A method for intelligent prediction of the weighing weight after tread extrusion production line

Abstract

This invention relates to an intelligent prediction method for the post-weighing weight of a tire extrusion production line. The method includes the following steps: 1) acquiring historical and real-time data of the rubber tire extruder production process; 2) constructing a preliminary similar dataset using a weighted similarity analysis method based on Euclidean distance; 3) determining the size of the current similar dataset using the Pearson correlation coefficient; 4) using the isolated forest algorithm to remove outliers from the current similar dataset; 5) constructing a local model based on a multi-head attention mechanism for immediate learning, training the model on a training set to obtain a post-weighing weight prediction model, inputting actual real-time data into the prediction model, and obtaining the post-weighing weight prediction result of the product from the rubber tire extrusion production line. Compared with existing technologies, this invention has advantages such as improving the efficiency of post-weighing weight prediction and reducing prediction lag.

Description

Technical Field

[0001] This invention relates to the field of product quality index prediction in the process industry, and in particular to an intelligent prediction method for the weighing weight after the tire tread extrusion production line. Background Technology

[0002] In industrial production, offline monitoring alone often fails to meet product quality requirements due to its inherent lag. Post-weighing weight measurement in rubber tire extrusion production lines is highly lagging and inaccurate. Real-time prediction of post-weighing weight can significantly reduce this lag, which is crucial for real-time parameter adjustments in the production process. However, global modeling is affected by product type and collected data, resulting in significant errors in indicator prediction and long model training and testing times. Therefore, employing a real-time learning model, which adjusts the local model in real-time based on input data to predict weight and adapt to changes in production processes, can effectively improve detection accuracy and enhance detection efficiency by adjusting the local model structure.

[0003] When building instant learning models, similarity analysis and outlier removal are typically used to obtain training data for local models. To address the overstretching and overcompression issues in DTW applications, the paper "Adaptively Constrained Dynamic Time Warping for Time Series Classification and Clustering" proposes an Adaptive Constrained DTW (ACDTW) algorithm. This algorithm introduces a new adaptive penalty function to more accurately calculate the distance between trajectories, adaptively adjusting the correspondence between two trajectories and achieving higher accuracy across different trajectories. The paper "Elastic Similarity and Distance Measures for Multivariate Time Series" proposes a metric called Developing Longest Common Subsequence (DLCSS) for mining LCSS-based time series. This metric defines two similarity thresholds to determine their values, eliminating the shortcomings of LCSS. It outperforms LCSS and DTW in determining better cluster representations, with a confidence level of at least 99.95%. The paper "A New Local Distance-Based Outlier Detection Approach for Scattered Real-World Data" proposes a Local Distance-Based Outlier Factor (LDOF) to measure outliers in sparse datasets. LDOF uses the relative position of an object to its neighbors to determine how far an object deviates from its neighbors. Experimental results show that its detection accuracy is superior to traditional KNN and LOF, especially since it is not sensitive to parameter values. The paper "Variational LSTM Enhanced Anomaly Detection for Industrial Big Data" proposes a Variational Long Short-Term Memory (VLSTM) learning model for intelligent anomaly detection based on reconstructed feature representations, significantly improving the accuracy and false positive rate of anomaly detection in industrial big data.

[0004] Furthermore, many researchers have designed real-time learning similarity analysis and local model structures using an overall framework. The paper "An ensemble JITL method based on multi-weighted similarity measures for cold rolling force prediction" proposes a novel real-time learning model for similarity analysis to improve the impact of rolling force prediction accuracy on product quality. First, relevant samples are selected using multi-weighted similarity measures. Then, local models are constructed and their output values ​​are queried. Finally, the outputs of each local model are integrated according to an ensemble learning strategy. In addition, a cumulative similarity factor is introduced to optimize the number of samples for local modeling, and a threshold is set to adaptively update the local models, achieving good results. Considering that traditional real-time learning modeling is not suitable for industrial processes exhibiting time-varying and non-Gaussian behavior, the paper "Adaptive Gaussian Mixture Model-Based Relevant Sample Selection for JITL Soft Sensor Development" proposes a GMM-based similarity metric to improve the prediction accuracy of JITL soft sensors. Through discussions involving numerical examples and case studies of industrial processes, the feasibility and effectiveness of the proposed method are demonstrated. Considering that traditional JITL methods mainly focus on equal sample sizes between process input variables and quality output variables, the paper "Semisupervised JITL Framework for Nonlinear Industrial Soft Sensing Based on Locally Semisupervised Weighted Principal Component Analysis (SWPPCR)" proposes a novel semisupervised JITL framework. By selecting the most relevant labeled and unlabeled samples and assigning them weights, a local SWPPCR can be constructed to estimate the output variables of the query sample. The effectiveness and flexibility of this method are demonstrated through case studies of its predictive power in industrial processes.

[0005] As mentioned above, many studies have demonstrated the effectiveness of instant learning models in real-time prediction. However, when applied to the field of weighing, most existing instant learning models do not consider the complexity of the algorithm. Due to the stacking of similarity analysis or the structural complexity of local models, the time spent on building datasets and training local models for prediction is long, which cannot significantly reduce prediction lag. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent prediction method for the post-weighing weight of a tire tread extrusion production line in order to improve the efficiency of post-weighing weight prediction and reduce prediction lag.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] Compared with the prior art, the present invention has the following beneficial effects:

[0009] A method for intelligent prediction of the weighing weight after a tire tread extrusion production line, the method comprising the following steps:

[0010] 1) Obtain historical data of the rubber tire extruder production process and real-time data for model prediction;

[0011] 2) By using the weighted Euclidean distance similarity analysis method, a preliminary similar dataset is constructed based on historical and real-time data to obtain a certain amount of similar data;

[0012] 3) Using the Pearson correlation coefficient, determine the size of the most correlated similar dataset based on the correlation of different sized subsets of the initial similar dataset, and use the subset of the most correlated similar dataset size as the current similar dataset;

[0013] 4) Use the Isolation Forest algorithm to remove outliers from the current similar dataset to obtain the training set;

[0014] 5) Based on the multi-head attention mechanism, a local model for real-time learning is constructed. The model is trained on the training set to obtain the post-weighing weight prediction model. The actual real-time data is input into the prediction model to obtain the post-weighing weight prediction result of the product of the rubber tire extrusion production line.

[0015] Furthermore,

[0016] 2) The specific steps are as follows:

[0017] 201) Obtain the Euclidean distance calculation formula;

[0018] 202) Mean centering of historical data;

[0019] 203) Perform dimensionality reduction projection and calculate the variance of each data point of the historical data after mean centering projected onto the low-dimensional coordinates;

[0020] 204) Maximize the variance by solving for the weighting factors using the Lagrange multiplier method;

[0021] 205) Calculate the weighted Euclidean distance similarity based on the weighting factor and the Euclidean distance formula;

[0022] 206) Based on real-time data, select a certain number of data with the highest similarity from historical data by weighted Euclidean distance similarity to construct a preliminary similarity dataset.

[0023] Furthermore, in 202), historical data is represented as X = [x1, x2, ..., x...]. m ], where m represents the total number of historical data points, and the historical data consists of m n-dimensional data points. The historical data after mean centering is:

[0024]

[0025] Where i represents the sequence number of the historical data.

[0026] Furthermore, the variance is:

[0027]

[0028] Where ω is the weighting factor and Var represents the variance.

[0029] Furthermore, the weighted Euclidean distance similarity is:

[0030]

[0031] Where, x i Represents various historical data, y j This represents each real-time data point, where k represents the dimension index, and x represents the data point index. ik For data x i The k-th dimension, y jk For data y j The k-th dimension, ω i This represents the magnitude of the weight factor for the i-th dimension of the data.

[0032] Furthermore, the specific steps in 3) include:

[0033] 301) Obtain subsets of different sizes from the initial similar dataset;

[0034] 302) Calculate the correlation between subsets of different sizes and real-time data using the Pearson correlation coefficient, and select the subset with the highest correlation. The size of this subset is the most suitable size for the similar dataset.

[0035] 303) Select the subset corresponding to the most suitable similar dataset size as the current similar dataset.

[0036] Furthermore, the specific steps of 4) are as follows:

[0037] 401) Construct an isolated forest algorithm model;

[0038] 402) Use the weights of the current similar datasets as input to train the Isolation Forest algorithm model, and obtain the trained algorithm model;

[0039] 403) Use the trained algorithm model to remove abnormal data from the current similar dataset to obtain the training set.

[0040] Furthermore, the specific steps of 5) are as follows:

[0041] 501) Construct a multi-head attention mechanism network structure model, i.e., a local model;

[0042] 502) Determine the input dimension and use the subsequent weighing weight as the output;

[0043] 503) Select the mean absolute error loss function as the loss function for the model;

[0044] 504) The model is trained based on the training set to obtain the weighing weight prediction model. The actual real-time data is input into the prediction model to obtain the weighing weight prediction result of the product of the rubber tire extrusion production line.

[0045] Furthermore, the loss function is:

[0046]

[0047] Among them, y i This refers to the true value, f(x) i () refers to the predicted value.

[0048] Furthermore, the network structure of the local model contains only three linear layers.

[0049] This invention, through the collection of historical datasets from rubber tire extrusion production, derives weighting factors that reduce the influence of physical dimensions of data across different dimensions. It also proposes a weighted Euclidean distance similarity analysis method. This novel similarity analysis method selects data to construct similar datasets, improving not only the quality of the selected data but also accelerating the efficiency of similarity analysis and increasing the training and prediction speed of the prediction model. Furthermore, by employing the Pearson correlation coefficient to measure the correlation between different sized subsets of the initial similar dataset, the optimal dataset size is determined, improving detection accuracy. The model in this invention uses a multi-head attention mechanism for post-weighing weight prediction, enabling real-time monitoring of quality indicators, reducing measurement lag, and facilitating the adjustment of production parameters. Attached Figure Description

[0050] Figure 1 This is a flowchart of the present invention;

[0051] Figure 2 This is a schematic diagram of the system operation architecture of the present invention;

[0052] Figure 3 This is a diagram showing the results of determining the dataset size in this invention.

[0053] Figure 4 This is a diagram showing the outlier removal results of the present invention;

[0054] Figure 5 This is a graph showing the predicted weight after weighing according to the present invention.

[0055] Figure 6 This is a comparison chart of the results of the present invention, wherein... Figure 6 (a) shows the comparison results between the present invention and global prediction models ANN and LSTM. Figure 6 (b) shows the comparison results between the present invention and JITL-ANN and WSJITL-ANN. Figure 6 (c) represents the box plot of the present invention compared with other models. Detailed Implementation

[0056] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0057] This invention proposes an intelligent prediction method for weighing after tread extrusion production line, the flowchart and block diagram are as follows. Figure 1 and Figure 2 As shown, the method includes the following steps:

[0058] 1) Obtain historical data of the rubber tire extruder production process and real-time data for model prediction;

[0059] 2) By using a weighted similarity analysis method based on Euclidean distance, a preliminary similar dataset is constructed by acquiring a certain amount of similar data based on historical and real-time data;

[0060] 3) Use the Pearson correlation coefficient to determine the most suitable size of the similar dataset based on the correlation of different sized subsets of the preliminary similar dataset, and use the subset of the most suitable size of the similar dataset as the current similar dataset;

[0061] 4) Use the Isolation Forest algorithm to remove outliers from the current similar dataset to obtain the training set;

[0062] 5) Based on the multi-head attention mechanism, a local model for real-time learning is constructed. The model is trained on the training set to obtain a weighing weight prediction model. The actual real-time data is input into the prediction model to obtain the weighing weight prediction results of the products of the rubber tire extrusion production line.

[0063] In step 1), historical data on the production process of different models of rubber tire extrusion production line are obtained. After removing some invalid physical quantities, the size is 30630×27. Each data point contains 27 physical quantities such as screw speed and current. Real-time data for model prediction is obtained. The dimensions are adjusted according to the size of the historical data to construct a real-time dataset with a size of 60×27.

[0064] In step 2), a preliminary similar dataset is constructed by obtaining a certain amount of similar data through a weighted similarity analysis method based on Euclidean distance. The specific implementation steps are as follows:

[0065] 201) Considering the efficiency of similarity analysis, Euclidean distance is chosen as the benchmark similarity analysis method. The similarity between two data points is determined by the magnitude of the Euclidean distance. The formula for Euclidean distance is:

[0066]

[0067] 202) Calculate the weighted similarity analysis weight factors. The main idea is to project high-dimensional historical data onto a one-dimensional plane through dimensionality reduction projection, and retain the original information by maximizing the variance, thereby calculating the projection matrix and obtaining the weight factors.

[0068] Suppose we have m n-dimensional data points, which can be represented as X = [x1, x2, ..., xn]. m By centering its mean, we can obtain:

[0069]

[0070] 203) Suppose we have a direction vector ω, and the coordinates of each data point projected onto this dimension are ω. t x i The variance can be expressed in the following form:

[0071]

[0072] 204) at ω T To maximize the variance given ω = 1, let... Solve for ω using the Lagrange multiplier method.

[0073] L(ω,λ)=ω T Cω+λ(1-ω T ω)

[0074] 205) Combining the weighting factors with the Euclidean distance formula yields the weighted Euclidean distance similarity analysis method. The weighting factors ω = [ω1, ω2, ..., ω] are calculated. n The weighted Euclidean distance similarity calculation formula is obtained by combining the weights and Euclidean distance:

[0075]

[0076] Where Dis(x) i ,y j ) represents data x i and y j The weighted Euclidean distance between them, ω i x represents the magnitude of the weight factor for the data in the i-th dimension. ik For data x i The k-th dimension, y jk For data y j The kth dimension.

[0077] 206) Using a weighted Euclidean distance similarity analysis method, the similarity between the data in the real-time dataset and the historical dataset is calculated, and a certain number of data are selected to construct a similar dataset. Considering that the amount of data collected for each type of rubber compound is approximately 1500-2000 records, 2000 records are selected as the baseline dataset size.

[0078] In step 3), the Pearson correlation coefficient is used to measure the correlation between subsets of similar datasets of different sizes to determine the most suitable size for the similar dataset. The specific implementation steps are as follows:

[0079] 301) Obtain subsets of different sizes from an initial similar dataset of size 2000. If we construct subsets of the dataset in increments of 100 data points, it can be represented as:

[0080] D = {D1, D2, ..., D} 20}

[0081] Here, the present invention selects 100 data points as an increment to determine the most suitable size of similar datasets.

[0082] 302) Calculate the correlation between subsets of different sizes and the original real-time data using the Pearson correlation coefficient, and select the subset with the highest correlation. The size of this subset is the optimal subset size. For example... Figure 3 As shown, each subset corresponds to a value between 0 and 1, representing the correlation between the subset of that size and the real-time data. This indicates that the dataset size is most suitable when it is 1800.

[0083] 303) Treat this subset as a similar dataset for further processing.

[0084] In step 4), the Isolation Forest algorithm is used to remove outliers from the similar datasets in step 3). The specific implementation steps are as follows:

[0085] 401) Construct an isolated forest algorithm model.

[0086] 402) The weights of similar datasets after weighing are used as input to train the Isolation Forest algorithm, resulting in the trained algorithm model.

[0087] 403) Use the trained model to remove outlier data from the dataset.

[0088] Outlier removal results are as follows Figure 4 As shown, the outlier ratio of the isolated forest rule is set to 10%. The algorithm predicts two results: 1 and -1. 1 indicates that the point is a normal value and is represented by a yellow dot, while -1 indicates that the point is an outlier and is represented by a purple dot, which will be removed.

[0089] In step 5), a local model for immediate learning is constructed based on a multi-head attention mechanism. This local model is then trained using the similar dataset from step 4), and the trained model is used to predict the subsequent weighing weight. The specific implementation steps are as follows:

[0090] 501) Construct a multi-head attention mechanism network structure model. The network structure contains only three linear layers, which meets the requirements of lightweight network and fast training and prediction.

[0091] 502) Select appropriate dimensions as inputs and determine which dimensions are in the three input layers; then weigh the weight as the output.

[0092] 503) Choose an appropriate loss function. Here, we choose mean absolute error as the loss function:

[0093]

[0094] 504) Complete model training and model prediction. Use the first 26 dimensions of the similar dataset obtained in 4) as input and the 27th dimension, i.e. the weight after weighing, as output to train the local model. After completing the model training, use the first 26 dimensions of the real-time data as input to predict the value of the weight after weighing.

[0095] like Figure 5 The figure shows the post-weighing weight prediction results of the real-time learning model of this invention. The root mean square error (RMSE) between the predicted and actual values ​​is 6.33, the mean absolute error (MAE) is 5.01, and the mean absolute percentage error (MAPE) is 0.26%.

[0096] like Figure 6 The image shown is a comparison chart of this model and other models. Among them, Figure 6 (a) is a comparison of the results of the present invention with global prediction models ANN and LSTM. It can be seen that the prediction results of the instant learning model have a significant improvement over the global modeling results of ANN and LSTM. Figure 6(b) is a comparison of the results of the present invention with JITL-ANN and WSJITL-ANN. It can be seen that the proposed network architecture has the best prediction stability and is closest to the true value of the predicted data, which can determine the effectiveness of local modeling. Figure 6 (c) is a box plot comparing the present invention with other models. The short yellow line in the box plot represents the median of the prediction error, and the green dashed line represents the mean of the prediction error. It can be seen that the median error of the model proposed in this invention is closest to 0, and the mean error is the smallest, basically 0. The black rectangles represent data with prediction errors within 50% of the overall data, which also shows that the overall prediction error of this model is relatively small. Table 1 is a quantitative description of the comparison results between the present invention and other models, from RMSE, R 2 The prediction performance of different models was compared from the perspectives of MAE and Training Time. The results show that the error of global modeling is relatively large. The effectiveness of the weighted similarity analysis proposed in this invention can be seen through JITL-ANN and WS-JITL-ANN, showing a significant improvement compared to the unweighted method. A comparison between WS-JITL-ANN and the Proposed model demonstrates the effectiveness of the network model used in this invention, showing improvements in both accuracy and detection efficiency. KWS-JITL-FBLS is a novel real-time learning network structure model, and the method in this invention also shows a significant improvement compared to the latest real-time learning models. Therefore, through comprehensive comparison, the effectiveness of the quality monitoring model of this invention has been verified.

[0097] Table 1. Quantitative description of the results of this invention compared with other models.

[0098]

[0099] The present invention has the following advantages:

[0100] (1) This invention collects historical datasets of rubber tire extrusion production, derives weighting factors that can reduce the influence of physical dimensions of data in different dimensions, and proposes a weighted Euclidean distance similarity analysis method. By using the new similarity analysis method to select data and construct similar datasets, it not only improves the quality of the selected data, but also speeds up the efficiency of similarity analysis.

[0101] (2) This invention uses the Pearson correlation coefficient to measure the correlation between different sized subsets of the initial similar dataset to determine the most suitable dataset size, thereby improving detection accuracy.

[0102] (3) The present invention uses a multi-head attention mechanism to predict the weight of the post-weighing, which enables real-time monitoring of quality indicators, helps to reduce measurement lag, and facilitates the adjustment of production parameters.

[0103] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for intelligent prediction of the weighing weight after a tire tread extrusion production line, characterized in that, The method includes the following steps: 1) Acquire historical data of the rubber tire extruder production process and real-time data for model prediction; the historical data and real-time data include screw speed and current; 2) Using the weighted Euclidean distance similarity analysis method, a preliminary similar dataset is constructed by obtaining a certain amount of similar data based on historical and real-time data; 3) Using the Pearson correlation coefficient, determine the size of the most correlated similar dataset based on the correlation of different sized subsets of the initial similar dataset, and use the subset of the most correlated similar dataset size as the current similar dataset; 4) Use the Isolation Forest algorithm to remove outliers from the current similar dataset to obtain the training set; 5) Based on the multi-head attention mechanism, a local model for real-time learning is constructed. The model is trained on the training set to obtain a post-weighing weight prediction model. Actual real-time data is input into the prediction model to obtain the post-weighing weight prediction result of the rubber tire extrusion production line product; 2) The specific steps are as follows: 201) Obtain the Euclidean distance calculation formula; 202) Mean centering of historical data; 203) Perform dimensionality reduction projection and calculate the variance of each data point of the historical data after mean centering projected onto the low-dimensional coordinates; 204) Maximize the variance by solving for the weighting factors using the Lagrange multiplier method; 205) Calculate the weighted Euclidean distance similarity based on the weighting factor and the Euclidean distance formula; 206) Based on real-time data, select a certain number of data with the highest similarity from historical data using weighted Euclidean distance similarity to construct a preliminary similarity dataset; In 202), historical data is represented as Where m represents the total number of historical data points, and the historical data consists of m n-dimensional data points. The historical data after mean centering is as follows: in i Indicates the sequence number of historical data; The variance is: in, As a weighting factor, Indicates variance; The weighted Euclidean distance similarity is: in, Representing various historical data, This represents each real-time data point, where k represents the dimension number. For data The k-th dimension, For data The k-th dimension, This represents the magnitude of the weight factor for the i-th dimension of the data.

2. The intelligent prediction method for weighing after tread extrusion production line according to claim 1, characterized in that, 3) The specific steps include: 301) Obtain subsets of different sizes from the initial similar dataset; 302) Calculate the correlation between subsets of different sizes and real-time data using the Pearson correlation coefficient, and select the subset with the highest correlation. The size of this subset is the most suitable size for the similar dataset. 303) Select the subset corresponding to the most suitable size of the similar dataset as the current similar dataset.

3. The intelligent prediction method for weighing after tread extrusion production line according to claim 1, characterized in that, 4) The specific steps are as follows: 401) Construct an isolated forest algorithm model; 402) Use the weights of the current similar datasets as input to train the Isolation Forest algorithm model, and obtain the trained algorithm model; 403) Use the trained algorithm model to remove abnormal data from the current similar dataset to obtain the training set.

4. The intelligent prediction method for weighing after tread extrusion production line according to claim 1, characterized in that, 5) The specific steps are as follows: 501) Construct a multi-head attention mechanism network structure model, i.e., a local model; 502) Determine the input dimension and use the subsequent weighing weight as the output; 503) Select the mean absolute error loss function as the loss function for the model; 504) The model is trained based on the training set to obtain the weighing weight prediction model. The actual real-time data is input into the prediction model to obtain the weighing weight prediction result of the product of the rubber tire extrusion production line.

5. The intelligent prediction method for weighing after tread extrusion production line according to claim 4, characterized in that, The loss function is: in, This refers to the actual value. This refers to the predicted value.

6. The intelligent prediction method for weighing after tread extrusion production line according to claim 5, characterized in that, The local model's network structure contains only three linear layers.

Images