Converter end point carbon content prediction method based on graph structure working condition division

Abstract

The invention discloses a converter end point carbon content prediction method based on graph structure working condition division, and belongs to the technical field of converter steelmaking, and the method comprises the following steps: constructing an association relationship between different heat numbers based on an attribute grade division mode to obtain a heat number graph structure; community division is carried out in the heat chart structure so as to mine different converter steelmaking working condition modes; for each divided community, taking internal heat information as training data, and training a preset neural network model to obtain an end point carbon content prediction model corresponding to the corresponding community; and for the current production plan heat, determining the community to which the current production plan heat belongs by adopting a voting mechanism, and predicting the endpoint carbon content according to the model corresponding to the community. According to the method, the hidden working condition mode mining method is combined with the neural network prediction model, different prediction models are adopted for different working conditions, and the end point carbon content prediction hit rate in the converter smelting process is increased.

Description

Technical Field

[0001] The invention relates to the technical field of converter steelmaking, and in particular to a method for predicting converter end point carbon content based on graph structure working condition division. Background Art

[0002] Converter steelmaking is one of the main steelmaking methods in the current steel production field. Depending on the actual type of steel being smelted, the content of each element in the molten steel and the temperature need to be controlled within the target range, that is, endpoint control. However, the high temperature and complex smelting environment of the converter makes it difficult to directly obtain the content of various elements inside it in real time through sensors. How to efficiently and accurately indirectly predict the element content or temperature of the end point of converter steelmaking based on measurable data, and ultimately achieve efficient and accurate endpoint control, is the key to achieving smelting target control.

[0003] With the rapid development of machine learning technology, many prediction methods have been proposed to solve the endpoint prediction problem, such as support vector regression (SVR), back propagation neural network (BPNN) and long short-term memory network (LSTM). At present, many methods improve the prediction accuracy by combining clustering algorithms, and their purpose is also to distinguish different hidden working conditions to train neural network models for specific working conditions.

[0004] However, most clustering methods rely on simple distance metrics, such as Euclidean distance, which is calculated as shown in formula (1).

[0005]

[0006] This method of dividing the operating mode based on clustering distance measurement is a traditional method that is sensitive to noise and outliers. Most of them require the specification of the number of clusters, and it is difficult to determine the weights of each attribute involved in the clustering calculation. In the actual production process, the division of the operating mode usually has obvious discrete characteristics, rather than simply evaluating it by the distance measurement between certain continuous process attribute values. Therefore, the existing endpoint prediction scheme for the converter steelmaking process in the steel production process cannot meet the actual production needs in terms of prediction hit rate. Summary of the invention

[0007] The present invention provides a method for predicting the carbon content at the converter endpoint based on graph structure working condition division, so as to solve the technical problem that the existing endpoint prediction scheme for the converter steelmaking process in the steel production process cannot meet the actual production needs in terms of prediction hit rate.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0009] On the one hand, the present invention provides a method for predicting the carbon content of a converter end point based on graph structure working condition division, and the method for predicting the carbon content of a converter end point based on graph structure working condition division comprises:

[0010] The association relationship between different heats is constructed based on the attribute level division method to obtain a heat graph structure; wherein, in the heat graph structure, each node represents a heat, the node describes the historical smelting information of different heats, and the edges between the nodes represent the correlation between different heats;

[0011] Performing community division in the heat graph structure to explore different converter steelmaking operating modes;

[0012] For each divided community, the historical smelting information of all furnaces contained in it is used as training data to train the preset neural network model to obtain the endpoint carbon content prediction model corresponding to the corresponding community;

[0013] For the current production plan furnace, a voting mechanism is used to determine the corresponding community of the converter steelmaking operating mode to which it belongs, and the endpoint carbon content prediction model corresponding to this community is used to predict the endpoint carbon content of the current production plan furnace.

[0014] Furthermore, the association relationship between different heats is constructed based on the attribute level division method to obtain the heat map structure, including:

[0015] Obtaining smelting information corresponding to each historical heat, recording it as historical smelting information, performing data quality assessment and data screening on the historical smelting information, removing abnormal data therein, and obtaining a historical smelting information data set;

[0016] Select multiple key attributes, based on the attribute level division method, combined with the historical smelting information data set, take each key attribute as the division basis in turn, build the association relationship between different heats, and after building the association relationship between different heats according to each key attribute, obtain the initial heat map structure;

[0017] Aggregate and optimize the edges in the initial heat graph structure to obtain a final heat graph structure.

[0018] Furthermore, the smelting information includes serial number, furnace age, brand, amount of scrap steel added, weight of molten iron entering the furnace, temperature of molten iron entering the furnace, carbon content in the molten iron, silicon content in the molten iron, manganese content in the molten iron, phosphorus content in the molten iron, sulfur content in the molten iron, iron adding time, blowing time, main blowing time, smelting cycle, amount of lime added, total oxygen consumption, secondary blowing oxygen consumption, TSC temperature, TSC carbon content, TSO temperature and TSO carbon content.

[0019] Furthermore, the key attributes include: temperature of molten iron entering the furnace, weight of molten iron entering the furnace, carbon content in the molten iron, silicon content in the molten iron, oxygen consumption, amount of scrap steel added, and amount of lime added.

[0020] Furthermore, the attribute level classification method is combined with the historical smelting information data set, and each key attribute is used as the classification basis in turn to build the association relationship between different heats, including:

[0021] For the current key attribute, its value is divided into multiple discrete intervals. If the values ​​of the current key attribute corresponding to two furnaces are in the same interval, the nodes corresponding to the two furnaces are connected by an undirected edge.

[0022] Furthermore, the edges in the initial furnace graph structure are aggregated and optimized, including:

[0023] Aggregate the edges between two nodes in the initial furnace graph structure into an edge with a weight value, where the weight value of the edge after aggregation is the number of edges between the corresponding two nodes before aggregation;

[0024] After the edges between the nodes in the initial furnace graph structure are aggregated, the edges with a weight of 1 are deleted.

[0025] Furthermore, community division is performed in the furnace graph structure, including:

[0026] The endpoint carbon content of the heat is used as the initialization label, and a label propagation algorithm is used to unsupervisedly detect different communities in the heat graph structure; wherein each community represents a converter steelmaking operating mode.

[0027] Furthermore, the preset neural network model uses a multi-layer perceptron as a backbone network to predict the carbon content of the converter end point, and uses an error back propagation algorithm to make the network gradually converge to achieve optimal performance.

[0028] Furthermore, when training the preset neural network model, the temperature of molten iron entering the furnace, the weight of molten iron entering the furnace, the carbon content in the molten iron, the silicon content in the molten iron, the manganese content in the molten iron, the phosphorus content in the molten iron, the sulfur content in the molten iron, the oxygen blowing time, the oxygen consumption, the weight of scrap steel, the weight of iron ore, the weight of lime and the weight of light-burned dolomite are used as inputs of the model, and the endpoint TSO carbon content in the molten steel is used as the true label.

[0029] Furthermore, for the current production plan heat, a voting mechanism is used to determine the corresponding community of the converter steelmaking operation mode to which it belongs, including:

[0030] For the current production plan heat, the association relationship between it and each heat in the heat graph structure is constructed based on the attribute level division method, and the community with the largest number of edges connected to the current production plan heat is taken as the community to which it belongs.

[0031] On the other hand, the present invention further provides an electronic device, comprising a processor and a memory; wherein the memory stores at least one instruction, and the instruction is loaded and executed by the processor to implement the above method.

[0032] In yet another aspect, the present invention further provides a computer-readable storage medium, wherein at least one instruction is stored in the storage medium, and the instruction is loaded and executed by a processor to implement the above method.

[0033] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0034] This method combines the hidden operating condition pattern mining method with the neural network prediction model, adopts different prediction models for different operating conditions, and improves the prediction hit rate of the terminal carbon content in the converter smelting process.

[0035] This method has strong portability, and only needs to add a community division step on the basis of the existing algorithm to realize the mining of operating mode. The MLP model used can also be replaced with other models such as support vector machine SVM, long short-term memory network LSTM, etc. At the same time, the converter furnace graph structure constructed by the present invention can meet the needs of continuous incremental learning. With the accumulation of production data, the number of nodes in the furnace graph structure continues to increase, and the community structure is more stable and clear. The accuracy can be further improved by regularly updating the prediction model corresponding to different communities or operating modes.

[0036] The graph-structure-based prediction method proposed in this method has strong versatility and can be extended to various tasks such as converter endpoint temperature and other endpoint component prediction in the future. It has a very broad application prospect. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0038] Figure 1 It is a schematic diagram of the execution flow of a method for predicting carbon content at the converter endpoint based on graph structure working condition division provided by an embodiment of the present invention;

[0039] Figure 2 It is a schematic diagram of an example of attribute grade classification based on carbon content provided by an embodiment of the present invention;

[0040] Figure 3 It is a detailed execution flow diagram of a method for predicting carbon content at the converter end point based on graph structure working condition division provided by an embodiment of the present invention;

[0041] Figure 4 This is a schematic diagram of node label initialization according to endpoint carbon content provided by an embodiment of the present invention;

[0042] Figure 5 is a schematic diagram of a community decision method based on a voting mechanism provided by an embodiment of the present invention;

[0043] Figure 6 It is a schematic diagram of the result of dividing the working mode of the converter furnace diagram structure provided by an embodiment of the present invention;

[0044] Figure 7 It is a system block diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0046] First of all, it should be noted that in the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "exemplary" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "exemplarily" is intended to present the concept in a concrete way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0047] First embodiment

[0048] This embodiment provides a method for predicting the carbon content of a converter end point based on graph structure working condition division for a converter steelmaking process in a steel production process. The method selects some important attributes through mechanism analysis, and constructs a graph structure of steelmaking furnaces by using attribute level division. Based on the label propagation algorithm, label propagation and community division are performed in the furnace graph structure, and different converter steelmaking working condition modes are mined, and different neural network models are trained accordingly. In actual application, a voting mechanism is used to determine the corresponding community of the working condition mode to which the current production plan furnace belongs, and the neural network model corresponding to the community is used to predict the end point carbon content.

[0049] The method for predicting the carbon content of the converter end point based on the graph structure working condition division can be implemented by an electronic device, which can be a terminal or a server. The execution flow of the method is as follows: Figure 1 As shown, the following steps are included:

[0050] S1, constructing the association relationship between different heats based on the attribute level division method to obtain a heat graph structure; wherein, in the heat graph structure, each node represents a heat, the node describes the historical smelting information of different heats, and the edges between the nodes represent the correlation between different heats;

[0051] It should be noted that the method of dividing the working condition mode is similar to the rating system. For example, steel products can be divided into high carbon steel ([C]>0.20%), medium carbon steel ([C]=0.06%~0.20%) and low carbon steel ([C]<0.06%) according to their carbon content. This method draws on the idea of ​​this attribute level division to discretely construct the correlation between different furnaces, and on this basis to mine hidden working condition modes. In this regard, in order to realize the mining of hidden working condition modes in the converter steelmaking process, we need to build a graph structure that describes the correlation between different furnaces. The nodes in the furnace structure graph describe the historical smelting information of different furnaces. The historical smelting information of a furnace includes information such as molten iron temperature, molten iron carbon content, molten iron weight, oxygen consumption, end point carbon, and end point temperature. Each node represents a furnace. The most critical challenge in constructing the graph structure of the converter furnace furnace is to model the correlation between different furnaces, that is, to determine the edges between furnaces. This embodiment proposes a method called "attribute level division" to establish the edge relationship between nodes in the furnace graph structure. For example, the carbon content of molten iron is used as the division basis, the carbon content value is divided into multiple discrete intervals, and the furnace nodes in the same interval are connected with an undirected edge, such as Figure 2 As shown, Figure 2 The carbon content values ​​of the molten iron of the two nodes Heat1 and Heat3 are both in the range of [4.4%~4.5%], so there is an edge connecting these two nodes.

[0052] Specifically, in this embodiment, Figure 3 As shown, the implementation process of the above S1 is as follows:

[0053] S11, obtaining smelting information corresponding to each historical furnace, recording it as historical smelting information, performing data quality assessment and data screening on the historical smelting information, removing abnormal data therein, and obtaining a historical smelting information data set;

[0054] S12, selecting multiple key attributes, based on the attribute level classification method, combining with the historical smelting information data set, taking each key attribute as the classification basis in turn, and constructing the association relationship between different heats, after constructing the association relationship between different heats according to each key attribute, respectively, obtaining the initial heat map structure;

[0055] Among them, the attribute level division method, combined with the historical smelting information data set, takes each key attribute as the division basis in turn to build the association relationship between different heats, specifically: for the current key attribute, its value is divided into multiple discrete intervals, if the values ​​of the current key attributes corresponding to two heats are in the same interval, the nodes corresponding to the two heats are connected by an undirected edge;

[0056] S13, aggregating and optimizing the edges in the initial furnace graph structure to obtain a final furnace graph structure;

[0057] Among them, it should be noted that, in addition to the carbon content of molten iron, this embodiment also uses molten iron weight, molten iron temperature, molten iron silicon content, oxygen consumption, scrap steel weight and lime weight as division attributes and is used to establish edge relationships in the furnace diagram, so there may be multiple edges between nodes after the process is completed. In order to facilitate the subsequent mining of hidden operating modes, this embodiment aggregates multiple edges between nodes into an edge with a weight value, and the size of the weight value is the number of edges between nodes. It has been found through practice that the edges in the furnace diagram obtained through the above process are too dense, which is not conducive to distinguishing different hidden operating modes. To solve the above problem, this embodiment deletes the edges with a weight of 1, thereby obtaining a furnace structure diagram with a moderate edge density and a clear structure.

[0058] S2, performing community division in the heat graph structure to explore different converter steelmaking operating modes;

[0059] It should be noted that the adjusted furnace graph structure contains different hidden operating modes, which are represented as several communities in the graph. In order to divide these hidden communities (each community can be considered as a hidden operating mode), this embodiment uses a label propagation algorithm to detect different communities in an unsupervised manner. The core idea of ​​the label propagation algorithm is: after initializing each node with a unique label, the algorithm repeatedly socializes the label of a node into the label that appears most frequently among the neighboring nodes of the node. When the label of each node appears most frequently among its neighboring nodes, the algorithm stops.

[0060] Specifically, the label propagation algorithm is mainly divided into the following five steps:

[0061] 1) Initialize the labels of all nodes in the graph. For a given node x, C x (0) = x.

[0062] 2) Set t=1.

[0063] 3) Arrange the nodes in the graph in a random order and set it to X.

[0064] 4) For each x∈X selected in a particular order, let The value of function f is the label with the highest frequency among the adjacent labels. If there are multiple labels with the highest frequency, a label is randomly selected.

[0065] 5) If each node has the largest number of labels among its neighboring nodes, stop the algorithm; otherwise, set t=t+1 and go to step 3).

[0066] Before executing the label propagation algorithm, it is necessary to determine the initialization label of each node. The simple approach is to use different numbers as the initialization label for each node. The main function of this embodiment is to predict the final carbon content of converter steelmaking, so the final carbon content of the historical furnace is used as the initialization label. The specific approach is to discretize the final carbon content into several intervals according to the attribute level division method, and assign an interval code to each interval. Figure 4 As shown in the figure, four communities are obtained according to the endpoint carbon content: 0, 1, 2, and 3. After executing several rounds of label propagation, a stable community structure is finally formed, and nodes in the same community have the same label.

[0067] S3, for each community, the historical smelting information of all furnaces contained in it is used as training data to train a preset neural network model to obtain a terminal carbon content prediction model corresponding to the corresponding community;

[0068] It should be noted that, Figure 3 As shown, after running the label propagation algorithm, a certain number of communities will be obtained, and each community corresponds to a certain operating mode. Therefore, after obtaining the execution result of the label propagation, it is necessary to train neural network models of different operating modes for different communities. In this regard, this embodiment uses a multi-layer perceptron (MLP) as the backbone network to predict the carbon content of the converter end point and uses the error back propagation algorithm to make the network gradually converge to achieve optimal performance. Since there are multiple communities, the neural network model corresponding to each community uses the furnace information within the community as training data. During training, the temperature of the molten iron entering the furnace, the weight of the molten iron entering the furnace, the carbon content in the molten iron, the silicon content in the molten iron, the manganese content in the molten iron, the phosphorus content in the molten iron, the sulfur content in the molten iron, the oxygen blowing time, the oxygen consumption, the weight of scrap steel, the weight of iron ore, the weight of lime and the weight of light-burned dolomite are used as the input of the model, and the end point TSO carbon content in the molten steel is used as the true label.

[0069] S4, for the current production plan furnace, a voting mechanism is used to determine the corresponding community of the converter steelmaking operating mode to which it belongs, and the endpoint carbon content prediction model corresponding to this community is used to predict the endpoint carbon content of the current furnace;

[0070] It should be noted that after obtaining the neural network models corresponding to different working modes, for the prediction of the final carbon content of the current furnace in the actual production process, this embodiment uses a voting mechanism to determine the community or working condition to which the current furnace belongs. The specific approach is: using attribute level division to build the association relationship between the current furnace and the historical furnaces in the graph, the community to which the current furnace belongs is the community with the largest number of edges connected to it, and the customized model corresponding to the community is used as the prediction model to guide the production of the current furnace.

[0071] The above process is as follows Figure 5 As shown, Figure 5 The heat named Heat* is assigned to the above community based on the voting mechanism and uses Model 1 as its prediction model to predict the final carbon content.

[0072] In summary, this embodiment provides a method for predicting the carbon content at the end of a converter based on the division of working conditions in a graph structure. The method adopts a combination of hidden working condition mode mining and neural network model to solve the problem of predicting the carbon content at the end of a converter. The historical furnace information is modeled in a graph structure, and the association between different furnaces is constructed based on the attribute level division. Different hidden working condition modes are mined by label propagation on the furnace structure diagram, and customized neural network models for specific working conditions are trained to achieve the effect of "divide and conquer". The test results show that the method for predicting the carbon content at the end of a converter provided in this embodiment is superior to the traditional single neural network model based on the full amount of data in multiple indicators such as the carbon hit rate, mean square error and mean absolute error of the end of a converter, thereby demonstrating the effectiveness of the method based on the division of working condition modes in the graph structure of the converter furnace. In addition, the graph structure of the converter furnace established by this method will continue to expand with the accumulation of smelting data. By continuously adding furnace data to update the graph structure, the prediction accuracy of the model can be further improved, and the continuous and incremental learning of the model can be realized. It can be applied to a variety of prediction tasks such as converter end carbon, end temperature and other end components.

[0073] Second embodiment

[0074] This embodiment uses a practical application example to illustrate the implementation process and effect of the method for predicting the final carbon content of a converter based on graph structure working condition division of the present invention. The method is based on the furnace graph structure and utilizes hidden working condition pattern mining to divide the historical furnace information into multiple communities, and then trains a neural network prediction model for specific working conditions for each community. To this end, the method first needs to obtain the available historical furnace information through data cleaning, and then construct a furnace graph according to the attribute level division method, and then update the furnace graph through edge aggregation and optimization and use it for label propagation to obtain the final community structure, and finally train different neural network models for different communities or working conditions and use them for prediction. For the current furnace in production, a voting mechanism is used to select the trained model for prediction, and update it to the furnace structure graph. The process is as follows Figure 3 shown.

[0075] Based on the above, the method mainly includes three parts: (1) Construction of converter heat graph structure based on attribute level division; (2) Label propagation and community division on converter heat graph structure; (3) Training of neural network prediction model and endpoint carbon prediction for working conditions. Each part is explained in detail below.

[0076] First of all, it should be noted that for practical applications, the training data should at least cover typical steel grades, with a training data set of no less than 1,500 and a test data set of no less than 300, of which 20% of the data in the training set is used to verify the training effect of the model. During model training, the model converges when the loss value does not change within several rounds, and the test process begins. The model hit rate is calculated. When the hit rate reaches more than 85% within the specified error range, it can be put into use, and the data of the current furnace can be used for actual prediction.

[0077] The following are the steps for constructing the structure of the converter heat diagram based on attribute level division:

[0078] The converter steelmaking process data is a collection of measurable data in the converter steelmaking process obtained through various methods, including serial number, furnace age, brand, scrap steel addition amount, molten iron weight, molten iron temperature, carbon (C) content in molten iron, silicon (Si) content in molten iron, manganese (Mn) content in molten iron, phosphorus (P) content in molten iron, sulfur (S) content in molten iron, iron addition time, blowing time, main blowing time, smelting cycle, lime addition amount, total oxygen consumption, secondary blowing oxygen consumption, TSC temperature, TSC carbon content, TSO temperature, and TSO carbon content.

[0079] This method performs data quality assessment and screening on the initial data, improves the data quality of the data set by removing abnormal data, and finally obtains the experimental data set. It is generally required that the data at least covers typical steel grades, and the amount of training data and test data is not less than 1500 and 300 respectively.

[0080] According to expert experience and metallurgical industry mechanism, in the above training data, this method uses the temperature of molten iron entering the furnace, the weight of molten iron entering the furnace, the carbon (C) content in the molten iron, the silicon (Si) content in the molten iron, the oxygen consumption, the amount of scrap steel added and the amount of lime added as the key attributes used for attribute classification and thus constructs the furnace graph. After the edge optimization aggregation operation, a furnace graph structure with 1603 nodes and 4504 edges is established.

[0081] The following are the steps of label propagation and community division on the converter heat graph structure:

[0082] The community division adopts the label propagation algorithm LPA, and uses the Neo4j graph database and the Neo4j Graph DataScience Library (GDS) framework to complete the label propagation process. Use the Cypher language to configure some configuration options of the LPA algorithm in GDS, such as initializing the label seed, the maximum number of iterations, etc. After completing the label propagation, the community information of the nodes in the graph can be obtained, and the training data of the nodes of different communities can be saved by community classification for subsequent verification and testing processes. Among them, 20% of the data is extracted from each community as the verification set of the customized model, and the remaining data is used for model training.

[0083] After executing the LPA algorithm, four communities can be obtained. The distribution is that community 0 has 788 nodes, community 1 has 368 nodes, community 2 has 301 nodes, and the remaining 146 nodes belong to community 3. Some of the results obtained by the division are as follows: Figure 6 shown.

[0084] The following are the steps for training the neural network prediction model and predicting the end point carbon for the working conditions:

[0085] 1. Prediction model training data preparation:

[0086] The method uses 13 input data for training the neural network, which are: temperature of molten iron entering the furnace X1, weight of molten iron entering the furnace X2, carbon (C) content in molten iron X3, silicon (Si) content in molten iron X4, manganese (Mn) content in molten iron X5, phosphorus (P) content in molten iron X6, sulfur (S) content in molten iron X7, oxygen blowing time X8, oxygen consumption X9, scrap steel weight X1, and the content of molten iron in molten iron X2. 10 、Iron ore weight X 11 , lime weight X 12 And light burned dolomite weight X 13 The endpoint TSO carbon content in the steel liquid is used as the true label.

[0087] 2. Forward calculation of prediction model:

[0088] The model training is based on the PyTorch deep learning framework. The training data features are used as input and first pass through a simple fully connected neural network with a network structure of 13×16×8×1, whose output is the predicted value of the endpoint carbon content.

[0089] 3. Back propagation of prediction model error:

[0090] After the forward pass, back propagation is required to adjust the network parameters for the training model. This method uses the mean square error MSE as the loss function used for back propagation as shown in the formula.

[0091]

[0092] Among them, y i and Represents the predicted value and true value of the i-th sample. The error calculated above is used to update the weights in the MLP through the error back propagation algorithm.

[0093] 4. Prediction model performance verification:

[0094] The data used for model verification comes from 20% of the sampled data of 1603 nodes, a total of 321 items. In actual production, hit rate is often used as an indicator to evaluate the prediction model, while in the field of deep learning, MSE is often used as an indicator to evaluate the performance of the prediction model. The hit rate is evaluated by taking the final endpoint carbon content prediction result y i The endpoint TSO carbon content in the steel liquid as the true label in the dataset To make a difference, we count the hits with errors in the narrow window error interval (-0.009, 0.009), the general window error interval (-0.012, 0.012) and the wide window error interval (-0.02, 0.02). The calculation method is shown in the formula

[0095]

[0096] Among them, bound is 0.009, 0.012 and 0.02 respectively, n is the amount of data, and count is the counting function. The prediction accuracy and MSE index of the model in the general window error range in each community are shown in Table 1.

[0097] Table 1 Prediction accuracy and MSE index of the model in the general window error range in each community

[0098]

[0099] *Single model means that communities are not divided, and all community data are used as training data to train a single prediction model.

[0100] 5. Use of prediction model:

[0101] For model testing, when the hit rate is evaluated within the error interval (-0.012, 0.012), if the hit rate reaches above 85%, the model can be put into actual prediction (use).

[0102] The model is used with single data for which prediction tasks need to be performed in the converter steelmaking process, i.e., 13×1 (13 is the number of input data columns, 1 is the number of test data) test data as input, and the final output result is the predicted value of the end point carbon content. Four representative data are selected from the test data as shown in Table 2.

[0103] Table 2 Representative data selected from the test data

[0104]

[0105] 6. Model hit rate improvement effect:

[0106] The model test uses 397 heat data. This paper uses the test set to evaluate the model performance as a whole. In the error intervals (-0.009, 0.009), (-0.012, 0.012) and (-0.02, 0.02), the prediction hit rates are 79.3%, 90.2% and 98.0% respectively. Compared with the use of a single MLP model, the hit rate is increased by about 0.5%, 0.8% and 1%, as shown in Table 3.

[0107] Table 3 Prediction hit rates of this model and single model in different error intervals

[0108]

[0109] It can be seen from the above that the method of the present invention improves the prediction hit rate of the end point carbon content in the converter smelting process.

[0110] Third embodiment

[0111] This embodiment provides an electronic device, such as Figure 7 As shown, the electronic device includes: a processor and a memory; wherein the processor and the memory can be connected via a communication bus; the memory stores at least one instruction, and the instruction is loaded and executed by the processor to implement the method of the first embodiment. In addition, the electronic device may also include a transceiver, the processor and the transceiver can be connected via a communication bus, and the transceiver is used to communicate with other devices.

[0112] Next, combine Figure 7 The following is a detailed introduction to the various components of the electronic device:

[0113] Among them, the processor is the control center of the electronic device, and the electronic device may include multiple processors, each of which may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The processor here may be a processor or a general term for multiple processing elements. For example, the processor is one or more central processing units (CPUs), or other general-purpose processors, application specific integrated circuits (ASICs), or one or more integrated circuits configured to implement an embodiment of the present invention, such as one or more microprocessors (digital signal processors, DSPs), or one or more field programmable gate arrays (field programmable gate arrays, FPGAs), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor may execute various functions of the electronic device by running or executing software programs stored in the memory and calling data stored in the memory.

[0114] In a specific implementation, as an embodiment, the processor may include one or more CPUs, such as Figure 7 The CPU0 and CPU1 shown in the figure are, of course, only exemplary.

[0115] The memory is used to store the software program for executing the solution of the present invention, and the execution is controlled by the processor. The specific implementation method can refer to the above method embodiment, which will not be repeated here.

[0116] Optionally, the memory may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory may be integrated with the processor or exist independently and accessed through the interface circuit ( Figure 7 (not shown) is coupled to the processor, which is not specifically limited in this embodiment of the present invention.

[0117] The transceiver may include a receiver and a transmitter ( Figure 7 The receiver is used to implement the receiving function, and the transmitter is used to implement the sending function. The transceiver can be integrated with the processor or exist independently and communicate with the electronic device through the interface circuit ( Figure 7 (not shown) is coupled to the processor, which is not specifically limited in this embodiment of the present invention.

[0118] In addition, it should be noted that Figure 7 The structure of the electronic device shown in the figure does not constitute a limitation on the device, and the actual device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently. In addition, the technical effects achieved by the electronic device when executing the method of the first embodiment above can refer to the technical effects described in the first embodiment above, so they are not repeated here.

[0119] Fourth embodiment

[0120] This embodiment provides a computer-readable storage medium, which stores at least one instruction, and the instruction is loaded and executed by a processor to implement the method of the first embodiment. The computer-readable storage medium may be a ROM, a random access memory, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc. The instructions stored therein may be loaded by a processor in a terminal to execute the method.

[0121] In addition, it should be noted that the present invention can be provided as a method, an apparatus or a computer program product. Therefore, the embodiment of the present invention can be in the form of a full or partial hardware embodiment, a full or partial software embodiment or an embodiment combining software and hardware. Moreover, when implemented using software, the embodiment of the present invention can be in the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program codes. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center containing one or more available media sets. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a DVD), or a semiconductor medium. The semiconductor medium may be a solid state hard disk.

[0122] The embodiments of the present invention 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 invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks 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, 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.

[0123] 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 1These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable terminal device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0124] It should also 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. The terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the 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 more restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements. In addition, the term "and / or" is only an association relationship describing the associated objects, indicating that there can be three relationships, for example, A and / or B, which can represent: A exists alone, A and B exist at the same time, and B exists alone, wherein A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding. "At least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can be represented by: a, b, c, ab, ac, bc or abc, where a, b, c can be single or plural.

[0125] In addition, it can be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0126] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0127] In several embodiments provided by the present invention, it should be understood that the disclosed equipment, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of functional modules / units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point, the coupling or direct coupling or communication connection between each other shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms. The unit described as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0128] If the method is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0129] Finally, it should be noted that the above is only a preferred embodiment of the present invention. It should be pointed out that although the preferred embodiment of the present invention has been described, for ordinary technicians in this technical field, once the basic creative concept of the present invention is known, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the protection scope of the present invention. Therefore, the attached claims are intended to be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the embodiments of the present invention.

Claims

1. A method for predicting the carbon content at the end point of a converter based on graph structure working condition division, characterized in that: include: The association relationship between different heats is constructed based on the attribute level division method to obtain a heat graph structure; wherein, in the heat graph structure, each node represents a heat, the node describes the historical smelting information of different heats, and the edges between the nodes represent the correlation between different heats; Performing community division in the heat graph structure to explore different converter steelmaking operating modes; For each divided community, the historical smelting information of all furnaces contained in it is used as training data to train the preset neural network model to obtain the endpoint carbon content prediction model corresponding to the corresponding community; For the current production plan furnace, a voting mechanism is used to determine the corresponding community of the converter steelmaking operating mode to which it belongs, and the endpoint carbon content prediction model corresponding to this community is used to predict the endpoint carbon content of the current production plan furnace.

2. The method for predicting the carbon content at the converter end point based on graph structure working condition division according to claim 1, characterized in that: The association relationship between different heats is constructed based on the attribute level division method to obtain the heat map structure, including: Obtaining smelting information corresponding to each historical heat, recording it as historical smelting information, performing data quality assessment and data screening on the historical smelting information, removing abnormal data therein, and obtaining a historical smelting information data set; Select multiple key attributes, based on the attribute level division method, combined with the historical smelting information data set, take each key attribute as the division basis in turn, build the association relationship between different heats, and after building the association relationship between different heats according to each key attribute, obtain the initial heat map structure; Aggregate and optimize the edges in the initial heat graph structure to obtain a final heat graph structure.

3. The method for predicting the carbon content at the converter end point based on graph structure working condition division according to claim 2 is characterized in that: The smelting information includes: serial number, furnace age, brand, amount of scrap steel added, weight of molten iron entering the furnace, temperature of molten iron entering the furnace, carbon content in the molten iron, silicon content in the molten iron, manganese content in the molten iron, phosphorus content in the molten iron, sulfur content in the molten iron, iron adding time, blowing time, main blowing time, smelting cycle, amount of lime added, total oxygen consumption, secondary blowing oxygen consumption, TSC temperature, TSC carbon content, TSO temperature and TSO carbon content.

4. The method for predicting the carbon content at the converter end point based on graph structure working condition division according to claim 2, characterized in that: The key attributes include: the temperature of the molten iron entering the furnace, the weight of the molten iron entering the furnace, the carbon content in the molten iron, the silicon content in the molten iron, the oxygen consumption, the amount of scrap steel added, and the amount of lime added.

5. The method for predicting the carbon content at the converter end point based on graph structure working condition division according to claim 2, characterized in that: The attribute level classification method, combined with the historical smelting information data set, takes each key attribute as the classification basis in turn to build the association relationship between different heats, including: For the current key attribute, its value is divided into multiple discrete intervals. If the values ​​of the current key attribute corresponding to two furnaces are in the same interval, the nodes corresponding to the two furnaces are connected by an undirected edge.

6. The method for predicting the carbon content at the converter end point based on graph structure working condition division according to claim 5, characterized in that: Aggregating and optimizing the edges in the initial furnace graph structure, including: Aggregate the edges between two nodes in the initial furnace graph structure into an edge with a weight value, where the weight value of the edge after aggregation is the number of edges between the corresponding two nodes before aggregation; After the edges between the nodes in the initial furnace graph structure are aggregated, the edges with a weight of 1 are deleted.

7. The method for predicting the carbon content at the converter end point based on graph structure working condition division according to claim 1, characterized in that: The community division is performed in the furnace graph structure, including: The endpoint carbon content of the heat is used as the initialization label, and a label propagation algorithm is used to unsupervisedly detect different communities in the heat graph structure; wherein each community represents a converter steelmaking operating mode.

8. The method for predicting the carbon content at the converter end point based on graph structure working condition division according to claim 1, characterized in that: The preset neural network model uses a multi-layer perceptron as a backbone network to predict the carbon content of the converter end point, and uses an error back propagation algorithm to make the network gradually converge to achieve optimal performance.

9. The method for predicting the carbon content at the converter end point based on graph structure working condition division according to claim 1, characterized in that: When training the preset neural network model, the temperature of molten iron entering the furnace, the weight of molten iron entering the furnace, the carbon content in the molten iron, the silicon content in the molten iron, the manganese content in the molten iron, the phosphorus content in the molten iron, the sulfur content in the molten iron, the oxygen blowing time, the oxygen consumption, the weight of scrap steel, the weight of iron ore, the weight of lime and the weight of light-burned dolomite are used as inputs of the model, and the endpoint TSO carbon content in the molten steel is used as the true label.

10. The method for predicting the carbon content at the end point of a converter based on graph structure working condition division according to claim 1, characterized in that: The voting mechanism is used to determine the corresponding community of the converter steelmaking working mode for the current production plan furnace, including: For the current production plan heat, the association relationship between it and each heat in the heat graph structure is constructed based on the attribute level division method, and the community with the largest number of edges connected to the current production plan heat is taken as the community to which it belongs.

Images