Slag rocking smelting control method and system

Abstract

The invention belongs to the technical field of metal refining control, and particularly relates to a slag rocking smelting control method and system.The method comprises the steps that firstly, the current slag composition and temperature are estimated through a model, and the initial viscosity is calculated through an RBF interpolation model; secondly, evaluating the confidence coefficient from two dimensions of data support and physical consistency, and starting boundary gradient projection correction when the confidence coefficient is low; and finally, carrying out dynamic weighted fusion on the initial estimated value and the corrected value to obtain reliable final viscosity, and carrying out closed-loop adjustment on parameters of the rocking smelting equipment so as to realize accurate control on the smelting process. The method solves the contradiction between a high-precision model and high reliability, and ensures that the system can always output a reasonable and stable estimation result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of metal refining control. More particularly, the present application relates to a molten slag stirring control method and system. BACKGROUND

[0002] In the process of refining molten steel, the viscosity of molten slag is a key physical parameter affecting the kinetics of slag-steel reaction, mass transfer efficiency and impurity removal effect; it is of great significance to optimize the smelting process and improve product quality by accurately obtaining the viscosity of molten slag in real time and adjusting the operating parameters of the stirring furnace accordingly.

[0003] However, due to the harsh environment of high temperature and corrosion, it is extremely difficult to directly measure the viscosity of molten slag online. Therefore, the existing technology usually uses a mathematical model based on offline data to estimate online.

[0004] Among them, the radial basis function (RBF) interpolation model becomes a common technical means because it can accurately approximate the nonlinear function relationship; this method constructs a mapping relationship by learning a large number of offline data points, estimates the viscosity of molten slag from a pre-established offline thermodynamic database according to the real-time temperature and estimated chemical composition, and then uses this viscosity value as the core basis for adjusting the stirring parameters.

[0005] In actual production, due to fluctuations in raw material composition, temperature measurement errors or abnormal feeding, the actual state point of the process is prone to deviate from the core area of the database, enter the boundary zone with sparse data, or even completely drift out of the database range; at this time, the RBF interpolation model performs extrapolation prediction, and the reliability of the estimation result will decrease sharply, resulting in a large error; in addition, as a pure data-driven black box model, the interpolation result of the RBF interpolation model lacks physical constraints and may output estimated values that violate basic metallurgical principles; these unreliable or incorrect estimated results will directly mislead the subsequent closed-loop control, affect the refining effect, and even cause production accidents. SUMMARY

[0006] To solve the above technical problem that when the actual working condition deviates from the pre-built database range, the RBF interpolation model is prone to produce incorrect viscosity outliers due to extrapolation, thereby endangering the stability of the control system, the present application provides solutions in the following aspects.

[0007] In a first aspect, the present invention provides a slag refining control method, comprising: acquiring real-time temperature and chemical composition vectors of slag in a refining furnace; inputting a current operating point composed of the real-time temperature and chemical composition vectors into a pre-constructed RBF interpolation model to obtain an initial viscosity estimate; calculating a data support domain index based on the proximity of the current operating point to data points in an offline database of the RBF interpolation model; calculating a physical consistency index based on the partial derivative of the initial viscosity estimate with respect to temperature; calculating the product of the data support domain index and the physical consistency index to obtain the total confidence level of the initial viscosity estimate; when the total confidence level is lower than a preset threshold, performing boundary gradient projection correction on the initial viscosity estimate to obtain a corrected viscosity estimate; calculating the final output viscosity based on the initial viscosity estimate, the corrected viscosity estimate, and the total confidence level; calculating the target refining angle of the refining furnace based on the final output viscosity, and controlling the refining furnace to operate according to the target refining angle.

[0008] This invention first obtains an initial estimated value through an RBF interpolation model. Then, it evaluates the confidence level of the estimated value from two dimensions: data support (distance between the operating point and the database) and physical consistency (viscosity variation with temperature). When the confidence level is low, gradient projection correction is performed using boundary information from the database to generate a highly reliable corrected value. Finally, the initial and corrected values ​​are dynamically weighted and fused based on the confidence level. This constructs a dual safety guarantee, which can effectively identify and handle unreliable outputs of the model while ensuring that the final output viscosity value always has both high accuracy and high stability. This fundamentally solves the contradiction between high-precision models and high reliability, and significantly improves the robustness and accuracy of intelligent control of the slag refining process.

[0009] Preferably, the method for obtaining the chemical composition vector includes: at the start of refining, calculating the initial absolute mass of each oxide component in the slag based on the weighing and composition report of the material fed into the furnace, and obtaining the initial chemical composition vector; when a material addition operation is detected, updating the mass of each oxide component in the slag based on the weight of the added material and the known composition; and during the continuous process, updating the chemical composition vector by linear interpolation based on the preset characteristic component change vector.

[0010] Preferably, the linear interpolation update of the chemical composition vector includes: in real-time production, at the start time when the process enters a certain stage. The model records the component vector at this moment. This value represents the updated quality; therefore, at any given time... ( Chemical composition vector Estimate using the following linear interpolation formula: In the formula, The starting point of a certain stage of the process Chemical composition vector, This represents the characteristic component change vector during the process stage. This refers to the characteristic duration of each process stage.

[0011] Preferably, the step of calculating the data support domain index based on the proximity of the current operating point to the data points in the offline database of the RBF interpolation model includes: based on the current operating point... In the offline database, the KD-tree nearest neighbor search algorithm is used to find the 10 nearest neighbors of the current operating point; the average nearest neighbor distance of the current operating point is calculated to obtain the average nearest neighbor distance of the current operating point. Then the data supports the domain index. The formula for calculation is: In the formula, To support the domain index for data; For reference distance; It is a natural exponential function.

[0012] This invention employs a KD-tree for efficient nearest neighbor search and designs a Gaussian decay function based on the average nearest neighbor distance. This transforms the geometric proximity relationship between the current working point and the database into a standardized confidence score ranging from 0 to 1. Compared to simple distance threshold judgment, this can more smoothly and sensitively reflect the transition state of the RBF interpolation model output from reliable interpolation to questionable extrapolation, providing a geometric basis for subsequent fusion weight calculation.

[0013] Preferably, the physical consistency index is calculated based on the partial derivative of the initial viscosity estimate with respect to temperature, including: In the formula, It is a physical consistency index; Sensitivity coefficient; This represents the local temperature gradient at the current operating point. It is the tangent function of a hyperbola.

[0014] This invention incorporates the fundamental metallurgical principle that the viscosity of slag decreases with increasing temperature into the model evaluation. By using a hyperbolic tangent function, the partial derivative of viscosity with respect to temperature is transformed into a standardized confidence score, adding a physical verification checkpoint to the purely data-driven RBF interpolation model. This effectively identifies and punishes artifacts that violate physical laws, enhancing the system's ability to identify and resist model errors.

[0015] Preferably, the initial viscosity estimate is corrected by boundary gradient projection to obtain a corrected viscosity estimate, which includes: finding the distance from the current operating point on the multidimensional convex hull formed by all data points in the offline database using a geometric algorithm. Nearest boundary point ; In the formula, To correct the viscosity estimate, This is the current operating point. Distance from the current operating point The nearest boundary point; For the RBF interpolation model at boundary points The viscosity value at that location; For the RBF interpolation model at boundary points Viscosity gradient vector at point; This represents the dot product operation of vectors.

[0016] This invention utilizes information from known data boundary points, namely viscosity values ​​and gradients, and performs local linear extrapolation through first-order Taylor expansion to achieve evidence-based inferences. It closely correlates with the local characteristics of the current operating point, thereby maximizing the estimation accuracy in areas with unknown data while ensuring reliability.

[0017] Preferably, the final output viscosity is calculated based on the initial viscosity estimate, the corrected viscosity estimate, and the total confidence level, including: using the total confidence level as the weight of the initial viscosity estimate, and using the difference between the value 1 and the total confidence level as the weight of the corrected viscosity estimate, performing a weighted summation to obtain the final output viscosity.

[0018] Preferably, calculating the target refining angle of the refining furnace based on the final output viscosity includes: based on the target mixing intensity index. The set refining frequency and the final output viscosity According to the calculation formula Obtain the target shaking angle .

[0019] Based on the physical principle of maintaining constant mixing intensity, this invention directly transforms the abstract viscosity value into a specific and executable target refining angle, enabling the control system to automatically adjust its operation according to the real-time changes in the slag state to overcome changes in flow resistance and achieve precise control of the mixing efficiency in the smelting process. This is a key link in realizing the final application of the entire intelligent control method.

[0020] Preferably, the target mixing intensity index The setting method includes: during the oxidation period, setting the target mixing intensity index. The first range is defined; during the reduction and desulfurization periods, a target mixing intensity index is set. This is the second range, and the values ​​in the second range are higher than those in the first range.

[0021] Secondly, the present invention provides a slag refining control system, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned slag refining control method is implemented.

[0022] By adopting the above technical solution, a computer program is generated from the above-mentioned slag refining control method and stored in a memory so that it can be loaded and executed by a processor. In this way, a terminal device can be made based on the memory and the processor for convenient use.

[0023] The beneficial effects of this invention are as follows: This invention first obtains an initial estimated value through an RBF interpolation model. Then, it evaluates the confidence level of the estimated value from two dimensions: data support (distance between the operating point and the database) and physical consistency (viscosity variation with temperature). When the confidence level is low, gradient projection correction is performed using boundary information from the database to generate a highly reliable corrected value. Finally, the initial and corrected values ​​are dynamically weighted and fused based on the confidence level. This constructs a dual safety guarantee, which can effectively identify and handle unreliable outputs of the model while ensuring that the final output viscosity value always has both high accuracy and high stability. This fundamentally solves the contradiction between high-precision models and high reliability, and significantly improves the robustness and accuracy of intelligent control of the slag refining process. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a slag refining control method according to the present invention; Figure 2 This is a flowchart illustrating step S3; Figure 3 It is a schematic diagram illustrating each stage of the process; Figure 4 This is a schematic diagram illustrating the final output viscosity; Figure 5 This is a schematic diagram illustrating the target swing angle. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] This invention discloses a method for controlling slag refining, referring to... Figure 1This includes steps S1-S5: S1: Obtain the real-time temperature and chemical composition vector of the slag in the refining furnace, and combine the real-time temperature and chemical composition vector to form the current operating point.

[0028] It should be noted that, in order to obtain a chemical composition estimate that can both respond quickly to operational events and reasonably reflect the smelting law while ensuring real-time performance, this scheme adopts a simplified model that combines instantaneous quality updates driven by feeding operation events with linear interpolation of composition driven by preset process stage characteristics.

[0029] Specifically, the real-time temperature of the molten slag is obtained through a temperature measuring device. .

[0030] Furthermore, based on the material balance model combining event-driven updates and stage feature interpolation, the current slag chemical composition vector is calculated; the method for obtaining the current slag chemical composition vector is as follows: (1) Initialization: At the start of refining, based on the accurate weighing and composition report of the materials fed into the furnace, calculate the initial absolute mass of each oxide component in the slag to form the initial mass vector. The initialization component vector is obtained through normalization. .

[0031] (2) Event-driven quality update: Real-time monitoring of material addition operations in the control system: When any auxiliary material (such as lime, deoxidizer, etc.) is added into the furnace, the model is triggered.

[0032] Specifically, at the time the event occurred The model calculates the mass of each oxide component in the added material based on its weight and known composition, and instantaneously adds it to the slag mass vector from the previous moment. Therefore, the composition after the feeding is complete... New quality , ; Components before feeding The quality; The components contained in the materials added this time The quality.

[0033] After the accumulation is completed, the new masses of each component after the feeding is completed are normalized, and the chemical composition vector is updated. This value will serve as the starting point for the next stage of calculation.

[0034] (3) The complete refining process can be divided into several stages with clear metallurgical tasks according to the production process sheet, including but not limited to the oxidation period, reduction period, and desulfurization period, as shown in the schematic diagram. Figure 3As shown; this method pre-calibrates the characteristics of each process stage by analyzing a large amount of historical data, thus replacing complex online reaction calculations; therefore, linear interpolation of components based on process stages includes: First, in the offline phase, by statistically analyzing historical production data, a characteristic component change vector is extracted and solidified for each standard process stage. and a feature duration This vector represents the typical average change in the percentage of each chemical component during this stage.

[0035] Then, in real-time production, the starting moment when the process enters a certain stage... The model records the component vector at this moment. This value is the latest value after event-driven quality updates; during this phase, assuming that the component changes caused by chemical reactions are approximately linear and uniform within a standard process phase, and that the overall trend is defined by the characteristic component change vector obtained from historical big data statistics, then at any given time... ( Chemical composition vector Estimate using the following linear interpolation formula: ; In the formula, The starting point of a certain stage of the process Chemical composition vector, This represents the characteristic component change vector during the process stage. This refers to the characteristic duration of each process stage.

[0036] It should be noted that the method for evaluating chemical composition vectors in this step is highly consistent with the actual production operation process, and can naturally combine discrete operations such as feeding and staged smelting with the estimation of continuous changes in composition.

[0037] Finally, the real-time temperature With the calculated chemical composition vector Combine the data to obtain the current operating point. .

[0038] It should be noted that combining event-driven instantaneous updates, i.e., feeding operations, with linear interpolation based on process stages can not only quickly respond to compositional abrupt changes caused by discrete operations such as feeding, but also reasonably simulate continuous compositional changes caused by chemical reactions during the smelting stage. While ensuring real-time performance, it obtains chemical composition data that is closer to actual working conditions at a lower computational cost, providing a more reliable input for subsequent accurate viscosity estimation.

[0039] S2: Input the current operating point into the pre-built RBF interpolation model to obtain the initial viscosity estimate.

[0040] It should be noted that the RBF interpolation model not only provides function values, but its analytical form also makes it easy to directly calculate the gradient. In addition, obtaining the initial viscosity value provides a basis for subsequent fusion, while obtaining its gradient with respect to temperature is a key step in assessing physical consistency.

[0041] Specifically, the current operating point Input into a pre-built offline RBF interpolation model In the process, the initial viscosity estimate is obtained. .

[0042] Furthermore, the RBF interpolation model At the current operating point Seeking information about temperature The partial derivatives are used to obtain the local temperature gradient at the current operating point. .

[0043] Radial basis function (RBF) interpolation is a mathematical method for approximating or interpolating scattered data points in a multidimensional space. In the technical solution of this invention, its core function is to construct a continuous and smooth function model based on a set of discrete pre-calculated data points, i.e., input temperature and multiple chemical components, and output viscosity. This model can quickly calculate the corresponding output value, i.e., viscosity, for any given new input point, i.e., temperature and components that are monitored and estimated in real time. This avoids complex and time-consuming thermodynamic equilibrium calculations in real time and meets the high timeliness requirements of online control.

[0044] Furthermore, the process of obtaining RBF interpolation is a completely offline, data-driven modeling process: (1) A large number of data points describing the mapping relationship between temperature, composition and viscosity of slag were obtained through thermodynamic software calculations and laboratory measurements, forming an offline database covering a wide range of working conditions. Due to the huge differences in the numerical range of each input dimension, the dataset must be normalized and preprocessed.

[0045] (2) The core idea of ​​the RBF interpolation model is that any function can be represented by a linear combination of a series of radial basis functions. Among them, the radial basis functions are Gaussian functions. By making the function expression pass through all the training data points in the working point viscosity training dataset, a linear equation system can be established and solved, thereby solving a set of optimal weights at once.

[0046] In the process of model building and training, a key hyperparameter is the shape parameter of the Gaussian function. :parameter Controlling the influence range of a single basis function has a significant impact on the smoothness and accuracy of the interpolation results; A value that is too small will cause the function to be too smooth and fail to capture details; a value that is too large will easily cause the function to oscillate wildly between data points, resulting in overfitting. Therefore, this parameter is optimized by using cross-validation, that is, trying different... The model is trained using a portion of the data, and then tested using a portion of the untrained data. The model's prediction error is then minimized, and the result that minimizes the test error is selected. The value serves as the final parameter of the model.

[0047] After training, the final RBF interpolation model consists of a set of determined center point coordinates and an optimal shape parameter. It consists of a set of corresponding weights.

[0048] S3: Calculate the data support domain index based on the proximity of the current operating point to the data points in the offline database of the RBF interpolation model, and calculate the physical consistency index based on the partial derivative of the initial viscosity estimate with respect to temperature.

[0049] Refer to the flowchart for step S3 Figure 2 The process includes steps S301 to S302, specifically as follows: S301: Calculate the data support domain index based on the proximity of the current operating point to the data points in the offline database of the RBF interpolation model.

[0050] It should be noted that in order to determine whether the current operating point is adequately supported by the existing data in the offline database, a data support domain index needs to be constructed: if an operating point is surrounded by dense known points in the database, its interpolation result has high reliability; conversely, if the surrounding data points are sparse or non-existent, the reliability is low.

[0051] Specifically, based on the current operating point In the offline database, the 10 closest data points are found using the KD-tree nearest neighbor search algorithm.

[0052] Furthermore, the average distance between the current operating point and its 10 nearest neighbors is calculated to obtain the average nearest neighbor distance of the current operating point. .

[0053] Furthermore, based on the average nearest neighbor distance Calculate the data support domain index The specific calculation formula is as follows: ; In the formula, The data support domain index has a value range of (0,1]. The average nearest neighbor distance for the current operating point; The reference distance is the critical distance threshold from dense to sparse distribution of data points. The value range is dynamically set to 2 to 3 times the average minimum neighbor distance of all data points in the offline database to reflect the density characteristics of the database itself. In this embodiment, it is set to 2.5 times the average minimum neighbor distance of the database. It is a natural exponential function.

[0054] Among them, the data support domain index It is about Gaussian decay function: at the current operating point When in a data-intensive area, much smaller The exponent's power approaches 0, making the data support domain exponent... The value is close to 1, indicating that the interpolation results of the RBF interpolation model have high geometric reliability; at the current operating point When moving to areas with sparse data or outside of those areas, As the value of increases, the power of the exponent increases negatively, making the data support domain exponent... The value decays smoothly and rapidly and approaches 0, indicating that the geometric reliability of the interpolation result is very low.

[0055] It should be noted that the construction of the data support domain index transforms complex geometric proximity relationships into a standardized confidence index. Its Gaussian decay characteristics ensure a sensitive response to changes in data density, providing reliable geometric weights for subsequent fusion.

[0056] S302: Calculate the physical consistency index based on the partial derivative of the initial viscosity estimate with respect to temperature.

[0057] It should be noted that in order to transform the temperature partial derivative, which may take any real value, into a consistency score that is also between 0 and 1, a centrally symmetric sigmoid function with a saturation region is needed to make a clear and smooth distinction between the three cases of compliance, criticality, and violation.

[0058] Specifically, based on the local temperature gradient at the current operating point. Calculate the physical consistency index The specific calculation formula is as follows: ; In the formula, This is the physical consistency index, and its value range is (0,1). This is a sensitivity coefficient used to adjust the intensity of the exponential response to a positive gradient, i.e., how large a positive gradient is considered unreliable. The value range of is [10, 100]. The larger the value, the steeper the function curve and the more severe the penalty for positive gradient. In this embodiment, it is set to 50 based on experience. This represents the local temperature gradient at the current operating point. It is the tangent function of a hyperbola.

[0059] The logical relationship of this formula is that the calculation formula for the physical consistency index is a smooth logical judgment function. According to metallurgical principles, the viscosity of slag decreases with increasing temperature, therefore its partial derivative with respect to temperature should be negative; where, when When it is a significantly negative value, The term approaches -1, making Approaching 1 indicates high physical consistency; when the partial derivative becomes positive, The term approaches 1, making A rapid approach to 0 indicates poor physical consistency, suggesting the model's output is questionable.

[0060] It should be noted that by incorporating prior physical knowledge (i.e., viscosity decreases monotonically with temperature) into the confidence assessment, a second independent verification mechanism is constructed, which greatly enhances the system's ability to identify model artifacts and errors.

[0061] S4: Calculate the product of the data support domain index and the physical consistency index to obtain the total confidence level of the initial viscosity estimate. When the total confidence level is lower than the preset threshold, perform boundary gradient projection correction on the initial viscosity estimate to obtain the corrected viscosity estimate.

[0062] It should be noted that when the model fails in the unknown region, the best strategy is not to revert to a global, coarse low-level model, but to use the edge information of the known world to make the most reasonable local linear inference about the unknown world. This is the core idea of ​​boundary gradient projection.

[0063] Specifically, based on the data-supported domain index Physical consistency index Calculate the total confidence level .

[0064] Furthermore, the total confidence level With the preset correction start threshold Comparison: Optionally, when If the current RBF interpolation model output is considered reliable and requires no correction, then the viscosity estimate can be corrected. equal to the initial viscosity estimate .

[0065] Among them, the modified start threshold This is the lowest confidence level at which the output of the RBF interpolation model can be directly accepted, and its value ranges from [0.8, 0.98]. In this embodiment, the activation threshold will be corrected. Set to 0.9.

[0066] Optionally, when At that time, the boundary gradient projection correction mechanism is activated to obtain the corrected viscosity estimate. The method to obtain it is as follows: First, on the multidimensional convex hull formed by all data points in the offline database, a geometric algorithm is used to find the distance to the current working point. Nearest boundary point .

[0067] Then, using the RBF interpolation model and its gradient, linear extrapolation is performed through a first-order Taylor expansion. The specific calculation formula is as follows: ; In the formula, To correct the viscosity estimate, This is the current operating point. Distance from the current operating point The nearest boundary point; For the RBF interpolation model at boundary points The viscosity value at that location; For the RBF interpolation model at boundary points Viscosity gradient vector at point; This represents the dot product operation of vectors.

[0068] The logical relationship of this formula is an inference based on the local linearization assumption: from reliable boundary points To the nearest current operating point The trend of viscosity function change can be approximated by the gradient at the boundary point, which is equivalent to constructing a tangent hyperplane at the boundary point and using the gradient corresponding to the boundary point. The position value serves as the best estimate of its viscosity.

[0069] It should be noted that the correction value generated by the extrapolation method in this step is closely related to the local characteristics of the current working condition, i.e., which edge it is closest to in the database. Therefore, its prediction accuracy and correlation are improved.

[0070] S5: Based on the initial viscosity estimate, the corrected viscosity estimate, and the total confidence level, the final output viscosity is calculated, and then the target refining angle of the refining furnace is calculated, and the refining furnace is controlled to operate according to the target refining angle.

[0071] It should be noted that by using the total confidence assessment result as the fusion weight, a seamless and smooth switching between different model estimates is achieved, ensuring the continuity and stability of the final output signal.

[0072] Specifically, based on the total confidence level For the initial viscosity estimate and corrected viscosity estimates Perform dynamic weighted fusion to calculate the final output viscosity. The specific calculation formula is as follows: ; In the formula, The final output viscosity; This is the initial viscosity estimate; To correct the viscosity estimate; This represents the total confidence level.

[0073] For example, a schematic diagram of the final output viscosity is shown below. Figure 4 As shown, the results include the dynamic weighted fusion of the initial viscosity estimate and the corrected viscosity estimate based on the total confidence level. When the confidence level is high, the final output viscosity coincides with the initial viscosity estimate, and the high-precision RBF estimation result is accepted. When the confidence level is low, the final output viscosity smoothly approaches the corrected viscosity estimate, avoiding drastic changes caused by model failure.

[0074] It should be noted that the final result The value is always limited to the initial viscosity estimate. and corrected viscosity estimates Between, its specific location is determined by the total confidence level Decision: When When the accuracy is high, the results are fully accepted based on the high precision. ;when At lower values, the results smoothly scale up to the corrected viscosity estimate obtained by projecting boundary information. This ensures that the changes in the output value are continuous and stable, without any sudden changes.

[0075] It should be noted that this fusion step resolves the contradiction between high-precision models and high reliability, ensuring that the system can always provide a reasonable and stable viscosity estimation result regardless of changes in operating conditions.

[0076] It should be further explained that the purpose of the refining process is to achieve effective material exchange at the slag-metal interface, and its intensity, i.e., the mixing intensity, should be maintained at an optimal constant level. The mixing intensity is directly proportional to the driving force applied by the refining action and inversely proportional to the flow resistance of the slag. The driving force is determined by the refining angle and frequency, and the flow resistance of the slag is determined by the viscosity. Therefore, based on the physical principle of constant mixing intensity, this invention transforms the final viscosity estimate into precise equipment operation instructions.

[0077] Specifically, the mixing intensity index ,in, For the angle of refining; The refining frequency; This is the final output viscosity.

[0078] For industrial-grade slag refining equipment, to ensure stable mixing of the molten metal rather than violent agitation, the refining frequency is typically in a low range: between 0.2 Hz and 1.5 Hz. 0.2 Hz means that completing one full refining cycle of "tilting forward, returning to center, tilting backward, returning to center" takes 5 seconds, while 1.5 Hz means that completing one full refining cycle takes approximately 0.67 seconds. In this embodiment, the refining frequency... Set to 0.8 Hz.

[0079] During the oxidation period, rapid dephosphorization and decarbonization are required to promote the oxidation reaction. This is controlled by the chemical reaction rate and mass transfer rate, and requires a high mixing intensity to violently agitate the slag-metal interface, rapidly transfer the oxidant, and remove the reaction products in time to maximize the reaction efficiency. Therefore, the target mixing intensity index is in a relatively high range [1.5,3]. During the reduction and desulfurization periods, deep desulfurization is required to remove oxide inclusions. Desulfurization is a typical slow reaction controlled by mass transfer and is the rate-limiting step in the entire refining process. It requires a very high mixing intensity to overcome the flow resistance caused by the potentially high viscosity of the white slag and to provide the strongest driving force for the migration of sulfur from steel to slag. Therefore, the target mixing intensity index is in a very high range [2,4].

[0080] To maintain a constant mixing strength, the calculation formula for the reeling angle is as follows: , The target mixing intensity index, Refining frequency, refining angle Adjustment and final output viscosity It is directly proportional; therefore, the final output viscosity will be obtained. The target refining angle is obtained by inputting the parameter setting module of the refining controller. Finally, the calculated target refining angle and refining frequency are used as set values ​​and sent to the drive system of the refining furnace for execution, thereby implementing closed-loop adjustment of the refining parameters.

[0081] Among them, when the viscosity of the slag When rising, the angle of the swing The mixing intensity is increased linearly in proportion to apply a greater driving force to overcome flow resistance, thereby maintaining a constant mixing intensity; conversely, when the viscosity decreases, the mixing angle is reduced accordingly to avoid over-stirring and save energy.

[0082] For example, a schematic diagram of the target swing angle is shown below. Figure 5As shown, the trend of the change in the shaking angle is highly consistent with the final viscosity, reflecting the control principle of constant mixing intensity in the scheme.

[0083] This invention also discloses a slag refining control system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a slag refining control method according to the present invention.

[0084] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A molten slag stirring control method, characterized by, The method comprises the following steps: acquiring a real-time temperature and a chemical composition vector of the slag in the rocking furnace; inputting a current working condition point composed of the real-time temperature and the chemical composition vector into a pre-constructed RBF interpolation model to obtain an initial viscosity estimation value; calculating a data support domain index according to a proximity of the current working condition point to data points in an offline database of the RBF interpolation model; and calculating a physical consistency index according to a partial derivative of the initial viscosity estimation value with respect to temperature; calculating a product of the data support domain index and the physical consistency index to obtain a total confidence of the initial viscosity estimation value; when the total confidence is lower than a preset threshold, performing boundary gradient projection correction on the initial viscosity estimation value to obtain a corrected viscosity estimation value; calculating a final output viscosity according to the initial viscosity estimation value, the corrected viscosity estimation value and the total confidence; calculating a target rocking angle of the rocking furnace according to the final output viscosity, and controlling the rocking furnace to operate according to the target rocking angle.

2. A molten slag stirring control method according to claim 1, characterized by, The method for acquiring the chemical composition vector comprises the following steps: at a refining start time, calculating initial absolute masses of oxide components in the slag according to a weight and a composition report of charging materials to obtain an initial chemical composition vector; when a material adding operation is monitored, updating the masses of the oxide components in the slag according to a weight and known compositions of added materials; and during a process stage, performing linear interpolation update on the chemical composition vector according to a preset characteristic composition change vector.

3. A molten slag stirring control method according to claim 2, characterized in that, The linear interpolation update on the chemical composition vector comprises the following steps: In real-time production, when a process enters the start of a certain phase , the model records the composition vector at that moment , the value is the updated quality; then the chemical composition vector at any time instant is estimated by the following linear interpolation algorithm:​ ； wherein is the starting time of the process into a certain phase is the chemical composition vector of the process, is the characteristic composition change vector of the process phase, is the characteristic time length of the process phase.

4. A molten slag stirring control method according to claim 1, wherein The calculation of the data support domain index according to the proximity of the current working condition point to the data points in the offline database of the RBF interpolation model comprises the following steps: According to the current operating point In the offline database, find the nearest 10 data by KD tree nearest neighbor search algorithm; calculate the average value of the distance between the current operating point and its 10 nearest neighbors to obtain the average nearest neighbor distance of the current operating point ; Data support domain index The formula for calculating the data support domain index is: ； wherein is the data support domain index; is the reference distance; is the natural exponential function.

5. A molten slag stirring control method according to claim 1, wherein The calculation of the physical consistency index according to the partial derivative of the initial viscosity estimation value with respect to temperature comprises the following steps: ； wherein is the physical consistency index; is the sensitivity coefficient; is the local temperature gradient at the current operating point; is the hyperbolic tangent function.

6. A molten slag stirring control method according to claim 1, wherein The boundary gradient projection correction on the initial viscosity estimation value to obtain the corrected viscosity estimation value comprises the following steps: on a multidimensional convex hull of all data points of the offline database, by a geometric algorithm, the distance of the current operating point the nearest boundary point ; ； wherein is the corrected viscosity estimate, is the current operating point, is the distance from the current operating point to the nearest boundary point; is the viscosity value of the RBF interpolation model at the boundary point ; is the viscosity gradient vector of the RBF interpolation model at the boundary point ; denotes the dot product operation of vectors.

7. A molten slag stirring control method according to claim 1, wherein The calculation of the final output viscosity according to the initial viscosity estimation value, the corrected viscosity estimation value and the total confidence comprises the following steps: The final output viscosity is obtained by weighted summation of the total confidence as a weight of the initial viscosity estimation value and a difference between a value 1 and the total confidence as a weight of the corrected viscosity estimation value.

8. A molten slag stirring control method according to claim 1, wherein The calculation of the target rocking angle of the rocking furnace according to the final output viscosity comprises the following steps: According to the target mixing intensity index , the set kneading frequency , and the final output viscosity , the target kneading angle is obtained according to the calculation formula .

9. A molten slag stirring control method according to claim 8, wherein the target mixed intensity index The setting method includes: during the oxidation period, setting the target mixed intensity index to a first range; and during the reduction period and the desulfurization period, setting the target mixed intensity index to a second range, the value of the second range being higher than the first range.

10. A molten slag stirring control system, characterized by, The method comprises the following steps: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a method for controlling slag rocking according to any one of claims 1-9 is implemented.

Images