A method for establishing a molten iron silicon content model based on multiple linear regression

Abstract

This invention relates to a method for establishing a model to predict the silicon content in molten iron based on multiple linear regression, comprising the following steps: 1) determining the variables affecting the mass percentage of Si in molten iron within the blast furnace; 2) performing multiple correlation analysis on the variables, retaining the corresponding parameters in the blast furnace body's apparent parameters or the heat balance, material balance, and chemical balance formulas for the two variables with a correlation coefficient greater than 0.80; 3) linearizing the variables with nonlinear relationships; 4) introducing the variable ΔSi; 5) calculating the lag time, excluding variables with a correlation coefficient of 0.010–0.040 with the same lag time as the Si content in molten iron; 6) performing z-score standardization on the data; 7) establishing a prediction model after standardizing the variables; 8) establishing the final linear prediction model after inverse standardization of the variables. This invention accurately establishes a prediction model for the Si content in molten iron through linear transformation and the introduction of the ΔSi variable, achieving the goal of obtaining the blast furnace temperature in advance.

Description

Technical Field

[0001] This invention relates to the field of iron and steel metallurgy technology, and in particular to a method for establishing a model for predicting silicon content in molten iron based on multiple linear regression. Background Technology

[0002] Blast furnace ironmaking is an extremely complex metallurgical physicochemical process carried out under high temperature and pressure, involving constant transfer of heat, mass, and momentum between the gas, solid, and liquid phases. The process is characterized by nonlinearity, large time delays, and high noise, making its prediction and control a major challenge in metallurgical automation. A blast furnace temperature prediction model is a key and difficult issue in achieving automation of the blast furnace production process. Blast furnace production involves complex chemical, kinetic, and thermodynamic processes under closed conditions; maintaining a reasonable furnace temperature is one of the key factors for stable and smooth operation. Excessive furnace temperature can affect the fluidity of slag and iron. This can affect the stable operation of the blast furnace. Excessively high furnace temperature means that some of the fuel consumed is wasted, which is detrimental to reducing the cost per ton of iron. Conversely, consistently low furnace temperature leads to heat loss in the hearth, worsens the gas flow distribution, and hinders the smooth operation of the blast furnace. The Si content in molten iron, as a concentrated reflection of the hearth's heat level, is of great significance in the daily parameter control of the blast furnace. Blast furnace operators often adjust the heat based on the Si content in the molten iron, establishing predictive models for the Si content in molten iron to obtain information about the blast furnace temperature in advance. Therefore, accurately obtaining the Si content in molten iron is a crucial condition for understanding the internal changes within the blast furnace.

[0003] Chinese patent CN101211383A discloses a feature analysis and prediction method for silicon content in blast furnace hot metal. Using blast furnace process parameters as input variables in a blast furnace hot metal silicon content prediction model, after preprocessing the sample data of the input variables with exponential weighted moving average filtering and normalization, an improved dynamic independent component analysis method is used to extract features from the sample data of the input variables, eliminating the correlation between production process parameters. A dynamic recursive model for predicting blast furnace hot metal silicon content is established using the least squares support vector method, and a genetic algorithm is introduced to optimize the model parameters. This method has universal applicability for predicting silicon content in blast furnace hot metal during the smelting process, achieving good prediction accuracy and improving the prediction hit rate. However, this patent only selects a subset of feature variables through its feature selection method. This cannot meet the needs of dynamic feature selection under different furnace conditions, and discarding unselected feature variables wastes monitoring data, reducing the accuracy and robustness of blast furnace hot metal silicon content prediction. Summary of the Invention

[0004] This invention provides a method for establishing a model to predict the silicon content of molten iron based on multiple linear regression. By linearizing the variables of nonlinear relationships through thermal balance, material balance, and chemical balance in the blast furnace area, the complex nonlinear problem is transformed into a simple linear problem to be solved. The method retains the variables of nonlinear relationships and comprehensively analyzes the mathematical relationships of various influencing factors. It avoids the complex influence of the disordered polynomial formed by the interaction of factors from the perspective of data analysis, and avoids the limitations and hollowness of the model. The method introduces the process variable factor ΔSi, and achieves the purpose of dynamically updating the model without changing other variables. By using multiple linear regression and inverse standardization to accurately establish a prediction model for the silicon content of molten iron, the method achieves the goal of obtaining the blast furnace temperature in advance, which is consistent with the actual situation of blast furnace production.

[0005] To achieve the above objectives, the present invention employs the following technical solution:

[0006] A method for establishing a model for predicting silicon content in molten iron based on multiple linear regression includes the following steps:

[0007] 1) Determine the variables in the blast furnace that affect the mass percentage of Si in the molten iron;

[0008] 2) Perform multiple correlation analysis on the variables. Two variables with a correlation coefficient greater than 0.80 are considered to be of the same type. Retain the corresponding parameters in the blast furnace body's apparent parameters or the formulas for heat balance, material balance and chemical balance in the two variables.

[0009] 3) Use scatter plots to distinguish between linearized and nonlinear influencing factors, and use scatter plots to determine whether each variable satisfies the linear relationship with the Si content in molten iron;

[0010] 4) Perform multivariate nonlinear regression analysis on variables that have a nonlinear relationship with the silicon content of molten iron, and linearize the variables with nonlinear relationships by means of heat balance, material balance and chemical balance in the blast furnace area;

[0011] (1) The formula for linear thermal balance modification in the blast furnace area is:

[0012]

[0013] Among them, V 风 For air volume, m 3 / min;

[0014] Oxygen enrichment rate;

[0015] t represents temperature, in °C;

[0016] h represents absolute humidity, expressed in g / m³. 3 ;

[0017] High-temperature zone heat expenditure item = U铁 ×Q e +U 渣 ×Q u +V 煤气 ×c 煤气 ×t 顶 +2890×10×ω[Fe]×rd+22960×10×ω[Si](2)

[0018] Among them, U 铁 Iron content, unit: kg / t, Q e Enthalpy of molten iron, unit: GJ / t;

[0019] U 渣 Q represents slag quantity, in kg / t. u Enthalpy of slag, in GJ / t;

[0020] V 煤气 This refers to the gas volume at the furnace top, in cubic meters (m³). 3 / min, c 煤气 Specific heat capacity of coal gas, unit: GJ / ℃·m 3 , t 顶 This refers to the top temperature, measured in °C.

[0021] ω[Fe] represents the iron content in pig iron, and ω[Si] represents the silicon content in molten iron.

[0022] (2) The formula for linear material balance modification is as follows:

[0023]

[0024] Wherein, ω[Si] represents the silicon content in the molten iron;

[0025] ω(CO) is the CO content in the top gas of the furnace;

[0026] ω(CO2) is the CO2 content in the top gas of the furnace;

[0027] V 炉顶煤气 Gas volume at the furnace top, unit: m³ 3 / min;

[0028] w Fe This refers to the iron content per unit of pig iron.

[0029] r d For direct reduction degree;

[0030] ω(C) 焦挥 The amount of carbon contained in the volatile matter of fuel;

[0031] ω(C) 熔 The amount of carbon decomposed in the flux;

[0032] (3) The formula for linearizing chemical equilibrium is as follows:

[0033]

[0034] Wherein, ω[Si] represents the Si content in the molten iron;

[0035] T is the physical heat of molten iron in °C;

[0036] P co CO partial pressure, kPa;

[0037] The activity of SiO2 in the slag;

[0038] f si is the activity coefficient of Si in molten iron;

[0039] 5) Calculate the difference ΔSi between the average silicon content of molten iron in the two hours preceding the current value, and introduce the variable ΔSi.

[0040] 6) Calculate the hysteresis time of each variable, exclude variables with the same hysteresis time and a correlation coefficient of 0.010 to 0.040 with the Si content of the molten iron, and reorganize the data table.

[0041] 7) Perform z-score standardization on the data processed in step 6);

[0042] 8) Perform multiple linear regression on the standardized variables, retaining those variables whose regression prediction contribution rate R-sp reaches 75%, and establish a prediction model for Si content in molten iron, where a i The regression coefficients of the multiple linear regression of variables X are given. i _1 represents the standardized variable data, C_1 represents the standardization constant, and the standardized multiple linear regression model is as follows:

[0043]

[0044] 9) The obtained prediction model is de-standardized to establish a linear model of Si content in molten iron and various variables, where A i X represents the regression coefficients of the multiple linear regression after inverse standardization of the variables. i The variables are destandardized, C is the destandardization constant, and the destandardized multiple linear regression prediction linear model is as follows:

[0045]

[0046] Furthermore, lag analysis was performed on each variable using Pearson correlation coefficient analysis. The lag time corresponding to the largest absolute value of the correlation coefficient was selected as the lag time between each variable and the silicon content, and the data corresponding to the lag time of each variable were used as new data.

[0047] Furthermore, the z-score standardization formula is as follows:

[0048]

[0049] Where x′ is the standardized input variable value, x is the input variable, u is the mean of the input variables, and σ is the input variable.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] 1) By linearizing the variables of the nonlinear relationship through the thermal balance, material balance and chemical balance of the blast furnace area, the complex nonlinear problem is transformed into a simple linear problem to be solved. The variables of the nonlinear relationship are retained, and the mathematical relationship of various influencing factors is analyzed in all aspects. This avoids the complicated influence of the disordered polynomial formed by the interaction of factors from the perspective of data analysis, and avoids the limitations of the model.

[0052] 2) By introducing the process variable factor ΔSi, the model can be dynamically updated without changing other variables. This avoids the disadvantages of traditional neural networks, such as long training time and easy getting trapped in local optima. At the same time, it avoids the hollowing out problem caused by the lack of important feature variables in the model, and obtains a silicon content prediction model that is more consistent with the production site.

[0053] 3) By using multiple linear regression and inverse standardization, an accurate prediction model for Si content in molten iron was established, which enabled the early determination of blast furnace temperature and was consistent with the actual situation of blast furnace production. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0055] Figure 2 This is a comparison chart of the predicted values ​​and actual values ​​obtained after applying the model of this invention. Detailed Implementation

[0056] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0057] See Figure 1 This is a schematic diagram of the method flow of the present invention. The present invention provides a method for establishing a model for predicting silicon content in molten iron based on multiple linear regression, comprising the following steps:

[0058] 1) Determine the variables affecting the mass percentage of Si in the molten iron within the blast furnace as model input variables, including CO, CO2, gas utilization rate, hot blast pressure, blast volume, permeability, top pressure, oxygen enrichment, oxygen enrichment rate, hourly pulverized coal injection rate, blast temperature, cold blast temperature, actual blast velocity, standard blast velocity, furnace belly gas volume, furnace top gas volume, furnace body soft water system feedwater main temperature, tuyere sleeve average water temperature difference, furnace bottom soft water system feedwater main temperature, and furnace bottom soft water system feedwater main flow rate. The following 42 variables were measured: flow rate of the main water supply pipe of the furnace body soft water system, temperature of the furnace hearth soft water system, inlet water flow rate of the tuyere sleeve, water temperature difference of the furnace body, water temperature difference of the furnace hearth, average top temperature, total pressure difference, outlet temperature of soft water system 1, outlet temperature of soft water system 2, outlet temperature of soft water system in the furnace hearth, inlet temperature of the filter bag, moisture content, atmospheric temperature, absolute moisture content, batch of material, batch of iron, hourly iron content, hourly slag content, hourly coke content, total heat load of the first section of the furnace body, total heat load of the second section of the furnace body, and total heat load of the third section of the furnace body.

[0059] 2) Multiple correlation analysis was performed on each variable. For variables with a correlation coefficient greater than 0.80, the displayed parameters of the blast furnace body or the corresponding parameters in the heat balance, material balance, and chemical balance formulas were retained. Specifically, the correlation coefficients between air volume and actual wind speed, standard wind speed, permeability, cold blast temperature, and furnace belly gas volume were 0.830, 1.000, 0.810, 1.000, and 0.992, respectively. Meanwhile, the correlation coefficient between oxygen enrichment and oxygen enrichment rate was 0.825, and the correlation coefficient between humidity and absolute humidity was 0.897. (The remaining text appears to be incomplete and requires further context.) The correlation coefficients between batch iron and hourly iron quantity were 0.918 and 0.804, respectively; the correlation coefficient between furnace top average temperature and bag filter inlet temperature was 0.832; the correlation coefficient between furnace body water temperature difference and soft water outlet temperature was 0.890; and the correlation coefficient between cold air temperature and atmospheric temperature was 0.828. Variables with correlation coefficients greater than 0.8 were excluded. Therefore, 12 variables were excluded: actual wind speed, standard wind speed, air permeability, cold air temperature, furnace belly gas quantity, oxygen enrichment, humidity, bag filter inlet temperature, furnace body water temperature difference, atmospheric temperature, material batch, and batch iron.

[0060] 3) Use scatter plots to distinguish between linear and nonlinear influencing factors. With the Si content in blast furnace hot metal as the output variable, draw a scatter plot and use the scatter plot to determine whether each variable satisfies the linear relationship with the Si content in hot metal.

[0061] 4) Using the Si content in blast furnace hot metal as the output variable and the variables retained in the previous step as input variables, the scatter plot shows that eight variables—air volume, hourly coal volume, hourly iron volume, hourly slag volume, hourly coke volume, total heat load of the first section of the furnace, total heat load of the second section of the furnace, and total heat load of the third section of the furnace—have a linear relationship with the Si content in the hot metal. Therefore, these eight linear variables are retained. A multivariate nonlinear regression analysis is performed on the remaining 22 variables. The 22 nonlinear variables are then modified using the linear transformation formulas for blast furnace heat balance, material balance, and chemical balance. Fourteen linear variables were generated, including air volume * oxygen enrichment rate, air volume * absolute humidity, air volume * air temperature, air volume * air temperature * absolute humidity, top gas * average top temperature, CO * top gas, CO2 * top gas, hourly iron production * gas utilization rate, 1 / top pressure^2 * CO^2, 1 / hot blast pressure^2 * CO^2, 1 / total pressure difference^2 * CO^2, average water temperature difference of tuyeres * tuyeres flow rate, flow rate of the main feedwater pipe of the furnace body soft water system * furnace hearth water temperature difference, and flow rate of the main feedwater pipe of the furnace bottom soft water system * furnace bottom soft water system water temperature difference.

[0062] (1) The formula for linear thermal balance modification in the blast furnace area is:

[0063]

[0064] Among them, V 风 For air volume, m 3 / min;

[0065] Oxygen enrichment rate;

[0066] t represents temperature, in °C;

[0067] h represents absolute moisture content, in g / m³ 3 ;

[0068] The above formula constructs linearized variables. V 风 ×h、V 风 ×t、V 风 ×t×h;

[0069] High-temperature zone heat expenditure item = U 铁 ×Q e +U 渣 ×Q u +V 煤气 ×c 煤气 ×t 顶 +2890×10×ω[Fe]×rd+22960×10×ω[Si]%(9)

[0070] Among them, U 铁 Iron content, unit: kg / t, Q e Enthalpy of molten iron, unit: GJ / t;

[0071] U 渣 Q represents slag quantity, in kg / t. u Enthalpy of slag, in GJ / t;

[0072] V 煤气 This refers to the gas volume at the furnace top, in cubic meters (m³). 3 / min, c 煤气 Specific heat capacity of coal gas, unit: GJ / ℃·m 3 , t 顶 This refers to the top temperature, measured in °C.

[0073] ω[Fe] represents the iron content in pig iron, and ω[Si] represents the silicon content in molten iron.

[0074] The above formula constructs the linear variable V 煤气 ×c 煤气 ×t 顶 ,ω[Fe]×r d U 铁 ×Q e U 渣 ×Q u ;

[0075] (2) The formula for linear material balance modification is as follows:

[0076]

[0077] Wherein, ω[Si] represents the silicon content in the molten iron;

[0078] ω(CO) is the CO content in the top gas of the furnace;

[0079] ω(CO2) is the CO2 content in the top gas of the furnace;

[0080] V 炉顶煤气 Gas volume at the furnace top, unit: m³ 3 / min;

[0081] w Fe This refers to the iron content per unit of pig iron.

[0082] r d For direct reduction degree;

[0083] ω(C) 焦挥 The amount of carbon contained in the volatile matter of fuel;

[0084] ω(C) 熔 The amount of carbon decomposed in the flux;

[0085] The above formula constructs the linear variable ω(CO)×V 炉顶煤气 ω(CO2)×V 炉顶煤气 , W Fe *r d ;

[0086] (3) The formula for linearizing chemical equilibrium is as follows:

[0087]

[0088] Wherein, ω[Si] represents the Si content in the molten iron;

[0089] T is the physical heat of molten iron in °C;

[0090] P co CO partial pressure, kPa;

[0091] The activity of SiO2 in the slag;

[0092] f si is the activity coefficient of Si in molten iron;

[0093] The above formula constructs a linear variable.

[0094] 5) Calculate the difference ΔSi between the silicon content of molten iron in the two hours preceding the current value, and introduce the variable factor ΔSi.

[0095] 6) Given the significant lag characteristic of blast furnace operation, data from a period of stable production during blast furnace operation were selected as the research object. Lag times of 1–8 hours were set according to the blast furnace smelting cycle. The Pearson correlation coefficients between input variables and silicon content were calculated for different lag times, and stepwise regression analysis was performed to determine the lag time between each variable and the Si content in molten iron. Variables with similar correlation coefficients and the same lag time were excluded. The data tables were then reorganized. The Pearson correlation coefficient formula is as follows:

[0096]

[0097] Where, r is the Pearson correlation coefficient, and x i For the input variable value, y i Here, N represents the silicon content value, and N is the number of data sets involved in the correlation analysis.

[0098] (1 / Top pressure) 2 *(CO) 2 (1 / Hot air pressure) 2 *(CO) 2 (1 / total differential pressure) 2 *(CO) 2The correlation coefficients of these three factors with the Si content in molten iron are close in magnitude, and all of them reach their maximum during the 3-hour hysteresis analysis. Therefore, these three factors belong to the same category. Only one result of the hysteresis analysis is retained, and the one with the highest absolute value of the correlation coefficient (1 / total pressure difference) is selected. 2 *(CO) 2 As input to the model.

[0099] Select the following parameters: ΔSi, total heat load of the first section of the furnace body, total heat load of the second section of the furnace body, total heat load of the third section of the furnace body, hourly pulverized coal injection rate, hourly iron production, hourly coke production, hourly slag production, air volume, air volume * oxygen enrichment rate, air volume * air temperature, air volume * absolute humidity, air volume * air temperature * absolute humidity, furnace top gas * CO, furnace top gas * CO2, flow rate of the main feedwater pipe of the furnace body soft water system * hearth water temperature difference, average water temperature difference of the tuyere sleeve * tuyere sleeve inlet flow rate, flow rate of the main feedwater pipe of the furnace bottom soft water system * furnace bottom soft water system water temperature difference, hourly iron production * utilization rate, (1 / total pressure difference). 2 *(CO) 2 Twenty-one variables, including the average top gas temperature and the top temperature of the furnace, are used as model inputs. Their corresponding lag times are 1 hour, 3 hours, 1 hour, 1 hour, 1 hour, 3 hours, 5 hours, 3 hours, 1 hour, 8 hours, 6 hours, 8 hours, 7 hours, 2 hours, 2 hours, 7 hours, 7 hours, 8 hours, 4 hours, 3 hours, and 7 hours, respectively.

[0100] 7) Based on the hysteresis analysis results, the data is reorganized to form a new training set data table, and the data is z-score standardized to eliminate the adverse effects of different units.

[0101]

[0102] Where x′ is the standardized input variable value, x is the input variable, u is the mean of the input variables, and σ is the input variable.

[0103] 8) Perform multiple linear regression on the standardized variables, retaining those variables whose regression prediction contribution rate R-sp reaches 75%, and establish a prediction model for Si content in molten iron, where a i The regression coefficients of the multiple linear regression of variables X are given. i _1 represents the standardized variable data, C_1 represents the standardization constant, and the standardized multiple linear regression model is as follows:

[0104]

[0105] Import the updated data into MINITAB, perform multivariate stepwise regression fitting, calculate the regression coefficients, and obtain the standardized prediction model for Si content in molten iron as follows:

[0106] Molten iron containing Si = 0.27726 + 0.01113ΔSi_1 + 0.02219 * 7-segment total heat load_1 + 0.01254 * 9-segment total heat load_1 - 0.01227 Hourly pulverized coal injection_1 - 0.00674 Hourly coke_1 + 0.01487 Air volume * air temperature_1 + 0.02315 Air volume * absolute humidity_1 + 0.02472 Air volume * air temperature * absolute humidity_1 + 0.01142 Furnace top gas * CO_1 - 0.02885 Furnace body soft water system feedwater main flow rate * hearth water temperature difference_1 - 0.02621 * (1 / total pressure difference) 2 *(CO) 2 _1-0.01158 Furnace top gas volume * average top temperature_1;

[0107] 9) The obtained prediction model is de-standardized to establish a linear model of Si content in molten iron and various variables, where A i X represents the regression coefficients of the multiple linear regression after inverse standardization of the variables. i The variables are destandardized, C is the destandardization constant, and the destandardized multiple linear regression prediction linear model is as follows:

[0108]

[0109] The linear prediction model for Si content in molten iron obtained by inverse standardization of the prediction model after regression is as follows:

[0110] Iron silicon content = -0.603 + 0.1737ΔSi + 0.0000427 Total heat load of section + 0.0000069 Total heat load of section - 0.00380 hours pulverized coal injection - 0.000537 hours coke production + 0.00000000113 Air volume * air temperature + 0.00000003199 Air volume * absolute humidity + 0.00000004849 Air volume * air temperature * absolute humidity + 0.000000258 Furnace top gas * CO - 0.000134 Furnace body soft water system feedwater main flow rate * hearth water temperature difference - 0.1722(1 / total pressure difference) 2 *(CO) 2 -0.00000000058 Furnace top gas volume * average top temperature.

[0111] See Figure 2 The chart shows a comparison between the actual Si content in molten iron and the model's predicted value. By comparing the model's predicted data with the actual production data, the accuracy rate for a Si content deviation of ±0.1% in molten iron is 92%, and the accuracy rate for a Si content deviation of ±0.05% in molten iron is 81%, achieving the goal of accurate prediction.

[0112] The above embodiments are implemented based on the technical solution of the present invention, providing detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the above embodiments. Unless otherwise specified, the methods used in the above embodiments are conventional methods.

Claims

1. A method for establishing a model for predicting silicon content in molten iron based on multiple linear regression, characterized in that, Includes the following steps: 1) Determine the variables in the blast furnace that affect the mass percentage of Si in the molten iron; 2) Perform multiple correlation analysis on the variables. Two variables with a correlation coefficient greater than 0.80 are considered to be of the same type. The corresponding parameters in the blast furnace body's apparent parameters or the formulas for heat balance, material balance and chemical balance are retained in the two variables. 3) Use scatter plots to distinguish between linearized and nonlinear influencing factors, and use scatter plots to determine whether each variable satisfies a linear relationship with the Si content in molten iron; 4) Perform multiple nonlinear regression analysis on variables with a nonlinear relationship to the silicon content of molten iron, and linearize the variables with nonlinear relationships by considering the heat balance, material balance, and chemical balance in the blast furnace area. The formula for linear heat balance modification in the blast furnace area is: Heat input in the high-temperature zone = 0.0468V 风 +0.630V 风 × -544.887×10 -6 V 风 ×h+85.374×10 -6 V 风 ×t+0.128×10 -6 V 风 ×t×h(1) Among them, V 风 For air volume, m 3 / min; Oxygen enrichment rate; t is the temperature, in °C; h represents absolute humidity, expressed in g / m³. 3 ; High-temperature zone heat expenditure item = U 铁 ×Q e + U 渣 ×Q u +V 煤气 ×c 煤气 ×t 顶 +2890×10×ω[Fe]×r d +22960×10×ω[Si](2) Among them, U 铁 Iron content, unit: kg / t, Q e Enthalpy of molten iron, unit: GJ / t; U 渣 Q represents slag quantity, in kg / t. u Enthalpy of slag, in GJ / t; V 煤气 This refers to the gas volume at the furnace top, in cubic meters (m³). 3 / min,c 煤气 Specific heat capacity of coal gas, unit: GJ / ℃▪m 3 , t 顶 This refers to the top temperature, measured in °C. ω[Fe] represents the iron content in pig iron, and ω[Si] represents the silicon content in molten iron. The formula for linearizing the material balance is as follows: 10× ω[Si]= (ω(CO)+ω(CO2))×V 炉顶煤气 - - w Fe *r d -ω (C) 焦挥 -ω (C) 熔 (3) Wherein, ω[Si] represents the silicon content in the molten iron; ω(CO) is the CO content in the top gas of the furnace; ω(CO2) is the CO2 content in the top gas of the furnace; V 炉顶煤气 Gas volume at the furnace top, unit: m³ 3 / min; w Fe This refers to the iron content per unit of pig iron. r d For direct reduction degree; ω(C) 焦挥 The amount of carbon contained in the volatile matter of fuel; ω(C) 熔 The amount of carbon decomposed in the flux; The formula for linearizing chemical equilibrium is as follows: （4） Wherein, ω[Si] represents the Si content in the molten iron; T is the physical heat of molten iron in °C; The partial pressure of CO is kPa. The activity of SiO2 in the slag; is the activity coefficient of Si in molten iron; 5) Calculate the difference ΔSi between the average silicon content of molten iron in the two hours preceding the current value, and introduce the variable ΔSi. 6) Calculate the lag time of each variable, exclude variables with the same lag time and correlation coefficient with the Si content of molten iron (0.010-0.040), and reorganize the data table; 7) Perform z-score standardization on the data processed in step 6); 8) Perform multiple linear regression on the standardized variables, retaining those variables whose regression prediction contribution rate R-sp reaches 75%, and establish a prediction model for Si content in molten iron. The regression coefficients for multiple linear regression of variables are: For standardized variable data, As a standardized constant, the standardized multiple linear regression model is as follows: （5） 9) The obtained prediction model is de-standardized to establish a linear model of Si content in molten iron and various variables, where... The regression coefficients of the multiple linear regression after inverse standardization of the variables are . The variables are the destandardized data, C is the destandardization constant, and the destandardized multiple linear regression prediction linear model is as follows: （6）。 2. The method for establishing a model for predicting silicon content in molten iron based on multiple linear regression according to claim 1, characterized in that, The Pearson correlation coefficient analysis method was used to perform lag analysis on each variable. The lag time corresponding to the largest absolute value of the correlation coefficient was selected as the lag time between each variable and the silicon content. The data corresponding to the lag time of each variable were used as the new data.

3. The method for establishing a model for predicting silicon content in molten iron based on multiple linear regression according to claim 1, characterized in that, The z-score standardization formula is as follows: （7） in, To standardize the input variable values, For input variables, The mean of the input variable. The standard deviation of the input variable.

Images