A method for predicting converter endpoint temperature based on XGBoost and finite element method
By combining XGBoost machine learning and Fluent finite element technology, a converter end-point temperature prediction model was established, which solved the problem of inaccurate end-point temperature control in converter steelmaking, achieved high-precision prediction and temperature field state analysis, and improved the quality and energy of molten steel Consumption management.
Patent Information
- Application Number
- CN202410962365.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-07-18
AI Technical Summary
Inaccurate end-point temperature control of converter steelmaking leads to extended smelting time, increased metal consumption and steel quality impacts. It is difficult for the prior art to accurately predict the end-point temperature of converter blowing, especially in complex multivariable and nonlinear systems.
Using XGBoost-based machine learning algorithm combined with Fluent finite element software, a converter end-point temperature prediction model is established through data preprocessing, model training and multiphase flow simulation, providing prediction and visual analysis of temperature field states.
It realizes high-precision converter end point temperature prediction, provides more comprehensive and specific temperature control guidance, and improves molten steel quality and energy consumption management.
Smart Images

Figure CN118504364B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of metallurgical technology, and in particular to a method for predicting converter endpoint temperature based on XGBoost and finite element combined drive. Background Art
[0002] Oxygen converter steelmaking is the main steelmaking method at present. Oxygen converter steelmaking not only has fast smelting speed, high production efficiency, strong raw material adaptability, can use about 20% of scrap steel and automatic control, but also overcomes the shortcomings of poor steel quality and few varieties during air blowing.
[0003] The determination of the converter blowing end point is an important link in the later stage of converter blowing. Improving the hit rate of converter blowing end point temperature can improve the quality of molten steel. Inaccurate control of the end point temperature will prolong the smelting time, reduce the life of the furnace lining, increase metal consumption, and affect the quality of steel. Since converter steelmaking is a complex system with multi-variable inputs, outputs, severe nonlinear relationships, and strong coupling, there are many factors that are difficult to quantify, so it cannot be well described linearly by mathematical equations. Although the actual measurement and experimental research methods can solve some problems, these methods require technicians to accumulate long-term work experience. With the rapid development of computer science and steel smelting technology, and at the same time, higher and higher requirements are put forward for the optimization of smelting processes. We need an accurate method for predicting the converter blowing end point temperature.
[0004] With the rapid development of artificial intelligence, there are more and more solutions to predict the converter endpoint temperature using machine learning big data algorithms. Compared with empirical modeling and mechanism modeling, data modeling does not rely on complex professional knowledge and years of experience accumulation. It is fast and efficient, but the prediction result of data modeling is a single value and cannot predict the macroscopic temperature field state. The present invention combines big data algorithms and finite element analysis, which not only meets the needs of fast and accurate prediction of the converter endpoint temperature, but also provides the temperature field state of the entire converter, ensuring that operators have a clearer understanding of temperature changes through data visualization. Summary of the invention
[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a converter endpoint temperature prediction method based on XGBoost and finite element combined drive.
[0006] To achieve the above object, the technical solution adopted by the present invention is: a converter endpoint temperature prediction method based on XGBoost and finite element joint drive, comprising the following steps:
[0007] S1: Collect converter blowing production process data, production plan data and sampling temperature measurement data;
[0008] S2: preprocess the data obtained in step S1 to remove duplicate items and null values in the data; standardize the data using the z-score method to eliminate the influence of the dimension and the variable's own variation size and the value size; and use the box plot detection method to delete abnormal points;
[0009] S3: Divide the data set obtained by step S2 into a training set and a test set;
[0010] S4: Use XGBoost to build a model in Python environment with the training set data, establish the relationship between input features and output features, and sort the contribution of input features to output features. Finally, remove features with a contribution score of less than 0.025 to the target index.
[0011] S5: Use the XGBoost model built in step S4 to verify the test set, using the mean square error MSE and determination coefficient R 2 As the evaluation criteria, if the MSE value and R 2 If the values are all within the preset range, the XGBoost model is judged to be qualified, and the qualified XGBoost model is used to predict the converter temperature; otherwise, it is unqualified, and the operations of step S2 to step S5 are repeated;
[0012] S6: Use Fluent finite element software to establish a multiphase flow model, specifically:
[0013] Apply boundary conditions corresponding to the working conditions of step S1, and repeatedly check the actual process results in step S1 with the simulated multiphase flow field results until the applied boundary conditions make the simulation results consistent with the actual process results;
[0014] S7: according to the multiphase flow model established in step S6, different converter endpoint temperatures are set to simulate the temperature field of the molten steel in the converter, and the molten steel temperature field data is exported;
[0015] S8: Predict the converter endpoint temperature value according to the qualified XGBoost model in step S5, select the corresponding molten steel temperature field data in step S7, and use matplotlib in the python environment to build a temperature field cloud map.
[0016] As a specific implementation, in step S5, the XGBoost model is set to a tree model gbtree, where:
[0017] The maximum depth of the tree is set to 10.
[0018] The learning rate cta is set to 0.056 in the model;
[0019] The number of gbtree classifiers is set to 1000;
[0020] The setting range of MSE value is less than 20, R 2 The setting range of the value is greater than 0.8, and the ratio of the converter temperature prediction value to the actual value within ±10℃ is greater than 90%.
[0021] As a specific implementation, the multiphase flow model established in step S6 is:
[0022] ,
[0023] Where: u, v, w are velocity vectors along the x, y, z directions, respectively; t is time; ρ is density; p is pressure; Fx, Fy, Fz are external forces on the unit volume of fluid in the x, y, z directions, respectively; μ is dynamic viscosity;
[0024] The heat conduction equation is:
[0025] ,
[0026] Where: T is temperature, °C; t is time, s; k is thermal conductivity, W / m·K; ρ is density, kg / m 3 ; C p is the specific heat, J / Kg•K.
[0027] As a specific implementation, in step S6, in the Fluent finite element method for establishing the multiphase flow model, the initial boundary conditions are set as follows:
[0028] The physical model is set according to the actual converter dimensions;
[0029] In the physical model-model module, set the material physical property parameters, set the time to steady state, select the energy equation, use the k-ε turbulence model for the viscosity model, and use the VOF model for the multiphase flow field model;
[0030] In the physical model-material module, set the density and viscosity of the fluid molten steel and oxygen, and the density, thermal conductivity, and specific heat capacity of the solid converter lining and shell materials;
[0031] Set the heat transfer coefficient between molten steel and furnace lining and the heat transfer coefficient between furnace shell and ambient air in the physical model-boundary conditions;
[0032] In the initialization conditions of the physical model-solution module, set the molten steel temperature to the converter end temperature.
[0033] As a specific implementation method, in step S7, different converter endpoint temperatures are set, the converter endpoint temperature range is 1500~1800℃, the setting value interval is n℃, and there are 300 / n+1 converter endpoint temperatures in total. Each converter endpoint temperature is substituted into the Fluent finite element multiphase flow model in step S6 to simulate the molten steel temperature field and export XYZ type data.
[0034] Due to the application of the above technical solution, the present invention has the following advantages compared with the prior art:
[0035] 1) The present invention builds a temperature prediction model through the XGBoost machine learning algorithm, which requires a small amount of data and has high prediction accuracy;
[0036] 2) The present invention combines the finite element multiphase flow temperature model to enrich the single prediction result of machine learning, calculates the node temperature of each part of the molten steel in the converter, and thus provides more accurate early warning and guidance for production operations;
[0037] 3) The present invention uses the calculation results of the data model as the initial conditions of the finite element model to calculate the node temperatures of various parts of the molten steel in the converter. The production operators have a more comprehensive and specific understanding of the temperature state of the molten steel at the converter end point, which facilitates the adoption of certain process measures to control the temperature, so as to obtain molten steel at an ideal temperature and save energy. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 It is a flow chart of a method for predicting converter endpoint temperature based on XGBoost and finite element joint drive according to the present invention;
[0039] Figure 2 This is a schematic diagram of the ranking of feature importance parameters;
[0040] Figure 3 This is a comparison chart between the measured value and the predicted value;
[0041] Figure 4 is the converter physical model;
[0042] Figure 5 This is the finite element temperature field cloud diagram of the converter;
[0043] Figure 6 This is the temperature field cloud diagram of the converter using python matplotlib. DETAILED DESCRIPTION
[0044] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.
[0045] The present invention provides a method for predicting the endpoint temperature of a converter based on the joint drive of XGBoost and finite element method. The converter steelmaking production process includes multiple process sections such as pretreatment, charging, melting, oxygen blowing and steel tapping. The equipment used for the converter steelmaking production process includes a converter, cooling equipment, dust removal equipment, blowing equipment, etc. The system used for the converter steelmaking production process includes a process control system and a production management system.
[0046] See also Figure 1 As shown, the prediction method comprises the following steps:
[0047] S1: Collecting converter steelmaking production data
[0048] Specifically, it includes collecting converter blowing production process data, production plan data and sampling temperature measurement data.
[0049] Among them, production process data refers to the data related to the entire converter steelmaking production process and each process interval.
[0050] The production process data is collected by the process control system, and the production process data includes but is not limited to charging data, converter blowing data, and converter status data. The charging, converter blowing, and converter status data all include multiple features.
[0051] Specifically, the charging data includes the ladle number (tank number), the time of charging the molten iron, the start time of blowing, the total charging amount, the temperature of the molten iron, the element content in the molten iron, the total scrap steel, the scrap steel ratio, the scrap steel gear, the slag material reference, the scrap steel added to the ladle, the temperature adjustment of the argon blowing station for the scrap steel for the furnace, the scrap steel, the recycled steel, the stamping material, the grinding chips, the shearing furnace charge, the screening slag steel, the Class A self-produced scrap steel, the heavy scrap, the first class of steelmaking pig iron, the crushed material, the packaged block, the alloy scrap steel, the second class of steelmaking pig iron, the high-grade magnetic separation powder, the pig iron slag, the magnetic separation slag steel, the steel bar cut head, the pure scrap steel, the self-produced steel slag, the limestone block, the dolomite for steelmaking, the quicklime block ash, etc. The element content in the molten iron includes C, Si, Mn, P, S, Cr, Ni, Cu, As, Sn, Sb, and Ti.
[0052] The converter blowing data include capping situation, oxygen supply time per pour, temperature per pour, C per pour, P per pour, S per pour, Cr per pour, Nb per pour, V per pour, number of oxygen supplies, number of pouring times, total oxygen supply time, theoretical oxygen consumption per pour, oxygen consumption per pour, total oxygen consumption, nitrogen consumption, steam recovery, slag time, smelting cycle, etc.
[0053] The converter status data includes ladle temperature, furnace age, gun (furnace) number, gun head gun age (furnace), gun body gun age (furnace), converter structure material, specific heat capacity, thermal conductivity, etc.
[0054] Production planning data includes furnace number, production shift, steel type, steel type classification, etc.
[0055] Sampling and temperature measurement data include detection protocol standards, sampling locations, sampler models, etc.
[0056] In this embodiment, data collection is based on the furnace batch number as process performance statistical information, that is, the data collection performance statistics of each furnace batch number are summarized into one record.
[0057] S2: Data preprocessing
[0058] The data collected in step S1 are preprocessed to remove duplicate items and null values in the data; the data are standardized using the z-score method to eliminate the influence of the dimension and the variable's own variation size and the value size; and the outliers are deleted using the box plot detection method.
[0059] S3: Divide the training set and test set
[0060] The data set obtained in step S2 is divided into a training set and a test set.
[0061] S4: Build an XGBoost model using the training set
[0062] The training set data is used in the Python environment to build a model using XGBoost, establish the relationship between input features and output features, and rank the contribution of input features to output features, such as Figure 2 As shown, finally, the features with a contribution score less than 0.025 to the target index are eliminated.
[0063] S5: Model Evaluation
[0064] Use the built XGBoost model to verify the test set, using the mean square error MSE and determination coefficient R 2 As the evaluation criteria, if the MSE value and R 2 If the MSE value and R 2 If any of the values is not within the preset range, the XGBoost model is judged to be unqualified, and the operations from step S2 to step S5 are repeated;
[0065] Here, the mean square error MSE and the coefficient of determination R 2 The specific formula is as follows:
[0066] ,
[0067] ,
[0068] Where: n is the number of sample predictions; is the model prediction value; y i is the actual value; is the average value; R 2 is the coefficient of determination; MSE is the mean square error.
[0069] The setting range of MSE value is less than 20, R 2 The setting range of the value is greater than 0.8, and the ratio of the converter temperature prediction value to the actual value within ±10℃ is greater than 90%. Figure 3 shown.
[0070] S6: Constructing a finite element multiphase flow model
[0071] The multiphase flow model is established using Fluent finite element software, specifically:
[0072] According to the applied boundary conditions corresponding to the working conditions in step S1, the actual process results in step S1 are repeatedly checked with the simulated multiphase flow field results until the applied boundary conditions make the simulation results consistent with the actual process results.
[0073] Here, the multiphase flow model is:
[0074] ,
[0075] Where: u, v, w are velocity vectors along the x, y, z directions respectively, t is time, ρ is density, p is pressure, Fx, Fy, Fz are external forces per unit volume of fluid in the x, y, z directions respectively, and μ is dynamic viscosity.
[0076] The heat conduction equation is:
[0077] ,
[0078] Where: T is temperature, °C; t is time, s; k is thermal conductivity, W / m·K; ρ is density, kg / m 3 ; C p is the specific heat, J / Kg•K.
[0079] The initial conditions for establishing the multiphase flow model in Fluent finite element are set as follows:
[0080] The physical model is set according to the actual converter size. At the same time, the model is simplified to facilitate simulation calculation. Figure 4 As shown;
[0081] In the physical model-model module, set the material physical property parameters, set the solver time to steady state, select the energy equation, use the k-ε turbulence model for the viscosity model, and use the VOF model for the multiphase flow model;
[0082] The density and viscosity of the fluid molten steel and oxygen are set in the physical model-material module, as shown in Table 1. The converter lining material is magnesia carbon brick, and the furnace shell is steel. Its density, thermal conductivity, and specific heat capacity are shown in Table 2.
[0083] Table 1 Density and viscosity of molten steel and oxygen
[0084]
[0085] Table 2 Density, specific heat capacity and thermal conductivity of furnace shell and lining
[0086]
[0087] The heat transfer coefficient between molten steel and furnace lining and the heat transfer coefficient between furnace shell and ambient air are set in the physical model-boundary conditions, as shown in Table 3;
[0088] Table 3 Heat transfer coefficient of furnace lining and furnace shell
[0089]
[0090] In the physical model-solution module initialization conditions, set the molten steel temperature to the S5 converter endpoint temperature.
[0091] S7: Simulation of the temperature field of molten steel in the converter
[0092] According to the multiphase flow model established in step S6, different converter end temperatures are set to simulate the temperature field of molten steel in the converter. Figure 5 As shown, and export the data;
[0093] Set different converter endpoint temperatures. Here, the converter endpoint temperature range is set to 1500~1800℃, the setting value interval is n℃, and there are 300 / n+1 converter endpoint temperatures in total. Here, 300 is the difference between the highest and lowest values in the converter endpoint temperature range.
[0094] Substitute each converter endpoint temperature into the Fluent finite element multiphase flow model in step S6 to simulate the molten steel temperature field and export XYZ data.
[0095] S8: Using python matplotlib to build a temperature field cloud map
[0096] According to the converter endpoint temperature prediction value obtained by using the qualified XGBoost model to predict the converter temperature in step S5, the corresponding molten steel temperature field XYZ data in step S7 is selected, and the temperature field cloud map is constructed using matplotlib in the python environment. Figure 6 shown. Embodiment 1
[0097] This example takes the converter steelmaking production data of Y Steel Plant in 2023 as an example to predict the converter endpoint temperature. The specific steps are as follows:
[0098] S1: Collect the converter steelmaking production data of Y Steel Plant from January to December 2023, including converter blowing production process data, production plan data and sampling temperature measurement data, a total of 8126 sets of data;
[0099] S2: Preprocess the data obtained in step S1, remove duplicates and null values in the data, standardize the data using the z-score method, eliminate the influence of the dimension and the variable's own variation size and the value size, use the box plot detection method to delete abnormal points, and 8000 groups of data remain;
[0100] S3: Divide the data set obtained in step S2 into a training set and a test set, and randomly divide the preprocessed data into 6400 sets of training data and 1600 sets of test data;
[0101] S4: Use XGBoost to build a model with the training set data in the python environment, establish the relationship between the input features and the output features, and sort the contribution of the input features to the output features. Finally, remove the features with an importance score less than 0.025 to the target index, and finally select the feature parameters, including: P, converter age, S, C, scrap ratio, molten steel weight, scrap steel addition, self-produced slag, limestone, smelting cycle, ladle temperature, converter molten iron temperature, Cr, heavy scrap, molten iron Mn, quicklime block ash, total oxygen, V, dolomite for steelmaking, and Nb;
[0102] The XGBoost model is set up as follows:
[0103] booster is set to tree model gbtree;
[0104] The maximum depth of the tree is set to 10.
[0105] The learning rate cta is set to 0.056 in the model;
[0106] The number of gbtree classifiers is set to 1000.
[0107] S5: Use the built XGBoost model to verify the test set, using the mean square error MSE and determination coefficient R 2 As an evaluation criterion, through calculation, when the number of classifiers of the XGBoost model is set to 1000, the MSE is 16.51 and R 2 The predicted value of the converter temperature is within ±10℃, which is 90.68%; the MSE value is less than the set value of 20, and R2 If the value is greater than the set value of 0.8 and the ratio of the difference between the predicted value and the actual value within ±10°C is greater than 90%, the model evaluation is considered qualified. After the model evaluation is qualified, the converter endpoint temperature is predicted;
[0108] S6: using Fluent finite element software to establish a multiphase flow model, setting materials, boundaries, and initial conditions accordingly according to the temperature measurement data of step S1 and the material data of the converter equipment, and performing multiple checks on the actual process results in step S1 and the simulated multiphase flow field results, so that the simulation results are consistent with the actual process results;
[0109] S7: according to the multiphase flow model established in S6, different converter endpoint temperatures are set, the temperature range is 1500-1800°C, the setting value interval is 5°C, and a total of 61 converter endpoint temperatures are set, and each converter endpoint temperature is substituted into the Fluent finite element multiphase flow model in step S6 to simulate the molten steel temperature field and export XYZ type data;
[0110] S8: According to the predicted value of converter terminal temperature obtained by using the qualified XGBoost model to predict the converter temperature in step S5, the corresponding molten steel temperature field data in step S7 is selected, and matplotlib is used in the python environment to build a temperature field cloud map.
[0111] The present invention is based on a converter endpoint temperature prediction method jointly driven by XGBoost and finite element, based on finite element, big data and machine learning technology, by collecting converter blowing production process data, production plan data and sampling temperature measurement data, establishing a data model and using the data model calculation results as the initial conditions of the finite element model, calculating the node temperatures of various parts of the molten steel in the converter, and the production operators have a more comprehensive and specific understanding of the converter endpoint molten steel temperature state, which is convenient for taking certain process measures to control the temperature, so as to obtain molten steel at an ideal temperature and save energy. The joint drive model combines the advantages of the finite element model and the data model, which can make the established model more representative and more adaptable, thereby providing more accurate early warning and guidance for production operations, thereby improving the quality of converter steelmaking and the stability of the production and manufacturing process.
[0112] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with the technology to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit of the present invention should be included in the protection scope of the present invention.
Claims
1. A converter endpoint temperature prediction method based on XGBoost and finite element joint drive, characterized in that: The following steps are involved: S1: Collect converter blowing production process data, production plan data and sampling temperature measurement data; S2: preprocess the data obtained in step S1 to remove duplicate items and null values in the data; standardize the data using the z-score method to eliminate the influence of the dimension and the variable's own variation size and the value size; and use the box plot detection method to delete abnormal points; S3: Divide the data set obtained by step S2 into a training set and a test set; S4: Use XGBoost to build a model in Python environment with the training set data, establish the relationship between input features and output features, and sort the contribution of input features to output features. Finally, remove features with a contribution score of less than 0.025 to the target index. S5: Use the XGBoost model built in step S4 to verify the test set, using the mean square error MSE and determination coefficient R 2 As the evaluation criteria, if the MSE value and R 2 If the values are all within the preset range, the XGBoost model is judged to be qualified, and the qualified XGBoost model is used to predict the converter temperature; otherwise, it is unqualified, and the operations of step S2 to step S5 are repeated; S6: Use Fluent finite element software to establish a multiphase flow model, specifically: Apply boundary conditions according to the working conditions of step S1, and repeatedly check the actual process results in step S1 with the simulated multiphase flow field results until the applied boundary conditions make the simulation results consistent with the actual process results; S7: according to the multiphase flow model established in step S6, different converter endpoint temperatures are set to simulate the temperature field of the molten steel in the converter, and the molten steel temperature field data is exported; S8: Predict the converter endpoint temperature value according to the qualified XGBoost model in step S5, select the corresponding molten steel temperature field data in step S7, and use matplotlib in the python environment to build a temperature field cloud map. The multiphase flow model established in step S6 is: , Where: u, v, w are velocity vectors along the x, y, z directions, respectively; t is time; ρ is density; p is pressure; Fx, Fy, Fz are external forces on the unit volume of fluid in the x, y, z directions, respectively; μ is dynamic viscosity; The heat conduction equation is: , Where: T is temperature, °C; t is time, s; k is thermal conductivity, W / m·K; ρ is density, kg / m 3 ; C p is the specific heat, J / Kg•K.
2. The method for predicting converter endpoint temperature based on XGBoost and finite element combined drive according to claim 1 is characterized in that: In step S5, the XGBoost model is set to the tree model gbtree, where: The maximum depth of the tree is set to 10. The learning rate cta is set to 0.056 in the model; The number of gbtree classifiers is set to 1000; The setting range of MSE value is less than 20, R 2 The setting range of the value is greater than 0.8, and the ratio of the converter temperature prediction value to the actual value within ±10℃ is greater than 90%.
3. The method for predicting converter endpoint temperature based on XGBoost and finite element combined drive according to claim 1 is characterized in that: In step S6, the initial boundary conditions are set as follows in the Fluent finite element method to establish the multiphase flow model: The physical model is set according to the actual converter dimensions; In the physical model-model module, set the material physical property parameters, set the time to steady state, select the energy equation, use the k-ε turbulence model for the viscosity model, and use the VOF model for the multiphase flow field model; In the physical model-material module, set the density and viscosity of the fluid molten steel and oxygen, and the density, thermal conductivity, and specific heat capacity of the solid converter lining and shell materials; Set the heat transfer coefficient between molten steel and furnace lining and the heat transfer coefficient between furnace shell and ambient air in the physical model-boundary conditions; In the initialization conditions of the physical model-solution module, set the molten steel temperature to the converter end temperature.
4. The method for predicting converter endpoint temperature based on XGBoost and finite element combined drive according to claim 1, characterized in that: In step S7, different converter endpoint temperatures are set, the converter endpoint temperature range is 1500~1800℃, the setting value interval is n℃, and there are 300 / n+1 converter endpoint temperatures in total. Each converter endpoint temperature is substituted into the Fluent finite element multiphase flow model in step S6 to simulate the molten steel temperature field and export XYZ type data.
