Material cladding layer performance prediction method and system based on big data

Abstract

The present invention discloses a material cladding layer performance prediction method and system based on big data, and the method includes: data collection, data preprocessing, feature information processing, building a material cladding layer performance prediction model and material cladding layer performance prediction. The present invention belongs to the field of data processing technology. The scheme generates anomaly values, calculates correlation factors, primary increments and advanced increments, designs dynamic fluctuation factors, updates individual positions, and determines the best feature subset according to the fitness value threshold and the maximum number of searches; generates nodes and sets initial weights and embedding vectors, calculates weighted Euclidean distance and average weighted similarity, sets update status labels and update status factors, updates important thresholds, screens out important neighbor nodes, updates the node embedding vectors and weights, trains weak classifiers and weighted combinations, finds the feature subset that is most conducive to material cladding layer performance prediction, and improves the accuracy and reliability of material cladding layer performance prediction.

Description

Technical Field

[0001] The present invention belongs to the field of data processing technology, and specifically refers to a material cladding layer performance prediction method and system based on big data. Background Art

[0002] The material cladding layer performance prediction method utilizes the powerful data analysis and processing capabilities of big data technology, collects and analyzes a large amount of data related to the material cladding layer, predicts the performance of the material cladding layer, and provides a scientific basis for the optimization of the material cladding process, which is beneficial to improving product quality and manufacturing efficiency. However, the existing material cladding layer performance prediction method has the problem that the material cladding layer data itself contains a large number of features, and the importance and correlation of different features are unclear, making it difficult to capture truly meaningful material cladding layer features, resulting in difficulty in accurately determining the optimal material cladding layer features and unable to accurately predict the actual material cladding layer performance; the existing material cladding layer performance prediction method has the problem that the material cladding layer data has a large number of complex nonlinear relationships, which are difficult to effectively capture, resulting in a large deviation between the prediction results and the actual situation, reducing the reliability of the model in predicting the cladding layer performance. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a material cladding layer performance prediction method and system based on big data. In view of the problem that the material cladding layer data itself contains a large number of features in the existing material cladding layer performance prediction method, and the importance and correlation of different features are unclear, it is difficult to capture the truly meaningful material cladding layer features, resulting in difficulty in accurately determining the optimal material cladding layer features and unable to accurately predict the actual material cladding layer performance, this scheme uses individual positions to represent feature subsets, which provides a wide range of feature selection possibilities for material cladding layer performance prediction; generates abnormal values, and calculates correlation factors, primary increments and advanced increments to finely capture the complex relationships and dynamic changes between material cladding layer features; designs dynamic fluctuation factors to better adapt to the characteristics and changes of material cladding layer data; updates individual positions to improve the quality of feature subsets and prediction accuracy; determines the optimal feature subset according to the fitness value threshold and the maximum number of searches, and constructs a prediction data The accuracy and reliability of the prediction of the performance of the material cladding layer are improved by collecting data from the data set; the existing material cladding layer performance prediction methods have a large number of complex nonlinear relationships in the material cladding layer data, which are difficult to effectively capture, resulting in a large deviation between the prediction results and the actual situation, and reducing the reliability of the cladding layer performance prediction. This scheme generates nodes and sets initial weights and embedding vectors. By calculating the weighted Euclidean distance and the average weighted similarity, the complex patterns in the material cladding layer data can be better captured; the update state label and update state factor are set to better adapt to the distribution and changes of different material cladding layer data; the important threshold is updated, the important neighbor nodes are screened out, and the key cladding layer characteristics and relationships are grasped; the embedding vectors and weights of the nodes are updated to gradually improve the prediction ability of the material cladding layer performance; the weak classifiers are trained, and the material cladding layer performance prediction model is obtained by weighted combination, which can better cope with the characteristics and changes of different material cladding layer data and improve the accuracy and effect of the cladding layer performance prediction.

[0004] The technical solution adopted by the present invention is as follows: The material cladding layer performance prediction method based on big data provided by the present invention comprises the following steps:

[0005] Step S1: data collection;

[0006] Step S2: data preprocessing;

[0007] Step S3: feature information processing;

[0008] Step S4: constructing a material cladding layer performance prediction model;

[0009] Step S5: predicting the performance of the material cladding layer.

[0010] Furthermore, in step S1, the data collection is to collect historical material cladding layer data; the historical material cladding layer data includes material composition data, process parameter data, environmental condition data, cladding layer geometry data and performance level; and the performance level is used as a data label.

[0011] Furthermore, in step S2, the data preprocessing is to perform data cleaning, data conversion and data normalization on the collected data; data cleaning includes processing noise, outliers, duplicate values ​​and missing values; data conversion is to convert the data into vector form; data normalization is to unify the data range based on the maximum and minimum normalization method.

[0012] Furthermore, in step S3, the feature information processing specifically includes the following steps:

[0013] Step S31: Initialize individual positions, establish a parameter search space based on the dimension information of the preprocessed material cladding layer data, randomly initialize N individual positions in the parameter search space, use each individual position to represent a feature subset scheme, and use the information gain of each feature subset scheme as the fitness value of the corresponding individual position. The individual dimension is equal to the material cladding layer data dimension. When the dimension value of the individual position is greater than 0.5, the feature subset scheme corresponding to the individual position contains the feature data corresponding to the dimension; otherwise, the feature subset scheme corresponding to the individual position does not contain the feature data corresponding to the dimension;

[0014] Step S32: Calculate the change value, generate a change value for each individual position, the dimension of the change value is equal to the individual dimension, and then calculate the correlation factor based on the change value , primary increment and Advanced Incremental , the formula used to generate the change value is as follows:

[0015] ;

[0016] Where μ is the control value, and are the abnormal values ​​corresponding to the w+1th and wth individual positions in the vth dimension during the tth search, , and are the correlation factor, primary increment, and advanced increment corresponding to the w-th individual position in the v-th dimension during the t-th search, respectively. t is the search number index, w is the individual position index, and v is the dimension index;

[0017] Step S33: Design a dynamic fluctuation factor, the formula used is as follows:

[0018] ;

[0019] In the formula, is the average value of the correlation factor corresponding to the wth individual position in all dimensions during the tth search, G is the adaptive factor, is the dynamic fluctuation factor corresponding to the wth individual position in the vth dimension during the tth search;

[0020] Step S34: Update individual positions, using the following formula:

[0021] ;

[0022] In the formula, and are the values ​​of the w-th individual position in the v-th dimension during the t+1-th and t-th searches, respectively. is the value of the global optimal position in the vth dimension during the tth search;

[0023] Step S35: Determine the optimal feature subset, pre-set the fitness value threshold and the maximum number of searches, update the fitness value of the individual position, and when there is an individual position whose fitness value is greater than the fitness value threshold, select the feature subset scheme corresponding to the individual position as the optimal feature subset scheme, and construct the prediction data set according to the optimal feature subset scheme; otherwise, if the maximum number of searches is reached, go to step S31; otherwise, add 1 to the search number and go to step S32.

[0024] Further, in step S4, the building of the material cladding layer performance prediction model specifically includes the following steps:

[0025] Step S41: Node initialization, including the following steps:

[0026] Step S411: Generate a node for each material cladding layer data in the prediction data set, and set the initial weight of each node to , the characteristic vector of the node is the corresponding material cladding layer data, and the node set S and the characteristic matrix Z are obtained. The characteristic matrix is ​​linearly transformed and activated to obtain the initial embedding vector of the node. The formula used is as follows:

[0027] ;

[0028] In the formula, is the initial embedding vector of the ith node, σ(•) is the activation function, W is the learnable weight matrix, is the feature vector of the i-th node in the feature matrix Z, i is the node index, and Q is the number of nodes;

[0029] Step S412: Calculate the Euclidean distance between any two nodes, and use the value obtained by subtracting the Euclidean distance from 1 as the similarity. Preset a similarity threshold. If the similarity between the two nodes is greater than the similarity threshold, there is an edge between the two nodes; otherwise, there is no edge between the two nodes; the number of edges is recorded as F; and a neighbor node set C of each node is constructed based on the nodes connected by the edges.

[0030] Step S42: Screening of important neighbor nodes, including the following steps:

[0031] Step S421: Calculate the weighted Euclidean distance between any two nodes based on the node weights and embedding vectors, subtract the weighted Euclidean distance from 1 as the weighted similarity, and calculate the average weighted similarity between all nodes and their neighboring nodes. The formula used is as follows:

[0032] ;

[0033] ;

[0034] In the formula, is the average weighted similarity between all nodes in the b-th convolutional layer and their neighbor nodes, is the number r in the b-th convolutional layer i and r j The similarity between Yes i The neighbor node set, b is the convolution layer index, r i and r j are the i-th and j-th nodes respectively, is the number r in the b-th convolutional layer i and r j The weighted Euclidean distance between and They are r in the b-th convolutional layer. i and r j The weight of and They are r in the b-th convolutional layer. i and r j The embedding vector of , j is the node index, is the L2 norm;

[0035] Step S422: Set the update status, pre-set the update status label and update status factor, the initial values ​​of which are g1 and 0 respectively; if , then the update state label of the b-th convolutional layer is reset to g1, and the value of the update state factor is reset to 1; otherwise, the update state label of the b-th convolutional layer is reset to g2, and the value of the update state factor is reset to -1; where, is the average similarity between all nodes in the b-1th convolutional layer and their neighboring nodes;

[0036] Step S423: Update the important threshold, pre-set a fixed value β, set the initial value of the important threshold of each convolutional layer to h, if the update status label of the b-th convolutional layer is g1, then modify the important threshold of the b-th convolutional layer to Otherwise, the important threshold of the b-th convolutional layer is modified to ;

[0037] Step S424: construct an important neighbor node set, and construct an important neighbor node set U for each node in each convolutional layer based on the nodes whose similarity is greater than the important threshold;

[0038] Step S43: Update the embedding vector of the node. Update the embedding vector of the node in each convolutional layer based on the set of important neighbor nodes. The formula used is as follows:

[0039] ;

[0040] In the formula, is the number r in the b+1th convolutional layer i The embedding vector of , MEAN(•) is the function to calculate the average value, is the number r in the b-th convolutional layer a The embedding vector signature, r a is the a-th node, a is the node index, is the number r in the b-th convolutional layer i The set of important neighbor nodes of is the union operator;

[0041] Step S44: Update the node weights and normalize the embedding vector of each node using the Softmax function to obtain the probability vector of the node belonging to each data label. , select the data label with the highest probability as the classification label of the node, and update the weight of the node. The formula used is as follows:

[0042] ;

[0043] In the formula, is the number r in the b+1th convolutional layer i The weight of , P is the number of data labels, Yes i The true data label, is the number of pixels in the b-th convolutional layer. i The classification label of is in the bth convolutional layer r i The probability vector belonging to each data label, , and They are respectively r in the b-th convolutional layer i The probability of data labels belonging to category 1, category p, and category P;

[0044] Step S45: training. Based on the updated node weights, return to step S42 to continue training the next convolution layer. There are B convolution layers in total. Each convolution layer is used as a weak classifier to obtain B weak classifiers. The weight of each weak classifier is calculated. The formula used is as follows:

[0045] ;

[0046] In the formula, α b is the weight of the b-th weak classifier, ε b is the error rate of the bth weak classifier;

[0047] Step S46: Combining, weighted combination of B weak classifiers to obtain a material cladding layer performance prediction model, taking the data label with the highest probability as the prediction label, the formula used is as follows:

[0048] ;

[0049] Where Y i is the output of the material cladding layer performance prediction model i The predicted label of .

[0050] Furthermore, in step S5, the material cladding layer performance prediction is to collect real-time material cladding layer data, and the real-time material cladding layer data includes material composition data, process parameter data, environmental condition data and cladding layer geometry data; after preprocessing, the data is input into the material cladding layer performance prediction model for processing, and the performance level of the material cladding layer is obtained based on the output prediction label.

[0051] The material cladding layer performance prediction system based on big data provided by the present invention includes a data acquisition module, a data preprocessing module, a feature information processing module, a material cladding layer performance prediction model building module and a material cladding layer performance prediction module;

[0052] The data acquisition module acquires historical material cladding layer data and sends the data to the data preprocessing module;

[0053] The data preprocessing module performs data cleaning, data conversion and data normalization on the collected data, and sends the data to the feature information processing module;

[0054] The feature information processing module uses individual positions to represent feature subset solutions, generates abnormal values, calculates correlation factors, primary increments and advanced increments, designs dynamic fluctuation factors, updates individual positions, determines the best feature subset according to the fitness value threshold and the maximum number of searches, constructs a prediction data set, and sends the data to a module for constructing a material cladding layer performance prediction model;

[0055] The module for constructing the material cladding layer performance prediction model generates nodes and sets initial weights and embedding vectors, calculates weighted Euclidean distance and average weighted similarity, sets update status labels and update status factors, updates important thresholds, screens out important neighbor nodes, updates the embedding vectors and weights of nodes, trains weak classifiers, obtains the material cladding layer performance prediction model through weighted combination, and sends the data to the material cladding layer performance prediction module;

[0056] The material cladding layer performance prediction module obtains the performance grade of the material cladding layer based on the prediction label output by the material cladding layer performance prediction model.

[0057] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0058] (1) In view of the problem that the existing material cladding layer performance prediction methods have the problem that the material cladding layer data itself contains a large number of features, and the importance and correlation of different features are unclear, it is difficult to capture the truly meaningful material cladding layer features, which makes it difficult to accurately determine the optimal material cladding layer features and cannot accurately predict the actual material cladding layer performance. This scheme uses individual positions to represent the feature subset scheme, initializes the individual positions and uses information gain as the fitness value, which provides a wide range of feature selection possibilities for material cladding layer performance prediction, and can improve the comprehensiveness and efficiency of feature selection; generates abnormal values, and calculates correlation factors, primary increments and advanced increments to finely capture the complex relationship and dynamic changes between material cladding layer features and better explore the feature space; designs dynamic fluctuation factors to better adapt to the characteristics and changes of material cladding layer data; updates individual positions, continuously optimizes the feature subset scheme, and improves the quality and prediction accuracy of the feature subset; determines the best feature subset according to the fitness value threshold and the maximum search number, and constructs a prediction data set, ensuring that the feature subset that is most conducive to the material cladding layer performance prediction can be found within a reasonable time and range, thereby improving the accuracy and reliability of the material cladding layer performance prediction.

[0059] (2) In view of the problem that the existing material cladding layer performance prediction methods have a large number of complex nonlinear relationships in the material cladding layer data, which are difficult to effectively capture, resulting in a large deviation between the prediction results and the actual situation, and reducing the reliability of the cladding layer performance prediction, this scheme generates nodes and sets initial weights and embedding vectors. By calculating the weighted Euclidean distance and the average weighted similarity, the weighted calculation can better highlight important features and relationships, thereby better capturing the complex patterns in the material cladding layer data; setting the update state label and update state factor, adjusting the state according to the similarity changes of the previous and next layers, can better adapt to different The distribution and changes of material cladding layer data; update important thresholds, screen out important neighbor nodes, more reasonably screen out neighbor nodes that have an important impact on performance prediction, and better grasp the key cladding layer characteristics and relationships; update the embedding vectors and weights of the nodes to better fit the data and reflect the actual situation, and gradually improve the prediction ability of the material cladding layer performance; train weak classifiers, and obtain the material cladding layer performance prediction model through weighted combination, which can better cope with the characteristics and changes of different material cladding layer data, improve the accuracy and effect of cladding layer performance prediction, and provide more powerful support for material cladding layer performance evaluation and optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A schematic diagram of the process of predicting the performance of a material cladding layer based on big data provided by the present invention;

[0061] Figure 2 A schematic diagram of a material cladding layer performance prediction system based on big data provided by the present invention;

[0062] Figure 3 is a schematic flow chart of step S3;

[0063] Figure 4 It is a schematic diagram of the process of step S4.

[0064] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only 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 ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0066] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0067] Example 1, see Figure 1 The present invention provides a material cladding layer performance prediction method based on big data, the method comprising the following steps:

[0068] Step S1: data collection, collecting historical material cladding layer data;

[0069] Step S2: data preprocessing, data cleaning, data conversion and data normalization processing of the collected data;

[0070] Step S3: feature information processing, using individual positions to represent feature subset solutions, generating change values, and calculating correlation factors, primary increments, and advanced increments, designing dynamic fluctuation factors, updating individual positions, determining the best feature subset based on the fitness value threshold and the maximum number of searches, and constructing a prediction data set;

[0071] Step S4: construct a material cladding layer performance prediction model, generate nodes and set initial weights and embedding vectors, calculate weighted Euclidean distance and average weighted similarity, set update status labels and update status factors, update important thresholds, filter out important neighbor nodes, update node embedding vectors and weights, train weak classifiers, and obtain a material cladding layer performance prediction model through weighted combination;

[0072] Step S5: predicting the performance of the material cladding layer, obtaining the performance grade of the material cladding layer based on the prediction label output by the material cladding layer performance prediction model.

[0073] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the historical material cladding layer data includes material composition data, process parameter data, environmental condition data, cladding layer geometry data and performance grade; the material composition data includes base material composition data and cladding material composition data; the process parameter data includes cladding process parameters, preheating process parameters and post-heating process parameters; the environmental condition data includes temperature, humidity, air pressure, electromagnetic environment and air quality; the cladding layer geometry data includes thickness, width, length and surface roughness; the performance grades include excellent, good, medium, qualified and poor, and the performance grades are used as data labels.

[0074] Example 3, see Figure 1This embodiment is based on the above embodiment. In step S2, data cleaning includes processing noise, outliers, duplicate values ​​and missing values; data conversion is to convert data into vector form; data normalization is to unify the data range based on the maximum and minimum normalization method.

[0075] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S3, the feature information processing specifically includes the following steps:

[0076] Step S31: Initialize individual positions, establish a parameter search space based on the dimension information of the preprocessed material cladding layer data, randomly initialize N individual positions in the parameter search space, use each individual position to represent a feature subset scheme, and use the information gain of each feature subset scheme as the fitness value of the corresponding individual position. The individual dimension is equal to the material cladding layer data dimension. When the dimension value of the individual position is greater than 0.5, the feature subset scheme corresponding to the individual position contains the feature data corresponding to the dimension; otherwise, the feature subset scheme corresponding to the individual position does not contain the feature data corresponding to the dimension;

[0077] Step S32: Calculate the change value, generate a change value for each individual position, the dimension of the change value is equal to the individual dimension, and then calculate the correlation factor based on the change value , primary increment and Advanced Incremental , the formula used to generate the change value is as follows:

[0078] ;

[0079] Where μ is the control value in the range of (0, 1), and are the abnormal values ​​corresponding to the w+1th and wth individual positions in the vth dimension during the tth search, , and are the correlation factor, primary increment, and advanced increment corresponding to the w-th individual position in the v-th dimension during the t-th search, respectively. t is the search number index, w is the individual position index, and v is the dimension index;

[0080] Step S33: Design a dynamic fluctuation factor, the formula used is as follows:

[0081] ;

[0082] In the formula, is the average value of the correlation factor corresponding to the wth individual position in all dimensions during the tth search, G is the adaptive factor, , is the dynamic fluctuation factor corresponding to the wth individual position in the vth dimension during the tth search;

[0083] Step S34: Update individual positions, using the following formula:

[0084] ;

[0085] In the formula, and are the values ​​of the w-th individual position in the v-th dimension during the t+1-th and t-th searches, respectively. is the value of the global optimal position in the vth dimension during the tth search. The global optimal position is the individual position with the largest fitness value;

[0086] Step S35: Determine the optimal feature subset, pre-set the fitness value threshold and the maximum number of searches, update the fitness value of the individual position, and when there is an individual position whose fitness value is greater than the fitness value threshold, select the feature subset scheme corresponding to the individual position as the optimal feature subset scheme, and construct the prediction data set according to the optimal feature subset scheme; otherwise, if the maximum number of searches is reached, go to step S31; otherwise, add 1 to the search number and go to step S32.

[0087] By performing the above operations, the existing material cladding layer performance prediction methods have the problem that the material cladding layer data itself contains a large number of features, and the importance and correlation of different features are unclear, making it difficult to capture truly meaningful material cladding layer features, resulting in difficulty in accurately determining the optimal material cladding layer features and unable to accurately predict the actual material cladding layer performance. This scheme uses individual positions to represent the feature subset scheme, initializes the individual positions and uses information gain as the fitness value, which provides a wide range of feature selection possibilities for material cladding layer performance prediction, and can improve the comprehensiveness and efficiency of feature selection; generates abnormal values, and calculates correlation factors, primary increments and advanced increments to finely capture the complex relationship and dynamic changes between material cladding layer features, and better explore the feature space; designs dynamic fluctuation factors to better adapt to the characteristics and changes of material cladding layer data; updates individual positions, continuously optimizes the feature subset scheme, and improves the quality and prediction accuracy of the feature subset; determines the best feature subset according to the fitness value threshold and the maximum search number, and constructs a prediction data set, ensuring that the feature subset that is most conducive to the material cladding layer performance prediction can be found within a reasonable time and range, and the accuracy and reliability of the material cladding layer performance prediction can be improved.

[0088] Example 5, see Figure 1 and Figure 4 Based on the above embodiment, in step S4, building a material cladding layer performance prediction model specifically includes the following steps:

[0089] Step S41: Node initialization, including the following steps:

[0090] Step S411: Generate a node for each material cladding layer data in the prediction data set, and set the initial weight of each node to , the feature vector of the node is the corresponding material cladding layer data, and the node set S and feature matrix Z are obtained. The feature matrix is ​​linearly transformed and activated to obtain the initial embedding vector of the node, and the original data is mapped to a low-dimensional representation space; the formula used is as follows:

[0091] ;

[0092] In the formula, is the initial embedding vector of the ith node, σ(•) is the ReLU activation function, W is the learnable weight matrix, is the feature vector of the i-th node in the feature matrix Z, i is the node index, and Q is the number of nodes;

[0093] Step S412: Calculate the Euclidean distance between any two nodes, and use the value obtained by subtracting the Euclidean distance from 1 as the similarity. Preset a similarity threshold. If the similarity between the two nodes is greater than the similarity threshold, there is an edge between the two nodes; otherwise, there is no edge between the two nodes; the number of edges is recorded as F; and a neighbor node set C of each node is constructed based on the nodes connected by the edges.

[0094] Step S42: Screening of important neighbor nodes, including the following steps:

[0095] Step S421: Calculate the weighted Euclidean distance between any two nodes based on the node weights and embedding vectors, subtract the weighted Euclidean distance from 1 as the weighted similarity, calculate the average weighted similarity between all nodes and their neighboring nodes, and more finely evaluate the correlation between nodes; the formula used is as follows:

[0096] ;

[0097] ;

[0098] In the formula, is the average weighted similarity between all nodes in the b-th convolutional layer and their neighbor nodes, is the number r in the b-th convolutional layer i and r j The similarity between Yes i The neighbor node set, b is the convolution layer index, r i and r j are the i-th and j-th nodes respectively, is the number r in the b-th convolutional layer i and r j The weighted Euclidean distance between and They are r in the b-th convolutional layer. i and r j The weight of and They are r in the b-th convolutional layer. i and r j The embedding vector of , j is the node index, is the L2 norm;

[0099] Step S422: Set the update status, pre-set the update status label and update status factor, the initial values ​​of which are g1 and 0 respectively; if , then the update state label of the b-th convolutional layer is reset to g1, and the value of the update state factor is reset to 1; otherwise, the update state label of the b-th convolutional layer is reset to g2, and the value of the update state factor is reset to -1; where, is the average similarity between all nodes in the b-1th convolutional layer and their neighbor nodes; adaptively identifies the most critical neighbor information for the current node;

[0100] Step S423: Update the important threshold, pre-set a fixed value β, set the initial value of the important threshold of each convolutional layer to h, if the update status label of the b-th convolutional layer is g1, then modify the important threshold of the b-th convolutional layer to Otherwise, the important threshold of the b-th convolutional layer is modified to ;

[0101] Step S424: construct an important neighbor node set, and construct an important neighbor node set U for each node in each convolutional layer based on the nodes whose similarity is greater than the important threshold;

[0102] Step S43: Update the embedding vector of the node. Update the embedding vector of the node in each convolutional layer based on the set of important neighbor nodes. Use the aggregation mechanism to further refine the node representation and capture more effective features. The formula used is as follows:

[0103] ;

[0104] In the formula, is the number r in the b+1th convolutional layer i The embedding vector of , MEAN(•) is the function to calculate the average value, is the number r in the b-th convolutional layer a The embedding vector signature, r a is the a-th node, a is the node index, is the number r in the b-th convolutional layer i The set of important neighbor nodes of is the union operator;

[0105] Step S44: Update the node weights and normalize the embedding vector of each node using the Softmax function to obtain the probability vector of the node belonging to each data label. , select the data label with the highest probability as the classification label of the node, and update the weight of the node. The formula used is as follows:

[0106] ;

[0107] In the formula, is the number r in the b+1th convolutional layer i The weight of , P is the number of data labels, Yes i The real data labels, is the number of pixels in the b-th convolutional layer. i The classification label of is in the bth convolutional layer r i The probability vector belonging to each data label, , and They are respectively r in the b-th convolutional layer i The probability of data labels belonging to category 1, category p, and category P;

[0108] Step S45: training. Based on the updated node weights, return to step S42 to continue training the next convolution layer. There are B convolution layers in total. Each convolution layer is used as a weak classifier to obtain B weak classifiers. The weight of each weak classifier is calculated. The formula used is as follows:

[0109] ;

[0110] In the formula, α b is the weight of the b-th weak classifier, ε b is the error rate of the bth weak classifier;

[0111] Step S46: Combining, weighted combination of B weak classifiers to obtain a material cladding layer performance prediction model, taking the data label with the highest probability as the prediction label, the formula used is as follows:

[0112] ;

[0113] Where Y i is the output of the material cladding layer performance prediction model i The predicted label of .

[0114] By performing the above operations, the existing material cladding layer performance prediction method has a problem that the material cladding layer data has a large number of complex nonlinear relationships, which are difficult to effectively capture, resulting in a large deviation between the prediction results and the actual situation, reducing the reliability of the cladding layer performance prediction. This solution generates nodes and sets initial weights and embedding vectors. By calculating the weighted Euclidean distance and the average weighted similarity, the weighted calculation can better highlight important features and relationships, thereby better capturing the complex patterns in the material cladding layer data; setting the update state label and the update state factor, adjusting the state according to the similarity changes of the previous and next layers, can better adapt to The distribution and changes of different material cladding layer data; update important thresholds, screen out important neighbor nodes, more reasonably screen out neighbor nodes that have an important impact on performance prediction, and better grasp the key cladding layer characteristics and relationships; update the embedding vectors and weights of the nodes to better fit the data and reflect the actual situation, and gradually improve the prediction ability of the material cladding layer performance; train weak classifiers, and obtain the material cladding layer performance prediction model through weighted combination, which can better cope with the characteristics and changes of different material cladding layer data, improve the accuracy and effect of cladding layer performance prediction, and provide more powerful support for material cladding layer performance evaluation and optimization.

[0115] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the material cladding layer performance prediction is to collect real-time material cladding layer data, and the real-time material cladding layer data includes material composition data, process parameter data, environmental condition data and cladding layer geometry data; after preprocessing, the data is input into the material cladding layer performance prediction model for processing, and the performance level of the material cladding layer is obtained based on the output prediction label.

[0116] Embodiment 7, see Figure 2 , this embodiment is based on the above embodiment, and the material cladding layer performance prediction system based on big data provided by the present invention includes a data acquisition module, a data preprocessing module, a feature information processing module, a material cladding layer performance prediction model building module and a material cladding layer performance prediction module;

[0117] The data acquisition module acquires historical material cladding layer data and sends the data to the data preprocessing module;

[0118] The data preprocessing module performs data cleaning, data conversion and data normalization on the collected data, and sends the data to the feature information processing module;

[0119] The feature information processing module uses individual positions to represent feature subset solutions, generates abnormal values, calculates correlation factors, primary increments and advanced increments, designs dynamic fluctuation factors, updates individual positions, determines the best feature subset according to the fitness value threshold and the maximum number of searches, constructs a prediction data set, and sends the data to a module for constructing a material cladding layer performance prediction model;

[0120] The module for constructing the material cladding layer performance prediction model generates nodes and sets initial weights and embedding vectors, calculates weighted Euclidean distance and average weighted similarity, sets update status labels and update status factors, updates important thresholds, screens out important neighbor nodes, updates the embedding vectors and weights of nodes, trains weak classifiers, obtains the material cladding layer performance prediction model through weighted combination, and sends the data to the material cladding layer performance prediction module;

[0121] The material cladding layer performance prediction module obtains the performance grade of the material cladding layer based on the prediction label output by the material cladding layer performance prediction model.

[0122] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0123] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0124] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A material cladding layer performance prediction method based on big data, characterized by: The method comprises the following steps: Step S1: data collection, collecting historical material cladding layer data; Step S2: data preprocessing; Step S3: feature information processing; Step S4: constructing a material cladding layer performance prediction model; Step S5: predicting the performance of the material cladding layer, obtaining the performance level of the material cladding layer based on the prediction label output by the material cladding layer performance prediction model; In step S4, the construction of the material cladding layer performance prediction model specifically includes the following steps: Step S41: node initialization; Step S42: Screening of important neighbor nodes; Step S43: Update the embedding vector of the node. Update the embedding vector of the node in each convolutional layer based on the set of important neighbor nodes. The formula used is as follows: ； In the formula, is the number r in the b+1th convolutional layer i The embedding vector of , MEAN(•) is the function for calculating the average value, and They are r in the b-th convolutional layer. i and r a The embedding vector signature, r i and r a are the i-th and a-th nodes respectively, i and a are node indexes, is the number r in the b-th convolutional layer i is the set of important neighbor nodes of , σ(•) is the activation function, W is the learnable weight matrix, b is the convolutional layer index, is the union operator; Step S44: Update the weight of the node; Step S45: training. Based on the updated node weights, return to step S42 to continue training the next convolution layer. There are B convolution layers in total. Each convolution layer is used as a weak classifier to obtain B weak classifiers. The weight of each weak classifier is calculated. The formula used is as follows: ； In the formula, α b is the weight of the b-th weak classifier, ε b is the error rate of the bth weak classifier; Step S46: Combining, weighted combination of B weak classifiers to obtain a material cladding layer performance prediction model, taking the data label with the highest probability as the prediction label, the formula used is as follows: ； Where Y i is the output of the material cladding layer performance prediction model i The predicted label of is in the bth convolutional layer r i A vector of probabilities belonging to each data label.

2. The material cladding layer performance prediction method based on big data according to claim 1 is characterized in that: In step S3, the feature information processing specifically includes the following steps: Step S31: Initialize individual positions, establish a parameter search space based on the dimension information of the preprocessed material cladding layer data, randomly initialize N individual positions in the parameter search space, use each individual position to represent a feature subset scheme, and use the information gain of each feature subset scheme as the fitness value of the corresponding individual position. The individual dimension is equal to the material cladding layer data dimension. When the dimension value of the individual position is greater than 0.5, the feature subset scheme corresponding to the individual position contains the feature data corresponding to the dimension; otherwise, the feature subset scheme corresponding to the individual position does not contain the feature data corresponding to the dimension; Step S32: Calculate the change value, generate a change value for each individual position, the dimension of the change value is equal to the individual dimension, and then calculate the correlation factor based on the change value , primary increment and Advanced Incremental , the formula used to generate the change value is as follows: ； Where μ is the control value, and are the abnormal values ​​corresponding to the w+1th and wth individual positions in the vth dimension during the tth search, , and are the correlation factor, primary increment, and advanced increment corresponding to the w-th individual position in the v-th dimension during the t-th search, respectively. t is the search number index, w is the individual position index, and v is the dimension index; Step S33: Design a dynamic fluctuation factor, the formula used is as follows: ； In the formula, is the average value of the correlation factor corresponding to the wth individual position in all dimensions during the tth search, G is the adaptive factor, is the dynamic fluctuation factor corresponding to the wth individual position in the vth dimension during the tth search; Step S34: Update individual positions, using the following formula: ； In the formula, and are the values ​​of the w-th individual position in the v-th dimension during the t+1-th and t-th searches, respectively. is the value of the global optimal position in the vth dimension during the tth search; Step S35: Determine the optimal feature subset, pre-set the fitness value threshold and the maximum number of searches, update the fitness value of the individual position, and when there is an individual position whose fitness value is greater than the fitness value threshold, select the feature subset scheme corresponding to the individual position as the optimal feature subset scheme, and construct the prediction data set according to the optimal feature subset scheme; otherwise, if the maximum number of searches is reached, go to step S31; otherwise, add 1 to the search number and go to step S32.

3. The material cladding layer performance prediction method based on big data according to claim 1 is characterized in that: In step S41, the node initialization specifically includes the following steps: Step S411: Generate a node for each material cladding layer data in the prediction data set, and set the initial weight of each node to , the characteristic vector of the node is the corresponding material cladding layer data, and the node set S and the characteristic matrix Z are obtained. The characteristic matrix is ​​linearly transformed and activated to obtain the initial embedding vector of the node. The formula used is as follows: ； In the formula, is the initial embedding vector of the ith node, σ(•) is the activation function, W is the learnable weight matrix, is the feature vector of the i-th node in the feature matrix Z, i is the node index, and Q is the number of nodes; Step S412: Calculate the Euclidean distance between any two nodes, and use the value obtained by subtracting the Euclidean distance from 1 as the similarity. Pre-set a similarity threshold. If the similarity between the two nodes is greater than the similarity threshold, there is an edge between the two nodes; otherwise, there is no edge between the two nodes; record the number of edges as F; and construct a neighbor node set C for each node based on the nodes connected by edges.

4. The material cladding layer performance prediction method based on big data according to claim 1 is characterized in that: In step S42, the important neighbor node screening specifically includes the following steps: Step S421: Calculate the weighted Euclidean distance between any two nodes based on the node weights and embedding vectors, subtract the weighted Euclidean distance from 1 as the weighted similarity, and calculate the average weighted similarity between all nodes and their neighboring nodes. The formula used is as follows: ； ； In the formula, is the average weighted similarity between all nodes in the b-th convolutional layer and their neighbor nodes, is the number r in the b-th convolutional layer i and r j The similarity between Yes i The neighbor node set, b is the convolution layer index, r i and r j are the i-th and j-th nodes respectively, is the number r in the b-th convolutional layer i and r j The weighted Euclidean distance between and They are r in the b-th convolutional layer. i and r j The weight of and They are r in the b-th convolutional layer. i and r j The embedding vector of , j is the node index, is the L2 norm; Step S422: Set the update status, pre-set the update status label and update status factor, the initial values ​​of which are g1 and 0 respectively; if , then the update state label of the b-th convolutional layer is reset to g1, and the value of the update state factor is reset to 1; otherwise, the update state label of the b-th convolutional layer is reset to g2, and the value of the update state factor is reset to -1; where, is the average similarity between all nodes in the b-1th convolutional layer and their neighboring nodes; Step S423: Update the important threshold, pre-set a fixed value β, set the initial value of the important threshold of each convolutional layer to h, if the update status label of the b-th convolutional layer is g1, then modify the important threshold of the b-th convolutional layer to Otherwise, the important threshold of the b-th convolutional layer is modified to ; Step S424: construct an important neighbor node set, and construct an important neighbor node set U for each node in each convolutional layer based on the nodes whose similarity is greater than the important threshold.

5. The material cladding layer performance prediction method based on big data according to claim 1 is characterized in that: In step S44, the weight of the updated node is normalized by the Softmax function on the embedding vector of each node to obtain the probability vector of the node belonging to each data label. , select the data label with the highest probability as the classification label of the node, and update the weight of the node. The formula used is as follows: ； In the formula, is the number r in the b+1th convolutional layer i The weight of , P is the number of data labels, Yes i The real data labels, is the number of pixels in the b-th convolutional layer. i The classification label of is in the bth convolutional layer r i The probability vector belonging to each data label, , and They are respectively r in the b-th convolutional layer i The probability of data labels belonging to class 1, class p, and class P.

6. The material cladding layer performance prediction method based on big data according to claim 1 is characterized by: In step S1, the data collection is to collect historical material cladding layer data; the historical material cladding layer data includes material composition data, process parameter data, environmental condition data, cladding layer geometry data and performance level; the performance level is used as a data label.

7. The material cladding layer performance prediction method based on big data according to claim 1 is characterized by: In step S5, the material cladding layer performance prediction is to collect real-time material cladding layer data, and the real-time material cladding layer data includes material composition data, process parameter data, environmental condition data and cladding layer geometry data; After preprocessing, the data is input into the material cladding layer performance prediction model for processing, and the performance level of the material cladding layer is obtained based on the output prediction label.

8. The material cladding layer performance prediction method based on big data according to claim 1 is characterized by: In step S2, the data preprocessing is to perform data cleaning, data conversion and data normalization on the collected data; data cleaning includes processing noise, outliers, duplicate values ​​and missing values; Data conversion is to convert data into vector form; data normalization is to unify the data range based on the maximum and minimum normalization method.

9. A material cladding layer performance prediction system based on big data, used to implement the material cladding layer performance prediction method based on big data as described in any one of claims 1 to 8, characterized in that: It includes a data acquisition module, a data preprocessing module, a feature information processing module, a module for building a material cladding layer performance prediction model, and a material cladding layer performance prediction module; The data acquisition module acquires historical material cladding layer data and sends the data to the data preprocessing module; The data preprocessing module performs data cleaning, data conversion and data normalization on the collected data, and sends the data to the feature information processing module; The feature information processing module uses individual positions to represent feature subset solutions, generates abnormal values, calculates correlation factors, primary increments and advanced increments, designs dynamic fluctuation factors, updates individual positions, determines the best feature subset according to the fitness value threshold and the maximum number of searches, constructs a prediction data set, and sends the data to a module for constructing a material cladding layer performance prediction model; The module for constructing the material cladding layer performance prediction model generates nodes and sets initial weights and embedding vectors, calculates weighted Euclidean distance and average weighted similarity, sets update status labels and update status factors, updates important thresholds, screens out important neighbor nodes, updates the embedding vectors and weights of nodes, trains weak classifiers, obtains the material cladding layer performance prediction model through weighted combination, and sends the data to the material cladding layer performance prediction module; The material cladding layer performance prediction module obtains the performance grade of the material cladding layer based on the prediction label output by the material cladding layer performance prediction model.

Images