CN112668234A - Intelligent control method for steelmaking endpoint of converter - Google Patents

Intelligent control method for steelmaking endpoint of converter Download PDF

Info

Publication number
CN112668234A
CN112668234A CN202011414366.5A CN202011414366A CN112668234A CN 112668234 A CN112668234 A CN 112668234A CN 202011414366 A CN202011414366 A CN 202011414366A CN 112668234 A CN112668234 A CN 112668234A
Authority
CN
China
Prior art keywords
end point
model
vector
subsystem
data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202011414366.5A
Other languages
Chinese (zh)
Other versions
CN112668234B (en
Inventor
高闯
李胜利
翟宝鹏
杨永辉
艾新港
李志刚
储茂祥
刘历铭
汪淼
孙悦
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Science and Technology Liaoning USTL
Original Assignee
University of Science and Technology Liaoning USTL
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Science and Technology Liaoning USTL filed Critical University of Science and Technology Liaoning USTL
Priority to CN202011414366.5A priority Critical patent/CN112668234B/en
Publication of CN112668234A publication Critical patent/CN112668234A/en
Application granted granted Critical
Publication of CN112668234B publication Critical patent/CN112668234B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P10/00—Technologies related to metal processing
    • Y02P10/25—Process efficiency

Landscapes

  • Carbon Steel Or Casting Steel Manufacturing (AREA)

Abstract

The invention provides an intelligent control method for a steelmaking endpoint of a converter, which is realized by the following subsystems: 1) the data preprocessing subsystem: collecting data from a database, preprocessing the data, determining the end point carbon content and the input variable of a temperature prediction subsystem model through independence and correlation analysis, and ensuring the model precision; 2) a molten steel end point prediction subsystem: predicting the end point carbon content and the end point temperature of converter steelmaking by adopting a non-parallel support vector regression algorithm based on wavelet weight; 3) oxygen blowing amount and auxiliary material calculating subsystem: calculating an optimization error according to output feedback of the prediction model by combining a whale swarm optimization algorithm and an incremental calculation method, and calculating the oxygen blowing amount, lime, light-burned dolomite and other auxiliary material addition amount required by the converting stage on the premise of ensuring the minimum optimization error; 4) the model updating subsystem: and updating and upgrading the prediction subsystem periodically according to the actual production condition. Can realize one-key steel making of the converter.

Description

Intelligent control method for steelmaking endpoint of converter
Technical Field
The invention relates to the technical field of converter steelmaking, in particular to an intelligent control method for a converter steelmaking endpoint.
Background
Converter steelmaking is the main steelmaking mode in China, and the quality of products is directly influenced by the control of end point components and temperature of molten steel. Because the smelting process of steel making is an extremely complex physical and chemical reaction process, the end point control of converter steel making is a research focus and a difficult point in the field of ferrous metallurgy, and the research on the control problem is carried out through several stages of empirical control, static control, dynamic control and intelligent control.
With the rapid development of computer technology and internet of things technology, a good foundation is laid for realizing one-key steel making. The intelligent control is a current research hotspot problem, the current mainstream control mode is to establish a model based on a neural network to predict the converter end point, and on the basis of the model, the oxygen blowing amount and the auxiliary material amount required by smelting are calculated by combining a mechanism or a mathematical method, so as to realize the end point control of the converter. The model has the defects that the model is easy to fall into a local minimum value, so that a global optimal solution cannot be obtained in the modeling process, inconvenience is brought to the searching of optimal parameters of the model, the modeling efficiency is low, and the realization of one-key steel making is not facilitated. A new modeling thought is explored, the learning performance and the efficiency of an intelligent model are improved, the full-automatic control of the converter is further realized, the field operation mode is effectively standardized, the splashing incidence rate is reduced, the converting time is shortened, the consumption of a slag former is reduced, the one-time carbon drawing rate of converter steelmaking is improved, and the method is a great innovation in the field of converter steelmaking.
Disclosure of Invention
In order to solve the technical problems in the background art, the invention provides an intelligent control method for a steelmaking endpoint of a converter, which can realize one-key steelmaking of the converter.
In order to achieve the purpose, the invention adopts the following technical scheme:
an intelligent control method for a converter steelmaking endpoint is realized by the following subsystems: the system comprises a data preprocessing subsystem, a molten steel end point predicting subsystem, an oxygen blowing amount and auxiliary material calculating subsystem and a model updating subsystem;
1) the data preprocessing subsystem: collecting data from a database, preprocessing the data, determining the end point carbon content and the input variable of a temperature prediction subsystem model through independence and correlation analysis, and ensuring the model precision;
2) a molten steel end point prediction subsystem: predicting the end point carbon content and the end point temperature of converter steelmaking by adopting a Wavelet weight-based non-parallel support vector regression algorithm (WTWWNPSVR);
3) oxygen blowing amount and auxiliary material calculating subsystem: calculating an optimization error according to output feedback of the prediction model by combining a whale swarm optimization algorithm and an incremental calculation method, and calculating the oxygen blowing amount, lime, light-burned dolomite and other auxiliary material addition amount required by the converting stage on the premise of ensuring the minimum optimization error;
4) the model updating subsystem: and updating and upgrading the prediction subsystem periodically according to the actual production condition.
Further, the data preprocessing method of the data preprocessing subsystem specifically includes the following steps:
step 1-1: reading n groups of converter data from a database, preprocessing the converter data, and removing irrelevant information such as smelting numbers, classes, furnace length names and the like to obtain n groups of converter data sets with m characteristic variables;
step 1-2: constructing an evaluation problem according to n groups of converter data and m characteristic variables, and determining a reference sequence and a comparison sequence; the original evaluation matrix is:
Figure BDA0002819690440000021
wherein, Fi＝[fi(1),...,fi(k),...,fi(n)]As a comparison sequence of the i-th characteristic variable, fi(k) Is the ith characteristic variable of the kth group of converter data;
determining a reference sequence R according to the evaluation purpose and the index condition0：
R0＝(r0(1),...,r0(k),...,r0(n)) (2)
For converter data, R0The output sequence of the finger model, i.e. end point carbon content data yCOr end point temperature data yT，r0(k) Is the output variable of the kth group of converter data;
step 1-3: normalizing reference sequence R0Comparing the sequence F to obtain matrix dimensionless data Y;
Figure BDA0002819690440000022
wherein, Y0＝(y0(1),...,y0(k),...,y0(n)) is a normalized reference sequence;
step 1-4: calculating a difference sequence omega; the difference sequence being the absolute value of the difference of the elements of each comparison sequence with the elements of the corresponding reference sequence, i.e.
Figure BDA0002819690440000031
Step 1-5: determining the maximum p and minimum v in the sequence of difference values, i.e.
Figure BDA0002819690440000032
Step 1-6: calculating the correlation coefficient, i.e. y, of each comparison sequence with the corresponding element of the reference sequencei(k) And y0(k) Coefficient of correlation between
Figure BDA0002819690440000037
：
Figure BDA0002819690440000033
Wherein rho epsilon (0,1) is an adjustable parameter;
step 1-7: calculating the average of the correlation coefficient of each feature to form a correlation sequence, i.e.
Figure BDA0002819690440000034
The larger the correlation coefficient is, the larger the influence of the corresponding input factor on the output variable is;
step 1-8: according to the grey correlation coefficient gamma0(i) Arranging the input variables from large to small, selecting l (l is less than or equal to m) features which have large influence on the output variables, and preliminarily taking the features as corresponding input variables;
step 1-9: performing independence analysis on the obtained input variable l by adopting a partial correlation analysis method to ensure that the input variables are independent or have small correlation; any two input variables xiAnd xjG (g is less than or equal to l-2) order partial correlation coefficient between
Figure BDA0002819690440000035
Can be calculated by the following formula:
Figure BDA0002819690440000036
wherein the right side of equation (8) represents the partial correlation coefficient of g-1; the partial correlation coefficient is a statistic which can really reflect the correlation between two variables; if the partial correlation coefficient between the two variables is small, the correlation between the two variables is small or even irrelevant;
step 1-10: through mechanism analysis and combined with correlation and independence analysis, d (d is less than or equal to l) influencing factors can be finally determined as the end point carbon content yCOr end point temperature yTInput variables of the prediction model are defined as x ═ x1,x2,...,xd]T∈Rd×1(ii) a The end point carbon content or end point temperature is used as an output variable of the prediction model.
Further, in the molten steel endpoint prediction subsystem, the method for predicting the endpoint carbon content and the endpoint temperature of converter steelmaking specifically comprises the following steps:
step 2-1: reading converter steelmaking input data, end point carbon content data and end point temperature data of Step 1-10, and determining the number L of training samples in a uniform sampling mode1(L1<n)；
Step 2-2: establishing a converter steelmaking endpoint information prediction model (WTWNPPSVR) with anti-noise performance; the model is based on NPSVR and introduces a parameter v1And v2Meanwhile, the weight among the samples is considered; assume a data set of
Figure BDA0002819690440000041
Is a Gaussian kernel function, order
Figure BDA0002819690440000042
To input a training sample, y ═ y1,...,yn]T∈RnIn order to output the training samples,
Figure BDA0002819690440000043
for the ith training sample, the objective function of the algorithm can be described as:
Figure BDA0002819690440000044
Figure BDA0002819690440000045
wherein, ci(i ≧ 1., 4) ≧ 0 is a penalty parameter, v1,v2,ε1,ε2Epsilon is more than or equal to 0 and is an adjustable parameter eta1，
Figure BDA0002819690440000047
ξ1，η2，
Figure BDA0002819690440000046
And xi2Is a relaxation variable, W ∈ Rn×1Is a weight vector of samples, [ w ]1；b1]And [ w2；b2]To augment the vector, e ═ 1,1]T∈Rn×1；
The target function (9) is taken as an example for explanation, and the target function (10) has similar explanation as the target function (9); the purpose of the first and second terms in the constraint is to determine two hyperplanes
Figure BDA0002819690440000058
And
Figure BDA0002819690440000059
so that the training samples are located as much as possible in the middle of the two hyperplanes; the first term of the objective function is a regularization term derived from a standard support vector regression algorithm; constraint the third term is to hyperplane the training samples to the lower boundary
Figure BDA0002819690440000051
Has a distance of at least epsilon1In other words, the training samples are positioned as far as possible
Figure BDA0002819690440000052
Upper side of (a); the second and third terms of the objective function are used to minimize the relaxation factor
Figure BDA0002819690440000053
And xi1And e1The width of the pipeline, and the coefficient vector W is a penalty vector of the relaxation factor;
by deriving the above model, the corresponding dual problem can be obtained:
Figure BDA0002819690440000054
Figure BDA0002819690440000055
wherein alpha is1＝[α11,....,α1n]T,
Figure BDA00028196904400000510
β1＝[β11,...,β1n]T,α2＝[α21,....,α2n]T,
Figure BDA00028196904400000511
And beta2＝[β21,...,β2n]TIs a lagrange multiplier vector;
step 2-3: based on wavelet transform theory, determining weight vector W ═ d of sample1,d2,···,dn]T，diRepresents a weight coefficient, and can be obtained by the following formula:
Figure BDA0002819690440000056
where exp (. cndot.) is an e exponential function, δ is the standard deviation of the Gaussian function, riIs the distance between the original data and the wavelet transformed denoised data;
step 2-4: initializing parameter c of WWNPSVR prediction modeli(i＝1,...,4),v1,v2,ε1,ε2ε and δ;
step 2-5: training by using a converter steelmaking training sample set; by solving equations (11) and (12), an optimal solution vector α can be obtained1，
Figure BDA00028196904400000512
β1，α2，
Figure BDA00028196904400000513
And beta2；
Step 2-6: by substituting the optimal solution into the following equations (14) and (15), w can be obtained1,b1And w2,b2；
Figure BDA0002819690440000061
Figure BDA0002819690440000062
Wherein, | SkL represents the number of support vectors;
step 2-7: will w1,b1And w2,b2The result of (3) is substituted into the formula (16), and the regression function f of the end point carbon content can be obtainedC(x) Or a temperature regression function fT(x)；
Figure BDA0002819690440000063
Step 2-8: substituting training samples into the regression function fC(x) Or fT(x) Obtaining a predicted value of the end point carbon content or the end point temperature; calculating indexes such as model precision and end point hit rate, completing model establishment if the indexes reach set values, or updating the parameter c of the WTWWNPSVR prediction model1,c2,c3,c4,v1,v2,ε1,ε2E, delta, repeating Step 2-5 to Step 2-7 until the index reaches the set value,and completing model building.
Further, the oxygen blowing amount and auxiliary material calculating method of the oxygen blowing amount and auxiliary material calculating subsystem specifically comprises the following steps:
step 3-1: the objective function of the following optimization problem of the oxygen material addition amount is established as follows:
Figure BDA0002819690440000064
wherein ffit (x) ═ fC/t (x) — DC/T)2Called fitness function, x is an input vector composed of variables such as oxygen blowing amount, lime addition amount and light-burned dolomite addition amount, and fC/T(x) Carbon temperature prediction for the prediction subsystem in Step2-8, DC/TA target value for end point carbon content or end point temperature;
step 3-2: setting the number j of whale groups and the maximum iteration number Nmax(ii) a Using the variables of oxygen blowing amount, lime addition amount and the like as optimization variables, carrying out normalization treatment on the variables, and mapping to [ 1-]Generating an equal amount of initial random solutions according to the set whale colony number j;
step 3-3: combining each group of initial solution variables with initial information of a furnace of molten iron to obtain x, respectively substituting into a prediction subsystem in Step2-8 to obtain a predicted value f of carbon content and temperatureC/T(x)；
Step 3-4: according to the fitness function f in equation (17)fit(x) Calculating the fitness of each group of solutions, and storing the optimal vector x with the minimum current fitness*；
Step 3-5: whale population optimization strategy using equation (18) if the current iteration number is less than NmaxThen a, r are updated1E, eta, k and p, determining a solution required by the next iteration, detecting whether a solution exceeding a search space exists, if so, mapping the solution to a random position in a feasible domain, and repeating the steps of Step3-4 and Step 3-5; otherwise, returning to the optimal solution, completing whale colony optimization of oxygen amount, and obtaining the optimal vector B of oxygen blowing amount and auxiliary material addition amount*＝invnorm(x*) Invnorm represents the inverse normalization process;
Figure BDA0002819690440000071
where t is the current iteration number, x (t) represents the position vector of the current whale, and λ ═ 2r1·xrand-x(t)|，xrandIs the random position of whale, r1Is [ -1,1 [ ]]Random vector within interval, η is a constant, E is coefficient vector, x*(t) represents the position vector of the current optimal solution, λzRepresenting the distance between whale and prey, a being a variable decreasing from 2 to 0, k being [ -1,1 [ -1 [ ]]Random vectors within the interval, z representing a probability variable;
step 3-6: by using the formula (17), an incremental calculation model is established, and the search interval of oxygen blowing amount is set to [ M ]1,N1]Step length of l1(ii) a The search interval for the amount of added adjuvant is set to [ M ]2,N2]Step length of l2；
Step 3-7: setting the initial values of oxygen blowing amount and auxiliary material addition amount in the same group of samples optimized by whale colony as M1,M2]Then combining the model input vector x with molten iron information to obtain a model input vector x, substituting the model input vector x into a prediction subsystem in Step2-8 to obtain a predicted value f of carbon content and temperatureC/T(x) (ii) a The fitness value is obtained according to the formula (17)
Figure BDA0002819690440000073
Saving vector C*＝x；
Step 3-8: according to the search interval l1And l2Gradually updating the values of the oxygen blowing amount and the auxiliary material addition amount to form a new model output variable xnewSubstituting the prediction subsystem in Step2-8 to obtain the predicted value f of carbon content and temperatureC/T(xnew) (ii) a The fitness value f is obtained according to the formula (17)fit(xnew) If, if
Figure BDA0002819690440000072
Then the vector C is saved*＝xnewAnd fitness value
Figure BDA0002819690440000081
When the oxygen blowing amount reaches N1The adding amount of the auxiliary materials reaches N2If so, executing Step 3-9, otherwise, repeating Step 3-8;
step 3-9: optimal solution B obtained from Step3-5 and Step3-8*And C*Extracting new vector composed of oxygen blowing amount and auxiliary material addition amount
Figure BDA0002819690440000082
And
Figure BDA0002819690440000083
substitution formula
Figure BDA0002819690440000084
Obtaining the optimal vector A of the oxygen blowing amount and the addition amount of the auxiliary materials, wherein K1And K2Is a weight coefficient;
step 3-10: by adjusting K1And K2The final oxygen blowing amount and the auxiliary material addition amount A are obtained*(ii) a And transmitting the final results of the oxygen blowing amount and the auxiliary material calculation subsystem to a database, reading data from the database by a PLC (programmable logic controller) system, determining a blowing mode and blowing stop time according to the optimal results, and finally sending a control instruction to a converter execution mechanism to complete the whole blowing process.
Further, the method for updating and upgrading the model updating subsystem specifically comprises the following steps:
step 4-1: when the newly added smelting data amount reaches L2Then, a population vector omega of the whale population is set0＝[Ω1,...,Ωl]TWherein Ω isi＝[c1,i,c2,i,c3,i,c4,i,ε1,i,ε2,i,εi,δi]TI 1.. times.l, a maximum number of iterations T is setmax(ii) a Defining a model parameter optimization fitness function as follows:
Figure BDA0002819690440000085
in the formula (I), the compound is shown in the specification,
Figure BDA0002819690440000086
size (U,1) represents vector x in matrix UiThe number of the (c) is,
Figure BDA0002819690440000087
is a predicted value of carbon content and temperature, y (x)i) Is the actual value of carbon content and temperature, τ is the upper error bound, L1The number of training samples determined in Step 2-1;
step 4-2: using equation (19), for each set of parameters
Figure BDA0002819690440000088
Calculating fitness
Figure BDA0002819690440000089
Obtaining the current optimal fitness f*(Ω*) And an optimum parameter omega*；
Step 4-3: if the current iteration number is less than TmaxUpdate omega*Obtaining the next set of parameters omegaiThe fitness f (Ω) is calculated using the formula (19)i)；
Step 4-4: if f is*(Ω*)≤f(Ωi) Then let Ω*＝ΩiAnd
Figure BDA00028196904400000810
returning to Step 4-3; otherwise, directly returning to Step 4-3;
step 4-5: if the current iteration number is equal to TmaxThen the global optimum parameter omega is output*(ii) a And completing the updating of the model.
Compared with the prior art, the invention has the beneficial effects that:
1. the molten steel end point prediction subsystem adopts a Wavelet weight-based non-parallel support vector machine algorithm (WTWWNPSVR) to realize the prediction of the end point carbon content and the end point temperature of converter steelmaking. WWNPSVR is a new pattern recognition method based on statistical learning theory, which seeks the best trade-off between model complexity and learning ability based on limited sample information to achieve the best generalization performance. The WWNPSVR algorithm can overcome the inherent defects of the neural network in the modeling process, and has the following advantages: 1) the method is specially aimed at the limited sample condition, and aims to obtain the optimal solution under the existing information, not only the optimal value when the number of samples tends to infinity; 2) the NPSVR algorithm is finally converted into two quadratic convex programming problems, theoretically, the obtained solution is necessarily a global optimal solution, and the problem that the neural network prediction model is easy to fall into a local minimum value in the modeling process is solved; 3) the algorithm converts the actual problem into a high-dimensional characteristic space through nonlinear transformation, and constructs a linear discriminant function in the high-dimensional space to replace the nonlinear discriminant function in the original space, and the special property skillfully solves the problem of dimension while ensuring that the model has better predictive performance, thereby avoiding dimension disaster. 4) WTWNPPSVR adopts a wavelet transform method to establish a prediction model, so that adverse effects of noise can be suppressed, the accuracy of end point prediction is improved, the yield of target products is improved, and energy conservation and emission reduction are realized.
2. The oxygen blowing amount and auxiliary material calculating subsystem adopts a combined calculating model, combines the advantages of whale colony optimization and incremental calculation, can further improve the calculation precision of the oxygen amount, improves the primary carbon drawing rate of converter steelmaking, and can save the labor cost for enterprises.
3. And the model updating subsystem adopts a Whale Optimization Algorithm (WOA) to optimize parameters in the prediction model, so that updating and upgrading of the prediction subsystem are realized. The parameter value has important significance on the generalization capability and stability of the model. Usually, a grid search method is adopted to select the optimal parameters of the model, but the grid search method is slow and has long running time. In order to overcome the problem, the method provided by the invention combines the advantages of high convergence speed of the WOA algorithm, simple adjustment parameters and the like, regularly updates and upgrades the molten steel terminal point prediction subsystem, can effectively avoid the problem of falling into local optimum, improves the updating efficiency of the model, and has reliability.
Drawings
FIG. 1 is a structural diagram of an intelligent control model for converter steelmaking according to the present invention;
FIG. 2 is a flow chart of a molten steel endpoint prediction subsystem of the present invention;
FIG. 3 is a flow chart of the model update subsystem of the present invention.
Detailed Description
The following detailed description of the present invention will be made with reference to the accompanying drawings.
As shown in fig. 1, an intelligent control method for a converter steelmaking endpoint is implemented by the following subsystems: the system comprises a data preprocessing subsystem, a molten steel end point predicting subsystem, an oxygen blowing amount and auxiliary material calculating subsystem and a model updating subsystem.
1) The data preprocessing subsystem: collecting data from a database, preprocessing the data, determining the end point carbon content and the input variable of a temperature prediction subsystem model through independence and correlation analysis, and ensuring the model precision; the method specifically comprises the following steps:
step 1-1: reading n groups of converter data from a database, preprocessing the converter data, and removing irrelevant information such as smelting numbers, classes, furnace length names and the like to obtain n groups of converter data sets with m characteristic variables;
step 1-2: constructing an evaluation problem according to n groups of converter data and m characteristic variables, and determining a reference sequence and a comparison sequence; the original evaluation matrix is:
Figure BDA0002819690440000101
wherein, Fi＝[fi(1),...,fi(k),...,fi(n)]As a comparison sequence of the i-th characteristic variable, fi(k) Is the ith characteristic variable of the kth group of converter data;
determining a reference sequence R according to the evaluation purpose and the index condition0：
R0＝(r0(1),...,r0(k),...,r0(n)) (2)
For converter data, R0The output sequence of the finger model, i.e. end point carbon content data yCOr end point temperature data yT，r0(k) Is the output variable of the kth group of converter data;
step 1-3: normalizing reference sequence R0Comparing the sequence F to obtain matrix dimensionless data Y;
Figure BDA0002819690440000102
wherein, Y0＝(y0(1),...,y0(k),...,y0(n)) is a normalized reference sequence;
step 1-4: calculating a difference sequence omega; the difference sequence being the absolute value of the difference of the elements of each comparison sequence with the elements of the corresponding reference sequence, i.e.
Figure BDA0002819690440000111
Step 1-5: determining the maximum p and minimum v in the sequence of difference values, i.e.
Figure BDA0002819690440000112
Step 1-6: calculating the correlation coefficient, i.e. y, of each comparison sequence with the corresponding element of the reference sequencei(k) And y0(k) Coefficient of correlation between
Figure BDA0002819690440000117
：
Figure BDA0002819690440000113
Wherein rho epsilon (0,1) is an adjustable parameter;
step 1-7: calculating the average of the correlation coefficient of each feature to form a correlation sequence, i.e.
Figure BDA0002819690440000114
The larger the correlation coefficient is, the larger the influence of the corresponding input factor on the output variable is;
step 1-8: according to the grey correlation coefficient gamma0(i) Arranging the input variables from large to small, selecting l (l is less than or equal to m) features which have large influence on the output variables, and preliminarily taking the features as corresponding input variables;
step 1-9: performing independence analysis on the obtained input variable l by adopting a partial correlation analysis method to ensure that the input variables are independent or have small correlation; any two input variables xiAnd xjG (g is less than or equal to l-2) order partial correlation coefficient between
Figure BDA0002819690440000115
Can be calculated by the following formula:
Figure BDA0002819690440000116
wherein the right side of equation (8) represents the partial correlation coefficient of g-1; the partial correlation coefficient is a statistic which can really reflect the correlation between two variables; if the partial correlation coefficient between the two variables is small, the correlation between the two variables is small or even irrelevant;
step 1-10: through mechanism analysis and combined with correlation and independence analysis, d (d is less than or equal to l) influencing factors can be finally determined as the end point carbon content yCOr end point temperature yTInput variables of the prediction model are defined as x ═ x1,x2,...,xd]T∈Rd×1(ii) a The end point carbon content or end point temperature is used as an output variable of the prediction model.
2) A molten steel end point prediction subsystem: predicting the end point carbon content and the end point temperature of converter steelmaking by adopting a Wavelet transformed weighted non-parallel support vector regression (WTWWNPSVR) algorithm; as shown in fig. 2, the method specifically includes the following steps:
step 2-1: reading converter steelmaking input data, end point carbon content data and end point temperature data of Step 1-10, and determining the number L of training samples in a uniform sampling mode1(L1<n)；
Step 2-2: establishing a converter steelmaking endpoint information prediction model (WTWNPPSVR) with anti-noise performance; the model is based on NPSVR and introduces a parameter v1And v2Meanwhile, the weight among the samples is considered; assume a data set of
Figure BDA0002819690440000121
Is a Gaussian kernel function, order
Figure BDA0002819690440000122
To input a training sample, y ═ y1,...,yn]T∈RnIn order to output the training samples,
Figure BDA0002819690440000123
for the ith training sample, the objective function of the algorithm can be described as:
Figure BDA0002819690440000124
Figure BDA0002819690440000125
wherein, ci(i ≧ 1., 4) ≧ 0 is a penalty parameter, v1,v2,ε1,ε2Epsilon is more than or equal to 0 and is an adjustable parameter eta1，
Figure BDA00028196904400001210
ξ1，η2，
Figure BDA0002819690440000129
And xi2Is a relaxation variable, W ∈ Rn×1Is a weight vector of samples, [ w ]1；b1]And [ w2；b2]To augment the vector, e ═ 1,1]T∈Rn×1；
The target function (9) is taken as an example for explanation, and the target function (10) has similar explanation as the target function (9); the purpose of the first and second terms in the constraint is to determine two hyperplanes
Figure BDA0002819690440000127
And
Figure BDA0002819690440000128
so that the training samples are located as much as possible in the middle of the two hyperplanes; the first term of the objective function is a regularization term, derived from the standard SVR; constraint the third term is to hyperplane the training samples to the lower boundary
Figure BDA0002819690440000131
Has a distance of at least epsilon1In other words, the training samples are positioned as far as possible
Figure BDA0002819690440000132
Upper side of (a); the second and third terms of the objective function are used to minimize the relaxation factor eta1，
Figure BDA0002819690440000138
And xi1And e1The width of the pipeline, and the coefficient vector W is a penalty vector of the relaxation factor;
by deriving the above model, the corresponding dual problem can be obtained:
Figure BDA0002819690440000134
Figure BDA0002819690440000135
wherein alpha is1＝[α11,....,α1n]T,
Figure BDA0002819690440000139
β1＝[β11,...,β1n]T,α2＝[α21,....,α2n]T,
Figure BDA00028196904400001310
And beta2＝[β21,...,β2n]TIs a lagrange multiplier vector;
step 2-3: based on wavelet transform theory, determining weight vector W ═ d of sample1,d2,···,dn]T，diRepresents a weight coefficient, and can be obtained by the following formula:
Figure BDA0002819690440000136
where exp (. cndot.) is an e exponential function, δ is the standard deviation of the Gaussian function, riIs the distance between the original data and the wavelet transformed denoised data;
step 2-4: initializing parameter c of WWNPSVR prediction modeli(i＝1,...,4),v1,v2,ε1,ε2ε and δ;
step 2-5: training by using a converter steelmaking training sample set; by solving equations (11) and (12), an optimal solution vector α can be obtained1，
Figure BDA00028196904400001311
β1，α2，
Figure BDA00028196904400001312
And beta2；
Step 2-6: by substituting the optimal solution into the following equations (14) and (15), w can be obtained1,b1And w2,b2；
Figure BDA0002819690440000141
Figure BDA0002819690440000142
Wherein, | SkL represents the number of support vectors;
step 2-7: will w1,b1And w2,b2The result of (3) is substituted into the formula (16), and the regression function f of the end point carbon content can be obtainedC(x) Or a temperature regression function fT(x)；
Figure BDA0002819690440000143
Step 2-8: substituting training samples into the regression function fC(x) Or fT(x) Obtaining a predicted value of the end point carbon content or the end point temperature; calculating indexes such as model precision and end point hit rate, completing model establishment if the indexes reach set values, or updating the parameter c of the WTWWNPSVR prediction model1,c2,c3,c4,v1,v2,ε1,ε2And e, repeating Step 2-5 to Step 2-7 until the index reaches a set value, and finishing model building.
3) Oxygen blowing amount and auxiliary material calculating subsystem: calculating an optimization error according to output feedback of the prediction model by combining a whale swarm optimization algorithm and an incremental calculation method, and calculating the oxygen blowing amount, lime, light-burned dolomite and other auxiliary material addition amount required by the converting stage on the premise of ensuring the minimum optimization error; the method specifically comprises the following steps:
step 3-1: the objective function of the following optimization problem of the oxygen material addition amount is established as follows:
Figure BDA0002819690440000144
in the formula (f)fit(x)＝(fC/T(x)-DC/T)2Called fitness function, x is an input variable composed of oxygen blowing amount, lime addition amount, light-burned dolomite addition amount and other variables, fC/T(x) Carbon temperature prediction for the prediction subsystem in Step2-8, DC/TA target value for end point carbon content or end point temperature;
step 3-2: setting the number j of whale groups and the maximum iteration number Nmax(ii) a Using the variables of oxygen blowing amount, lime addition amount and the like as optimization variables, carrying out normalization treatment on the variables, and mapping to [ 1-]Generating an equal amount of initial random solutions according to the set whale colony number j;
step 3-3: combining each group of initial solution variables with initial information of a furnace of molten iron to obtain x, respectively substituting into a prediction subsystem in Step2-8 to obtain a predicted value f of carbon content and temperatureC/T(x)；
Step 3-4: according to the fitness function f in equation (17)fit(x) Calculating the fitness of each group of solutions, and storing the optimal vector x with the minimum current fitness*；
Step 3-5: whale population optimization strategy using equation (18) if the current iteration number is less than NmaxThen a, r are updated1E, eta, k and p, determining a solution required by the next iteration, detecting whether a solution exceeding a search space exists, if so, mapping the solution to a random position in a feasible domain, and repeating the steps of Step3-4 and Step 3-5; otherwise, returning to the optimal solution, completing whale colony optimization of oxygen amount, and obtaining the optimal vector B of oxygen blowing amount and auxiliary material addition amount*＝invnorm(x*) Invnorm represents the inverse normalization process;
Figure BDA0002819690440000151
where t is the current iteration number, x (t) represents the position vector of the current whale, and λ ═ 2r1·xrand-x(t)|，xrandIs the random position of whale, r1Is [ -1,1 [ ]]Random direction within intervalQuantity, eta is a constant, E is a coefficient vector, x*(t) represents the position vector of the current optimal solution, λzRepresenting the distance between whale and prey, a being a variable decreasing from 2 to 0, k being [ -1,1 [ -1 [ ]]Random vectors within the interval, z representing a probability variable;
step 3-6: by using the formula (17), an incremental calculation model is established, and the search interval of oxygen blowing amount is set to [ M ]1,N1]Step length of l1(ii) a The search interval for the amount of added adjuvant is set to [ M ]2,N2]Step length of l2；
Step 3-7: setting the initial values of oxygen blowing amount and auxiliary material addition amount in the same group of samples optimized by whale colony as M1,M2]Then combining the model input vector x with molten iron information to obtain a model input vector x, substituting the model input vector x into a prediction subsystem in Step2-8 to obtain a predicted value f of carbon content and temperatureC/T(x) (ii) a The fitness value is obtained according to the formula (17)
Figure BDA0002819690440000153
Saving vector C*＝x；
Step 3-8: according to the search interval l1And l2Gradually updating the values of the oxygen blowing amount and the auxiliary material addition amount to form a new model output variable xnewSubstituting the prediction subsystem in Step2-8 to obtain the predicted value f of carbon content and temperatureC/T(xnew) (ii) a The fitness value f is obtained according to the formula (17)fit(xnew) If, if
Figure BDA0002819690440000152
Then the vector C is saved*＝xnewAnd fitness value
Figure BDA0002819690440000161
When the oxygen blowing amount reaches N1The adding amount of the auxiliary materials reaches N2If so, executing Step 3-9, otherwise, repeating Step 3-8;
step 3-9: optimal solution B obtained from Step3-5 and Step3-8*And C*Extracting new vector composed of oxygen blowing amount and auxiliary material addition amount
Figure BDA0002819690440000162
And
Figure BDA0002819690440000163
substitution formula
Figure BDA0002819690440000164
Obtaining the optimal vector A of the oxygen blowing amount and the addition amount of the auxiliary materials, wherein K1And K2Is a weight coefficient;
step 3-10: by adjusting K1And K2The final oxygen blowing amount and the auxiliary material addition amount A are obtained*(ii) a And transmitting the final results of the oxygen blowing amount and the auxiliary material calculation subsystem to a database, reading data from the database by a PLC (programmable logic controller) system, determining a blowing mode and blowing stop time according to the optimal results, and finally sending a control instruction to a converter execution mechanism to complete the whole blowing process.
4) The model updating subsystem: updating and upgrading the prediction subsystem periodically according to the actual production condition; as shown in fig. 3, the specific steps include the following:
step 4-1: when the newly added smelting data amount reaches L2Then, a population vector omega of the whale population is set0＝[Ω1,...,Ωl]TWherein Ω isi＝[c1,i,c2,i,c3,i,c4,i,ε1,i,ε2,i,εi,δi]TI 1.. times.l, a maximum number of iterations T is setmax(ii) a Defining a model parameter optimization fitness function as follows:
Figure BDA0002819690440000165
in the formula (I), the compound is shown in the specification,
Figure BDA0002819690440000166
size (U,1) represents vector x in matrix UiThe number of the (c) is,
Figure BDA0002819690440000167
is a predicted value of carbon content and temperature, y (x)i) Is the actual value of carbon content and temperature, τ is the upper error bound, L1The number of training samples determined in Step 2-1;
step 4-2: using equation (19), for each set of parameters
Figure BDA0002819690440000168
Calculating fitness
Figure BDA0002819690440000169
Obtaining the current optimal fitness f*(Ω*) And an optimum parameter omega*；
Step 4-3: if the current iteration number is less than TmaxUpdate omega*Obtaining the next set of parameters omegaiThe fitness f (Ω) is calculated using the formula (19)i)；
Step 4-4: if f is*(Ω*)≤f(Ωi) Then let Ω*＝ΩiAnd
Figure BDA00028196904400001610
returning to Step 4-3; otherwise, directly returning to Step 4-3;
step 4-5: if the current iteration number is equal to TmaxThen the global optimum parameter omega is output*(ii) a And completing the updating of the model.
The above embodiments are implemented on the premise of the technical solution of the present invention, and detailed embodiments and specific operation procedures are given, but the scope of the present invention is not limited to the above embodiments. The methods used in the above examples are conventional methods unless otherwise specified.

Claims (5)

1. The intelligent control method for the steelmaking endpoint of the converter is characterized by being realized by the following subsystems: the system comprises a data preprocessing subsystem, a molten steel end point predicting subsystem, an oxygen blowing amount and auxiliary material calculating subsystem and a model updating subsystem;
1) the data preprocessing subsystem: collecting data from a database, preprocessing the data, determining the end point carbon content and the input variable of a temperature prediction subsystem model through independence and correlation analysis, and ensuring the model precision;
2) a molten steel end point prediction subsystem: predicting the end point carbon content and the end point temperature of converter steelmaking by adopting a non-parallel support vector regression algorithm based on wavelet weight;
3) oxygen blowing amount and auxiliary material calculating subsystem: calculating an optimization error according to output feedback of the prediction model by combining a whale swarm optimization algorithm and an incremental calculation method, and calculating the oxygen blowing amount, lime, light-burned dolomite and other auxiliary material addition amount required by the converting stage on the premise of ensuring the minimum optimization error;
4) the model updating subsystem: and updating and upgrading the prediction subsystem periodically according to the actual production condition.
2. The intelligent control method for the steelmaking endpoint of the converter as claimed in claim 1, wherein the data preprocessing method of the data preprocessing subsystem specifically comprises the following steps:
step 1-1: reading n groups of converter data from a database, preprocessing the converter data, and removing irrelevant information such as smelting numbers, classes, furnace length names and the like to obtain n groups of converter data sets with m characteristic variables;
step 1-2: constructing an evaluation problem according to n groups of converter data and m characteristic variables, and determining a reference sequence and a comparison sequence; the original evaluation matrix is:
Figure FDA0002819690430000011
wherein, Fi＝[fi(1),...,fi(k),...,fi(n)]As a comparison sequence of the i-th characteristic variable, fi(k) Is the ith characteristic variable of the kth group of converter data;
determining a reference sequence R according to the evaluation purpose and the index condition0：
R0＝(r0(1),...,r0(k),...,r0(n)) (2)
For converter data, R0The output sequence of the finger model, i.e. end point carbon content data yCOr end point temperature data yT，r0(k) Is the output variable of the kth group of converter data;
step 1-3: normalizing reference sequence R0Comparing the sequence F to obtain matrix dimensionless data Y;
Figure FDA0002819690430000021
wherein, Y0＝(y0(1),...,y0(k),...,y0(n)) is a normalized reference sequence;
step 1-4: calculating a difference sequence omega; the difference sequence being the absolute value of the difference of the elements of each comparison sequence with the elements of the corresponding reference sequence, i.e.
Figure FDA0002819690430000022
Step 1-5: determining the maximum p and minimum v in the sequence of difference values, i.e.
Figure FDA0002819690430000023
Step 1-6: calculating the correlation coefficient, i.e. y, of each comparison sequence with the corresponding element of the reference sequencei(k) And y0(k) Coefficient of correlation between
Figure FDA0002819690430000026
Figure FDA0002819690430000024
Wherein rho epsilon (0,1) is an adjustable parameter;
step 1-7: calculating the average of the correlation coefficient of each feature to form a correlation sequence, i.e.
Figure FDA0002819690430000025
The larger the correlation coefficient is, the larger the influence of the corresponding input factor on the output variable is;
step 1-8: according to the grey correlation coefficient gamma0(i) Arranging the input variables from large to small, selecting l (l is less than or equal to m) features which have large influence on the output variables, and preliminarily taking the features as corresponding input variables;
step 1-9: performing independence analysis on the obtained input variable l by adopting a partial correlation analysis method to ensure that the input variables are independent or have small correlation; any two input variables xiAnd xjG (g is less than or equal to l-2) order partial correlation coefficient between
Figure FDA0002819690430000031
Can be calculated by the following formula:
Figure FDA0002819690430000032
wherein the right side of equation (8) represents the partial correlation coefficient of g-1; the partial correlation coefficient is a statistic which can really reflect the correlation between two variables; if the partial correlation coefficient between the two variables is small, the correlation between the two variables is small or even irrelevant;
step 1-10: through mechanism analysis and combined with correlation and independence analysis, d (d is less than or equal to l) influencing factors can be finally determined as the end point carbon content yCOr end point temperature yTInput variables of the prediction model are defined as x ═ x1,x2,...,xd]T∈Rd×1(ii) a The end point carbon content or end point temperature is used as an output variable of the prediction model.
3. The intelligent control method for the converter steelmaking endpoint according to claim 1, wherein in the molten steel endpoint prediction subsystem, the method for predicting the endpoint carbon content and the endpoint temperature of converter steelmaking specifically comprises the following steps:
step 2-1: reading converter steelmaking input data, end point carbon content data and end point temperature data of Step 1-10, and determining the number L of training samples in a uniform sampling mode1(L1<n)；
Step 2-2: establishing a converter steelmaking endpoint information prediction model with anti-noise performance; the model is based on NPSVR and introduces a parameter v1And v2Meanwhile, the weight among the samples is considered; assume a data set of
Figure FDA0002819690430000033
Figure FDA0002819690430000034
Is a Gaussian kernel function, order
Figure FDA0002819690430000035
To input a training sample, y ═ y1,...,yn]T∈RnIn order to output the training samples,
Figure FDA0002819690430000036
for the ith training sample, the objective function of the algorithm can be described as:
Figure FDA0002819690430000037
Figure FDA0002819690430000041
wherein, ci(i＝1,..,4) ≧ 0 is a penalty parameter, v1,v2,ε1,ε2Epsilon is more than or equal to 0 and is an adjustable parameter eta1,
Figure FDA0002819690430000042
ξ1,η2,
Figure FDA0002819690430000043
And xi2Is a relaxation variable, W ∈ Rn×1Is a weight vector of samples, [ w ]1；b1]And [ w2；b2]To augment the vector, e ═ 1,1]T∈Rn×1；
The target function (9) is taken as an example for explanation, and the target function (10) has similar explanation as the target function (9); the purpose of the first and second terms in the constraint is to determine two hyperplanes
Figure FDA0002819690430000044
And
Figure FDA0002819690430000045
so that the training samples are located as much as possible in the middle of the two hyperplanes; the first term of the objective function is a regularization term, derived from the standard SVR; constraint the third term is to hyperplane the training samples to the lower boundary
Figure FDA0002819690430000046
Has a distance of at least epsilon1In other words, the training samples are positioned as far as possible
Figure FDA0002819690430000047
Upper side of (a); the second and third terms of the objective function are used to minimize the relaxation factor eta1,
Figure FDA0002819690430000048
And xi1And e1The width of the pipeline, coefficient vector W being penalty of relaxation factorA penalty vector;
by deriving the above model, the corresponding dual problem can be obtained:
Figure FDA0002819690430000049
Figure FDA00028196904300000410
wherein alpha is1＝[α11,....,α1n]T,
Figure FDA00028196904300000411
β1＝[β11,...,β1n]T,α2＝[α21,....,α2n]T,
Figure FDA00028196904300000412
And beta2＝[β21,...,β2n]TIs a lagrange multiplier vector;
step 2-3: based on wavelet transform theory, determining weight vector W ═ d of sample1,d2,···,dn]T，diRepresents a weight coefficient, and can be obtained by the following formula:
Figure FDA0002819690430000051
where exp (. cndot.) is an e exponential function, δ is the standard deviation of the Gaussian function, riIs the distance between the original data and the wavelet transformed denoised data;
step 2-4: parameter c of converter steelmaking endpoint information prediction model for initializing Step 2-2i(i＝1,...,4),v1,v2,ε1,ε2ε and δ;
step 2-5: utilize converter steelmaking training sample setTraining is carried out; by solving equations (11) and (12), an optimal solution vector α can be obtained1,
Figure FDA0002819690430000052
β1,α2,
Figure FDA0002819690430000053
And beta2；
Step 2-6: by substituting the optimal solution into the following equations (14) and (15), w can be obtained1,b1And w2,b2；
Figure FDA0002819690430000054
Figure FDA0002819690430000055
Wherein, | SkL represents the number of support vectors;
step 2-7: will w1,b1And w2,b2The result of (3) is substituted into the formula (16), and the regression function f of the end point carbon content can be obtainedC(x) Or a temperature regression function fT(x)；
Figure FDA0002819690430000056
Step 2-8: substituting training samples into the regression function fC(x) Or fT(x) Obtaining a predicted value of the end point carbon content or the end point temperature; calculating indexes such as model precision and end point hit rate, completing model establishment if the indexes reach set values, or updating the parameter c of the prediction model of the end point information of the converter steelmaking of Step 2-21,c2,c3,c4,v1,v2,ε1,ε2And e, repeating Step 2-5 to Step 2-7 until the index reaches a set value, and finishing model building.
4. The intelligent control method for the steelmaking endpoint of the converter as claimed in claim 1, wherein the method for calculating the oxygen blowing amount and the auxiliary materials of the oxygen blowing amount and auxiliary material calculating subsystem specifically comprises the following steps:
step 3-1: the objective function of the following optimization problem of the oxygen material addition amount is established as follows:
Figure FDA0002819690430000061
in the formula (f)fit(x)＝(fC/T(x)-DC/T)2Called fitness function, x is an input vector composed of variables such as oxygen blowing amount, lime addition amount and light-burned dolomite addition amount, and fC/T(x) Carbon temperature prediction for the prediction subsystem in Step2-8, DC/TA target value for end point carbon content or end point temperature;
step 3-2: setting the number j of whale groups and the maximum iteration number Nmax(ii) a Using the variables of oxygen blowing amount, lime addition amount and the like as optimization variables, carrying out normalization treatment on the variables, and mapping to [ 1-]Generating an equal amount of initial random solutions according to the set whale colony number j;
step 3-3: combining each group of initial solution variables with initial information of a furnace of molten iron to obtain x, respectively substituting into a prediction subsystem in Step2-8 to obtain a predicted value f of carbon content and temperatureC/T(x)；
Step 3-4: according to the fitness function f in equation (17)fit(x) Calculating the fitness of each group of solutions, and storing the optimal vector x with the minimum current fitness*；
Step 3-5: whale population optimization strategy using equation (18) if the current iteration number is less than NmaxThen a, r are updated1E, eta, k and p, determining a solution required by the next iteration, detecting whether a solution exceeding a search space exists, if so, mapping the solution to a random position in a feasible domain, and repeating the steps of Step3-4 and Step 3-5; otherwise, returning to the optimal solution to complete whale colony optimization of the oxygen material amount to obtainTo the optimal vector B of the oxygen blowing amount and the addition amount of auxiliary materials*＝invnorm(x*) Invnorm represents the inverse normalization process;
Figure FDA0002819690430000062
where t is the current iteration number, x (t) represents the position vector of the current whale, and λ ═ 2r1·xrand-x(t)|，xrandIs the random position of whale, r1Is [ -1,1 [ ]]Random vector within interval, η is a constant, E is coefficient vector, x*(t) represents the position vector of the current optimal solution, λzRepresenting the distance between whale and prey, a being a variable decreasing from 2 to 0, k being [ -1,1 [ -1 [ ]]Random vectors within the interval, z representing a probability variable;
step 3-6: by using the formula (17), an incremental calculation model is established, and the search interval of oxygen blowing amount is set to [ M ]1,N1]Step length of l1(ii) a The search interval for the amount of added adjuvant is set to [ M ]2,N2]Step length of l2；
Step 3-7: setting the initial values of oxygen blowing amount and auxiliary material addition amount in the same group of samples optimized by whale colony as M1,M2]Then combining the model input vector x with molten iron information to obtain a model input vector x, substituting the model input vector x into a prediction subsystem in Step2-8 to obtain a predicted value f of carbon content and temperatureC/T(x) (ii) a The fitness value is obtained according to the formula (17)
Figure FDA0002819690430000071
Saving vector C*＝x；
Step 3-8: according to the search interval l1And l2Gradually updating the values of the oxygen blowing amount and the auxiliary material addition amount to form a new model output variable xnewSubstituting the prediction subsystem in Step2-8 to obtain the predicted value f of carbon content and temperatureC/T(xnew) (ii) a The fitness value f is obtained according to the formula (17)fit(xnew) If, if
Figure FDA0002819690430000072
Then the vector C is saved*＝xnewAnd fitness value
Figure FDA0002819690430000073
When the oxygen blowing amount reaches N1The adding amount of the auxiliary materials reaches N2If so, executing Step 3-9, otherwise, repeating Step 3-8;
step 3-9: optimal solution B obtained from Step3-5 and Step3-8*And C*Extracting new vector composed of oxygen blowing amount and auxiliary material addition amount
Figure FDA0002819690430000074
And
Figure FDA0002819690430000075
substitution formula
Figure FDA0002819690430000076
Obtaining the optimal vector A of the oxygen blowing amount and the addition amount of the auxiliary materials, wherein K1And K2Is a weight coefficient;
step 3-10: by adjusting K1And K2The final oxygen blowing amount and the auxiliary material addition amount A are obtained*(ii) a And transmitting the final results of the oxygen blowing amount and the auxiliary material calculation subsystem to a database, reading data from the database by a PLC (programmable logic controller) system, determining a blowing mode and blowing stop time according to the optimal results, and finally sending a control instruction to a converter execution mechanism to complete the whole blowing process.
5. The intelligent control method for the steelmaking endpoint of the converter as claimed in claim 1, wherein the updating and upgrading method of the model updating subsystem specifically comprises the following steps:
step 4-1: when the newly added smelting data amount reaches L2Then, a population vector omega of the whale population is set0＝[Ω1,...,Ωl]TWherein Ω isi＝[c1,i,c2,i,c3,i,c4,i,ε1,i,ε2,i,εi,δi]TI 1.. times.l, a maximum number of iterations T is setmax(ii) a Defining a model parameter optimization fitness function as follows:
Figure FDA0002819690430000077
in the formula (I), the compound is shown in the specification,
Figure FDA0002819690430000078
size (U,1) represents vector x in matrix UiThe number of the (c) is,
Figure FDA0002819690430000079
is a predicted value of carbon content and temperature, y (x)i) Is the actual value of carbon content and temperature, τ is the upper error bound, L1The number of training samples determined in Step 2-1;
step 4-2: using equation (19), for each set of parameters
Figure FDA0002819690430000081
Calculating fitness
Figure FDA0002819690430000082
Obtaining the current optimal fitness f*(Ω*) And an optimum parameter omega*；
Step 4-3: if the current iteration number is less than TmaxUpdate omega*Obtaining the next set of parameters omegaiThe fitness f (Ω) is calculated using the formula (19)i)；
Step 4-4: if f is*(Ω*)≤f(Ωi) Then let Ω*＝ΩiAnd
Figure FDA0002819690430000083
returning to Step 4-3; otherwise, directly returning to Step 4-3;
step 4-5: if it is notThe number of current iterations is equal to TmaxThen the global optimum parameter omega is output*(ii) a And completing the updating of the model.
CN202011414366.5A 2020-12-07 2020-12-07 Intelligent control method for converter steelmaking end point Active CN112668234B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011414366.5A CN112668234B (en) 2020-12-07 2020-12-07 Intelligent control method for converter steelmaking end point

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011414366.5A CN112668234B (en) 2020-12-07 2020-12-07 Intelligent control method for converter steelmaking end point

Publications (2)

Publication Number Publication Date
CN112668234A true CN112668234A (en) 2021-04-16
CN112668234B CN112668234B (en) 2023-08-15

Family

ID=75401327

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011414366.5A Active CN112668234B (en) 2020-12-07 2020-12-07 Intelligent control method for converter steelmaking end point

Country Status (1)

Country Link
CN (1) CN112668234B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115058555A (en) * 2022-07-19 2022-09-16 北京博谦工程技术有限公司 Intelligent soft measurement method and system for measuring converter endpoint carbon content
CN115456264A (en) * 2022-09-01 2022-12-09 北京科技大学 Method for predicting end point carbon content and end point temperature of small and medium-sized converters
CN117851816A (en) * 2024-03-07 2024-04-09 北京科技大学 Method for predicting steelmaking end point component of converter

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090193936A1 (en) * 2008-02-01 2009-08-06 Honeywell International, Inc. Methods and apparatus for an oxygen furnace quality control system
CN102766723A (en) * 2012-06-29 2012-11-07 石家庄钢铁有限责任公司 Limestone slagging steelmaking lance position control method for basic oxygen furnaces
CN103778466A (en) * 2013-12-30 2014-05-07 浙江中控研究院有限公司 Converter steel-making temperature modeling forecasting method based on vector error and system thereof
CN106148637A (en) * 2015-04-10 2016-11-23 南京理工大学 The pneumatic steelmaking carbon content dynamic detection system of high stable
CN108676955A (en) * 2018-05-02 2018-10-19 中南大学 A kind of BOF Steelmaking Endpoint carbon content and temprature control method
CN109447346A (en) * 2018-10-26 2019-03-08 冶金自动化研究设计院 Based on gray prediction and neural network ensemble model converter oxygen consumption prediction technique
CN109886456A (en) * 2019-01-14 2019-06-14 西安工程大学 Debris flow prediction method based on PCA Yu mixed kernel function LSSVR
US20200133254A1 (en) * 2018-05-07 2020-04-30 Strong Force Iot Portfolio 2016, Llc Methods and systems for data collection, learning, and streaming of machine signals for part identification and operating characteristics determination using the industrial internet of things
CN111915080A (en) * 2020-07-30 2020-11-10 中冶华天工程技术有限公司 Raw fuel cost optimal proportioning method based on molten iron quality constraint

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090193936A1 (en) * 2008-02-01 2009-08-06 Honeywell International, Inc. Methods and apparatus for an oxygen furnace quality control system
CN102766723A (en) * 2012-06-29 2012-11-07 石家庄钢铁有限责任公司 Limestone slagging steelmaking lance position control method for basic oxygen furnaces
CN103778466A (en) * 2013-12-30 2014-05-07 浙江中控研究院有限公司 Converter steel-making temperature modeling forecasting method based on vector error and system thereof
CN106148637A (en) * 2015-04-10 2016-11-23 南京理工大学 The pneumatic steelmaking carbon content dynamic detection system of high stable
CN108676955A (en) * 2018-05-02 2018-10-19 中南大学 A kind of BOF Steelmaking Endpoint carbon content and temprature control method
US20200133254A1 (en) * 2018-05-07 2020-04-30 Strong Force Iot Portfolio 2016, Llc Methods and systems for data collection, learning, and streaming of machine signals for part identification and operating characteristics determination using the industrial internet of things
CN109447346A (en) * 2018-10-26 2019-03-08 冶金自动化研究设计院 Based on gray prediction and neural network ensemble model converter oxygen consumption prediction technique
CN109886456A (en) * 2019-01-14 2019-06-14 西安工程大学 Debris flow prediction method based on PCA Yu mixed kernel function LSSVR
CN111915080A (en) * 2020-07-30 2020-11-10 中冶华天工程技术有限公司 Raw fuel cost optimal proportioning method based on molten iron quality constraint

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
JOVAN PHULL等: "An Application of Decision Tree-Based Twin Support Vector Machines to Classify Dephosphorization in BOF Steelmaking", 《METALS》, pages 1 - 15 *
LIMING LIU等: "End-point prediction of 260 tons basic oxygen furnace (BOF) steelmaking based on WNPSVR and WOA", 《JOURNAL OF INTELLIGENT & FUZZY SYSTEMS》, pages 2923 - 2937 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115058555A (en) * 2022-07-19 2022-09-16 北京博谦工程技术有限公司 Intelligent soft measurement method and system for measuring converter endpoint carbon content
CN115456264A (en) * 2022-09-01 2022-12-09 北京科技大学 Method for predicting end point carbon content and end point temperature of small and medium-sized converters
CN115456264B (en) * 2022-09-01 2023-07-25 北京科技大学 Endpoint carbon content and endpoint temperature prediction method for small and medium-sized converter
CN117851816A (en) * 2024-03-07 2024-04-09 北京科技大学 Method for predicting steelmaking end point component of converter

Also Published As

Publication number Publication date
CN112668234B (en) 2023-08-15

Similar Documents

Publication Publication Date Title
CN112668234B (en) Intelligent control method for converter steelmaking end point
US20200118015A1 (en) Fuzzy curve analysis based soft sensor modeling method using time difference Gaussian process regression
CN105740984A (en) Product concept performance evaluation method based on performance prediction
CN108845501B (en) Blast furnace molten iron quality self-adaptive optimization control method based on lazy learning
CN109242188B (en) Long-term interval prediction and structure learning method for steel gas system
CN110598929B (en) Wind power nonparametric probability interval ultrashort term prediction method
CN107085371A (en) Crude(oil)unit economic model forecast control method based on data-driven
Fu et al. A building energy consumption prediction method based on integration of a deep neural network and transfer reinforcement learning
CN110245398B (en) Soft measurement deep learning method for thermal deformation of air preheater rotor
CN116307149A (en) Blast furnace performance optimization method based on attention LSTM and KBNSGA
Li et al. Dual ensemble online modeling for dynamic estimation of hot metal silicon content in blast furnace system
CN113204918B (en) Modeling method for cement kiln production energy consumption
CN110033126A (en) Shot and long term memory network prediction technique based on attention mechanism and logistic regression
CN111798023B (en) Comprehensive coke ratio prediction method in steelmaking sintering production
Liu et al. End-point prediction of 260 tons basic oxygen furnace (BOF) steelmaking based on WNPSVR and WOA
CN111639821A (en) Cement kiln production energy consumption prediction method and system
CN109101683B (en) Model updating method for pyrolysis kettle of coal quality-based utilization and clean pretreatment system
CN116738123A (en) Wellhead pressure control method based on improved longhorn beetle whisker algorithm optimization fuzzy PID
Wan et al. A modeling method of wide random forest multi-output soft sensor with attention mechanism for quality prediction of complex industrial processes
CN113223634B (en) Blast furnace molten iron silicon content prediction method based on two-dimensional self-attention enhanced GRU model
CN114943277A (en) Method for identifying operation condition and predicting operation performance of comprehensive energy system
CN113593657B (en) Cement free calcium soft measurement system with quality target as guiding semi-supervised learning
CN113723015A (en) Catalytic cracking unit optimization method based on mechanism model and big data technology
CN116127345B (en) Converter steelmaking process mode design method based on deep clustering generation countermeasure network
Wang A new variable selection method for soft sensor based on deep learning

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant