Rh decarburization parameter prediction method based on coupling of data-driven algorithm and mechanism model

Abstract

The present application belongs to the technical field of steel metallurgy, and relates to a RH decarburization parameter prediction method based on coupling of a data-driven algorithm and a mechanism model, which comprises the following steps: collecting actual industrial parameters of a RH refining process; establishing a mechanism model; setting an initial value range of fitting parameters in the mechanism model; sampling to generate multiple groups of parameters in the initial value range, inputting the mechanism model for calculation to obtain an initial data set of carbon and oxygen mass fractions changing with time; preprocessing the initial data set; training a data-driven algorithm by using the preprocessed data; optimizing hyperparameters of the data-driven algorithm to obtain an optimal data-driven algorithm; inputting the industrial measured carbon and oxygen mass fractions into the optimal data-driven algorithm to output the optimal value of the predicted fitting parameters. The method has the beneficial effect that, through coupling of the mechanism model and the data-driven algorithm, bidirectional optimization of fitting parameter prediction and decarburization curve simulation is realized, and the prediction accuracy is significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of iron and steel metallurgy technology, specifically relating to a method for predicting RH decarburization parameters based on the coupling of data-driven algorithms and mechanism models. Background Technology

[0002] Ultra-low carbon steel is widely used in the automotive, construction, and home appliance industries due to its high strength and excellent deep-drawing performance. The RH refining process is a key step in the production of low-carbon and ultra-low carbon steel, such as... Figure 2 As shown, its core principle is to achieve efficient decarbonization by utilizing the carbon-oxygen reaction under vacuum conditions.

[0003] Currently, various RH (Ruhrstahl-Heraeus) decarburization models are used to predict the change of carbon content in molten steel over time. Traditional mechanistic models are mainly based on metallurgical reaction kinetics and mass transfer theory, comprehensively considering the carbon-oxygen reaction and mass transfer processes in multiphase media such as molten steel, slag layer, and vacuum chamber. They calculate the decarburization rate and the carbon content in molten steel during the decarburization process through decarburization thermodynamic equilibrium equations and decarburization kinetic equations. Examples include kinetic models based on mass conservation and reaction interface mass transfer models. However, due to the complexity of operating parameters and dynamic changes in operating conditions during the RH refining process, pure mechanistic models cannot accurately reflect the unsteady-state characteristics of actual production. Mechanistic models constructed under fixed vacuum conditions and oxygen blowing times are difficult to adapt to the flexible and varied operational requirements on site.

[0004] Meanwhile, some empirical formulas or models are also used to estimate RH decarburization parameters. These methods generally use regression analysis on a large amount of historical smelting data to provide a qualitative relationship between carbon content and process fitting parameters. Empirical models can quickly predict results under limited operating conditions, but they rely on sample data and empirical coefficients, lack physical interpretation, and do not have good generalization ability. Pure data-driven algorithms (such as neural networks, SVR, decision trees, etc.) use machine learning to fit refining endpoint temperatures, compositions, etc., for predicting the composition and temperature of RH vacuum refining molten steel. Although these methods can capture the patterns hidden in the data, their predictive ability for new and abnormal operating conditions is limited due to the lack of metallurgical mechanism constraints. Summary of the Invention

[0005] Technical problems to be solved

[0006] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a method for predicting RH decarbonization parameters based on the coupling of data-driven algorithms and mechanism models, which solves the problems of existing RH vacuum refining decarbonization process parameter prediction relying on empirical formulas, insufficient accuracy of calculation results, and inability to adapt to complex working conditions.

[0007] Technical solution

[0008] To achieve the above objectives, the main technical solutions adopted by the present invention include:

[0009] This invention provides a method for predicting RH decarbonization parameters based on the coupling of data-driven algorithms and mechanistic models, comprising the following steps:

[0010] Step 1: Collect actual industrial parameters for the RH refining process;

[0011] Step 2: Establish a mechanism model based on the characteristics of the RH vacuum refining process;

[0012] Step 3: Set the initial range of the fitting parameters in the mechanistic model;

[0013] Step 4: Sample and generate multiple sets of parameters within the initial value range, input them into the mechanism model for calculation, and obtain an initial dataset of carbon and oxygen mass fraction changes over time;

[0014] Step 5: Preprocess the initial dataset;

[0015] Step 6: Train the data-driven algorithm using the preprocessed data;

[0016] Step 7: Optimize the hyperparameters of the data-driven algorithm to obtain the optimal data-driven algorithm;

[0017] Step 8: Input the industrially measured carbon and oxygen mass fractions into the optimal data-driven algorithm, and output the optimal values ​​of the predicted fitting parameters.

[0018] As a further improvement of the present invention, the mechanism model includes actual industrial parameters;

[0019] The actual industrial parameters include: steel composition when entering RH, steel temperature when entering RH, ladle capacity, vacuum level in vacuum chamber, circulation flow rate, oxygen blowing flow rate, and argon blowing time.

[0020] As a further improvement of the present invention, step 2, establishing the mechanism model, includes the following steps:

[0021] Step 2.1: Divide the RH system into a ladle molten steel zone, a vacuum chamber molten steel zone, and a vacuum chamber gas zone;

[0022] Step 2.2: Based on the assumptions that the molten steel is completely mixed, the reaction occurs only in the vacuum chamber, and the rate is controlled by mass transfer, establish a set of governing equations for the carbon-oxygen reaction and mass transfer.

[0023] As a further improvement of the present invention, the set of control equations includes: the carbon content variation equation in the ladle molten steel zone, the oxygen content variation equation in the ladle molten steel zone, the carbon content variation equation in the vacuum chamber molten steel zone, the oxygen content variation equation in the vacuum chamber molten steel zone under oxygen blowing conditions, the carbon-oxygen balance relationship at the gas-liquid interface, the stoichiometric relationship of the carbon-oxygen reaction, and the molten steel circulation flow rate equation.

[0024] The change in carbon content in the molten steel zone of the ladle is expressed by the following formula:

[0025]

[0026] The change in oxygen content in the molten steel zone of the ladle is expressed by the following formula:

[0027]

[0028] The change in carbon content in the molten steel zone of the vacuum chamber is expressed by the following formula:

[0029]

[0030] The change in oxygen content in the molten steel zone of the vacuum chamber is expressed by the following formula:

[0031]

[0032] Under oxygen blowing conditions, the change in oxygen content in the molten steel zone of the vacuum chamber can be expressed by the following formula:

[0033]

[0034] The equilibrium relationship between carbon and oxygen at the gas-liquid interface is expressed by the following equation:

[0035]

[0036] The stoichiometric relationship of the carbon-oxygen reaction is expressed by the following equation:

[0037]

[0038] The circulation flow rate of molten steel is expressed by the following formula:

[0039]

[0040] In the formula, W L and w represent the mass of molten steel in the ladle and vacuum chamber, respectively, in kg; Q represents the molten steel circulation flow rate, in kg / s; w[C] V and w[C] L The mass fractions of dissolved carbon in the molten steel in the vacuum chamber and ladle are respectively expressed as 10. -4 % represents; w[O] V and w[O] LThe mass fractions of dissolved oxygen in the molten steel in the vacuum chamber and ladle are respectively expressed as 10⁻⁶. -4 % indicates w[C] S The mass fraction of dissolved carbon at chemical reaction equilibrium is expressed in units of 10. -4 % represents; w[O] S The mass fraction of dissolved oxygen at chemical equilibrium is expressed in units of 10. -4 % represents; ρ l This indicates the density of molten steel, expressed in kg / m³. 3 ; 'a' represents the effective area of ​​the gas-molten steel interface, in meters (m²). 2 ;k C and k O These represent the mass transfer coefficients of carbon and oxygen, respectively, in units of m. 3 / s;P CO P1 represents the partial pressure of carbon monoxide in the furnace gas of the RH vacuum chamber, in Pa; P2 represents the vacuum chamber pressure, in Pa; T represents the temperature of the molten steel, in K; M C M represents the molar mass of carbon, expressed in kg / mol. O The value of Q represents the molar mass of oxygen, expressed in kg / mol; D represents the inner diameter of the impregnation tube, expressed in m; Q represents the inner diameter of the tube. g This indicates the argon flow rate, expressed in L / min.

[0041] As a further improvement of the present invention, the fitting parameters include the carbon volumetric mass transfer coefficient ak. C oxygen volumetric mass transfer coefficient ak O , and oxygen absorption efficiency β under forced decarbonization conditions;

[0042] The initial range of the fitting parameters is determined based on one or more of the following: cross-sectional area of ​​the vacuum chamber container, surface area of ​​molten steel, circulation flow rate, stirring power, and ladle capacity.

[0043] The initial range of the fitting parameters is determined by the following formula:

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051] In the formula, S is the surface area of ​​the molten steel in the vacuum container, in m². 2 G represents the circulating flow rate of molten steel, in meters per second (m³). 3 / s; ɛ represents the stirring power in the degassing container, in W / t; W is the ladle capacity, in t; a is a constant representing the effective gas-liquid contact area, with a value of 10A; A is the cross-sectional area of ​​the vacuum container, in m². 2 .

[0052] As a further improvement of the present invention, in step 4, the Latin hypercube sampling method is used to generate N sets of parameter combinations within the initial value range of the fitting parameters; each set of parameter combinations is input into the mechanism model to calculate the data of carbon mass fraction and oxygen mass fraction in molten steel changing with time;

[0053] Based on the time-varying data, the carbon mass fraction and oxygen mass fraction at multiple specific time points under at least one operating condition are extracted to construct an initial dataset for training the data-driven algorithm.

[0054] As a further improvement to the method of the present invention, the preprocessing in step 5 includes normalization, outlier removal, and k-fold cross-validation to divide the training set and the validation set.

[0055] As a further improvement to the method of the present invention, in step 7, a combination of random search and cross-validation is used to optimize the hyperparameters.

[0056] As a further improvement to the method of the present invention, step 9 is also included: substituting the optimal value of the fitting parameters back into the mechanism model, calculating the carbon and oxygen concentration change curves, verifying by comparing the mean square error with the measured data, and selecting the data-driven algorithm model with the smallest prediction error as the final model.

[0057] The beneficial effects of this invention are:

[0058] Improved accuracy: By coupling the mechanistic model with the data-driven algorithm, the two-way optimization of fitting parameter prediction and decarbonization curve simulation is achieved, which significantly improves the prediction accuracy compared with simple empirical formulas.

[0059] High adaptability: By using Latin hypercube sampling, N sets of parameter combinations are generated within the initial value range of the fitting parameters. The time series data of carbon and oxygen mass fractions in molten steel are calculated by inputting them into the mechanism model. The carbon and oxygen mass fractions at multiple specific time points under at least one working condition are extracted to construct the initial dataset. Only a small amount of carbon and oxygen concentration information at a time point is needed to quickly deduce the mechanism fitting parameters. It is adaptable to different working conditions such as natural decarburization and forced decarburization, and is also suitable for situations where oxygen and carbon are not synchronized, or where the number of oxygen and carbon decarburization cycles is not equal.

[0060] Good real-time performance: The data-driven algorithm learns the inverse mapping of the mechanism model, which can quickly predict key parameters on site and meet the real-time needs of industrial production.

[0061] Versatility: The method is not only applicable to the RH vacuum refining decarburization process, but can also be extended to the prediction of reaction kinetic parameters in other metallurgical processes. Attached Figure Description

[0062] Figure 1 This is a flowchart illustrating a method for predicting RH decarbonization parameters based on the coupling of data-driven algorithms and mechanism models, provided in an embodiment of the present invention.

[0063] Figure 2 This is a schematic diagram illustrating the principles of natural decarbonization and forced decarbonization in embodiments of the present invention;

[0064] Figure 3 This is a curve comparing the measured data and simulated values ​​of carbon mass fraction in the ladle under natural decarburization conditions in this embodiment of the invention.

[0065] Figure 4 This is a curve comparing the measured data and simulated values ​​of oxygen mass fraction in the ladle under natural decarburization conditions in this embodiment of the invention.

[0066] Figure 5 This is a curve comparing the measured data and simulated values ​​of carbon mass fraction in the ladle under forced decarburization conditions in this embodiment of the invention.

[0067] Figure 6 This is a curve comparing the measured data and simulated values ​​of oxygen mass fraction in the ladle under forced decarburization conditions in this embodiment of the invention. Detailed Implementation

[0068] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0069] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.

[0070] The purpose of this invention is to provide a method for predicting RH decarburization parameters based on the coupling of data-driven algorithms and mechanistic models. This method combines modern machine learning algorithms with classical metallurgical decarburization mechanism models, utilizing data-driven approaches to correct and optimize mechanistic parameters, thereby significantly improving the prediction accuracy of key RH decarburization parameters and broadening their applicability. Simultaneously, this invention enhances the model's responsiveness to dynamic RH decarburization conditions, enabling simultaneous prediction of carbon and oxygen content in molten steel.

[0071] like Figure 1 As shown, a method for predicting RH decarbonization parameters based on the coupling of data-driven algorithms and mechanistic models includes the following steps:

[0072] Step 1: Collect actual industrial parameters for the RH refining process;

[0073] Specifically, the actual industrial parameters involved in the RH steelmaking process are collected and the data is analyzed.

[0074] The mechanistic model includes actual industrial parameters;

[0075] The actual industrial parameters include: the composition of the molten steel when entering the RH chamber, the temperature of the molten steel when entering the RH chamber, the capacity of the ladle, the vacuum level of the vacuum chamber, the circulation flow rate, the oxygen blowing flow rate, and the argon blowing time.

[0076] Step 2: Establish a mechanism model based on the characteristics of the RH vacuum refining process;

[0077] Specifically, based on the characteristics of the RH vacuum refining process, a mechanism model for carbon-oxygen reaction and mass transfer is established;

[0078] Step 2.1: Divide the RH system into a ladle molten steel zone, a vacuum chamber molten steel zone, and a vacuum chamber gas zone;

[0079] Specifically, due to the thorough and instantaneous nature of steel molten mixing, carbon and oxygen in the molten steel are evenly distributed in the ladle and vacuum chamber; the vacuum chamber is the only place for the decarburization reaction; the mass fraction of carbon and oxygen in the molten steel is in equilibrium with the partial pressure of CO gas at the gas-steel interface; the carbon-oxygen reaction in the molten steel is a rapid chemical reaction, and the decarburization rate depends on the mass transfer of carbon and oxygen in the molten steel.

[0080] Step 2.2: Based on the assumptions that the molten steel is completely mixed, the reaction occurs only in the vacuum chamber, and the rate is controlled by mass transfer, establish a set of governing equations for the carbon-oxygen reaction and mass transfer.

[0081] The set of control equations includes: the equation for the change of carbon content in the ladle molten steel zone, the equation for the change of oxygen content in the ladle molten steel zone, the equation for the change of carbon content in the vacuum chamber molten steel zone, the equation for the change of oxygen content in the vacuum chamber molten steel zone under oxygen blowing conditions, the carbon-oxygen balance relationship at the gas-liquid interface, the stoichiometric relationship of the carbon-oxygen reaction, and the equation for the steel circulation flow rate.

[0082] The change in carbon content in the molten steel zone of the ladle is expressed by the following formula:

[0083]

[0084] The change in oxygen content in the molten steel zone of the ladle is expressed by the following formula:

[0085]

[0086] The change in carbon content in the molten steel zone of the vacuum chamber is expressed by the following formula:

[0087]

[0088] The change in oxygen content in the molten steel zone of the vacuum chamber is expressed by the following formula:

[0089]

[0090] Under oxygen blowing conditions, the change in oxygen content in the molten steel zone of the vacuum chamber can be expressed by the following formula:

[0091]

[0092] The equilibrium relationship between carbon and oxygen at the gas-liquid interface is expressed by the following equation:

[0093]

[0094] The stoichiometric relationship of the carbon-oxygen reaction is expressed by the following equation:

[0095]

[0096] The circulation flow rate of molten steel is expressed by the following formula:

[0097]

[0098] In the formula, W L and w represent the mass of molten steel in the ladle and vacuum chamber, respectively, in kg; Q represents the molten steel circulation flow rate, in kg / s; w[C] V and w[C]L The mass fractions of dissolved carbon in the molten steel in the vacuum chamber and ladle are respectively expressed as 10. -4 % represents; w[O] V and w[O] L The mass fractions of dissolved oxygen in the molten steel in the vacuum chamber and ladle are respectively expressed as 10⁻⁶. -4 % indicates w[C] S The mass fraction of dissolved carbon at chemical reaction equilibrium is expressed in units of 10. -4 % represents; w[O] S The mass fraction of dissolved oxygen at chemical reaction equilibrium is expressed in units of 10. -4 % represents; ρ l This indicates the density of molten steel, expressed in kg / m³. 3 ; 'a' represents the effective area of ​​the gas-molten steel interface, in meters (m²). 2 ;k C and k O The mass transfer coefficient between carbon and oxygen is expressed in m. 3 / s;P CO P1 represents the partial pressure of carbon monoxide in the furnace gas of the RH vacuum chamber, in Pa; P2 represents the vacuum chamber pressure, in Pa; T represents the temperature of the molten steel, in K; M C M represents the molar mass of carbon, expressed in kg / mol. O The value of Q represents the molar mass of oxygen, expressed in kg / mol; D represents the inner diameter of the impregnation tube, expressed in m; Q represents the inner diameter of the tube. g This indicates the argon flow rate, expressed in L / min.

[0099] Step 3: Set the initial range of the fitting parameters in the mechanistic model;

[0100] Specifically, the fitting parameters include the carbon volumetric mass transfer coefficient ak. C oxygen volumetric mass transfer coefficient ak O And the oxygen absorption efficiency β under forced decarbonization conditions;

[0101] The initial range of the fitting parameters is determined based on one or more of the following: cross-sectional area of ​​the vacuum chamber container, surface area of ​​molten steel, circulation flow rate, stirring power, and ladle capacity.

[0102] The initial range of the fitting parameters is determined by the following formula:

[0103] (9)

[0104] (10)

[0105] (11)

[0106] (12)

[0107] (13)

[0108] (14)

[0109] (15)

[0110] In the formula, S is the surface area of ​​the molten steel in the vacuum container, in m². 2 G represents the circulating flow rate of molten steel, in meters per second (m³). 3 / s; ɛ represents the stirring power in the degassing container, in W / t; W is the ladle capacity, in t; a is a constant representing the effective gas-liquid contact area, with a value of 10A; A is the cross-sectional area of ​​the vacuum container, in m². 2 .

[0111] As shown in Table 1, the initial values ​​and ranges of the fitting parameters are obtained based on the 280 t RH industrial data and formulas (9)-(15).

[0112] Table 1 Calculation of fitting parameters for the RH decarbonization process

[0113]

[0114] Table 1 shows that there are two key fitting parameters for the natural decarbonization process: the carbon volumetric mass transfer coefficient ak. C oxygen volumetric mass transfer coefficient ak O Forced decarbonization requires increasing oxygen uptake efficiency β, which is ak. C ak O and β.

[0115] In this embodiment, the initial value range of the key fitting parameters is determined as follows:

[0116] During natural decarbonization: ak C =0.1~0.6m 3 / s;ak O =0.01∼0.4 m 3 / s.

[0117] Under forced decarbonization conditions: ak C =0.1~0.5m 3 / s ;ak O =0.05~0.3 m 3 / s; β=0.3∼0.8.

[0118] Step 4: Sample and generate multiple sets of parameters within the initial value range, input them into the mechanism model for calculation, and obtain an initial dataset of carbon and oxygen mass fraction changes over time;

[0119] Specifically, within a given initial range of fitting parameters, N sets of parameters (e.g., N = 10000) are generated using Latin hypercube (LHS) sampling and input into the constructed mechanistic model. Multiple calculations are performed using the mechanistic model to obtain data on the changes in carbon and oxygen mass fractions in molten steel over time. Based on the data on changes over time, carbon and oxygen mass fractions at multiple specific time points under at least one operating condition are extracted to construct an initial dataset for training the data-driven algorithm.

[0120] In this embodiment, the mechanistic model uses ak C and ak O (Forced decarbonization increases β) as input, with the carbon content w[C] at two sampling times. t1 w[C] t2 And the oxygen content w[O] at the three constant oxygen times. t1 w[O] t2 w[O] t3 As the output, the initial dataset is constructed. Only a small amount of carbon and oxygen concentration information at a few time points is needed to quickly deduce the mechanism fitting parameters. It is adaptable to different working conditions such as natural decarbonization and forced decarbonization, and is also suitable for situations where oxygen and carbon determination are asynchronous or the number of oxygen and carbon determinations is unequal. In particular, the two sampling times and the three oxygen determination times can be different, and the number of sampling and oxygen determinations can also be different. Any combination can be used for training and prediction.

[0121] Step 5: Preprocess the initial dataset;

[0122] Specifically, the data in the initial dataset are normalized, outlier removal is performed, and k-fold cross-validation is used to divide the training set and validation set.

[0123] In this implementation, the training set and the validation set are divided in an 8:2 ratio.

[0124] Step 6: Train the data-driven algorithm using the preprocessed data;

[0125] Using the preprocessed data from the initial dataset output by the mechanistic model as input, the data-driven algorithm is trained to infer the key fitting parameters of the mechanistic model.

[0126] Specifically, in this step, the input to the data-driven algorithm is the carbon mass fraction w[C] at two different sampling times measured in industrial practice. t1 w[C] t2 The oxygen mass fraction w[O] at three constant oxygen times. t1 w[O] t2 w[O] t3 ;

[0127] The output consists of the fitting parameters for the mechanistic model, including:

[0128] During natural decarbonization: ak C and ak O ;

[0129] Under forced decarbonization conditions: ak C ak O And oxygen absorption efficiency β.

[0130] The data-driven algorithm uses the time series values ​​output by the mechanistic model as input features, which include the carbon mass fraction w[C] at two different sampling times. t1 w[C] t2 And the oxygen mass fraction w[O] at three constant oxygen times. t1 w[O] t2 and w[O] t3 The output target is the fitting parameters of the mechanistic model: ak under natural decarbonization conditions. C and ak O Under forced decarbonization conditions, it is ak. C ak O And oxygen absorption efficiency β.

[0131] The intelligent model can use machine learning algorithms such as BPNN (Backpropagation Neural Network), XGBoost (eXtreme Gradient Boosting), decision tree, and random forest. The above input and output data are respectively fed into the three intelligent models of BPNN, XGBoost, and random forest for training.

[0132] Step 7: Optimize the hyperparameters of the data-driven algorithm to obtain the optimal data-driven algorithm;

[0133] Specifically, the optimal data-driven algorithm is obtained by optimizing hyperparameters (such as the number of neural network layers, tree model depth, etc.) through random search and cross-validation.

[0134] A multi-model parallel evaluation strategy is adopted, and the candidate model algorithms include: BPNN, XGBoost regression, and random forest regression.

[0135] Among them, random search combined with k-fold cross-validation (k=5) is used to search for key hyperparameters, and the models are evaluated on the validation set to determine the optimal hyperparameter configuration for each intelligent model.

[0136] If the optimal conditions are not met, the parameter range of the random search is readjusted or the number of cross-validation folds is increased before performing hyperparameter optimization again.

[0137] If the termination condition is met, the current optimized hyperparameter combination is set as the final parameter configuration of the data-driven algorithm.

[0138] In this embodiment, the optimal parameter settings for each intelligent model are obtained through hyperparameter search as follows:

[0139] The BPNN is configured with two hidden layers (150 nodes each), the activation function is ReLU, the optimizer is stochastic gradient descent, and the learning rate is 0.09.

[0140] XGBoost is set with a learning rate of 0.05, a maximum tree depth of 9, a subsampling rate of 0.8, and 190 weak learners.

[0141] The random forest contains 167 trees, has a maximum depth of 8, and uses a feature selection method of log2.

[0142] Step 8: Input the industrially measured carbon and oxygen mass fractions into the optimal data-driven algorithm, and output the optimal values ​​of the predicted fitting parameters;

[0143] The measured carbon and oxygen mass fractions from industrial settings are input into the data-driven algorithm, which outputs the predicted fitting parameters ak. C ak O The optimal values ​​of β and β.

[0144] The carbon mass fraction at two different sampling times and the oxygen mass fraction at three constant oxygen times measured in the industrial experiment were input into the data-driven intelligent algorithm with optimal parameters. The fitting parameters predicted by each algorithm model are shown in Table 2.

[0145] Table 2 Fitting parameters predicted by each intelligent model

[0146]

[0147] Step 9: Substitute the optimal values ​​of the fitted parameters back into the mechanism model to calculate the carbon and oxygen concentration change curves. Verify the model by comparing the mean square error with the measured data, and select the data-driven algorithm model with the smallest prediction error as the final model.

[0148] Specifically, to ensure the accuracy of the fitted parameter predictions, the optimal values ​​of the fitted parameters are substituted back into the mechanistic model to calculate the carbon and oxygen concentration change curves over time. The mean square error (MSE) is calculated and compared with experimental and industrial measured data for verification. The data-driven algorithm model with the smallest prediction error is selected as the final model.

[0149] The carbon-oxygen concentration curve mentioned here was obtained by establishing a set of governing equations for the mass fraction of carbon and oxygen in molten steel and by numerical integration using the fourth-order Runge–Kutta method.

[0150] In this embodiment, the optimal data-driven algorithm is the Random Forest algorithm, and its best-fit parameters are as follows:

[0151] During natural decarbonization: ak C ≈0.1738m 3 / s, ak O ≈0.1132m 3 / s; Under forced decarbonization conditions: ak C ≈0.1417m 3 / s, ak O ≈0.0984m 3 / s, β≈0.5688.

[0152] Substitute the fitting parameters obtained from formulas (9)-(15) and the fitting parameters predicted by the intelligent model into the mechanism model, calculate the carbon and oxygen content curves, and display and compare them to evaluate the degree of improvement of the present invention compared with the traditional empirical formula calculation.

[0153] like Figure 3 and Figure 4 The figure shows a comparison curve between the measured and simulated values ​​of carbon mass fraction and oxygen mass fraction under natural decarbonization conditions.

[0154] like Figure 5 and Figure 6 The figure shows a comparison curve between the measured and simulated values ​​of carbon mass fraction and oxygen mass fraction under forced decarbonization conditions.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.

Claims

1. A method for predicting RH decarbonization parameters based on the coupling of data-driven algorithms and mechanistic models, characterized in that, Includes the following steps: Step 1: Collect actual industrial parameters for the RH refining process; The actual industrial parameters include: molten steel composition when entering RH, molten steel temperature when entering RH, ladle capacity, vacuum level of vacuum chamber, circulation flow rate, oxygen blowing flow rate, and argon blowing time; Step 2: Establish a mechanism model based on the characteristics of the RH vacuum refining process; The mechanism model includes actual industrial parameters. Step 2, establishing the mechanism model, includes the following steps: Step 2.1: Divide the RH system into a ladle molten steel zone, a vacuum chamber molten steel zone, and a vacuum chamber gas zone; Step 2.2: Based on the assumptions that the molten steel is completely mixed, the reaction occurs only in the vacuum chamber, and the rate is controlled by mass transfer, establish a set of governing equations for the carbon-oxygen reaction and mass transfer; Step 3: Set the initial range of the fitting parameters in the mechanistic model; The fitting parameters include the carbon volumetric mass transfer coefficient ak. C oxygen volumetric mass transfer coefficient ak O , and oxygen absorption efficiency β under forced decarbonization conditions; The initial range of the fitting parameters is determined based on one or more of the following: cross-sectional area of ​​the vacuum chamber container, surface area of ​​molten steel, circulation flow rate, stirring power, and ladle capacity. The initial range of the fitting parameters is determined by the following formula: ； ； ； ； ； ； ； In the formula, Q represents the circulating flow rate of molten steel, with units of kg / s; w[C] V The mass fraction of carbon dissolved in molten steel in a vacuum chamber is expressed in 10⁻⁶. -4 % represents; w[O] V The mass fraction of dissolved oxygen in molten steel in a vacuum chamber is expressed as 10⁻⁶. -4 % represents the surface area of ​​the molten steel in the vacuum container, in meters. 2 G represents the circulating flow rate of molten steel, in meters per second (m³). 3 / s; ɛ represents the stirring power in the degassing container, in W / t; W is the ladle capacity, in t; a is a constant representing the effective gas-liquid contact area, with a value of 10A; A is the cross-sectional area of ​​the vacuum container, in m². 2 P2 represents the vacuum chamber pressure, in Pa. Step 4: Sample and generate multiple sets of parameters within the initial value range, input them into the mechanism model for calculation, and obtain an initial dataset of carbon and oxygen mass fraction changes over time; In step 4, the Latin hypercube sampling method is used to generate N sets of parameter combinations within the initial value range of the fitting parameters; each set of parameter combinations is input into the mechanism model to calculate the data on the change of carbon mass fraction and oxygen mass fraction in molten steel over time. Based on the time-varying data, the carbon mass fraction and oxygen mass fraction at multiple specific time points under at least one working condition are extracted to construct an initial dataset for training the data-driven algorithm. Step 5: Preprocess the initial dataset; Step 6: Train the data-driven algorithm using the preprocessed data; Step 7: Optimize the hyperparameters of the data-driven algorithm to obtain the optimal data-driven algorithm; Step 8: Input the industrially measured carbon and oxygen mass fractions into the optimal data-driven algorithm, and output the optimal values ​​of the predicted fitting parameters.

2. The RH decarbonization parameter prediction method based on the coupling of data-driven algorithms and mechanism models according to claim 1, characterized in that: The set of control equations includes: the equation for the change of carbon content in the ladle molten steel zone, the equation for the change of oxygen content in the ladle molten steel zone, the equation for the change of carbon content in the vacuum chamber molten steel zone, the equation for the change of oxygen content in the vacuum chamber molten steel zone under oxygen blowing conditions, the carbon-oxygen balance relationship at the gas-liquid interface, the stoichiometric relationship of the carbon-oxygen reaction, and the equation for the steel circulation flow rate. The change in carbon content in the molten steel zone of the ladle is expressed by the following formula: ； The change in oxygen content in the molten steel zone of the ladle is expressed by the following formula: ； The change in carbon content in the molten steel zone of the vacuum chamber is expressed by the following formula: ； The change in oxygen content in the molten steel zone of the vacuum chamber is expressed by the following formula: ； Under oxygen blowing conditions, the change in oxygen content in the molten steel zone of the vacuum chamber can be expressed by the following formula: ； The equilibrium relationship between carbon and oxygen at the gas-liquid interface is expressed by the following equation: ； The stoichiometric relationship of the carbon-oxygen reaction is expressed by the following equation: ； The circulation flow rate of molten steel is expressed by the following formula: ； In the formula, W L and w represent the mass of molten steel in the ladle and vacuum chamber, respectively, in kg; Q represents the molten steel circulation flow rate, in kg / s; w[C] V and w[C] L The mass fractions of dissolved carbon in the molten steel in the vacuum chamber and ladle are respectively expressed as 10. -4 % represents; w[O] V and w[O] L The mass fractions of dissolved oxygen in the molten steel in the vacuum chamber and ladle are respectively expressed as 10⁻⁶. -4 % indicates w[C] S The mass fraction of dissolved carbon at chemical reaction equilibrium is expressed in units of 10. -4 % represents; w[O] S The mass fraction of dissolved oxygen at chemical equilibrium is expressed in units of 10. -4 % represents; ρ l This indicates the density of molten steel, expressed in kg / m³. 3 ; 'a' represents the effective area of ​​the gas-molten steel interface, in meters (m²). 2 ;k C and k O These represent the mass transfer coefficients of carbon and oxygen, respectively, in units of m. 3 / s;P CO P1 represents the partial pressure of carbon monoxide in the furnace gas of the RH vacuum chamber, in Pa; P2 represents the vacuum chamber pressure, in Pa; T represents the temperature of the molten steel, in K; M C M represents the molar mass of carbon, expressed in kg / mol. O The value of Q represents the molar mass of oxygen, expressed in kg / mol; D represents the inner diameter of the impregnation tube, expressed in m; Q represents the inner diameter of the tube. g This indicates the argon flow rate, expressed in L / min.

3. The RH decarbonization parameter prediction method based on the coupling of data-driven algorithms and mechanism models according to claim 1, characterized in that: The preprocessing in step 5 includes normalization, outlier removal, and k-fold cross-validation to divide the training set and validation set.

4. The RH decarbonization parameter prediction method based on the coupling of data-driven algorithms and mechanism models according to claim 1, characterized in that: In step 7, the hyperparameters are optimized using a combination of random search and cross-validation.

5. The RH decarbonization parameter prediction method based on the coupling of data-driven algorithms and mechanism models according to claim 1, characterized in that: The method also includes step 9: substituting the optimal value of the fitting parameters back into the mechanism model, calculating the carbon and oxygen concentration change curves, verifying the results by comparing the mean square error with the measured data, and selecting the data-driven algorithm model with the smallest prediction error as the final model.

Images