Production line design parameter incidence relation reasoning method and system

Abstract

The invention relates to a production line design parameter incidence relation reasoning method and system. The method comprises the following steps: constructing a production line design parameter incidence relation graph, and abstracting four-dimensional parameters such as configuration and dynamic pattern into an id node and a label edge; constructing a feature extraction model, and respectively outputting a first feature matrix containing node self and adjacent information and a second feature matrix containing four-type information through multi-head image attention self feature extraction and image attention four-type feature extraction; through third data preprocessing, a bidirectional triangular graph neural network model is built and trained to obtain an incidence relation reasoning model, and according to the incidence relation reasoning model, the parameter incidence relation is reasoned and added into the graph. The system comprises an atlas construction module, a feature extraction module and an incidence relation reasoning module. According to the method, the graph attention mechanism and the bidirectional triangular graph neural network are utilized, the problems of poor asymmetric relation modeling, causal identification loss and the like in the production line design parameter association relation are solved, and the parameter association relation can be effectively analyzed.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial production lines, and particularly relates to a method and system for inferring the correlation relationship of production line design parameters. Background Art

[0002] With the rapid development of the manufacturing industry, production lines have become complex manufacturing systems integrating high-cost investment and engineering technology intensiveness. Their complexity is mainly reflected in the multi-dimensional dynamic correlation characteristics among the four major dimensions of configuration design, motion design, control design, and optimization design. During the production line design process, a large number of parameters are involved, such as equipment speed, material flow rate, temperature, production rhythm, etc. However, the dependency relationships between these parameters are often not fully modeled, and there are even some unknown implicit correlations. Especially in the context of the current rise of the multi-variety and small-batch customized production mode, production lines need to be able to quickly adjust parameter configurations, which urgently requires accurate parameter correlation relationships to guide dynamic optimization.

[0003] In traditional production line design, excessive reliance on engineers' experience is used to handle parameter correlation relationships. This method cannot operate efficiently when facing complex parameter systems and is difficult to achieve large-scale application. At the same time, in complex production lines, it is very difficult for humans to comprehensively and accurately analyze the non-linear coupling relationships between parameters, which leads to a long trial-and-error cycle and causes a large amount of resource waste.

[0004] In terms of inferring the correlation relationships between design parameters, although it has received extensive attention in recent years, many scholars have developed various models and methods to model design parameters and infer relationships. However, most methods have limitations. For example, few scholars use systematic means, such as using graph neural networks to model design parameters and verifying the relationships through a new deep learning network architecture for two-way reasoning, to improve the correlation relationships of design parameters and make them more reliable.

[0005] Currently, as one of the existing inference methods, parameter inference driven by knowledge graphs has obvious disadvantages. First, it has insufficient ability to mine implicit relationships, has poor modeling ability for asymmetric relationships (such as "parameter A affects B, but B does not affect A"), and is difficult to capture indirect correlations across processes. Second, there are functional shortcomings at the inference mechanism level, such as the lack of reverse causality identification, the inability to effectively learn from data, and the lack of self-verification ability.

[0006] Therefore, the present invention provides a method and system for inferring the correlation relationship of production line design parameters to solve the above technical problems. Summary of the Invention

[0007] In view of the above problems, the purpose of the present invention is to provide a method and system for inferring the correlation relationship of production line design parameters, which are used to solve problems such as poor modeling of asymmetric relationships and loss of causal identification in the correlation relationship of production line design parameters.

[0008] The present invention provides a method for inferring the correlation relationship of production line design parameters, including: S1, constructing a correlation relationship graph of production line design parameters; using a visualization tool to abstract the design parameters of the four dimensions of the configuration, motion type, control type, and optimization type of the production line into nodes with unique ids, and labeling the classification attributes of each node according to the four-type category; and abstracting the mutual correlation relationships between the design parameters into edges with labels; based on the node ids, four-type category attributes, and edges with labels, construct a production line correlation relationship graph containing node feature attributes and edge relationship labels;

[0009] S2, constructing a feature extraction model, including:

[0010] Self-feature extraction of multi-head graph attention:

[0011] First data preprocessing: Sort all nodes in ascending order of node ids and mark labels, and construct a first adjacency matrix according to the correlation relationship between the head node and the tail node;

[0012] First model construction: Adopt a multi-head attention network, convert the node ids into a matrix through an embedding layer, and construct four layers of multi-head graph attention convolutional layers as the first initial model; among them, the activation functions connected to the first three convolutional layers of the four layers of multi-head graph attention convolutional layers are LeakyRelu, and the activation function of the fourth convolutional layer of the four layers of multi-head graph attention convolutional layers is Softmax;

[0013] First model training: Input the sorted node ids and the first adjacency matrix into the first initial model, and train the first initial model based on the cross-entropy loss function and the Adam optimizer to obtain a self-feature extraction model of the multi-head graph attention mechanism. The self-feature extraction model of the multi-head graph attention mechanism outputs a first feature matrix containing the node's own features and the correlation information of adjacent nodes;

[0014] Four-type feature extraction of graph attention:

[0015] Second data preprocessing: Mark labels according to the four-type category to which the nodes belong, and construct a second adjacency matrix;

[0016] Second model construction: Adopt a general graph attention mechanism, and construct five layers of graph attention convolutional layers as the second initial model; among them, the activation functions connected to the first four convolutional layers of the five layers of graph attention convolutional layers are LeakyRelu, and the activation function of the fifth convolutional layer of the five layers of graph attention convolutional layers is Softmax;

[0017] Second model training: Input the first feature matrix and the second adjacency matrix into the second initial model, and train the second initial model based on the cross-entropy loss function and the Adam optimizer to obtain a type-four feature extraction model of the graph attention mechanism. The type-four feature extraction model of the graph attention mechanism outputs a second feature matrix containing type-four information;

[0018] S3. Inference on the correlation relationship of production line design parameters:

[0019] Third data preprocessing: Rearrange the second feature matrix in ascending order of node id, and abstract the first adjacency matrix and the second adjacency matrix into undirected graphs; Traverse the undirected graph through the breadth-first search algorithm (BFS) to obtain the shortest path with a path length greater than 2 between any two nodes. Determine the intermediate nodes based on the shortest path, extract the second feature matrices corresponding to the head-intermediate-tail three nodes and splice them in order into a feature vector; Use the feature vector containing the labeled correlation relationship between the head node and the tail node as learning data, and the rest as inference data; Perform label marking, feature matrix swapping, and reverse label calculation on the learning data, and divide the learning data into a training set, a validation set, and a test set according to a ratio of 8:1:1. Synchronously perform reverse feature matrix processing on the inference data;

[0020] Third model construction: Extract the second feature matrices corresponding to the positions of the head node, the tail node, and the intermediate node from the learning data; Calculate the node feature values and weight coefficients based on the extracted second feature matrices; Provide a bidirectional triangular graph neural network model composed of seven fully connected layers and one pooling layer as the third initial model; Input the second feature matrices of the head node, the tail node, and the intermediate node into three independent fully connected layers respectively. The activation function connected by the three fully connected layers is LeakyRelu; Input the matrices output by the three fully connected layers into the pooling layer, and introduce the correction coefficients of the two end nodes and the correction coefficient of the intermediate node in the pooling layer to enhance the feature importance of the head node and the tail node. After pooling, generate a single matrix; Pass the pooled single matrix through four fully connected layers in sequence. Among them, the activation functions connected by the first three fully connected layers of the four fully connected layers are LeakyRelu, and the activation function connected by the fourth fully connected layer is Softmax;

[0021] Third model training: Input the training set into the third initial model, and train the third initial model based on the cross-entropy loss function, the Adam optimizer, the StepLR learning rate decay strategy, the Dropout regularization method, and the LayerNorm regularization method to obtain a correlation relationship inference model;

[0022] Association relationship reasoning: Input the reasoning data and the reasoning data processed by the inverse feature matrix into the association relationship reasoning model respectively, obtain the results and add them. If the sum of the two is N (N is the type of association relationship), then there is an association relationship with the corresponding label between the two nodes, otherwise there is no such relationship; Infer the head node id, tail node id and relationship name with existing relationships through the association relationship reasoning model, and add them to the graph spectrum using the function statements in the visualization tool.

[0023] Preferably, the expression of the activation function LeakyRelu is: LeakyReLU(x) = max(0, x) + α·min(0, x); where α is the slope controlling the negative activation value and x is the input matrix.

[0024] Preferably, the expression of the activation function Softmax is: where, X i is the i-th element in the input matrix, X j is the j-th element in the input matrix, and K is the number of columns of the input matrix.

[0025] Preferably, the expression of the cross-entropy loss function is: where, N is the total number of categories, y i is the encoding of the true label, and P i is the probability value of the i-th category predicted by the model.

[0026] Preferably, the calculation expression of the eigenvalue is: T = <x.x>, where X is the second feature matrix of the node; the calculation expression of the weight coefficient is: where T i is the eigenvalue of the i-th node.

[0027] Preferably, the expression of the fully connected layer is: y = W·x + b; where, W is the weight matrix, x is the input matrix, and b is the bias vector; the expression of the pooling layer is: y = a ht ·w h ·x h + α m ·w m ·x m + α ht ·w t ·x t ; where, α ht is the correction coefficient of the two end nodes equal to 0.45, α m is the correction coefficient of the middle node equal to 0.1, W h is the weight coefficient of the head node, W m is the weight coefficient of the middle node, W t is the weight coefficient of the tail node, and x his the second feature matrix corresponding to the head node, x m is the second feature matrix corresponding to the intermediate node, x t is the second feature matrix corresponding to the tail node.

[0028] Preferably, the expression of the StepLR learning rate decay strategy is: where lr initial is the initial learning rate, γ is the learning rate decay factor, e is the index of the number of training times, s is the learning rate decay step size; the expression of the Dropout regularization method is: where p is the dropout probability; m is the mask matrix, taking values of 0 or 1, following the Bernoulli distribution; x is the input matrix; the expression of the LayerNorm regularization method is: where γ1 is the learnable parameter; μ is the mean of x, is the standard deviation with numerical stability terms, and β is the learnable offset coefficient.

[0029] Preferably, the visualization tool is neo4j.

[0030] The present invention also provides a system for analyzing the correlation relationship of production line design parameters, including:

[0031] A graph construction module for performing step S1 in claim 1 to construct a correlation relationship graph of production line design parameters;

[0032] A feature extraction module for performing step S2 in claim 1 to construct a feature extraction model, including a multi-head graph attention self-feature extraction unit and a graph attention type-four feature extraction unit;

[0033] A correlation relationship reasoning module for performing step S3 in claim 1 to perform correlation relationship reasoning on production line design parameters, including a third data preprocessing unit, a third model building unit, a third model training unit, and a correlation relationship reasoning unit.

[0034] Compared with the related technologies, a method and system for inferring the correlation relationship of production line design parameters provided by the present invention have the following beneficial effects: 1. Strong ability to model asymmetric relationships: Utilizing the graph attention mechanism and the bidirectional triangular graph neural network, it breaks through the modeling limitations of traditional methods for asymmetric parameter relationships and accurately captures the complex dependencies between parameters; 2. Complete identification of causal relationships: Combining multi-head graph attention and four-type feature extraction, systematically integrating the node itself and four-dimensional category information to avoid the loss of causal relationships in the analysis; 3. Excellent self-verification and generalization ability: Through reverse feature matrix processing and two-way input verification of inference data, ensuring the reliability of the model results; 4. Combining regularization methods such as Dropout and LayerNorm to enhance the model's ability to learn from data and generalize to different scenarios; 5. Efficient analysis and visualization: Based on the neo4j visualization graph, realizing the intuitive presentation and dynamic update of parameter relationships, reducing the complexity of the correlation analysis of production line design parameters, improving the design optimization efficiency, and providing strong support for the collaborative design of intelligent manufacturing production line parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic flow chart of a method for inferring the correlation relationship of production line design parameters of the present invention;

[0036] Figure 2 It is a schematic diagram of the correlation relationship graph described in the present invention;

[0037] Figure 3 It is a schematic diagram of the structure of the self-feature extraction model of the multi-head graph attention mechanism described in the present invention

[0038] Figure 4 It is a schematic diagram of the structure of the four-type feature extraction model of the graph attention mechanism described in the present invention;

[0039] Figure 5 It is a schematic diagram of the head, tail nodes and intermediate nodes described in the present invention;

[0040] Figure 6 It is a schematic diagram of the structure of the bidirectional triangular graph neural network inference model described in the present invention;

[0041] Figure 7 It is a schematic diagram of the structure of an analysis system for the correlation relationship of production line design parameters of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] The present invention provides a method and system for inferring the correlation relationship of production line design parameters, aiming to solve the problems such as poor modeling of asymmetric relationships and loss of causal recognition in the existing correlation relationship of production line design parameters.

[0043] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0044] Please refer to the attached Figure 1-7 As shown, the present invention provides a method for inferring the correlation relationship of production line design parameters, including: S1, constructing a correlation relationship graph of production line design parameters; using a visualization tool to abstract the design parameters in the four dimensions of the production line configuration, motion type, control type, and optimization type into nodes with unique ids, and labeling the classification attributes of each node according to the four-type category; and abstracting the mutual correlation relationships between the design parameters into edges with labels; based on the node ids, four-type category attributes, and edges with labels, construct a production line correlation relationship graph including node feature attributes and edge relationship labels.

[0045] In this embodiment, the visualization tool uses the neo4j graph database to complete the construction and visualization display of the production line correlation relationship graph. Of course, the visualization tool described in the present invention is not limited to neo4j, and any graph database or visualization platform (such as TigerGraph, Graphviz, etc.) that can implement node attribute annotation, edge relationship labeling, and graph dynamic update is applicable to the technical solution of the present invention.

[0046] S2, constructing a feature extraction model, including multi-head graph attention self-feature extraction and graph attention four-type feature extraction. The aim is to transform the original design parameter data into a low-dimensional space mapping with stronger representation ability for subsequent correlation relationship inference.

[0047] It should be noted that the purpose of multi-head graph attention self-feature extraction is to map the id data of the nodes in the graph to a low-dimensional space, thereby increasing the information it can accommodate; at the same time, considering the correlation relationship between the node and its adjacent nodes, in order to facilitate subsequent inference, some useful information of the adjacent nodes needs to be added to its own features. However, after the extraction of the node's own features, since the four-type information of the node is not included in the feature matrix of the node, and the correlation relationship of the design parameters is affected by the four-type, after the multi-head graph attention self-feature extraction is completed, it is also necessary to adopt the graph attention four-type feature extraction method to extract the four-type features of the node.

[0048] Specifically, the steps of extracting the self-features of the multi-head graph attention include: First data preprocessing: All nodes are sorted in ascending order of node ID and labeled, and a first adjacency matrix is constructed according to the association relationship between the head node and the tail node; First model construction: A multi-head attention network is adopted, and the node ID is converted into a matrix through an embedding layer, and four layers of multi-head graph attention convolutional layers are built as the first initial model; among them, the activation function connected to the first three convolutional layers of the four layers of multi-head graph attention convolutional layers is LeakyRelu, and the activation function of the fourth convolutional layer of the four layers of multi-head graph attention convolutional layers is Softmax; First model training: The sorted node ID and the first adjacency matrix are input into the first initial model, and the first initial model is trained based on the cross-entropy loss function and the Adam optimizer to obtain a self-feature extraction model of the multi-head graph attention mechanism. The self-feature extraction model of the multi-head graph attention mechanism outputs a first feature matrix containing the self-features of the nodes and the association information of adjacent nodes.

[0049] Further, the steps of extracting the four-type features of the graph attention include: Second data preprocessing: Labels are marked according to the four-type categories to which the nodes belong, and a second adjacency matrix is constructed; Second model construction: A common graph attention mechanism is adopted, and five layers of graph attention convolutional layers are built as the second initial model; among them, the activation function connected to the first four convolutional layers of the five layers of graph attention convolutional layers is LeakyRelu, and the activation function of the fifth convolutional layer of the five layers of graph attention convolutional layers is Softmax; Second model training: The first feature matrix and the second adjacency matrix are input into the second initial model, and the second initial model is trained based on the cross-entropy loss function and the Adam optimizer to obtain a four-type feature extraction model of the graph attention mechanism. The four-type feature extraction model of the graph attention mechanism outputs a second feature matrix containing four-type information.

[0050] In this embodiment, the expression of the activation function LeakyRelu is: LeakyReLU(x) = max(0, x) + α·min(0, x); where α is the slope controlling the negative activation value, and x is the input matrix; the expression of the activation function Softmax is: where, X i is the i-th element in the input matrix, X j is the j-th element in the input matrix, K is the number of columns of the input matrix, and the Softmax function can convert an unnormalized matrix (usually the output of the model) into a probability distribution, so that each value in the output is between 0 and 1, and the sum of all outputs is 1; the expression of the cross-entropy loss function is: where, N is the total number of categories, y i is the encoding of the true label, P iis the probability value of the i-th class predicted by the model, that is, the output after Softmax normalization. In addition, the Adam optimizer is used to accelerate the convergence rate of the model loss value.

[0051] S3, Inference of the correlation relationship of production line design parameters. The function of this step is to realize the inference of the correlation relationship between parameters through a bidirectional triangular graph neural network, focusing on solving the problems of the influence of intermediate nodes and the neglect of relationship directivity in traditional graph neural networks.

[0052] Further, the steps of reasoning about the correlation relationship of the production line design parameters specifically include: S31 Third data preprocessing: Rearrange the second feature matrix in ascending order of node id, and abstract the first adjacency matrix and the second adjacency matrix into undirected graphs; Traverse the undirected graph through the breadth-first search algorithm (BFS) to obtain the shortest path with a path length greater than 2 between any two nodes, determine the intermediate nodes based on the shortest path, extract the second feature matrices corresponding to the head-middle-tail three nodes and splice them in order into a feature vector; Use the feature vector containing the labeled correlation relationship between the head node and the tail node as learning data, and the rest as inference data; Perform label marking, feature matrix swapping, and reverse label calculation on the learning data, divide the learning data into a training set, a validation set, and a test set according to a ratio of 8:1:1, and synchronously perform reverse feature matrix processing on the inference data; S32, Third model construction: Extract the second feature matrices corresponding to the positions of the head node, the tail node, and the intermediate node from the learning data; Calculate the eigen values and weight coefficients of each node based on the extracted second feature matrices; Provide a bidirectional triangular graph neural network model composed of seven fully connected layers and one pooling layer as the third initial model; Input the second feature matrices of the head node, the tail node, and the intermediate node into three independent fully connected layers respectively, and the activation function connected by the three fully connected layers is LeakyRelu; Input the matrices output by the three fully connected layers into the pooling layer, introduce the correction coefficients of the two end nodes and the correction coefficient of the intermediate node in the pooling layer to enhance the feature importance of the head node and the tail node, and generate a single matrix after pooling; Pass the pooled single matrix through four fully connected layers in sequence, where the activation functions connected by the first three fully connected layers of the four fully connected layers are LeakyRelu, and the activation function connected by the fourth fully connected layer is Softmax; S33, Third model training: Input the training set into the third initial model, and train the third initial model based on the cross-entropy loss function, the Adam optimizer, the StepLR learning rate decay strategy, the Dropout regularization method, and the LayerNorm regularization method to obtain a correlation relationship inference model; S34, Correlation relationship inference: Input the inference data and the inference data processed by the reverse feature matrix into the correlation relationship inference model respectively, add the obtained results, if the sum of the two is N (N is the type of correlation relationship), then there is a correlation relationship with the corresponding label between the two nodes, otherwise there is no such relationship; Infer the head node id, tail node id, and relationship name with an existing relationship through the correlation relationship inference model, and add them to the graph spectrum using the function statements in the visualization tool.

[0053] In this embodiment, the expression of reverse label calculation is: y -= 2N1 - 1 - y2, where: y- is the reverse label, N1 is the label type, and y2 is the label value. The calculation expression of the eigenvalue is: T = <x.x>, where X is the second feature matrix of the node; the calculation expression of the weight coefficient is: where Ti is the eigenvalue of the i-th node. The expression of the fully connected layer is: y = W·x + b; where, W is the weight matrix, x is the input matrix, and b is the bias vector; the expression of the pooling layer is: y = α ht ·w h ·x h +α m ·w m ·x m +α ht ·w t ·x t ; where, α ht is the correction coefficient of the two end nodes equal to 0.45, α m is the correction coefficient of the middle node equal to 0.1, W h is the weight coefficient of the head node, W m is the weight coefficient of the middle node, W t is the weight coefficient of the tail node, x h is the second feature matrix corresponding to the head node, x m is the second feature matrix corresponding to the middle node, x t is the second feature matrix corresponding to the tail node.

[0054] It is worth mentioning that the expression of the StepLR learning rate decay strategy is: where, lr initial is the initial learning rate, γ is the learning rate decay factor, e is the index of the number of training times, and s is the learning rate decay step; the expression of the Dropout regularization method is: where, p is the dropout probability; m is the mask matrix, taking values of 0 or 1, following the Bernoulli distribution; x is the input matrix; the expression of the LayerNorm regularization method is: where, γ1 is the learnable parameter; μ is the mean of x, is the standard deviation with a numerical stability term, and β is the learnable offset coefficient.

[0055] The present invention also provides a production line design parameter correlation relationship analysis system 100, including:

[0056] A graph construction module 10, configured to execute step S1 in claim 1 to construct a correlation relationship graph of production line design parameters;

[0057] The feature extraction module 20 is used to execute step S2 in claim 1, construct a feature extraction model, including a multi-head graph attention self-feature extraction unit 201 and a graph attention four-type feature extraction unit 202;

[0058] The association relationship reasoning module 30 is used to execute step S3 in claim 1, perform the association relationship reasoning of production line design parameters, including a third data preprocessing unit 301, a third model building unit 302, a third model training unit 303, and an association relationship reasoning unit 304.

[0059] Compared with the related technologies, the association relationship reasoning method and system for production line design parameters provided by the present invention have the following beneficial effects: 1. Strong asymmetric relationship modeling ability: Using the graph attention mechanism and the bidirectional triangular graph neural network, it breaks through the modeling limitations of traditional methods for asymmetric parameter relationships and accurately captures the complex dependencies between parameters; 2. Complete causal relationship recognition: Combining multi-head graph attention and four-type feature extraction, systematically integrating node self and four-dimensional category information to avoid the loss of causal relationships in the analysis; 3. Excellent self-verification and generalization ability: Through reverse feature matrix processing and bidirectional input verification of inference data, ensure the reliability of the model results; 4. Combine regularization methods such as Dropout and LayerNorm to enhance the model's ability to learn from data and generalize to different scenarios; 5. Efficient analysis and visualization: Based on the neo4j visualization graph, achieve an intuitive presentation and dynamic update of parameter relationships, reduce the complexity of the association analysis of production line design parameters, improve the design optimization efficiency, and provide strong support for the collaborative design of intelligent manufacturing production line parameters.

[0060] The above-described embodiments should be understood as illustrative and not limiting the protection scope of the present invention. The protection scope of the present invention is subject to the claims. For those skilled in the art, without departing from the essence and scope of the present invention, some non-essential improvements and adjustments made to the present invention still fall within the protection scope of the present invention.

Claims

1. A method for inferring the correlation relationship of production line design parameters, characterized in that The method includes: S1. Construct an association relationship graph of production line design parameters; use a visualization tool to abstract the design parameters of the four dimensions of the production line configuration, motion type, control type, and optimization type into nodes with unique IDs, and label the classification attributes of each node according to the four-type category; and abstract the mutual association relationships between the design parameters into edges with labels; based on the node IDs, four-type category attributes, and edges with labels, construct a production line association relationship graph containing node feature attributes and edge relationship labels; S2. Construction of a feature extraction model, including: Self-feature extraction of multi-head graph attention: First data preprocessing: Arrange all nodes in ascending order of node ID and mark the labels, and construct a first adjacency matrix according to the association relationship between the head node and the tail node; First model construction: Adopt a multi-head attention network, convert the node ID into a matrix through an embedding layer, and construct four-layer multi-head graph attention convolutional layers as the first initial model; among them, the activation functions connected to the first three convolutional layers of the four-layer multi-head graph attention convolutional layers are LeakyRelu, and the activation function of the fourth convolutional layer of the four-layer multi-head graph attention convolutional layers is Softmax; First model training: Input the sorted node IDs and the first adjacency matrix into the first initial model, and train the first initial model based on the cross-entropy loss function and the Adam optimizer to obtain a self-feature extraction model of the multi-head graph attention mechanism. The self-feature extraction model of the multi-head graph attention mechanism outputs a first feature matrix containing the node's own features and the associated information of adjacent nodes; Four-type feature extraction of graph attention: Second data preprocessing: Mark the labels according to the four-type category to which the node belongs, and construct a second adjacency matrix; Second model construction: Adopt a general graph attention mechanism and construct five-layer graph attention convolutional layers as the second initial model; among them, the activation functions connected to the first four convolutional layers of the five-layer graph attention convolutional layers are LeakyRelu, and the activation function of the fifth convolutional layer of the five-layer graph attention convolutional layers is Softmax; Second model training: Input the first feature matrix and the second adjacency matrix into the second initial model, and train the second initial model based on the cross-entropy loss function and the Adam optimizer to obtain a four-type feature extraction model of the graph attention mechanism. The four-type feature extraction model of the graph attention mechanism outputs a second feature matrix containing four-type information; S3. Association relationship reasoning of production line design parameters Third data preprocessing: Rearrange the second feature matrix in ascending order of node IDs, and abstract the first adjacency matrix and the second adjacency matrix into undirected graphs; traverse the undirected graphs through the breadth-first search algorithm (BFS) to obtain the shortest paths with a path length greater than 2 between any two nodes, determine intermediate nodes based on the shortest paths, extract the second feature matrices corresponding to the head-middle-tail three nodes and splice them in order into feature vectors; use the feature vectors containing the labeled association relationships between the head node and the tail node as learning data, and the rest as inference data; perform label marking, feature matrix swapping, and reverse label calculation on the learning data, divide the learning data into a training set, a validation set, and a test set according to a ratio of 8:1:1, and perform reverse feature matrix processing on the inference data synchronously; Third model construction: Extract the second feature matrices corresponding to the head node, tail node, and intermediate node positions from the learning data; calculate the node eigenvalues and weight coefficients based on the extracted second feature matrices; provide a bidirectional triangular graph neural network model composed of seven fully connected layers and one pooling layer as the third initial model; input the second feature matrices of the head node, tail node, and intermediate node into three independent fully connected layers respectively, and the activation function connected by the three fully connected layers is LeakyRelu; input the matrices output by the three fully connected layers into the pooling layer, introduce the correction coefficients of the two end nodes and the correction coefficient of the intermediate node in the pooling layer to enhance the feature importance of the head node and the tail node, and generate a single matrix after pooling; pass the pooled single matrix through four fully connected layers in sequence, where the activation functions connected by the first three of the four fully connected layers are LeakyRelu, and the activation function connected by the fourth fully connected layer is Softmax; Third model training: Input the training set into the third initial model, and train the third initial model based on the cross-entropy loss function, Adam optimizer, StepLR learning rate decay strategy, Dropout regularization method, and LayerNorm regularization method to obtain an association relationship inference model; Association relationship inference: Input the inference data and the inference data processed by the reverse feature matrix into the association relationship inference model respectively, add the obtained results, if the sum of the two is N (N is the type of association relationship), then there is an association relationship with the corresponding label between the two nodes, otherwise there is no such relationship; infer the head node ID, tail node ID, and relationship name with an existing relationship through the association relationship inference model, and add them to the graph through the function statements in the visualization tool.

2. The inference method for the correlation relationship of production line design parameters according to claim 1, wherein The expression of the activation function LeakyRelu is: LeakyReLU(x) = max(0, x) + α·min(0, x); where α is the slope controlling the negative activation value, and x is the input matrix.

3. A method for inferring the correlation relationship of production line design parameters according to claim 1, characterized in that The expression of the activation function Softmax is as follows: where, X i is the i-th element in the input matrix, and X j is the j-th element in the input matrix, and K is the number of columns of the input matrix.

4. A method for inferring the correlation relationship of production line design parameters according to claim 3, characterized in that The expression of the cross-entropy loss function is: where N is the total number of categories, y i is the encoding of the true label, and P i is the probability value of the i-th category predicted by the model.

5. A method for inferring the correlation relationship of production line design parameters according to claim 4, characterized in that The calculation expression of the eigenvalue is: T = <x.x>, where X is the second feature matrix of the node; the calculation expression of the weight coefficient is: where T i is the eigenvalue of the i-th node.

6. The inference method for the correlation relationship of production line design parameters according to claim 5, wherein , the expression of the fully connected layer is: y = W x + b; where W is the weight matrix, x is the input matrix, and b is the bias vector; the expression of the convergence layer is: y = α ht w t ·x h +α m w m ·x m +α ht w t ·x t ; Among them, α ht The correction coefficient of the two end nodes is equal to 0.45, α m The correction factor for the middle node is equal to 0.1, W h is the first node weight coefficient, W m is the weight coefficient of the intermediate node, W t is the tail node weight coefficient, x h is the second characteristic matrix corresponding to the first node, x m is the second characteristic matrix corresponding to the intermediate node, x t is the second characteristic matrix corresponding to the tail node.

7. A method for inferring the correlation relationship of production line design parameters according to claim 6, characterized in that The expression of the StepLR learning rate decay strategy is: Among them, lr initial is the initial learning rate, γ is the learning rate attenuation factor, e is the index of the number of training times, and s is the learning rate attenuation step size; the expression of the Dropout regularization method is: Where p is the drop probability; m is the mask matrix, which takes a value of 0 or 1 and obeys the Bernoulli distribution; x is the input matrix; the expression of the LayerNorm regularization method is: Among them, γ1 is a learnable parameter; μ is the mean of x, is the standard deviation of the numerical stability term, and β is the learnable offset coefficient.

8. A method for inferring the correlation relationship of production line design parameters according to claim 1, characterized in that The visualization tool is neo4j.

9. An analysis system for the correlation relationship of production line design parameters, characterized in that Including: A graph construction module for performing step S1 in claim 1 to construct an association relationship graph of production line design parameters; A feature extraction module, configured to execute step S2 of claim 1 and construct a feature extraction model, including a multi-head graph attention self-feature extraction unit and a graph attention four-type feature extraction unit; An association relationship reasoning module is used to execute step S3 in claim 1 to perform association relationship reasoning of production line design parameters, including a third data preprocessing unit, a third model building unit, a third model training unit and an association relationship reasoning unit.

Images