Soft-sensing model of oil absorption value of glassine paper based on SA-PSO-LSSVM

Abstract

This invention addresses the issue that while there is a corresponding relationship between the components and oil absorption value in the glassine paper coating process, this relationship is non-linear and cannot be easily predicted. Therefore, it provides a soft measurement model based on SA-PSO-LSSVM to predict the oil absorption value of glassine paper. This model enables accurate detection of the oil absorption value in the glassine paper coating process, providing a quick method for optimizing the glassine paper coating process. It uses real daily production data from paper mills and is applicable to the actual production activities of most paper mills.

Description

Technical Field

[0001] This invention belongs to the field of coating process technology for paper manufacturing enterprises, specifically involving a soft measurement model of oil absorption value of glassine paper based on SA-PSO-LSSVM. Background Technology

[0002] Glassine paper, also known as translucent glass backing paper, is a type of paper made from glassine base paper that has been coated to achieve a certain level of transparency, strength, and special oil resistance. The coating process of the glassine base paper determines the level of oil resistance of the glassine paper. Due to the high cost of traditional glassine paper, manufacturers are committed to finding low-cost, high-performance coating processes. Since the influence of each component in the coating process on the oil absorption value varies and they can interact with each other, it is particularly important to study the relationship between each coating component and the performance indicators of glassine paper. Among them, the oil absorption value is a key indicator for measuring the quality of glassine paper. The traditional oil absorption value is measured after the paper processing steps are completed, which has a certain time delay. Therefore, rapid and forward-looking oil absorption value measurement is of great significance for the production of glassine paper and its process adjustment.

[0003] Therefore, the technical problem to be solved by this invention is: there is a corresponding relationship between each component and the oil absorption value in the glassine paper coating process, but since the relationship is non-linear, it is not possible to simply infer and predict its value. Existing technologies cannot achieve accurate detection of the oil absorption value in the glassine paper coating process, and this invention provides a quick method for optimizing the glassine paper coating process. Summary of the Invention

[0004] To address the technical problem that there is a corresponding relationship between the components and the oil absorption value in the glassine paper coating process, but because the relationship is non-linear, it is not possible to simply infer and predict the value. Existing technologies cannot achieve accurate detection of the oil absorption value in the glassine paper coating process, and to provide a quick method for optimizing the glassine paper coating process, this invention provides a soft measurement model for the oil absorption value of glassine paper based on SA-PSO-LSSVM.

[0005] The specific plan is as follows:

[0006] A soft measurement model for the oil absorption value of glassine paper based on SA-PSO-LSSVM

[0007] The establishment method includes the following steps:

[0008] S1: In daily production or testing, select oil absorption value data corresponding to different glassine paper processes, use four components in the glassine paper coating process that affect the oil absorption value as input, and use the oil absorption value as the output of the model.

[0009] S2: Data processing: including outlier removal, random error elimination, data standardization, and nonlinear partial least squares (PLSR) simplification of data structure;

[0010] S3: Divide the data processed in S2 into two groups: one group is used as training data and the other group is used as test data. Build an oil absorption value prediction model based on LSSVM, use the root mean square error (RMSE) to reflect the accuracy of the model, and output the oil absorption value data of the LSSVM oil absorption value prediction model.

[0011] S4: Use the PSO algorithm to find the optimal combination of regularization parameters and kernel function of the LSSVM oil absorption value prediction model, so that it reaches the global optimum and outputs the oil absorption value data of the PSO-LSSVM oil absorption value prediction model.

[0012] S5: The PSO-LSSVM oil absorption value prediction model was established using the simulated annealing algorithm (SA). The simulated annealing algorithm adopted the Metropolis criterion. The accuracy of the SA-PSO-LSSVM oil absorption value prediction model was tested using training data and test data respectively. The oil absorption value data of the SA-PSO-LSSVM oil absorption value prediction model was output, and the prediction accuracy of different algorithms was analyzed and compared.

[0013] Step S2 includes:

[0014] S201: Outliers exist in the extracted production or experimental data. The normality test is used to remove outliers from historical data, and the result is calculated as follows: Where x is the data to be detected, μ is the sample mean, σ is the standard deviation of the sample, and points that do not satisfy this formula are outliers.

[0015] S202: Extracted daily production or experimental data undergoes smoothing to eliminate random errors caused by the environment. A five-point cubic smoothing method is used to eliminate random errors caused by the environment.

[0016]

[0017]

[0018]

[0019]

[0020]

[0021] The smooth polynomial is shown in the above equation, Y i Y represents consecutive points with equal spacing. i * is Y iThe improved values ​​are calculated symmetrically at both ends of the data points using (1)(2)(4)(5) and the middle data points using formula (3). The data after removing outliers using the normal distribution test is then processed using the five-point cubic smoothing method to eliminate random errors.

[0022] S203: The four glassine paper coating process components selected have different dimensions and orders of magnitude, and the levels between the variables are too different. The data will be scaled proportionally to a certain specified interval, and the data will be standardized. The Min-Max normalization method will be used to unify the dimensions. The normalization function mapminmax in MATLAB will be used to normalize to [0,1], [x*,PS]=mapminmax(x,0,1), where x* is the normalized variable; x is the variable before normalization; PS is the mapping form of the training data;

[0023] S204: The PLSR method simplifies the data structure and its effectiveness in explaining independent and dependent variables:

[0024]

[0025]

[0026] Variable projection analysis:

[0027]

[0028] Where x is the independent variable, y is the dependent variable, and t h It is a principal component, p is the number of independent variables, Rd is the explanatory power, and Rd(y:t) is the explanatory power. i ) = r 2 (y:t i The explanatory power of a component on the dependent variable is the ability of the independent variable to explain the dependent variable. It is the k-th component of the weight vector, VIP k As an indicator for selecting variables, the larger the value of a variable, the greater its role in explaining the dependent variable; the smaller the value of a variable, the weaker its explanatory power on the dependent variable, that is, the smaller its influence.

[0029] Step S3 includes:

[0030] S301: Divide the data processed in step S2 into a training set and a test set. The training set includes 41 sets of sample data, and the test set includes 14 sets of sample data.

[0031] S302: A least squares support vector machine (LSSVM) model for predicting the oil absorption value of glassine paper is established. The root mean square error (RMSE) is used to express the model accuracy. Appropriate regularization parameters γ and kernel functions σ are selected to maximize the model's accuracy.

[0032]

[0033] In the above formula, y represents the predicted value from the LSSVM model. i is the measured value, and n is the sample size.

[0034] Step S4 includes:

[0035] S401: Initialize a group of particles with a population size of N, and initialize their positions and velocities;

[0036] S402: Calculate the initial fitness value for each particle; record the optimal position of the individual and the optimal position of the population;

[0037] S403: Update speed and location;

[0038] V i =ω*V i +c1*r1*(pBest[i]-X i )+c2*r2*(gBest[i]-X i )

[0039] X i =X i +V i

[0040] Among them, V i Let ω be the velocity vector, c1 and c2 be the learning factors, r1 and r2 be random numbers uniformly distributed between 0 and 1, pBest be the individual extreme value, gBest be the global extreme value, and X be the position vector.

[0041] S404: Compare the current fitness value with the previous fitness value to check whether the minimum error or the maximum number of iterations has been reached;

[0042] S405: If the maximum number of iterations is reached, extract the optimized LSSVM parameters and establish a PSO-LSSVM oil absorption value prediction model.

[0043] S406: If not achieved, return to step S402;

[0044] S407: Train the training set data using the obtained PSO-LSSVM oil absorption value prediction model;

[0045] S408: Predict the oil absorption value of the trained PSO-LSSVM oil absorption value using the test set data;

[0046] S409: Output the prediction results and errors for the training and test sets.

[0047] Step S5 includes:

[0048] S501: Set the maximum number of iterations, initial temperature T, initial values ​​of parameters C1 and C2 of the particle swarm optimization algorithm and the search range, randomly generate the population and initialize the velocity;

[0049] S502: Calculate fitness value;

[0050] S503: Calculate the predicted values ​​for the training set and the prediction set, denormalize the predicted data, and calculate the mean square error.

[0051] S504: Obtain the individual optimal value and the population optimal value, and update the current position and velocity of the particles;

[0052] S505: Calculate the objective value of the function in the neighborhood b near the given neighborhood a, and calculate the fitness values ​​f(a) and f(b);

[0053] S506: Calculate whether Δf = f(b) - f(a) is less than or equal to 0. If yes, accept the new solution; if no, determine whether it meets the Metropolis criterion. If yes, accept the new solution; if no, update the iteration count and return to step S505.

[0054] S507: Check if the final termination condition is met. If yes, terminate and output the optimal solution; otherwise, perform annealing and return to step S502.

[0055] The beneficial effects of this invention are as follows:

[0056] (1) The soft measurement modeling data of the present invention comes from the daily production data of real paper manufacturing enterprises and is applicable to the actual production activities of general paper manufacturing enterprises.

[0057] (2) This invention uses the normal distribution test to remove outliers, the five-point cubic smoothing method to eliminate random errors caused by environmental influences, and the Min-Max normalization method to normalize the data to a uniform range of values ​​to eliminate the influence of dimensions. Through multiple means, the error influence caused by the modeling data is reduced, and the oil absorption value prediction is more accurate.

[0058] (3) The modeling method selected in this invention is based on LSSVM, which is suitable for small sample data; the PLSR method is used to simplify the data structure of the model and eliminate multiple correlations between variables.

[0059] (4) This invention uses the SA algorithm to optimize the PSO algorithm and introduces the Metropolis criterion into the PSO algorithm, which can improve the particle diversity of the particle swarm, avoid local optima, and increase the convergence accuracy of the algorithm.

[0060] (5) In this invention, the SA algorithm optimizes the PSO-LSSVM oil absorption value prediction model and selects test sample data from actual production. The root mean square error (RMSE) between the predicted oil absorption value and the actual oil absorption value is extremely small. Attached Figure Description

[0061] Figure 1 This is a flowchart of the soft measurement model for predicting the oil absorption value of glassine paper, based on SA-PSO-LSSVM, as described in this invention.

[0062] Figure 2 This is a schematic diagram of the prediction results of the LSSVM oil absorption value prediction model on the training set data.

[0063] Figure 3 This is a schematic diagram of the prediction results of the LSSVM oil absorption value prediction model on the test set data.

[0064] Figure 4 This is a schematic diagram of the optimization process of the PSO-LSSVM oil absorption value prediction model.

[0065] Figure 5 This is a schematic diagram of the prediction results of the PSO-LSSVM oil absorption value prediction model on the training set data.

[0066] Figure 6 This is a schematic diagram of the prediction results of the PSO-LSSVM oil absorption value prediction model on the test set data.

[0067] Figure 7 This is a flowchart of the SA algorithm improving the PSO algorithm.

[0068] Figure 8 This is a schematic diagram illustrating the prediction results of the SA algorithm-improved PSO-LSSVM oil absorption value prediction model on the training set data.

[0069] Figure 9 This is a schematic diagram illustrating the prediction results of the SA algorithm-improved PSO-LSSVM oil absorption value prediction model on the test set data. Detailed Implementation

[0070] The technical solutions in the embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the implementation of the present invention, and not all of it. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0071] like Figure 1As shown, a soft sensor model for predicting the oil absorption value of glassine paper, based on SA-PSO-LSSVM, is applied to predict the oil absorption value in the glassine paper production process in paper manufacturing enterprises. The establishment method includes the following steps:

[0072] S1: In daily production or testing, select oil absorption value data corresponding to different glassine paper processes, use four components in the glassine paper coating process that affect the oil absorption value as input, and use the oil absorption value as the output of the model.

[0073] S2: Data processing: including outlier removal, random error elimination, data standardization, and nonlinear partial least squares (PLSR) simplification of data structure;

[0074] S3: Divide the data processed in S2 into two groups: one group is used as training data and the other group is used as test data. Build an oil absorption value prediction model based on LSSVM, use the root mean square error (RMSE) to reflect the accuracy of the model, and output the oil absorption value data of the LSSVM oil absorption value prediction model.

[0075] S4: Use the PSO algorithm to find the optimal combination of regularization parameter γ and kernel function σ of the LSSVM oil absorption value prediction model, so that it reaches the global optimum and outputs the oil absorption value data of the PSO-LSSVM oil absorption value prediction model.

[0076] S5: The PSO-LSSVM oil absorption value prediction model was established using the simulated annealing algorithm (SA). The simulated annealing algorithm adopted the Metropolis criterion. The accuracy of the SA-PSO-LSSVM oil absorption value prediction model was tested using training data and test data respectively. The oil absorption value data of the SA-PSO-LSSVM oil absorption value prediction model was output, and the prediction accuracy of different algorithms was analyzed and compared.

[0077] Step S2 includes:

[0078] S201: Outliers exist in the extracted production or test data. These outliers are mainly caused by machine equipment factors and will affect the next step of data analysis and modeling. They need to be removed. Based on mathematical statistics, we can control the overall characteristics of the sample and find data points that deviate from most samples. This invention selects the normal distribution test to remove outliers in historical data.

[0079] The basis of the normality test is that the data as a whole falls within a range of 3 standard deviations, with a probability of 99.76%. Other cases are considered outliers as they do not conform to the overall distribution of the sample.

[0080] Where x is the data to be detected, μ is the sample mean, σ is the sample standard deviation, and points that do not satisfy this formula are outliers.

[0081] S202: In addition to the influence of machinery and equipment, another important factor affecting the extracted daily production or test data is the change of environment. Therefore, the data needs to be smoothed to eliminate random errors caused by the environment. This invention uses the five-point cubic smoothing method to eliminate random errors caused by the environment. The principle of the five-point cubic smoothing method is: when the number of nodes is greater than or equal to 5, the data of 5 consecutive reading points with equal spacing is smoothed using a cubic least squares polynomial.

[0082]

[0083]

[0084]

[0085]

[0086]

[0087] The smooth polynomial is shown in the above equation, Y i Y represents consecutive points with equal spacing. i * is Y i The improved values ​​are calculated symmetrically at both ends of the data points using (1)(2)(4)(5) and at the middle data points using (3). The data after removing outliers using the normal distribution test is then processed using the five-point cubic smoothing method to eliminate random errors.

[0088] S203: The four glassine paper coating process components selected have different dimensions and orders of magnitude, and the levels of the variables differ too much. If historical data is used directly to build the model, the effect of variables with higher values ​​may be very obvious in the comprehensive analysis, which may relatively weaken the effect of variables with lower values. In order to ensure that the data results are more reliable, the original data is standardized, that is, the data is scaled proportionally to fit within a certain specified range. This invention adopts the Min-Max normalization method to unify the dimensions and uses the normalization function mapminmax in MATLAB to normalize to [0,1].

[0089] [x*,PS] = mapminmax(x,0,1),

[0090] Where x* is the normalized variable; x is the variable before normalization; and PS is the mapping form of the training data.

[0091] S204: The PLSR method simplifies the data structure and its effectiveness in explaining independent and dependent variables:

[0092]

[0093]

[0094] Variable projection analysis:

[0095]

[0096] Where x is the independent variable, y is the dependent variable, and t h It is a principal component, p is the number of independent variables, Rd is the explanatory power, and Rd(y:t) is the explanatory power. i ) = r 2 (y:t i The explanatory power of a component on the dependent variable is the ability of the independent variable to explain the dependent variable. It is the k-th component of the weight vector, VIP k As an indicator for selecting variables, the larger the value of a variable, the greater its role in explaining the dependent variable; the smaller the value of a variable, the weaker its explanatory power on the dependent variable, that is, the smaller its influence.

[0097] Step S3 includes:

[0098] S301: Divide the data processed in step S2 into a training set and a test set. The training set includes 41 sets of sample data, and the test set includes 14 sets of sample data.

[0099] S302: The least squares support vector machine (LSSVM) is used to establish a prediction model for the oil absorption value of glassine paper. The root mean square error (RMSE) is used to express the accuracy of the model. Appropriate regularization parameters γ and kernel functions σ are selected to maximize the accuracy of the model.

[0100]

[0101] In the above formula, y represents the predicted value from the LSSVM model. i is the measured value, and n is the sample size.

[0102] Least Squares Support Vector Machine (LSSVM) improves its generalization ability by controlling the optimization objective to minimize structured risk rather than empirical risk. Compared to Support Vector Machine (SVM), LSSVM requires fewer parameters, replaces the objective function inequality constraint with an equality constraint, and uses the sum of squared errors instead of the hinge loss function. Therefore, LSSVM is developed from SVM and was initially used for classification problems. Later, it is frequently used for nonlinear fitting and is suitable for solving problems with small sample data. When solving nonlinear function problems, LSSVM provides a certain improvement in the generalization ability and accuracy of the prediction model compared to SVM.

[0103] This invention uses Least Squares Support Vector Machine (LSSVM) to establish an oil absorption value prediction model for sample data after cleaning treatment, wherein... Figure 2 This chart compares the predicted oil absorption values ​​of the LSSVM oil absorption value prediction model with the actual values ​​on the training set sample data. The root mean square error (RMSE) is 0.081394. Figure 3 This is a comparison chart of the predicted oil absorption values ​​of the LSSVM oil absorption value prediction model and the actual values ​​of the test set sample data. The root mean square error (RMSE) is 0.123320.

[0104] S4: The PSO algorithm is used to find the optimal combination of regularization parameter γ and kernel function σ in the LSSVM oil absorption value prediction model to achieve the global optimum. The optimization steps are as follows:

[0105] S401: Initialize a group of particles with a population size of N, and initialize their positions and velocities;

[0106] S402: Calculate the initial fitness value for each particle; record the optimal position of the individual and the optimal position of the population;

[0107] S403: Update speed and location;

[0108] V i =ω*V i +c1*r1*(pBest[i]-X i )+c2*r2*(gBest[i]-X i )

[0109] X i =X i +V i

[0110] Among them, V i Let X be the velocity vector, ω be the inertia weight coefficient, c1 and c2 be the learning factors, r1 and r2 be uniformly distributed random numbers between 0 and 1, pBest be the individual extreme value, gBest be the global extreme value, and X be the velocity vector. i This is a position vector.

[0111] S404: Compare the current fitness value with the previous fitness value to check whether the minimum error or the maximum number of iterations has been reached;

[0112] S405: If the maximum number of iterations is reached, extract the optimized LSSVM parameters and establish a PSO-LSSVM oil absorption value prediction model.

[0113] S406: If not achieved, return to step S3032;

[0114] S407: Train the training set data using the obtained PSO-LSSVM oil absorption value prediction model;

[0115] S408: Predict the oil absorption value of the trained PSO-LSSVM oil absorption value using the test set data;

[0116] S409: Output the prediction results and errors for the training and test sets.

[0117] in, Figure 4 The diagram shows the optimization process of the PSO-LSSVM oil absorption value prediction model. Figure 5 The PSO-LSSVM oil absorption value prediction model predicts the oil absorption value on the training set data, with a root mean square error (RMSE) of 0.071472. Figure 6 The PSO-LSSVM oil absorption value prediction model predicts the test set data, with a root mean square error (RMSE) of 0.094937.

[0118] like Figure 7 As shown in Figure S5: The PSO-LSSVM oil absorption value prediction model optimized using the simulated annealing algorithm (SA). The simulated annealing algorithm adopts the Metropolis criterion. The accuracy of the SA-PSO-LSSVM model is tested using training and test data respectively, and the prediction accuracy of different algorithms is analyzed and compared. The following are the specific steps of the SA algorithm to improve the accuracy of the PSO-LSSVM prediction model:

[0119] S501: Set the maximum number of iterations, initial temperature T, initial values ​​of parameters C1 and C2 of the particle swarm optimization algorithm and the search range, randomly generate the population and initialize the velocity;

[0120] S502: Calculate fitness value;

[0121] S503: Calculate the predicted values ​​for the training set and the prediction set, denormalize the predicted data, and calculate the mean square error.

[0122] S504: Obtain the individual optimal value and the population optimal value, and update the current position and velocity of the particles;

[0123] S505: Calculate the objective value of the function in the neighborhood b near the given neighborhood a, and calculate the fitness values ​​f(a) and f(b);

[0124] S506: Calculate whether Δf = f(a) - f(b) is less than or equal to 0. If yes, accept the new solution; if no, determine whether it meets the Metropolis criterion. If yes, accept the new solution; if no, update the iteration count and return to step S505.

[0125] S507: Check if the final termination condition is met. If yes, terminate and output the optimal solution; otherwise, perform annealing and return to step S502.

[0126] in, Figure 8The SA algorithm was used to improve the prediction results of the PSO-LSSVM oil absorption value prediction model on the training set data. The root mean square error (RMSE) was 0.017607. Figure 9 The root mean square error (RMSE) of the improved PSO-LSSVM oil absorption value prediction model based on the SA algorithm for the test set data is 0.04744.

[0127] RMSE value LSSVM model PSO-LSSVM SA-PSO-LSSVM model training set 0.081394 0.071472 0.017607 test set 0.123320 0.094937 0.04744

[0128] The table above shows the root mean square error (RMSE) values ​​of three oil absorption value prediction models. In the glassine paper coating process, the relationship between each component and the oil absorption value is non-linear, making it impossible to simply infer and predict the value. Therefore, this invention proposes a soft measurement model based on SA-PSO-LSSVM for predicting the oil absorption value of glassine paper. This model is applied to the prediction of oil absorption value in the glassine paper production process of paper mills. The LSSVM, PSO-LSSVM, and SA-PSO-LSSVM oil absorption value prediction models are used to predict the selected sample data. Comparison reveals that the SA-PSO-LSSVM oil absorption value prediction model has the smallest RMSE, the highest prediction accuracy, and is convenient and fast.

[0129] The technical means disclosed in this invention are not limited to those disclosed in the above embodiments, but also include technical solutions composed of any combination of the above technical features. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications are also considered within the scope of protection of this invention.

Claims

1. A soft measurement model for the oil absorption value of glassine paper based on SA-PSO-LSSVM, characterized in that: The establishment method includes the following steps: S1: In daily production or testing, select oil absorption value data corresponding to different glassine paper processes, use four components in the glassine paper coating process that affect the oil absorption value as input, and use the oil absorption value as the output of the model. S2: Data processing: including outlier removal, random error elimination, data standardization, and nonlinear partial least squares (PLSR) simplification of data structure; S3: Divide the data processed in S2 into two groups: one group is used as training data and the other group is used as test data. Build an oil absorption value prediction model based on LSSVM, use the root mean square error (RMSE) to reflect the accuracy of the model, and output the oil absorption value data of the LSSVM oil absorption value prediction model. S4: Use the PSO algorithm to find the optimal combination of regularization parameters and kernel function of the LSSVM oil absorption value prediction model, so that it reaches the global optimum and outputs the oil absorption value data of the PSO-LSSVM oil absorption value prediction model. S5: The PSO-LSSVM oil absorption value prediction model was established using the simulated annealing algorithm (SA). The simulated annealing algorithm adopted the Metropolis criterion. The accuracy of the SA-PSO-LSSVM oil absorption value prediction model was tested using training data and test data respectively. The oil absorption value data of the SA-PSO-LSSVM oil absorption value prediction model was output, and the prediction accuracy of different algorithms was analyzed and compared. Step S2 includes: S201: Outliers exist in the extracted production or experimental data. The normality test is used to remove outliers from historical data, and the result is calculated as follows: , where x is the data to be detected, μ is the sample mean, σ is the standard deviation of the sample, and points that do not satisfy this formula are outliers; S202: Extracted daily production or experimental data undergoes smoothing to eliminate random errors caused by the environment. A five-point cubic smoothing method is used to eliminate random errors caused by the environment. ； The smooth polynomial is shown in the above equation, Y i Y represents consecutive points at equal intervals. i * is Y i The improved value is calculated symmetrically at both ends of the data point using (1)(2)(4)(5), and the middle data point is calculated using formula (3). The data after removing outliers by the normal distribution test is then processed by the five-point cubic smoothing method to eliminate random errors. S203: The four glassine paper coating process components selected have different dimensions and orders of magnitude, and the levels between the variables differ too much. The data will be scaled proportionally to a specified interval, and standardized. The Min-Max normalization method will be used to unify the dimensions, and the normalization function mapminmax in MATLAB will be used to normalize the data to [0,1]. [ x *, PS =mapminmax(x,0,1), where x * represents the normalized variable, x represents the variable before normalization, and PS represents the mapping form of the training data; S204: The PLSR method simplifies the data structure and its effectiveness in explaining independent and dependent variables: ； Variable projection analysis: ； Where x is the independent variable and y is the dependent variable. t h It is a principal component, and p is the number of independent variables. Rd It is explanatory ability. Rd ( y : t i )= r 2 ( y : t i ω represents the explanatory power of the independent variable over the dependent variable. 2 ik It is the k-th component of the weight vector. VIP k As an indicator for selecting variables, the larger the value of a variable, the greater its role in explaining the dependent variable; the smaller the value of a variable, the weaker its explanatory power on the dependent variable, that is, the smaller its influence.

2. The soft measurement model for oil absorption value of glassine paper based on SA-PSO-LSSVM as described in claim 1, characterized in that: Step S3 includes: S301: Divide the data processed in step S2 into a training set and a test set. The training set includes 41 sets of sample data, and the test set includes 14 sets of sample data. S302: A least squares support vector machine (LSSVM) model for predicting the oil absorption value of glassine paper is established. The root mean square error (RMSE) is used to express the model accuracy. Appropriate regularization parameters γ and kernel functions σ are selected to maximize the model's accuracy. ； In the above formula, ŷ i These are the predicted values ​​from the LSSVM model. y i These are measured values. n This represents the number of samples.

3. The soft measurement model for oil absorption value of glassine paper based on SA-PSO-LSSVM as described in claim 1, characterized in that: Step S4 includes: S401: Initialize a group of particles with a population size of N, and initialize their positions and velocities; S402: Calculate the initial fitness value for each particle; record the optimal position of the individual and the optimal position of the population; S403: Update speed and location; ； Among them, V i Let ω be the velocity vector, c1 and c2 be the learning factors, r1 and r2 be random numbers uniformly distributed between 0 and 1, pBest be the individual extreme value, gBest be the global extreme value, and X be the position vector. S404: Compare the current fitness value with the previous fitness value to check whether the minimum error or the maximum number of iterations has been reached; S405: If the maximum number of iterations is reached, extract the optimized LSSVM parameters and establish a PSO-LSSVM oil absorption value prediction model. S406: If not achieved, return to step S402; S407: Train the training set data using the obtained PSO-LSSVM oil absorption value prediction model; S408: Predict the oil absorption value of the trained PSO-LSSVM oil absorption value using the test set data; S409: Output the prediction results and errors for the training and test sets.

4. The soft measurement model for oil absorption value of glassine paper based on SA-PSO-LSSVM as described in claim 1, characterized in that: Step S5 includes: S501: Set the maximum number of iterations, initial temperature T, initial values ​​of parameters C1 and C2 of the particle swarm optimization algorithm and the search range, randomly generate the population and initialize the velocity; S502: Calculate fitness value; S503: Calculate the predicted values ​​for the training set and the prediction set, denormalize the predicted data, and calculate the mean square error. S504: Obtain the individual optimal value and the population optimal value, and update the current position and velocity of the particles; S505: Calculate the objective value of the function in the neighborhood b near the given neighborhood a; calculate the fitness values ​​f(a) and f(b). S506: Calculate whether ∆f=f(b)-f(a) is less than or equal to 0. If yes, accept the new solution; if no, determine whether it meets the Metropolis criterion. If yes, accept the new solution; if no, update the iteration count and return to step S505. S507: Check if the final termination condition is met. If yes, terminate and output the optimal solution; otherwise, perform annealing and return to step S502.

Images