CN117894397B - Continuous casting mold flux viscosity forecasting method based on machine learning - Google Patents
Continuous casting mold flux viscosity forecasting method based on machine learning Download PDFInfo
- Publication number
- CN117894397B CN117894397B CN202410298431.4A CN202410298431A CN117894397B CN 117894397 B CN117894397 B CN 117894397B CN 202410298431 A CN202410298431 A CN 202410298431A CN 117894397 B CN117894397 B CN 117894397B
- Authority
- CN
- China
- Prior art keywords
- continuous casting
- mold flux
- model
- casting mold
- viscosity
- 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.)
- Active
Links
- 230000004907 flux Effects 0.000 title claims abstract description 97
- 238000009749 continuous casting Methods 0.000 title claims abstract description 92
- 238000010801 machine learning Methods 0.000 title claims abstract description 60
- 238000013277 forecasting method Methods 0.000 title claims abstract description 19
- 239000002893 slag Substances 0.000 claims abstract description 32
- 238000012549 training Methods 0.000 claims abstract description 21
- 238000000547 structure data Methods 0.000 claims abstract description 20
- -1 oxygen ions Chemical class 0.000 claims abstract description 16
- 229910052760 oxygen Inorganic materials 0.000 claims abstract description 14
- 239000001301 oxygen Substances 0.000 claims abstract description 14
- 238000012360 testing method Methods 0.000 claims abstract description 10
- 229910052731 fluorine Inorganic materials 0.000 claims abstract description 7
- 239000011737 fluorine Substances 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 33
- 238000013528 artificial neural network Methods 0.000 claims description 19
- 238000000034 method Methods 0.000 claims description 16
- 230000014509 gene expression Effects 0.000 claims description 15
- 239000013598 vector Substances 0.000 claims description 12
- 238000005457 optimization Methods 0.000 claims description 11
- 238000005266 casting Methods 0.000 claims description 8
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 claims description 6
- 238000003066 decision tree Methods 0.000 claims description 6
- 238000009826 distribution Methods 0.000 claims description 6
- 230000006872 improvement Effects 0.000 claims description 6
- 239000000843 powder Substances 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 238000004140 cleaning Methods 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 5
- 150000001768 cations Chemical class 0.000 claims description 4
- 239000003795 chemical substances by application Substances 0.000 claims description 4
- 150000002500 ions Chemical class 0.000 claims description 4
- 239000000203 mixture Substances 0.000 claims description 4
- 229910011763 Li2 O Inorganic materials 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 3
- PNEYBMLMFCGWSK-UHFFFAOYSA-N aluminium oxide Inorganic materials [O-2].[O-2].[O-2].[Al+3].[Al+3] PNEYBMLMFCGWSK-UHFFFAOYSA-N 0.000 claims description 3
- 229910052593 corundum Inorganic materials 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 229910001845 yogo sapphire Inorganic materials 0.000 claims description 3
- VTLYFUHAOXGGBS-UHFFFAOYSA-N Fe3+ Chemical compound [Fe+3] VTLYFUHAOXGGBS-UHFFFAOYSA-N 0.000 claims description 2
- KRHYYFGTRYWZRS-UHFFFAOYSA-M Fluoride anion Chemical compound [F-] KRHYYFGTRYWZRS-UHFFFAOYSA-M 0.000 claims description 2
- KKCBUQHMOMHUOY-UHFFFAOYSA-N Na2O Inorganic materials [O-2].[Na+].[Na+] KKCBUQHMOMHUOY-UHFFFAOYSA-N 0.000 claims description 2
- 230000001186 cumulative effect Effects 0.000 claims description 2
- 238000005315 distribution function Methods 0.000 claims description 2
- 238000011156 evaluation Methods 0.000 claims description 2
- JEIPFZHSYJVQDO-UHFFFAOYSA-N iron(III) oxide Inorganic materials O=[Fe]O[Fe]=O JEIPFZHSYJVQDO-UHFFFAOYSA-N 0.000 claims description 2
- 229910052751 metal Inorganic materials 0.000 abstract description 3
- 239000002184 metal Substances 0.000 abstract description 3
- 229910000831 Steel Inorganic materials 0.000 description 7
- 239000010959 steel Substances 0.000 description 7
- 239000000463 material Substances 0.000 description 6
- 230000000694 effects Effects 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 2
- BASFCYQUMIYNBI-UHFFFAOYSA-N platinum Chemical compound [Pt] BASFCYQUMIYNBI-UHFFFAOYSA-N 0.000 description 2
- ZOKXTWBITQBERF-UHFFFAOYSA-N Molybdenum Chemical compound [Mo] ZOKXTWBITQBERF-UHFFFAOYSA-N 0.000 description 1
- 229910004742 Na2 O Inorganic materials 0.000 description 1
- 229910004298 SiO 2 Inorganic materials 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000012535 impurity Substances 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 230000001050 lubricating effect Effects 0.000 description 1
- 238000005461 lubrication Methods 0.000 description 1
- 239000011733 molybdenum Substances 0.000 description 1
- 229910052750 molybdenum Inorganic materials 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003647 oxidation Effects 0.000 description 1
- 238000007254 oxidation reaction Methods 0.000 description 1
- 229910052697 platinum Inorganic materials 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
- G16C20/30—Prediction of properties of chemical compounds, compositions or mixtures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
- G16C20/60—In silico combinatorial chemistry
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Chemical & Material Sciences (AREA)
- Computing Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Crystallography & Structural Chemistry (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- General Health & Medical Sciences (AREA)
- Medicinal Chemistry (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Continuous Casting (AREA)
Abstract
The invention belongs to the technical field of metal continuous casting, in particular to a continuous casting mold flux viscosity forecasting method based on machine learning, which collects continuous casting mold flux viscosity data of known components at different temperatures; converting the component data of the continuous casting mold flux into slag structure data based on the characteristics of oxygen ions and fluorine ions, taking the slag structure data and the temperature as model input characteristic values, and taking the viscosity of the mold flux as model output characteristic values; normalizing slag structure data, temperature and viscosity data of the continuous casting mold flux, and dividing the data into a training set and a testing set; the composition-structure-viscosity-based stacking integrated machine learning model is constructed, and a Bayesian algorithm is used for optimizing the machine learning algorithm in the stacking integrated machine learning model, so that the optimized continuous casting mold flux viscosity prediction model is used for predicting the viscosity of the continuous casting mold flux at different temperatures, and the viscosity of the continuous casting mold flux can be accurately predicted.
Description
Technical Field
The invention relates to the technical field of metal continuous casting, in particular to a continuous casting mold flux viscosity forecasting method based on machine learning.
Background
The continuous casting process can continuously produce metal blanks, can more effectively utilize raw materials and energy sources while improving the production efficiency, and is an important component of the current steel manufacturing flow. The continuous casting covering slag provides enough lubrication for continuous casting of steel in a continuous casting crystallizer, absorbs impurities in molten steel, controls heat conduction in the continuous casting crystallizer and prevents oxidation of the molten steel, and is an indispensable auxiliary material for continuous casting. The physical and chemical properties of the continuous casting mold flux directly affect the smooth running of continuous casting and the quality of casting blanks, and the continuous casting mold flux performance meeting the continuous casting production requirements is one of key elements for producing high-quality steel materials. The change of factors such as different steel grades, different drawing speeds and sections of the same steel grade all needs the covering slag with different performances to be matched with the covering slag. The viscosity of the mold flux is one of key performances in the design and production of the mold flux, the lubricating performance is reduced by using the mold flux with high viscosity, the tendency of slag coiling is increased by using the mold flux with low viscosity, so that a liquid slag film between a casting blank and a crystallizer wall cannot be uniformly distributed, and the quality of the casting blank and the smooth running of continuous casting cannot be effectively ensured. Therefore, knowledge of the viscosity of the mold flux is an indispensable step in the development and application of the continuous casting mold flux.
Usually, the viscosity of the casting powder is measured at high temperature by adopting a rotary viscometer, one day is usually required for measuring one viscosity, and a crucible and a rotor which are made of platinum materials or molybdenum under the protection of inert atmosphere are required for reducing errors, so that the cost of manpower and material resources is high, and a plurality of models for forecasting the viscosity of the continuous casting powder are generated. The existing prediction model of the viscosity of the continuous casting mold flux is mainly an empirical model and a semi-empirical model, and although a few models begin to consider a simple melt structure, parameters of the model are still empirical parameters, so that the reliability and the universality are poor. The crystallizer casting powder mainly comprises CaO, siO2, mgO, al2O3, na2O, mnO, li2O, fe O3, F and other components, and the components are complex; and because the viscosity is influenced by a plurality of factors and is related to the content and the structure, the complex relation between the components and the viscosity cannot be directly and accurately described by adopting a mathematical formula, so that the problems of poor generalization capability, large prediction error and the like of the existing model exist, and accurate guidance cannot be provided for the research and development design of the continuous casting mold flux.
Disclosure of Invention
In order to solve the problems in the prior art, the main purpose of the invention is to provide a continuous casting mold flux viscosity forecasting method based on machine learning, which can forecast the viscosity of mold flux more accurately.
In order to solve the technical problems, according to one aspect of the present invention, the following technical solutions are provided:
a continuous casting mold flux viscosity forecasting method based on machine learning comprises the following steps:
s1, collecting viscosity data of continuous casting mold flux of known components at different temperatures, and cleaning and preprocessing the data;
S2, according to the relation between the components of the continuous casting mold flux and the slag structure, converting the component data of the continuous casting mold flux into slag structure data based on the characteristics of oxygen ions and fluorine ions, wherein the slag structure data and the temperature are used as model input characteristic values together, and the mold flux viscosity is used as model output characteristic values;
S3, normalizing slag structure data, temperature and viscosity data of the continuous casting mold flux, and dividing the data into a training set and a testing set;
s4, constructing a stacking integrated machine learning model based on composition-structure-viscosity, and optimizing a machine learning algorithm in the stacking integrated machine learning model by using a Bayesian algorithm to obtain an optimized continuous casting mold flux viscosity prediction model;
s5, forecasting the viscosity of the continuous casting mold flux at different temperatures by using an optimized continuous casting mold flux viscosity forecasting model.
As a preferable scheme of the continuous casting mold flux viscosity forecasting method based on machine learning, the invention comprises the following steps: in the step S1, preprocessing the data includes cleaning the data to remove errors.
As a preferable scheme of the continuous casting mold flux viscosity forecasting method based on machine learning, the invention comprises the following steps: in the step S2, the composition data of the continuous casting mold flux is converted into slag structure data based on the characteristics of oxygen ions and fluorine ions according to the relationship between the composition of the continuous casting mold flux and the slag structure, specifically as expressed in the expressions (1) to (14),
（1）
（2）
（3）
（4）
（5）
（6）
（7）
（8）
（9）
（10）
（11）
（12）
（13）
（14）
Wherein x i is the mole fraction of the mold flux component i; i comprises CaO, siO 2、Al2O3、Na2O、F、MgO、MnO、Fe2O3、Li2 O; the amount of non-bridging oxygen for the cation j to be connected with Si in the mold flux component; /(I) The amount of non-bridging oxygen connected with cations j and Al in the casting powder component after the k ions participate in charge compensation; /(I)The number of bridging oxygens connected with Al after the k ions participate in charge compensation; /(I)Is the number of bridging oxygens attached to Si; /(I)Is the amount of bridging oxygen to which Fe (III) is attached; /(I)Is the number of fluoride ions.
As a preferable scheme of the continuous casting mold flux viscosity forecasting method based on machine learning, the invention comprises the following steps: in the step S3, the slag structure data, the temperature and the viscosity data of the continuous casting mold flux are normalized so as to eliminate the difference between different orders of magnitude, and 80-85% of the data are selected as training set data, and 15-20% of the data are selected as test set data.
As a preferable scheme of the continuous casting mold flux viscosity forecasting method based on machine learning, the invention comprises the following steps: in the step S3, the normalization processing is performed by adopting an expression (15),
（15）
Wherein, X is any data value in the dataset, min and Max are the minimum and maximum data values in the dataset, respectively, and X N is the normalized value of X.
As a preferable scheme of the continuous casting mold flux viscosity forecasting method based on machine learning, the invention comprises the following steps: in the step S4, a two-layer stacked integrated machine learning model is constructed, wherein the first layer is a base learner and comprises an artificial neural network model, a gradient lifting tree model and a support vector regression model, and the second layer uses a decision tree model as a meta learner; the learners used in the stacked model all employ supervised machine learning regression algorithms.
As a preferable scheme of the continuous casting mold flux viscosity forecasting method based on machine learning, the invention comprises the following steps: in the step S4, when the stacking integrated machine learning training is performed, the prediction results of the viscosity of the continuous casting mold flux are obtained through different base learners, then the results of the plurality of base learners are used as the input of the meta learner, and the output of the meta learner is the final prediction result of the viscosity of the continuous casting mold flux.
As a preferable scheme of the continuous casting mold flux viscosity forecasting method based on machine learning, the invention comprises the following steps: in the step S4, a bayesian algorithm is used to optimize a machine learning algorithm in the stacked integrated machine learning model, specifically:
S41, selecting a Gaussian process as a proxy model optimization algorithm by defining a random function So that for any input x the corresponding output/>Is a random variable whose joint distribution is subject to a multivariate gaussian distribution, the expression is as in formula (16),
（16）
Wherein,For/>Mean function of/>Represents the covariance of x;
S42, using the expected lifting function as a collection function in Bayesian optimization, wherein the expressions are shown as formulas (17) - (18),
（17）
（18）
Wherein,To boost the collection function, x is the input quantity of the sample point to be evaluated, μ is the average of the observation points,/>Is the existing maximum value,/>A normal cumulative distribution function; z represents a normalized improvement amount for measuring the improvement of the function value at the observation point (x) with respect to the currently known optimal function value; sigma is the standard deviation of the observation point,/>Representing a normal distribution probability density function.
S43, randomly obtaining a group of hyper-parameter vectors x when using Bayes to optimize a given hyper-parameter space, substituting the hyper-parameter vectors x into a machine learning algorithm, and obtaining a next evaluation point by maximizing a collection function based on a Gaussian process agent modelThe expressions are shown as formulas (19) - (20),
（19）
（20）
Wherein,Representing selection of the point with the greatest expected improvement,/>For the interval in which the input x exists,Is the posterior mean value of Gaussian agent model in the step t+1,/>For the model to be optimized,/>In order to input the training data,And/>Characteristic values and label values of training data;
S44, evaluating the model, integrating the data and updating the proxy model;
S45, optimizing the machine learning model by using an iterative Bayesian algorithm until the error of the model reaches the requirement.
As a preferable scheme of the continuous casting mold flux viscosity forecasting method based on machine learning, the invention comprises the following steps: in the step S4, the super parameters used for the bayesian algorithm to optimize the artificial neural network model include a learning rate, an activation function of the neural network, a structural parameter of the neural network, an initialization of a weight, a regularization parameter, a training batch size and a selection and a parameter of a neural network optimizer; super parameters for the Bayesian algorithm optimized gradient lifting tree model include learning rate, number of learners, depth of tree, minimum split sample number of tree, minimum leaf node sample number of tree, sub-sample ratio, column sampling ratio and regularization parameters; the super parameters for the bayesian algorithm optimized support vector regression model include the kernel function of the model, the penalty function, the parameters of the kernel function, the tolerance of the penalty function, the sample weights, and the buffer size.
As a preferable scheme of the continuous casting mold flux viscosity forecasting method based on machine learning, the invention comprises the following steps: in the step S4, the super parameters used for the bayesian algorithm to optimize the decision tree model are the depth of the tree, the minimum number of split samples, the minimum number of leaf node samples, the maximum feature number during splitting, and the minimum amount of reduction of the unrepeace of node splitting.
The beneficial effects of the invention are as follows:
The invention provides a continuous casting mold flux viscosity forecasting method based on machine learning, which is used for collecting continuous casting mold flux viscosity data of known components at different temperatures; converting the component data of the continuous casting mold flux into slag structure data based on the characteristics of oxygen ions and fluorine ions, wherein the slag structure data and the temperature are used as model input characteristic values together, and the viscosity of the mold flux is used as model output characteristic values; normalizing slag structure data, temperature and viscosity data of the continuous casting mold flux, and dividing the data into a training set and a testing set; the method comprises the steps of constructing a stacking integrated machine learning model based on components, structures and viscosity, optimizing the machine learning algorithm in the stacking integrated machine learning model by using a Bayesian algorithm, obtaining an optimized continuous casting mold flux viscosity prediction model, predicting the viscosity of the continuous casting mold flux at different temperatures, accurately predicting the viscosity of the continuous casting mold flux at different temperatures and different components, greatly reducing the experiment quantity, saving a large amount of manpower and material resources, and having good guiding significance for the design and development of the continuous casting mold flux.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to the results shown in the drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic flow chart of the present invention.
FIG. 2 is a flow chart of the Bayesian optimization algorithm of the present invention.
FIG. 3 is a graph showing the effect of the number of neurons in different hidden layers on the model error of the artificial neural network according to the present invention.
FIG. 4 is a graph showing the effect of the number of learners on the model error of the gradient lift tree according to the present invention.
FIG. 5 is a graph comparing the predicted and measured results of the test set data of the present invention.
The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.
Detailed Description
The following description will be made clearly and fully with reference to the technical solutions in the embodiments, and it is apparent that the described embodiments are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Compared with a method for measuring the viscosity of the continuous casting mold flux through a large number of experiments, the continuous casting mold flux viscosity forecasting method based on machine learning is simple and quick, saves time, labor and material resource cost, and has smaller deviation from experimental results; compared with the prior experience and semi-experience model for forecasting the viscosity of the continuous casting mold flux, the viscosity forecasting effect based on the machine learning method is more accurate, has stronger generalization capability, is suitable for forecasting the viscosity of more different continuous casting mold fluxes, and can meet the viscosity forecasting requirements under the variety of the prior mold flux development.
As shown in fig. 1, the invention provides a continuous casting mold flux viscosity forecasting method based on machine learning, which comprises the following steps:
s1, collecting viscosity data of continuous casting mold flux of known components at different temperatures, and cleaning and preprocessing the data;
S2, according to the relation between the components of the continuous casting mold flux and the slag structure, converting the component data of the continuous casting mold flux into slag structure data with obvious physicochemical significance and based on the characteristics of oxygen ions and fluorine ions, wherein the slag structure data and the temperature are used as model input characteristic values together, and the viscosity of the mold flux is used as model output characteristic values;
S3, normalizing slag structure data, temperature and viscosity data of the continuous casting mold flux, and dividing the data into a training set and a testing set;
s4, constructing a stacking integrated machine learning model based on composition-structure-viscosity, and optimizing a machine learning algorithm in the stacking integrated machine learning model by using a Bayesian algorithm to obtain an optimized continuous casting mold flux viscosity prediction model;
s5, forecasting the viscosity of the continuous casting mold flux at different temperatures by using an optimized continuous casting mold flux viscosity forecasting model.
The collection and processing of the viscosity data of the casting powder before the framework stacking and integrating the learning model is indispensable. The covering slag component comprises CaO and SiO 2、Al2O3、Na2 O, F, and may also comprise one or more of MgO, mnO, fe 2O3、Li2 O. After the viscosity data of all continuous casting mold flux is obtained, abnormal data in the viscosity data are cleaned and removed, the results of multiple tests on the same slag system are averaged, the component data of the mold flux are converted into slag structure data based on oxygen ion and fluoride ion characteristics through expressions (1) - (14) and are used as input characteristics of a model together with temperature, and in order to avoid the problems of unstable subsequent modeling and too slow model convergence speed caused by overlarge fluctuation range of a data set, normalization processing is carried out on the data used for modeling through the expression (15).
The step S4 specifically includes the following:
(1) Two-layer stacked integrated learning model
Constructing a two-layer stacked integrated machine learning model, wherein the first layer is a base learner and comprises an artificial neural network model, a gradient lifting tree model and a support vector regression model, and the second layer uses a decision tree model as a meta learner; the learners used in the stacked model all employ supervised machine learning regression algorithms.
(2) Training and optimization of base learner
And writing Python codes to construct an artificial neural network model, a gradient lifting tree model and a support vector regression model, importing the preprocessed 1200 sets of training set data into a base learner, training the model and optimizing the base learner based on expressions (16) - (20) by using a Bayesian optimization algorithm as shown in fig. 2.
The super parameters for the Bayesian algorithm optimization artificial neural network model comprise learning rate, activation function of the neural network, structural parameters of the neural network, initialization of weights, regularization parameters, training batch size, selection and parameters of a neural network optimizer and the like; super parameters for the Bayesian algorithm optimized gradient lifting tree model comprise learning rate, number of learners, depth of tree, minimum split sample number of tree, minimum leaf node sample number of tree, sub-sample proportion, column sampling proportion, regularization parameter and the like; the super parameters used for the Bayesian algorithm optimization support vector regression model comprise kernel functions of the model, penalty functions, parameters of the kernel functions, tolerance degree of the loss functions, sample weights, buffer sizes and the like. The model is optimized until the error meets the requirements. The number of hidden layer neurons of the neural network model and the number of learners of the gradient lifting tree are core parameters of the model, the influence of different hidden layer neurons on the model error of the artificial neural network is shown in fig. 3, and the influence of different learners on the model error of the gradient lifting tree is shown in fig. 4. And selecting root mean square error to evaluate the prediction reliability of the model, wherein the expression is shown as the formula (21).
（21）
Where n is the number of samples,For the forecast value,/>Is the actual measurement value.
(3) Training and optimization of meta learner
And forming a new database by using the forecast result of the base learner as a new input characteristic training element learner, wherein the element learner selects a decision tree model, and super parameters for Bayesian algorithm optimization are the depth of the tree, the minimum splitting sample number, the minimum leaf node sample number, the maximum characteristic number during splitting, the minimum unrepeace reduction of node splitting and the like. The model is optimized until the error meets the requirements.
And inputting 283 groups of test data into the trained model, and evaluating the forecasting effect of the model by adopting the decision coefficient shown in the expression (22).
（22）
Where n is the number of samples,For the forecast value,/>For the measured value,/>Is the average of the measured values.
The decision coefficients of the artificial neural network model, the gradient lifting tree model and the support vector regression model are respectively 0.907, 0.921 and 0.897, and the decision coefficient of the stacked integrated learning model is 0.971, so that the prediction reliability is highest. The prediction result and the actual measurement result of the stacked integrated machine learning model are shown in fig. 5, and the comparison shows that the prediction value and the actual measurement value are more consistent, so that the viscosity of the continuous casting mold flux can be predicted by using the stacked integrated machine learning model, and the method has higher accuracy.
The foregoing description is only of the preferred embodiments of the present invention and is not intended to limit the scope of the invention, and all equivalent structural changes made by the content of the present invention or direct/indirect application in other related technical fields are included in the scope of the present invention.
Claims (8)
1. The continuous casting mold flux viscosity forecasting method based on machine learning is characterized by comprising the following steps:
s1, collecting viscosity data of continuous casting mold flux of known components at different temperatures, and cleaning and preprocessing the data;
S2, according to the relation between the components of the continuous casting mold flux and the slag structure, converting the component data of the continuous casting mold flux into slag structure data based on the characteristics of oxygen ions and fluorine ions, wherein the slag structure data and the temperature are used as model input characteristic values together, and the mold flux viscosity is used as model output characteristic values;
S3, normalizing slag structure data, temperature and viscosity data of the continuous casting mold flux, and dividing the data into a training set and a testing set;
S4, constructing a stacking integrated machine learning model based on composition-structure-viscosity, and optimizing a machine learning algorithm in the stacking integrated machine learning model by using a Bayesian algorithm to obtain an optimized continuous casting mold flux viscosity prediction model; when stacking integrated machine learning training is carried out, firstly, obtaining a prediction result of continuous casting mold flux viscosity through different base learners, and then taking the results of a plurality of base learners as input of a meta-learner, wherein the output of the meta-learner is a final prediction result of continuous casting mold flux viscosity; the machine learning algorithm in the stacked integrated machine learning model is optimized by using a Bayes algorithm, and specifically comprises the following steps:
S41, selecting a Gaussian process as a proxy model optimization algorithm by defining a random function So that for any input x the corresponding output/>Is a random variable whose joint distribution is subject to a multivariate gaussian distribution, the expression is as in formula (16),
（16）
Wherein,For/>Mean function of/>Represents the covariance of x;
S42, using the expected lifting function as a collection function in Bayesian optimization, wherein the expressions are shown as formulas (17) - (18),
（17）
（18）
Wherein,For the desired lifting of the collection function, x is the input quantity for the sample point to be evaluated, μ is the mean of the observation points,Is the existing maximum value,/>A normal cumulative distribution function; z represents a normalized improvement amount for measuring the improvement of the function value at the observation point (x) with respect to the currently known optimal function value; sigma is the standard deviation of the observation point,/>Representing a normal distribution probability density function;
S43, randomly obtaining a group of hyper-parameter vectors x when using Bayes to optimize a given hyper-parameter space, substituting the hyper-parameter vectors x into a machine learning algorithm, and obtaining a next evaluation point by maximizing a collection function based on a Gaussian process agent model The expressions are shown as formulas (19) - (20),
（19）
（20）
Wherein,Representing selection of the point with the greatest expected improvement,/>For the interval where the input x exists,/>Is the posterior mean value of Gaussian agent model in the step t+1,/>For the model to be optimized,/>For input training data,/>And/>Characteristic values and label values of training data;
S44, evaluating the model, integrating the data and updating the proxy model;
S45, optimizing a machine learning model by using an iterative Bayesian algorithm until the error of the model reaches the requirement;
s5, forecasting the viscosity of the continuous casting mold flux at different temperatures by using an optimized continuous casting mold flux viscosity forecasting model.
2. The machine learning-based continuous casting mold flux viscosity prediction method according to claim 1, wherein the preprocessing of the data in step S1 includes cleaning erroneous data.
3. The machine learning-based continuous casting mold flux viscosity prediction method according to claim 1, wherein in the step S2, the composition data of the continuous casting mold flux is converted into slag structure data based on oxygen ion and fluoride ion characteristics according to the relationship between the composition of the continuous casting mold flux and the slag structure, specifically as expressed in expressions (1) - (14),
（1）
（2）
（3）
（4）
（5）
（6）
（7）
（8）
（9）
（10）
（11）
（12）
（13）
（14）
Wherein x i is the mole fraction of the mold flux component i; i comprises CaO, siO 2、Al2O3、Na2O、F、MgO、MnO、Fe2O3、Li2 O; the amount of non-bridging oxygen for the cation j to be connected with Si in the mold flux component; /(I) The amount of non-bridging oxygen connected with cations j and Al in the casting powder component after the k ions participate in charge compensation; /(I)The number of bridging oxygens connected with Al after the k ions participate in charge compensation; /(I)Is the number of bridging oxygens attached to Si; /(I)Is the amount of bridging oxygen to which Fe (III) is attached; /(I)Is the number of fluoride ions.
4. The machine learning-based continuous casting mold flux viscosity prediction method according to claim 1, wherein in the step S3, the slag structure data, the temperature and the viscosity data of the continuous casting mold flux are normalized, 80-85% of the data are selected as training set data, and 15-20% of the data are selected as test set data.
5. The machine learning-based continuous casting mold flux viscosity prediction method according to claim 4, wherein in the step S3, the normalization process is performed using expression (15),
（15）
Wherein, X is any data value in the dataset, min and Max are the minimum and maximum data values in the dataset, respectively, and X N is the normalized value of X.
6. The machine learning-based continuous casting mold flux viscosity prediction method according to claim 1, wherein in the step S4, a two-layer stacked integrated machine learning model is constructed, the first layer is a base learner including an artificial neural network model, a gradient lifting tree model and a support vector regression model, and the second layer uses a decision tree model as a meta learner; the learners used in the stacked model all employ supervised machine learning regression algorithms.
7. The machine learning-based continuous casting mold flux viscosity prediction method according to claim 1, wherein in the step S4, the super parameters for optimizing the artificial neural network model by using the bayesian algorithm include learning rate, activation function of the neural network, structural parameters of the neural network, initialization of weights, regularization parameters, batch size of training, and selection and parameters of the neural network optimizer; super parameters for the Bayesian algorithm optimized gradient lifting tree model include learning rate, number of learners, depth of tree, minimum split sample number of tree, minimum leaf node sample number of tree, sub-sample ratio, column sampling ratio and regularization parameters; the super parameters for the bayesian algorithm optimized support vector regression model include the kernel function of the model, the penalty function, the parameters of the kernel function, the tolerance of the penalty function, the sample weights, and the buffer size.
8. The machine learning-based continuous casting mold flux viscosity prediction method according to claim 1, wherein in the step S4, the super parameters used for optimizing the decision tree model by the bayesian algorithm are the depth of the tree, the minimum number of split samples, the minimum number of leaf node samples, the maximum feature number at the split time, and the minimum amount of unrepeace reduction of node split.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410298431.4A CN117894397B (en) | 2024-03-15 | 2024-03-15 | Continuous casting mold flux viscosity forecasting method based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410298431.4A CN117894397B (en) | 2024-03-15 | 2024-03-15 | Continuous casting mold flux viscosity forecasting method based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117894397A CN117894397A (en) | 2024-04-16 |
| CN117894397B true CN117894397B (en) | 2024-05-28 |
Family
ID=90652182
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410298431.4A Active CN117894397B (en) | 2024-03-15 | 2024-03-15 | Continuous casting mold flux viscosity forecasting method based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117894397B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021007812A1 (en) * | 2019-07-17 | 2021-01-21 | 深圳大学 | Deep neural network hyperparameter optimization method, electronic device and storage medium |
| CN114925620A (en) * | 2022-06-21 | 2022-08-19 | 国网福建省电力有限公司宁德供电公司 | Short-term wind power prediction method and system based on ensemble learning algorithm |
| CN116434893A (en) * | 2023-06-12 | 2023-07-14 | 中才邦业(杭州)智能技术有限公司 | Concrete compressive strength prediction model, construction method, medium and electronic equipment |
| CN116451322A (en) * | 2023-04-11 | 2023-07-18 | 哈尔滨工业大学 | Bayesian optimization-based LSTM deep learning network mechanical prediction method |
| CN117455320A (en) * | 2023-12-25 | 2024-01-26 | 江苏恒力化纤股份有限公司 | QGMVPN model-based multi-kettle polymerization process melt quality index online prediction method |
-
2024
- 2024-03-15 CN CN202410298431.4A patent/CN117894397B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021007812A1 (en) * | 2019-07-17 | 2021-01-21 | 深圳大学 | Deep neural network hyperparameter optimization method, electronic device and storage medium |
| CN114925620A (en) * | 2022-06-21 | 2022-08-19 | 国网福建省电力有限公司宁德供电公司 | Short-term wind power prediction method and system based on ensemble learning algorithm |
| CN116451322A (en) * | 2023-04-11 | 2023-07-18 | 哈尔滨工业大学 | Bayesian optimization-based LSTM deep learning network mechanical prediction method |
| CN116434893A (en) * | 2023-06-12 | 2023-07-14 | 中才邦业(杭州)智能技术有限公司 | Concrete compressive strength prediction model, construction method, medium and electronic equipment |
| CN117455320A (en) * | 2023-12-25 | 2024-01-26 | 江苏恒力化纤股份有限公司 | QGMVPN model-based multi-kettle polymerization process melt quality index online prediction method |
Non-Patent Citations (1)
| Title |
|---|
| 基于卷积神经网络的热轧带钢力学性能预报;胡石雄;李维刚;杨威;;武汉科技大学学报;20180929(第05期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117894397A (en) | 2024-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110414788B (en) | Electric energy quality prediction method based on similar days and improved LSTM | |
| CN114611844B (en) | Method and system for determining alloy addition amount in converter tapping process | |
| CN110929347A (en) | Hot continuous rolling strip steel convexity prediction method based on gradient lifting tree model | |
| CN107358318A (en) | Based on GM（1,1）The urban power consumption Forecasting Methodology of model and Grey Markov chain predicting model | |
| CN111444942B (en) | Intelligent forecasting method and system for silicon content of blast furnace molten iron | |
| CN109523077B (en) | Wind power prediction method | |
| CN115293366B (en) | Model training method, information prediction method, device, equipment and medium | |
| CN114239400A (en) | Multi-working-condition process self-adaptive soft measurement modeling method based on local double-weighted probability hidden variable regression model | |
| CN112053019B (en) | Method for realizing intellectualization of optical fiber preform deposition process | |
| CN110717281B (en) | Simulation model credibility evaluation method based on hesitation cloud language term set and cluster decision | |
| CN117139380A (en) | Camber control method based on self-learning of regulation experience | |
| Zhao et al. | Prediction of mechanical properties of cold rolled strip based on improved extreme random tree | |
| CN117894397B (en) | Continuous casting mold flux viscosity forecasting method based on machine learning | |
| US20240002964A1 (en) | Method and system for determining converter tapping quantity | |
| CN113408192A (en) | Intelligent electric meter error prediction method based on GA-FSVR | |
| Liu et al. | XGBoost-based model for predicting hydrogen content in electroslag remelting | |
| CN108665090B (en) | Urban power grid saturation load prediction method based on principal component analysis and Verhulst model | |
| CN114282658B (en) | Method, device and medium for analyzing and predicting flow sequence | |
| CN110648023A (en) | Method for establishing data prediction model based on quadratic exponential smoothing improved GM (1,1) | |
| CN105069214A (en) | Process reliability evaluation method based on nonlinear correlation analysis | |
| CN114655074A (en) | Electric automobile actual driving energy consumption estimation method based on Bayesian regression | |
| CN116324323A (en) | Method for generating learned prediction model for predicting energy efficiency of melting furnace, method for predicting energy efficiency of melting furnace, and computer program | |
| CN106447065A (en) | Method for predicting coagulation bath link performance index in carbon fiber precursor production process | |
| CN107977742B (en) | Construction method of medium-long term power load prediction model | |
| CN117497087B (en) | Oxide glass performance prediction method based on interpretable high-dimensional spatial prediction model |
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 |