A method and system for constructing a prediction model of decarburization reaction kinetics in a converter
By constructing a coupled network model and introducing control parameters, the problems of disconnection and coupling effects in the converter steelmaking model were solved, enabling precise dynamic control and optimization of the converter decarburization reaction and improving the model's adaptability and accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- AUTOMATION RES & DESIGN INST OF METALLURGICAL IND
- Filing Date
- 2026-02-11
- Publication Date
- 2026-06-05
AI Technical Summary
Existing converter steelmaking models suffer from several problems, including a disconnect between the mechanistic model and actual control, a lack of process guidance in data-driven models, an inability to form closed-loop optimization, and insufficient modeling of the coupling effects of multiple control parameters. These issues result in inadequate accuracy, interpretability, and online adaptability of the models.
A coupled network model based on BP neural network, random forest model and XGBoost model is constructed. Combined with the decarburization reaction kinetics theory, control parameters such as lance height, molten iron quality and slagging agent lime quality are introduced. Through iterative optimization, a converter decarburization reaction kinetics prediction model is established, forming a closed-loop optimization mechanism of prediction-feedback-correction.
It enables the model to directly respond to operational commands, improves the accuracy and long-term adaptability of decarbonization rate prediction, and can adapt to raw material fluctuations and equipment aging, ensuring the model's continued leading performance.
Smart Images
Figure CN122154422A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent control and modeling technology for iron and steel metallurgical processes, and in particular to a method and system for constructing a predictive model of converter decarburization reaction kinetics. Background Technology
[0002] Converter steelmaking is a nonlinear physicochemical reaction process carried out under multi-phase, high-temperature conditions. During the blowing process, it is difficult to obtain accurate and continuous dynamic information in real time, such as changes in temperature and composition. As a key process in steel production, the core challenge of converter smelting lies in achieving dynamic process control. Currently, actual production mainly focuses on controlling the final composition and temperature. The basic requirement is to reduce the carbon content to within the required range for the steel grade at the end of oxygen blowing, remove harmful elements such as phosphorus, sulfur, and manganese, and ensure the temperature reaches the tapping standard.
[0003] In related technologies, converter process prediction models are key to optimizing production and improving operational accuracy and efficiency. However, existing models generally suffer from the following problems: First, the mechanistic model is disconnected from actual control; traditional kinetic model parameters rely heavily on empirical constants, making it difficult to correct them online using actual controllable variables. Second, data-driven models lack process guidance capabilities and fail to effectively establish the engineering correlation between carbon content and operating parameters. Third, they cannot form a closed-loop logic of prediction, feedback, and correction, lacking a mechanism to continuously optimize model parameters using measured TSC / TSO data, resulting in poor adaptability and difficulty in coping with changes in operating conditions caused by raw material fluctuations and equipment aging. Fourth, there is insufficient modeling of the coupling effects between multiple control parameters; factors are treated in isolation, and their quantitative relationship with kinetic coefficients is not established, leading to a lack of uniformity in accuracy, interpretability, operational guidance, and online adaptability. Summary of the Invention
[0004] Based on the above analysis, the present invention aims to provide a method and system for constructing a predictive model of converter decarburization reaction kinetics, in order to solve the problems of existing technical models being disconnected from actual control, lacking process guidance, being unable to achieve closed-loop optimization, and ignoring the coupling effects of multiple control parameters.
[0005] On one hand, embodiments of the present invention provide a method for constructing a predictive model of converter decarburization reaction kinetics, including: Based on valid sample data of the converter process in steel enterprises, a TSO prediction model for the final steel composition at the converter stage is constructed. Based on the actual production data of each heat and the TSO prediction model for molten steel composition, the predicted results of the TSO content of molten steel at the converter endpoint for each heat are obtained. The preset decarbonization reaction kinetics model is modified by introducing control parameters; the predicted TSO content of each item is used as the endpoint constraint to perform parameter iteration optimization on the preset decarbonization reaction kinetics model; and based on the modified decarbonization reaction kinetics model and the optimal parameter results, the optimized converter decarbonization reaction kinetics prediction model is obtained.
[0006] Furthermore, the input data in the effective sample data includes at least: oxygen supply time, total blowing time, carbon removal oxygen consumption, molten iron temperature entering the furnace, actual molten iron quantity, actual scrap steel quantity, lime quality, lightly calcined dolomite quality, pellet quality, dynamic pellet quality, and molten iron carbon content. The output data in the valid sample data includes at least the predicted content of C, Si, Mn and P elements at the converter endpoint.
[0007] Furthermore, the TSO prediction model is a coupled network model trained based on a BP neural network, a random forest model, and an XGBoost model; The prediction results of the TSO content of the converter steel at the end of each heat, based on the actual production data of each heat and the TSO prediction model, include: The actual production data of each furnace batch is input into the coupled network model, and the prediction results of the BP neural network, random forest model and XGBoost model are weighted and averaged to obtain the TSO content prediction results for each furnace batch.
[0008] Furthermore, the fundamental model of decarbonization reaction kinetics specifically includes calculation formulas for the overall decarbonization rate, the decarbonization rate in the jet impact zone, and the decarbonization rate in the emulsion zone; wherein, The formulas for calculating the overall decarbonization rate include: ; Where, overall represents the sum of all reaction zones; iz represents the jet impact zone; em represents the emulsion zone; W m For the quality of molten steel; C Cm m is the mass percentage of carbon; m is the mass. The formulas for calculating the decarburization rate in the jet impact zone include: ; Where [%C] represents the mass percentage of carbon in the molten steel; A iz V represents the area of the reaction interface in the jet impact zone. m For the quality of molten steel; To correct the oxygen mass transfer coefficient of carbon elements in the jet impact zone; This represents the mass percentage of carbon in the jet impact zone at equilibrium. The formulas for calculating the decarburization rate in the emulsion zone include: ; in, To correct the mass transfer coefficient of carbon in the emulsion region; A d m is the area of the metal droplets in the emulsion. d This indicates the mass of the metal droplets in the emulsion; ρ is the density of molten steel; m is the mass; This represents the percentage of carbon by mass at equilibrium.
[0009] Furthermore, the control parameters include at least: lance height h, molten iron mass M0, and lime mass M (the slag-forming agent). Lime ; The basic model of decarburization reaction kinetics is constructed based on the theory of decarburization reaction kinetics and is used to simulate the carbon element reaction kinetics process in the jet impact zone and emulsion zone of the converter decarburization reaction. The introduction of control parameters to modify the preset decarbonization reaction kinetics model includes: The oxygen mass transfer coefficient of carbon in the jet impact zone is corrected based on the lance height h and the mass of molten iron M0 entering the furnace; and based on the mass of lime, the slag-forming agent M... Lime The mass transfer coefficient of carbon in the emulsion region is corrected.
[0010] Furthermore, the correction formula for the oxygen mass transfer coefficient of carbon in the jet impact zone includes: ; in, The corrected oxygen mass transfer coefficient of carbon in the jet impact zone; is the diffusion coefficient of carbon in a liquid metal droplet; The circulating renewal rate of the molten steel in the pool under top blowing conditions is given by: Q = lance flow rate; g = gravitational acceleration; H = 1 / 2. bath r is the depth of the molten pool. cm Where M is the impact radius; M0 is the mass of molten iron entering the furnace; a is a constant; h is the lance height; b is the adjustment constant for the lance height h; This is the correction factor for carbon elements in the jet impact zone.
[0011] Furthermore, the correction formula for the mass transfer coefficient of carbon element in the emulsion region includes: ; in, This is the corrected mass transfer coefficient of carbon in the emulsion region; ρ is the diffusion coefficient of carbon in the liquid steel phase; u is the droplet velocity; d is the average diameter of the metal droplet; M LimeFor the quality of lime used as a slag-forming agent; This is an index used to adjust the quality of lime when slag-forming agents are added.
[0012] Furthermore, the step of using the predicted TSO content results as endpoint constraints to iteratively optimize the parameters of the preset decarbonization reaction kinetics model includes: The predicted TSO content for each heat is used as the endpoint constraint and substituted into the basic model of decarburization reaction kinetics. The percentage content of molten steel composition calculated at each time step is used as the initial value for the next time step. Compare the calculated percentage of molten steel composition at a specific measurement time for each heat, and the error between them and the corresponding true value of molten steel composition, until the optimal parameter result with the smallest error is obtained.
[0013] Furthermore, the method also includes: Based on the real-time control variables of each heat and the converter decarburization reaction kinetic prediction model, the predicted decarburization rate of each heat during the blowing process is calculated, and the carbon element content change curve corresponding to each heat is obtained based on the predicted decarburization rate. The predicted decarbonization rate includes at least the following: the decarbonization rate in the emulsion zone, the decarbonization rate in the jet impact zone, and the total decarbonization rate for each heat during the blowing process.
[0014] On the other hand, embodiments of the present invention provide a system for constructing a converter decarburization reaction kinetics prediction model, comprising: The TSO prediction model building module is used to build a TSO prediction model for the final steel composition of the converter based on valid sample data of the converter process in steel enterprises. The converter endpoint TSO prediction module is used to predict the TSO content of the molten steel at the converter endpoint for each heat based on the actual production data of each heat and the steel composition TSO prediction model. The kinetic model optimization module is used to introduce control parameters to modify the preset decarbonization reaction kinetic basic model; the predicted TSO content of each item is used as the endpoint constraint to perform parameter iterative optimization on the preset decarbonization reaction kinetic basic model, and the optimized converter decarbonization reaction kinetic prediction model is obtained based on the modified decarbonization reaction kinetic basic model and the optimal parameter results.
[0015] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects: First, unlike related technologies where the mechanism model is disconnected from actual control, this invention introduces the following parameters: lance height h, molten iron mass M0, and lime mass M (the slag-forming agent). LimeThe control parameters, which can be observed and adjusted in real time by on-site operators, can accurately capture the impact of minute changes in control parameters on the kinetic reaction of elements removed during converter smelting, and realize the model's direct response to operating commands.
[0016] Second, unlike related technologies which suffer from insufficient modeling of the coupling effects of multiple control parameters, this invention does not treat a single parameter in isolation, but incorporates multiple key factors into the calculation of the mass transfer coefficient simultaneously. This reflects the coupling effect of multiple control parameters on the decarbonization rate, accurately distinguishes decarbonization behavior under different operating conditions, and improves the accuracy of decarbonization rate prediction.
[0017] Third, unlike related technologies where data-driven models lack process guidance capabilities, this invention uses the TSO content prediction results output by the TSO prediction model to replace unrealistic thermodynamic equilibrium assumptions and uses them as the endpoint constraint of the kinetic differential equation, making the simulation path of the entire reaction closer to the actual blowing process; it retains the interpretability and extrapolation capability of the mechanism model, and inherits the fitting advantage of the data model for complex nonlinear relationships, forming a complementary advantage.
[0018] Fourth, unlike related technologies that cannot achieve closed-loop optimization, this invention establishes a "prediction-feedback-correction" closed loop, which significantly improves the long-term adaptability and robustness of the converter decarburization reaction kinetic prediction model. By using the newly acquired measured TSC / TSO data for each furnace, the correction coefficients are continuously updated iteratively, enabling the model to automatically adapt to long-term operating condition drifts such as fluctuations in raw material composition, equipment aging, and changes in operating habits. This ensures that each iteration can find the globally optimal or near-optimal parameter combination, thereby guaranteeing the model's continued leading performance.
[0019] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description
[0020] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Figure 1 is a flowchart of the method for constructing a converter decarburization reaction kinetics prediction model according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the converter steelmaking area according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the construction process of the BP neural network according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of the random forest model according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the main modules of the converter decarburization reaction kinetic prediction model construction system according to an embodiment of the present invention. Detailed Implementation
[0021] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.
[0022] A specific embodiment of the present invention discloses a method for constructing a predictive model of converter decarburization reaction kinetics, such as... Figure 1 As shown, it includes the following steps S1 to S3: Step S1: Based on the valid sample data of the converter process in steel enterprises, construct a TSO prediction model for the final steel composition of the converter.
[0023] In implementation, this invention combines data-driven modeling with kinetic mechanism modeling. First, a data-driven approach is used to accurately predict the carbon content at the converter endpoint, then reliable input is provided to the kinetic process model. It can be understood that the converter endpoint refers to the moment when converter blowing ends, at which point the composition and temperature of the molten steel must meet the tapping standards required for the target steel grade. TSO refers to the composition of the molten steel at the converter endpoint.
[0024] Specifically, in the construction phase of the TSO prediction model, firstly, the actual production data of the converter process of the steel enterprise is preprocessed to obtain effective sample data; then, based on the effective sample data, a data-driven modeling method is used to establish a basic model for predicting the content of each element at the converter endpoint, which is the TSO prediction model.
[0025] Preferably, during the training phase, the input data in the effective sample data includes at least: oxygen supply time, total blowing time, carbon removal oxygen consumption, molten iron temperature entering the furnace, actual molten iron quantity, actual scrap steel quantity, lime quality, lightly calcined dolomite quality, pellet quality, dynamic pellet quality, and molten iron carbon content; the output data in the effective sample data includes at least: the predicted content results of C, Si, Mn, and P elements at the converter endpoint.
[0026] In some implementations, the TSO prediction model is a coupled network model trained based on a BP neural network, a random forest model, and an XGBoost model.
[0027] For example, based on valid sample data, three base models—BP neural network, random forest, and XGBoost model—are trained respectively. Then, the three trained models are coupled through ensemble learning methods. The prediction results can be fused using a weighted average method to form the final coupled network model. The specific coupling process can be implemented by referring to relevant technical means such as ensemble learning frameworks in this field, which will not be elaborated here.
[0028] It can be understood that a BP (Back Propagation) neural network is a type of multi-layer feedforward neural network. Its model structure consists of an input layer, a hidden layer, and an output layer. Neurons in adjacent layers are fully connected, while neurons within the same layer are independent of each other. Specifically, in this embodiment, the BP neural network initially uses the 11 influencing factors from the aforementioned effective sample data as the data variables required for the input nodes. It also undergoes data normalization, backpropagation algorithm, and weight adjustment processes. The specific modeling flowchart can be found in [reference needed]. Figure 3 .
[0029] The main parameters for training the BP neural network in this embodiment include: (1) Maximum number of iterations and learning rate: The learning rate is the size of the weight modification when the error is fed back each time. Its value is between 0.01 and 0.8, preferably 0.01; the maximum number of iterations is generally between 1000 and 10000, preferably 2000. (2) Activation function: The preferred activation function is ReLU.
[0030] (3) Number of input, hidden and output nodes: The number of nodes is selected based on empirical methods, enumeration methods and manual setting methods. (4) Error function: mainly by studying the mean square error between the true value and the predicted value as the basis for judging the accuracy of the model.
[0031] Random Forest is an ensemble learning method. Its core idea is to use the predictions from multiple decision trees to reduce the model's variance, thereby improving its overall generalization ability. For the specific structure of the Random Forest model, please refer to [link / reference needed]. Figure 4 .
[0032] The main parameters for training the random forest model in this embodiment include: (1) Number of decision trees: Increasing the number of decision trees can improve the stability and accuracy of the model, but it will also increase the computation time and memory consumption. Therefore, the number of decision trees in this embodiment is preferably 300.
[0033] (2) Maximum depth of decision tree: It is used to control the complexity of the tree and prevent overfitting; the maximum depth in this embodiment is preferably 40.
[0034] (3) Minimum number of samples required for internal node re-division: This is used to control the minimum number of samples required for splitting nodes and prevent overfitting; in this embodiment, the minimum number of samples for re-division is preferably 2.
[0035] (4) Minimum number of samples required for leaf nodes: This is used to control the minimum number of samples required for leaf nodes to prevent overfitting; in this embodiment, the minimum number of samples for leaf nodes is preferably 4.
[0036] (5) Maximum number of features considered for each split node: This is used to control the number of features considered for each split node and reduce the variance of the model; common values are auto (all features), sqrt (square root feature number), and log2 (logarithmic feature number). In this embodiment, log2 is preferred.
[0037] The XGBoost model is an improved gradient boosting decision tree algorithm (GBDT) that enhances simulation accuracy and efficiency. The main parameters for training the XGBoost model in this embodiment include: (1) Number of base estimators: that is, the number of decision trees to be iteratively trained; in this embodiment, the preferred value is 200.
[0038] (2) Contribution of each weak learner: The lower the learning rate, the smaller the influence of each type of individual learner, and therefore more learners are needed to achieve the same effect. In this embodiment, the preferred learning rate is 0.01.
[0039] (3) Specify the type of weak learner to be used: Different weak learners and their configurations (such as the maximum depth of the decision tree) will affect the complexity and generalization ability of the model. In this embodiment, the decision tree regressor is selected as the weak learner, and its maximum depth (max_depth) is adjusted to balance the model complexity and generalization ability.
[0040] (4) Setting the loss function: In this embodiment, the exponential loss function is selected to measure the prediction error and optimize the model training process.
[0041] In addition to the three data-driven models mentioned above, this embodiment may also employ combinations of other machine learning or deep learning models as alternatives to construct a TSO prediction model, specifically including: (1) Replace the XGBoost model or random forest model with one or more of the following models: LightGBM, CatBoost, and Support Vector Regression (SVR).
[0042] (2) Use deep neural networks (DNN) or long short-term memory networks (LSTM) to construct TSO prediction models; among them, LSTM networks are particularly suitable for processing blowing process data with time-series characteristics (such as the curve of oxygen flow rate changing with time), and can directly model dynamic processes.
[0043] (3) Use ensemble learning frameworks, such as Stacking or Blending, to input the prediction results of multiple base models as meta-features into a secondary model (such as logistic regression) for final prediction.
[0044] Of course, the various network structures and parameter settings shown above are only for ease of understanding and simplification of description. This invention does not impose specific limitations on the construction and training process of the TSO prediction model.
[0045] Step S2: Based on the actual production data of each heat and the TSO prediction model for molten steel composition, the predicted TSO content of molten steel at the converter endpoint for each heat is obtained.
[0046] It is understandable that the real-time dynamic characterization of carbon content during converter smelting is the core objective for achieving precise control of molten steel composition. The TSO content of molten steel at the end of converter smelting is the core initial constraint for subsequent iterative calculations of the kinetic prediction model. Only by clearly defining the target value at the end of the converter can the carbon content change curve at each moment during the blowing process be derived and fitted in subsequent steps, thus achieving full-time characterization "from the end point to the process".
[0047] In other words, the prediction results output by the TSO prediction model will serve as the input conditions for subsequent kinetic process prediction, providing precise endpoint constraints for the kinetic calculation formula. Then, through iterative calculation, the carbon element composition changes of each furnace at each moment during the blowing process will be simulated, ultimately achieving real-time dynamic characterization of carbon content throughout the converter smelting process, and providing precise process guidance for on-site operation adjustments.
[0048] In practice, the process of outputting the TSO content prediction results specifically includes: inputting the actual production data of each batch into the coupled network model, and performing a weighted average of the prediction results of the BP neural network, random forest model, and XGBoost model to obtain the TSO content prediction results corresponding to each batch; the weighted average refers to the weighted summation of the predicted values output by the BP neural network, random forest model, and XGBoost model based on the prediction accuracy of each network on the validation set or the pre-set weight coefficients to obtain the final prediction result after fusion, which will not be elaborated here.
[0049] The TSO prediction model for the final stage of converter smelting in this invention can predict the content of elements such as C, Si, Mn, and P at the final stage of converter smelting. Establishing this TSO prediction model is a prerequisite for high-precision prediction of the content of various elements in the reaction process of molten steel, and also an important foundation for achieving stable control of molten steel composition. In this embodiment, real data from each heat of steelmaking from a steel enterprise is input into the trained TSO prediction model, thereby obtaining the predicted TSO content of the final stage of converter smelting for each heat. This predicted TSO content is then used as the input condition for a subsequent kinetic prediction model to optimize the kinetic prediction model.
[0050] Step S3: Introduce control parameters to modify the preset decarbonization reaction kinetics model; use the predicted TSO content results as endpoint constraints to perform parameter iteration optimization on the preset decarbonization reaction kinetics model; and obtain the optimized converter decarbonization reaction kinetics prediction model based on the modified decarbonization reaction kinetics model and the optimal parameter results.
[0051] Specifically, the core of establishing a kinetic prediction model for converter decarburization reaction lies in simulating the dynamic evolution of carbon content throughout the entire blowing process through iterative calculations, based on initial conditions and reaction mechanisms. Since the converter smelting process involves multiple reaction environments and interfaces, when thermodynamic equilibrium and mass transfer conditions change dynamically, droplets ejected from the molten pool generate numerous dispersed interfaces, causing the refining reaction to occur between the emulsion droplets and the slag. Its kinetics are closely related to the interface area, droplet residence time, slag physicochemical conditions, and droplet formation rate.
[0052] To facilitate understanding of the establishment process of the kinetic prediction model in the embodiments of the present invention, the technical principles of the decarbonization reaction kinetic theory will be explained below.
[0053] Decarburization is a key process in metallurgy where carbon in molten metal reacts with dissolved oxygen [O] and oxidants such as FeO in slag to generate CO gas. Combined with... Figure 2 As shown, this invention establishes a mathematical method for processing multi-zone reaction kinetics to simulate the time-varying rate parameters in molten steel and the overall refining process between various elements. Based on different reaction environments and mass transfer conditions, the main reaction zones are divided into: (1) Jet impact zone (iz): In this zone, the oxygen blown into the converter oxygen lance reacts directly with the molten steel at a temperature of 2100 to 2600℃.
[0054] (2) Slag-steel-gas emulsion zone (em): also known as the emulsion zone, in which the oxygen jet blown in by the converter oxygen lance impacts the molten steel in the pool, causing the metal droplets to splash through the slag layer and form an emulsion phase with the slag and gas.
[0055] (3) Steel-slag contact reaction zone (sm): In this zone, the converter slag and molten steel come into contact over a large area to carry out mass transfer and reaction.
[0056] Based on the three different regions mentioned above, the overall rate of removal of various elements in the decarbonization reaction kinetics theory can be expressed as follows: (1) Where, j represents various chemical elements such as P (phosphorus), Si (silicon), Mn (manganese), C (carbon), etc.; overall represents the sum of all reaction zones; iz, sm, and em represent the jet impact zone, the steel-slag contact reaction zone, and the slag-steel-gas emulsion zone, respectively; W m The mass of molten steel is expressed in tons (t); C jm Let j be the mass percentage of element j.
[0057] Based on the above decarburization reaction theory, this invention constructs a corresponding basic model of decarburization reaction kinetics and a predictive model of converter decarburization reaction kinetics to simulate the time-varying rate parameters in molten steel and the overall reaction process of each element.
[0058] The following section elaborates on the construction process of the basic model for decarbonization reaction kinetics.
[0059] In this embodiment of the invention, the basic decarburization reaction kinetic model is constructed based on the theory of decarburization reaction kinetics and is used to simulate the carbon element reaction kinetics process in the jet impact zone and the emulsion zone during the converter decarburization reaction. Specifically, the basic decarburization reaction kinetic model includes calculation formulas for the total decarburization rate, the decarburization rate in the jet impact zone, and the decarburization rate in the emulsion zone.
[0060] First, since the carbon removal behavior in converter smelting is significantly regionally dominant, the contribution of the slag-steel reaction zone (sm) to decarburization is weak and negligible. Therefore, the focus is only on the two core zones: the jet impact zone (iz) and the emulsion zone (em), rather than covering all three zones. Thus, the formula for calculating the total decarburization rate is a simplification of formula (1), specifically including: (2) Where, overall represents the sum of all regions; iz represents the jet impact zone; em represents the emulsion zone; W m The mass of molten steel is expressed in tons (t); C Cm denoted as , where is the mass percentage of carbon; m is the mass.
[0061] Secondly, since the high temperature in the jet impact zone significantly promotes oxygen dissolution in the molten steel, the oxidation rate in this region can be expressed according to the first-order rate law under the assumption that the reaction is controlled by mass transfer in the molten steel. Meanwhile, the gas-phase mass transfer coefficient is much higher than the molten steel phase mass transfer coefficient (usually several orders of magnitude higher), and the chemical reaction itself is extremely rapid under high-temperature conditions. Based on this, the formula for calculating the decarburization rate (carbon oxidation rate) in the jet impact zone includes: (3) Where [%C] represents the mass percentage of carbon in the molten steel; A iz The area of the reaction interface in the jet impact zone, in meters. 2 V m The mass of molten steel is expressed in kg. The oxygen mass transfer coefficient of carbon elements in the jet impact zone is corrected, with units of m / s; This represents the percentage of carbon by mass at equilibrium in the jet impact zone.
[0062] Meanwhile, the oxygen mass transfer coefficient of carbon in the jet impact zone was corrected. The calculation formula can be expressed as follows: (4) in, The diffusion coefficient of carbon in a liquid metal droplet is given in m. 2 / s; Q is the circulation and renewal velocity of the molten steel in the pool under top blowing conditions, in m / s; Q is the lance flow rate, in Nm³. 3 / h; g is the acceleration due to gravity, in m / s². 2 H bath r is the depth of the molten pool, in meters (m). cm The impact radius is expressed in meters (m).
[0063] Finally, the oxidation rate of a single droplet element moving in the emulsion zone can be described by the rate equation controlled by mass transfer on the molten steel side, that is, the calculation formula for the decarburization rate in the emulsion zone includes: (5) in, To correct for the mass transfer coefficient of carbon in the emulsion region, the unit is m / s; A d The area of the metal droplets in the emulsion, in m². 2 ;m d This indicates the mass of metal droplets in the emulsion, expressed in kg. ρ is the density of molten steel; m is the mass; This represents the mass percentage of carbon at equilibrium. Meanwhile, the mass transfer coefficient of carbon element in the emulsion zone was corrected. It can be represented as follows: (6) in, This is the diffusion coefficient of carbon in the liquid steel phase, expressed in m. 2 / s; u is the droplet velocity in m / s; d is the average diameter of the metal droplet in m.
[0064] In summary, formulas (2) to (6) are the kinetic calculation formulas in the basic model of decarbonization reaction kinetics.
[0065] It can be understood that parameters that can be directly controlled and adjusted by operators on the production site are called control parameters, such as the lance height h, the mass of molten iron entering the furnace M0, and the mass of lime added as slagging agent M. Lime These are four key control parameters affecting the mass transfer coefficient of the decarburization reaction kinetics in converter smelting; in addition, the lance flow rate Q can also be directly controlled and adjusted by on-site operators. Therefore, this invention considers integrating the basic decarburization reaction kinetics model with the control parameters and related correction coefficients, and optimizing the model parameters based on the TSO content prediction results to obtain an optimized converter decarburization reaction kinetics prediction model. This allows on-site technical operators to continuously correct the kinetics prediction model by directly adjusting the control parameters. This model can better match the actual product content requirements on-site, thereby effectively avoiding abnormal operating conditions and product quality problems, and significantly improving production efficiency.
[0066] The following section elaborates on the modification process of the aforementioned basic model of decarbonization reaction kinetics.
[0067] Preferably, the control parameters include at least: lance height h, molten iron mass M0, and lime mass M (the slag-forming agent). Lime The above control parameters are all derived from actual production data for each batch.
[0068] The introduction of control parameters to modify the preset decarburization reaction kinetics model includes: modifying the oxygen mass transfer coefficient of carbon in the jet impact zone based on the lance height h and the mass of molten iron M0; and modifying the mass of lime, the slag-forming agent M... Lime The mass transfer coefficient of carbon in the emulsion region is corrected.
[0069] On the one hand, fitting analysis of actual field data revealed that the quality of molten iron directly affects the physicochemical properties of the molten pool (such as temperature field distribution and molten metal flow behavior), thereby altering the thickness of the mass transfer boundary layer and the element diffusion path. When the quality of molten iron increases, the heat capacity of the molten pool increases, and the jet impact depth and energy distribution change during the injection process. Therefore, it is necessary to dynamically compensate the mass transfer coefficient in the jet impact zone by introducing the molten iron quality M0, in order to quantify the impact of different batches of molten iron on the mass transfer efficiency and enhance the model's robustness to raw material fluctuations.
[0070] Meanwhile, in actual production, the spray gun height often needs to be adjusted due to real-time changes in raw material conditions, production process requirements, and other factors. Therefore, this embodiment introduces the spray gun height h, which allows operators to adjust the model parameters online based on the spray gun height data detected in real time at the production site, so as to dynamically adapt to different working conditions. In addition, considering that even small changes in spray gun height can significantly affect the kinetic conditions and mass transfer efficiency of the reaction system, thereby changing the rate of the entire decarbonization reaction, this model places the spray gun height h in the exponential term of the formula to more sensitively reflect its nonlinear influence on the reaction process.
[0071] Based on this, the corrected formula for the oxygen mass transfer coefficient of carbon in the jet impact zone includes: (7) in, The corrected oxygen mass transfer coefficient of carbon in the jet impact zone; The diffusion coefficient of carbon in a liquid metal droplet is given in m. 2 / s; Q is the circulation and renewal velocity of the molten steel in the pool under top blowing conditions, in m / s; Q is the lance flow rate, in Nm³. 3 / h; g is the acceleration due to gravity, in m / s². 2 H bath r is the depth of the molten pool, in meters (m). cm M0 is the impact radius in meters; M0 is the mass of molten iron entering the furnace in kilograms; a is a constant; h is the lance height in meters; b is the adjustment constant for the lance height h. This is the correction factor for the carbon element introduced into the jet impact zone.
[0072] On the other hand, based on the thermodynamics and transport principles of slag-steel reaction, it is known that the amount of lime (CaO) added, as the main slagging agent, directly determines the slag basicity, which in turn affects the slag-steel interfacial tension (reducing interfacial tension increases the contact area) and slag viscosity (low viscosity promotes element diffusion). Therefore, this embodiment introduces the mass M of lime as a slagging agent. LimeAs a dynamic correction factor for the mass transfer coefficient in the emulsion zone, it can quantify the dynamic influence of slagging agent on the mass transfer process. Thus, it not only considers the synergistic effect of stirring energy, but also resolves the contradiction between the fixed alkalinity assumption in the traditional model and the actual production fluctuations.
[0073] Based on this, the corrected formula for the mass transfer coefficient of carbon in the emulsion zone includes: (8) in, The corrected mass transfer coefficient of carbon in the emulsion region is expressed in m / s. ρ is the diffusion coefficient of carbon in the liquid steel phase, in m / s; u is the droplet velocity, in m / s; d is the average diameter of the metal droplet, in m; M Lime The mass of lime, the slag-forming agent, is expressed in tons (t). This is an index used to adjust the quality of lime when slag-forming agents are added.
[0074] Furthermore, the preset decarburization reaction kinetics basic model is subjected to parameter iterative optimization, including: substituting the predicted TSO content for each heat as an endpoint constraint into the decarburization reaction kinetics basic model, and using the percentage content of molten steel composition calculated at each time step as the initial value for the next time step; for example, firstly, the reaction rate of each element is calculated based on the current state ([C], [P], temperature T, ...); secondly, the reduction of each element within this step (Δ[C], Δ[P]...); then, the content of each element at the next moment is updated ([C](t+Δt) = [C](t)-Δ[C]); finally, the updated content of each element is used as the initial value for the next time step.
[0075] Then, the calculated percentage of molten steel composition for each heat at a specific measurement time (such as the TSC / TSO measurement time) is compared with the error between the calculated result and the corresponding true value of molten steel composition composition, until the optimal result with the smallest error is obtained.
[0076] Specifically, in each heat, based on the real-time acquired actual production data, the time difference between the "blowing start time" and the "TSO measurement time" is taken as the total reaction time, and the TSO content prediction result obtained from the aforementioned steps is used as the endpoint constraint of the decarburization reaction kinetics basic model, and substituted into the aforementioned formulas (3) and (5) for iterative calculation; in the error comparison stage, the percentage content of molten steel calculated for the "TSC measurement time" (i.e., the first measurement time when the converter oxygen blowing rate is about 80%) is compared with the actual value of the percentage content of molten steel at the TSC measurement time in the actual production data to obtain the accuracy of various elements within the error range. Similarly, kinetic calculations are also performed for the "TSO measurement time" (i.e., the measurement time at the converter endpoint), and the calculation results are compared with the actual value of the percentage content of molten steel at that time.
[0077] Based on the above comparison results, the parameters involved in the mass transfer coefficient function of carbon in each reaction region are automatically optimized through iterative fitting to continuously improve the model's prediction accuracy. This fitting process relies on the global optimization tool forest_minimize (from the skopt library) of the random forest surrogate model. Iterative search finds the parameter combination that minimizes the prediction error, and this optimal parameter result is taken as the result. For example, the optimized model parameters are shown in Table 1: Table 1. Optimal model parameters required for carbon removal kinetics calculation.
[0078] Based on this, the results of the optimal model parameters obtained from the debugging are substituted into the basic model of decarbonization reaction kinetics after the fusion control parameters are obtained, that is, substituted into formulas (7) and (8), and the operation guidance formula with the key control parameters on site as input variables can be derived.
[0079] For example, by substituting the optimal parameter results in Table 1 into the formula (7) for the jet impact zone and the formula (8) for the emulsion zone, formulas (9) and (10) with the key on-site control parameters as input variables can be derived, as shown below: (9) (10) In summary, based on the modified decarbonization reaction kinetics basic model and the optimal parameter results, the optimized converter decarbonization reaction kinetics prediction model can be obtained; finally, the complete kinetics prediction model of this invention mainly includes formulas (2), (3), (5), (7), and (8), as well as formulas (9) and (10) with the optimal parameter results substituted.
[0080] Therefore, it can be seen that during the smelting process, production technicians can adjust the spray gun flow rate Q, spray gun height h, molten iron quality M0, and slag-forming agent lime quality M0. Lime This allows for continuous adjustment and optimization of the dynamic prediction model, thereby avoiding abnormal operating conditions and improving production efficiency.
[0081] Preferably, the method further includes: calculating the predicted decarburization rate of each heat during the blowing process based on the real-time control variables of each heat and the converter decarburization reaction kinetic prediction model, and obtaining the carbon element content change curve corresponding to each heat based on the predicted decarburization rate. The predicted decarbonization rate includes at least the following: the decarbonization rate in the emulsion zone, the decarbonization rate in the jet impact zone, and the total decarbonization rate for each heat during the blowing process.
[0082] It can be understood that the real-time control variables for each furnace are the variables in formulas (9) and (10), namely, the lance flow rate Q, the lance height h, the molten iron mass M0, and the slag-forming agent lime mass M. Lime In actual production, the real values of the real-time control variables obtained above are substituted into the above-optimized and adjusted kinetic prediction model to calculate the predicted decarburization rate at each moment during the blowing process of each furnace. Specifically, the decarburization rate of the emulsion zone can be calculated based on formula (5), the decarburization rate of the jet impact zone can be calculated based on formula (3), and the total decarburization rate can be calculated based on formula (2). At the same time, based on the above three rates, the carbon content change curve at each moment during the blowing process of each furnace is fitted.
[0083] It is understood that the above embodiments are only for ease of understanding and simplification of description, and should not be construed as limiting the present invention. The present invention does not impose specific limitations on the selection of control parameters, TSO prediction model, dynamic prediction model and dynamic formula derivation process.
[0084] Therefore, it can be seen that the embodiments of the present invention can achieve one of the following beneficial effects: First, this invention introduces the following parameters: lance height h, molten iron mass M0, and slag-forming agent lime mass M... Lime The control parameters, which can be observed and adjusted in real time by on-site operators, can accurately capture the impact of minute changes in control parameters on the kinetic reaction of elements removed during converter smelting, and realize the model's direct response to operating commands.
[0085] Second, this invention does not treat a single parameter in isolation, but incorporates multiple key factors into the calculation of the mass transfer coefficient simultaneously, reflecting the coupling effect of multiple control parameters on the decarbonization rate. It can accurately distinguish the decarbonization behavior under different operating conditions and improve the accuracy of decarbonization rate prediction.
[0086] Third, this invention uses the TSO content prediction results output by the TSO prediction model to replace the thermodynamic equilibrium assumption and uses it as the endpoint constraint of the kinetic differential equation, making the simulation path of the entire reaction closer to the actual blowing process; it retains the interpretability and extrapolation ability of the mechanism model, and inherits the fitting advantage of the data model for complex nonlinear relationships, forming a complementary advantage.
[0087] Fourth, this invention establishes a "prediction-feedback-correction" closed loop, which significantly improves the long-term adaptability and robustness of the converter decarburization reaction kinetic prediction model. By using the newly acquired measured TSC / TSO data for each furnace, the correction coefficients are continuously updated iteratively, enabling the model to automatically adapt to long-term operating condition drifts such as fluctuations in raw material composition, equipment aging, and changes in operating habits. This ensures that each iteration can find the globally optimal or near-optimal parameter combination, thereby guaranteeing the model's continued leading performance.
[0088] In another embodiment of the present invention, a system for constructing a converter decarburization reaction kinetic prediction model is proposed, such as... Figure 5 As shown, it specifically includes the following modules: The TSO prediction model building module is used to build a TSO prediction model for the final steel composition of the converter based on valid sample data of the converter process in steel enterprises. The converter endpoint TSO prediction module is used to predict the TSO content of the molten steel at the converter endpoint for each heat based on the actual production data of each heat and the steel composition TSO prediction model. The kinetic model optimization module is used to introduce control parameters to modify the preset decarbonization reaction kinetic basic model; the predicted TSO content of each item is used as the endpoint constraint to perform parameter iterative optimization on the preset decarbonization reaction kinetic basic model, and the optimized converter decarbonization reaction kinetic prediction model is obtained based on the modified decarbonization reaction kinetic basic model and the optimal parameter results.
[0089] The above-described method and system embodiments are based on the same principles, and their related aspects can be referenced from each other to achieve the same technical effects. For specific implementation processes, please refer to the foregoing embodiments, which will not be repeated here.
[0090] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.
[0091] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for constructing a predictive model of converter decarburization reaction kinetics, characterized in that, include: Based on valid sample data of the converter process in steel enterprises, a TSO prediction model for the final steel composition at the converter stage is constructed. Based on the actual production data of each heat and the TSO prediction model for molten steel composition, the predicted results of the TSO content of molten steel at the converter endpoint for each heat are obtained. The preset decarbonization reaction kinetics model is modified by introducing control parameters; the predicted TSO content of each item is used as the endpoint constraint to perform parameter iteration optimization on the preset decarbonization reaction kinetics model; and based on the modified decarbonization reaction kinetics model and the optimal parameter results, the optimized converter decarbonization reaction kinetics prediction model is obtained.
2. The construction method according to claim 1, characterized in that, The input data in the valid sample data shall include at least: oxygen supply time, total blowing time, carbon removal oxygen consumption, molten iron temperature, actual molten iron quantity, actual scrap steel quantity, lime quality, lightly calcined dolomite quality, pellet quality, dynamic pellet quality, and molten iron carbon content. The output data in the valid sample data includes at least the content of C, Si, Mn and P elements at the converter endpoint.
3. The construction method according to claim 2, characterized in that, The TSO prediction model is a coupled network model trained based on BP neural network, random forest model and XGBoost model. The prediction results of the TSO content of the converter steel at the end of each heat, based on the actual production data of each heat and the TSO prediction model, include: The actual production data of each furnace batch is input into the coupled network model, and the prediction results of the BP neural network, random forest model and XGBoost model are weighted and averaged to obtain the TSO content prediction results for each furnace batch.
4. The construction method according to claim 1, characterized in that, The fundamental model of decarbonization reaction kinetics specifically includes calculation formulas for the total decarbonization rate, the decarbonization rate in the jet impact zone, and the decarbonization rate in the emulsion zone; among which, The formulas for calculating the overall decarbonization rate include: ； Where, overall represents the sum of all reaction zones; iz represents the jet impact zone; em represents the emulsion zone; W m For the quality of molten steel; C Cm m is the mass percentage of carbon; m is the mass. The formulas for calculating the decarburization rate in the jet impact zone include: ； Where [%C] represents the mass percentage of carbon in the molten steel; A iz V represents the area of the reaction interface in the jet impact zone. m For the quality of molten steel; To correct the oxygen mass transfer coefficient of carbon elements in the jet impact zone; This represents the mass percentage of carbon in the jet impact zone at equilibrium. The formulas for calculating the decarburization rate in the emulsion zone include: ； in, To correct the mass transfer coefficient of carbon in the emulsion region; A d m is the area of the metal droplets in the emulsion. d This indicates the mass of the metal droplets in the emulsion; ρ is the density of molten steel; m is the mass; This represents the percentage of carbon by mass at equilibrium.
5. The construction method according to claim 4, characterized in that, The control parameters include at least: lance height h, molten iron mass M0, and lime mass of slag-forming agent M. Lime ; The basic model of decarburization reaction kinetics is constructed based on the theory of decarburization reaction kinetics and is used to simulate the carbon element reaction kinetics process in the jet impact zone and emulsion zone of the converter decarburization reaction. The introduction of control parameters to modify the preset decarbonization reaction kinetics model includes: The oxygen mass transfer coefficient of carbon in the jet impact zone is corrected based on the lance height h and the mass of molten iron M0 entering the furnace; and based on the mass of lime, the slag-forming agent M... Lime The mass transfer coefficient of carbon in the emulsion region is corrected.
6. The construction method according to claim 5, characterized in that, The correction formula for the oxygen mass transfer coefficient of carbon in the jet impact zone includes: ； in, The corrected oxygen mass transfer coefficient of carbon in the jet impact zone; is the diffusion coefficient of carbon in a liquid metal droplet; The circulating renewal rate of the molten steel in the pool under top blowing conditions is given by: Q = lance flow rate; g = gravitational acceleration; H = 1 / 2. bath r is the depth of the molten pool. cm Where M is the impact radius; M0 is the mass of molten iron entering the furnace; a is a constant; h is the lance height; b is the adjustment constant for the lance height h; This is the correction factor for carbon elements in the jet impact zone.
7. The construction method according to claim 6, characterized in that, The correction formula for the mass transfer coefficient of carbon in the emulsion zone includes: ； in, This is the corrected mass transfer coefficient of carbon in the emulsion region; ρ is the diffusion coefficient of carbon in the liquid steel phase; u is the droplet velocity; d is the average diameter of the metal droplet; M Lime For the quality of lime used as a slag-forming agent; This is an index used to adjust the quality of lime when slag-forming agents are added.
8. The construction method according to claim 7, characterized in that, The step of using the predicted TSO content results as endpoint constraints to iteratively optimize the parameters of the preset decarbonization reaction kinetics model includes: The predicted TSO content for each heat is used as the endpoint constraint and substituted into the basic model of decarburization reaction kinetics. The percentage content of molten steel composition calculated at each time step is used as the initial value for the next time step. Compare the calculated percentage of molten steel composition at a specific measurement time for each heat, and the error between them and the corresponding true value of molten steel composition, until the optimal parameter result with the smallest error is obtained.
9. The construction method according to claim 8, characterized in that, The method further includes: Based on the real-time control variables of each heat and the converter decarburization reaction kinetic prediction model, the predicted decarburization rate of each heat during the blowing process is calculated, and the carbon element content change curve corresponding to each heat is obtained based on the predicted decarburization rate. The predicted decarbonization rate includes at least the following: the decarbonization rate in the emulsion zone, the decarbonization rate in the jet impact zone, and the total decarbonization rate for each heat during the blowing process.
10. A system for constructing a predictive model of converter decarburization reaction kinetics, characterized in that, include: The TSO prediction model building module is used to build a TSO prediction model for the final steel composition of the converter based on valid sample data of the converter process in steel enterprises. The converter endpoint TSO prediction module is used to predict the TSO content of the molten steel at the converter endpoint for each heat based on the actual production data of each heat and the steel composition TSO prediction model. The kinetic model optimization module is used to introduce control parameters to modify the preset decarbonization reaction kinetic basic model; the predicted TSO content of each item is used as the endpoint constraint to perform parameter iterative optimization on the preset decarbonization reaction kinetic basic model, and the optimized converter decarbonization reaction kinetic prediction model is obtained based on the modified decarbonization reaction kinetic basic model and the optimal parameter results.