Converter molten iron judgment method and device, storage medium and computer equipment

Abstract

The invention relates to the technical field of ferrous metallurgy, and discloses a converter molten iron judgment method and device, a storage medium and computer equipment.The converter molten iron judgment method comprises the steps that firstly, original molten iron data are obtained and preprocessed, an original data set is constructed, and then molten iron category labels are added to part of data in the original data set according to a preset molten iron category division rule; the method comprises the steps of obtaining a target data set, constructing a graph convolutional network based on the target data set, training and testing the graph convolutional network by using the target data set, generating a graph convolutional neural network model, and finally judging the collected to-be-detected molten iron data by using the graph convolutional neural network model to obtain a molten iron judgment result. According to the method, the type of the molten iron can be judged before smelting is started, then pre-judgment and early warning of dangerous accidents possibly generated in actual production are achieved, the casualty rate caused by splashing and dry returning accidents in the actual production can be reduced, corresponding economic losses are reduced, and production safety is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of iron and steel metallurgy, and in particular to a method, a device, a storage medium and a computer device for determining molten iron in a converter. Background Art

[0002] In the traditional converter steelmaking process, splashing and back drying seriously restrict production efficiency. Once it occurs, it will not only lead to a large loss of steel materials, affect gas recovery efficiency, shorten the service life of equipment, cause significant economic losses, but also pose a serious threat to the safety of personnel. According to statistics, splashing can cause metal material losses of up to 10%, and 80% of furnace scalding accidents are related to splashing.

[0003] Considering that the production process of converter steelmaking has obvious "black box" characteristics, that is, it is difficult to directly observe and control the production process, the existing technology also lacks a direct and effective prediction method for splashing and back-drying phenomena, and considering that on-site operations mainly rely on the operator's experience, reaction speed and emergency response capabilities, the potential safety hazards are further increased.

[0004] The published patent CN202410355686.X provides a method for identifying converter splashing using transfer learning and a residual network with a dual attention mechanism. Specifically, an industrial camera is used to collect images and a residual network model with transfer learning and a dual attention mechanism is used to determine whether splashing occurs during the smelting process. However, this technical solution can only provide real-time warnings and is unable to effectively determine and predict the key factors of splashing and drying back from the source, thereby failing to achieve a comprehensive prediction of the entire furnace process, which results in the inability of on-site personnel to effectively respond to the occurrence of risks in a short period of time. Summary of the invention

[0005] In view of this, the present application provides a method, device, storage medium and computer equipment for determining molten iron in a converter, the main purpose of which is to solve the technical problem in the prior art that it is impossible to effectively determine the splashing and drying of molten iron from the source, and thus it is impossible to achieve a comprehensive prediction of the entire furnace process.

[0006] According to a first aspect of the present invention, there is provided a method for determining molten iron in a converter, comprising: Acquire original molten iron data, preprocess the original molten iron data, and construct an original data set based on the preprocessed original molten iron data; Adding molten iron category labels to part of the data in the original data set according to a preset molten iron category classification rule to obtain a target data set; Building a graph convolutional network based on the target data set, and using the target data set to train and test the graph convolutional network to generate a graph convolutional neural network model; The molten iron data to be tested is collected, and the molten iron data to be tested is judged using the graph convolutional neural network model to obtain a molten iron judgment result.

[0007] Optionally, the obtaining of raw molten iron data and preprocessing the raw molten iron data includes: Obtaining original molten iron data, and calculating the mean of the original molten iron data and the standard deviation of each of the original molten iron data; Calculate a standard score corresponding to each piece of the original molten iron data based on the mean value of the original molten iron data and the standard deviation of each piece of the original molten iron data; Determining the standard score according to a preset standard score threshold; When the standard score of the original molten iron data does not exceed the standard score threshold, marking the original molten iron data as normal data; When the standard score of the original molten iron data exceeds the standard score threshold, the original molten iron data is marked as abnormal data, and the abnormal data is replaced with the mean value of the original molten iron data.

[0008] Optionally, adding molten iron category labels to part of the data in the original data set according to a preset molten iron category classification rule includes: Randomly extracting some data points from the original data set, marking the extracted data points as target data points, and obtaining the molten iron composition of each target data point; Based on the molten iron composition of the target data point, the molten iron category of the target data point is determined using a preset molten iron category classification rule, and a molten iron category label is added to the target data point according to the molten iron category.

[0009] Optionally, constructing a graph convolutional network based on the target data set includes: Dividing the target data set into a training data set and a test data set, wherein all data points in the training data set carry the molten iron category label; Defining a node feature matrix and setting an adjacency matrix based on a threshold connection method, and defining an input layer based on the node feature matrix and the adjacency matrix; Defining multiple graph convolutional layers and selecting activation functions, wherein the graph convolutional layers include a weight matrix; Standardizing the adjacency matrix, determining graph convolution propagation features based on the standardized adjacency matrix, the node feature matrix, the activation function, and the weight matrix, and defining a hidden layer according to the convolution propagation features and the plurality of graph convolution layers; Defining a loss function and an optimizer, constructing an initial graph convolutional network based on the input layer, the hidden layer, the loss function and the optimizer, and training the initial graph convolutional network using the training data set to update the parameters of the initial graph convolutional network, wherein the loss function is a cross entropy loss function; The test data set is used to evaluate the accuracy of the initial graph convolutional network after updating the parameters, and the initial graph convolutional network that passes the accuracy evaluation is selected as the final generated graph convolutional network.

[0010] Optionally, the using the target data set to train and test the graph convolutional network to generate a graph convolutional neural network model includes: The graph convolutional network is globally trained using the training data set, and the trained graph convolutional network is tested using the test data set, and a graph convolutional neural network model is generated based on the graph convolutional network that passes the test.

[0011] Optionally, the globally training the graph convolutional network using the training data set includes: Based on a preset number of iterations, the graph convolutional network is globally trained using the training data set, the network weights of the graph convolutional network are updated using the optimizer during the global training process, and the loss function in the graph convolutional network is L2 regularized to minimize the value of the loss function, wherein the global training includes forward propagation and backward propagation; After the graph convolutional network completes global training, the network weights of the graph convolutional network are saved based on a preset format.

[0012] Optionally, the adjacency matrix is ​​set based on the threshold connection method, including: Determine a plurality of initial sample nodes, and calculate the Euclidean distance between each initial sample node and the remaining initial sample nodes; For each initial sample node, according to the order of the multiple Euclidean distances from small to large, a preset number of Euclidean distances are selected as target Euclidean distances, and a target initial sample node corresponding to the target Euclidean distance is determined; An adjacency matrix is ​​established based on each initial sample node and the corresponding target initial sample node.

[0013] According to a second aspect of the present invention, there is provided a device for determining molten iron in a converter, the device comprising: A data acquisition module, used to acquire original molten iron data, preprocess the original molten iron data, and construct an original data set based on the preprocessed original molten iron data; A data processing module, used to add molten iron category labels to part of the data in the original data set according to a preset molten iron category classification rule to obtain a target data set; A model building module, used to build a graph convolutional network based on the target data set, and use the target data set to train and test the graph convolutional network to generate a graph convolutional neural network model; The result generation module is used to collect the molten iron data to be tested, and use the graph convolutional neural network model to judge the molten iron data to be tested to obtain the molten iron judgment result.

[0014] According to a third aspect of the present invention, there is provided a storage medium having a computer program stored thereon, which implements the above-mentioned converter molten iron determination method when the program is executed by a processor.

[0015] According to a fourth aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the above-mentioned converter molten iron determination method is implemented when the processor executes the program.

[0016] The converter molten iron determination method, device, storage medium and computer equipment provided by the present invention first obtain the original molten iron data, and pre-process the original molten iron data, construct an original data set based on the pre-processed original molten iron data, and then add molten iron category labels to part of the data in the original data set according to the preset molten iron category classification rules to obtain a target data set, and then construct a graph convolution network based on the target data set, and use the target data set to train and test the graph convolution network to generate a graph convolution neural network model, and finally collect the molten iron data to be tested, and use the graph convolution neural network model to determine the molten iron data to be tested, and obtain the molten iron determination result. The above method collects the original molten iron data in the early stage, and marks part of the data according to the preset molten iron category classification rules, and establishes a graph convolution neural network model to classify all data. The determination of the molten iron category can be realized before the smelting has not yet begun, and then the prediction and early warning of dangerous accidents that may occur in actual production can be realized, which helps to reduce the casualties caused by splashing and back-drying accidents in actual production, and reduce the corresponding economic losses, and improve the safety of production.

[0017] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 A schematic flow chart of a method for determining molten iron in a converter provided by an embodiment of the present invention is shown; Figure 2 A schematic diagram of a flow chart of generating a graph convolutional neural network model in a method for determining molten iron in a converter provided by an embodiment of the present invention is shown; Figure 3 A bar chart showing the frequency of occurrence of different molten iron categories in a method for determining molten iron in a converter provided by an embodiment of the present invention is shown; Figure 4 A dotted line graph of a cross entropy loss function in a method for determining molten iron in a converter provided by an embodiment of the present invention is shown; Figure 5 A dot-line graph showing the accuracy of a method for determining molten iron in a converter provided by an embodiment of the present invention is shown; Figure 6 A schematic structural diagram of a converter molten iron determination device provided by an embodiment of the present invention is shown; Figure 7 A schematic diagram of the device structure of a computer device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0019] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0020] In one embodiment, Figure 1 As shown, a method for determining molten iron in a converter is provided, comprising the following steps: 101. Obtain original molten iron data, preprocess the original molten iron data, and construct an original data set based on the preprocessed original molten iron data.

[0021] Specifically, the application scenarios of the technical solution of the present application are mainly concentrated in the converter smelting process in steel production, wherein the converter is an important equipment in steel production, which is used to remove impurities in molten iron and adjust the chemical composition of the molten iron to meet the required specifications. The content of elements such as carbon, silicon, manganese, phosphorus and sulfur in the molten iron not only has an important influence on the quality of the final product, but also determines whether splashing and drying will occur during the converter steelmaking process. Therefore, precise control of the molten iron composition is very critical.

[0022] In this embodiment, an OPC data acquisition program (OLE for Process Control) is pre-established, and raw molten iron data is collected in real time from the MES system (Manufacturing Execution System), the raw molten iron data including the content of various elements, collection time, furnace number and other information; then, a converter smelting production report is obtained from the factory's production management system, the report usually including detailed production records, and the collected raw molten iron data is compared with the data in the production report to ensure that the two are consistent. If a difference is found, it is necessary to further investigate the cause and correct the data to ensure the accuracy of the collected raw molten iron data and the data in the report; finally, a SQL Server database table is created to store the molten iron composition data, and the table structure includes fields such as furnace number, content of each element, collection time, and then the real-time collected raw molten iron data is inserted into the SQL Server database table to ensure that the data is correctly stored.

[0023] Among them, the OPC data acquisition program is written in C# and interacts with the on-site industrial automation software.

[0024] In an optional embodiment, step 101 can be implemented in the following manner: obtain original molten iron data, calculate the mean of the original molten iron data and the standard deviation of each original molten iron data; calculate the standard score corresponding to each original molten iron data based on the mean of the original molten iron data and the standard deviation of each original molten iron data; determine the standard score according to a preset standard score threshold; when the standard score of the original molten iron data does not exceed the standard score threshold, mark the original molten iron data as normal data; when the standard score of the original molten iron data exceeds the standard score threshold, mark the original molten iron data as abnormal data, and replace the abnormal data with the mean of the original molten iron data.

[0025] In this embodiment, the Z-score (standard score) method is used to process the outliers in the original molten iron data. First, the mean of all the original molten iron data is calculated. , and the standard deviation of each raw molten iron data , and then calculate the Z-score of each original molten iron data, the formula is as follows:

[0026] Among them, X is a single original molten iron data; is the mean of all original molten iron data; is the standard deviation of the single original molten iron data; Then, a preset standard score threshold is obtained, specifically ±3, and the original molten iron data that does not exceed the standard score threshold is marked as normal data, and the original molten iron data that exceeds the standard score threshold is marked as abnormal data, and the abnormal data is replaced with the mean of the original molten iron data; after completing the replacement of the abnormal data in the original molten iron data, all data are deduplicated, all duplicate entries are cleaned up, and each record is ensured to be unique. Finally, the original data set is jointly constructed based on the normal data and the mean of the original molten iron data with which the abnormal data has been replaced.

[0027] 102. Add molten iron category labels to part of the data in the original data set according to the preset molten iron category classification rule to obtain the target data set.

[0028] Among them, the preset molten iron classification rules can determine whether the molten iron is prone to splashing or drying up based on the molten iron data in the original molten iron data.

[0029] In this embodiment, the original data set is generated by preprocessing the original molten iron data, and then molten iron category labels are added to part of the data in the original data set according to the preset molten iron category classification rules to identify molten iron that is prone to splashing or drying in advance, thereby effectively improving the data quality in the data set. It also helps to enhance the generalization ability of the model, optimize feature extraction, improve model training and improve the interpretability of the model in the subsequent process, and ultimately improve the accuracy of the judgment of converter molten iron.

[0030] 103. Build a graph convolutional network based on the target dataset, and use the target dataset to train and test the graph convolutional network to generate a graph convolutional neural network model.

[0031] Specifically, Graph Convolution Networks (GCNs) methods are divided into two categories, spectral domain-based methods and spatial domain-based methods; spectral domain-based methods define graph convolution by introducing filters from the perspective of graph signal processing, where the graph convolution operation is interpreted as removing noise from the graph signal; while spatial domain-based methods represent graph convolution as aggregating feature information from neighbors.

[0032] In this embodiment, the graph convolutional network can capture the complex relationship between nodes, accurately understand the interaction between the components of molten iron, and can effectively capture global information rather than just local information through multi-layer convolution operations; and the graph convolutional network uses the information of neighboring nodes to update the node representation, which helps to improve the classification accuracy, especially when the node features are insufficient, and by aggregating neighbor information, the graph convolutional network has a certain robustness to noise and outliers. In summary, by constructing a graph convolutional network based on the target data set and using the target data set for training and testing, an efficient graph convolutional neural network model can be generated, which can effectively capture the complex relationship between nodes, improve classification accuracy, and provide good interpretability and flexibility.

[0033] 104. Collect the molten iron data to be tested, use the graph convolutional neural network model to judge the molten iron data to be tested, and obtain the molten iron judgment result.

[0034] In this embodiment, after completing the construction of the graph convolutional neural network model, molten iron determination can be performed. The preliminary data collection process can refer to the method of step 101, and the OPC data collection program is used to collect the molten iron data to be tested from the MES system in real time, and the molten iron data to be tested is preprocessed to remove noise and outliers, and then the pre-trained graph convolutional neural network model is used to classify the molten iron data to be tested. The graph convolutional neural network model can capture the complex relationship between nodes and improve the classification accuracy. The final determination result of the molten iron is easy to splash molten iron, easy to splash molten iron and it is recommended that the molten steel is over-refined, easy to return to dry molten iron, easy to return to dry molten iron and it is recommended that the molten steel is over-refined, normal molten iron and normal molten iron and it is recommended that the molten steel is over-refined; after obtaining the determination result of the molten iron, the determination result of the molten iron can be displayed on the secondary operation interface of the steel plant. The secondary operation interface can be a graphical user interface or other forms of visual interface. If the determination result is abnormal, an alarm prompt can be given on the interface to remind the operator to take corresponding measures.

[0035] Among them, due to the problem of timeliness of information transmission by the MES system, it is often necessary to judge the molten iron data in advance before the converter enters the smelting stage. Therefore, this application designs a judgment time. In the judgment process, the moment when the oxygen gun position is higher than the oxygen opening point and the oxygen flow rate is zero is selected as the trigger condition for the judgment. The molten iron data in the MES system is judged once every 2 seconds, and the data and judgment results are updated in real time, and the judgment results are associated with the specific furnace by the furnace number.

[0036] The method for determining molten iron in a converter provided in this embodiment first obtains the original molten iron data, and pre-processes the original molten iron data, constructs an original data set based on the pre-processed original molten iron data, and then adds molten iron category labels to part of the data in the original data set according to the preset molten iron category classification rules to obtain a target data set, and then constructs a graph convolution network based on the target data set, and uses the target data set to train and test the graph convolution network to generate a graph convolution neural network model, and finally collects the molten iron data to be tested, and uses the graph convolution neural network model to determine the molten iron data to be tested, and obtains the molten iron determination result. The above method collects the original molten iron data in the early stage, and labels part of the data according to the preset molten iron category classification rules, and classifies all the data by establishing a graph convolution neural network model, so that the determination of the molten iron category can be realized before the smelting begins, and then the prediction and early warning of dangerous accidents that may occur in actual production can be realized, which helps to reduce the casualties caused by splashing and back-drying accidents in actual production, and reduce the corresponding economic losses, and improve the safety of production.

[0037] Furthermore, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the implementation process of this embodiment, a method for generating a graph convolutional neural network model is provided, such as Figure 2 As shown, the method comprises the following steps: 201. Add molten iron category labels to part of the data in the original data set according to the preset molten iron category classification rules.

[0038] Specifically, some data points are randomly extracted from the original data set, and the extracted data points are marked as target data points to obtain the molten iron composition of each target data point; and based on the molten iron composition of the target data point, the molten iron category of the target data point is determined using a preset molten iron category classification rule, and a molten iron category label is added to the target data point according to the molten iron category.

[0039] In this embodiment, if Figure 3As shown in the figure, the classification of molten iron is mainly based on the characteristics of molten iron and treatment suggestions. There are six specific categories. Each category reflects the different properties of molten iron and some specific suggestions in the subsequent treatment process. They are: Category 1 molten iron: Highly splashing molten iron (HS-ESP). This type of molten iron is prone to splashing during the smelting process. During the smelting process, it is necessary to avoid safety problems and reduced production efficiency caused by splashing; Category 2 molten iron: Highly splashing molten iron and recommended over-refining of steel (HS-ESP-RF). This type of molten iron is not only prone to splashing, but also requires additional treatment in the subsequent refining process. In addition to paying attention to the splashing problem, the molten steel also needs to be over-refined to ensure the quality of the final product; Category 3 molten iron: Highly return-drying molten iron (HS-ERD). This type of molten iron is prone to The phenomenon of drying out occurs, which leads to operational difficulties. Measures need to be taken to prevent drying out, such as adjusting the charge ratio and controlling the temperature. The fourth category of molten iron: molten iron that is easy to dry out and it is recommended to over-refine the molten steel (HS-ERD-RF). This type of molten iron is not only easy to dry out, but also requires additional treatment in the subsequent refining process, that is, the molten steel is over-refined to ensure the quality of the final product. The fifth category of molten iron: normal molten iron (HS-NOR). This type of molten iron has no obvious special properties and belongs to conventional molten iron. It can be processed according to conventional smelting and refining processes. The sixth category of molten iron: normal molten iron and it is recommended to over-refine the molten steel (HS-NOR-RF). Although this type of molten iron belongs to normal molten iron, it needs additional treatment in the subsequent refining process, that is, the molten steel is over-refined to improve product quality.

[0040] Among them, whether the molten iron is prone to splashing or drying can be judged by the accumulated experience of personnel, or it can be achieved through some quantitative indicators, mainly including the chemical composition of the molten iron, as well as temperature, slag properties and other factors.

[0041] Among them, the selection of part of the data in the original data set can be carried out in the following way, that is, all the data in the original data set are randomly divided into six clusters, corresponding to six molten iron modes, and sampling is performed at fixed intervals in each cluster, and then the extracted data are labeled.

[0042] 202. Standardize the data to obtain a target data set.

[0043] In this embodiment, after adding molten iron category labels to part of the data in the original data set, the data still needs to be standardized, which can effectively improve the efficiency of subsequent model training, accelerate convergence, avoid numerical problems, improve model performance, enhance generalization ability, and reduce the risk of overfitting.

[0044] 203. Construct an initial graph convolutional network, train the initial graph convolutional network, and evaluate the generated graph convolutional network.

[0045] In an optional embodiment, step 203 can be implemented in the following manner: divide the target data set into a training data set and a test data set, wherein the data points in the training data set all carry molten iron category labels; define a node feature matrix and set the adjacency matrix based on a threshold connection method, and define an input layer based on the node feature matrix and the adjacency matrix; define multiple graph convolution layers and select an activation function, wherein the graph convolution layer includes a weight matrix; standardize the adjacency matrix, determine the graph convolution propagation characteristics based on the standardized adjacency matrix, node feature matrix, activation function and weight matrix, and define a hidden layer according to the convolution propagation characteristics and multiple graph convolution layers; define a loss function and an optimizer, construct an initial graph convolution network based on the input layer, hidden layer, loss function and optimizer, and train the initial graph convolution network using the training data set to update the parameters of the initial graph convolution network, wherein the loss function is a cross entropy loss function; use the test data set to evaluate the accuracy of the initial graph convolution network after updating the parameters, and select the initial graph convolution network that passes the accuracy evaluation as the final generated graph convolution network.

[0046] In this embodiment, a graph convolutional network is used to cluster graph structure data. The core of the network is that it can perform convolution operations through the adjacency relationship of the graph and update the node feature representation layer by layer. The specific operation steps include: first, defining an input layer, the input layer includes a node feature matrix and an adjacency matrix, wherein the node feature is defined, and its matrix shape is (N, D), wherein N is the number of nodes, and D is the feature dimension. Based on the original molten iron data, N=2000 and D=7 are selected, and then the shape of the adjacency matrix A is set to (2, 17504); after completing the definition of the input layer, a hidden layer is defined, and a graph convolution layer is defined, wherein the number of neurons is 32, and then the features of neighbor nodes are aggregated through a convolution operation. Each layer of the graph convolution layer is processed by H (l+1) = (H (l) W (l) ), where is the standardized adjacency matrix; H(l) is the node feature of the lth layer; W(l) is the learnable weight matrix; σ is the activation function, and ReLU (Rectified Linear Unit) is selected here, which can be specifically expressed as ReLU(x) = max(0, x). Compared with Sigmoid and Tanh activation functions, ReLU activation function does not have the gradient vanishing problem during gradient calculation, so it can effectively accelerate neural network training. As long as the input of ReLU is greater than 0, it can directly output the input value, and when the input is less than 0, the output is 0, which can be specifically expressed as:

[0047] Then define the loss function and optimizer. Specifically, use the cross entropy loss function to classify the data points with molten iron category labels in the training data set. The cross entropy loss function point line graph is as follows: Figure 4 As shown, the loss function is as follows:

[0048] Among them, y i is the true label of node i, is the predicted output.

[0049] Then, the test data set is used to evaluate the accuracy, so as to evaluate the accuracy of the initial graph convolutional network after updating the parameters and perform model evaluation, such as Figure 5 The accuracy is shown as a dotted line graph, and the accuracy is calculated as follows:

[0050]

[0051] in, is the predicted value of the i-th sample; y i is the true label of the i-th sample; is an indicator function, which is 1 when the predicted value is equal to the true label, otherwise it is 0; i∈train_mask is that only samples with labeled labels are considered; correct is the number of samples predicted correctly; |train_mask| is the number of labeled samples in the training data set.

[0052] In an optional embodiment, step 203 can also be implemented in the following manner: determine multiple initial sample nodes, calculate the Euclidean distance between each initial sample node and the remaining initial sample nodes; for each initial sample node, select a preset number of Euclidean distances as the target Euclidean distance in the order of multiple Euclidean distances from small to large, and determine the target initial sample node corresponding to the target Euclidean distance; establish an adjacency matrix based on each initial sample node and the corresponding target initial sample node.

[0053] In this embodiment, the adjacency matrix is ​​designed by selecting the threshold connection method in the adjacency matrix initialization. By calculating the Euclidean distance, the 10 smallest samples for each sample are selected to form the adjacency matrix. The Euclidean distance formula is as follows:

[0054] Among them, P and Q are two points in n-dimensional space.

[0055] 204. Use the target dataset to train and test the graph convolutional network to generate a graph convolutional neural network model.

[0056] In an optional embodiment, step 204 can be implemented in the following manner: based on a preset number of iterations, the graph convolution network is globally trained using a training data set, the network weights of the graph convolution network are updated using an optimizer during the global training process, and the loss function in the graph convolution network is L2 regularized to minimize the value of the loss function, wherein the global training includes forward propagation and backward propagation; after the graph convolution network completes the global training, the network weights of the graph convolution network are saved based on a preset format.

[0057] In this embodiment, the training data set is first used for global training, and the Adam optimizer is used to continuously update the network. At the same time, L2 regularization is performed to prevent overfitting. The parameters are set to minimize the value of the loss function. Specifically, in the global training process, the number of iterations epoch is set, specifically epoch=200, and the regularization parameter is set to 10 -3 , then save the network weights, select the .pt format to save the network weights, and finally test it on the test data set to obtain the graph convolutional neural network model.

[0058] Specifically, the calculation formula of L2 regularization is as follows:

[0059] Where L is the loss function of the network training on the original pre-processed molten iron, and w i is the network weight, λ is the regularization coefficient, is the sum of the squares of all weights.

[0060] The graph convolutional neural network model generation method provided in this embodiment can effectively capture the local dependencies between nodes and the global structural characteristics by defining convolution operations on the graph structure, and can process graph data of different types and scales, and has strong semi-supervised learning capabilities.

[0061] Further, as Figures 1 to 5 The specific implementation of the method shown in the embodiment provides a converter molten iron determination device, such as Figure 6 As shown, the device includes: a data acquisition module 301, a data processing module 302, a model building module 303 and a result generation module 304, wherein: The data acquisition module 301 is used to acquire original molten iron data, pre-process the original molten iron data, and construct an original data set based on the pre-processed original molten iron data; The data processing module 302 is used to add molten iron category labels to part of the data in the original data set according to the preset molten iron category classification rules to obtain a target data set; A model building module 303 is used to build a graph convolutional network based on a target data set, and train and test the graph convolutional network using the target data set to generate a graph convolutional neural network model; The result generation module 304 is used to collect the molten iron data to be tested, and use the graph convolutional neural network model to judge the molten iron data to be tested to obtain the molten iron judgment result.

[0062] In a specific application scenario, the data acquisition module 301 can be specifically used to acquire original molten iron data, calculate the mean of the original molten iron data and the standard deviation of each original molten iron data; calculate the standard score corresponding to each original molten iron data based on the mean of the original molten iron data and the standard deviation of each original molten iron data; determine the standard score according to a preset standard score threshold; when the standard score of the original molten iron data does not exceed the standard score threshold, mark the original molten iron data as normal data; when the standard score of the original molten iron data exceeds the standard score threshold, mark the original molten iron data as abnormal data, and replace the abnormal data with the mean of the original molten iron data.

[0063] In a specific application scenario, the data processing module 302 can be used to randomly extract some data points from the original data set, mark the extracted data points as target data points, and obtain the molten iron composition of each target data point; based on the molten iron composition of the target data point, the molten iron category of the target data point is determined using a preset molten iron category classification rule, and a molten iron category label is added to the target data point according to the molten iron category.

[0064] In a specific application scenario, the model construction module 303 can be specifically used to divide the target data set into a training data set and a test data set, wherein the data points in the training data set all carry molten iron category labels; define a node feature matrix and set an adjacency matrix based on a threshold connection method, and define an input layer based on the node feature matrix and the adjacency matrix; define multiple graph convolution layers and select an activation function, wherein the graph convolution layer includes a weight matrix; standardize the adjacency matrix, determine the graph convolution propagation characteristics based on the standardized adjacency matrix, node feature matrix, activation function and weight matrix, and define a hidden layer according to the convolution propagation characteristics and multiple graph convolution layers; define a loss function and an optimizer, construct an initial graph convolution network based on the input layer, hidden layer, loss function and optimizer, and train the initial graph convolution network using the training data set to update the parameters of the initial graph convolution network, wherein the loss function is a cross entropy loss function; use the test data set to evaluate the accuracy of the initial graph convolution network after updating the parameters, and select the initial graph convolution network that passes the accuracy evaluation as the final generated graph convolution network.

[0065] In a specific application scenario, the model building module 303 can be used to perform global training on the graph convolutional network using a training data set, and to test the trained graph convolutional network using a test data set, and to generate a graph convolutional neural network model based on the graph convolutional network that passes the test.

[0066] In a specific application scenario, the model building module 303 can be specifically used to perform global training on the graph convolutional network based on a preset number of iterations using a training data set, update the network weights of the graph convolutional network using an optimizer during the global training process, and perform L2 regularization on the loss function in the graph convolutional network to minimize the value of the loss function, wherein the global training includes forward propagation and backward propagation; after the graph convolutional network completes the global training, the network weights of the graph convolutional network are saved based on a preset format.

[0067] In a specific application scenario, the model building module 303 can be specifically used to determine multiple initial sample nodes, calculate the Euclidean distance between each initial sample node and the remaining initial sample nodes; for each initial sample node, select a preset number of Euclidean distances as the target Euclidean distance in the order of multiple Euclidean distances from small to large, and determine the target initial sample node corresponding to the target Euclidean distance; establish an adjacency matrix based on each initial sample node and the corresponding target initial sample node.

[0068] It should be noted that for other corresponding descriptions of the functional units involved in the converter molten iron determination device provided in this embodiment, reference can be made to Figure 1 and Figure 2 The corresponding description in will not be repeated here.

[0069] Based on the above Figure 1 The method shown, accordingly, this embodiment also provides a storage medium, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned converter molten iron determination method is implemented.

[0070] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product. The software product to be identified can be stored in a non-volatile storage medium (which can be a CD-ROM, USB flash drive, mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the converter molten iron determination method in various implementation scenarios of the present application.

[0071] Based on the above Figure 1 and Figure 2 The method shown, and Figure 6 The converter molten iron determination device shown in the figure is used to achieve the above-mentioned purpose. Figure 7As shown, this embodiment also provides a physical device for determining the molten iron in a converter, which includes a communication bus, a processor, a memory and a communication interface, and may also include an input and output interface and a display device, wherein each functional unit can communicate with each other through the bus. The memory stores a computer program, and the processor is used to execute the program stored in the memory and execute the method for determining the molten iron in the converter in the above embodiment.

[0072] Optionally, the physical device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a WI-FI interface), etc.

[0073] Those skilled in the art will appreciate that the structure of a converter molten iron determination entity equipment provided in this embodiment does not constitute a limitation on the entity equipment, and may include more or fewer components, or a combination of certain components, or different component arrangements.

[0074] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware of the physical device and the software resources to be identified, and supports the operation of the information processing program and other software and / or programs to be identified. The network communication module is used to realize the communication between the components inside the storage medium, and the communication with other hardware and software in the information processing physical device.

[0075] Through the description of the above implementation methods, the technicians in this field can clearly understand that the present application can be implemented by means of software plus necessary general hardware platforms, or by hardware. First, the original molten iron data is obtained, and the original molten iron data is preprocessed, and the original data set is constructed based on the preprocessed original molten iron data, and then the molten iron category label is added to part of the data in the original data set according to the preset molten iron category classification rule to obtain the target data set, and then the graph convolution network is constructed based on the target data set, and the graph convolution network is trained and tested using the target data set to generate a graph convolution neural network model, and finally the molten iron data to be tested is collected, and the molten iron data to be tested is judged using the graph convolution neural network model to obtain the molten iron judgment result. The above method collects the original molten iron data in the early stage, and marks part of the data according to the preset molten iron category classification rules, and classifies all the data by establishing a graph convolution neural network model, so that the determination of the molten iron category can be realized before the smelting begins, and then the prediction and early warning of dangerous accidents that may occur in actual production are realized, which helps to reduce the casualties caused by splashing and back-drying accidents in actual production, and reduce the corresponding economic losses, and improve the safety of production.

[0076] Those skilled in the art will appreciate that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily necessary for implementing the present application. Those skilled in the art will appreciate that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the description of the implementation scenario, or can be changed accordingly and located in one or more devices different from the present implementation scenario. The modules of the above-mentioned implementation scenario can be combined into one module, or can be further split into multiple submodules.

[0077] The above serial numbers of this application are only for description and do not represent the advantages and disadvantages of the implementation scenarios. The above disclosure is only a few specific implementation scenarios of this application, but this application is not limited to them, and any changes that can be thought of by technicians in this field should fall within the scope of protection of this application.

Claims

1. A method for determining molten iron in a converter, characterized in that: include: Acquire original molten iron data, preprocess the original molten iron data, and construct an original data set based on the preprocessed original molten iron data; Adding molten iron category labels to part of the data in the original data set according to a preset molten iron category classification rule to obtain a target data set; Building a graph convolutional network based on the target data set, and using the target data set to train and test the graph convolutional network to generate a graph convolutional neural network model; The molten iron data to be tested is collected, and the molten iron data to be tested is judged using the graph convolutional neural network model to obtain a molten iron judgment result.

2. The method according to claim 1, characterized in that The obtaining of raw molten iron data and preprocessing of the raw molten iron data includes: Obtaining original molten iron data, and calculating the mean of the original molten iron data and the standard deviation of each of the original molten iron data; Calculate a standard score corresponding to each piece of the original molten iron data based on the mean value of the original molten iron data and the standard deviation of each piece of the original molten iron data; Determining the standard score according to a preset standard score threshold; When the standard score of the original molten iron data does not exceed the standard score threshold, marking the original molten iron data as normal data; When the standard score of the original molten iron data exceeds the standard score threshold, the original molten iron data is marked as abnormal data, and the abnormal data is replaced with the mean value of the original molten iron data.

3. The method according to claim 1, characterized in that The adding molten iron category labels to part of the data in the original data set according to the preset molten iron category classification rule includes: Randomly extracting some data points from the original data set, marking the extracted data points as target data points, and obtaining the molten iron composition of each target data point; Based on the molten iron composition of the target data point, the molten iron category of the target data point is determined using a preset molten iron category classification rule, and a molten iron category label is added to the target data point according to the molten iron category.

4. The method according to claim 1, characterized in that The step of constructing a graph convolutional network based on the target data set includes: Dividing the target data set into a training data set and a test data set, wherein all data points in the training data set carry the molten iron category label; Defining a node feature matrix and setting an adjacency matrix based on a threshold connection method, and defining an input layer based on the node feature matrix and the adjacency matrix; Defining multiple graph convolutional layers and selecting activation functions, wherein the graph convolutional layers include a weight matrix; Standardizing the adjacency matrix, determining graph convolution propagation features based on the standardized adjacency matrix, the node feature matrix, the activation function, and the weight matrix, and defining a hidden layer according to the convolution propagation features and the plurality of graph convolution layers; Defining a loss function and an optimizer, constructing an initial graph convolutional network based on the input layer, the hidden layer, the loss function and the optimizer, and training the initial graph convolutional network using the training data set to update the parameters of the initial graph convolutional network, wherein the loss function is a cross entropy loss function; The test data set is used to evaluate the accuracy of the initial graph convolutional network after updating the parameters, and the initial graph convolutional network that passes the accuracy evaluation is selected as the final generated graph convolutional network.

5. The method according to claim 4, characterized in that The using the target data set to train and test the graph convolutional network to generate a graph convolutional neural network model includes: The graph convolutional network is globally trained using the training data set, and the trained graph convolutional network is tested using the test data set, and a graph convolutional neural network model is generated based on the graph convolutional network that passes the test.

6. The method according to claim 5, characterized in that The globally training the graph convolutional network using the training data set includes: Based on a preset number of iterations, the graph convolutional network is globally trained using the training data set, the network weights of the graph convolutional network are updated using the optimizer during the global training process, and the loss function in the graph convolutional network is L2 regularized to minimize the value of the loss function, wherein the global training includes forward propagation and backward propagation; After the graph convolutional network completes global training, the network weights of the graph convolutional network are saved based on a preset format.

7. The method according to claim 4, characterized in that The method of setting the adjacency matrix based on the threshold connection includes: Determine a plurality of initial sample nodes, and calculate the Euclidean distance between each initial sample node and the remaining initial sample nodes; For each initial sample node, according to the order of the multiple Euclidean distances from small to large, a preset number of Euclidean distances are selected as target Euclidean distances, and a target initial sample node corresponding to the target Euclidean distance is determined; An adjacency matrix is ​​established based on each initial sample node and the corresponding target initial sample node.

8. A converter molten iron determination device, characterized in that: The device comprises: A data acquisition module, used to acquire original molten iron data, preprocess the original molten iron data, and construct an original data set based on the preprocessed original molten iron data; A data processing module, used to add molten iron category labels to part of the data in the original data set according to a preset molten iron category classification rule to obtain a target data set; A model building module, used to build a graph convolutional network based on the target data set, and use the target data set to train and test the graph convolutional network to generate a graph convolutional neural network model; The result generation module is used to collect the molten iron data to be tested, and use the graph convolutional neural network model to judge the molten iron data to be tested to obtain the molten iron judgment result.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images