NL2030465B1 - Continuous casting slab quality predicting method based on extreme learning machine - Google Patents

Continuous casting slab quality predicting method based on extreme learning machine Download PDF

Info

Publication number
NL2030465B1
NL2030465B1 NL2030465A NL2030465A NL2030465B1 NL 2030465 B1 NL2030465 B1 NL 2030465B1 NL 2030465 A NL2030465 A NL 2030465A NL 2030465 A NL2030465 A NL 2030465A NL 2030465 B1 NL2030465 B1 NL 2030465B1
Authority
NL
Netherlands
Prior art keywords
continuous casting
extreme learning
learning machine
data
quality
Prior art date
Application number
NL2030465A
Other languages
Dutch (nl)
Other versions
NL2030465A (en
Inventor
Yang Jianping
Liu Qing
Chen Hengzhi
Guan Min
Original Assignee
Univ Beijing Science & Technology
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 Univ Beijing Science & Technology filed Critical Univ Beijing Science & Technology
Priority to NL2030465A priority Critical patent/NL2030465B1/en
Publication of NL2030465A publication Critical patent/NL2030465A/en
Application granted granted Critical
Publication of NL2030465B1 publication Critical patent/NL2030465B1/en

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/10—Numerical modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00—Details relating to the application field
    • G06F2113/22—Moulding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Continuous Casting (AREA)

Abstract

The present disclosure provides a continuous casting slab guality predicting method based on extreme learning machine, which belongs to a technical field, of iron and steel metallurgy. The method first collects the data involved in an actual production of continuous casting slab, finds influencing factors affecting the continuous casting slab guality, then. preprocesses the selected sample data, determines a number of input nodes and output nodes of the extreme learning machine according to these data, and then inputs the training data set into the extreme learning machine to complete a training of the extreme learning' machine, finally, inputs the remaining sample data to complete the defect grade classification of the continuous casting slab quality. The method has the advantages of fast training speed, high prediction accuracy and good adaptability.

Description

P2992 /NLpd
CONTINUOUS CASTING SLAB QUALITY PREDICTING METHOD BASED ON EXTREME LEARNING MACHINE
TECHNICAL FIELD The present disclosure relates to a technical field of iron and steel metallurgy, in particular to a continuous casting slab quality predicting method based on extreme learning machine.
BACKGROUND ART In recent decades, with the rapid development of iron and steel enterprises, continuous casting has become the most active research field of iron and steel enterprises due to the hot charg- ing and hot delivery of continuous casting slab and direct rolling technology. The development of this technology not only greatly reduces the production cost and equipment investment of iron and steel, but also improves the competitiveness of products. However, the current continuous casting technology can not completely elim- inate the generation of defects, resulting in defects will inevi- tably occur in the subsequent hot charging, hot delivery and di- rect rolling process, which will affect the quality of steel prod- ucts. If the continuous casting slabs with these defects can be Judged and sorted out in time, it can not only improve the quality of continuous casting slabs/products, but also be conducive to the continuous production of continuous casting-continuous rolling process. How to predict and detect slab quality timely and accu- rately is an important problem to be solved in the process of sus- tainable development of iron and steel enterprises. At present, methods of continuous casting slab quality pre- diction mainly include statistical methods and non-statistical methods. The statistical methods include linear regression, non- linear regression, etc. The non-statistical methods include expert system, BP neural network, etc. In a process of continuous casting slab production, molten steel undergoes a series of complex physi- cal and chemical changes such as solidification, crystallization and phase transformation, which involves many production equipment and parameters, and each influencing factor has a strong nonlinear relationship with continuous casting slab defects. The adaptabil- ity and generalization ability of the continuous casting slab quality prediction model based on statistical method are weak, which has certain limitations for the continuous casting slab quality prediction of multiple steel grades and small batch pro- duction mode in iron and steel enterprises. The continuous casting slab quality prediction model based on the non-statistical method has strong adaptability and generalization ability. Because there are many influences on continuous casting slab quality, and there is a strong nonlinear relationship between various influencing factors and continuous casting slab defects, the strong nonlinear approximation ability of artificial neural network can well solve this problem, and BP neural network is one of the most widely used neural networks at present. Similarly, this method has also been successfully applied to the quality prediction of continuous cast- ing slab in the field of iron and steel. However, the model based on this method needs a lot of time in the training process, is easy to fall into the local optimal value, and a large number of network training parameters need to be set in the training pro- cess, and the prediction accuracy is low. It is difficult to accu- rately determine the quality of continuous casting slab in iron and steel enterprises quickly and in time, which is not conducive to the efficient production of high-quality steel in iron and steel enterprises. Therefore, developing a continuous casting slab quality predicting method with strong adaptability, fast operation speed and high prediction accuracy is of great significance to im- prove "narrow window" automatic control level of continuous cast- ing slab quality in steelmaking-continuous casting process.
SUMMARY The technical problem to be solved by the present disclosure is to provide a continuous casting slab quality predicting method based on extreme learning machine.
The method includes the following steps: (1) selection of input variables of extreme learning machine: performing correlation analysis by Pearson correlation coefficient to find influencing factors of continuous casting slab quality ac- cording to the influencing factors of continuous casting slab quality and continuous casting slab defects; (2) collecting data of the influencing factors of continuous casting slab quality, pre-processing the data, determining a num- ber of input nodes and output nodes of the extreme learning ma- chine, and determining sample data used to establish a model; (3) normalizing input data of the extreme learning machine, and selecting two-thirds of the collected sample data to input in- to the extreme learning machine to complete a training of the ex- treme learning machine; (4) inputting remaining one-third of the sample data to the extreme learning machine to verify an accuracy of the model and setting a reasonable number of hidden layer nodes and hidden layer activation functions by comprehensively considering a training fitting degree of extreme learning machine and the accuracy of judging defect grade, to ensure an optimization of network struc- ture and complete a defect grade classification of continuous casting slab quality.
Wherein, the data pre-processing method in the step (2) is to pre-process abnormal data through an elimination and data smooth- ing technology.
A selection range of normalization processing of data in the step (3) is [0.1, 0.9]. Due to a large magnitude difference be- tween variable data, it will affect the classification accuracy of the extreme learning machine on the defect level of the continuous casting slab. The range of data normalization is selected as [0.1,
0.9], which can avoid the impact on the prediction accuracy of the continuous casting slab quality prediction model based on the ex- treme learning machine due to the large magnitude difference of the data, and can maintain the original information of each varia- ble data.
The establishment of a whole model in the step (2) includes model training, model verification and model parameter selection.
The selection of historical data in the step (3) is to ran- domly select two-thirds of the sample data.
In this method, the industrial control computer and process database are used to realize the real-time prediction of the con- tinuous casting slab quality. Wherein, the industrial control com-
puter is used to judge and feedback the continucus casting slab quality in real time; and the process database is connected to the industrial control computer to collect and record the continuous casting slab production process data in real time, so as to pro- vide data support for the operation of the industrial control com- puter. The beneficial effects of the above technical scheme of the present disclosure are as follows: In the above scheme, the poor adaptability, long training time and easy to fall into local optimal value of the continuous casting slab quality prediction model based on statistical method, expert system and BP neural network can be avoided. The method does not need to set a large number of neural network parameters and find the optimal network structure parameters in the training process, and the classification accuracy and operation speed of the continuous casting slab quality judgment model are obviously improved. The method of the present disclosure classifies the defect grade of continuous casting slab through the extreme learning ma- chine. In the training process, it is not necessary to adjust the input weight of the network and the offset of the hidden layer, and only numbers of hidden layer nodes of the network need to be set to produce a unique optimal solution. Moreover, the model has fast training speed, high prediction accuracy and good adaptabil- ity. Compared with the continuous casting slab quality prediction model based on statistical method, expert system and BP neural network, the prediction accuracy and calculation speed have been significantly improved, and the quality of the continuous casting slab can be judged timely and accurately.
BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a schematic diagram of a composition of a continu- ous casting slab quality predicting method based on extreme learn- ing machine of the present disclosure; Fig. 2 is a calculation flowchart of each operating module of an industrial control computer involved in the present disclosure.
DETAILED DESCRIPTION OF THE EMBODIMENTS In order to make the technical problems, technical solutions and advantages to be solved by the present disclosure clearer, a detailed description will be given below with reference to the ac- companying drawings and specific embodiments.
The present disclosure provides a continuous casting slab 5 quality predicting method based on extreme learning machine, as shown in Fig. 1, which is a schematic diagram of the method.
As shown in Fig. 2, the specific steps of the method are as follows: (1) selection of input variables of extreme learning machine: performing correlation analysis by Pearson correlation coefficient to find influencing factors of continuous casting slab quality ac- cording to the influencing factors of continuous casting slab quality and continuous casting slab defects; (2) collecting data of the influencing factors of continuous casting slab quality, pre-processing the data, determining a num- ber of input nodes and output nodes of the extreme learning ma- chine, and determining sample data used to establish a model; (3) normalizing input data of the extreme learning machine, and selecting two-thirds of the collected sample data to input in- to the extreme learning machine to complete a training of the ex- treme learning machine; (4) inputting remaining one-third of the sample data to the extreme learning machine to verify an accuracy of the model and completing a defect grade classification of continuous casting slab quality.
In a specific application, the effectiveness of the model is mainly verified by three typical defects (central porosity, cen- tral segregation, shrinkage cavity) of 180mmx180mm continuous casting slab 60Si2Mn spring steel from a domestic special steel plant. Firstly, prediction processing and correlation analysis are performed on the collected sample data to determine the required data for the model; secondly, two-thirds of the data samples are used to establish a continuous casting slab quality prediction model based on extreme learning machine, and then the established model is used to process one-third of the data sample to be clas- sified; finally, the recognition result is compared with the actu- al result based on the output of the model to draw a conclusion,
and the time required for the model operation is counted. The training process does not need to set a large number of neural network structure parameters, and just need to select and deter- mine an appropriate number of hidden layer nodes to obtain the op- timal solution of the model, which greatly improves the calcula- tion speed and classification accuracy of the model.
The specific application process is as follows: (1) Firstly, historical data of the 180mmx180mm continuous casting slab 60Si2Mn spring steel produced by the domestic special steel plant is collected, and influencing factor sets affecting the continuous casting slab quality is determined. The present disclosure takes the three typical defects of central porosity, central segregation and shrinkage cavity defects of continuous casting slab as an example, and a number of input variables of the model are determined by using Pearson correlation coefficient for correlation analysis of the collected data and Professor Cai Kaike's induction and summary.
(2) The factors affecting the continuous casting slab quality are preprocessed, and the abnormal data are preprocessed to deter- mine the final sample data set.
(3) The data sample set obtained above are normalized, and the data normalization range is [0.1, 0.9], at the same time, classification labels for the defect grade of continuous casting slab corresponding to the sample are created.
{4) The extreme learning machine is constructed. According to the data processed in (3), two-thirds of the data are randomly se- lected to train the model, and the remaining one-third of the data is randomly selected to verify the effectiveness of the model. The reasonable number of hidden layer nodes and hidden layer activa- tion function are set by comprehensively considering the training fit of the extreme learning machine and the accuracy of judging defect level.
(5) After tundish molten steel pouring starts, the database system collects and records the pouring information in real time (tundish molten steel composition, mold equipment and process pa- rameters, molten steel temperature data, etc.), the data obtained above are normalized, and the data processing range is [0.1, 0.9].
According to the data provided by the process database system, the industrial control computer inputs the established continuous casting slab quality prediction model based on extreme learning machine, and conducts real-time judgment and classification of the three typical defects (central porosity, center segregation, shrinkage cavity) of spring steel produced on site.
(6) The classification results are predicted according to the extreme learning machine. If it is judged to be a defective con- tinuous casting slab, it will be sorted out in time; if the quali- ty of the judged continuous casting slab is a normal continuous casting slab, the continuous casting slab will directly enter a next process; which not only provides auxiliary information for the operator to control the continuous casting slab quality, but also provides a basis for the operator to judge the continuous casting slab quality.
The continuous casting slab quality prediction model based on extreme learning machine proposed by the present disclosure has been applied to the classification and quality evaluation of 60Si2Mn spring steel 180mmx180mm continuous casting slab defects of the domestic special steel plant. Taking the three typical de- fects of center porosity, center segregation, and shrinkage cavity of the continuous casting slab of the steel plant as an example to illustrate the experimental effect of the present disclosure. Ta- ble 1 shows the test results after the implementation of the pre- sent disclosure. The results show: compared with the continuous casting slab quality prediction model based on statistical method, expert system, and BP neural network, the continuous casting slab quality prediction model based on extreme learning machine has further improved the classification accuracy and operation speed of three typical defects: central porosity, central segregation and shrinkage cavity, and the accuracy of classifying the three typical defect levels of center porosity, center segregation, and shrinkage cavity of the of continuous casting slab can reach 85%,
82.5%, and 70% respectively, and the method is extremely fast in the process of determining the defect level, only need 0.1s. The method provided by the present disclosure can quickly and effi- ciently determine the continuous casting slab quality, thereby as-
sisting the operator to accurately control the continuous casting slab quality, and continuously improving control level of the con- tinuous casting slab quality.
Table 1: test results after implementation of the present disclosure orders steel actual actual actual predicted pradioted predicted grades center venta! shriskage central canta] shrinkage porosity segregation cavity porosity segregation cavity defects defects deïents defects defecte defects eN eis 3 BONS i 3 3 i 3 i i SSRI i i i 1 3 i § SININS & X 3 3 3 § 13 BONISss 3 3 i 3 3 2 3 SONIA 3 4 i 3 $ x 8 SNe 3 ¥ 3 3 3 & is SHIN i i 3 t 3 3 1 SONS 3 1 3 i 3 2 ® SORIA i 3 3 2 8 3 zi WSJ 3 i 1 i 1 i 43 FRI 3 3 3 3 d i 23 SIM 3 3 3 3 ¥ 3 23 SIR 3 4 3 3 3 1 za HIME & 3 3 3 3 3 hE SINNER 3 i 3 3 3 ï 3 BINIMe 3 1 3 2 3 3 Ee SORIA i i 3 i 3 i 38 DURE dn 3 3 3 i 3 3 3 SOSENe 2 ¥ 3 § X § 3 SIN i i 3 ¥ i 3 38 HENNE 8 i 3 & 3 2 IF SUSAN i x 3 3 1 oo 3E BTR i 1 & 3 i 2 23 GIST 3 3 i 2. 2 1 48 HOSEA i i 5 3 i 3 Note: since the extreme learning machine needs to set classi-
fication labels, the "1" in the table indicates the normal contin- uous casting slab, the "2" in the table indicates that the contin- uous casting slab defect grade is 0.5, and the "3" in the table indicates that the continuous casting slab defect grade is 1.0, the "4" in the table indicates that the continuous casting slab defect grade is 1.5, and the "5" in the table indicates that the continuous casting slab defect grade is 2.0. The continuous cast- ing slab defect grades increases from small to large, indicating that the defect grade is getting more and more serious.
The above are the preferred embodiments of the present dis- closure. It should be pointed out that for ordinary technicians in the technical field, several improvements and refinements can be made without departing from the principles of the present disclo- sure, and these improvements and refinements should also be re- garded as the protection scope of the present disclosure.

Claims (6)

CONCLUSIESCONCLUSIONS 1. Werkwijze voor het voorspellen van de kwaliteit van een met een continue gietmethode verkregen plaat op basis van een extreme learning-machine, waarbij de specifieke stappen als volgt zijn: (1) selectie van invoervariabelen van extreme learning-machine: het uitvoeren van correlatieanalyse door Pearson- correlatiecoëfficiënt om beïnvloedende factoren van de kwaliteit van een met een continue gietmethode verkregen plaat te vinden in overeenstemming met de beïnvloedende factoren van de kwaliteit van een met een continue gietmethode verkregen plaat en de defecten van een met een continue gietmethode verkregen plaat; (2) het verzamelen van gegevens van de beïnvloedende factoren van de kwaliteit van een met een continue gietmethode verkregen plaat, het voorbewerken van de gegevens, het bepalen van een aantal in- voerknooppunten en uitvoerknooppunten van de extreme learning- machine, en het bepalen van voorbeeldgegevens die worden gebruikt om een model tot stand te brengen; (3) het normaliseren van invoergegevens van de extreme learning- machine en het selecteren van tweederden van de verzamelde voor- beeldgegevens om in de extreme learning-machine in te voeren om een training van de extreme learning-machine te voltooien; (4) het invoeren van de resterende eenderde van de voorbeeldgege- vens in de extreme learning-machine om de nauwkeurigheid van het model te verifiëren en een classificatie van defecten van de kwa- liteit van een met een continue gietmethode verkregen plaat te voltooien.1. Method for predicting the quality of a slab obtained by a continuous casting method based on an extreme learning machine, the specific steps of which are as follows: (1) selection of input variables of extreme learning machine: conducting correlation analysis by Pearson correlation coefficient to find factors influencing the quality of a continuous casting plate according to the factors influencing the quality of a continuous casting plate and the defects of a continuous casting plate; (2) collecting data on the factors influencing the quality of a slab obtained by a continuous casting method, pre-processing the data, determining a number of input nodes and output nodes of the extreme learning machine, and determining sample data used to create a model; (3) normalizing input data from the extreme learning machine and selecting two-thirds of the collected sample data for input to the extreme learning machine to complete an extreme learning machine workout; (4) entering the remaining one-third of the sample data into the extreme learning machine to verify the accuracy of the model and complete a classification of defects on the quality of a continuous cast slab. 2. Werkwijze voor het voorspellen van de kwaliteit van een met een continue gietmethode verkregen plaat op basis van een extreme learning-machine volgens conclusie 1, waarbij de werkwijze voor voorverwerking van gegevens in stap (2) is om abnormale gegevens vooraf te verwerken door middel van een technologie voor elimina- tie en gegevensafvlakking.The method for predicting the quality of a slab obtained by a continuous casting method based on an extreme learning machine according to claim 1, wherein the data pre-processing method in step (2) is to pre-process abnormal data by means of of an elimination and data smoothing technology. 3. Werkwijze voor het voorspellen van de kwaliteit van een met een continue gietmethode verkregen plaat op basis van een extreme learning-machine volgens conclusie 1, waarbij een selectiebereik van normalisatieverwerking van gegevens in stap (3) [0,1, 0,9] is.The method for predicting the quality of a slab obtained by a continuous casting method based on an extreme learning machine according to claim 1, wherein a selection range of normalization processing of data in step (3) is [0.1, 0.9] is. 4. Werkwijze voor het voorspellen van de kwaliteit van een met een continue gietmethode verkregen plaat op basis van een extreme learning-machine volgens conclusie 1, waarbij het tot stand bren- gen van een heel model in stap (2) modeltraining, modelverificatie en modelparameterselectie omvat.The method for predicting the quality of a continuous casting machine based slab based on an extreme learning machine according to claim 1, wherein the establishment of a whole model in step (2) comprises model training, model verification and model parameter selection includes. 5. Werkwijze voor het voorspellen van de kwaliteit van een met een continue gietmethode verkregen plaat op basis van een extreme learning-machine volgens conclusie 1, waarbij de selectie van his- torische gegevens in stap (3) het willekeurig selecteren van twee- derden van de voorbeeldgegevens is.The method for predicting the quality of an extreme learning machine slab obtained by a continuous casting method according to claim 1, wherein the selection of historical data in step (3) includes randomly selecting two-thirds of is the sample data. 6. Werkwijze voor het voorspellen van de kwaliteit van een met een continue gietmethode verkregen plaat op basis van een extreme learning-machine volgens conclusie 1, waarbij de werkwijze gebruik maakt van industriële besturingscomputers en procesdatabases om realtime voorspelling van de kwaliteit van een met een continue gietmethode verkregen plaat te realiseren.The method of predicting the quality of a continuous casting machine based plate based on an extreme learning method according to claim 1, the method using industrial control computers and process databases to provide real-time prediction of the quality of a continuous casting casting method to realize the plate obtained.
NL2030465A 2022-01-07 2022-01-07 Continuous casting slab quality predicting method based on extreme learning machine NL2030465B1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
NL2030465A NL2030465B1 (en) 2022-01-07 2022-01-07 Continuous casting slab quality predicting method based on extreme learning machine

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
NL2030465A NL2030465B1 (en) 2022-01-07 2022-01-07 Continuous casting slab quality predicting method based on extreme learning machine

Publications (2)

Publication Number Publication Date
NL2030465A NL2030465A (en) 2022-11-15
NL2030465B1 true NL2030465B1 (en) 2022-11-24

Family

ID=84046021

Family Applications (1)

Application Number Title Priority Date Filing Date
NL2030465A NL2030465B1 (en) 2022-01-07 2022-01-07 Continuous casting slab quality predicting method based on extreme learning machine

Country Status (1)

Country Link
NL (1) NL2030465B1 (en)

Also Published As

Publication number Publication date
NL2030465A (en) 2022-11-15

Similar Documents

Publication Publication Date Title
CN114329940A (en) Continuous casting billet quality prediction method based on extreme learning machine
Cemernek et al. Machine learning in continuous casting of steel: A state-of-the-art survey
CN109675935B (en) Online fault diagnosis method for IPCA rolling process with variable control limit
US20200341460A1 (en) Production process control method and production process control device
CN116307289B (en) Textile processing procedure parameter detection and prediction method, system and storage medium
CN106600464A (en) Power grid construction new project cost rapid estimation method based on fuzzy mathematics
CN110991916A (en) Casting blank quality judgment system and method
WO2021007871A1 (en) Alumina production operation optimization system and method employing cloud-side collaboration
CN111950577A (en) Point inspection method and device
CN115533059A (en) Method for optimizing cutting length of casting blank
CN113192568A (en) Refining furnace desulfurization terminal point forecasting method and system
CN116050678A (en) Die-casting product processing test system and method based on cloud computing
NL2030465B1 (en) Continuous casting slab quality predicting method based on extreme learning machine
CN110930065A (en) Casting blank quality determination method and system
CN114692983A (en) Automatic gear shifting prediction method and system for special vehicle
CN116703254B (en) Production information management system for mechanical parts of die
CN110855519A (en) Network flow prediction method
CN109840309A (en) A kind of calculation method of iron melt desulfurizing agent dosage
Tokova et al. Inductive modelling as a basis of informational support of decisions in foundry production
CN114800049A (en) Grating ruler processing operation signal error compensation system
CN108364095B (en) Molten steel quality diagnosis method in steelmaking production process based on data mining
CN112758106A (en) Vehicle running track prediction method and device
TW201812586A (en) A machine maintenance forecasting system and the methods
CN115034370B (en) BP network model-based method for predicting hearth activity of blast furnace
CN112099353B (en) Divergence constraint kernel discriminant analysis-based continuous casting billet subsurface slag inclusion defect prediction method