A method for predicting slag viscosity based on hybrid linear programming algorithm

The slag viscosity sample data is processed through a hybrid linear programming algorithm, which solves the slag viscosity measurement hysteresis problem, real-time prediction and adjustment are achieved, and blast furnace production efficiency and equipment life are improved.

CN113780627BActive Publication Date: 2024-11-29WISDRI ENG & RES INC LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202110934758.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-08-16
Publication Date
2024-11-29
Estimated Expiration
2041-08-16

AI Technical Summary

Technical Problem

In the prior art, the slag viscosity measurement method has a large hysteresis, and it is impossible to predict viscosity based on the slag composition in time, affecting the blast furnace anterotravel and pig iron quality, and it is difficult to adjust the viscosity in time to extend the blast furnace life.

Method used

Using a method based on a hybrid linear programming algorithm, the slag viscosity sample data is collected and processed, the fitting formula is defined, the distance coefficient is optimized using the least squares method and the gradient descent method, the weight factor is calculated, and the slag viscosity is calculated by selecting the optimal combination through the cost function. Predicted value.

Benefits of technology

It realizes the real prediction of the slag viscosity after obtaining the slag component data, timely adjusting the viscosity, improving the blast furnace inverted and pig iron quality, and extending the blast furnace life.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113780627B_ABST
    Figure CN113780627B_ABST
Patent Text Reader

Abstract

The present invention discloses a slag viscosity prediction method based on a hybrid linear programming algorithm, comprising the following steps: defining a fitting formula for data in a viscosity sample library, and using the least square method to obtain an initial value and an intercept value of a distance coefficient; using the data of each sample point in the viscosity sample library to iteratively optimize the distance coefficient in the direction of gradient descent, and selecting the optimal distance coefficient according to the root mean square error value; selecting a number of sample points closest to the current input point from the viscosity sample library, selecting 4 sample points from them, and using the determinant to calculate the weight factor of each combination; defining a cost function of the weight factor, selecting a combination with the smallest cost function and using it to calculate the predicted value of the slag viscosity. The present invention solves the problem of large hysteresis in measuring the slag viscosity, and can immediately predict the current slag viscosity after obtaining the composition data of the slag, so that timely adjustment measures can be taken to always control the viscosity within a suitable range.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of blast furnace automatic control, and in particular to a slag viscosity prediction method based on a hybrid linear programming algorithm. Background Art

[0002] The viscosity of blast furnace slag refers to the internal friction coefficient between two layers of liquid with different speeds. It is the inverse of fluidity and is directly related to the fluidity of slag. Slag fluidity directly affects the smooth operation of blast furnace and the quality of pig iron. Therefore, it is a slag property indicator that blast furnace workers are most concerned about.

[0003] The viscosity of slag has a great influence on blast furnace smelting. First, the viscosity affects the permeability of the material column below the slag zone. If the viscosity of the slag is too high, it cannot flow smoothly in the dripping zone, reducing the porosity of the coke skeleton, increasing the resistance of the coal gas, and affecting the smooth operation of the blast furnace. Secondly, the viscosity affects the desulfurization ability of the slag. Slag with low viscosity and good fluidity is conducive to desulfurization, while slag with high viscosity and poor fluidity is not conducive to desulfurization. Thirdly, the viscosity of the slag affects the slag discharge operation. Slag with too high viscosity is prone to sticking to the groove and slag condensation at the slag mouth, which will cause slag discharge difficulties. Finally, the viscosity of the slag also affects the life of the blast furnace. Slag with high viscosity is easy to form slag skin in the furnace, which plays a role in protecting the lining, while slag with too low viscosity and good fluidity will wash the lining and shorten the life of the blast furnace.

[0004] The viscosity of slag is generally measured on site using an internal cylinder rotation viscometer. First, the slag sample needs to be sampled and made, and then the measurement is performed. During the measurement, the graphite probe is immersed in the slag melt. When the motor drives the graphite probe to rotate through the hanging wire and the graphite rod, the hanging wire generates a torsion angle due to the viscosity of the slag. The viscosity measurement value can be obtained by processing the torsion angle. Measuring the viscosity of slag is time-consuming, and the result is generally obtained several hours after sampling, which has a large lag.

[0005] There are many factors that affect slag viscosity, two of which are slag temperature and composition. In the blast furnace production process, when the furnace temperature is relatively stable, the slag viscosity is mainly affected by its composition. If the slag viscosity can be predicted in time according to the slag composition, adjustment measures can be taken in advance to control the viscosity to a suitable range, improve the quality of blast furnace operation and pig iron, promote desulfurization reaction, and extend the life of the blast furnace. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a slag viscosity prediction method based on a hybrid linear programming algorithm in view of the defects in the prior art, so as to predict the slag viscosity value immediately after the slag composition is measured.

[0007] The technical solution adopted by the present invention to solve its technical problem is:

[0008] The present invention provides a slag viscosity prediction method based on a hybrid linear programming algorithm, the method comprising the following steps:

[0009] Step S10, collecting viscosity sample data of slag, forming a viscosity sample library, defining a fitting formula for the data in the viscosity sample library, and using the least square method to obtain an initial value and an intercept value of the distance coefficient;

[0010] Step S20, using the data of each sample point in the viscosity sample library to iteratively optimize the distance coefficient in the direction of gradient descent, and selecting the optimal distance coefficient according to the root mean square error value;

[0011] Step S30, selecting a number of sample points closest to the current input point from the viscosity sample library, selecting 4 sample points from the sample points, and using the determinant to calculate the weight factor of each combination;

[0012] Step S40, defining a cost function of weight factors, selecting a combination with the minimum cost function and using it to calculate a predicted value of slag viscosity.

[0013] Furthermore, the method for defining the fitting formula for the data in the viscosity sample library in step S10 is:

[0014] The viscosity sample data measured by the viscometer are collected to form a sample library. Each sample point contains the molar fraction values ​​of the four components of the slag and the viscosity value obtained by the test. After taking the natural logarithm of the viscosity value, there is a better linear relationship between the two. The fitting formula for defining the data in the viscosity sample library is:

[0015] lnN=σ Ca ×C Ca +σ Mg ×C Mg +σ Si ×C Si +σ Al ×C Al +β0

[0016] Where N is the viscosity value tested, σ Ca and C Ca are the distance coefficient and mole fraction of CaO component, σ Mg and C Mg are the distance coefficient and mole fraction of MgO component, σ Si and C Si are the distance coefficient and mole fraction of SiO2 component, σ Al and C Al are the distance coefficient and mole fraction of Al2O3 components, and β0 is the intercept value in the linear fitting formula. Ca , σ Mg , σ Si , σAl , β0 is the parameter to be identified.

[0017] Furthermore, the method of using the least square method to obtain the initial value of the distance coefficient in step S10 is:

[0018] The least squares parameter estimation method of the linear regression model is used to calculate the initial value of the distance coefficient. The calculation formula is as follows:

[0019]

[0020] In the formula, represents the initial value of the kth distance coefficient, 1≤k≤4, N i represents the viscosity value of the i-th sample in the sample library, S N is the number of samples in the sample library, C k,i represents the kth mole fraction of the ith sample, For S N The average value of the kth mole fractions.

[0021] Furthermore, the method for obtaining the intercept value according to the initial value of the distance coefficient in step S10 is:

[0022] According to the initial value of the distance coefficient, the intercept value is calculated as follows:

[0023]

[0024] Furthermore, the method of sequentially using the data of each sample point in the step S20 to iteratively optimize the distance coefficient in the direction of gradient descent is:

[0025] Starting from the first sample in the sample library, the estimated error of each sample is calculated sequentially. Assuming that the current sample is the mth sample, the estimated error of the sample is:

[0026]

[0027] In the formula, is the kth distance coefficient after updating with the m-1th sample in the sample library, 1≤k≤4, C k,m is the kth mole fraction of the mth sample, N m is the viscosity value of the mth sample.

[0028] Update the estimated distance coefficient according to the error in the direction of gradient descent:

[0029]

[0030] In the formula, is the kth coefficient after m updates, R Sis the learning rate parameter of the gradient descent method.

[0031] Furthermore, the method of calculating the root mean square error according to the updated distance coefficient in step S20 and judging whether it is less than the error threshold is:

[0032] MSE (root mean square error) is used as the optimization objective function, so the objective function after the mth update is:

[0033]

[0034] Set a smaller value as the error threshold. If the MSE index after the coefficient update is lower than the error threshold, the optimization algorithm ends and the current estimated coefficient is taken as the final value of the distance coefficient. Right now:

[0035] If the MSE index after the coefficient update is higher than the error threshold, the next sample data is used to execute the previous step and this step in a loop.

[0036] Furthermore, the method of sorting the MSE index of all sample points in step S20 and selecting the optimal distance coefficient value is:

[0037] If all sample point data are used and all calculated MSE indicators are higher than the error threshold, the MSE indicators of all sample points are sorted, and the distance coefficient with the smallest MSE is taken as the final value of the distance coefficient. Assume that the minimum MSE corresponds to the jth sample point, then

[0038] Furthermore, the method of selecting a number of sample points closest to the current input point from the viscosity sample library in step S30 is:

[0039] Calculate the distance between all sample points in the viscosity sample library and the current input point in a loop as follows:

[0040]

[0041] In the formula, D(x,i) is the distance between the current input point x and the point i in the sample library, C x,m is the molar fraction of the mth component at point x, C i,m is the molar fraction of the mth component at point i, σ m,final is the distance coefficient of the mth component, calculated in step S20.

[0042] After finding the distances between all sample points and the current input point, select the smallest N points (N>4) and store them in a two-dimensional array for later use.

[0043] Furthermore, the method of selecting 4 points from N sample points in step S30 is:

[0044] Calculate all combinations of 4 points selected from N sample points, and store the data of the 4 points contained in each combination into a two-dimensional array for later use.

[0045] Furthermore, the method of calculating the quaternary system weight factor of each combination using determinant loop in step S30 is:

[0046] The weight factor is calculated based on the mole fractions of the four components at each point. Suppose the mole fractions of the four components at the current input point are [C N,Ca ,C N,Mg ,C N,Si ,C N,Al ]; and the molar fractions of the four components of the four sample points in a certain combination are [C 1,Ca ,C 1,Mg ,C 1,Si ,C 1,Al ],[C 2,Ca ,C 2,Mg ,C 2,Si ,C 2,Al ],[C 3,Ca , C 3,Mg , C 3,Si , C 3,Al ],[C 4,Ca ,C 4,Mg ,C 4,Si ,C 4,Al ].

[0047] Calculate the determinant values ​​of the four components of the four sample points of the combination, set it as D4, that is:

[0048]

[0049] The weight factor of each sample point is calculated as follows:

[0050]

[0051]

[0052]

[0053]

[0054] Loop through the calculations to calculate the weight factor for each combination and store it in a two-dimensional array for later use.

[0055] Furthermore, the method of defining the cost function in step S40 and calculating the cost function value of each combination is:

[0056] Define the cost function as follows:

[0057] Cost(k) = a × abs(W sum - 1.0) + (1.0 - a) × ∑abs(W neg ) for 1 ≤ k ≤ 15

[0058] In the formula, Cost(k) is the cost function value of the k-th combination, where 1 ≤ k ≤ 15; a is the penalty coefficient, 0.0 < a < 1.0; abs is the absolute value operator, W sum is the weight sum of 4 components, W neg is the negative weight, and ∑abs(W neg ) means adding up the absolute values of all negative weights. The first term on the right side is used to penalize the difference between the weight sum and 1.0, and the second term is used to penalize the negative weights. The penalty coefficient a is used to adjust the proportion of these two penalty methods.

[0059] Furthermore, the method for sorting the cost functions of each combination and selecting the combination with the minimum cost function in step S40 is as follows:

[0060] Sort the cost functions of each combination, and select the combination with the minimum cost function as the optimal solution. Let the weight factors of the 4 sample points included in the optimal solution be b 1,opt , b 2,opt , b 3,opt , b 4,opt , and the viscosity values of these 4 points are S1, S2, S3, and S4.

[0061] Furthermore, the method for obtaining the viscosity prediction value of the input point in step S40 is as follows:

[0062] Calculate the predicted viscosity value of the input point according to the following formula:

[0063] S pre = b 1,opt × S1 + b 2,opt × S2 + b 3,opt × S3 + b 4,opt × S4

[0064] In this way, according to the above steps, the slag viscosity value can be immediately calculated based on the analyzed slag composition. If the slag viscosity is not within the suitable range, it can be adjusted in time to optimize the process performance of blast furnace production.

[0065] The beneficial effects of the present invention are as follows: the slag viscosity prediction method based on the hybrid linear programming algorithm of the present invention solves the problem of large hysteresis in the existing slag viscosity measurement method. After obtaining the slag composition data, the current slag viscosity can be immediately predicted, so that adjustment measures can be taken in time to always control the viscosity within a suitable range, thereby improving the quality of blast furnace operation and pig iron, promoting desulfurization reaction, and extending the life of the blast furnace. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0067] Figure 1 It is a flow chart of a slag viscosity prediction method based on a hybrid linear programming algorithm provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0069] like Figure 1 As shown, an embodiment of the present invention provides a slag viscosity prediction method based on a hybrid linear programming algorithm, comprising the following steps:

[0070] Step S10, defining a fitting formula for the data in the viscosity sample library, and using the least squares method to obtain an initial value and an intercept value of the distance coefficient.

[0071] Step S101, defining a fitting formula for data in a viscosity sample library.

[0072] The viscosity sample data measured by the viscometer are collected to form a sample library. Each sample point contains the molar fraction values ​​of the four components of the slag and the viscosity value obtained by the test. After taking the natural logarithm of the viscosity value, there is a better linear relationship between the two. In this embodiment, the fitting formula for defining the data in the viscosity sample library is:

[0073] lnN=σ Ca ×C Ca +σ Mg ×C Mg +σ Si ×C Si +σ Al ×C Al +β0

[0074] Where N is the viscosity value tested, σ Ca and CCa are the distance coefficient and mole fraction of CaO component, σ Mg and C Mg are the distance coefficient and mole fraction of MgO component, σ Si and C Si are the distance coefficient and mole fraction of SiO2 component, σ Al and C Al are the distance coefficient and mole fraction of Al2O3 components, and β0 is the intercept value in the linear fitting formula. Ca , σ Mg , σ Si , σ Al , β0 is the parameter to be identified.

[0075] Step S102: using the least square method to obtain the initial value and intercept value of the distance coefficient.

[0076] The least squares parameter estimation method of the linear regression model is used to calculate the initial value of the distance coefficient. The calculation formula is as follows:

[0077]

[0078] In the formula, represents the initial value of the kth distance coefficient, 1≤k≤4, N i represents the viscosity value of the i-th sample in the sample library, S N is the number of samples in the sample library, C k,i represents the kth mole fraction of the ith sample, For S N The average value of the kth mole fractions.

[0079] Step S103, calculating the intercept value according to the initial value of the distance coefficient.

[0080] According to the initial value of the distance coefficient, the intercept value is calculated as follows:

[0081]

[0082] Step S20, using the data of each sample point to iteratively optimize the distance coefficient in the direction of gradient descent, and selecting the optimal distance coefficient according to the root mean square error value.

[0083] After using the least squares method to obtain the initial value and intercept value of the distance coefficient, the gradient descent method can be used to further optimize the distance coefficient. Since the gradient descent method continuously searches for a better value in the direction of the gradient descent of the identification coefficient, and the intercept β0 is a constant and has no gradient, the gradient descent method can only optimize the distance coefficient but not the intercept coefficient.

[0084] Step S201, sequentially use the data of each sample point to iteratively optimize the distance coefficient in the direction of gradient descent.

[0085] Starting from the first sample in the sample library, the estimated error of each sample is calculated sequentially. Assuming that the current sample is the mth sample, the estimated error of the sample is:

[0086]

[0087] In the formula, is the kth distance coefficient after updating with the m-1th sample in the sample library, 1≤k≤4, C k,m is the kth mole fraction of the mth sample, N m is the viscosity value of the mth sample.

[0088] Update the estimated distance coefficient according to the error in the direction of gradient descent:

[0089]

[0090] In the formula, is the kth coefficient after m updates, R S is the learning rate parameter of the gradient descent method, which is set to 0.001 in this example.

[0091] Step S202, calculating the root mean square error according to the updated distance coefficient, and determining whether it is less than the error threshold.

[0092] In this embodiment, MSE (root mean square error) is used as the optimization objective function, so the objective function after the mth update is:

[0093]

[0094] Set a smaller value as the error threshold. In this example, it is set to 0.1. If the MSE index after the coefficient update is lower than the error threshold, the optimization algorithm ends and the current estimated coefficient is taken as the final value of the distance coefficient. Right now:

[0095] If the MSE index after the coefficient update is higher than the error threshold, the next sample data is used to execute steps S201 to S202 in a loop.

[0096] Step S203, sorting the MSE indicators of all sample points and selecting the optimal distance coefficient value.

[0097] If all sample point data are used and all calculated MSE indicators are higher than the error threshold, the MSE indicators of all sample points are sorted, and the distance coefficient with the smallest MSE is taken as the final value of the distance coefficient. Assume that the minimum MSE corresponds to the jth sample point, then

[0098] Step S30, selecting a number of sample points closest to the current input point from the viscosity sample library, selecting 4 sample points from the sample points, and using the determinant to calculate the weight factor of each combination.

[0099] Step S301, select several sample points closest to the current input point from the viscosity sample library, and store the mole fraction of each component at each point into the two-dimensional array D6 min .

[0100] Calculate the distance between all sample points in the viscosity sample library and the current input point in a loop as follows:

[0101]

[0102] In the formula, D(x,i) is the distance between the current input point x and the point i in the sample library, C x,m is the molar fraction of the mth component at point x, C i,m is the molar fraction of the mth component at point i, σ m,final is the distance coefficient of the mth component, calculated in step S20.

[0103] After finding the distance between all sample points and the current input point, select N points with the smallest distance (N>4). In this embodiment, select N=6, that is, select 6 points with the smallest distance and store them in a 6*4 two-dimensional array D6. min , each row represents a sample point, and each column represents the mole fraction of a component.

[0104] Step S302, select 4 points from the N sample points, and store the selected sample point data into a two-dimensional array I15.

[0105] According to the principle of the quaternary system method, when a new point that needs to be predicted for viscosity is input, four close sample points are found from the sample library, and then the viscosity values ​​of the samples are used to predict the viscosity of the new input point.

[0106] It is easy to think of selecting the 4 points closest to the current input point, but in actual use, it is found that sometimes the weight factors calculated by using the 4 points closest to the current input point are unreasonable. For example, negative values ​​may appear, and one weight value may be very large while the other weight values ​​are very small. If such weights are used to predict the viscosity value, it is obviously inaccurate. To avoid this situation, in this embodiment, the 6 points closest to the current input point are pre-selected, and then 4 points are selected from the 6 points, for a total of There are 15 combinations, and the serial numbers of the 4 points included in these 15 combinations are stored in a two-dimensional array I15 of 15*4 elements.

[0107] Step S303, using the determinant to calculate the quaternion system weight factor of each combination.

[0108] The weight factor is calculated based on the mole fractions of the four components at each point. Suppose the mole fractions of the four components at the current input point are [C N,Ca ,C N,Mg ,C N,Si ,C N,Al ]; and the molar fractions of the four components of the four sample points in a certain combination of I15 are [C 1,Ca ,C 1,Mg ,C 1,Si , C 1,Al ],[C 2,Ca , C 2,Mg , C 2,Si , C 2,Al ],[C 3,Ca ,C 3,Mg ,C 3,Si , C 3,Al ],[C 4,Ca , C 4,Mg , C 4,Si ,C 4,Al ].

[0109] Calculate the determinant values ​​of the four components of the four sample points of the combination, set it as D4, that is:

[0110]

[0111] The weight factor of each sample point is calculated as follows:

[0112]

[0113]

[0114]

[0115]

[0116] The weight factors of 15 combinations are calculated in a loop and stored in a two-dimensional array B15 of 15*4 elements.

[0117] Step S40, defining a cost function of a weight factor, selecting a combination with a minimum cost function and using it to calculate a predicted value of slag viscosity.

[0118] Step S401, define a cost function, and calculate the cost function value of each combination.

[0119] A negative value in the weights indicates that the current input point is not within the space enclosed by the four points. The larger the absolute value of the negative value, the farther the deviation. Obviously, this is very不利 to the prediction result. Moreover, it is better for the sum of the weights of the four components to be close to 1.0. If the sum of the weights deviates too far from 1.0, it also indicates that the prediction result is not可取.

[0120] Based on the above considerations, the cost function is defined as:

[0121] Cost(k) = a × abs(W sum -1.0) + (1.0 - a) × ∑abs(W neg ) 1 ≤ k ≤ 15

[0122] In the formula, Cost(k) is the cost function value of the k-th combination, 1 ≤ k ≤ 15; a is the penalty coefficient, 0.0 < a < 1.0; abs is the operator for taking the absolute value, W sum is the sum of the weights of the four components, W neg is the negative weight, ∑abs(W neg ) represents the sum of the absolute values of all negative weights. The first term on the right side is used to penalize the difference between the sum of the weights and 1.0, and the second term is used to penalize the negative weights. The penalty coefficient a is used to adjust the proportion of these two penalty methods.

[0123] Step S402, sort the cost functions of each combination and select the combination with the smallest cost function.

[0124] A small cost function indicates few negative weights and the sum of the weights of the four components is close to 1.0. Then, the prediction effect using this combination will be relatively good; conversely, if the cost function is large, it means that the current input point deviates far from the selected four sample points, and the prediction effect will be very poor.

[0125] Sort the cost functions of each combination and select the combination with the smallest cost function as the optimal solution. Let the weight factors of the four sample points included in the optimal solution be b 1,opt 、b 2,opt 、b 3,opt 、b 4,opt , and the viscosity values of these four points are S1, S2, S3, S4.

[0126] Step S403, obtain the viscosity prediction value of the input point.

[0127] Calculate the predicted viscosity value of the input point according to the following formula:

[0128] S pre = b 1,opt × S1 + b 2,opt × S2 + b 3,opt × S3 + b 4,opt × S4

[0129] After the above steps, when the current slag composition data is obtained, the viscosity value of the slag can be calculated immediately, thereby solving the lag problem of slag viscosity measurement and completing the slag viscosity prediction method based on the hybrid linear programming algorithm.

[0130] In summary, the slag viscosity prediction method based on the hybrid linear programming algorithm provided by the embodiment of the present invention realizes the function of predicting the slag viscosity according to the composition data of the slag. The method first defines the fitting formula of the data in the viscosity sample library, and uses the least squares method to obtain the initial value and intercept value of the distance coefficient. Then, the data of each sample point is used to iteratively optimize the distance coefficient in the direction of gradient descent, and the optimal distance coefficient is selected according to the root mean square error value. Then, a number of sample points closest to the current input point are selected from the viscosity sample library, and 4 sample points are selected from them, and the weight factor of each combination is calculated using the determinant. Finally, the cost function of the weight factor is defined, and the combination with the smallest cost function is selected and used to calculate the predicted value of the slag viscosity. This method solves the problem of large hysteresis in the determination of slag viscosity. After obtaining the composition data of the slag, the current slag viscosity can be predicted immediately, so that adjustment measures can be taken in time to always control the viscosity within a suitable range, improve the quality of the blast furnace and pig iron, promote the desulfurization reaction, and extend the life of the blast furnace.

[0131] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A slag viscosity prediction method based on a hybrid linear programming algorithm, characterized in that: The method comprises the following steps: Step S10, collecting viscosity sample data of slag, forming a viscosity sample library, defining a fitting formula for the data in the viscosity sample library, and using the least squares method to obtain an initial value and an intercept value of the distance coefficient; Step S20, using the data of each sample point in the viscosity sample library to iteratively optimize the distance coefficient in the direction of gradient descent, and selecting the optimal distance coefficient according to the root mean square error value; Step S30, selecting a number of sample points closest to the current input point from the viscosity sample library, selecting 4 sample points from the sample points, and using the determinant to calculate the weight factor of each combination; Step S40, defining a cost function of the weight factor, selecting a combination with the minimum cost function and using it to calculate a predicted value of slag viscosity; The method for defining the fitting formula of the data in the viscosity sample library in step S10 is: The viscosity sample data measured by the viscometer are collected to form a sample library. Each sample point contains the molar fraction values ​​of the four components of the slag and the viscosity value obtained by the test. After taking the natural logarithm of the viscosity value, there is a better linear relationship between the two. The fitting formula for the data in the viscosity sample library is defined as: lnN=σ Ca ×C Ca +s Mg ×C Mg +s Si ×C Si +s Al ×C Al +β0 Where N is the viscosity value tested, σ Ca and C Ca are the distance coefficient and mole fraction of CaO component, σ Mg and C Mg are the distance coefficient and mole fraction of MgO component, σ Si and C Si are the distance coefficient and mole fraction of SiO2 component, σ Al and C Al are the distance coefficient and mole fraction of Al2O3 components, β0 is the intercept value in the linear fitting formula; Ca , σ Mg , σ Si , σ Al , β0 is the parameter to be identified; The method of sequentially using the data of each sample point in the step S20 to iteratively optimize the distance coefficient in the direction of gradient descent is: Starting from the first sample in the sample library, the estimated error of each sample is calculated sequentially. Assuming that the current sample is the mth sample, the estimated error of the sample is: In the formula, is the kth distance coefficient after updating with the m-1th sample in the sample library, 1≤k≤4, C k,m is the kth mole fraction of the mth sample, N m is the viscosity value of the mth sample; Update the estimated distance coefficient according to the error in the direction of gradient descent: In the formula, is the kth coefficient after m updates, R S is the learning rate parameter of the gradient descent method; The method of calculating the root mean square error according to the updated distance coefficient in step S20 and judging whether it is less than the error threshold is: The MSE root mean square error is used as the optimization objective function, so the objective function after the mth update is: Set the error threshold. If the MSE index after the coefficient update is lower than the error threshold, the optimization algorithm ends and the current estimated coefficient is taken as the final value of the distance coefficient. Right now: is the kth coefficient after m updates; If the MSE index after the coefficient update is higher than the error threshold, the next sample data will be used to execute the cycle; The method for sorting the MSE index of all sample points in step S20 and selecting the optimal distance coefficient value is: If all sample point data are used and all calculated MSE indicators are higher than the error threshold, the MSE indicators of all sample points are sorted, and the distance coefficient with the smallest MSE is taken as the final value of the distance coefficient. Assume that the minimum MSE corresponds to the jth sample point, then The method for defining the cost function in step S40 and calculating the cost function value of each combination is: The cost function is defined as: Cost(k)=a×abs(W sum -1.0)+(1.0-a)×∑abs(W neg )1≤k≤15 Where Cost(k) is the cost function value of the kth combination, 1≤k≤15; a is the penalty coefficient, 0.0<a<1.0; abs is the absolute value operator, W sum is the weighted sum of the four components, W neg is a negative weight, ∑abs(W neg ) means adding the absolute values ​​of all negative weights; the first term on the right is used to penalize the difference between the weight and 1.0, the second term is used to penalize negative weights, and the penalty coefficient a is used to adjust the proportion of these two penalty methods; The method of sorting the cost function of each combination in step S40 and selecting the combination with the smallest cost function is: Sort the cost function of each combination, select the combination with the smallest cost function as the optimal solution, and set the weight factor of the 4 sample points contained in the optimal solution to b 1,opt , b 2,opt , b 3,opt , b 4,opt , and the viscosity values ​​of these four points are S1, S2, S3, and S4; The method for obtaining the viscosity prediction value of the input point in step S40 is: The predicted viscosity value for the input point is calculated as follows: S pre =b 1,opt ×S1+b 2,opt ×S2+b 3,opt ×S3+b 4,opt ×S4 In this way, by following the above steps, the slag viscosity value can be immediately calculated based on the tested slag composition. If the slag viscosity is not within a suitable range, it can be adjusted in time to optimize the process performance of blast furnace production.

2. The slag viscosity prediction method based on the hybrid linear programming algorithm according to claim 1 is characterized in that: The method of using the least square method to obtain the initial value of the distance coefficient in step S10 is: The least squares parameter estimation method of the linear regression model is used to calculate the initial value of the distance coefficient. The calculation formula is as follows: In the formula, represents the initial value of the kth distance coefficient, 1≤k≤4, N i represents the viscosity value of the i-th sample in the sample library, S N is the number of samples in the sample library, C k,i represents the kth mole fraction of the ith sample, For S N The average value of the k-th mole fraction; The method for obtaining the intercept value according to the initial value of the distance coefficient in step S10 is: According to the initial value of the distance coefficient, the intercept value is calculated as follows:

3. The slag viscosity prediction method based on hybrid linear programming algorithm according to claim 1, characterized in that: The method of selecting a number of sample points closest to the current input point from the viscosity sample library in step S30 is: Calculate the distance between all sample points in the viscosity sample library and the current input point in a loop as follows: In the formula, D(x,i) is the distance between the current input point x and the point i in the sample library, C x,m is the molar fraction of the mth component at point x, C i,m is the molar fraction of the mth component at point i, σ m,final is the distance coefficient of the mth component, calculated in step S20; After finding the distances between all sample points and the current input point, select the smallest N points, N>4, and store them in a two-dimensional array for later use; The method of selecting 4 points from N sample points in step S30 is: Calculate all combinations of 4 points selected from N sample points, and store the data of the 4 points contained in each combination into a two-dimensional array for later use.

4. The slag viscosity prediction method based on hybrid linear programming algorithm according to claim 3 is characterized in that: The method of using determinant loop to calculate the weight factor of each combination of quaternary system method in step S30 is: The weight factor is calculated based on the mole fractions of the four components at each point. Suppose the mole fractions of the four components at the current input point are [C N,Ca ,C N,Mg ,C N,Si ,C N,Al ]; and the molar fractions of the four components of the four sample points in a certain combination are [C 1,Ca ,C 1,Mg ,C 1,Si ,C 1,Al ],[C 2,Ca ,C 2,Mg ,C 2,Si ,C 2,Al ],[C 3,Ca ,C 3,Mg ,C 3,Si ,C 3,Al ],[C 4,Ca ,C 4,Mg ,C 4,Si ,C 4,Al ]; Calculate the determinant values ​​of the four components of the four sample points of the combination, set it as D4, that is: The weight factor of each sample point is calculated as follows: Loop through the calculations to calculate the weight factor for each combination and store it in a two-dimensional array for later use.

Citation Information

Patent Citations

  • Economic dispatching method based on general wind power forecasting error model

    CN105846425A

  • Efficient converter protection method for predicting converter final slag viscosity model

    CN110991772A