Data processing method, device, equipment and storage medium

Abstract

The embodiments of the present application provide a data processing method, apparatus, device and storage medium to solve the problem of poor prediction performance of soft measurement models in the prior art. The method comprises: obtaining the data to be predicted of the target transaction; inputting the data to be predicted into the target predictor to obtain a prediction result; wherein the prediction result comprises at least two dependent variables; the data to be predicted comprises the independent variable of the dependent variable; the target predictor comprises a graph convolution layer, the graph convolution layer comprises a graph structure obtained by training based on sample data, and the nodes of the graph structure comprise sample dependent variables in the sample data. The embodiments of the present application are based on a graph structure with multiple dependent variables as nodes, and use graph convolution operations to connect nodes with correlations, so that different dependent variables influence each other in the prediction process, so as to improve the comprehensive soft measurement performance of multi-dependent variable prediction.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a data processing method and device, an electronic device and a storage medium. Background Art

[0002] Multivariable predictive control refers to a control method with multiple inputs or outputs, which uses control strategies such as multi-step testing, rolling optimization and feedback correction. Multivariable predictive control systems, also known as multi-input and multi-output systems, are widely used in industrial production, big data processing, and artificial intelligence. In practical applications, multivariable predictive control systems usually use a prediction model, input independent variables, and predict dependent variables through the idea of ​​closed-loop feedback control. However, in the prediction process, there are a large number of controlled variables (i.e., dependent variables) that are difficult to measure, cannot be measured, and have a large lag in measurement. Traditional detection methods cannot provide feedback information in the closed-loop circuit, making feedback control impossible to achieve. Therefore, in the prior art, the prediction performance of multivariable predictive control systems is poor and the accuracy is low.

[0003] Taking the field of industrial production as an example, with the continuous advancement of automation technology, modern industry has gradually entered the stage of fully automated production. In industrial production, thanks to the core idea of ​​closed-loop feedback control in automation technology, the production process can be regulated by sensing the controlled variables in industrial production, so as to effectively overcome the disturbances in the production process and realize the automated control of the production process.

[0004] The emergence of soft measurement technology provides a way to solve the above problems. Since there is a mutual coupling relationship between many variables in the entire industrial production process, the core idea of ​​soft measurement is to use variables that are easy to measure directly (i.e., independent variables) to establish a model, and then predict variables that are difficult to measure (i.e., dependent variables). The performance of closed-loop feedback control based on soft measurement technology depends to a large extent on the prediction performance of the dependent variable. Therefore, building a soft measurement model with good real-time performance and accurate prediction can provide an effective perception basis for subsequent process control and optimization.

[0005] At present, soft measurement technology is generally divided into two categories: model-based measurement technology and data-based measurement technology. Since the data-based soft measurement method does not require a lot of knowledge about the mechanism of industrial processes, it only relies on a large amount of data collected on site to build a model to characterize the characteristics of the object, so it has been widely studied and developed, and some soft measurement models of technical data are also widely used in the field of industrial measurement.

[0006] In actual industrial applications, soft sensor models usually need to predict multiple dependent variables, resulting in poor prediction performance. Summary of the invention

[0007] The embodiment of the present application provides a data processing method to improve the prediction performance of a multivariable predictive control system.

[0008] Correspondingly, an embodiment of the present application also provides a data processing device, an electronic device and a storage medium to ensure the implementation and application of the above method.

[0009] In order to solve the above problems, the present application discloses a data processing method, which includes:

[0010] Acquiring data to be predicted of a target transaction; the target transaction includes a production transaction or a parameter prediction transaction executed according to the data to be predicted;

[0011] Inputting the data to be predicted into a target predictor to obtain a prediction result; wherein the prediction result includes at least two dependent variables; the data to be predicted includes an independent variable of the dependent variable;

[0012] The target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample dependent variables in the sample data.

[0013] The present application also discloses a data processing device, which includes:

[0014] Obtaining independent variable data of the target production transaction; the independent variable data includes the starting element addition amount, or the independent variable data includes the starting element addition amount and the process control variable;

[0015] Inputting the independent variable data into the target predictor to obtain a prediction result; wherein the prediction result includes at least two dependent variables; the dependent variable includes the end point element content;

[0016] The target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample dependent variables in the sample data.

[0017] The present application discloses a data processing method, which includes:

[0018] Obtaining independent variable data of the target production transaction; the independent variable data includes the starting element addition amount, or the independent variable data includes the starting element addition amount and the process control variable;

[0019] Inputting the independent variable data into the target predictor to obtain a prediction result; wherein the prediction result includes at least two dependent variables; the dependent variable includes the end point element content;

[0020] The target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample dependent variables in the sample data.

[0021] The present application also discloses a data processing device, which includes:

[0022] A data acquisition module, used to acquire the data to be predicted of the target transaction; the target transaction includes a production transaction or a parameter prediction transaction executed according to the data to be predicted;

[0023] A data prediction module, used for inputting the data to be predicted into a target predictor to obtain a prediction result; wherein the prediction result includes at least two dependent variables; and the data to be predicted includes an independent variable of the dependent variable;

[0024] The target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample dependent variables in the sample data.

[0025] The present application discloses a data processing method, which includes:

[0026] Obtain industrial parameters for industrial production affairs;

[0027] The industrial parameters are input into a target predictor to obtain a prediction result; wherein the prediction result includes at least two prediction parameters; the target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample prediction parameters in the sample data.

[0028] The present application also discloses a data processing device, which includes:

[0029] A parameter acquisition module is used to obtain industrial parameters of industrial production affairs;

[0030] A parameter prediction module is used to input the industrial parameters into a target predictor to obtain a prediction result; wherein the prediction result includes at least two prediction parameters; the target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample prediction parameters in the sample data.

[0031] An embodiment of the present application further discloses an electronic device, including: a processor; and a memory, on which executable code is stored. When the executable code is executed, the processor executes one or more methods described in the embodiments of the present application.

[0032] The embodiments of the present application also disclose one or more machine-readable storage media on which executable codes are stored. When the executable codes are executed, the processor executes one or more methods described in the embodiments of the present application.

[0033] Compared with the prior art, the embodiments of the present application have the following advantages:

[0034] In an embodiment of the present application, data to be predicted of a target transaction is obtained, and the data to be predicted is input into a target predictor to obtain a prediction result; the prediction result includes at least two dependent variables, and the target predictor includes a graph convolution layer, which is a graph structure with multiple dependent variables as nodes, and the graph structure uses a graph convolution operation to connect dependent variables with related relationships to each other, so that different dependent variables influence each other in the prediction process; based on the coupling relationship between the dependent variables, the performance of the target predictor on some dependent variables with poor prediction performance is improved, the comprehensive soft measurement performance of multi-dependent variable prediction is improved, and the prediction accuracy is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is one of the application schematic diagrams of a data processing method according to an embodiment of the present application;

[0036] Figure 2 This is the second application schematic diagram of a data processing method according to an embodiment of the present application;

[0037] Figure 3 is a flowchart of steps of an embodiment of a data processing method of the present application;

[0038] Figure 4 is a flowchart of steps of another data processing method embodiment of the present application;

[0039] Figure 5 is a flowchart of steps of another data processing method embodiment of the present application;

[0040] Figure 6 is a structural block diagram of an embodiment of a data processing device of the present application;

[0041] Figure 7 is a structural block diagram of an optional embodiment of a data processing device of the present application;

[0042] Figure 8 is a structural block diagram of another data processing device embodiment of the present application;

[0043] Fig. 9 is a structural block diagram of another data processing device embodiment of the present application;

[0044] Fig.10 It is a schematic diagram of the structure of a device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0045] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0046] The terms "first", "second", etc. in the specification and claims of this application are used to distinguish similar objects, rather than to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein.

[0047] The embodiments of the present application can be applied to variable prediction fields such as industrial production. In the prediction process of multiple dependent variables, the relationship between related variables is mined through the graph convolution layer; based on the graph structure with multiple dependent variables as nodes, the graph convolution operation is used to connect the nodes with related relationships, so that different dependent variables can influence each other in the prediction process, so as to improve the comprehensive soft measurement performance of multi-dependent variable prediction.

[0048] like Figure 1 As shown, Figure 1 A schematic diagram of an application scenario of an embodiment of the present application is shown. Figure 1 The diagram in the figure is a scenario diagram for predicting the dependent variable of a target transaction, where the target transaction can be industrial production, big data processing, etc. The input data is the data to be predicted, i.e., the independent variable. The data to be predicted is input into the target predictor to obtain the prediction result, which is the output data of the dependent variable. Figure 1 It can be seen that the output data includes multiple dependent variables, that is, the target predictor is used for the prediction of multiple dependent variables. The applicant found that since multiple dependent variables are in the same control process, the fluctuations of each dependent variable affect each other, so there is a strong coupling relationship between each dependent variable; while the soft measurement model in the prior art usually only considers the influence of the independent variable on the dependent variable in the prediction process, and ignores the coupling relationship between the dependent variables.

[0049] Taking the soft measurement model in the field of industrial production as an example, there are many data-based soft measurement models in the prior art, such as multivariate statistical analysis methods such as principal component regression (PCR), partial least squares regression (PLS), and Gaussian process regression (GPR); in addition, machine learning models such as support vector regression (SVR), random forest, and artificial neural network (ANN) are also gradually widely used in the field of industrial soft measurement. In the prediction process, the existing soft measurement models usually do not consider the correlation between multiple dependent variables, resulting in poor prediction performance.

[0050] In the embodiment of the present application, the dependent variables are predicted in combination with the correlation between the dependent variables. Specifically, Figure 1 In step 101 of the embodiment of the present application, firstly, the data to be predicted of the target transaction is obtained; the target transaction includes a production transaction or a parameter prediction transaction executed according to the data to be predicted.

[0051] The target transaction can be an industrial production transaction, such as steelmaking. With the continuous advancement of automation technology, modern industry has gradually entered the stage of fully automated production. In industrial production, thanks to the core idea of ​​closed-loop feedback control in automation technology, the production process can be regulated by sensing the controlled variables in industrial production, so as to effectively overcome the disturbance in the production process and realize the automated control of the production process.

[0052] The target task can also be other prediction tasks, such as temperature prediction, sewage pollution prediction, etc. The data to be predicted is the independent variable of the prediction process, which is usually a variable that is easy to measure directly, and the dependent variable is predicted based on the data to be measured. The dependent variable is the prediction result, which is usually a variable that is difficult to predict.

[0053] As a first example, if the target transaction is steelmaking, the independent variables include parameters such as alloy addition, process variables, and molten steel related data; alloy addition refers to the addition amount of each alloy; process variables include temperature, time, and other variables. The dependent variable is the end-point element content, such as acid-soluble aluminum (Als), carbon (C), chromium (Cr), molybdenum (Mo), etc.; since the end-point element content is difficult to measure and control directly, and some elements between the end-point elements will affect each other, the dependent variable is predicted based on the independent variables.

[0054] As a second example, the target transaction is a temperature prediction transaction, and the independent variables include humidity percentage, dew point temperature, wind speed, cloud coverage, etc., as well as the sampled temperature value; the dependent variable is the temperature of each unit time within the target prediction time period. For example, if the target prediction time period is 1 day, then the dependent variable is every hour of the day, and the temperature of each hour has a certain correlation, that is, there is a certain coupling between the dependent variables.

[0055] As a third example, the target task is a sewage pollution prediction task. The independent variables include variables such as temperature, flow, pH, dissolved oxygen concentration, redox potential, turbidity and conductivity; the dependent variables include chemical oxygen demand (COD), nitrate, phosphate concentration, etc. The detection of each dependent variable takes a long time and the process is complicated; therefore, the dependent variable is predicted by the independent variable.

[0056] In step 102, the data to be predicted is input into a target predictor to obtain a prediction result; wherein the prediction result includes at least two dependent variables; and the data to be predicted includes an independent variable of the dependent variable.

[0057] Among them, the prediction result includes at least two dependent variables, that is, the embodiment of the present application is a multi-dependent variable prediction process, and if there is a certain correlation between the multiple dependent variables in the actual production application process, they will affect each other.

[0058] In order to introduce the correlation information between multiple dependent variables, the graph convolution network is a machine learning model for mining the mutual relationship in the graph structure that has developed rapidly in recent years. In the embodiment of the present application, the target predictor is a pre-trained predictor, and the correlation information between multiple dependent variables is introduced during the training process. The target predictor includes a graph convolution layer, and the graph convolution layer includes a graph structure obtained by training according to sample data. The nodes of the graph structure include sample dependent variables in the sample data, and the sample dependent variables are the dependent variables in the sample data, that is, the known dependent variable values; in the sample data, each group of data includes sample independent variables and sample dependent variables; the sample independent variables are input into the predictor to obtain the predicted dependent variables, and then the predictor is reversely optimized according to the gap between the known sample dependent variables and the predicted dependent variables, and the predictor that finally meets the accuracy requirements is the target predictor.

[0059] Graph convolutional networks have been widely used in social network relationship analysis, file classification, traffic prediction and other fields due to their ability to conveniently perform operations in non-Euclidean spaces. The graph convolutional layer is a graph structure with sample dependent variables as nodes. The graph structure is usually composed of nodes and edges, and the relationships between nodes are described by edges. Figure 1 As shown, Figure 1Each node N in corresponds to a feature of the graph convolution layer, and the edge between adjacent nodes N represents the relationship between two adjacent nodes. It can be understood that Figure 1 Only the graph convolution layer is shown in the figure. The actual number of nodes may not be Figure 1 Through the graph convolution operation, the weighted sum of the features of the adjacent nodes of each node is used as the feature of the node; therefore, through the graph convolution operation, the features of the nodes can be extracted based on the mutual relationship described under the graph structure.

[0060] After the data to be predicted is input into the target predictor as input data, it is subjected to feature extraction through the graph convolution layer to obtain output data as the prediction result; Figure 1 As shown, the output data includes n dependent variables, each dependent variable corresponds to a channel, and in the graph convolution layer, the channels of each dependent variable with mutual relationship are connected, so that the graph convolution layer refers to this mutual relationship when extracting features, so that the mutual relationship between multiple dependent variables affects the prediction results.

[0061] In the prior art, graph convolutional networks usually regard samples as nodes and mine the relationships between samples instead of the relationships between non-variables. Therefore, it is impossible to effectively introduce the correlation information between multiple dependent variables. In the embodiments of the present application, for industrial soft measurement scenarios with multiple dependent variables, the correlation information between multiple dependent variables is introduced to improve the prediction performance.

[0062] In addition, in the prior art graph convolutional networks, the relationship between nodes is usually calculated using distance metrics, and distance cannot accurately describe the correlation information between nodes. In the embodiment of the present application, each channel in the target predictor can use the independent variable information to independently learn different dependent variables, and at the same time, multiple dependent variables are used as different nodes in the graph convolution layer, and the strength of the correlation between nodes is described through the edges in the graph convolution layer, accurately measuring the correlation between nodes.

[0063] Referring to the first example above, taking steelmaking as an example, the independent variables include alloy addition, process variables, molten steel related data, etc.; the independent variables are input into the target predictor, and the target predictor extracts features from the independent variables. Through the graph convolution layer of the target predictor, the correlation between the dependent variables is integrated in the prediction process to obtain the dependent variable as the prediction result; the dependent variable is the end-point element content, and the end-point element is, for example, acid-soluble aluminum (Als). Elements such as carbon (C), chromium (Cr), molybdenum (Mo), etc., the prediction result includes the content of each end-point element; since the target predictor is a pre-trained predictor, its accuracy meets the preset accuracy requirements, so the accuracy of the end-point element content predicted by the target predictor is also high.

[0064] In the above embodiment of the present application, the data to be predicted of the target transaction is obtained, and the data to be predicted is input into the target predictor to obtain a prediction result; the prediction result includes at least two dependent variables, and the target predictor includes a graph convolution layer, which is a graph structure with multiple dependent variables as nodes, and the graph structure uses graph convolution operations to connect dependent variables with correlations to each other, so that different dependent variables affect each other in the prediction process; based on the coupling relationship between the dependent variables, the performance of the target predictor on some dependent variables with poor prediction performance is improved, the comprehensive soft measurement performance of multi-dependent variable prediction is improved, and the prediction accuracy is improved. The embodiment of the present application solves the problem of poor prediction performance of soft measurement models in the prior art.

[0065] like Figure 2 As shown, in an optional embodiment, the target predictor further includes a target fully connected layer, and the target fully connected layer includes a first fully connected layer ( Figure 2 Middle fully connected layer 1), the second fully connected layer ( Figure 2 The middle fully connected layer 2) and the third fully connected layer ( Figure 2 Middle fully connected layer 3);

[0066] Among them, the first fully connected layer, the graph convolutional layer, the second fully connected layer and the third fully connected layer are connected in sequence; it can be understood that the number of fully connected layers included in the target fully connected layer can also be other data besides 3. In the embodiment of the present application, the number of fully connected layers is only 3 for schematic illustration.

[0067] In the target predictor, each layer extracts features from the input data, and the extracted features are the output data;

[0068] The output data of the previous layer is the input data of the next layer;

[0069] The input data of the first fully connected layer is the data to be predicted. After the first fully connected layer extracts features from the data to be predicted, the extracted features are input to the graph convolution layer as input data; then the graph convolution layer extracts features from the input data, and the extracted features are input to the second fully connected layer as input data; the second fully connected layer continues to extract features from the input data, and inputs the extracted features to the third fully connected layer, and the output data of the third fully connected layer is the prediction result. Referring to the first example above, the data to be predicted includes parameters such as alloy addition amount, process variables, and molten steel related data. The above parameters are input to the target predictor, and feature extraction is performed through each layer of the target predictor to finally obtain the prediction result, which includes the end point element content of this steelmaking, such as acid-soluble aluminum (Als), carbon (C), chromium (Cr), molybdenum (Mo) and other element contents.

[0070] In an optional embodiment, each of the target fully connected layers includes N channels, where N is the number of the dependent variables, and the channel processes one of the dependent variables.

[0071] In the target predictor, each dependent variable corresponds to a channel, such as Figure 2 As shown in , dependent variable 1 corresponds to channel 1, dependent variable 2 corresponds to channel 2, ..., dependent variable n corresponds to channel n; when inputting the data to be predicted, each group of data to be predicted corresponds to a channel, that is, each channel inputs a complete set of data to be predicted; each group of data to be predicted is independently feature extracted in each channel of each fully connected layer, and a graph convolution operation is performed in the graph convolution layer based on the correlation between the dependent variables. Finally, each channel of the third fully connected layer outputs a dependent variable.

[0072] In an optional embodiment, in the graph structure, the characteristic value of each node is a result of weighted summation of adjacent nodes;

[0073] The weight in the graph structure is the weight between each two adjacent nodes, which is used to measure the relationship between two adjacent nodes; in the graph structure, the weight is the element in the adjacency matrix of the sample dependent variable; in the graph convolutional network, the eigenvalue of each node is obtained by the weighted sum of all its adjacent nodes, and the weight is determined by the relationship between two adjacent nodes.

[0074] The weight is an element in the adjacency matrix, which is generated according to the sample dependent variable. The adjacency matrix is ​​used to identify the relationship between each two adjacent nodes. For example, Aij represents the relationship between node i and node j. If Aij is 0, it means that node i has no relationship with node j, and correspondingly, node i and node j have no edge in the graph convolution network; when Aij is not 0, it represents the weight, that is, the degree of association between node i and node j, and correspondingly, node i and node j have an edge in the graph convolution network, and different weights can be displayed through different edges. The adjacency matrix is ​​extracted from the sample data of the target predictor. After the target predictor is trained and formed, when the target predictor is used to predict transactions, the elements in the adjacency matrix are no longer adjusted. In this way, based on the weights between nodes, the mutual relationship between dependent variables can be measured more accurately, while the graph convolution network in the prior art usually uses distance to measure the mutual relationship, and the distance cannot well describe the correlation information between nodes.

[0075] In an optional embodiment, before the step of obtaining the target transaction data to be predicted, the target predictor needs to be trained. The process of training the target predictor mainly includes the following steps:

[0076] Step 11, obtaining the sample data; wherein the sample data includes training samples and test samples.

[0077] Step 12, training an initial predictor according to the training samples to obtain a predictor to be tested;

[0078] Step 13: Test the predictor to be tested according to the test sample to obtain the target predictor.

[0079] The sample data is the sample data of the target transaction. In step 11, the sample data of the target transaction is obtained. For example, in order to improve the prediction accuracy of the target predictor, a large amount of sample data can be obtained for training. The sample data includes training samples and test samples. The training samples are used to train the predictor, and the test samples are used to test the accuracy of the trained predictor to obtain a target predictor that meets the accuracy requirements. The sample data includes independent variables (hereinafter referred to as sample independent variables) and dependent variables (hereinafter referred to as sample dependent variables). That is to say, the dependent variables in the sample data are known. During the training of the predictor, the sample dependent variables in the training samples are used to reversely optimize the predictor; during the testing process, the sample dependent variables in the test samples are used to judge the accuracy of the prediction results output by the predictor.

[0080] Specifically, in step 12, the training sample is input into the initial predictor to obtain the prediction result, and the predictor is reversely optimized according to the known sample dependent variable, and then the next training sample is iterated to finally obtain the predictor to be tested that meets the preset requirements. In step 13, the predictor to be tested is tested according to the test sample to obtain the target predictor that meets the accuracy requirements. In combination with the first example above, sample data of steelmaking affairs are obtained, and the sample data are divided into training samples and test samples according to a certain ratio; optionally, the allocation ratio can be 8:2, or other allocation ratios; the sample data includes actual data in historical production tasks, that is, actual independent variables and corresponding dependent variable values, the actual independent variables are parameters such as alloy addition amount, process variables, molten steel related data, and the actual dependent variable is the actual content of the endpoint element; the actual independent variables are input into the initial predictor to obtain the predicted content of the endpoint element, and then the predictor is reversely optimized according to the difference between the predicted content and the actual content, and the next sample data is iterated; this cycle is iterated until a predictor to be tested that meets the accuracy requirements is obtained, and then the test sample is used for testing to obtain a target predictor whose accuracy in predicting the content of the endpoint element meets the requirements.

[0081] In an optional embodiment, before step 12, the method further includes:

[0082] The sample data is normalized; the dimensional differences between different variables are eliminated by normalization. Optionally, the min-max normalization method can be used to convert the values ​​of all variables into the interval [0, 1] through linear transformation.

[0083] In an optional embodiment, the sample in the sample data includes a sample independent variable and the sample dependent variable;

[0084] Step 12 includes:

[0085] Step 121: input a first sample independent variable of a first training sample in the training samples into an initial predictor to obtain a first prediction content.

[0086] The first training sample is any sample in the training samples, which is input into the initial predictor to obtain the first prediction content. The first training sample includes a first sample independent variable and a first sample dependent variable.

[0087] Step 122: reversely optimize the initial predictor according to the first prediction content and the first sample dependent variable of the first training sample to obtain an optimized predictor.

[0088] If there is a difference between the first prediction content and the first sample dependent variable, the initial predictor is reversely optimized according to the difference to offset the difference so as to improve the prediction accuracy of the predictor.

[0089] Step 123, iteratively inputting the second sample independent variable in the training sample into the optimized predictor, and performing the reverse optimization until the number of iterations meets the first preset requirement and / or the loss function meets the second preset requirement, to obtain the predictor to be tested.

[0090] The second sample independent variable is iterated to the optimized predictor, and the predictor is continuously optimized inversely according to the dependent variable corresponding to the second sample independent variable, ..., and this cycle is repeated until the loop iteration termination condition is reached to obtain the predictor to be tested.

[0091] The loop iteration termination condition includes at least one of the number of iterations satisfying a first preset requirement and the loss function satisfying a second preset requirement. Specifically, the first preset requirement may be that the number of iterations reaches a preset value; the second preset requirement may be that the loss value calculated by the loss function is lower than a preset value, and the loss function is the difference between the value predicted by the predictor and the actual value. For example, in the first example, the loss function is the difference between the predicted value of the endpoint element and the actual value.

[0092] In an optional embodiment, the initial predictor includes an initial fully connected layer and an initial graph convolutional layer; each of the initial fully connected layers includes N channels;

[0093] The initial fully connected layer includes a fourth fully connected layer, a fifth fully connected layer and a sixth fully connected layer; the initial fully connected layer corresponds to the target fully connected layer in the target predictor; accordingly, the fourth fully connected layer corresponds to the first fully connected layer after the predictor training is completed; the fifth fully connected layer corresponds to the second fully connected layer after the predictor training is completed; the sixth fully connected layer corresponds to the third fully connected layer after the predictor training is completed; the initial graph convolution layer corresponds to the target fully connected layer after the predictor training is completed.

[0094] Among them, the fourth fully connected layer, the initial graph convolution layer, the fifth fully connected layer and the sixth fully connected layer are connected in sequence, each layer extracts features of the input data, and the extracted features are the output data; the output data of the previous layer is the input data of the next layer; according to the aforementioned correspondence, the function of each layer in the initial fully connected layer is also the same as the function of the corresponding target fully connected layer, such as the fourth fully connected layer and the first fully connected layer, the two use the same algorithm to extract features of the input data, and the parameters of the algorithm may be different. During the predictor training process, the parameters are continuously optimized, and finally the first fully connected layer is obtained; the other fully connected layers and graph convolution layers are the same, and the embodiments of the present application will not be repeated here.

[0095] In an optional embodiment, step 121 includes:

[0096] Step 1211, inputting a first sample independent variable of a first training sample in the training samples into an initial predictor;

[0097] Step 1212, at the initial fully connected layer of each of the initial predictors, calculating output data according to the first data relationship; and at the initial graph convolution layer of the initial predictor, calculating output data according to the second data relationship;

[0098] Step 1213: Obtain output data of the sixth fully connected layer as the first prediction content.

[0099] Combination Figure 2 , Figure 2 The fully connected layer 1 is the fourth fully connected layer in this embodiment, the fully connected layer 2 is the fifth fully connected layer in this embodiment, the fully connected layer 3 is the sixth fully connected layer, and the graph convolution layer is the initial graph convolution layer. In several fully connected layers in the initial fully connected layer, the first data relationship is used for feature extraction; it should be noted that the first data relationship of each fully connected layer is not exactly the same, and there may be different parameters; and the first data relationship of different channels in the same fully connected layer is not exactly the same, and there may be different parameters.

[0100] In the initial graph convolution layer, the data inputted by the fourth fully connected layer is feature extracted according to the second data relationship. For each node, the weighted sum of its adjacent nodes is performed according to the weight to obtain the feature value of the node. The weight is an element in the adjacency matrix. The adjacency matrix is ​​generated according to the sample dependent variable. The adjacency matrix is ​​used to represent the relationship between each two adjacent nodes. For example, Aij represents the relationship between node i and node j. If Aij is 0, it means that node i has no relationship with node j, and there is no edge between node i and node j in the graph convolution network. When Aij is not 0, it represents the weight, that is, the degree of association between node i and node j, and there is an edge between node i and node j in the graph convolution network, and different weights can be displayed through different edges. The adjacency matrix is ​​extracted from the sample data of the target predictor. After the target predictor is trained, the elements in the adjacency matrix are no longer adjusted when the target predictor is used to predict transactions.

[0101] The data output by the sixth fully connected layer is the first predicted content. During the subsequent reverse optimization, the initial predictor is reversely optimized according to the difference between the first content and the first sample dependent variable.

[0102] In an optional embodiment, the first data relationship is:

[0103] H1 (i) =σ1[H1 (i-1) W1 (i) +b (i) ]

[0104] Among them, H1 (i) Represents the output data of the i-th layer; H1 (i-1) Represents the output data of the i-1th layer; W1 (i) represents the first weight of the i-th layer, b (i) represents the bias parameter of the i-th layer; σ1(x) represents the first activation function of x.

[0105] Among them, combined Figure 2 , with the direction from input data to output data as the positive direction, each layer (fully connected layer and graph convolutional layer) is numbered sequentially, then i can be 3 or 4; when i = 1, that is, the fourth fully connected layer (or the first fully connected layer in the target predictor), H1 (1) =X, X is the input data; take i as 3 as an example, where H1 (3) =σ1[H1 (2) W1 (3) +b (3) ], the characteristic value (i.e. output data) of this layer is H1 (3) .

[0106] W1 (i) represents the first weight of the i-th layer, b (i)Represents the bias parameter of the i-th layer; in the predictor (initial predictor and target predictor), W1 (i) 、b (i) It can be the same or different in each layer; different W1 can also be used in different channels of the same layer. (i) , and different b (i) In the initial predictor, for each W1 (i) 、b (i) The initial values ​​are randomly set, and during the predictor training process, the values ​​of the two are continuously adjusted through reverse optimization to obtain the result of global optimization, that is, the value with the minimum global loss function is taken as the final value of the two.

[0107] σ1(x) represents the first activation function of x, that is, σ1[H1 (i-1) W1 (i) +b (i) ] means [H1 (i-1) W1 (i) +b (i) ] is an activation function, optionally, the activation function can be a linear rectification function (Rectified Linear Unit, ReLU) function.

[0108] In an optional embodiment, in step 1212, the step of calculating output data according to the second data relationship at the initial graph convolution layer of the initial predictor includes:

[0109] In the graph convolution layer, an adjacency matrix of the first sample dependent variable of the first training sample is generated;

[0110] Calculate output data according to the adjacency matrix and the second data relationship;

[0111] The second data relationship is:

[0112]

[0113] Among them, H2 (i) Represents the output data of the i-th layer, combined with Figure 2 , taking the direction from input data to output data as the positive direction, number each layer (fully connected layer and graph convolutional layer) sequentially, then i is 2.

[0114] H2 (i-1) Represents the output data of the i-1th layer, that is, the output data of the fourth fully connected layer (or the first fully connected layer), that is, H1 (1) .

[0115] W (i) represents the weight of the i-th layer, W (i) That is W (2)During the training of the predictor, W is continuously optimized through reverse optimization. (2) The value of is adjusted to obtain the result of global optimization, that is, the value of the global loss function with the minimum value is taken as W (2) The final value of

[0116] σ2(x) represents the second activation function of x, that is , optionally, the activation function may be a ReLU function; wherein σ2(x) and σ1(x) may use the same activation function or different ones, which is not limited in the embodiments of the present application.

[0117] A represents the adjacency matrix, where represents the adjacency matrix with self-connection, I q Represents the q*q unit matrix; q is the number of nodes in the graph convolutional layer, q=N, that is, the number of nodes is the same as the number of channels mentioned above, and both are the number of dependent variables.

[0118] D represents the degree matrix of A, which is the measurement matrix, so that A can be matrixed.

[0119] The left and right multiplications of are used to normalize the adjacency matrix.

[0120] In an optional embodiment, the step of generating an adjacency matrix of the first sample dependent variable of the first training sample includes:

[0121] Generate an adjacency matrix of the first sample dependent variable according to the first sample dependent variable and the third data relationship;

[0122] The third data relationship is:

[0123]

[0124] stA=A T

[0125] A ii =0, A ij ≥0, (i,j=1,2,...,q)

[0126] Among them, q represents the number of nodes in the graph convolution layer; q=N, that is, the number of nodes is the same as the number of channels mentioned above, and both are the number of dependent variables.

[0127] A represents the adjacency matrix; A T represents the transposed matrix of A;

[0128] Indicates (Y train -Y train A)2 The second norm of , the norm is a function with the concept of length, which is used to measure the size of elements in the matrix space.

[0129] ‖A‖1 represents the first norm of A; λ is a preset square coefficient. In the embodiment of the present application, the first norm refers to the L1 norm, which is the sum of the absolute values ​​of each element in the matrix; the second norm is the L2 norm, which refers to taking the square root of the total sum of squares in the matrix, and the total sum of squares is the sum of the squares of each element.

[0130] The third data relationship is the objective function of establishing the adjacency matrix. In the process of constructing the graph structure, the sparse coding method is used to learn the sparse representation between multiple dependent variables to obtain the adjacency matrix, so that the edges in the graph structure can describe the strength of the correlation between the nodes. It means that the first sample dependent variable matrix is ​​reconstructed as much as possible through the adjacency matrix A, and the diagonal elements of the adjacency matrix A are constrained to be zero, so that the elements of each column in the adjacency matrix A can represent the constructed relationship of other dependent variables to a certain dependent variable, so as to describe the correlation between multiple dependent variables.

[0131] In addition, in order to avoid the cancellation of positive and negative correlations in the subsequent adjacency matrix standardization process, the correlation between multiple dependent variables is assumed to be symmetric and non-negative, and the ‖A‖1 norm term can compress some smaller elements in the adjacency matrix to zero, making the adjacency matrix sparse. Optionally, the above objective function can be solved by a convex optimization method, which will not be described in detail in the embodiments of this application.

[0132] In the process of constructing the adjacency matrix, the graph structure is obtained through sparse coding learning, and a multi-channel graph convolutional network structure is designed to effectively utilize the correlation information between multiple dependent variables, improve the performance of the model on some dependent variables with poor prediction performance and the comprehensive performance of soft measurement.

[0133] In an optional embodiment, the process of training the predictor requires calculating the loss value of the predictor, and step 122 includes:

[0134] The loss value of the optimized predictor is calculated according to the first prediction content, the first sample dependent variable of the first training sample, and a fourth data relationship, wherein the fourth data relationship is:

[0135]

[0136] Among them, the loss function of the fourth data relation target predictor training process, taking the training optimization predictor as an example, represents the loss value of the optimized predictor, that is, the difference between the actual dependent variable and the predicted dependent variable, Y1 represents the first sample dependent variable, Represents the first prediction content; p represents the number of the first sample dependent variables, m represents the number of the first training samples, and when calculating the loss function, the differences of all training samples are counted. In the later process of reverse optimization of the predictor, the parameters in the feature extraction data relationship of each layer of the predictor are adjusted to minimize the value of the loss function, that is, the difference between the actual value and the predicted value is minimized, thereby improving the prediction accuracy of the target predictor; γ represents the preset balance coefficient; W3 represents the regularization term of the graph convolution layer of the optimized predictor, which is the second norm of the graph convolution layer, that is, the square root of the total sum of squares in the feature matrix of the graph convolution layer, which is used to avoid overfitting of the target predictor.

[0137] In an optional embodiment, the predictor to be tested is tested by a test sample, and step 13 includes:

[0138] Inputting the second sample independent variable of each of the test samples into the predictor to be tested to obtain a second prediction content;

[0139] The regression parameters of the predictor to be tested are determined according to the second sample dependent variable of the test sample and the second prediction content; the regression parameters are regression coefficients, which are used to represent the influence of the independent variable on the dependent variable; the regression parameters are determined according to the difference between the second sample dependent variable and the second prediction content.

[0140] If the regression parameter meets the third preset requirement, the predictor to be tested is determined to be the target predictor. The regression parameter may include multiple types of parameters, and the third preset requirement includes specific requirements for each parameter; specifically, in an optional embodiment, the step of determining the regression parameter of the predictor to be tested according to the second sample dependent variable of the test sample and the second prediction content includes: a first step and a second step, and it should be noted that the first step and the second step can be performed simultaneously.

[0141] The first step is to determine the first regression parameter of the predictor to be tested according to the second sample dependent variable of the test sample, the second prediction content and the fifth data relationship, wherein the fifth data relationship is:

[0142]

[0143] Among them, R 2 represents the first regression parameter, n represents the number of the test samples, Y2 represents the second sample dependent variable, represents the second predicted content, represents the mean of the dependent variable of the second sample; R 2 That is, the coefficient of determination, when the second prediction content is compared with the mean of the second sample dependent variable, through R 2The coefficient of determination reflects how much of the fluctuation of the dependent variable can be explained by the fluctuation of the independent variable, and its value range is usually between 0 and 1; R 2 The larger the value, the higher the degree of explanation of the independent variable on the dependent variable, indicating that the performance of the current predictor to be tested is better; therefore, the third preset requirement can include R 2 The lowest threshold value, when R 2 When it is greater than or equal to the threshold, the predictor meets the third preset requirement.

[0144] In the second step, according to the second sample dependent variable of the test sample, the second prediction content and the fifth data relationship, the second regression parameter of the predictor to be tested is determined, and the sixth data relationship is:

[0145]

[0146] Among them, RMSE represents the second regression parameter, and RMSE is the root mean square error. The difference between the second prediction content and the second sample dependent variable is first squared, then the mean is calculated, and finally the square root is taken; the smaller the RMSE value, the better the performance of the predictor to be tested and the lower the prediction difference.

[0147] Combined with the first example, taking steelmaking as an example, the independent variables include parameters such as alloy addition amount, process variables, and molten steel related data; and the dependent variable is the end point element content, such as acid-soluble aluminum (Als), carbon (C), chromium (Cr), molybdenum (Mo), etc.; since the end point element content is difficult to measure and control directly, and some elements between the end point elements will affect each other, the dependent variable is predicted based on the independent variable. Combined with Table 1 and Table 2, the actual prediction results of the target predictor in one embodiment of the present application are compared with the results of other prediction models. The evaluation index of Table 1 is R 2 ,The evaluation index of Table 2 is RMSE; among them, the table marked with (X) is the best prediction performance of the corresponding endpoint element.

[0148] Other prediction models include the partial least squares model, support vector regression model, fully connected network model, extreme learning machine model, and graph convolutional network model in the prior art, and the last one is the target predictor in the embodiment of the present application.

[0149] Table 1:

[0150]

[0151]

[0152] Table 2:

[0153]

[0154]

[0155] Based on Table 1 and Table 2, it can be seen that most elements have the best evaluation indicators of the target predictor in the embodiment of the present application. At the same time, in terms of average results and average rankings, the target predictor in the embodiment of the present application is significantly better than other comparison models; in addition, since the correlation between multiple dependent variables is considered in the process of establishing the target predictor, the prediction performance of the target predictor on elements Als, P, and V is significantly better than other elements, and the features extracted from the elements with better prediction performance can be used to help predict elements with poor performance. Therefore, the best performance is achieved on these three elements. Even if the performance on individual elements is not the best, the comprehensive performance is also improved to a certain extent compared with other models.

[0156] In the above embodiment of the present application, the data to be predicted of the target transaction is obtained, and the data to be predicted is input into the target predictor to obtain a prediction result; the prediction result includes at least two dependent variables, and the target predictor includes a graph convolution layer, which is a graph structure with multiple dependent variables as nodes, and the graph structure uses a graph convolution operation to connect the dependent variables with correlation to each other, so that different dependent variables influence each other in the prediction process; based on the coupling relationship between the dependent variables, the performance of the target predictor on some dependent variables with poor prediction performance is improved, the comprehensive soft measurement performance of the multi-dependent variable prediction is improved, and the prediction accuracy is improved.

[0157] On the basis of the above embodiments, the embodiment of the present application also provides a data processing method. In the prediction process of multiple dependent variables, the relationship between related variables is mined through a graph convolution layer; based on a graph structure with multiple dependent variables as nodes, a graph convolution operation is used to connect the nodes with related relationships, so that different dependent variables influence each other in the prediction process, so as to improve the comprehensive soft measurement performance of multi-dependent variable prediction.

[0158] Reference Figure 3 , shows a flowchart of a data processing method according to an embodiment of the present application, the method comprising:

[0159] Step 301, obtaining data to be predicted of a target transaction; the target transaction includes a production transaction or a parameter prediction transaction executed according to the data to be predicted.

[0160] The target transaction can be an industrial production transaction, such as steelmaking. With the continuous advancement of automation technology, modern industry has gradually entered the stage of fully automated production. In industrial production, thanks to the core idea of ​​closed-loop feedback control in automation technology, the production process can be regulated by sensing the controlled variables in industrial production, so as to effectively overcome the disturbance in the production process and realize the automated control of the production process.

[0161] The target task can also be other prediction tasks, such as temperature prediction, sewage pollution prediction, etc. The data to be predicted is the independent variable of the prediction process, which is usually a variable that is easy to measure directly, and the dependent variable is predicted based on the data to be measured. The dependent variable is the prediction result, which is usually a variable that is difficult to predict.

[0162] As a first example, if the target transaction is steelmaking, the independent variables include parameters such as alloy addition, process variables, and molten steel related data; alloy addition refers to the addition amount of each alloy; process variables include temperature, time, and other variables. The dependent variable is the end-point element content, such as acid-soluble aluminum (Als), carbon (C), chromium (Cr), molybdenum (Mo), etc.; since the end-point element content is difficult to measure and control directly, and some elements between the end-point elements will affect each other, the dependent variable is predicted based on the independent variables.

[0163] As a second example, the target transaction is a temperature prediction transaction, and the independent variables include humidity percentage, dew point temperature, wind speed, cloud coverage, etc., as well as the sampled temperature value; the dependent variable is the temperature of each unit time within the target prediction time period. For example, if the target prediction time period is 1 day, then the dependent variable is every hour of the day, and the temperature of each hour has a certain correlation, that is, there is a certain coupling between the dependent variables.

[0164] As a third example, the target task is a sewage pollution prediction task. The independent variables include variables such as temperature, flow, pH, dissolved oxygen concentration, redox potential, turbidity and conductivity; the dependent variables include chemical oxygen demand (COD), nitrate, phosphate concentration, etc. The detection of each dependent variable takes a long time and the process is complicated; therefore, the dependent variable is predicted by the independent variable.

[0165] Step 302, input the data to be predicted into the target predictor to obtain a prediction result; wherein the prediction result includes at least two dependent variables; the data to be predicted includes the independent variable of the dependent variable;

[0166] The target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample dependent variables in the sample data.

[0167] Among them, the prediction result includes at least two dependent variables, that is, the embodiment of the present application is a multi-dependent variable prediction process, and if there is a certain correlation between the multiple dependent variables in the actual production application process, they will affect each other.

[0168] In order to introduce the correlation information between multiple dependent variables, the graph convolution network is a machine learning model for mining the mutual relationship in the graph structure that has developed rapidly in recent years. In the embodiment of the present application, the target predictor is a pre-trained predictor, and the correlation information between multiple dependent variables is introduced during the training process. The target predictor includes a graph convolution layer, and the graph convolution layer includes a graph structure obtained by training according to sample data. The nodes of the graph structure include sample dependent variables in the sample data, and the sample dependent variables are the dependent variables in the sample data, that is, the known dependent variable values; in the sample data, each group of data includes sample independent variables and sample dependent variables; the sample independent variables are input into the predictor to obtain the predicted dependent variables, and then the predictor is reversely optimized according to the gap between the known sample dependent variables and the predicted dependent variables, and the predictor that finally meets the accuracy requirements is the target predictor.

[0169] Graph convolutional networks have been widely used in social network relationship analysis, file classification, traffic prediction and other fields due to their ability to conveniently perform operations in non-Euclidean spaces. The graph convolutional layer is a graph structure with sample dependent variables as nodes. The graph structure is usually composed of nodes and edges, and the relationships between nodes are described by edges. Figure 1 As shown, Figure 1 Each node N in corresponds to a feature of the graph convolution layer, and the edge between adjacent nodes N represents the relationship between two adjacent nodes. It can be understood that Figure 1 Only the graph convolution layer is shown in the figure. The actual number of nodes may not be Figure 1 Through the graph convolution operation, the weighted sum of the features of the adjacent nodes of each node is used as the feature of the node; therefore, through the graph convolution operation, the features of the nodes can be extracted based on the mutual relationship described under the graph structure.

[0170] After the data to be predicted is input into the target predictor as input data, it is subjected to feature extraction through the graph convolution layer to obtain output data as the prediction result; Figure 1 As shown, the output data includes n dependent variables, each dependent variable corresponds to a channel, and in the graph convolution layer, the channels of each dependent variable with mutual relationship are connected, so that the graph convolution layer refers to this mutual relationship when extracting features, so that the mutual relationship between multiple dependent variables affects the prediction results.

[0171] In the prior art, graph convolutional networks usually regard samples as nodes and mine the relationships between samples instead of the relationships between non-variables. Therefore, it is impossible to effectively introduce the correlation information between multiple dependent variables. In the embodiments of the present application, for industrial soft measurement scenarios with multiple dependent variables, the correlation information between multiple dependent variables is introduced to improve the prediction performance.

[0172] In addition, in the prior art graph convolutional networks, the relationship between nodes is usually calculated using distance metrics, and distance cannot accurately describe the correlation information between nodes. In the embodiment of the present application, each channel in the target predictor can use the independent variable information to independently learn different dependent variables, and at the same time, multiple dependent variables are used as different nodes in the graph convolution layer, and the strength of the correlation between nodes is described through the edges in the graph convolution layer, accurately measuring the correlation between nodes.

[0173] Referring to the first example above, taking steelmaking as an example, the independent variables include alloy addition, process variables, molten steel related data, etc.; the independent variables are input into the target predictor, and the target predictor extracts features from the independent variables. Through the graph convolution layer of the target predictor, the correlation between the dependent variables is integrated in the prediction process to obtain the dependent variable as the prediction result; the dependent variable is the end-point element content, and the end-point element is, for example, acid-soluble aluminum (Als). Elements such as carbon (C), chromium (Cr), molybdenum (Mo), etc., the prediction result includes the content of each end-point element; since the target predictor is a pre-trained predictor, its accuracy meets the preset accuracy requirements, so the accuracy of the end-point element content predicted by the target predictor is also high.

[0174] In an optional embodiment, the target predictor further includes a target fully connected layer, and the target fully connected layer includes a first fully connected layer, a second fully connected layer, and a third fully connected layer;

[0175] Wherein, the first fully connected layer, the graph convolutional layer, the second fully connected layer and the third fully connected layer are connected in sequence;

[0176] Each layer extracts features from the input data, and the extracted features are the output data;

[0177] The output data of the previous layer is the input data of the next layer;

[0178] The input data of the first fully connected layer is the data to be predicted, and the output data of the third fully connected layer is the prediction result.

[0179] In an optional embodiment, each of the target fully connected layers includes N channels, where N is the number of the dependent variables, and the channel processes one of the dependent variables.

[0180] In an optional embodiment, in the graph structure, the characteristic value of each node is a result of weighted summation of adjacent nodes;

[0181] The weight in the graph structure is the weight between each two adjacent nodes.

[0182] The weights are elements in an adjacency matrix, and the adjacency matrix is ​​generated according to the sample dependent variable.

[0183] In an optional embodiment, before the step of obtaining the data to be predicted of the target transaction, the method includes:

[0184] Acquire the sample data; wherein the sample data includes a training sample and a test sample;

[0185] According to the training samples, an initial predictor is trained to obtain a predictor to be tested;

[0186] The predictor to be tested is tested according to the test sample to obtain the target predictor.

[0187] In an optional embodiment, the sample in the sample data includes a sample independent variable and the sample dependent variable;

[0188] The step of training the initial predictor according to the training sample to obtain the predictor to be tested comprises:

[0189] Inputting a first sample independent variable of a first training sample in the training samples into an initial predictor to obtain a first prediction content;

[0190] According to the first prediction content and the first sample dependent variable of the first training sample, reversely optimize the initial predictor to obtain an optimized predictor;

[0191] The second sample independent variable in the training sample is iteratively input into the optimized predictor, and the reverse optimization is performed until the number of iterations meets the first preset requirement and / or the loss function meets the second preset requirement, so as to obtain the predictor to be tested.

[0192] In an optional embodiment, the initial predictor includes an initial fully connected layer and an initial graph convolutional layer; each of the initial fully connected layers includes N channels;

[0193] The initial fully connected layer includes a fourth fully connected layer, a fifth fully connected layer and a sixth fully connected layer;

[0194] Among them, the fourth fully connected layer, the initial graph convolution layer, the fifth fully connected layer and the sixth fully connected layer are connected in sequence, each layer extracts features from the input data, and the extracted features are output data; the output data of the previous layer is the input data of the next layer.

[0195] In an optional embodiment, the step of inputting the first sample independent variable of the first training sample in the training samples into the initial predictor to obtain the first prediction content includes:

[0196] Inputting a first sample independent variable of a first training sample among the training samples into an initial predictor;

[0197] At an initial fully connected layer of each of the initial predictors, output data is calculated according to the first data relationship; and at an initial graph convolution layer of the initial predictor, output data is calculated according to the second data relationship;

[0198] The output data of the sixth fully connected layer is obtained as the first prediction content.

[0199] In an optional embodiment, the first data relationship is:

[0200] H1 (i) =σ1[H1 (i-1) W1 (i) +b (i) ]

[0201] Among them, H1 (i) Represents the output data of the i-th layer; H1 (i-1) Represents the output data of the i-1th layer; W1 (i) represents the first weight of the i-th layer, b (i) represents the bias parameter of the i-th layer; σ1(x) represents the first activation function of x.

[0202] In an optional embodiment, the step of calculating output data according to the second data relationship at the initial graph convolution layer of the initial predictor comprises:

[0203] In the graph convolution layer, an adjacency matrix of the first sample dependent variable of the first training sample is generated;

[0204] Calculate output data according to the adjacency matrix and the second data relationship;

[0205] The second data relationship is:

[0206]

[0207] Among them, H2 (i) Represents the output data of the i-th layer; H2 (i-1) Represents the output data of the i-1th layer; W (i) represents the weight of the i-th layer; σ2(x) represents the second activation function of x;

[0208] A represents the adjacency matrix, and D represents the degree matrix of A;

[0209] W2 (i) represents the second weight of the i-th layer.

[0210] In an optional embodiment, the step of generating an adjacency matrix of the first sample dependent variable of the first training sample includes:

[0211] Generate an adjacency matrix of the first sample dependent variable according to the first sample dependent variable and the third data relationship;

[0212] The third data relationship is:

[0213]

[0214] stA=A T

[0215] A ii =0, A ij ≥0, (i,j=1,2,...,q)

[0216] Wherein, q represents the number of nodes in the graph convolutional layer;

[0217] A represents the adjacency matrix; Y train represents the first sample dependent variable;

[0218] Indicates (Y train -Y train A) 2 The second norm of ;

[0219] ‖A‖1 represents the first norm of A; λ is the preset square coefficient.

[0220] In an optional embodiment, the step of reversely optimizing the initial predictor according to the first prediction content and the first sample dependent variable of the first training sample comprises:

[0221] The loss value of the optimized predictor is calculated according to the first prediction content, the first sample dependent variable of the first training sample, and a fourth data relationship, wherein the fourth data relationship is:

[0222]

[0223] in, represents the loss value of the optimized predictor, Y1 represents the first sample dependent variable, represents the first prediction content, p represents the number of the first sample dependent variables, m represents the number of the first training samples; γ represents the preset balance coefficient; W3 represents the regularization term of the graph convolution layer of the optimized predictor.

[0224] In an optional embodiment, the step of testing the predictor to be tested according to the test sample to obtain the target predictor includes:

[0225] Inputting the second sample independent variable of each of the test samples into the predictor to be tested to obtain a second prediction content;

[0226] Determining the regression parameters of the predictor to be tested according to the second sample dependent variable of the test sample and the second prediction content;

[0227] If the regression parameter meets the third preset requirement, the predictor to be tested is determined to be the target predictor.

[0228] In an optional embodiment, the step of determining the regression parameter of the predictor to be tested according to the second sample dependent variable and the second prediction content of the test sample includes:

[0229] The first regression parameter of the predictor to be tested is determined according to the second sample dependent variable of the test sample, the second prediction content and the fifth data relationship, wherein the fifth data relationship is:

[0230]

[0231] Among them, R 2 represents the first regression parameter, n represents the number of the test samples, Y2 represents the second sample dependent variable, represents the second predicted content, represents the mean of the dependent variable of the second sample;

[0232] And determining the second regression parameter of the predictor to be tested according to the second sample dependent variable of the test sample, the second prediction content and the sixth data relationship, the sixth data relationship being:

[0233]

[0234] Among them, RMSE represents the second regression parameter.

[0235] In an optional embodiment, before the step of training the initial predictor according to the training samples, the method further comprises:

[0236] The sample data is normalized.

[0237] In the above embodiment of the present application, the data to be predicted of the target transaction is obtained, and the data to be predicted is input into the target predictor to obtain a prediction result; the prediction result includes at least two dependent variables, and the target predictor includes a graph convolution layer, which is a graph structure with multiple dependent variables as nodes, and the graph structure uses a graph convolution operation to connect the dependent variables with correlation to each other, so that different dependent variables influence each other in the prediction process; based on the coupling relationship between the dependent variables, the performance of the target predictor on some dependent variables with poor prediction performance is improved, the comprehensive soft measurement performance of the multi-dependent variable prediction is improved, and the prediction accuracy is improved.

[0238] On the basis of the above embodiments, the embodiments of the present application also provide a data processing method. In the production field, in the prediction process of multiple dependent variables, the relationship between related variables is mined through a graph convolution layer; based on a graph structure with multiple dependent variables as nodes, a graph convolution operation is used to connect the nodes with related relationships, so that different dependent variables influence each other in the prediction process, so as to improve the comprehensive soft measurement performance of multi-dependent variable prediction.

[0239] Reference Figure 4 , shows a flowchart of a data processing method according to an embodiment of the present application, the method comprising:

[0240] Step 401, obtaining independent variable data of a target production transaction; the independent variable data includes the amount of a starting element added, or the independent variable data includes the amount of a starting element added and a process control variable.

[0241] With the continuous advancement of automation technology, modern industry has gradually entered the stage of fully automated production. In industrial production, thanks to the core idea of ​​closed-loop feedback control in automation technology, the production process can be regulated by sensing the controlled variables in industrial production, so as to effectively overcome the disturbances in the production process and realize the automated control of the production process.

[0242] The target production business is such as steelmaking, the starting element addition amount is such as element addition amount, alloy addition amount; process control variables are such as temperature, time and other variables. The dependent variable is the end element content, such as acid-soluble aluminum (Als), carbon (C), chromium (Cr), molybdenum (Mo), etc.; since the end element content is difficult to measure and control directly, and some elements between the end elements will affect each other, the dependent variable is predicted based on the independent variable.

[0243] Step 402, inputting the independent variable data into the target predictor to obtain a prediction result; wherein the prediction result includes at least two dependent variables; the dependent variable includes the end point element content;

[0244] The target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample dependent variables in the sample data.

[0245] Among them, the prediction result includes at least two dependent variables, that is, the embodiment of the present application is a multi-dependent variable prediction process, and if there is a certain correlation between the multiple dependent variables in the actual production application process, they will affect each other.

[0246] In order to introduce the correlation information between multiple dependent variables, the graph convolution network is a machine learning model for mining the mutual relationship in the graph structure that has developed rapidly in recent years. In the embodiment of the present application, the target predictor is a pre-trained predictor, and the correlation information between multiple dependent variables is introduced during the training process. The target predictor includes a graph convolution layer, and the graph convolution layer includes a graph structure obtained by training according to sample data. The nodes of the graph structure include sample dependent variables in the sample data, and the sample dependent variables are the dependent variables in the sample data, that is, the known dependent variable values; in the sample data, each group of data includes sample independent variables and sample dependent variables; the sample independent variables are input into the predictor to obtain the predicted dependent variables, and then the predictor is reversely optimized according to the gap between the known sample dependent variables and the predicted dependent variables, and the predictor that finally meets the accuracy requirements is the target predictor.

[0247] Graph convolutional networks have been widely used in social network relationship analysis, file classification, traffic prediction and other fields due to their ability to conveniently perform operations in non-Euclidean spaces. The graph convolutional layer is a graph structure with sample dependent variables as nodes. The graph structure is usually composed of nodes and edges, and the relationships between nodes are described by edges. Figure 1 As shown, Figure 1 Each node N in corresponds to a feature of the graph convolution layer, and the edge between adjacent nodes N represents the relationship between two adjacent nodes. It can be understood that Figure 1 Only the graph convolution layer is shown in the figure. The actual number of nodes may not be Figure 1 Through the graph convolution operation, the weighted sum of the features of the adjacent nodes of each node is used as the feature of the node; therefore, through the graph convolution operation, the features of the nodes can be extracted based on the mutual relationship described under the graph structure.

[0248] After the independent variable data is input into the target predictor as input data, it is subjected to feature extraction through the graph convolution layer to obtain the output data as the prediction result; Figure 1 As shown, the output data includes n dependent variables, each dependent variable corresponds to a channel, and in the graph convolution layer, the channels of each dependent variable with mutual relationship are connected, so that the graph convolution layer refers to this mutual relationship when extracting features, so that the mutual relationship between multiple dependent variables affects the prediction results.

[0249] In the prior art, graph convolutional networks usually regard samples as nodes and mine the relationships between samples instead of the relationships between non-variables. Therefore, it is impossible to effectively introduce the correlation information between multiple dependent variables. In the embodiments of the present application, for industrial soft measurement scenarios with multiple dependent variables, the correlation information between multiple dependent variables is introduced to improve the prediction performance.

[0250] In addition, in the prior art graph convolutional networks, the relationship between nodes is usually calculated using distance metrics, and distance cannot accurately describe the correlation information between nodes. In the embodiment of the present application, each channel in the target predictor can use the independent variable information to independently learn different dependent variables, and at the same time, multiple dependent variables are used as different nodes in the graph convolution layer, and the strength of the correlation between nodes is described through the edges in the graph convolution layer, accurately measuring the correlation between nodes.

[0251] Still taking steelmaking as an example, the independent variables include alloy addition, process variables, molten steel related data, etc.; the independent variables are input into the target predictor, the target predictor extracts features from the independent variables, and through the graph convolution layer of the target predictor, the correlation between the dependent variables is integrated in the prediction process to obtain the dependent variable as the prediction result; the dependent variable is the end-point element content, such as acid-soluble aluminum (Als). Elements such as carbon (C), chromium (Cr), molybdenum (Mo), etc., the prediction result includes the content of each end-point element; since the target predictor is a pre-trained predictor, its accuracy meets the preset accuracy requirements, so the accuracy of the end-point element content predicted by the target predictor is also high.

[0252] In an optional embodiment, after the step of obtaining the prediction result, the method includes:

[0253] According to the endpoint element content of the predicted result, the independent variable data is adjusted. For example, if the content of some elements in the endpoint element content is different from the user's expectation, the independent variable can be adjusted in the direction of offsetting the difference, such as adjusting some starting element contents or adjusting process variables. For example, the user can pre-set the target endpoint content of a certain element, and after obtaining the endpoint element content in the predicted result, compare the difference between the target endpoint content and the endpoint element content in the predicted result, and then adjust the independent variable randomly or according to the pre-set adjustment rules, and after obtaining the target endpoint content, output the corresponding independent variable.

[0254] The steps of this embodiment are similar to those of the above embodiment, and the details can be referred to the description of the above embodiment. That is, after step 402, the method provided by this embodiment includes other steps in the method provided by the above embodiment, which will not be described in detail in this embodiment.

[0255] In an embodiment of the present application, independent variable data of a target production transaction is obtained, and the independent variable data is input into a target predictor to obtain a prediction result; the prediction result includes at least two dependent variables; the dependent variable includes an end point element content, and the target predictor includes a graph convolution layer, the graph convolution layer is a graph structure with multiple dependent variables as nodes, and graph convolution operations are used in the graph structure to interconnect dependent variables with related relationships, so that different dependent variables influence each other in the prediction process; based on the coupling relationship between the dependent variables, the performance of the target predictor on some dependent variables with poor prediction performance is improved, the comprehensive soft measurement performance of multi-dependent variable prediction is improved, and the prediction accuracy is improved.

[0256] On the basis of the above embodiments, the embodiment of the present application further provides a data processing method. In the field of industrial production, in the process of predicting multiple prediction parameters (production results), the relationship between related variables is mined through a graph convolution layer; based on a graph structure with multiple prediction parameters as nodes, a graph convolution operation is used to connect the nodes with related relationships, so that different prediction parameters (in the prediction process) influence each other, so as to improve the comprehensive soft measurement performance of multi-prediction parameter prediction.

[0257] Reference Figure 5 , shows a flowchart of a data processing method according to an embodiment of the present application, the method comprising:

[0258] Step 501, obtaining industrial parameters of industrial production affairs.

[0259] With the continuous advancement of automation technology, modern industry has gradually entered the stage of fully automated production. In industrial production, thanks to the core idea of ​​closed-loop feedback control in automation technology, the production process can be regulated by sensing the controlled variables in industrial production, so as to effectively overcome the disturbances in the production process and realize the automated control of the production process.

[0260] Industrial production affairs such as steelmaking, for example, industrial parameters include starting element addition and process control variables. Starting element addition includes element addition and alloy addition; process control variables include temperature, time and other variables. The prediction parameters include endpoint element content, such as acid-soluble aluminum (Als), carbon (C), chromium (Cr), molybdenum (Mo), etc.; since the endpoint element content is difficult to measure and control directly, and some elements between endpoint elements will affect each other, prediction parameters are obtained based on industrial parameters.

[0261] Step 502, input the industrial parameters into a target predictor to obtain a prediction result; wherein the prediction result includes at least two prediction parameters; the target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample prediction parameters in the sample data.

[0262] Among them, the prediction result includes at least two prediction parameters, that is, the embodiment of the present application is a multi-prediction parameter prediction process, and if there is a certain correlation between the multiple prediction parameters in the actual production application process, they will affect each other.

[0263] In order to introduce the correlation information between multiple prediction parameters, the graph convolution network is a machine learning model for mining the mutual relationship in the graph structure that has developed rapidly in recent years. In the embodiment of the present application, the target predictor is a pre-trained predictor, and the correlation information between multiple prediction parameters is introduced during the training process. The target predictor includes a graph convolution layer, and the graph convolution layer includes a graph structure obtained by training according to sample data. The nodes of the graph structure include sample prediction parameters in the sample data, and the sample prediction parameters are prediction parameters in the sample data, that is, known prediction parameter values; in the sample data, each group of data includes sample industrial parameters and sample prediction parameters; the sample industrial parameters are input into the predictor to obtain predicted prediction parameters, and then the predictor is reversely optimized according to the gap between the known sample prediction parameters and the predicted prediction parameters, and the predictor that finally meets the accuracy requirements is the target predictor.

[0264] Graph convolutional networks have been widely used in social network relationship analysis, file classification, traffic prediction and other fields due to their ability to conveniently perform operations in non-Euclidean spaces. The graph convolutional layer is a graph structure with sample prediction parameters as nodes. The graph structure is usually composed of nodes and edges, and the relationships between nodes are described by edges. Figure 1 As shown, Figure 1 Each node N in corresponds to a feature of the graph convolution layer, and the edge between adjacent nodes N represents the relationship between two adjacent nodes. It can be understood that Figure 1 Only the graph convolution layer is shown in the figure. The actual number of nodes may not be Figure 1 Through the graph convolution operation, the weighted sum of the features of the adjacent nodes of each node is used as the feature of the node; therefore, through the graph convolution operation, the features of the nodes can be extracted based on the mutual relationship described under the graph structure.

[0265] After the industrial parameters are input into the target predictor as input data, they are extracted through the graph convolution layer to obtain the output data as the prediction result; Figure 1 As shown, the output data includes n prediction parameters, each prediction parameter corresponds to a channel, and in the graph convolution layer, each channel of the prediction parameters with mutual relationship is connected, so that the graph convolution layer refers to this mutual relationship when performing feature extraction, so that the mutual relationship between multiple prediction parameters affects the prediction result.

[0266] In the prior art, graph convolutional networks usually regard samples as nodes and mine the relationships between samples instead of the relationships between non-variables. Therefore, it is impossible to effectively introduce the correlation information between multiple prediction parameters. In the embodiments of the present application, for industrial soft measurement scenarios with multiple prediction parameters, the correlation information between multiple prediction parameters is introduced to improve the prediction performance.

[0267] In addition, in the prior art graph convolutional networks, the relationship between nodes is usually calculated using distance metrics, and distance cannot accurately describe the correlation information between nodes. In the embodiment of the present application, each channel in the target predictor can independently learn different prediction parameters using industrial parameter information, and at the same time, multiple prediction parameters are respectively used as different nodes in the graph convolution layer, and the strength of the correlation between nodes is described through the edges in the graph convolution layer, accurately measuring the correlation between nodes.

[0268] Still taking steelmaking as an example, industrial parameters include alloy addition, process variables, molten steel related data, etc.; the industrial parameters are input into the target predictor, and the target predictor extracts features from the industrial parameters. Through the graph convolution layer of the target predictor, the correlation between the various prediction parameters is integrated in the prediction process to obtain the prediction parameters as the prediction results; the prediction parameters are the end-point element content, such as acid-soluble aluminum (Als). Elements such as carbon (C), chromium (Cr), molybdenum (Mo), etc., and the prediction results include the content of each end-point element; since the target predictor is a pre-trained predictor, its accuracy meets the preset accuracy requirements, so the accuracy of the end-point element content predicted by the target predictor is also high.

[0269] In an optional embodiment, after the step of obtaining the prediction result, the method includes:

[0270] According to the endpoint element content of the prediction result, the industrial parameter data is adjusted. For example, if the content of some elements in the endpoint element content is different from the user's expectation, the industrial parameters can be adjusted in the direction of offsetting the difference, such as adjusting some starting element contents or adjusting process variables. For example, the user can pre-set the target endpoint content of a certain element, and after obtaining the endpoint element content in the prediction result, compare the difference between the target endpoint content and the endpoint element content in the prediction result, and then adjust the industrial parameters randomly or according to the pre-set adjustment rules, and after obtaining the target endpoint content, output the corresponding industrial parameters.

[0271] The steps of this embodiment are similar to those of the above embodiment, and the details can be referred to the description of the above embodiment. That is, after step 502, the method provided by this embodiment includes other steps in the method provided by the above embodiment, which will not be described in detail in this embodiment.

[0272] In an embodiment of the present application, industrial parameters of industrial production affairs are obtained, and the industrial parameters are input into a target predictor to obtain a prediction result; the prediction result includes at least two prediction parameters, and the target predictor includes a graph convolution layer, and the graph convolution layer is a graph structure with multiple prediction parameters as nodes. The graph structure uses a graph convolution operation to connect prediction parameters with related relationships to each other, so that different prediction parameters influence each other in the prediction process; based on the coupling relationship between the prediction parameters, the performance of the target predictor on some prediction parameters with poor prediction performance is improved, the comprehensive soft measurement performance of multi-prediction parameter prediction is improved, and the prediction accuracy is improved.

[0273] On the basis of the above embodiments, this embodiment further provides a data processing device, which is applied to electronic devices such as terminal devices and servers.

[0274] Reference Figure 6 , shows a structural block diagram of a data processing device embodiment of the present application, which may specifically include the following modules:

[0275] The data acquisition module 601 is used to acquire the data to be predicted of the target transaction; the target transaction includes a production transaction or a parameter prediction transaction executed according to the data to be predicted.

[0276] The target transaction can be an industrial production transaction, such as steelmaking. With the continuous advancement of automation technology, modern industry has gradually entered the stage of fully automated production. In industrial production, thanks to the core idea of ​​closed-loop feedback control in automation technology, the production process can be regulated by sensing the controlled variables in industrial production, so as to effectively overcome the disturbance in the production process and realize the automated control of the production process.

[0277] The target task can also be other prediction tasks, such as temperature prediction, sewage pollution prediction, etc. The data to be predicted is the independent variable of the prediction process, which is usually a variable that is easy to measure directly, and the dependent variable is predicted based on the data to be measured. The dependent variable is the prediction result, which is usually a variable that is difficult to predict.

[0278] As a first example, if the target business is steelmaking, the independent variables include parameters such as alloy addition, process variables, and molten steel related data; alloy addition refers to the addition of each alloy; process variables include temperature, time, and other variables. The dependent variable is the end-point element content, such as acid-soluble aluminum (Als), carbon (C), chromium (Cr), molybdenum (05o), etc.; since the end-point element content is difficult to measure and control directly, and some elements between the end-point elements will affect each other, the dependent variable is predicted based on the independent variables.

[0279] As a second example, the target transaction is a temperature prediction transaction, and the independent variables include humidity percentage, dew point temperature, wind speed, cloud coverage, etc., as well as the sampled temperature value; the dependent variable is the temperature of each unit time within the target prediction time period. For example, if the target prediction time period is 1 day, then the dependent variable is every hour of the day, and the temperature of each hour has a certain correlation, that is, there is a certain coupling between the dependent variables.

[0280] As a third example, the target task is a sewage pollution prediction task. The independent variables include variables such as temperature, flow, pH, dissolved oxygen concentration, redox potential, turbidity and conductivity; the dependent variables include chemical oxygen demand (COD), nitrate, phosphate concentration, etc. The detection of each dependent variable takes a long time and the process is complicated; therefore, the dependent variable is predicted by the independent variable.

[0281] The data prediction module 602 is used to input the data to be predicted into the target predictor to obtain a prediction result; wherein the prediction result includes at least two dependent variables; the data to be predicted includes an independent variable of the dependent variable;

[0282] The target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample dependent variables in the sample data.

[0283] Among them, the prediction result includes at least two dependent variables, that is, the embodiment of the present application is a multi-dependent variable prediction process, and if there is a certain correlation between the multiple dependent variables in the actual production application process, they will affect each other.

[0284] In order to introduce the correlation information between multiple dependent variables, the graph convolution network is a machine learning model for mining the mutual relationship in the graph structure that has developed rapidly in recent years. In the embodiment of the present application, the target predictor is a pre-trained predictor, and the correlation information between multiple dependent variables is introduced during the training process. The target predictor includes a graph convolution layer, and the graph convolution layer includes a graph structure obtained by training according to sample data. The nodes of the graph structure include sample dependent variables in the sample data, and the sample dependent variables are the dependent variables in the sample data, that is, the known dependent variable values; in the sample data, each group of data includes sample independent variables and sample dependent variables; the sample independent variables are input into the predictor to obtain the predicted dependent variables, and then the predictor is reversely optimized according to the gap between the known sample dependent variables and the predicted dependent variables, and the predictor that finally meets the accuracy requirements is the target predictor.

[0285] Graph convolutional networks have been widely used in social network relationship analysis, file classification, traffic prediction and other fields due to their ability to conveniently perform operations in non-Euclidean spaces. The graph convolutional layer is a graph structure with sample dependent variables as nodes. The graph structure is usually composed of nodes and edges, and the relationships between nodes are described by edges. Figure 1 As shown, Figure 1 Each node N in corresponds to a feature of the graph convolution layer, and the edge between adjacent nodes N represents the relationship between two adjacent nodes. It can be understood that Figure 1 Only the graph convolution layer is shown in the figure. The actual number of nodes may not be Figure 1 Through the graph convolution operation, the weighted sum of the features of the adjacent nodes of each node is used as the feature of the node; therefore, through the graph convolution operation, the features of the nodes can be extracted based on the mutual relationship described under the graph structure.

[0286] After the data to be predicted is input into the target predictor as input data, it is subjected to feature extraction through the graph convolution layer to obtain output data as the prediction result; Figure 1 As shown, the output data includes n dependent variables, each dependent variable corresponds to a channel, and in the graph convolution layer, the channels of each dependent variable with mutual relationship are connected, so that the graph convolution layer refers to this mutual relationship when extracting features, so that the mutual relationship between multiple dependent variables affects the prediction results.

[0287] In the prior art, graph convolutional networks usually regard samples as nodes and mine the relationships between samples instead of the relationships between non-variables. Therefore, it is impossible to effectively introduce the correlation information between multiple dependent variables. In the embodiments of the present application, for industrial soft measurement scenarios with multiple dependent variables, the correlation information between multiple dependent variables is introduced to improve the prediction performance.

[0288] In addition, in the prior art graph convolutional networks, the relationship between nodes is usually calculated using distance metrics, and distance cannot accurately describe the correlation information between nodes. In the embodiment of the present application, each channel in the target predictor can use the independent variable information to independently learn different dependent variables, and at the same time, multiple dependent variables are used as different nodes in the graph convolution layer, and the strength of the correlation between nodes is described through the edges in the graph convolution layer, accurately measuring the correlation between nodes.

[0289] Referring to the first example above, taking steelmaking as an example, the independent variables include alloy addition, process variables, molten steel related data, etc.; the independent variables are input into the target predictor, and the target predictor extracts features from the independent variables. Through the graph convolution layer of the target predictor, the correlation between the dependent variables is integrated in the prediction process to obtain the dependent variable as the prediction result; the dependent variable is the end-point element content, and the end-point element is, for example, acid-soluble aluminum (Als). Elements such as carbon (C), chromium (Cr), molybdenum (Mo), etc., the prediction result includes the content of each end-point element; since the target predictor is a pre-trained predictor, its accuracy meets the preset accuracy requirements, so the accuracy of the end-point element content predicted by the target predictor is also high.

[0290] In an optional embodiment, the target predictor further includes a target fully connected layer, and the target fully connected layer includes a first fully connected layer, a second fully connected layer, and a third fully connected layer;

[0291] Wherein, the first fully connected layer, the graph convolutional layer, the second fully connected layer and the third fully connected layer are connected in sequence;

[0292] Each layer extracts features from the input data, and the extracted features are the output data;

[0293] The output data of the previous layer is the input data of the next layer;

[0294] The input data of the first fully connected layer is the data to be predicted, and the output data of the third fully connected layer is the prediction result.

[0295] In an optional embodiment, each of the target fully connected layers includes N channels, where N is the number of the dependent variables, and the channel processes one of the dependent variables.

[0296] In an optional embodiment, in the graph structure, the characteristic value of each node is a result of weighted summation of adjacent nodes;

[0297] The weight in the graph structure is the weight between each two adjacent nodes.

[0298] The weights are elements in an adjacency matrix, and the adjacency matrix is ​​generated according to the sample dependent variable.

[0299] See also Figure 6 In an optional embodiment, the device comprises:

[0300] The sample data acquisition module 603 is used to acquire the sample data; wherein the sample data includes training samples and test samples;

[0301] A predictor training module 604 is used to train an initial predictor according to the training samples to obtain a predictor to be tested;

[0302] The predictor testing module 605 is used to test the predictor to be tested according to the test sample to obtain the target predictor.

[0303] In an optional embodiment, the sample in the sample data includes a sample independent variable and the sample dependent variable;

[0304] See also Figure 6 , the predictor training module 604 includes:

[0305] A first input submodule 6041 is used to input a first sample independent variable of a first training sample in the training samples into an initial predictor to obtain a first prediction content;

[0306] An optimization submodule 6042 is used to reversely optimize the initial predictor according to the first prediction content and the first sample dependent variable of the first training sample to obtain an optimized predictor;

[0307] The iterative processing submodule 6043 is used to iteratively input the second sample independent variable in the training sample into the optimized predictor and perform the reverse optimization until the number of iterations meets the first preset requirement and / or the loss function meets the second preset requirement, thereby obtaining the predictor to be tested.

[0308] In an optional embodiment, the initial predictor includes an initial fully connected layer and an initial graph convolutional layer; each of the initial fully connected layers includes N channels;

[0309] The initial fully connected layer includes a fourth fully connected layer, a fifth fully connected layer and a sixth fully connected layer;

[0310] Among them, the fourth fully connected layer, the initial graph convolution layer, the fifth fully connected layer and the sixth fully connected layer are connected in sequence, each layer extracts features from the input data, and the extracted features are output data; the output data of the previous layer is the input data of the next layer.

[0311] See also Figure 6 In an optional embodiment, the first input submodule 6041 includes:

[0312] An input unit, used for inputting a first sample independent variable of a first training sample among the training samples into an initial predictor;

[0313] A computing unit, configured to compute output data at an initial fully connected layer of each of the initial predictors according to the first data relationship; and to compute output data at an initial graph convolution layer of the initial predictors according to the second data relationship;

[0314] An acquisition unit is used to acquire output data of the sixth fully connected layer as the first prediction content.

[0315] In an optional embodiment, the first data relationship is:

[0316] H1 (i) =σ1[H1 (i-1) W1 (i) +b (i) ]

[0317] Among them, H1 (i) Represents the output data of the i-th layer; H1 (i-1) Represents the output data of the i-1th layer; W1 (i) represents the first weight of the i-th layer, b (i) represents the bias parameter of the i-th layer; σ1(x) represents the first activation function of x.

[0318] In an optional embodiment, the computing unit is used to:

[0319] In the graph convolution layer, an adjacency matrix of the first sample dependent variable of the first training sample is generated;

[0320] Calculate output data according to the adjacency matrix and the second data relationship;

[0321] The second data relationship is:

[0322]

[0323] Among them, H2 (i) Represents the output data of the i-th layer; H2 (i-1) Represents the output data of the i-1th layer; W (i) represents the weight of the i-th layer; σ2(x) represents the second activation function of x;

[0324] A represents the adjacency matrix, and D represents the degree matrix of A;

[0325] W2 (i) represents the second weight of the i-th layer.

[0326] In an optional embodiment, the computing unit is used to:

[0327] Generate an adjacency matrix of the first sample dependent variable according to the first sample dependent variable and the third data relationship;

[0328] The third data relationship is:

[0329]

[0330] stA=A T

[0331] A ii =0, A ij ≥0, (i,j=1,2,...,q)

[0332] Wherein, q represents the number of nodes in the graph convolutional layer;

[0333] A represents the adjacency matrix; Y train represents the first sample dependent variable;

[0334] Indicates (Y train -Y train A) 2 The second norm of ;

[0335] ‖A‖1 represents the first norm of A; λ is the preset square coefficient.

[0336] In an optional embodiment, the optimization submodule 6042 is used to:

[0337] The loss value of the optimized predictor is calculated according to the first prediction content, the first sample dependent variable of the first training sample, and a fourth data relationship, wherein the fourth data relationship is:

[0338]

[0339] in, represents the loss value of the optimized predictor, Y1 represents the first sample dependent variable, represents the first prediction content, p represents the number of the first sample dependent variables, 05 represents the number of the first training samples; γ represents the preset balance coefficient; W3 represents the regularization term of the graph convolution layer of the optimized predictor.

[0340] See also Figure 6 In an optional embodiment, the predictor test module 605 includes:

[0341] A second input submodule 6051 is used to input the second sample independent variable of each test sample into the predictor to be tested to obtain a second prediction content;

[0342] A first determination submodule 6052, configured to determine the regression parameters of the predictor to be tested according to the second sample dependent variable of the test sample and the second prediction content;

[0343] The second determining submodule 6053 is configured to determine the predictor to be tested as a target predictor if the regression parameter meets a third preset requirement.

[0344] In an optional embodiment, the first determining submodule 6052 includes:

[0345] The first determining unit is used to determine the first regression parameter of the predictor to be tested according to the second sample dependent variable of the test sample, the second prediction content and the fifth data relationship, wherein the fifth data relationship is:

[0346]

[0347] Among them, R 2 represents the first regression parameter, n represents the number of the test samples, Y2 represents the second sample dependent variable, represents the second predicted content, represents the mean of the dependent variable of the second sample;

[0348] and a second determining unit, configured to determine a second regression parameter of the predictor to be tested according to the second sample dependent variable of the test sample, the second prediction content and the fifth data relationship, wherein the sixth data relationship is:

[0349]

[0350] Among them, RMSE represents the second regression parameter.

[0351] See also Figure 6 In an optional embodiment, the device further comprises:

[0352] The normalization processing module 606 is used to perform normalization processing on the sample data.

[0353] In the above embodiment of the present application, the data acquisition module 601 acquires the data to be predicted of the target transaction, and the data prediction module 602 inputs the data to be predicted into the target predictor to obtain a prediction result; the prediction result includes at least two dependent variables, and the target predictor includes a graph convolution layer, and the graph convolution layer is a graph structure with multiple dependent variables as nodes. The graph structure uses a graph convolution operation to connect dependent variables with related relationships to each other, so that different dependent variables influence each other in the prediction process; based on the coupling relationship between the dependent variables, the performance of the target predictor on some dependent variables with poor prediction performance is improved, the comprehensive soft measurement performance of multi-dependent variable prediction is improved, and the prediction accuracy is improved.

[0354] On the basis of the above embodiments, this embodiment further provides a data processing device, which is applied to electronic devices such as terminal devices and servers.

[0355] Reference Figure 8 , shows a structural block diagram of a data processing device embodiment of the present application, which may specifically include the following modules:

[0356] The independent variable data acquisition module 801 is used to acquire the independent variable data of the target production transaction; the independent variable data includes the starting element addition amount, or the independent variable data includes the starting element addition amount and the process control variable.

[0357] With the continuous advancement of automation technology, modern industry has gradually entered the stage of fully automated production. In industrial production, thanks to the core idea of ​​closed-loop feedback control in automation technology, the production process can be regulated by sensing the controlled variables in industrial production, so as to effectively overcome the disturbances in the production process and realize the automated control of the production process.

[0358] The target production business is such as steelmaking, the starting element addition amount is such as element addition amount, alloy addition amount; process control variables are such as temperature, time and other variables. The dependent variable is the end element content, such as acid-soluble aluminum (Als), carbon (C), chromium (Cr), molybdenum (Mo), etc.; since the end element content is difficult to measure and control directly, and some elements between the end elements will affect each other, the dependent variable is predicted based on the independent variable.

[0359] The independent variable prediction module 802 is used to input the independent variable data into the target predictor to obtain a prediction result; wherein the prediction result includes at least two dependent variables; the dependent variable includes the end point element content;

[0360] The target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample dependent variables in the sample data.

[0361] Among them, the prediction result includes at least two dependent variables, that is, the embodiment of the present application is a multi-dependent variable prediction process, and if there is a certain correlation between the multiple dependent variables in the actual production application process, they will affect each other.

[0362] In order to introduce the correlation information between multiple dependent variables, the graph convolution network is a machine learning model for mining the mutual relationship in the graph structure that has developed rapidly in recent years. In the embodiment of the present application, the target predictor is a pre-trained predictor, and the correlation information between multiple dependent variables is introduced during the training process. The target predictor includes a graph convolution layer, and the graph convolution layer includes a graph structure obtained by training according to sample data. The nodes of the graph structure include sample dependent variables in the sample data, and the sample dependent variables are the dependent variables in the sample data, that is, the known dependent variable values; in the sample data, each group of data includes sample independent variables and sample dependent variables; the sample independent variables are input into the predictor to obtain the predicted dependent variables, and then the predictor is reversely optimized according to the gap between the known sample dependent variables and the predicted dependent variables, and the predictor that finally meets the accuracy requirements is the target predictor.

[0363] Graph convolutional networks have been widely used in social network relationship analysis, file classification, traffic prediction and other fields due to their ability to conveniently perform operations in non-Euclidean spaces. The graph convolutional layer is a graph structure with sample dependent variables as nodes. The graph structure is usually composed of nodes and edges, and the relationships between nodes are described by edges. Figure 1 As shown, Figure 1 Each node N in corresponds to a feature of the graph convolution layer, and the edge between adjacent nodes N represents the relationship between two adjacent nodes. It can be understood that Figure 1 Only the graph convolution layer is shown in the figure. The actual number of nodes may not be Figure 1 Through the graph convolution operation, the weighted sum of the features of the adjacent nodes of each node is used as the feature of the node; therefore, through the graph convolution operation, the features of the nodes can be extracted based on the mutual relationship described under the graph structure.

[0364] After the data to be predicted is input into the target predictor as input data, it is subjected to feature extraction through the graph convolution layer to obtain output data as the prediction result; Figure 1 As shown, the output data includes n dependent variables, each dependent variable corresponds to a channel, and in the graph convolution layer, the channels of each dependent variable with mutual relationship are connected, so that the graph convolution layer refers to this mutual relationship when extracting features, so that the mutual relationship between multiple dependent variables affects the prediction results.

[0365] In the prior art, graph convolutional networks usually regard samples as nodes and mine the relationships between samples instead of the relationships between non-variables. Therefore, it is impossible to effectively introduce the correlation information between multiple dependent variables. In the embodiments of the present application, for industrial soft measurement scenarios with multiple dependent variables, the correlation information between multiple dependent variables is introduced to improve the prediction performance.

[0366] In addition, in the prior art graph convolutional networks, the relationship between nodes is usually calculated using distance metrics, and distance cannot accurately describe the correlation information between nodes. In the embodiment of the present application, each channel in the target predictor can use the independent variable information to independently learn different dependent variables, and at the same time, multiple dependent variables are used as different nodes in the graph convolution layer, and the strength of the correlation between nodes is described through the edges in the graph convolution layer, accurately measuring the correlation between nodes.

[0367] Still taking steelmaking as an example, the independent variables include alloy addition, process variables, molten steel related data, etc.; the independent variables are input into the target predictor, the target predictor extracts features from the independent variables, and through the graph convolution layer of the target predictor, the correlation between the dependent variables is integrated in the prediction process to obtain the dependent variable as the prediction result; the dependent variable is the end-point element content, such as acid-soluble aluminum (Als). Elements such as carbon (C), chromium (Cr), molybdenum (Mo), etc., the prediction result includes the content of each end-point element; since the target predictor is a pre-trained predictor, its accuracy meets the preset accuracy requirements, so the accuracy of the end-point element content predicted by the target predictor is also high.

[0368] In an optional embodiment, the device further comprises:

[0369] The adjustment module is used to adjust the independent variable data according to the endpoint element content of the prediction result.

[0370] Other modules included in this embodiment are similar to those in the above embodiment. For details, please refer to the description of the above embodiment, and this embodiment will not be described in detail here.

[0371] In an embodiment of the present application, the independent variable data acquisition module 801 acquires the independent variable data of the target production transaction, and the independent variable prediction module 802 inputs the independent variable data into the target predictor to obtain a prediction result; the prediction result includes at least two dependent variables; the dependent variable includes the endpoint element content, and the target predictor includes a graph convolution layer, the graph convolution layer is a graph structure with multiple dependent variables as nodes, and the graph structure uses a graph convolution operation to connect dependent variables with correlation to each other, so that different dependent variables influence each other in the prediction process; based on the coupling relationship between the dependent variables, the performance of the target predictor on some dependent variables with poor prediction performance is improved, the comprehensive soft measurement performance of multi-dependent variable prediction is improved, and the prediction accuracy is improved.

[0372] On the basis of the above embodiments, this embodiment further provides a data processing device, which is applied to electronic devices such as terminal devices and servers.

[0373] Reference Fig. 9 , shows a structural block diagram of a data processing device embodiment of the present application, which may specifically include the following modules:

[0374] The parameter acquisition module 901 is used to acquire industrial parameters of industrial production affairs.

[0375] With the continuous advancement of automation technology, modern industry has gradually entered the stage of fully automated production. In industrial production, thanks to the core idea of ​​closed-loop feedback control in automation technology, the production process can be regulated by sensing the controlled variables in industrial production, so as to effectively overcome the disturbances in the production process and realize the automated control of the production process.

[0376] Industrial production affairs such as steelmaking, for example, industrial parameters include starting element addition and process control variables. Starting element addition includes element addition and alloy addition; process control variables include temperature, time and other variables. The prediction parameters include endpoint element content, such as acid-soluble aluminum (Als), carbon (C), chromium (Cr), molybdenum (Mo), etc.; since the endpoint element content is difficult to measure and control directly, and some elements between endpoint elements will affect each other, prediction parameters are obtained based on industrial parameters.

[0377] The parameter prediction module 902 is used to input the industrial parameters into the target predictor to obtain a prediction result; wherein the prediction result includes at least two prediction parameters; the target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample prediction parameters in the sample data.

[0378] Among them, the prediction result includes at least two prediction parameters, that is, the embodiment of the present application is a multi-prediction parameter prediction process, and if there is a certain correlation between the multiple prediction parameters in the actual production application process, they will affect each other.

[0379] In order to introduce the correlation information between multiple prediction parameters, the graph convolution network is a machine learning model for mining the mutual relationship in the graph structure that has developed rapidly in recent years. In the embodiment of the present application, the target predictor is a pre-trained predictor, and the correlation information between multiple prediction parameters is introduced during the training process. The target predictor includes a graph convolution layer, and the graph convolution layer includes a graph structure obtained by training according to sample data. The nodes of the graph structure include sample prediction parameters in the sample data, and the sample prediction parameters are prediction parameters in the sample data, that is, known prediction parameter values; in the sample data, each group of data includes sample industrial parameters and sample prediction parameters; the sample industrial parameters are input into the predictor to obtain predicted prediction parameters, and then the predictor is reversely optimized according to the gap between the known sample prediction parameters and the predicted prediction parameters, and the predictor that finally meets the accuracy requirements is the target predictor.

[0380] Graph convolutional networks have been widely used in social network relationship analysis, file classification, traffic prediction and other fields due to their ability to conveniently perform operations in non-Euclidean spaces. The graph convolutional layer is a graph structure with sample prediction parameters as nodes. The graph structure is usually composed of nodes and edges, and the relationships between nodes are described by edges. Figure 1 As shown, Figure 1 Each node N in corresponds to a feature of the graph convolution layer, and the edge between adjacent nodes N represents the relationship between two adjacent nodes. It can be understood that Figure 1 Only the graph convolution layer is shown in the figure. The actual number of nodes may not be Figure 1 Through the graph convolution operation, the weighted sum of the features of the adjacent nodes of each node is used as the feature of the node; therefore, through the graph convolution operation, the features of the nodes can be extracted based on the mutual relationship described under the graph structure.

[0381] After the industrial parameters are input into the target predictor as input data, they are extracted through the graph convolution layer to obtain the output data as the prediction result; Figure 1 As shown, the output data includes n prediction parameters, each prediction parameter corresponds to a channel, and in the graph convolution layer, each channel of the prediction parameters with mutual relationship is connected, so that the graph convolution layer refers to this mutual relationship when performing feature extraction, so that the mutual relationship between multiple prediction parameters affects the prediction result.

[0382] In the prior art, graph convolutional networks usually regard samples as nodes and mine the relationships between samples instead of the relationships between non-variables. Therefore, it is impossible to effectively introduce the correlation information between multiple prediction parameters. In the embodiments of the present application, for industrial soft measurement scenarios with multiple prediction parameters, the correlation information between multiple prediction parameters is introduced to improve the prediction performance.

[0383] In addition, in the prior art graph convolutional networks, the relationship between nodes is usually calculated using distance metrics, and distance cannot accurately describe the correlation information between nodes. In the embodiment of the present application, each channel in the target predictor can independently learn different prediction parameters using industrial parameter information, and at the same time, multiple prediction parameters are respectively used as different nodes in the graph convolution layer, and the strength of the correlation between nodes is described through the edges in the graph convolution layer, accurately measuring the correlation between nodes.

[0384] Still taking steelmaking as an example, industrial parameters include alloy addition, process variables, molten steel related data, etc.; the industrial parameters are input into the target predictor, and the target predictor extracts features from the industrial parameters. Through the graph convolution layer of the target predictor, the correlation between the various prediction parameters is integrated in the prediction process to obtain the prediction parameters as the prediction results; the prediction parameters are the end-point element content, such as acid-soluble aluminum (Als). Elements such as carbon (C), chromium (Cr), molybdenum (Mo), etc., and the prediction results include the content of each end-point element; since the target predictor is a pre-trained predictor, its accuracy meets the preset accuracy requirements, so the accuracy of the end-point element content predicted by the target predictor is also high.

[0385] In an embodiment of the present application, the parameter acquisition module 901 acquires the industrial parameters of the industrial production affairs, and the parameter prediction module 902 inputs the industrial parameters into the target predictor to obtain a prediction result; the prediction result includes at least two prediction parameters, and the target predictor includes a graph convolution layer, and the graph convolution layer is a graph structure with multiple prediction parameters as nodes. The graph structure uses a graph convolution operation to connect the prediction parameters with related relationships to each other, so that different prediction parameters influence each other in the prediction process; based on the coupling relationship between the prediction parameters, the performance of the target predictor on some prediction parameters with poor prediction performance is improved, the comprehensive soft measurement performance of the multi-prediction parameter prediction is improved, and the prediction accuracy is improved.

[0386] The embodiment of the present application also provides a non-volatile readable storage medium, which stores one or more modules (programs). When the one or more modules are applied to a device, the device can execute instructions (instructions) of each method step in the embodiment of the present application.

[0387] The present application embodiment provides one or more machine-readable storage media, on which instructions are stored, and when executed by one or more processors, the electronic device executes one or more of the methods described in the above embodiments. In the present application embodiment, the electronic device includes various types of devices such as terminal devices and servers (clusters).

[0388] The embodiments of the present disclosure may be implemented as a device configured as desired using any appropriate hardware, firmware, software, or any combination thereof, and the device may include electronic devices such as terminal devices and servers (clusters). Fig.10 An exemplary apparatus 1000 that can be used to implement various embodiments described in this application is schematically illustrated.

[0389] For one embodiment, Fig.10An exemplary apparatus 1000 is shown having one or more processors 1002, a control module (chip set) 1004 coupled to at least one of the (one or more) processors 1002, a memory 1006 coupled to the control module 1004, a non-volatile memory (NVM) / storage device 1008 coupled to the control module 1004, one or more input / output devices 1010 coupled to the control module 1004, and a network interface 1012 coupled to the control module 1004.

[0390] The processor 1002 may include one or more single-core or multi-core processors, and the processor 1002 may include any combination of general-purpose processors or special-purpose processors (such as graphics processors, application processors, baseband processors, etc.). In some embodiments, the device 1000 can be used as a terminal device, server (cluster), etc. described in the embodiments of the present application.

[0391] In some embodiments, the apparatus 1000 may include one or more computer-readable media (e.g., memory 1006 or NVM / storage device 1008) having instructions 1014 and one or more processors 1002 combined with the one or more computer-readable media and configured to execute the instructions 1014 to implement a module to perform the actions described in the present disclosure.

[0392] For one embodiment, the control module 1004 may include any suitable interface controller to provide any suitable interface to at least one of the processor(s) 1002 and / or any suitable device or component in communication with the control module 1004 .

[0393] The control module 1004 may include a memory controller module to provide an interface to the memory 1006. The memory controller module may be a hardware module, a software module, and / or a firmware module.

[0394] The memory 1006 may be used, for example, to load and store data and / or instructions 1014 for the device 1000. For one embodiment, the memory 1006 may include any suitable volatile memory, such as a suitable DRAM. In some embodiments, the memory 1006 may include double data rate type four synchronous dynamic random access memory (DDR4 SDRAM).

[0395] For one embodiment, control module 1004 may include one or more input / output controllers to provide an interface to NVM / storage device 1008 and input / output device(s) 1010 .

[0396] For example, NVM / storage 1008 may be used to store data and / or instructions 1014. NVM / storage 1008 may include any suitable non-volatile memory (e.g., flash memory) and / or may include any suitable non-volatile storage device(s) (e.g., one or more hard disk drives (HDDs), one or more compact disk (CD) drives, and / or one or more digital versatile disk (DVD) drives).

[0397] NVM / storage device 1008 may include storage resources that are physically part of the device on which apparatus 1000 is installed, or it may be accessible to the device without being part of the device. For example, NVM / storage device 1008 may be accessed via input / output device(s) 1010 over a network.

[0398] (One or more) input / output devices 1010 may provide an interface for the apparatus 1000 to communicate with any other appropriate device, and the input / output device 1010 may include a communication component, an audio component, a sensor component, etc. The network interface 1012 may provide an interface for the apparatus 1000 to communicate through one or more networks, and the apparatus 1000 may wirelessly communicate with one or more components of a wireless network according to any of one or more wireless network standards and / or protocols, for example, accessing a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G, 5G, etc., or a combination thereof for wireless communication.

[0399] For one embodiment, at least one of the processor(s) 1002 may be packaged together with the logic of one or more controllers (e.g., a memory controller module) of the control module 1004. For one embodiment, at least one of the processor(s) 1002 may be packaged together with the logic of one or more controllers of the control module 1004 to form a system-in-package (SiP). For one embodiment, at least one of the processor(s) 1002 may be integrated on the same die with the logic of one or more controllers of the control module 1004. For one embodiment, at least one of the processor(s) 1002 may be integrated on the same die with the logic of one or more controllers of the control module 1004 to form a system-on-chip (SoC).

[0400] In various embodiments, the apparatus 1000 may be, but is not limited to, a terminal device such as a server, a desktop computing device, or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet computer, a netbook, etc.). In various embodiments, the apparatus 1000 may have more or fewer components and / or a different architecture. For example, in some embodiments, the apparatus 1000 includes one or more cameras, a keyboard, a liquid crystal display (LCD) screen (including a touch screen display), a non-volatile memory port, multiple antennas, a graphics chip, an application-specific integrated circuit (ASIC), and a speaker.

[0401] Among them, the main control chip can be used as a processor or control module in the detection device, sensor data, location information, etc. are stored in a memory or NVM / storage device, the sensor group can be used as an input / output device, and the communication interface may include a network interface.

[0402] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0403] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0404] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0405] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0406] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0407] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present application.

[0408] Finally, 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 terminal 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 terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0409] The data processing method and device, an electronic device and a storage medium provided by the present application are introduced in detail above. The principles and implementation methods of the present application are explained in this article using specific examples. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A data processing method, characterized in that: The method comprises: Acquire the data to be predicted of the target transaction; the target transaction includes a production transaction or a parameter prediction transaction executed according to the data to be predicted; the data to be predicted is an independent variable of the prediction process; the independent variable includes the amount of the starting element added, or the independent variable includes the amount of the starting element added and the process control variable; The data to be predicted is input into a target predictor to obtain a prediction result; wherein the prediction result includes at least two dependent variables; the data to be predicted includes an independent variable of the dependent variable; the target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample dependent variables in the sample data; the graph structure uses a graph convolution operation to connect the dependent variables with correlation to each other; in the target predictor, each dependent variable corresponds to a channel.

2. The data processing method according to claim 1, characterized in that: The target predictor also includes a target fully connected layer, and the target fully connected layer includes a first fully connected layer, a second fully connected layer, and a third fully connected layer; Among them, the first fully connected layer, the graph convolutional layer, the second fully connected layer and the third fully connected layer are connected in sequence.

3. The data processing method according to claim 2, characterized in that: Each of the target fully connected layers includes N channels, where N is the number of the dependent variables, and the channel processes one of the dependent variables.

4. The data processing method according to claim 2, characterized in that: In the graph structure, the characteristic value of each node is the result of weighted summation of adjacent nodes; The weight in the graph structure is the weight between each two adjacent nodes; The weights are elements in an adjacency matrix, and the adjacency matrix is ​​generated according to the sample dependent variable.

5. The data processing method according to claim 1, characterized in that: Before the step of obtaining the target transaction's data to be predicted, the method includes: Acquire the sample data; wherein the sample data includes a training sample and a test sample; According to the training samples, an initial predictor is trained to obtain a predictor to be tested; The predictor to be tested is tested according to the test sample to obtain the target predictor.

6. The data processing method according to claim 5, characterized in that: The sample in the sample data includes a sample independent variable and the sample dependent variable; The step of training the initial predictor according to the training sample to obtain the predictor to be tested comprises: Inputting a first sample independent variable of a first training sample in the training samples into an initial predictor to obtain a first prediction content; According to the first prediction content and the first sample dependent variable of the first training sample, reversely optimize the initial predictor to obtain an optimized predictor; The second sample independent variable in the training sample is iteratively input into the optimized predictor, and the reverse optimization is performed until the number of iterations meets the first preset requirement and / or the loss function meets the second preset requirement, so as to obtain the predictor to be tested.

7. The data processing method according to claim 6, characterized in that: The initial predictor includes an initial fully connected layer and an initial graph convolutional layer; each of the initial fully connected layers includes N channels; The initial fully connected layer includes a fourth fully connected layer, a fifth fully connected layer, and a sixth fully connected layer.

8. The data processing method according to claim 7, characterized in that: The step of inputting the first sample independent variable of the first training sample in the training samples into the initial predictor to obtain the first prediction content includes: Inputting a first sample independent variable of a first training sample among the training samples into an initial predictor; At an initial fully connected layer of each of the initial predictors, output data is calculated according to the first data relationship; and at an initial graph convolution layer of the initial predictor, output data is calculated according to the second data relationship; The output data of the sixth fully connected layer is obtained as the first prediction content.

9. The data processing method according to claim 8, characterized in that: The first data relationship is: H1 (i) =σ1[H1 (i-1) W1 (i) +b (i) ] Among them, H1 (i) Represents the output data of the i-th layer; H1 (i-1) Represents the output data of the i-1th layer; W1 (i) represents the first weight of the i-th layer, b (i) represents the bias parameter of the i-th layer; σ1(x) represents the first activation function of x.

10. The data processing method according to claim 8, characterized in that: The step of calculating output data according to the second data relationship at the initial graph convolution layer of the initial predictor comprises: In the graph convolution layer, an adjacency matrix of the first sample dependent variable of the first training sample is generated; Calculate output data according to the adjacency matrix and the second data relationship; The second data relationship is: Among them, H2 (i) Represents the output data of the i-th layer; H2 (i-1) Represents the output data of the i-1th layer; W (i) represents the weight of the i-th layer; σ2(x) represents the second activation function of x; A represents the adjacency matrix, and D represents the degree matrix of A; 11. The data processing method according to claim 10, characterized in that: The step of generating an adjacency matrix of the first sample dependent variable of the first training sample comprises: Generate an adjacency matrix of the first sample dependent variable according to the first sample dependent variable and the third data relationship; The third data relationship is: s.t.A=A T A ii = 0, A ij ≥0, among them, i, j=1, 2, ......, q Wherein, q represents the number of nodes in the graph convolutional layer; A represents the adjacency matrix; Y train represents the first sample dependent variable; Indicates (Y train -Y train A) 2 The second norm of ; ||A||1 represents the first norm of A; λ is the preset square coefficient.

12. The data processing method according to claim 6, characterized in that: The step of reversely optimizing the initial predictor according to the first prediction content and the first sample dependent variable of the first training sample comprises: The loss value of the optimized predictor is calculated according to the first prediction content, the first sample dependent variable of the first training sample, and a fourth data relationship, wherein the fourth data relationship is: in, represents the loss value of the optimized predictor, Y1 represents the first sample dependent variable, represents the first prediction content, p represents the number of the first sample dependent variables, m represents the number of the first training samples; γ represents the preset balance coefficient; W3 represents the regularization term of the graph convolution layer of the optimized predictor.

13. The data processing method according to claim 5, characterized in that: The step of testing the predictor to be tested according to the test sample to obtain the target predictor comprises: Inputting the second sample independent variable of each of the test samples into the predictor to be tested to obtain a second prediction content; Determining the regression parameters of the predictor to be tested according to the second sample dependent variable of the test sample and the second prediction content; If the regression parameter meets the third preset requirement, the predictor to be tested is determined to be the target predictor.

14. The data processing method according to claim 13, characterized in that: The step of determining the regression parameter of the predictor to be tested according to the second sample dependent variable of the test sample and the second prediction content comprises: The first regression parameter of the predictor to be tested is determined according to the second sample dependent variable of the test sample, the second prediction content and the fifth data relationship, wherein the fifth data relationship is: Among them, R 2 represents the first regression parameter, n represents the number of the test samples, Y2 represents the second sample dependent variable, represents the second predicted content, represents the mean of the dependent variable of the second sample; And determining the second regression parameter of the predictor to be tested according to the second sample dependent variable of the test sample, the second prediction content and the sixth data relationship, the sixth data relationship being: Among them, RMSE represents the second regression parameter.

15. The data processing method according to claim 5, characterized in that: Before the step of training the initial predictor according to the training samples, the method further comprises: The sample data is normalized.

16. A data processing method, characterized in that: The method comprises: Obtaining independent variable data of the target production transaction; the independent variable data includes the starting element addition amount, or the independent variable data includes the starting element addition amount and the process control variable; The independent variable data is input into a target predictor to obtain a prediction result; wherein the prediction result includes at least two dependent variables; the dependent variable includes the end point element content; the target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample dependent variables in the sample data; in the graph structure, the dependent variables with correlation are connected to each other using a graph convolution operation; in the target predictor, each dependent variable corresponds to a channel.

17. The data processing method according to claim 16, characterized in that: After the step of obtaining the prediction result, the method comprises: The independent variable data is adjusted according to the endpoint element content of the prediction result.

18. A data processing method, characterized in that: The method comprises: Acquiring industrial parameters of industrial production affairs; the industrial parameters include the amount of starting element added, or the industrial parameters include the amount of starting element added and process control variables; The industrial parameters are input into a target predictor to obtain a prediction result; wherein the prediction result includes at least two prediction parameters; the target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample prediction parameters in the sample data; in the graph structure, graph convolution operations are used to connect dependent variables with correlation to each other; in the target predictor, each dependent variable corresponds to a channel.

19. A data processing device, characterized in that: The device comprises: A data acquisition module is used to acquire the data to be predicted of the target transaction; the target transaction includes a production transaction or a parameter prediction transaction executed according to the data to be predicted; the data to be predicted is an independent variable of the prediction process; the independent variable includes the amount of the starting element added, or the independent variable includes the amount of the starting element added and the process control variable; A data prediction module, used for inputting the data to be predicted into a target predictor to obtain a prediction result; wherein the prediction result includes at least two dependent variables; and the data to be predicted includes an independent variable of the dependent variable; The target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample dependent variables in the sample data; the dependent variables with correlation are connected to each other using a graph convolution operation in the graph structure; in the target predictor, each dependent variable corresponds to a channel.

20. A data processing device, characterized in that: The device comprises: An independent variable data acquisition module, used to acquire independent variable data of a target production transaction; the independent variable data includes the amount of the starting element added, or the independent variable data includes the amount of the starting element added and a process control variable; An independent variable prediction module is used to input the independent variable data into a target predictor to obtain a prediction result; wherein the prediction result includes at least two dependent variables; the dependent variable includes the end point element content; The target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample dependent variables in the sample data; the dependent variables with correlation are connected to each other using a graph convolution operation in the graph structure; in the target predictor, each dependent variable corresponds to a channel.

21. A data processing device, characterized in that: The device comprises: A parameter acquisition module, used to acquire industrial parameters of industrial production affairs; the industrial parameters include the amount of starting element added, or the industrial parameters include the amount of starting element added and process control variables; A parameter prediction module is used to input the industrial parameter into a target predictor to obtain a prediction result; wherein the prediction result includes at least two prediction parameters; the target predictor includes a graph convolution layer, the graph convolution layer includes a graph structure trained according to sample data, and the nodes of the graph structure include sample prediction parameters in the sample data; in the target predictor, each dependent variable corresponds to a channel.

22. An electronic device, characterized in that: include: processor; and A memory having executable code stored thereon, which, when executed, causes the processor to perform the method as claimed in one or more of claims 1 to 18.

23. One or more machine-readable storage media having executable code stored thereon, characterized in that: When the executable code is executed, the processor is caused to perform the method as claimed in one or more of claims 1 to 18.

Images