Metallurgical slag viscosity prediction method and system

Abstract

The embodiment of the invention provides a metallurgical slag viscosity prediction method and system, and the method can obtain an initial sample set through random sampling after setting a constraint condition. Calculating viscosity values by using a preset viscosity calculation model to generate a sample viscosity value set; and constructing a feature sample set containing subinterval endpoint values according to a partial dependency graph generated by the sample viscosity value set, obtaining a theoretical value set and a true value set based on the feature sample set, and calculating an error set so as to use the error set to correct the sample viscosity value set and generate a corrected sample set. And constructing a slag viscosity prediction model based on the corrected sample set, and predicting the slag viscosity value by using the slag viscosity prediction model. According to the method, theoretical viscosity data corrected by a test and machine learning modeling are combined, nonlinear relations among the slag viscosity, the compound components and the temperature are captured, and the prediction precision can be improved.

Description

Technical Field

[0001] The present application relates to the field of metallurgical process control, and in particular to a method and system for predicting the viscosity of metallurgical slag. Background Art

[0002] The viscosity of metallurgical slag is a key physical parameter in the smelting process, which can affect the fluidity of the slag, the metal reduction efficiency and the reaction process in the furnace. In the smelting process, the smelting process parameters need to be set according to the viscosity of the metallurgical slag. However, under high temperature conditions, the slag viscosity measurement has problems such as poor composition uniformity, high time consumption and feedback lag, which limits the application of the viscosity of metallurgical slag in the production process. Therefore, it is necessary to predict the viscosity of metallurgical slag.

[0003] In order to predict the viscosity of metallurgical slag, a viscosity prediction model can be constructed based on metallurgical slag samples. The viscosity prediction model can include empirical models, structural models, thermodynamic models, and models based on machine learning. Based on the constructed viscosity prediction model, the viscosity of metallurgical slag can be predicted, so that metallurgical process parameters can be set in advance according to the prediction results.

[0004] For example, a viscosity prediction model can be constructed based on the Arrhenius equation, that is, by selecting a range of slag components, designing a quadratic regression orthogonal test for the range of slag components, and then using analytically pure reagents to prepare slag of the above components, and using the rotating cylinder method to test the slag viscosity to obtain the viscosity value, and then according to the slag components and viscosity values, the viscosity parameter equation is built according to the Arrhenius equation to construct a slag viscosity prediction model. However, when using the viscosity prediction model for viscosity prediction, it is easy to generate errors when the data range is exceeded, making the slag viscosity prediction method low in accuracy and poor in real-time performance. Summary of the invention

[0005] In view of this, an embodiment of the present application provides a metallurgical slag viscosity prediction method and system to solve the problem of low accuracy and poor real-time performance of the slag viscosity prediction method.

[0006] According to one aspect of the present application, a method for predicting metallurgical slag viscosity is provided, the method comprising: An initial sample set is obtained by random sampling within preset constraints, wherein the constraints include a mass fraction interval and a temperature interval of the slag-based compound of the metallurgical slag to be predicted; the initial sample set includes a plurality of parameter points; the parameter points are a combination of mass fraction values ​​and temperature values ​​obtained by random sampling; Calculating the viscosity values ​​corresponding to the parameter points in the initial sample set using a preset viscosity calculation model to generate a sample viscosity value set; Generating a partial dependence graph according to the sample viscosity value set, and constructing a characteristic sample set based on the endpoint values ​​of the partial dependence graph, wherein the partial dependence graph is used to characterize the response of a prediction variable to a viscosity value, wherein the prediction variable is a slag-based compound of a metallurgical slag to be predicted; Acquire a theoretical value set, a true value set, and an error set, wherein the theoretical value set includes viscosity values ​​obtained by calculating the characteristic sample set using the viscosity calculation model; the true value set includes viscosity values ​​obtained by testing the characteristic sample set through experiments; and the error set includes error values ​​obtained by performing difference calculation on the theoretical value set and the true value set; Correcting the set of sample viscosity values ​​using the initial sample set and the error set to generate a corrected sample set; A slag viscosity prediction model is constructed based on the corrected sample set, and the slag viscosity value is predicted using the slag viscosity prediction model.

[0007] Optionally, using a preset viscosity calculation model to calculate the viscosity values ​​corresponding to the parameter points in the initial sample set to generate a sample viscosity value set includes: Calling a preset viscosity calculation model, wherein the viscosity calculation model is one of an empirical model, a structural model, a molecular dynamics model, and a thermodynamic application model; Inputting the parameter points in the initial sample set into the viscosity calculation model one by one to obtain the viscosity value output by the viscosity calculation model; The viscosity value is added to the initial sample set to generate the sample viscosity value set, in which the viscosity value has an associated relationship with the combination of the mass fraction value and the temperature value corresponding to the parameter point.

[0008] Optionally, generating a partial dependence graph according to the set of sample viscosity values ​​includes: Applying a plurality of machine learning algorithms to the sample viscosity value set to perform modeling to obtain a plurality of first candidate models; Obtaining quantifiable indicators of the first candidate model, wherein the quantifiable indicators include one or more combinations of root mean square error, mean square error, determination coefficient, mean absolute percentage error, and mean absolute error; Selecting the optimal model from the plurality of candidate models according to the quantifiable indicators; A partial dependence plot of all predictor variables on viscosity values ​​was generated using the optimal model.

[0009] Optionally, constructing a feature sample set based on the endpoint values ​​of the partial dependence graph includes: Based on a preset interval partitioning algorithm, the partial dependence graph is divided into a plurality of sub-intervals, wherein the preset interval partitioning algorithm includes one or more combinations of an average distribution method, a response distribution method, and a random distribution method; Record the endpoint value of each sub-interval where the predictor variable is located; The endpoint values ​​satisfying the constraint conditions are combined to construct the feature sample set.

[0010] Optionally, based on a preset interval partitioning algorithm, the partial dependency graph is divided into a plurality of sub-intervals, including: Obtaining a curve function in the partial dependence graph and a definition domain of the curve function; Extracting the monotonic interval of the curve function in the definition domain; In the monotonic interval, the domain is evenly divided into a first number of subintervals, and the length of each subinterval is calculated; A segmentation point is calculated according to the subinterval length, and the monotonic interval is divided into a first number of subintervals using the segmentation point.

[0011] Optionally, based on a preset interval partitioning algorithm, the partial dependency graph is divided into a plurality of sub-intervals, including: Obtain the curve function in the partial dependence graph and the domain and range of the curve function; Extracting the monotonic interval of the curve function in the definition domain; In the monotonic interval, the value range is evenly divided into a second number of sub-intervals, and the length of each sub-interval is calculated; Calculate the split point according to the sub-interval length; The predicted variable value corresponding to the segmentation point is calculated, and the monotonic interval is divided into a second number of sub-intervals according to the predicted variable value.

[0012] Optionally, based on a preset interval partitioning algorithm, the partial dependency graph is divided into a plurality of sub-intervals, including: Obtaining a curve function in the partial dependence graph and a definition domain of the curve function; Extracting the monotonic interval of the curve function in the definition domain; In the monotonic interval, determining the number of subintervals, and randomly generating a third number of segmentation points, where the third number is equal to the number of subintervals minus 1; The monotonic interval is divided into a third number of sub-intervals according to the segmentation points.

[0013] Optionally, using the initial sample set and the error set to correct the sample viscosity value set to generate a corrected sample set includes: Extracting parameter points from the initial sample set; Searching for a target point in the feature sample set based on the parameter point, the target point being an endpoint value closest to the parameter point; Based on the error set, recording the error value corresponding to the target point; Calculating the arithmetic mean of multiple error values ​​corresponding to the same target point, and establishing a slag sample viscosity value error set according to the arithmetic mean; The slag sample viscosity value error set is added to the viscosity values ​​in the sample viscosity value set to obtain a corrected sample set.

[0014] Optionally, constructing a slag viscosity prediction model based on the corrected sample set includes: Dividing the modified sample set into a training set and a validation set; Applying a plurality of machine learning algorithms to the training sets in the revised sample set to perform modeling to obtain a plurality of second candidate models; Calculate a quantifiable index of the second candidate model using a validation set in the revised sample set, wherein the quantifiable index includes a root mean square error; The slag viscosity prediction model is extracted based on the quantifiable index, and the slag viscosity prediction model is a model with the smallest root mean square error among multiple second alternative models.

[0015] According to another aspect of the present application, a metallurgical slag viscosity prediction system is provided, the system comprising: A sampling module, used for obtaining an initial sample set by random sampling within preset constraints, wherein the constraints include a mass fraction interval and a temperature interval of the slag-based compound of the metallurgical slag to be predicted; the initial sample set includes a plurality of parameter points; the parameter points are a combination of a mass fraction value and a temperature value obtained by random sampling; A theoretical viscosity calculation module, used to calculate the viscosity values ​​corresponding to the parameter points in the initial sample set using a preset viscosity calculation model to generate a sample viscosity value set; A feature extraction module, used for generating a partial dependence graph according to the sample viscosity value set, and constructing a feature sample set based on the endpoint values ​​of the partial dependence graph, wherein the partial dependence graph is used for characterizing the response of a prediction variable to a viscosity value, wherein the prediction variable is a slag-based compound of the metallurgical slag to be predicted; an error calculation module, for obtaining a theoretical value set, a real value set and an error set, wherein the theoretical value set includes viscosity values ​​obtained by calculating the characteristic sample set using the viscosity calculation model; the real value set includes viscosity values ​​obtained by testing the characteristic sample set through experiments; and the error set includes error values ​​obtained by performing difference calculation on the theoretical value set and the real value set; A correction module, configured to correct the sample viscosity value set using the initial sample set and the error set to generate a corrected sample set; The viscosity prediction module is used to construct a slag viscosity prediction model based on the corrected sample set, and use the slag viscosity prediction model to predict the slag viscosity value.

[0016] According to another aspect of the present application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor implements the above-mentioned metallurgical slag viscosity prediction method when executing the program.

[0017] According to another aspect of the present application, a storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned metallurgical slag viscosity prediction method is implemented.

[0018] By means of the above technical scheme, the embodiment of the present application provides a metallurgical slag viscosity prediction method and system, wherein the method can obtain an initial sample set by random sampling after setting constraints. Then, the viscosity value is calculated using a preset viscosity calculation model to generate a sample viscosity value set. Then, a feature sample set containing subinterval endpoint values ​​is constructed based on the partial dependency graph generated by the sample viscosity value set, and a theoretical value set and a true value set are obtained based on the feature sample set, and an error set is calculated to correct the sample viscosity value set using the error set to generate a corrected sample set. Thus, a slag viscosity prediction model is constructed based on the corrected sample set, and the slag viscosity value is predicted using the slag viscosity prediction model. The method can improve the prediction accuracy by combining the experimentally corrected theoretical viscosity data with machine learning modeling to capture the nonlinear relationship between slag viscosity and compound composition and temperature. In addition, a sample parameter set is generated by random sampling and the modeling process is optimized to reduce redundant calculations and experimental complexity, and combined with the optimal model screening, the prediction efficiency can be improved. The method is compatible with multiple metallurgical slag systems and multi-dimensional changes in their chemical composition and temperature. By introducing experimental correction and multi-level optimization, it is suitable for viscosity prediction under complex production conditions and provides reliable technical support for metallurgical process optimization and control. The method can also be adjusted according to the needs of different scenarios by flexibly selecting a variety of machine learning models and evaluation indicators, and has good generalization ability and application potential.

[0019] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A schematic diagram of a process for predicting the viscosity of metallurgical slag provided in an embodiment of the present application; Figure 2 A schematic diagram of the viscosity prediction model construction and viscosity prediction process provided in the embodiments of the present application; Figure 3 A schematic diagram of a partial dependence diagram of CaO provided in an embodiment of the present application; Figure 4 A schematic diagram of a partial dependence diagram of SiO2 provided in an embodiment of the present application; Figure 5 A schematic diagram of a partial dependence diagram of MgO provided in an embodiment of the present application; Figure 6 A schematic diagram of a partial dependence diagram of Al2O3 provided in an embodiment of the present application; Figure 7 A schematic diagram of a partial dependence diagram of temperature provided in an embodiment of the present application; Figure 8 A schematic diagram of the structure of a metallurgical slag viscosity prediction system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0022] In the embodiment of the present application, slag is a by-product produced in the metal smelting process. Slag is formed by chemical reactions between impurities in metal ore and flux under high temperature conditions. Slag, as a mixture in a molten state, can include a variety of compounds. Taking the steelmaking process as an example, slag can include oxides in metallurgical raw materials or oxides generated in the metallurgical process, such as CaO, FeO, MnO, MgO, Al2O3, SiO2, P2O5, Fe2O3, etc. The slag may contain a small amount of other types of compounds, such as fluoride (CaF2), chloride (NaCl), sulfide (CaS, MnS), sulfate, etc.

[0023] As a fluid mixture, slag has a certain viscosity. The viscosity of metallurgical slag is a key physical parameter in the smelting process, which can affect the fluidity of slag, metal reduction efficiency and reaction process in the furnace.

[0024] The viscosity of slag can be affected by many factors, including the chemical composition, temperature and structure of the slag. For example, in copper smelting slag, the viscosity of solid-phase slag can be calculated by combining the Roscoe equation with the multivariate multiphase equilibrium calculation and pure liquid slag viscosity calculation functions of FactSage software. The viscosity of copper smelting slag is related to the content of iron oxides, that is, the viscosity of copper smelting slag decreases as the content of iron oxides increases. When the mass fraction of SiO2 is lower than 25% or higher than 40%, a large amount of solid phase substances will appear in the system, causing the viscosity of the slag to increase sharply.

[0025] In the metal smelting process, the smelting process parameters can be set according to the viscosity adaptability of the metallurgical slag. However, under high temperature conditions, the slag viscosity measurement has problems such as poor composition uniformity, high time consumption and feedback lag, which limits the application of the viscosity of the metallurgical slag in the production process. Therefore, it is necessary to predict the viscosity of the metallurgical slag. In some embodiments, the viscosity of the metallurgical slag can be predicted based on empirical models, structural models, thermodynamic models and machine learning models. However, when using the viscosity prediction model for viscosity prediction, it is easy to generate errors when the data range is exceeded, making the slag viscosity prediction method low in accuracy and poor in real time.

[0026] In order to solve the problem of low accuracy and poor real-time performance of the slag viscosity prediction method, a metallurgical slag viscosity prediction method is provided in this embodiment. Figure 1 , Figure 2 As shown, the method includes: S100, obtaining an initial sample set by random sampling within preset constraints.

[0027] When constructing a viscosity prediction model, constraints may be specified first. The constraints may include the type of slag-based compounds, mass fraction range, and temperature range of the metallurgical slag to be predicted. The types of compounds of the metallurgical slag-based compounds may include, but are not limited to: SiO2, Al2O3, CaO, MgO, MnO, FeO, Fe2O3, NiO, PbO, ZnO, Li2O, Na2O, B2O3, TiO2, and Ti2O3.

[0028] The mass fraction range of slag compounds in the constraint conditions can be set for each compound according to the specific chemical composition of the metallurgical slag to be predicted. And the mass fraction of each compound must satisfy the sum of the mass fractions of all compounds to be 100%. The temperature range in the constraint conditions can be set according to the melting temperature range of the metallurgical slag to be predicted. Different temperature ranges can be set for different metallurgical slags.

[0029] In some embodiments, the preset constraints may also include the ratio of basic oxides to acidic oxides, including binary basicity, ternary basicity, quaternary basicity, etc. The ratio of basic oxides to acidic oxides can reflect the material's resistance to acidity and chemical properties. Among them, binary basicity is the mass ratio of the main basic oxides (such as calcium oxide CaO) to the main acidic oxides (such as silicon oxide SiO2) in the slag during the ironmaking process. Binary basicity can be used to determine the fluidity, viscosity and interaction between the slag and the molten metal. Ternary basicity is based on binary basicity and also considers magnesium oxide (MgO), that is, the ratio of the alkaline components of calcium oxide CaO and magnesium oxide (MgO) to silicon oxide. Compared with binary basicity, ternary basicity can more comprehensively describe the alkalinity strength and reaction characteristics of the slag system. Quaternary basicity refers to the ratio of the total mass of calcium oxide (CaO) and magnesium oxide (MgO) in the slag to the mass percentage content of silicon dioxide (SiO2) and aluminum oxide (Al2O3). Quaternary basicity can be applied to the situation where the magnesium oxide (MgO) content in the metallurgical slag is high.

[0030] In addition to the ratios of basic oxides to acidic oxides such as binary basicity, ternary basicity, and quaternary basicity mentioned above, the preset constraints can also define the ratio range between several specific compounds according to specific analysis needs, such as the mass fraction ratio of calcium oxide (CaO) and magnesium oxide (MgO).

[0031] When setting the compound type of the metallurgical slag system, one or more compounds among the above compounds can be selected. In some embodiments, different constraints can be set according to the specific application field and component content of the metallurgical slag.

[0032] For example, according to the specific application scenario of metallurgical slag, the preset constraints include compound type, compound mass fraction range, temperature range and binary basicity range. That is, the compound types of the metallurgical slag system to be predicted can be specified as: CaO, SiO2, MgO, Al2O3. The mass fraction range of each compound is: w (CaO): 20%~60%, w (SiO2): 20%~60%, w (MgO): 5%~20%, w (Al2O3): 5%~20%. Slag system temperature range: 1400~1700℃. Other constraints are: binary basicity w (CaO) / w (SiO2) range: 0.9~1.8.

[0033] After setting the constraints, random sampling can be performed within the value range set by the constraints to obtain an initial sample set. That is, within the compound composition, compound mass range, and temperature range specified in the constraints, a number of mass fraction values ​​and temperature values ​​containing all specified types of compounds are extracted using random sampling. The extracted mass fraction values ​​and temperature values ​​are combined to form multiple parameter points, thereby constructing an initial sample set A based on the multiple parameter points. The initial sample set A is used to characterize the slag sample parameters, so the initial sample set includes multiple parameter points, and the parameter points are combinations of mass fraction values ​​and temperature values ​​obtained by random sampling.

[0034] In some embodiments, when random sampling is performed within the value range set by the constraint condition, it can be based on a specific random sampling method, which includes but is not limited to: simple random sampling method, Dirichlet distribution sampling method, low-discrepancy sequence method, etc. Through any of the above random sampling methods, an initial sample set can be obtained that at least includes a point set including all compound mass fractions and corresponding slag system temperatures, that is: ; Wherein, A represents the initial sample set, including multiple slag sample parameters; w i Indicates i The mass fraction of the compound, T Indicates the corresponding slag system temperature.

[0035] For example, within the compound composition and compound mass fraction range specified by the above constraints, 3000 mass fraction values ​​and temperature values ​​of all types of compounds are extracted using a simple random sampling method to construct an initial sample set A, as shown in Table 1: Table 1, initial sample set A;

[0036] In some embodiments, during the sampling process using the random sampling method, it is also possible to determine whether the extracted parameter points exceed the compound composition, compound mass fraction interval, and temperature interval specified by the constraint conditions. If the random sampling results exceed the interval range specified in the constraint conditions, the group of data can be ignored and the next group of samples can be performed until the number of the initial sample set A reaches the preset sample capacity.

[0037] S200: Calculate the viscosity values ​​corresponding to the parameter points in the initial sample set using a preset viscosity calculation model to generate a sample viscosity value set.

[0038] After setting the constraint conditions and obtaining the initial sample set by random sampling within the preset constraint conditions, the preset viscosity calculation model can be called and the viscosity values ​​corresponding to the parameter points in the initial sample set can be calculated using the preset viscosity calculation model.

[0039] Among them, the preset viscosity calculation model is a model or tool for performing viscosity calculation of compound composition and temperature coupling. In some embodiments, the preset viscosity calculation model is one of an empirical model, a structural model, a molecular dynamics model, and a thermodynamic application model. For example, the preset viscosity calculation model is a composition-temperature coupling viscosity calculation model constructed based on the National Physical Laboratory (NPL). For another example, the viscosity calculation model is an empirical model used in the metallurgical field to predict the viscosity values ​​of slag systems with different components, such as the Riboud model and the Urbain model. The Riboud model is based on the Weymann-Frenkel viscosity formula and the Urbain model, combined with the blast furnace slag viscosity data, to derive the relationship between slag composition and viscosity value. The Urbain model is suitable for describing the relationship between the viscosity and temperature of silicate melts, supercooled liquids, and glasses. For another example, the viscosity calculation model is a prediction tool constructed based on thermodynamic software such as FactSage.

[0040] After calculating the viscosity value of each parameter point in the initial sample set A by the preset viscosity calculation model, the viscosity value can be combined with the initial sample set A to obtain a sample viscosity value set B. The sample viscosity value set may include a combination of the mass fraction value of each compound and the corresponding temperature value, and a viscosity value calculated by the viscosity calculation model under the combination of the mass fraction value of the compound and the corresponding temperature value. That is, the sample viscosity value set B is the calculated viscosity value corresponding to all parameter points of the initial sample set A, that is: ; Wherein, B represents the set of sample viscosity values; A represents the initial sample set; η Indicates the calculated viscosity value.

[0041] Therefore, in some embodiments, when using a preset viscosity calculation model to calculate the viscosity values ​​corresponding to the parameter points in the initial sample set to generate a sample viscosity value set, the preset viscosity calculation model can be called first, and then the parameter points in the initial sample set are input into the viscosity calculation model one by one to obtain the viscosity values ​​output by the viscosity calculation model. The viscosity values ​​are then added to the initial sample set to generate the sample viscosity value set. It can be seen that in the sample viscosity value set, the viscosity value has an association relationship with the mass fraction value and temperature value combination corresponding to the parameter point. In the sample viscosity value set, the association relationship between the parameter point and the predicted viscosity value can be represented by setting them in the same row or column.

[0042] For example, the NPL method is used to calculate the viscosity values ​​of all parameter points in the initial sample set A, thereby establishing the sample viscosity value set B, as shown in Table 2: Table 2, sample viscosity value set B;

[0043] Among them, in the process of calculating the viscosity value based on the NPL method, the viscosity value can be calculated according to the following formula: ; in, μ represents the viscosity of the system, which is a measure of the internal resistance of the fluid; exp() represents the exponential function, which is the power of the base of the natural logarithm, e; A Represents a constant related to the exponential function, as the coefficient of the exponential function; B represents a constant related to the system; T Represents the thermodynamic temperature.

[0044] Among them, the coefficients of the exponential function A It can be calculated according to the following formula: ; System-dependent constants B It can be calculated according to the following formula: ; In the formula, Λ To correct the optical alkalinity, it can be calculated by the following formula: ; in, X i For the i The corrected mole fraction of the component; n i For the i The number of oxygen in the components; Λ i For the iThe optical basicity of the components, illustratively, is as follows: CaO is 1; SiO2 is 0.48; MgO is 0.78; Al2O3 is 0.6.

[0045] It should be noted that in order to accurately obtain the correlation between compound composition, temperature value and viscosity value, when calculating the viscosity value through the composition-temperature coupled viscosity calculation model or tool, the selected calculation model should ensure that the viscosity calculation can be performed on all parameter points in the constructed initial sample set A, and only one model or tool can be selected for all viscosity calculations within one calculation cycle.

[0046] S300, generating a partial dependence graph according to the sample viscosity value set, and constructing a feature sample set based on the endpoint values ​​of the partial dependence graph.

[0047] After obtaining the sample viscosity value set B, regression analysis can be performed based on the sample viscosity value set B to determine the response relationship function of the predicted variable to the predicted value. The response relationship function can be represented in the form of a partial dependence graph. For the metallurgical slag viscosity prediction process, the predicted variable is the mass fraction of the slag-based compound in the slag, and the predicted value is the viscosity value calculated by the calculation model. Therefore, the partial dependence graph can include the independent variable coordinates and the dependent variable coordinates, the independent variable coordinates are the mass fraction value or temperature value of a slag-based compound, and the dependent variable coordinates are the viscosity value.

[0048] In order to determine the response relationship function of the prediction variables to the predicted values, in some embodiments, a variety of machine learning methods can be applied to model, and the optimal model can be selected based on quantifiable evaluation indicators, so that the selected optimal model can generate a partial dependence diagram of the response of all prediction variables to the predicted values.

[0049] That is, in some embodiments, when generating a partial dependence graph based on the sample viscosity value set, multiple machine learning algorithms may be applied to the sample viscosity value set to perform modeling to obtain multiple first candidate models, and then quantifiable indicators of the first candidate models are obtained, and the optimal model is selected from the multiple candidate models according to the quantifiable indicators, and a partial dependence graph of all predicted variables responding to viscosity values ​​is generated through the optimal model.

[0050] The multiple machine learning algorithms may include but are not limited to support vector machine (SVM) model, Gaussian process regression model, efficient linear model, kernel model, integrated model, neural network model, tree model, linear regression model, stepwise linear regression model and sub-models of each model with different hyperparameters; the quantifiable evaluation indicators may include but are not limited to root mean square error (RMSE), mean square error (MSE), determination coefficient ( R 2 ), Mean Absolute Percentage Error (MAPE), and Mean Absolute Error (MAE).

[0051] For example, for the sample viscosity value set B shown in Table 2, a variety of machine learning methods can be applied to modeling, including: SVM model, Gaussian process regression model, efficient linear model, kernel model, integrated model, neural network model, tree model, linear regression model, stepwise linear regression model and sub-models with different hyperparameters for each model, totaling 28 models. The 5-fold cross-validation method is used for machine learning, so that the optimal model is selected according to the RMSE result calculated on the validation set, and the optimal model generates a partial dependence diagram of all predictor variables on the predicted response. The RMSE results of the validation set of each model are shown in Table 3: Table 3, RMSE results of model validation set;

[0052] According to the minimum RMSE principle, model No. 8 is selected as the optimal model, and the partial dependence diagram of the mass fraction values ​​of CaO, SiO2, MgO, Al2O3 and temperature values ​​on the predicted viscosity values ​​is obtained, as shown in Figure 3 , Figure 4 , Figure 5 , Figure 6 , Figure 7 shown.

[0053] After the partial dependence graph is generated, the partial dependence graph may be divided into a plurality of sub-intervals based on a preset interval partitioning algorithm, wherein the preset interval partitioning algorithm includes one or more combinations of an average distribution method, a response distribution method, and a random distribution method.

[0054] In some embodiments, when the partial dependence graph is divided into a plurality of subintervals using the average distribution method, the curve function in the partial dependence graph and the domain of the curve function may be first obtained, and then the monotonic interval of the curve function in the domain may be extracted. Within the monotonic interval, the domain is evenly divided into a first number of subintervals, and the length of each subinterval is calculated. Then, a split point is calculated based on the subinterval length, and the monotonic interval is divided into the first number of subintervals using the split point.

[0055] For example, for each partial dependence graph curve function f i ( x ), and its domain D i =[ a i , b i ]. You can first determine f i ( x )exist D i All monotonic intervals on I i,m =[ c i,m , d i,m ]( m =1,2,……, M i ). In each monotonic interval I i,m The domain is divided equally into k i,m subintervals, the length of each subinterval is: ; in, c i,m is the left endpoint of the monotonic interval, d i,m is the right endpoint of the monotonic interval; i,m is the number of sub-interval divisions.

[0056] Based on the length of each subinterval and the number of subinterval divisions, the split point can be calculated, that is: ; in, x i,m,j Indicates i The compound j Split points; c i,m is the left endpoint of the monotonic interval; hi,m is the length of the subinterval.

[0057] Thus, according to the calculated segmentation points, I i,m Split into 𝑘 i,m subintervals, that is, each subinterval can be expressed as: c i,m , x i,m,1 ],[ x i,m,1 , x i,m,2 ],…,[ x i,m,ki,m-1 , d i,m ]; then record the predicted variable value set of all sub-interval endpoints, and obtain the following endpoint value set: ; in, c i,m is the left endpoint of the monotonic interval; x i,m,j Indicates i The compound j endpoints; d i,m is the right endpoint of the monotonic interval; i,m is the number of sub-intervals, that is, based on the average distribution method, the curve function in the partial dependence diagram can be evenly divided according to the domain. Within the monotonic interval of the domain, 𝑘 can be recorded for each compound. i,m-1 endpoint values.

[0058] In some embodiments, when the response distribution method is used to divide the partial dependence graph into multiple sub-intervals, after obtaining the curve function in the partial dependence graph and the definition domain and value range of the curve function, the monotonic interval of the curve function in the definition domain can be extracted. And within the monotonic interval, the value range is evenly divided into a second number of sub-intervals, and the length of each sub-interval is calculated. Then, the segmentation point is calculated according to the sub-interval length, and the predicted variable value corresponding to the segmentation point is calculated, so that the monotonic interval is divided into the second number of sub-intervals according to the predicted variable value.

[0059] For example, for each partial dependence graph curve function f i ( x ) and its domain D i = [ a i , b i ], you can first determinef i ( x )exist D i All monotonic intervals on I i,m =[ c i,m , d i,m ]( m =1,2,……, M i ). In each monotonic interval I i,m In, get the value range corresponding to the curve function, that is: ; in, f i ( I i,m ) indicates that the curve function is in the monotonic interval I i,m The value range within y i,m min Indicates that on a monotonic interval I i,m Inside f i ( I i,m ) minimum value; y i,m max Indicates that on a monotonic interval I i,m Inside f i ( I i,m ). By dividing the value range corresponding to the curve function into 𝑘 i,m subintervals, the length of each subinterval can be determined as: ; In the formula, Δ y i,m represents the length of each subinterval, 𝑘 i,m Indicates the number of sub-intervals. According to the length of each sub-interval Δ y i,m The split point can be calculated, namely: ; in, y i,m,j Indicates i The compound j Split points; y i,mmin Indicates that on a monotonic interval I i,m Inside f i ( I i,m ), which is the lower end point of the range; Δ y i,m Indicates the length of each subinterval.

[0060] And calculate the corresponding split point x value: ; Thus, according to the sub-interval segmentation method provided in the above embodiment, according to the corresponding x value, I i,m Split into k i,m subintervals, and collect the sets of predictor values ​​at the endpoints of all subintervals to obtain the set of endpoint values.

[0061] In some embodiments, when a partial dependence graph is divided into multiple subintervals using a random distribution method, after obtaining the curve function and the definition domain of the curve function in the partial dependence graph and extracting the monotonic interval of the curve function in the definition domain, the number of subintervals can be determined within the monotonic interval, and a third number of segmentation points can be randomly generated. The third number is equal to the number of subintervals minus 1. The monotonic interval is then divided into the third number of subintervals according to the segmentation points.

[0062] For example, for each partial dependence graph the curve function f i ( x ) and its domain D i = [ a i , b i ], the curve function can be determined f i ( x ) in the domain D i All monotonic intervals on I i,m =[ c i,m , d i,m ]( m =1,2,……, M i ). In each monotonic interval I i,mWithin, determine the number of subintervals 𝑘 i,m , then randomly generate 𝑘 in this interval based on random distribution methods such as uniform distribution and normal distribution i,m -1 split point, namely: ; in, x i,m,j Indicates i The compound j The split points can satisfy c i,m < x i,m,1 < x i,m,2 <……< x i,m,ki,m-1 < d i,m Then, according to the randomly generated split points, I i,m Split into k i,m subintervals, and a set of endpoint 𝑥 values ​​that collects all subintervals.

[0063] By dividing the partial dependence graph into multiple sub-intervals through the sub-interval division method provided in the above embodiment, the endpoint values ​​of each sub-interval where the prediction variable is located can be recorded, and the endpoint values ​​that meet the constraint conditions can be combined to construct the feature sample set.

[0064] That is, after dividing the sub-intervals, the values ​​of each prediction variable at the endpoints of each interval can be recorded, and the endpoint values ​​of all prediction variables can be combined within the compound component mass fraction interval and temperature interval specified in the preset constraints to construct a feature sample set C. The feature sample set C is used to characterize the characteristic slag sample parameters of the metallurgical slag to be predicted.

[0065] The characteristic sample set C is the combination of each prediction variable at the end point of the partition interval, and the point set that satisfies the mass fraction interval range of the compound components specified in the preset constraints, that is: ; Where C represents the feature sample set; x represents the element in the feature sample set; w i Indicates the mass fraction value of the compound; T represents the temperature value; then each w i ( i =1, 2, …, n )and T is divided into several interval endpoints, denoted as { ai1 , a i2 , …, a ikj}and{ b 1, b 2, ..., b kT}.

[0066] For example, after obtaining the partial dependence graph of CaO, SiO2, MgO, Al2O3 content and temperature on the predicted viscosity value, each partial dependence graph is divided into 4 intervals according to the response distribution method, and the value of each predicted variable at the endpoint of each interval is recorded. The endpoint values ​​of all predicted variables are combined within the compound component mass fraction interval and temperature interval specified by the preset constraints to construct the characteristic sample set C, as shown in Table 4: Table 4, Feature sample set C;

[0067] According to the feature sample set C shown in Table 4, the endpoint values ​​of each predictor variable in each sub-interval can be determined as shown in Table 5: Table 5, endpoint values ​​of the predictor variables in each subinterval;

[0068] S400, obtaining a theoretical value set, a true value set, and an error set.

[0069] After constructing the characteristic sample set, theoretical value calculation and analysis test can be performed on the constructed characteristic sample set to obtain a theoretical value set and a real value set. The theoretical value set includes viscosity values ​​calculated by using the viscosity calculation model on the characteristic sample set.

[0070] In order to calculate the theoretical value, the same viscosity calculation model as step S200 can be used, and the viscosity calculation model can be used to calculate the theoretical value according to the data points in the feature sample set C. That is, in some embodiments, a preset viscosity calculation model can be called first. Among them, the viscosity calculation model is one of an empirical model, a structural model, a molecular dynamics model, and a thermodynamic application model. Then, the data points in the feature sample set C are input into the viscosity calculation model one by one to obtain the theoretical viscosity value output by the viscosity calculation model, and then the theoretical viscosity value is added to the feature sample set C to generate a theoretical value set D.

[0071] Similarly, in the theoretical value set D, the theoretical viscosity value has an associated relationship with the combination of the mass fraction endpoint value and the temperature endpoint value corresponding to the data point. It can be seen that the theoretical value set D is a point set containing the calculated viscosity values ​​corresponding to all data points of the feature sample set C, that is: ; Where D represents the theoretical value set; C represents the feature sample set; η Indicates the calculated theoretical viscosity value.

[0072] For example, based on the component-temperature coupled viscosity calculation model or tool, the viscosity values ​​of all points in the viscosity value calculation set C are calculated to establish a theoretical value set D, as shown in Table 6: Table 6, theoretical value set D;

[0073] The true value set includes viscosity values ​​obtained by testing the characteristic sample set. The viscosity values ​​of all data points in the characteristic sample set C can be obtained by experimental methods, and a true value set E is established, which is used to characterize the true viscosity values ​​of the characteristic slag samples.

[0074] The test methods include capillary method, oscillating container method, rotation method and oscillating plate method, etc. The true value set E is a point set containing the test viscosity values ​​corresponding to all data points of the feature sample set C, that is: ; Among them, E represents the true value set; C represents the feature sample set; Indicates the test viscosity value.

[0075] For example, the rotation method is used to obtain the true viscosity values ​​of all data points in the characteristic sample set C, and the true value set E shown in Table 7 is established: Table 7, true value set E;

[0076] After obtaining the theoretical value set and the true value set according to the method provided in the above embodiment, error calculation can also be performed based on the theoretical value set and the true value set to obtain an error set. That is, the error set includes error values ​​obtained by performing difference calculation on the theoretical value set and the true value set.

[0077] According to the theoretical value set D and the true value set E, a mathematical method can be used to establish an error set F of the viscosity value of the characteristic slag sample. The mathematical method can be to subtract the experimental viscosity value in the true value set E from the calculated viscosity value in the theoretical value set D, or to subtract the calculated viscosity value in the theoretical value set D from the experimental viscosity value in the true value set E. The error set F obtained by calculation is used to characterize the error of the viscosity value of the slag sample. Therefore, the error set F is a point set containing the difference values ​​of the viscosity values ​​obtained by mathematical processing corresponding to all data points of the characteristic sample set C, that is: ; Among them, F represents the error set; C represents the feature sample set; Δ η It represents the difference of viscosity values ​​obtained by mathematical processing.

[0078] For example, the experimental viscosity values ​​in the true value set E are subtracted from the calculated viscosity values ​​in the theoretical value set D to establish the error set F of the viscosity values ​​of the characteristic slag samples, as shown in Table 8: Table 8, Error set F;

[0079] S500: Correct the sample viscosity value set using the initial sample set and the error set to generate a corrected sample set.

[0080] After the error set F is obtained, the sample viscosity value set B may be corrected based on the error set F and the initial sample set A, so as to correct the prediction results of the viscosity values ​​in the sample viscosity value set B and obtain a corrected sample set.

[0081] In some embodiments, in order to generate a corrected sample set, when the sample viscosity value set is corrected using the initial sample set and the error set, a parameter point can be extracted from the initial sample set. And based on the parameter point, a target point is searched in the feature sample set. Wherein, the target point is the endpoint value closest to the parameter point. And based on the error set, the error value corresponding to the target point is recorded. Then the arithmetic mean of multiple error values ​​corresponding to the same target point is calculated, and a slag sample viscosity value error set is established based on the arithmetic mean. Thereby, the slag sample viscosity value error set is added to the viscosity value in the sample viscosity value set to obtain a corrected sample set.

[0082] After obtaining the error set F, all parameter points in the initial sample set A can be searched in the feature sample set C for the closest point to the parameter point. k point and record this k The points correspond to the error values ​​in the error set F.k The arithmetic mean of the error values ​​is taken to establish the slag sample viscosity value error set G. Then the sample viscosity value set B is corrected correspondingly to the slag sample viscosity value error set G to establish the corrected sample set H.

[0083] Among them, the search distance is the closest k The method of points can be the nearest neighbor algorithm (k-NN), and k The value of the parameter should not be less than the number of compound types. Therefore, the slag sample viscosity value error set G is the point set of viscosity value errors corresponding to all parameter points of the initial sample set A after processing, that is: ; Among them, G is the error set of slag sample viscosity values; A is the initial sample set; Δ η Indicates the error of the viscosity value at that point.

[0084] Then, the sample viscosity value set B is corrected according to the corresponding correction method, so as to obtain the corrected sample set H. The sample correction set H includes the point set of the corrected viscosity values ​​of all data points in the sample viscosity value set B, that is: ; Among them, H represents the sample correction set; A represents the initial sample set; η 'Indicates the corrected viscosity value at that point.

[0085] For example, for all points in the initial sample set A, search for the 6 points closest to the point in the feature sample set C in turn, and record the error values ​​of these 6 points in the error set F. Take the arithmetic mean of the 6 error values ​​of each point in the initial sample set A, add the arithmetic mean to the viscosity value of the corresponding point in the sample viscosity value set B, and obtain the corrected sample set H of the slag, as shown in Table 9: Table 9, modified sample set H;

[0086] S600: constructing a slag viscosity prediction model based on the corrected sample set, and using the slag viscosity prediction model to predict a slag viscosity value.

[0087] After obtaining the corrected sample set, a slag viscosity prediction model can be constructed based on the corrected sample set, that is, the corrected sample set H is modeled using a machine learning method.

[0088] In some embodiments, a plurality of machine learning methods may be applied to the modified sample set H for modeling, and the optimal model may be selected based on quantifiable evaluation indicators to obtain the optimal slag viscosity prediction model within the compound component mass fraction interval and the temperature interval under specified constraints. To this end, the modified sample set may be divided into a training set and a validation set, and then a plurality of machine learning algorithms may be applied to the training set in the modified sample set for modeling to obtain a plurality of second candidate models. The quantifiable indicators of the second candidate model are calculated using the validation set in the modified sample set to extract the slag viscosity prediction model based on the quantifiable indicators. The quantifiable indicators may include a root mean square error, and the corresponding slag viscosity prediction model is the model with the smallest root mean square error among the plurality of second candidate models.

[0089] For example, multiple machine learning methods are applied to the modified sample set H for modeling, and the optimal model is selected according to the RMSE result of the validation set. The optimal model is the optimal slag viscosity prediction model under the specified constraints. Among them, the various machine learning methods include: SVM model, Gaussian process regression model, efficient linear model, kernel model, integrated model, neural network model, tree model, linear regression model, stepwise linear regression model and sub-models with different hyperparameters for each model, totaling 28 models, and the 5-fold cross-validation method is used for machine learning. The RMSE results of the validation set of each second candidate model are shown in Table 10: Table 10, RMSE results of the validation set of the second alternative model;

[0090] It can be seen that based on the minimum RMSE principle, model No. 16 can be selected as the optimal model, that is, this model is the slag viscosity prediction model.

[0091] It should be noted that, since it is also necessary to evaluate the model based on a variety of machine learning methods and quantifiable evaluation indicators when determining the slag viscosity prediction model, the quantifiable evaluation indicators used in determining the slag viscosity prediction model can be the same as the multiple machine learning methods and quantifiable evaluation indicators used when determining the response relationship function of the predicted variables to the predicted values. However, other machine learning methods and quantifiable evaluation indicators may also be selected when necessary.

[0092] After the slag viscosity prediction model is constructed, the slag viscosity prediction model can be used to predict the slag viscosity value. In some embodiments, in order to predict the slag viscosity value, input data can be first obtained, wherein the input data includes mass fraction values ​​and temperature values ​​of multiple types of compounds within the compound type, mass fraction interval and temperature interval corresponding to the preset constraint conditions. The input data is then input into the slag viscosity prediction model to obtain the viscosity value output by the slag viscosity prediction model.

[0093] By applying the technical solution of this embodiment, the method can combine the experimentally corrected theoretical viscosity data with machine learning modeling to capture the nonlinear relationship between slag viscosity and compound composition and temperature, which can improve the prediction accuracy. In addition, by randomly sampling to generate a set of sample parameters and optimizing the modeling process, redundant calculations and experimental complexity are reduced, and combined with the optimal model screening, the prediction efficiency can be improved. The method can be compatible with a variety of metallurgical slag systems and multi-dimensional changes in their chemical composition and temperature. By introducing experimental corrections and multi-level optimization, it is suitable for viscosity prediction under complex production conditions, and provides reliable technical support for metallurgical process optimization and control. It can also flexibly select a variety of machine learning models and evaluation indicators so that the method can be adjusted according to the needs of different scenarios, and has good generalization ability and application potential.

[0094] Further, as a specific implementation of the metallurgical slag viscosity prediction method described in the above embodiment, the present application embodiment provides a metallurgical slag viscosity prediction system, such as Figure 8 As shown, the system includes: A sampling module, used for obtaining an initial sample set by random sampling within preset constraints, wherein the constraints include a mass fraction interval and a temperature interval of the slag-based compound of the metallurgical slag to be predicted; the initial sample set includes a plurality of parameter points; the parameter points are a combination of a mass fraction value and a temperature value obtained by random sampling; A theoretical viscosity calculation module, used to calculate the viscosity values ​​corresponding to the parameter points in the initial sample set using a preset viscosity calculation model to generate a sample viscosity value set; A feature extraction module, used for generating a partial dependence graph according to the sample viscosity value set, and constructing a feature sample set based on the endpoint values ​​of the partial dependence graph, wherein the partial dependence graph is used for characterizing the response of a prediction variable to a viscosity value, wherein the prediction variable is a slag-based compound of the metallurgical slag to be predicted; an error calculation module, for obtaining a theoretical value set, a real value set and an error set, wherein the theoretical value set includes viscosity values ​​obtained by calculating the characteristic sample set using the viscosity calculation model; the real value set includes viscosity values ​​obtained by testing the characteristic sample set through experiments; and the error set includes error values ​​obtained by performing difference calculation on the theoretical value set and the real value set; A correction module, configured to correct the sample viscosity value set using the initial sample set and the error set to generate a corrected sample set; The viscosity prediction module is used to construct a slag viscosity prediction model based on the corrected sample set, and use the slag viscosity prediction model to predict the slag viscosity value.

[0095] By applying the technical solution of this embodiment, a metallurgical slag viscosity prediction system provided in the above embodiment can obtain an initial sample set by random sampling after setting constraints. Then use the preset viscosity calculation model to calculate the viscosity value to generate a sample viscosity value set. Then construct a feature sample set containing sub-interval endpoint values ​​based on the partial dependence graph generated by the sample viscosity value set, and obtain a theoretical value set and a true value set based on the feature sample set, and calculate the error set to use the error set to correct the sample viscosity value set and generate a corrected sample set. Thus, a slag viscosity prediction model is constructed based on the corrected sample set, and the slag viscosity prediction model is used to predict the slag viscosity value. The system captures the nonlinear relationship between slag viscosity and compound composition and temperature by combining experimentally corrected theoretical viscosity data with machine learning modeling, which can improve prediction accuracy. In addition, by randomly sampling to generate a sample parameter set and optimizing the modeling process, redundant calculations and experimental complexity are reduced, and combined with optimal model screening, prediction efficiency can be improved.

[0096] It should be noted that, for other corresponding descriptions of the functional units involved in a metallurgical slag viscosity prediction system provided in an embodiment of the present application, reference can be made to the corresponding descriptions in the metallurgical slag viscosity prediction method provided in the above embodiment, and will not be repeated here.

[0097] The embodiment of the present application also provides a computer device, which can be a personal computer, a server, a network device, etc. The computer device includes a bus, a processor, a memory and a communication interface, and can also include an input and output interface and a display device. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store location information. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the steps in each method embodiment are implemented.

[0098] Those skilled in the art will appreciate that the structure of the above-mentioned computer device is only a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components, or combine certain components, or have a different arrangement of components.

[0099] In one embodiment, a computer-readable storage medium is also provided. The computer-readable storage medium may be non-volatile or volatile, and stores a computer program thereon. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0100] In one embodiment, a computer program product is also provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0101] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0102] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods.

[0103] Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc.

[0104] Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0105] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0106] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0107] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for predicting the viscosity of metallurgical slag, characterized in that: The method comprises: An initial sample set is obtained by random sampling within preset constraints, wherein the constraints include a mass fraction interval and a temperature interval of the slag-based compound of the metallurgical slag to be predicted; the initial sample set includes a plurality of parameter points; the parameter points are a combination of mass fraction values ​​and temperature values ​​obtained by random sampling; Calculating the viscosity values ​​corresponding to the parameter points in the initial sample set using a preset viscosity calculation model to generate a sample viscosity value set; Generating a partial dependence graph according to the sample viscosity value set, and constructing a characteristic sample set based on the endpoint values ​​of the partial dependence graph, wherein the partial dependence graph is used to characterize the response of the prediction variable to the viscosity value, wherein the prediction variable is the mass fraction of the slag-based compound and the slag-based temperature of the metallurgical slag to be predicted; Acquire a theoretical value set, a true value set, and an error set, wherein the theoretical value set includes viscosity values ​​obtained by calculating the characteristic sample set using the viscosity calculation model; the true value set includes viscosity values ​​obtained by testing the characteristic sample set through experiments; and the error set includes error values ​​obtained by performing difference calculation on the theoretical value set and the true value set; Correcting the set of sample viscosity values ​​using the initial sample set and the error set to generate a corrected sample set; A slag viscosity prediction model is constructed based on the corrected sample set, and the slag viscosity value is predicted using the slag viscosity prediction model.

2. The method according to claim 1, characterized in that Calculating the viscosity values ​​corresponding to the parameter points in the initial sample set using a preset viscosity calculation model to generate a sample viscosity value set, including: Calling a preset viscosity calculation model, wherein the viscosity calculation model is one of an empirical model, a structural model, a molecular dynamics model, and a thermodynamic application model; Inputting the parameter points in the initial sample set into the viscosity calculation model one by one to obtain the viscosity value output by the viscosity calculation model; The viscosity value is added to the initial sample set to generate the sample viscosity value set, in which the viscosity value has an associated relationship with the combination of the mass fraction value and the temperature value corresponding to the parameter point.

3. The method according to claim 1, characterized in that Generating a partial dependence graph according to the set of sample viscosity values ​​includes: Applying a plurality of machine learning algorithms to the sample viscosity value set to perform modeling to obtain a plurality of first candidate models; Obtaining quantifiable indicators of the first candidate model, wherein the quantifiable indicators include one or more combinations of root mean square error, mean square error, determination coefficient, mean absolute percentage error, and mean absolute error; Selecting the optimal model from the plurality of candidate models according to the quantifiable indicators; A partial dependence plot of all predictor variables on viscosity values ​​was generated using the optimal model.

4. The method according to claim 3, characterized in that: Constructing a feature sample set based on the endpoint values ​​of the partial dependence graph includes: Based on a preset interval partitioning algorithm, the partial dependence graph is divided into a plurality of sub-intervals, wherein the preset interval partitioning algorithm includes one or more combinations of an average distribution method, a response distribution method, and a random distribution method; Record the endpoint value of each sub-interval where the predictor variable is located; The endpoint values ​​satisfying the constraint conditions are combined to construct the feature sample set.

5. The method according to claim 4, characterized in that Based on a preset interval partitioning algorithm, the partial dependency graph is divided into a plurality of sub-intervals, including: Obtaining a curve function in the partial dependence graph and a definition domain of the curve function; Extracting the monotonic interval of the curve function in the definition domain; In the monotonic interval, the domain is evenly divided into a first number of subintervals, and the length of each subinterval is calculated; A segmentation point is calculated according to the subinterval length, and the monotonic interval is divided into a first number of subintervals using the segmentation point.

6. The method according to claim 4, characterized in that Based on a preset interval partitioning algorithm, the partial dependency graph is divided into a plurality of sub-intervals, including: Obtain the curve function in the partial dependence graph and the domain and range of the curve function; Extracting the monotonic interval of the curve function in the definition domain; In the monotonic interval, the value range is evenly divided into a second number of sub-intervals, and the length of each sub-interval is calculated; Calculate the split point according to the sub-interval length; The predicted variable value corresponding to the segmentation point is calculated, and the monotonic interval is divided into a second number of sub-intervals according to the predicted variable value.

7. The method according to claim 4, characterized in that Based on a preset interval partitioning algorithm, the partial dependency graph is divided into a plurality of sub-intervals, including: Obtaining a curve function in the partial dependence graph and a definition domain of the curve function; Extracting the monotonic interval of the curve function in the definition domain; In the monotonic interval, determining the number of subintervals, and randomly generating a third number of segmentation points, where the third number is equal to the number of subintervals minus 1; The monotonic interval is divided into a third number of sub-intervals according to the segmentation points.

8. The method according to claim 1, characterized in that: Correcting the sample viscosity value set using the initial sample set and the error set to generate a corrected sample set includes: Extracting parameter points from the initial sample set; Searching for a target point in the feature sample set based on the parameter point, the target point being an endpoint value closest to the parameter point; Based on the error set, recording the error value corresponding to the target point; Calculating the arithmetic mean of multiple error values ​​corresponding to the same target point, and establishing a slag sample viscosity value error set according to the arithmetic mean; The slag sample viscosity value error set is added to the viscosity values ​​in the sample viscosity value set to obtain a corrected sample set.

9. The method according to claim 1, characterized in that: A slag viscosity prediction model is constructed based on the corrected sample set, including: Dividing the modified sample set into a training set and a validation set; Applying a plurality of machine learning algorithms to the training sets in the revised sample set to perform modeling to obtain a plurality of second candidate models; Calculate a quantifiable index of the second candidate model using a validation set in the revised sample set, wherein the quantifiable index includes a root mean square error; The slag viscosity prediction model is extracted based on the quantifiable index, and the slag viscosity prediction model is a model with the smallest root mean square error among multiple second alternative models.

10. A metallurgical slag viscosity prediction system, characterized in that: The system comprises: A sampling module, used for obtaining an initial sample set by random sampling within preset constraints, wherein the constraints include a mass fraction interval and a temperature interval of the slag-based compound of the metallurgical slag to be predicted; the initial sample set includes a plurality of parameter points; the parameter points are a combination of a mass fraction value and a temperature value obtained by random sampling; A theoretical viscosity calculation module, used to calculate the viscosity values ​​corresponding to the parameter points in the initial sample set using a preset viscosity calculation model to generate a sample viscosity value set; A feature extraction module, used for generating a partial dependence graph according to the sample viscosity value set, and constructing a feature sample set based on the endpoint values ​​of the partial dependence graph, wherein the partial dependence graph is used for characterizing the response of a prediction variable to a viscosity value, wherein the prediction variable is a slag-based compound of the metallurgical slag to be predicted; an error calculation module, for obtaining a theoretical value set, a real value set and an error set, wherein the theoretical value set includes viscosity values ​​obtained by calculating the characteristic sample set using the viscosity calculation model; the real value set includes viscosity values ​​obtained by testing the characteristic sample set through experiments; and the error set includes error values ​​obtained by performing difference calculation on the theoretical value set and the real value set; A correction module, configured to correct the sample viscosity value set using the initial sample set and the error set to generate a corrected sample set; The viscosity prediction module is used to construct a slag viscosity prediction model based on the corrected sample set, and use the slag viscosity prediction model to predict the slag viscosity value.

Images