] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for optimizing the production process of the carbon fiber raw yarn washing tank.
[0013] Fourthly, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method for optimizing the production process of the carbon fiber raw filament washing tank.
[0014] The present invention has the following beneficial effects: 1. Based on production demand data, an objective function and a decision function are constructed that balance prediction accuracy, computational efficiency, and process constraints. This enables the sparse LS-SVM model built based on the objective function and decision function to accurately predict DMSO quality scores. The core parameters of the initial sparse LS-SVM model are iteratively optimized by improving the NOA algorithm. By constructing constraint violation degree and dynamic adjustment factor, the process constraints are quantitatively considered. A comprehensive fitness function is constructed by combining root mean square error, constraint violation degree, and dynamic adjustment factor to achieve efficient optimization of parameter combinations and significantly improve the prediction accuracy of DMSO quality scores.
[0015] 2. The objective function integrates multiple requirements such as prediction accuracy, computational efficiency, and process constraints. Through flexible configuration of weight vectors, L1 sparse penalty coefficients, and L2 regularization penalty coefficients, high accuracy in DMSO quality score prediction is ensured. The L1 sparse penalty coefficient achieves model lightweighting, while the L2 regularization penalty coefficient avoids overfitting, balancing the model's accuracy and generalization ability. A weighted fusion of RBF kernel functions and multinomial kernel functions is used to construct the decision function, addressing the difficulty of a single kernel function in achieving both local accuracy and global consistency, adapting to the fitting requirements of different intervals in the washing process. The sparsed LS-SVM model constructed based on the objective function and decision function balances prediction accuracy, computational efficiency, and process constraints, achieving high-precision prediction of DMSO quality scores.
[0016] 3. The improved NOA algorithm expands the search range by simulating the foraging and storage behavior of crows. A comprehensive fitness function, constructed by combining root mean square error, constraint violation degree, and dynamic adjustment factor, achieves precise selection of parameter combinations, significantly improving the model parameter fit. During optimization, the number of key support vectors is controlled by the L1 sparse penalty coefficient, simplifying the model structure and reducing computational load while ensuring prediction accuracy. This balances high accuracy with high computational efficiency, meeting the needs of real-time prediction and rapid optimization in industrial scenarios. Through dynamic adjustment factor and constraint violation quantification mechanism, the optimized model parameters strictly conform to the parameter range constraints and adjustment rules of the washing process, further improving the prediction accuracy of DMSO quality fraction. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a structural diagram of the device according to an embodiment of the present invention. Detailed Implementation
[0019] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.
[0020] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.
[0021] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.
[0022] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.
[0023] Reference Figure 1 This invention provides an optimized method for the production process of carbon fiber precursor washing tanks, comprising: The S100 acquires production demand data and historical production data from the carbon fiber precursor washing tank production system, preprocesses the historical production data, and obtains sample process data and label DMSO quality scores.
[0024] In some embodiments, the preprocessing of historical production data to obtain sample process data and label DMSO quality scores includes: Obtain historical process data from historical production data, as well as the corresponding historical DMSO quality scores; Historical process data and historical DMSO quality scores were cleaned and normalized to obtain sample process data and labeled DMSO quality scores.
[0025] In some embodiments, the process data includes four indicators: washing tank temperature, washing water flow rate, washing time, and raw yarn speed. These indicators cover the key factors affecting the washing effect, ensuring the completeness of the input information while avoiding excessive dimensionality that could increase model complexity. The applicable range is 1-10 process parameters, which can be expanded according to the actual scenario.
[0026] The DMSO (dimethyl sulfoxide) mass fraction includes the DMSO mass fraction at the outlet of the washing tank and the DMSO mass fraction of the carbon fiber precursor, totaling two indicators. These indicators comprehensively evaluate the washing quality and are directly related to the performance of the carbon fiber precursor products. The applicable range is 1-5 indicators, which can be adapted to the quality control needs of different processes.
[0027] Historical production data from the past 6-12 months were extracted from the database of the carbon fiber precursor washing tank production system. The historical process data included four core parameters: washing tank temperature (°C), washing water flow rate (m³ / h), washing time (min), and precursor speed (m / min). The historical DMSO mass fraction for the corresponding period was extracted simultaneously, including the DMSO mass fraction (%) at the washing tank outlet and the residual DMSO mass fraction (%) in the carbon fiber precursor, ensuring that each set of process data corresponds one-to-one with the DMSO mass fraction.
[0028] Outlier detection algorithms are used to remove outliers from the process data that exceed a reasonable range; a small amount of missing data is supplemented by missing value imputation; if the missing data rate exceeds 10% in a certain period, the process-quality data group corresponding to that period is removed; duplicate records are removed to avoid data redundancy affecting model training.
[0029] The cleaned historical process data and historical DMSO quality scores were normalized using a min-max normalization method, mapping the data to the [0,1] interval. Normalization yielded standardized sample process data and labeled DMSO quality scores, ensuring uniformity of dimensions across different data dimensions and improving model convergence speed and prediction accuracy.
[0030] S200 constructs an objective function and a decision function based on production demand data, and then constructs an initial sparse LS-SVM model based on the objective function and the decision function.
[0031] In some embodiments, the production demand data includes prediction accuracy demand data, computational efficiency demand data, washing process constraint data, local fitting demand data, and global fitting demand data.
[0032] In some embodiments, the required accuracy data is used to define the error threshold for predicting the mass fraction of DMSO, such as setting the root mean square error of the predicted mass fraction of DMSO at the outlet of the washing tank to ≤0.05% and the absolute error of the predicted mass fraction of DMSO remaining in the raw yarn to ≤0.03%. The accuracy requirements are adjusted according to the production quality level. For high-end products, the error threshold needs to be reduced by 30%. At the same time, the sample size requirement for accuracy verification is marked, such as ≥500 sets of measured data.
[0033] Computational efficiency requirements are used to set the upper limit of the time taken for a single prediction by the model, such as ≤0.5 seconds / set of data in industrial real-time scenarios and ≤5 seconds / set of data in offline optimization scenarios; limit the proportion of key support vectors, such as ≤20% of the total number of samples, to avoid the model being too complex and causing computational delays; and clarify the maximum threshold for the number of iterations for optimization, such as ≤100 iterations for improving the NOA algorithm, to balance accuracy and efficiency.
[0034] The washing process constraint data is used to define the legal range of core process parameters, such as washing tank temperature 50-90℃, washing water flow rate 1-5m³ / h, washing time 3-10min, and raw fiber speed 2-8m / min; set the qualified threshold for DMSO mass fraction, such as ≤1.5% at the outlet and ≤0.3% of raw fiber residue; and clarify parameter adjustment constraints, such as ≤5℃ for a single temperature adjustment, to avoid drastic process fluctuations.
[0035] The local fitting requirement data is used to set the local fitting priority for process-sensitive intervals, requiring the prediction error in this interval to be 20% lower than the global error; a reference range for the RBF kernel width is given, such as RBF kernel width σ∈[0.1,2.0]. The higher the proportion of the sensitive interval, the smaller the kernel width value should be, such as RBF kernel width σ∈[0.1,0.8], to enhance the ability to capture local features.
[0036] The global fitting requirement data is used to clarify the consistency requirements for fitting the process parameters across the entire range, with the global prediction error fluctuation coefficient ≤15%; the range of polynomial kernel parameters is set, such as the polynomial order d∈[2,4], i.e., 3-4 order for complex processes and 2 order for simple processes, the kernel offset c∈[0.5,2.0], and the kernel function weighting coefficient μ∈[0.3,0.7], to ensure that there is no significant deviation in the global trend fitting.
[0037] In some embodiments, the process of constructing the objective function includes: Based on the prediction accuracy requirement data, set the weight vector, error penalty coefficient and regression error term; based on the computational efficiency requirement data, set the L1 sparsity penalty coefficient and L2 regularization penalty coefficient to obtain the importance coefficient of the sample process data. The first target value is obtained by calculating the weight vector, the second target value is obtained by calculating the error penalty coefficient and the regression error term, the third target value is obtained by calculating the L1 sparsity penalty coefficient and the importance coefficient, and the fourth target value is obtained by calculating the L2 regularization penalty coefficient and the importance coefficient. Add the first target value, the second target value, the third target value, and the fourth target value to obtain the total target value; Based on the constraints of the water washing process, set the constraints, and under the constraints, minimize the total objective value as the objective function.
[0038] In some embodiments, the objective function M is expressed as:
[0039] Where i is the ID of the sample process data, and n is the number of sample process data. The first target value, The second target value, The third target value, This is the fourth target value.
[0040] w is the weight vector of the sparse LS-SVM model, used to measure the contribution of input features to the output, corresponding to the contribution weights of the input process parameters of the washing tank to the DMSO quality fraction. For example, if temperature has the greatest impact on the DMSO removal effect, then the weight value of the temperature dimension in w will be larger.
[0041] e i Let e be the regression error term for the process data of the i-th sample group, that is, the deviation between the DMSO mass fraction predicted by the model and the actual measured value under the i-th set of process parameters. For example, if the actual DMSO mass fraction of a certain sample group is 0.5%, and the model predicts it to be 0.45%, then e i =0.05%.
[0042] a iLet be the importance coefficient of the i-th sample process data for the sparsed LS-SVM model. In the sparsed LS-SVM model, a i The samples with a value of 0 are key support vectors, corresponding to the combinations of process parameters that have the greatest impact on the DMSO mass fraction, such as critical temperature and flow rate range.
[0043] r is the error penalty coefficient, used to control the degree of penalty for regression error and the model's tolerance for prediction error. In water washing tank prediction, if high accuracy is required for the prediction of DMSO mass fraction, such as an error ≤0.1%, then r needs to be increased.
[0044] This is the L1 sparsity penalty coefficient, used to make most Lagrange multipliers approach 0, thus sparsifying the model and controlling the degree of sparsity, i.e., the number of key support vectors. In water washing tank prediction, The larger the value, the more accurate the sample of key process parameters can be selected.
[0045] This is the L2 regularization penalty coefficient, used to avoid overfitting of the model due to L1 regularization. In the prediction of the washing tank, if... If the value is too large, the model may only focus on a few key samples and ignore the overall process rules. In this case, the value needs to be increased. .
[0046] The L2 norm of the weight vector w represents the magnitude of the weight vector w, which reflects the overall contribution of the input features to the output. The smaller the value, the simpler the model and the stronger its generalization ability.
[0047] The expression for the constraint is:
[0048] This represents the key support vector of the i-th group of sample process data. High-dimensional eigenvectors mapped to a high-dimensional feature space. In the context of a water washing tank, these are used to capture the complex nonlinear relationship between process parameters and DMSO mass fraction, such as the nonlinear law that the DMSO removal effect drops sharply after the temperature rises to a critical value.
[0049] This represents the normalized true value of the predicted DMSO quality fraction corresponding to the process data of the i-th sample, which corresponds to the normalized result of the DMSO quality fraction of the raw yarn or the DMSO quality fraction at the outlet of the water washing tank in the water washing tank scenario.
[0050] 'b' represents the bias term in the sparse LS-SVM model, used to adjust the position of the decision function. Specifically, it adjusts the baseline value of the prediction result, addressing the issue of non-zero prediction values when all input features are zero. In water washing tank prediction, it can offset the systematic bias between process parameters and DMSO mass fraction.
[0051] In some embodiments, the process of constructing the decision function includes: Set the RBF kernel width based on the local fitting requirements data, and set the polynomial kernel offset, polynomial order, and kernel function weighting coefficients based on the global fitting requirements data. Obtain key support vectors from the sample process data, construct an RBF kernel function based on the sample process data, key support vectors, and RBF kernel width, and construct a polynomial kernel function based on the sample process data, key support vectors, polynomial kernel offset, and polynomial order. Decision functions are constructed based on the kernel function weighting coefficients, RBF kernel function, and polynomial kernel function.
[0052] In some embodiments, decision function The expression is:
[0053] in, The input feature vector for the sample process data, For the i-th key support vector, These are the weighting coefficients for the kernel function.
[0054] The RBF kernel function, or radial basis function, represents the input feature vector of the sample process data. With the i-th key support vector In the RBF kernel function, the closer the similarity value is to 1, the more similar the process parameters of the two devices are. This is expressed as:
[0055] in, The width of the RBF kernel is used to control the local fitting ability of the kernel function.
[0056] Let be a polynomial kernel function, representing the input feature vector of the sample process data. With the i-th key support vector The similarity under the polynomial kernel function is used to capture the global nonlinear relationship of process parameters, and its expression is:
[0057] Where c is the polynomial kernel offset, used to adjust the nonlinearity of the kernel function. If c=0, the kernel function is a homogeneous polynomial kernel; if c>0, the kernel function is a non-homogeneous polynomial kernel, which can enhance the model's ability to fit low-similarity samples. In water washing tank prediction, c=1 is usually set to avoid the kernel function relying too much on the inner product and improve the model's generalization ability. d is the polynomial order, used to control the global fitting ability of the kernel function. The smaller d is, the lower the nonlinearity of the kernel function, and the more the model focuses on linear relationships; the larger d is, the higher the nonlinearity of the kernel function, and the model can capture more complex relationships, but it is prone to overfitting. In water washing tank prediction, d=2 or d=3 is usually set to balance the fitting ability and generalization ability.
[0058] The LS-SVM model is a least squares support vector machine. The weight vector w, bias term b, and initial key support vector obtained by solving are substituted into the preset RBF kernel function and polynomial kernel function. Combined with the weighting coefficient μ of the kernel function, they are concatenated to form a complete decision function. Finally, the optimization logic of the objective function and the prediction logic of the decision function are integrated in the least squares support vector machine to form the initial sparse LS-SVM model.
[0059] S300 uses sample process data and label DMSO quality scores to iteratively optimize the objective and decision functions of the initial sparse LS-SVM model using an improved NOA algorithm, thus obtaining the final sparse LS-SVM model.
[0060] In some embodiments, the step of iteratively optimizing the objective function and decision function of the initial sparse LS-SVM model using sample process data and label DMSO quality scores, to obtain the final sparse LS-SVM model, includes: By improving the NOA algorithm, the error penalty coefficient, L1 sparsity penalty coefficient, and L2 regularization penalty coefficient of the objective function of the initial sparse LS-SVM model, as well as the RBF kernel width, polynomial kernel offset, polynomial order, and kernel function weighting coefficient of the decision function, are iteratively optimized to obtain an optimized sparse LS-SVM model. The sample process data is input into the optimized sparse LS-SVM model to predict the DMSO quality score, and the predicted DMSO quality score is obtained. Obtain the number of key support vectors for predicting DMSO quality scores, as well as the total number of samples in the sample process data; The optimal fitness value is calculated based on the predicted DMSO quality score, the labeled DMSO quality score, the number of key support vectors, and the total number of samples. The initial sparse LS-SVM model was optimized multiple times, and the optimized sparse LS-SVM model with the smallest fitness value was taken as the final sparse LS-SVM model.
[0061] In some embodiments, the sample process data and the label DMSO quality scores are divided into training sample data and test sample data in a 7:3 ratio. The training sample data is divided into k sample sets on average. Each sample set includes n sample process data and corresponding n label DMSO quality scores. Each normalized sample process parameter data forms an input feature vector. The n input feature vectors form an n×m sample input matrix, where m is the process parameter dimension. The n label DMSO quality scores form an n×2 output matrix, where 2 corresponds to two DMSO quality score indicators.
[0062] The improved NOA algorithm (improved Star Raven algorithm) is launched to optimize the core parameters of the sparse LS-SVM model. The parameters to be optimized include: error penalty coefficient r, L1 sparse penalty coefficient λ1, L2 regularization penalty coefficient λ2, RBF kernel width σ, polynomial kernel offset c, polynomial order d, and kernel function weighting coefficient μ.
[0063] Error value of the i-th iteration The calculation formula is:
[0064] in, This represents the process data of the j-th sample in the i-th iteration. This represents the DMSO quality score of the j-th label in the i-th iteration. Let represent the j-th predicted DMSO quality score in the i-th iteration, and n be the total number of samples in the sample process data.
[0065] Calculate the average error after k iterations. k is the current iteration number.
[0066] Simultaneously calculate the proportion of support vectors of the model after each iteration. , where |S| is the number of key support vectors.
[0067] Optimize fitness value ,in, and The weight parameters are ω1=0.8 and ω2=0.2.
[0068] By iteratively updating the parameter combination using the improved NOA algorithm, the parameter combination with the smallest optimized fitness function value is selected as the optimal parameter combination for the sparse LS-SVM model, thus completing the model training.
[0069] In some embodiments, the iterative optimization of the error penalty coefficient, L1 sparsity penalty coefficient, and L2 regularization penalty coefficient of the objective function of the initial sparse LS-SVM model, as well as the RBF kernel width, polynomial kernel offset, polynomial order, and kernel function weighting coefficients of the decision function, using the improved NOA algorithm to obtain an optimized sparse LS-SVM model, includes: A set of error penalty coefficients, L1 sparsity penalty coefficients, L2 regularization penalty coefficients, RBF kernel width, polynomial kernel offset, polynomial order, and kernel function weighting coefficients are used as the positions of individual star crows to construct a star crow population. Randomly perturb the Star Crow population and update the position of each Star Crow individual; The root mean square error, constraint violation degree, and dynamic adjustment factor of the individual Star Raven are calculated based on the updated position. The comprehensive fitness value is then calculated based on the root mean square error, constraint violation degree, and dynamic adjustment factor. The population of ravens was subjected to multiple random perturbations. The position with the lowest comprehensive fitness value was taken as the optimal position. The parameters of the initial sparse LS-SVM model were configured according to the optimal position to obtain the optimized sparse LS-SVM model.
[0070] In some embodiments, the core parameters of the sparse LS-SVM model are optimized by improving the NOA algorithm, thereby achieving accurate prediction of key indicators in the carbon fiber precursor washing tank and providing data support for the optimization of the washing process. The specific objective is to minimize the root mean square error between the predicted and actual values, while ensuring that the optimization results meet the actual constraints of the washing process.
[0071] The improved NOA algorithm achieves parameter search by simulating the foraging-storage ecological behavior of jays. Individual jays search within a pre-defined two-dimensional parameter space. Each search includes two phases: foraging (exploring new parameter regions) and storage (retaining high-quality parameters). The individual's position is updated based on the fitness value. This process is repeated until the optimal overall fitness value is reached, and the corresponding optimal parameter combination is output.
[0072] The updated positions are then substituted into the sparsed LS-SVM model, and the root mean square error, constraint violation degree, and dynamic adjustment factor are calculated. The formula for calculating the root mean square error is as follows: =
[0073] in, Represents sample process data The root mean square error, where M is the total number of samples in the sample process data. For the DMSO quality fraction of the label, To predict the DMSO quality fraction.
[0074] The formula for calculating the degree of constraint violation is:
[0075] in, Represents sample process data The degree of constraint violation Represents sample process data Let m be the j-th inequality constraint function, and m be the number of inequality constraint functions. Represents sample process data The k-th equality constraint function, where n is the number of equality constraint functions. =0 indicates that the constraint is fully satisfied. The larger the value, the more severe the violation, providing a quantitative basis for subsequent penalty mechanisms.
[0076] Dynamic adjustment factor The calculation formula is:
[0077] Where t is the iteration number, t∈[1,T], and T is the maximum iteration number; θ(t) increases from 1 to 2 as the iteration number increases, realizing a lenient penalty in the early stage and a strict penalty in the later stage. In the early stage of iteration, some parameters that violate the constraints are allowed to be explored to expand the search range, and in the later stage, the convergence to a feasible solution is forced, avoiding the problem of insufficient exploration or unfeasible solution caused by traditional fixed penalty.
[0078] The formula for calculating the overall fitness value is:
[0079] in, Represents sample process data The overall fitness value; The base penalty coefficient is used to adjust the weight of constraint satisfaction and prediction accuracy. The larger the coefficient, the higher the penalty weight, and the algorithm prioritizes ensuring that the parameters meet the process constraints. The smaller the coefficient, the more emphasis is placed on optimizing prediction accuracy. The coefficient can be preset according to the constraint complexity.
[0080] The S400 acquires the process data to be tested from the carbon fiber precursor washing tank production system, inputs the process data to be tested into the final sparse LS-SVM model to predict the DMSO mass fraction, obtains the final DMSO mass fraction, and optimizes the production process of the carbon fiber precursor washing tank based on the final DMSO mass fraction.
[0081] In some embodiments, the final DMSO mass fraction is compared with a preset qualified threshold to identify three categories: a DMSO mass fraction > 1.5% at the outlet of the washing tank or a DMSO mass fraction > 0.3% for raw fiber residue is considered exceeding the standard; an outlet ≤ 1.5% and raw fiber residue ≤ 0.3% is considered qualified; and an outlet ≤ 0.8% and raw fiber residue ≤ 0.1% is considered optimal.
[0082] If the temperature exceeds the limit, prioritize increasing the water washing tank temperature, ensuring a single temperature increase of ≤5℃ and not exceeding 90℃, or increase the water washing flow rate, increasing by +0.5m per cycle. 3 / h, not exceeding 5m 3 / h; if it still does not meet the standard, extend the washing time by +1min per wash, not exceeding 10min, or reduce the raw yarn speed by -0.5m / min per wash, not less than 2m / min.
[0083] When the result is acceptable but not optimal, the sensitive process range parameters corresponding to the RBF core width are fine-tuned. For example, when the temperature is 60-70℃, the parameters are fine-tuned by ±2℃ to balance energy consumption and performance.
[0084] When optimal, maintain the current process parameters and record them as baseline parameters for subsequent production reference.
[0085] After parameter adjustment, new process data to be tested is collected and input into the model for re-prediction. If three consecutive sets of data meet the standard, the parameters are fixed; if they still do not meet the standard, the above adjustment process is repeated until the DMSO quality fraction meets the target requirements.
[0086] Reference Figure 2 This invention provides a carbon fiber precursor washing tank production process optimization device 20, used to implement a method for optimizing the carbon fiber precursor washing tank production process. The device includes: Data acquisition module 21 is used to acquire production demand data and historical production data of the carbon fiber precursor washing tank production system, preprocess the historical production data, and obtain sample process data and label DMSO quality score. The sparse LS-SVM model building module 22 is used to build objective functions and decision functions based on production demand data, and to build an initial sparse LS-SVM model based on the objective functions and decision functions; The sparse LS-SVM model training module 23 is used to iteratively optimize the objective function and decision function of the initial sparse LS-SVM model using the sample process data and the label DMSO quality score, and obtain the final sparse LS-SVM model. The parameter optimization module 24 is used to acquire the process data to be tested in the carbon fiber precursor washing tank production system, input the process data to be tested into the final sparse LS-SVM model to predict the DMSO mass fraction, obtain the final DMSO mass fraction, and optimize the production process of the carbon fiber precursor washing tank based on the final DMSO mass fraction.
[0087] This application provides an electronic device, including a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the carbon fiber raw material washing tank production process optimization method of any of the above schemes.
[0088] Specifically, the processor may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor may also include onboard memory for caching purposes. The processor may be a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this application.
[0089] Memory can be any medium capable of containing, storing, transmitting, propagating, or transmitting instructions. For example, memory can include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, instruments, or propagation media. Specific examples of memory include: magnetic storage devices such as magnetic tape or hard disk drives (HDDs); optical storage devices such as optical discs (CD-ROMs); and also random access memory (RAM) or flash memory; and / or wired / wireless communication links.
[0090] This application also provides a computer-readable medium storing a computer program that, when executed by a processor, implements the carbon fiber precursor washing tank production process optimization method described above. This computer-readable medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into that device / apparatus / system. The aforementioned computer-readable medium carries one or more programs, which, when executed, implement the method as described in the embodiments of this application.
[0091] According to embodiments of this application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may 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, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may 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 application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may 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, wired, optical fiber, radio frequency signals, etc., or any suitable combination thereof.
[0092] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments, but should be defined not only by the appended claims, but also by their equivalents. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. An optimized method for the production process of a carbon fiber precursor washing tank, characterized in that, include: Obtain production demand data and historical production data of the carbon fiber precursor washing tank production system, preprocess the historical production data, and obtain sample process data and label DMSO quality score. Based on production demand data, construct objective and decision functions, and then construct an initial sparse LS-SVM model based on the objective and decision functions. Using sample process data and label DMSO quality scores, the objective function and decision function of the initial sparse LS-SVM model are iteratively optimized using the improved NOA algorithm to obtain the final sparse LS-SVM model. The process data to be tested in the carbon fiber precursor washing tank production system is obtained. The process data to be tested is input into the final sparse LS-SVM model to predict the DMSO mass fraction and obtain the final DMSO mass fraction. Based on the final DMSO mass fraction, the production process of the carbon fiber precursor washing tank is optimized.
2. The optimized method for the production process of the carbon fiber precursor washing tank according to claim 1, characterized in that, The preprocessing of historical production data to obtain sample process data and label DMSO quality scores includes: Obtain historical process data from historical production data, as well as the corresponding historical DMSO quality scores; Historical process data and historical DMSO quality scores were cleaned and normalized to obtain sample process data and labeled DMSO quality scores.
3. The method for optimizing the production process of the carbon fiber precursor washing tank according to claim 1, characterized in that: The production demand data includes prediction accuracy demand data, computational efficiency demand data, washing process constraint data, local fitting demand data, and global fitting demand data.
4. The optimized production process of the carbon fiber precursor washing tank according to claim 3, characterized in that, The process of constructing the objective function includes: Based on the prediction accuracy requirement data, set the weight vector, error penalty coefficient and regression error term; based on the computational efficiency requirement data, set the L1 sparsity penalty coefficient and L2 regularization penalty coefficient to obtain the importance coefficient of the sample process data. The first target value is obtained by calculating the weight vector, the second target value is obtained by calculating the error penalty coefficient and the regression error term, the third target value is obtained by calculating the L1 sparsity penalty coefficient and the importance coefficient, and the fourth target value is obtained by calculating the L2 regularization penalty coefficient and the importance coefficient. Add the first target value, the second target value, the third target value, and the fourth target value to obtain the total target value; Based on the constraints of the water washing process, set the constraints, and under the constraints, minimize the total objective value as the objective function.
5. The optimized production process of the carbon fiber precursor washing tank according to claim 3, characterized in that, The process of constructing the decision function includes: Set the RBF kernel width based on the local fitting requirements data, and set the polynomial kernel offset, polynomial order, and kernel function weighting coefficients based on the global fitting requirements data. Obtain key support vectors from the sample process data, construct an RBF kernel function based on the sample process data, key support vectors, and RBF kernel width, and construct a polynomial kernel function based on the sample process data, key support vectors, polynomial kernel offset, and polynomial order. Decision functions are constructed based on the kernel function weighting coefficients, RBF kernel function, and polynomial kernel function.
6. The optimized method for the production process of the carbon fiber precursor washing tank according to claim 1, characterized in that, The process involves iteratively optimizing the objective and decision functions of the initial sparse LS-SVM model using sample process data and DMSO quality scores, employing an improved NOA algorithm to obtain the final sparse LS-SVM model, including: By improving the NOA algorithm, the error penalty coefficient, L1 sparsity penalty coefficient, and L2 regularization penalty coefficient of the objective function of the initial sparse LS-SVM model, as well as the RBF kernel width, polynomial kernel offset, polynomial order, and kernel function weighting coefficient of the decision function, are iteratively optimized to obtain an optimized sparse LS-SVM model. The sample process data is input into the optimized sparse LS-SVM model to predict the DMSO quality score, and the predicted DMSO quality score is obtained. Obtain the number of key support vectors for predicting DMSO quality scores, as well as the total number of samples in the sample process data; The optimal fitness value is calculated based on the predicted DMSO quality score, the labeled DMSO quality score, the number of key support vectors, and the total number of samples. The initial sparse LS-SVM model was optimized multiple times, and the optimized sparse LS-SVM model with the smallest fitness value was taken as the final sparse LS-SVM model.
7. The optimized production process of the carbon fiber precursor washing tank according to claim 6, characterized in that, The improved NOA algorithm iteratively optimizes the error penalty coefficient, L1 sparsity penalty coefficient, and L2 regularization penalty coefficient of the objective function of the initial sparse LS-SVM model, as well as the RBF kernel width, polynomial kernel offset, polynomial order, and kernel function weighting coefficients of the decision function, to obtain an optimized sparse LS-SVM model, including: A set of error penalty coefficients, L1 sparsity penalty coefficients, L2 regularization penalty coefficients, RBF kernel width, polynomial kernel offset, polynomial order, and kernel function weighting coefficients are used as the positions of individual star crows to construct a star crow population. Randomly perturb the Star Crow population and update the position of each Star Crow individual; The root mean square error, constraint violation degree, and dynamic adjustment factor of the individual Star Raven are calculated based on the updated position. The comprehensive fitness value is then calculated based on the root mean square error, constraint violation degree, and dynamic adjustment factor. The population of ravens was subjected to multiple random perturbations. The position with the lowest comprehensive fitness value was taken as the optimal position. The parameters of the initial sparse LS-SVM model were configured according to the optimal position to obtain the optimized sparse LS-SVM model.
8. A device for optimizing the production process of a carbon fiber precursor washing tank, used to implement the method for optimizing the production process of a carbon fiber precursor washing tank as described in any one of claims 1 to 7, characterized in that, The device includes: The data acquisition module is used to acquire production demand data and historical production data of the carbon fiber precursor washing tank production system, preprocess the historical production data, and obtain sample process data and label DMSO quality score. The sparse LS-SVM model building module is used to construct the objective function and decision function based on production demand data, and to construct the initial sparse LS-SVM model based on the objective function and decision function. The sparse LS-SVM model training module is used to iteratively optimize the objective function and decision function of the initial sparse LS-SVM model using sample process data and label DMSO quality scores, and obtain the final sparse LS-SVM model. The parameter optimization module is used to acquire the process data to be tested in the carbon fiber precursor washing tank production system. The process data to be tested is input into the final sparse LS-SVM model to predict the DMSO mass fraction, obtain the final DMSO mass fraction, and optimize the production process of the carbon fiber precursor washing tank based on the final DMSO mass fraction.
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 carbon fiber precursor washing tank production process optimization method as described in any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the carbon fiber precursor washing tank production process optimization method as described in any one of claims 1 to 7.