Optimization and Adjustment Method of High-Speed ​​Fitting Adhesive Viscosity and Solid Content Based on BO-RF-JITL

Abstract

This invention belongs to the field of process optimization for high-speed cigarette filter adhesive production, specifically involving a method for optimizing and adjusting the viscosity and solid content of high-speed cigarette filter adhesive based on BO-RF-JITL. First, historical production data of the high-speed cigarette filter adhesive is preprocessed to extract features related to viscosity and solid content. Correlation analysis and feature importance analysis are used to select features with a significant impact on viscosity and solid content, constructing a feature vector. Then, the RF algorithm is used to train a model based on the preprocessed data, and the BO algorithm is used to fine-tune the hyperparameters, establishing a mapping relationship between viscosity and solid content and the feature vector, and evaluating the model's predictive performance. Finally, the JITL algorithm is used to find the optimal adjustment amount of the process formulation based on the set target viscosity and solid content to achieve the production target of the high-speed cigarette filter adhesive. This solution can effectively improve the control accuracy of viscosity and solid content in the production of high-speed cigarette filter adhesive, reduce production costs, and increase production efficiency.

Description

Technical Field

[0001] This invention belongs to the field of production process optimization technology for high-speed cigarette filter adhesive, specifically involving a method for optimizing and adjusting the viscosity and solid content of high-speed cigarette filter adhesive based on BO-RF-JITL. Background Technology

[0002] High-speed filter tip adhesive is a key material used in cigarette production to connect the filter tip and the tobacco body. Its performance directly affects cigarette production efficiency, product quality, and consumer experience. In cigarette production, the viscosity and solids content of the filter tip adhesive are critical quality parameters. Viscosity determines the fluidity of the adhesive, affecting its coating and bonding effects. Excessively high viscosity may make the adhesive difficult to apply, while excessively low viscosity may result in too much adhesive being applied to unwanted areas, affecting the strength and appearance of the filter tip. Solids content affects the drying rate and film-forming effect of the adhesive. Too low a solids content leads to incomplete curing, while too high a content may result in poor uniformity during coating, thus affecting the filter tip's performance.

[0003] In high-speed production processes, the adhesive for splices must cure within an extremely short time, thus requiring extremely precise control over its viscosity and solids content. Traditional adhesive formulation adjustment methods rely primarily on manual experience and experimental adjustments. Production personnel typically adjust the adhesive ratio based on historical experience to meet the needs of different production conditions. However, when the production line is running at high speed, manual experience cannot ensure that each formulation adjustment accurately meets the desired adhesive performance; repeated trials and manual adjustments are not only time-consuming but also lead to significant waste in the production process. Each adjustment requires time to evaluate the results, and the production process often cannot provide real-time feedback, nor can it dynamically monitor and adjust the adhesive performance in real time.

[0004] In recent years, machine learning has achieved remarkable results in industrial production, particularly in formula optimization, process control, and quality prediction. However, in cigarette production, practical applications still face several challenges, especially in dynamically adjusting production parameters: real-time performance and computational complexity limit the model's rapid response; insufficient data quality and real-time updates affect prediction accuracy; the nonlinearity and dynamic changes in the production process are difficult for static models to adapt to in a timely manner; insufficient model interpretability and decision support limit practical application; and environmental changes and unforeseen events increase the difficulty of adjustment. These problems need to be addressed by optimizing algorithms, improving data quality, and building decision support systems. Summary of the Invention

[0005] To address the limitations of existing traditional adjustment methods, such as the lack of real-time feedback and dynamic adjustment mechanisms, this invention proposes a high-speed nozzle adhesive viscosity and solids content optimization adjustment method based on BO-RF-JITL. By combining machine learning algorithms with a real-time feedback mechanism, the key parameters in the adhesive formulation can be dynamically adjusted, thereby optimizing the adhesive performance during the production process.

[0006] This invention is achieved using the following technical solution: a method for optimizing and adjusting the viscosity and solid content of high-speed nozzle adhesive based on BO-RF-JITL, comprising the following steps:

[0007] Step A: Preprocess historical production data of high-speed connector adhesive:

[0008] Acquire the dataset, examine the historical production data of high-speed nozzle adhesive, remove or correct outliers and missing values ​​to ensure data quality and integrity; extract features related to viscosity and solid content from the raw data, and use Z-Score standardization to standardize the numerical features and eliminate the influence of units.

[0009] Step B, Feature Engineering:

[0010] Pearson correlation analysis was used to calculate the Pearson correlation coefficient between features and target variables (viscosity and solid content) to assess the influence of features on the target variables. The random forest algorithm in machine learning was used to calculate the contribution of each feature to the model prediction results to assess the importance of the features. Based on the results of feature importance analysis and correlation analysis, features that contribute significantly to the model prediction results and have high correlation were selected, while features that contribute little to the model prediction results and are irrelevant to the target variables were removed.

[0011] Step C: Train the random forest prediction model:

[0012] A random forest regression (RF) algorithm was used to build a predictive model for viscosity and solids content. By constructing multiple decision trees and combining their outputs, the complex nonlinear relationships in the data were effectively captured, and the risk of overfitting was reduced. During training, the hyperparameters of the model were tuned using the Bayesian optimization (BO) algorithm, including the number of trees, the maximum depth of the trees, and the minimum number of samples per node, to obtain the optimal model parameters. Mean squared error (MSE) and relative error were used as evaluation metrics to verify the predictive performance of the model and ensure that it could accurately reflect the relationship between viscosity and solids content and each input feature. During model training and validation, feature selection and model parameters were continuously optimized to improve prediction accuracy and stability.

[0013] Step D: Establish an instant learning optimization and adjustment model and output the solution results:

[0014] First, based on the dataset in the database, the initial water volume and aqueous solution E content are calculated using Euclidean distance. A trained random forest regression model is then used to predict viscosity and solid content based on current production input conditions (such as key parameters like water volume and aqueous solution E content). The difference between the predicted viscosity and the target viscosity (i.e., the expected value) is calculated to assess the deviation between the current production parameters and the target viscosity. Dynamic adjustments are made based on the viscosity deviation. After each adjustment, the prediction model is used again to predict viscosity and solid content, and the effect of the adjustment is checked. If the target value is reached, the adjustment stops. This process is continuously optimized through a real-time feedback mechanism until the difference between the predicted viscosity and the expected viscosity meets the predetermined optimization termination condition. Finally, the optimized parameters such as water volume, aqueous solution E content, predicted viscosity, and predicted solid content are output, and the relative error is calculated to evaluate the effectiveness of the optimization process.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0016] This solution combines the prediction model of the random forest algorithm with the real-time learning optimization algorithm, and uses the Bayesian optimization algorithm to perform hyperparameter tuning for the prediction model. By predicting the viscosity and solid content of the glue in real time during the production process based on the machine learning algorithm, and dynamically adjusting the production parameters according to the difference between the prediction results and the expected values, this real-time feedback and optimization mechanism makes the production process more accurate, significantly improves the stability of product quality, and reduces errors and fluctuations caused by human intervention.

[0017] In addition, the automated optimization and adjustment process simplifies the cumbersome operations in the traditional production process. It can not only greatly improve the accuracy of glue performance adjustment, but also significantly shorten the test and adjustment time. It avoids the shortcomings of relying on human experience in traditional methods, thereby improving production efficiency, optimizing glue formulation and ensuring product quality stability, and adapting to the needs of modern large-scale production. Attached Figure Description

[0018] Figure 1 This is a block diagram illustrating the principle of optimizing and adjusting the viscosity and solid content of the adhesive in a high-speed connector according to an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of the prediction results of the glue solid content of the high-speed nozzle according to an embodiment of the present invention; wherein, (a) is a scatter plot of the relative error of the solid content; and (b) is a distribution plot of the actual value and the predicted value of the solid content.

[0020] Figure 3 This is a schematic diagram of the predicted viscosity of the high-speed nozzle adhesive according to an embodiment of the present invention; (a) is a scatter plot of the relative error of viscosity; (b) is a distribution diagram of the actual and predicted viscosity values.

[0021] Figure 4This is a schematic diagram of the Pearson correlation analysis results for the viscosity and solid content of the high-speed connector adhesive in an embodiment of the present invention.

[0022] Figure 5 This is a schematic diagram illustrating the characteristic importance of the viscosity of the high-speed connector adhesive in an embodiment of the present invention.

[0023] Figure 6 This is a schematic diagram illustrating the characteristic importance of the adhesive content of high-speed nozzles in an embodiment of the present invention. Detailed Implementation

[0024] To better understand the above-described objects, features, and advantages of the present invention, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. Many specific details are set forth in the following description to provide a thorough understanding of the present invention; however, the present invention may be practiced in other ways than those described herein, and therefore, the present invention is not limited to the specific embodiments disclosed below.

[0025] An example of a method for optimizing and adjusting the viscosity and solids content of high-speed nozzle adhesive based on BO-RF-JITL includes the following steps:

[0026] Step A: Perform data preprocessing on historical production data of high-speed nozzle adhesive to extract characteristic variables related to viscosity and solid content;

[0027] Step B: Through correlation analysis and feature importance analysis, select the feature variables that have a significant impact on viscosity and solid content, and construct the feature vector;

[0028] Step C: Constructing a Random Forest Prediction Model: The Random Forest (RF) algorithm is used to train the prediction model based on preprocessed historical production data. The hyperparameters of the prediction model are then tuned using the Bayesian Optimization (BO) algorithm. The mapping relationship between viscosity and solid content and feature vectors is established, and the prediction performance of the prediction model is evaluated.

[0029] Step D: Combining Just-In-Time Learning (JITL) optimization algorithm, based on the set target viscosity and solid content, find the optimal adjustment amount of the splice adhesive formulation to meet the high-speed splice adhesive production target.

[0030] To better understand the present invention, the principles of the invention will be introduced, such as... Figure 1As shown, firstly, production indicators and expected target parameters for high-speed nozzle adhesive are collected as input data, including viscosity and solid content. Then, this data is input into a tobacco adhesive production database for storage and processing, which is regularly updated to maintain data validity. Next, machine learning algorithms are used to predict and analyze the data, and the generated prediction results are stored in the training prediction module. Outputs include predicted values ​​for viscosity and solid content, as well as adjustments for water and aqueous solution E. To improve prediction accuracy, a real-time optimization algorithm is used to adjust the model in real time. Finally, the differences between the predictions and actual production data are compared, and a feedback mechanism is used to continuously optimize the model, thereby achieving precise control of the nozzle adhesive production process and continuous improvement of product quality.

[0031] The specific implementation methods for each step are described in detail below:

[0032] Step A involves preprocessing the historical production data of the high-speed connector adhesive, including cleaning and data standardization.

[0033] This example collected 150 historical data entries. Characteristic variables included emulsion A content, emulsion A viscosity, emulsion A solid content, emulsion F content, emulsion F viscosity, emulsion F solid content, aqueous solution E content, aqueous solution E solid content, aqueous solution F content, aqueous solution F solid content, water content, other formulation content, other formulation solid content, and total weight. Emulsion A, emulsion F, aqueous solution E, aqueous solution F, and other formulations are all material designations for producing high-speed nozzle adhesive; specific limitations are not specified here due to variations in formulations. The target variables are the viscosity and solid content of the nozzle adhesive. The amount of aqueous solution E and the water content are variables that need to be optimized and adjusted in actual production. Adjusting these two variables ensures the specified viscosity and solid content of the nozzle adhesive.

[0034] (1) When cleaning the data, first check the historical data of the high-speed nozzle adhesive to remove or correct outliers and missing values.

[0035] Outlier handling is determined by calculating the mean and standard deviation of each feature. A value of a feature greater than the mean ± 3 times the standard deviation is considered an outlier and is deleted or corrected. Let the dataset be... ,in Indicates the first one sample Each feature value.

[0036] Outlier detection criteria:

[0037] ;

[0038] in, For the first The mean of each feature, For the first The standard deviation of each feature is considered an outlier if it exceeds this range.

[0039] For missing values ​​in the data, methods such as mean imputation and regression imputation can be used. This example uses mean imputation, so for missing values... It can be set to the mean of this feature:

[0040] ;

[0041] (2) Standardize the numerical features using the Z-Score standardization method to eliminate the influence between different units, specifically:

[0042] ;

[0043] in, It is the standardized value. It is the original value. It is the mean of this feature. It is the standard deviation of that feature.

[0044] II. Step B, Feature Engineering: Through correlation analysis and feature importance analysis, select feature variables that have a significant impact on viscosity and solid content, and construct feature vectors;

[0045] In feature engineering, feature importance is calculated using Pearson correlation analysis and random forest algorithm, and features that have a significant impact on the target variables (viscosity and solid content) are selected.

[0046] 1. Analyze the correlation between the features and the target variable. The formula for calculating the Pearson correlation coefficient is:

[0047] ;

[0048] in, Features and target variable The Pearson correlation coefficient between them and The mean of the feature and target variables. (By...) Figure 4 It can be seen that the solid content of aqueous solution E, the solid content of aqueous solution F, and other solid contents have a correlation of 0 with viscosity and solid content, meaning that from a data perspective, these three characteristics are unrelated to viscosity and solid content. The viscosity of emulsion A and the viscosity of emulsion F are positively correlated with viscosity, while the other indicators are negatively correlated with them. The solid content of emulsion A, the solid content of emulsion F, the content of emulsion A, the content of emulsion F, the content of other formulations, and the content of aqueous solution F are positively correlated with solid content, while the other indicators are negatively correlated with them.

[0049] Feature importance calculation in the random forest algorithm:

[0050] In this invention, the random forest algorithm is used to calculate the importance of features through information gain. In each decision tree, when selecting a feature-split node, the Gini index brought by that feature is calculated as the importance of the feature. The formula for calculating the Gini index is:

[0051] ;

[0052] in, For the current dataset, For dataset Medium category The sample proportion.

[0053] Information gain Indicates the degree to which a certain feature reduces the Gini index:

[0054] ;

[0055] in, It is the dataset before the partition (the dataset of the parent node). It is the first after the division A subset It is the first The number of samples in each subset This is the total number of samples. The results, obtained by calculating the importance of each feature, are shown below. Figure 5 , Figure 6 .

[0056] 3. Feature selection: Based on the Pearson correlation coefficient between the input variables and the target variable, and the calculated feature importance, select features that are highly correlated with viscosity and solid content;

[0057] Set the threshold for Pearson correlation coefficient (Take a value of 0.5 to 0.8) for the correlation coefficient The features selected were chosen as input features, and feature importance was used for validation. Features with little contribution to the model's prediction results and those irrelevant to the target variable were removed. This process ensured that the selected features were both highly important and minimized the impact of multicollinearity on the model. Therefore, the following features were selected to train the model for predicting viscosity: emulsion A viscosity, emulsion F viscosity, aqueous solution E content, water content, emulsion A solid content, and emulsion F solid content. The following features were selected to train the model for predicting solid content: emulsion A solid content, emulsion F solid content, water content, emulsion A viscosity, aqueous solution E content, and emulsion F viscosity. This feature selection conclusion was based on example data; different data may yield different feature selection results.

[0058] 4. Dataset partitioning: After data preprocessing, there are a total of 148 data entries. These are divided into a training set of 120 data entries and a test set of 28 data entries to train the machine learning model.

[0059] Step C: Train the random forest prediction model;

[0060] 1. This embodiment uses the Random Forest (RF) algorithm to train the prediction model. First, multiple subsets are generated from the original dataset using sampling with replacement. Each subset is used to train a decision tree. During the construction of each decision tree, for each node, only a subset of features are randomly selected from all features to determine the optimal split point. This introduces additional randomness, increasing the model's generalization ability. For new input data... Each tree in a random forest A predicted value will be given. The final prediction result It is the average of all decision tree predictions:

[0061] ;

[0062] in, It is the number of decision trees. It is the first The predicted value for each tree.

[0063] (1) Construction of the random forest prediction model for viscosity:

[0064] Input characteristics: viscosity of emulsion A, viscosity of emulsion F, content of aqueous solution E, water content, solid content of emulsion A, and solid content of emulsion F;

[0065] Output: Predicted viscosity value;

[0066] (2) Construction of the random forest prediction model for solid content:

[0067] Input features: Emulsion A solid content, emulsion solid content, water content, emulsion A viscosity, aqueous solution E content, emulsion F viscosity

[0068] Output: Predicted solids content

[0069] 2. Bayesian optimization of hyperparameter tuning:

[0070] To further improve the performance of the random forest prediction model, this invention employs Bayesian optimization (BO) to fine-tune the hyperparameters, reducing the computational overhead required by traditional cross-validation methods. Bayesian optimization establishes a surrogate model of the objective function and guides the search process step-by-step using the acquisition function, finding the optimal hyperparameter combination within a limited number of experiments. The optimization objective is to minimize the relative prediction error. The specific implementation steps are as follows:

[0071] (1) Construction of the proxy model:

[0072] The optimization method of this invention represents the objective function by constructing a Gaussian Process (GP) model. A Gaussian process is a non-parametric Bayesian model that can be based on historical observation data. To predict the objective function, the mean and variance of each input point are given. The mathematical expression for the Gaussian process is:

[0073] ;

[0074] (2) Selection of acquisition function:

[0075] This invention uses Expected Improvement (EI) as the acquisition function, which balances exploration (exploring underexplored regions) and exploitation (focusing on searching the region near the current optimal solution). The mathematical formula for Expected Improvement is as follows:

[0076] ;

[0077] in, It is the currently known optimal solution. Is the objective function at point The predicted value at the current location. Improvements are expected in the calculation of the value at the current location. The expected improvement of the objective function relative to the current optimal solution is used as the basis for selecting the next evaluation point.

[0078] (3) Iterative process of Bayesian optimization:

[0079] Initialize training data: Randomly select a set of hyperparameter combinations. And use cross-validation to calculate the corresponding model performance values. (The model's prediction error);

[0080] Building a proxy model: based on initial data points Train a Gaussian process model so that it can predict the distribution of the objective function;

[0081] Optimize the acquisition function: Based on the current agent model, calculate the expected improvement value (EI) for each candidate point, and select the point with the largest expected improvement value as the next evaluation point;

[0082] Update the model: Feed the results of the new evaluation points back to the surrogate model and retrain the model, continuing to optimize the acquisition function using the new training data;

[0083] Repeat the steps: Repeat the optimization of the acquisition function and the model update operation until the predetermined stopping condition is met, that is, the objective function converges.

[0084] (4) Definition of hyperparameter search space:

[0085] In this invention, the following search space is set for the hyperparameters of the viscosity and solids content models:

[0086] Number of decision trees (n_estimators): The search space is [100, 500], indicating that the minimum number of decision trees is 100 and the maximum number is 500.

[0087] Maximum depth (max_depth): The search space is [5, 30], which means that the maximum depth of the tree is at least 5 and at most 30.

[0088] Minimum number of sample splits (min_samples_split): The search space is [2, 10], indicating that the minimum number of sample splits for each node is between 2 and 10.

[0089] By using the Bayesian optimization algorithm, these hyperparameter combinations are searched and optimized step by step to achieve the best model performance.

[0090] 3. Evaluate the model's effectiveness:

[0091] For evaluating the prediction model, this invention uses relative error and mean square error to assess the predictive effectiveness of the viscosity and solid content of the high-speed nozzle adhesive. Relative error measures the ratio of the error between the predicted and actual values ​​to the actual values.

[0092] ;

[0093] Mean squared error (MSE) is calculated by averaging the squared error between the predicted and actual values ​​over all samples.

[0094] ;

[0095] This is used to measure the accuracy of the random forest model's predictions. The evaluation results for the viscosity and solids content prediction models are shown below. Figures 2-3 .

[0096] IV. Step D: Establish an instant learning optimization and adjustment model and output the solution results:

[0097] 1. Determine the initial values ​​for water volume and the E content of the aqueous solution:

[0098] In the production and formulation of high-speed connector adhesive, the specified viscosity and solid content of the adhesive are obtained by adjusting the input amounts of water and aqueous solution E. Therefore, while the data for other materials are known, the amounts of water and aqueous solution E are unknown. In the established real-time learning optimization adjustment model, the initial values ​​for the water quantity and the content of aqueous solution E must first be determined.

[0099] This invention establishes an instant search algorithm to find existing data records in a multi-dimensional feature space that most closely match the material data features input by the user. Using this algorithm, suitable water and aqueous solution E ratios can be quickly recommended based on existing data.

[0100] First, extract the feature variables from the input for computation. Then, iterate through each row of the dataset in the database and calculate the Euclidean distance between the input feature and the feature value of the current row:

[0101] ;

[0102] in, It is the input feature vector. It is a vector composed of the feature variables of a certain row in the dataset.

[0103] Finally, select the 5 data points with the smallest calculated Euclidean distance, extract and return the values ​​of "water volume" and "water solution E content", and average them to obtain the initial values ​​of water volume and water solution E content.

[0104] 2. Establish the model and solve it.

[0105] In the production of high-speed connector adhesive, the relationship between water volume and aqueous solution E content and viscosity is as follows: when water volume increases and aqueous solution E content decreases, viscosity decreases; conversely, when water volume decreases and aqueous solution E content increases, viscosity increases. However, it is difficult to replicate this pattern using machine learning combined with general optimization algorithms. Therefore, this invention designs a Just-In-Time Learning (JITL) optimization adjustment algorithm. Based on the aforementioned relationship between water volume and aqueous solution E content and viscosity, and combined with a machine learning random forest algorithm, it adjusts the water volume and aqueous solution E content.

[0106] First, based on the pre-trained random forest prediction model, the test set feature variable data, including the initial values ​​of water and aqueous solution E, are imported. The desired target viscosity and target solid content are then input to predict the viscosity of the high-speed nozzle adhesive under the current input conditions. and solid content ;

[0107] Secondly, the difference between the predicted viscosity and the expected viscosity is calculated. And dynamically adjust the input parameters based on this difference: water volume and aqueous solution E amount To narrow the gap between the predicted results and the expected values, the adjustment rules are as follows:

[0108] If the viscosity is too high Then update the water volume and the volume of the aqueous solution E: ;

[0109] If the viscosity is too low Then update the water volume and the volume of the aqueous solution E: ;

[0110] Then, during the adjustment process, if the water volume or the value of the aqueous solution E exceeds the predetermined range (e.g.) or If the error occurs, the exception handling mechanism will be triggered, and the optimization process will be stopped.

[0111] Next, the prediction and adjustment are repeated until the difference between the predicted viscosity and the desired viscosity meets the optimization termination condition. .

[0112] Finally, output the optimized water volume. Aqueous solution E amount Predicting viscosity Predicting solid content The relative error was calculated and is shown in Table 1.

[0113] Table 1 Optimization and Adjustment Results

[0114] Initial water volume Initial aqueous solution E amount Optimized water volume Optimized aqueous solution E amount Expected viscosity Predicted viscosity Actual viscosity in production Predicting solid content viscosity difference Relative error (%) 85.60 250.00 79.60 244.00 5000.00 5187.40 5170 0.53 187.4 3.75 71.60 240.00 71.60 240.00 5000.00 4960.56 5040 0.53 39.4 0.79 69.60 245.00 75.60 242.00 5000.00 5175.43 5140 0.53 175.4 3.51 69.60 245.00 69.60 245.00 5000.00 4815.73 4870 0.53 184.3 3.69 68.60 250.00 68.60 250.00 5000.00 4823.43 4840 0.53 176.6 3.53 63.60 255.00 63.60 255.00 5000.00 4865.27 5010 0.53 134.7 2.69 83.60 255.00 79.60 244.00 5000.00 5187.40 5200 0.53 187.4 3.75 53.60 255.00 53.60 255.00 5000.00 5122.80 5050 0.53 122.8 2.46 73.60 235.00 62.60 247.00 5000.00 4874.23 4890 0.53 125.8 2.52 72.60 235.00 62.60 247.00 5000.00 4874.23 4950 0.53 125.8 2.52

[0115] Through the above process, this solution establishes a model based on the random forest prediction algorithm and the real-time learning optimization algorithm to predict the viscosity and solid content of the high-speed nozzle adhesive in real time, and dynamically adjusts production parameters (such as water volume and aqueous solution E volume) according to the prediction results to optimize the production process. This solution can improve production accuracy, reduce manual intervention, reduce production costs, and ensure the stability of product quality.

[0116] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments for application in other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for optimizing and adjusting the viscosity and solid content of high-speed nozzle adhesive based on BO-RF-JITL, characterized in that, Includes the following steps: Step A: Perform data preprocessing on historical production data of high-speed nozzle adhesive to extract characteristic variables related to viscosity and solid content; Step B: Through correlation analysis and feature importance analysis, select input feature variables that have a significant impact on viscosity and solid content, and construct feature vectors; Step C: Constructing a Random Forest Prediction Model: The Random Forest algorithm (RF) is used to train the prediction model based on preprocessed historical production data. The hyperparameters of the prediction model are then tuned using the Bayesian optimization algorithm (BO). The mapping relationship between viscosity and solid content and feature vectors is established, and the prediction performance of the prediction model is evaluated. Step D: Calculate the initial values ​​of water and aqueous solution E. Using a trained random forest prediction model combined with an instant learning optimization algorithm, find the optimal adjustment amount of the splice adhesive formulation based on the set target viscosity and solid content to meet the production target of high-speed splice adhesive. In the production and formulation of high-speed splice adhesive, the specified viscosity and solid content of the high-speed splice adhesive are obtained by adjusting the input amount of water and aqueous solution E. Therefore, it is first necessary to determine the initial values ​​of water volume and aqueous solution E content, as follows: First, extract feature variables. Then, iterate through each row of the dataset and calculate the Euclidean distance between the input feature and the feature value of the current row: in, It is the input feature vector. It is a vector composed of the feature variables of a certain row in the dataset; Finally, select the data points with small Euclidean distances, extract and return the values ​​of water volume and aqueous solution E content, and average them to obtain the initial values ​​of water volume and aqueous solution E content. In step D, a Just-In-Time (JITL) learning optimization algorithm combining a real-time feedback mechanism is proposed. By dynamically adjusting the input parameters water and aqueous solution E, the adjustment amount of the high-speed nozzle adhesive formulation is optimized in real time based on the deviation between the predicted viscosity and solid content and the expected values. (1) Based on the pre-trained random forest prediction model, import the test set feature variable data, including the initial values ​​of water and aqueous solution E, input the desired target viscosity and target solid content, and predict the viscosity of the high-speed nozzle adhesive under the current input conditions. and solid content ; (2) Calculate the difference between the viscosity prediction result and the expected viscosity. And dynamically adjust the input parameters based on this difference: water volume and aqueous solution E amount The adjusted rules are as follows: If the viscosity is high If the value is greater than 0, then update the water volume and the amount of aqueous solution E: If the viscosity is low If the value is less than 0, then update the water volume and the volume of the aqueous solution E: ； (3) During the adjustment process, if the water volume or the value of the aqueous solution E exceeds the predetermined range, i.e. or If this occurs, the exception handling mechanism will be triggered, and the optimization process will be stopped. (4) Repeat the prediction and adjustment until the difference between the predicted viscosity and the desired viscosity meets the optimization termination condition. Finally, the optimized water volume is output. Aqueous solution E amount Predicting viscosity Predicting solid content And calculate the relative error.

2. The method for optimizing and adjusting the viscosity and solid content of high-speed nozzle adhesive based on BO-RF-JITL according to claim 1, characterized in that: Step A is specifically implemented in the following way: obtain historical production data of high-speed connector adhesive and check it, remove or correct outliers and missing values; extract feature variables related to viscosity and solid content from it, and standardize the numerical feature variables. The feature variables include emulsion A content, emulsion A viscosity, emulsion A solid content, emulsion F content, emulsion F viscosity, emulsion F solid content, aqueous solution E content, aqueous solution E solid content, aqueous solution F content, aqueous solution F solid content, water volume, other formulation content, other formulation solid content, and total weight.

3. The method for optimizing and adjusting the viscosity and solid content of high-speed nozzle adhesive based on BO-RF-JITL according to claim 1, characterized in that: In step B, Pearson correlation analysis and the Random Forest (RF) algorithm are used to calculate the importance of feature variables. Specifically: Step B1: Perform Pearson correlation analysis on the characteristic variable and the target variable, and calculate the Pearson correlation coefficient between the characteristic variable and the target variable. To assess the extent to which features influence the target variable; Step B2: Using the information gain-based feature importance in the Random Forest algorithm (RF), calculate the contribution of each feature variable to the prediction results of the Random Forest prediction model to evaluate the importance of the feature variables. In each decision tree, when selecting a feature to split a node, the change in Gini index or entropy brought about by that feature is calculated as the importance of the feature; where information gain represents the degree to which a feature reduces the Gini index. Step B3: Select features that are highly correlated with viscosity and solid content based on Pearson correlation coefficient and feature importance; Set the threshold for Pearson correlation coefficient For correlation coefficient The feature variables are selected as input features, and feature importance is used as the validation factor to remove features that contribute little to the prediction results of the random forest prediction model and features that are irrelevant to the target variable.

4. The method for optimizing and adjusting the viscosity and solid content of high-speed nozzle adhesive based on BO-RF-JITL according to claim 1, characterized in that: In step C, when training the prediction model, the Bayesian optimization algorithm (BO) is used to iteratively optimize the hyperparameters. Specifically: (1) Initially, several initial values ​​are randomly selected. And evaluate the objective function. ; (2) Based on the initial evaluation results, a Gaussian process model is constructed to predict the behavior of the objective function at other points; the mathematical expression of the Gaussian process is: in, It is the mean function. It is the covariance function, used to calculate the similarity between input points. By training a Gaussian process, the predicted distribution of the objective function at each point is obtained. (3) Calculate the expected improvement value for each candidate point using the acquisition function, and select the point with the largest expected improvement for a new evaluation; specifically, use the expected improvement EI as the acquisition function, and the mathematical formula for the expected improvement is as follows: in, It is the currently known optimal solution. Is the objective function at point The predicted value at the current location is expected to improve the calculation. The expected improvement of the objective function relative to the current optimal solution is used as the basis for selecting the next evaluation point; (4) Feed the new evaluation results back to the Gaussian process model, use the trained Gaussian process model to predict the mean and variance of the objective function at the candidate points, and then calculate the expected improvement value for each candidate point. Select the point with the largest expected improvement value as the next evaluation point and update the model parameters; (5) Repeat steps (3) and (4) until the stopping condition is met, that is, the maximum number of evaluations is reached or the optimal solution of the objective function is converged.

5. The method for optimizing and adjusting the viscosity and solid content of high-speed nozzle adhesive based on BO-RF-JITL according to claim 1, characterized in that: In step C, the prediction model is evaluated using relative error and mean square error to assess the predictive effectiveness of the viscosity and solid content of the high-speed nozzle adhesive.

Images