CN107377634B - A kind of hot-strip outlet Crown Prediction of Media method - Google Patents
A kind of hot-strip outlet Crown Prediction of Media method Download PDFInfo
- Publication number
- CN107377634B CN107377634B CN201710588439.4A CN201710588439A CN107377634B CN 107377634 B CN107377634 B CN 107377634B CN 201710588439 A CN201710588439 A CN 201710588439A CN 107377634 B CN107377634 B CN 107377634B
- Authority
- CN
- China
- Prior art keywords
- strip
- particle
- exports
- crown prediction
- hot
- 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
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B21—MECHANICAL METAL-WORKING WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
- B21B—ROLLING OF METAL
- B21B37/00—Control devices or methods specially adapted for metal-rolling mills or the work produced thereby
- B21B37/28—Control of flatness or profile during rolling of strip, sheets or plates
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B21—MECHANICAL METAL-WORKING WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
- B21B—ROLLING OF METAL
- B21B38/00—Methods or devices for measuring, detecting or monitoring specially adapted for metal-rolling mills, e.g. position detection, inspection of the product
- B21B38/02—Methods or devices for measuring, detecting or monitoring specially adapted for metal-rolling mills, e.g. position detection, inspection of the product for measuring flatness or profile of strips
Abstract
The hot-strip of the present invention exports Crown Prediction of Media method：Layering Cai Ji be in hot-strip production process strip creation data；Noise reduction process is carried out to creation data；Creation data after noise reduction is divided into training set and test set；Creation data after noise reduction is subjected to dimension-reduction treatment；Using the normalized matrix after dimensionality reduction as the input of supporting vector machine model, the parameter of supporting vector machine model is optimized using the particle swarm optimization algorithm based on hybridization；Crown Prediction of Media model is exported using best parameter group construction support vector machines strip；Forecasting model is trained with training set, the Generalization Capability of forecasting model is tested with test set.The forecasting procedure of the present invention determines the optimal parameter of support vector machines by Crossbreeding Particle Swarm optimizing, and the precision that the support vector machines strip established based on support vector machines exports Crown Prediction of Media model is made to be improved.Forecasting model is based on mass production data, and the acquisition of creation data is easily operated, and the Generalization Ability of model is stronger.
Description
Technical field
The present invention relates to a kind of hot-strip quality control technologies more particularly to a kind of hot-strip to export Crown Prediction of Media side
Method.
Background technology
Hot strip rolling has highly important status, the about half of world steel total amount all sources in steel and iron industry
With hot continuous rolling production line.One hot continuous rolling production line includes the equipment of many precisions, complicated Hybrid Control Model and evil
Bad working environment, this is all that the raising of product quality brings difficulty.But with the development of science and technology, every profession and trade, each portion
Door is more and more to the demand of strip, while user is also more and more high to the quality of strip, especially to household electrical appliances steel plate, automobile
The plates shape such as steel plate, tin plate and electric steel plate is proposed very high request.If crown of strip is bad, occur excessive
Convexity, local crowning, wedge shape etc. will all seriously affect quality and the service life of consumer products.Continuous hot-rolling mill is-it is a non-linear, big
Time lag, multivariable, the dynamical system of close coupling.There are many factor for influencing strip outlet convexity, such as：Roll-force, bending roller force, roller
The thermal expansion of shape and roll, roller diameter, incoming profile, plate be wide, milling train time lag, mill speed, the variation of milling train rhythm, band
The temperature fluctuation of material and cooling water, the variation etc. of rolling mill screwdown amount.Therefore, to realize that accurately controlling for the system is one difficult
Task.Traditional method is to set up strip crown relational model using traditional mathematical tool according to rolling therory, and analysis is rolled
The flexure of state bottom roll processed, flattening, situations such as thermally expanding.For the ease of modeling, the complexity of system need to be simplified, provided perhaps
More assumed conditions, but to reduce model accuracy as cost.Requirement with Modern Manufacturing Technology to strip shape precision is gradually increased,
Improving the task of model or control accuracy becomes very urgent.For this reason, it may be necessary to which new method is found to carry out more rolling machine system
Accurate prediction and modeling export convexity to achieve the purpose that accurately control strip.
Invention content
The embodiment of the present invention proposes that a kind of hot-strip exports Crown Prediction of Media method, and this method passes through Crossbreeding Particle Swarm
Optimizing determines the optimal parameter of support vector machines, and the support vector machines strip established based on support vector machines is made to export Crown Prediction of Media
The precision of model is improved.
The present invention provides a kind of hot-strip outlet Crown Prediction of Media method, includes the following steps：
Step 1：Layering Cai Ji be in hot-strip production process each piece of strip p creation data and with a p
Dimensional vector is indicated, and layer is not divided according to steel grade, finish to gauge strip width and finish to gauge belt steel thickness；
Step 2：Noise reduction process is carried out to the other creation data of each layer using 3 σ principles of statistics；
Step 3：Creation data after noise reduction is divided into two set of training set and test set, set according to a certain percentage
The consistency of data distribution will be kept by dividing；
Step 4：The other creation data of each layer after noise reduction is constituted into observation matrix, and standard is carried out to observation matrix
Change transformation and dimension-reduction treatment, obtains the normalized matrix after dimensionality reduction；
Step 5：Using the normalized matrix after dimensionality reduction as the input of supporting vector machine model, using the particle based on hybridization
Colony optimization algorithm optimizes the parameter of supporting vector machine model；
Step 6：The best parameter group construction support vector machines strip obtained using optimization exports Crown Prediction of Media model；
Step 7：Crown Prediction of Media model is exported with training set Training Support Vector Machines strip, supporting vector is tested with test set
Machine strip exports the Generalization Capability of Crown Prediction of Media model；
Step 8：Using coefficient of determination R2, mean absolute error MAE, average absolute percent error MAPE, root-mean-square error
RMSE come evaluate support vector machines strip export Crown Prediction of Media model overall performance.
The hot-strip outlet Crown Prediction of Media method of the present invention at least has the advantages that：Present invention employs one kind
Artificial intelligence approach exports Crown Prediction of Media model to establish support vector machines strip.Model is based on mass production data, and produces
The acquisition of data is easily operated, so the Generalization Ability of model is stronger.Seek to influence heat in addition, having broken away from during model foundation
Complicated mathematical physics relationship between strip outlet each variable of convexity is rolled, solves strong coupling between each input variable well
Close, it is non-linear the problems such as.It can effectively be carried out using the method for the present invention after strip sample data by reasonably screening and handling
Hot-strip exports Crown Prediction of Media, lays a good foundation for being precisely controlled for convexity of outlet.
Description of the drawings
Fig. 1 is the flow of the hot-strip outlet Crown Prediction of Media method based on Crossbreeding Particle Swarm Support Vector Machines Optimized
Figure；
Fig. 2 is design sketch of the creation data after Principal Component Analysis dimensionality reduction；
Fig. 3 is fitness value and average fitness change during Crossbreeding Particle Swarm Support Vector Machines Optimized structural parameters
Change figure；
Fig. 4 is the prediction effect figure of model strip outlet convexity on training set；
Fig. 5 is the prediction effect figure of model strip outlet convexity on test set.
Specific implementation mode
The present invention is described in further detail with reference to the accompanying drawings and examples.
It is pre- for carrying out strip outlet convexity with the Final Stand Rolling data of certain 1780 hot tandem in the present embodiment
Report, the flow that hot-strip exports Crown Prediction of Media method are as shown in Figure 1.Forecasting procedure includes the following steps：
Step 1：Layering Cai Ji be in hot-strip production process each piece of strip p creation data and with a p
Dimensional vector is indicated, and layer is not divided according to steel grade, finish to gauge strip width and finish to gauge belt steel thickness.
When it is implemented, acquiring the creation data of the last rack of certain 1780 hot tandem rolling mill, the creation data of acquisition includes
The inlet temperature T of each rack, outlet temperature t, inlet thickness H, exit thickness h, strip width W were rolled in the operation of rolling
Journey bending roller force FB, roll shifting amount S, rolling force FR, roll cooling water flow QmWith entrance convexity CH.The acquisition of data is other by layer
It carries out, layer is not divided according to steel grade and strip steel specification.Division methods are as shown in table 1.
Table 1 is that data collection layer is other in the present embodiment.
| Steel grade | Sample size | Finish to gauge thickness (mm) | Strip width (mm) |
| SPHC-B | 100 | 4.2 | 1411 |
| SPA-H | 100 | 6.0 | 1199 |
| 65Mn-YT | 100 | 5.0 | 1391 |
| SS400B-1 | 74 | 2.95 | 1332 |
| SS400B-1 | 100 | 3.6 | 1293 |
Step 2：Noise reduction process is carried out to the other creation data of each layer using 3 σ principles of statistics.
When it is implemented, obtaining the manufacturing parameter of 474 blocks of steel using 3 σ principle cancelling noise data of statistics.Part producing
Parameter is as shown in table 2.
Table 2 is the manufacturing parameter of part strip.
Step 3：Creation data after noise reduction is divided into two set of training set A and test set B, collection according to a certain percentage
Close the consistency for dividing and keeping data distribution.
When it is implemented, choose each specification strip steel data 80% (380 pieces) be used as training set, remaining 20% (94
Block) it is used as test set.Training set data=100*80%+100*80%+100*80%+74*80%+100*80%=380 is surveyed
Examination collection data=100*20%+100*20%+100*20%+74*20%+100*20%=94.
Step 4：The other creation data of each layer after noise reduction is constituted into observation matrix, and standard is carried out to observation matrix
Change transformation and dimension-reduction treatment, obtains the normalized matrix after dimensionality reduction.
Step 4 specifically includes：
Step 4.1：The other creation data of each layer after noise reduction is constituted into observation matrix, and to observation matrix into rower
Standardization transformation obtains normalized matrix, specially：
The creation data of each piece of strip is regarded as a p dimensional vector X=(X1,X2,…,Xp), p is the number of manufacturing parameter
It measures, p=10 in the present embodiment obtains the creation data of 474 pieces of strips, creation data X=(X altogether after noise reduction1,X2,…,X10)
Observation matrix be expressed as：
Normalized matrix is obtained after normalized transformation to be expressed as：
Standardized formula is：
Wherein,
Step 4.2：Dimension-reduction treatment is carried out to normalized matrix using Principal Component Analysis.It specifically includes：
Step 4.2.1：Calculate the correlation matrix R of collected creation data：
In the present embodiment, the matrix element of correlation matrix R is listed by table 3.
The matrix element list of 3 correlation matrix R of table：
Wherein,
Step 4.2.2：Calculate the characteristic value (λ of correlation matrix R1,λ2,…,λ10) and corresponding feature vector, and make
It is ranked sequentially by size, λ1≥λ2≥…λp≥0；Calculate separately character pair value λiFeature vector be expressed as：
ei=(ei,1,ei,2,…,ei,474), i=1,2 ..., 10 (5)
Calculate separately character pair value λiFeature vector, make | | ei| |=1, i.e.,Wherein eijIndicate vector ei
J-th of component.In the present embodiment the characteristic value of correlation matrix R be (3.8836,2.1965,1.4238,1.0119,
0.9145,0.3324,0.1555,0.0587,0.0229,0.0003)
Step 4.2.3：Important principal component is selected, and writes out principal component expression formula, according to each principal component contribution rate of accumulative total
Size choose k principal component, contribution rate η refers to the proportion that the variance of some principal component accounts for whole variances, is also related herein
Some characteristic value of coefficient matrix R accounts for the total proportion of All Eigenvalues, i.e.,：
Contribution rate of accumulative total is：
Calculate each principal component contributor rate and contribution rate of accumulative total.
4 each principal component contributor rate of table and contribution rate of accumulative total：
| Principal component | Characteristic value | Contribution rate (%) | Contribution rate of accumulative total (%) |
| 1 | 3.8836 | 0.3884 | 0.3884 |
| 2 | 2.1965 | 0.2197 | 0.608 |
| 3 | 1.4238 | 0.1424 | 0.7504 |
| 4 | 1.0119 | 0.1012 | 0.8516 |
| 5 | 0.9145 | 0.0915 | 0.943>0.90 |
| 6 | 0.3324 | 0.0332 | 0.9763>0.95 |
| 7 | 0.1555 | 0.0156 | 0.9918 |
| 8 | 0.0587 | 0.0059 | 0.9977 |
| 9 | 0.0229 | 0.0023 | 1 |
| 10 | 0.0003 | 0 | 1 |
Contribution rate is bigger, illustrates that the information for the original variable that the principal component is included is stronger, in the present invention, contribution rate of accumulative total
Reach 90% or more, just can guarantee most information of original variable.Become 5 dimension datas from 10 dimension datas after dimensionality reduction,
Principal component is as shown in Figure 2.
Step 5：Using the normalized matrix after dimensionality reduction as the input of supporting vector machine model, using the particle based on hybridization
Colony optimization algorithm optimizes the parameter of supporting vector machine model；
When it is implemented, using the manufacturing parameter after dimensionality reduction as the input of supporting vector machine model, using based on hybridization
Particle swarm optimization algorithm optimizes the parameter of supporting vector machine model, these parameters include the penalty coefficient of support vector machines
C, kernel functional parameter σ and loss function value ε.Optimization Steps include the following steps：
Step 5.1：Particle cluster algorithm is initialized, the position and speed of each particle in population scale, population is carried out just
Beginningization；
Specially：One p dimensions search space of definition, the quantity for the manufacturing parameter that wherein p is acquired by each piece of strip,
There are the molecular population X=(X of n grain in p dimensions search space1,X2,...,Xn), wherein i-th of particle be expressed as p dimension to
Measure Xi=[xi1,xi2,…,xip]T, position of i-th of particle in p ties up search space is represented, the speed of i-th of particle is Vi=
[Vi1,Vi2,…,Vip]T, individual extreme value is Pi=[Pi1,Pi2,…,Pip]T, the global extremum of population is Pg=[Pg1,Pg2,…,
Pgp]T.P=5 in the present embodiment.
Step 5.2：Calculate the fitness value of each particle in population；It specifically includes：
Step 5.2.1：It determines fitness function, the predicted value and reality of convexity is exported using strip under the conditions of cross validation
For mean square error MSE between value as fitness function, fitness function expression formula is as follows：
Wherein,The predicted value of convexity, y are exported for stripiThe actual value of convexity is exported for strip；
Step 5.2.2：The fitness value of each particle is calculated according to fitness function.
Step 5.3：The fitness value of each particle and individual extreme value are compared, if its fitness value is more than individual
Extreme value then uses its fitness value as new individual extreme value；The fitness value of each particle and global extremum are compared, if its
Fitness value then uses its fitness value as new global extremum more than global extremum.
Step 5.4：According to the position and speed of new individual extreme value and new global extremum more new particle；
Wherein, the more new formula of particle rapidity is：
The more new formula of particle position is：
Wherein, ω is inertia weight；D=1,2 ..., p；I=1,2 ..., n；K is current iteration number；VidFor particle
Speed；c1,c2For acceleration factor；r1,r2Random number between being 0~1.In the present embodiment, ω=0.72, c are taken1=c2=
1.19, the broken number of population quantity 20, maximum iteration 100, cross-validation method is 5.The search ranges C are that 0~100, σ is searched
Rope ranging from 0~100, ε searches ranging from 0~1.
Step 5.5：Population is reinitialized, choosing certain amount of particle according to probability of crossover puts it into hybridization pond
In, hybridization generates same number of filial generation particle to the parent particle in pond two-by-two at random, constitutes new population；
Wherein, the speed of the position of filial generation particle and filial generation particle is expressed as：
Wherein, mxFor the position of parent particle, nxFor the position of filial generation particle, mvFor the speed of parent particle, nvFor filial generation
The speed of particle, i are random number between 0~1.
Step 5.6：Step 5.2 is repeated to step 5.4 more new individual extreme value and global extremum, and then updates filial generation particle
Position and speed.
Step 5.7：When iterations reach setting value, stopping optimizes and exports optimum results, searched out in the present embodiment
Best parameter group be (16.75,0.097,0.0473).
The ginseng to supporting vector machine model using the particle swarm optimization algorithm based on hybridization is illustrated in figure 3 in the present embodiment
Fitness value variation diagram during number optimizes, c1=c2=1.19, population quantity 20, maximum iteration 100.
Step 6：It is convex using best parameter group (C, σ, ε) the construction support vector machines strip outlet searched out in step 5
Forecasting model is spent, is specifically included：
Step 6.1：The actual value that collected creation data and strip are exported to convexity constitutes data setxiFor the creation data for influencing strip and exporting convexity of selection, yiFor
Strip exports the actual value of convexity, defines decision plane f (x)=wTφ (x)+b is that support vector machines strip exports Crown Prediction of Media
Model, support vector machines strip outlet Crown Prediction of Media model the problem of expression formula be defined as：
Wherein, φ (xi) it is high-dimensional feature space i=1 ..., m, w are the adjustable weight vector of decision plane, and b is decision
The offset of the biasing of plane, i.e. decision plane relative to origin；
C is penalty coefficient, C>0, ε indicates f (x) and yiBetween maximum deviation, lεFor insensitive loss function,
Step 6.2：Introduce slack variableThe problem of Crown Prediction of Media model is exported to support vector machines strip expression formula
It is rewritten, obtains revised problem expression formula：
Step 6.3：Introduce Lagrange multiplier α, α*,μ,μ*, obtain Lagrangian such as following formula：
Step 6.4：Enable L (w, b, α, α*,ξ,ξ*,μ,μ*) to w, b, ξ, ξ*Partial derivative is zero：
Step 6.5：Lagrangian is substituted into revised problem expression formula, obtains dual problem expression formula：
Wherein, Q=φ (xi)Tφ(xj)；
It solves dual problem and obtains w, the solution of b：
Step 6.6：It obtains support vector machines strip and exports Crown Prediction of Media model：
Wherein,σ is kernel functional parameter.
Step 7：Crown Prediction of Media model is exported with the support vector machines strip constructed in training set A training steps 6, with test
Collect B test support vector machines strips and exports Crown Prediction of Media model generalization performance；
Step 8：Using coefficient of determination R2, mean absolute error MAE, average absolute percent error MAPE, root-mean-square error
RMSE come evaluate support vector machines strip export Crown Prediction of Media model overall performance.Their calculation formula are as follows：
5 model error result of calculation of table.
| Training set | Test set | |
| MAE | 1.7426 | 2.1042 |
| MAPE (%) | 2.8704 | 3.2767 |
| RMSE | 2.1374 | 1.3018 |
Forecasting model on training set prediction effect as shown in figure 4, the prediction effect on test set is as shown in Figure 5.
The hot-strip outlet Crown Prediction of Media method of the present invention at least has the advantages that：Present invention employs one kind
Artificial intelligence approach establishes the strip crown forecasting model of hot-strip.Model is based on mass production data, and creation data
Acquire it is easily operated, so the Generalization Ability of model is stronger.Seek to influence hot-strip in addition, having broken away from during model foundation
Complicated mathematical physics relationship, solves close coupling between each input variable well between outlet each variable of convexity, non-thread
The problems such as property.Hot-strip can be effectively carried out using the method for the present invention after strip sample data by reasonably screening and handling
Crown Prediction of Media is exported, is laid a good foundation for being precisely controlled for convexity.
The foregoing is merely the preferable embodiments of the present invention, are not intended to limit the invention, all the present invention's
Any modification made by within spirit and principle, equivalent replacement and improvement etc., should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of hot-strip exports Crown Prediction of Media method, which is characterized in that include the following steps：
Step 1：Layering Cai Ji be in hot-strip production process each piece of strip p creation data and with a p tie up to
Amount is indicated, and layer is not divided according to steel grade, finish to gauge strip width and finish to gauge belt steel thickness；
Step 2：Noise reduction process is carried out to the other creation data of each layer using 3 σ principles of statistics；
Step 3：Creation data after noise reduction is divided into two set of training set and test set according to a certain percentage, set divides
Keep the consistency of data distribution；
Step 4：The other creation data of each layer after noise reduction is constituted into observation matrix, and change is standardized to observation matrix
It changes and dimension-reduction treatment, obtains the normalized matrix after dimensionality reduction；
Step 5：It is excellent using the population based on hybridization using the normalized matrix after dimensionality reduction as the input of supporting vector machine model
Change algorithm to optimize the parameter of supporting vector machine model；
Step 6：The best parameter group construction support vector machines strip obtained using optimization exports Crown Prediction of Media model；
Step 7：Crown Prediction of Media model is exported with training set Training Support Vector Machines strip, support vector machines band is tested with test set
Steel exports the Generalization Capability of Crown Prediction of Media model；
Step 8：Using coefficient of determination R2, mean absolute error MAE, average absolute percent error MAPE, root-mean-square error RMSE are next
Evaluate the overall performance of support vector machines strip outlet Crown Prediction of Media model.
2. hot-strip as described in claim 1 exports Crown Prediction of Media method, which is characterized in that the step 4 includes：
Step 4.1：The other creation data of each layer after noise reduction is constituted into observation matrix, and observation matrix is standardized
Transformation obtains normalized matrix；
Step 4.2：Dimension-reduction treatment is carried out to normalized matrix using Principal Component Analysis.
3. hot-strip as claimed in claim 2 exports Crown Prediction of Media method, which is characterized in that the step 4.1 is specially：
The creation data of each piece of strip is regarded as a p dimensional vector X=(X1,X2,…,Xp), obtain N block bands altogether after noise reduction
The creation data of steel, creation data X=(X1,X2,…,Xp) observation matrix be expressed as：
Normalized matrix is obtained after normalized transformation to be expressed as：
Standardized formula is：
Wherein,
4. hot-strip as claimed in claim 3 exports Crown Prediction of Media method, which is characterized in that the step 4.2 includes：
Step 4.2.1：Calculate the correlation matrix R of collected creation data：
Wherein,
Step 4.2.2：Calculate the characteristic value (λ of correlation matrix R1,λ2,…λp), and it is made to be ranked sequentially by size, λ1≥λ2
≥…λp≥0；Calculate separately character pair value λiFeature vector ei(i=1,2 ..., p), makes ‖ ei‖=1, i.e.,Its
Middle eijIndicate vector eiJ-th of component；
Step 4.2.3：Important principal component is selected, and writes out principal component expression formula, according to the big of each principal component contribution rate of accumulative total
Small k principal component of selection, contribution rate η refer to the proportion that the variance of some principal component accounts for whole variances, are also related coefficient herein
Some characteristic value of matrix R accounts for the total proportion of All Eigenvalues, i.e.,：
Contribution rate of accumulative total is：
Contribution rate is bigger, illustrates that the information for the original variable that the principal component is included is stronger, and contribution rate of accumulative total reaches 90% or more,
It just can guarantee most information of original variable.
5. hot-strip as described in claim 1 exports Crown Prediction of Media method, which is characterized in that the step 5 includes：
Step 5.1：Particle cluster algorithm is initialized, the position and speed of each particle in population scale, population is initialized；
Step 5.2：Calculate the fitness value of each particle in population；
Step 5.3：The fitness value of each particle and individual extreme value are compared, if its fitness value is more than individual extreme value
Then use its fitness value as new individual extreme value；The fitness value of each particle and global extremum are compared, if it is adapted to
Angle value then uses its fitness value as new global extremum more than global extremum；
Step 5.4：According to the position and speed of new individual extreme value and new global extremum more new particle；
Step 5.5：Population is reinitialized, choosing certain amount of particle according to probability of crossover puts it into hybridization pond, pond
In parent particle hybridization generates same number of filial generation particle two-by-two at random, constitute new population；
Step 5.6：Step 5.2 is repeated to step 5.4 more new individual extreme value and global extremum, and then updates the position of filial generation particle
And speed；
Step 5.7：When iterations reach setting value, stopping optimizes and exports the optimized parameter of supporting vector machine model.
6. hot-strip as claimed in claim 5 exports Crown Prediction of Media method, which is characterized in that the step 5.1 is specially：
It defines a p and ties up search space, the quantity for the manufacturing parameter that wherein p is acquired by each piece of strip ties up search space in p
There are the molecular population X=(X of n grain1,X2,...,Xn), wherein i-th of particle is expressed as the vectorial X of p dimensionsi=[xi1,
xi2,…,xip]T, position of i-th of particle in p ties up search space is represented, the speed of i-th of particle is Vi=[Vi1,
Vi2,…,Vip]T, individual extreme value is Pi=[Pi1,Pi2,…,Pip]T, the global extremum of population is Pg=[Pg1,Pg2,…,Pgp]T。
7. hot-strip as claimed in claim 6 exports Crown Prediction of Media method, which is characterized in that the step 5.2 is specially：
Step 5.2.1：Determine fitness function, using under the conditions of cross validation strip export convexity predicted value and actual value it
Between mean square error MSE as fitness function, fitness function expression formula is as follows：
Wherein,The predicted value of convexity, y are exported for stripiThe actual value of convexity is exported for strip；
Step 5.2.2：The fitness value of each particle is calculated according to fitness function.
8. hot-strip as claimed in claim 6 exports Crown Prediction of Media method, which is characterized in that particle in the step 5.4
The more new formula of speed is：
The more new formula of particle position is：
Wherein, ω is inertia weight；D=1,2 ..., p；I=1,2 ..., n；K is current iteration number；VidFor the speed of particle；
c1,c2For acceleration factor；r1,r2Random number between being 0~1.
9. hot-strip as claimed in claim 6 exports Crown Prediction of Media method, which is characterized in that step 5.5 generation of neutrons
The position of particle and the speed of filial generation particle are expressed as：
Wherein, mxFor the position of parent particle, nxFor the position of filial generation particle, mvFor the speed of parent particle, nvFor filial generation particle
Speed, i be 0~1 between random number.
10. hot-strip as described in claim 1 exports Crown Prediction of Media method, which is characterized in that the step 6 includes：
Step 6.1：The actual value that collected creation data and strip are exported to convexity constitutes data setxiFor the creation data for influencing strip and exporting convexity of selection, yiFor
Strip exports the actual value of convexity, defines decision plane f (x)=wTφ (x)+b is that support vector machines strip exports Crown Prediction of Media
Model, support vector machines strip outlet Crown Prediction of Media model the problem of expression formula be defined as：
Wherein, φ (xi) it is high-dimensional feature space i=1 ..., m, w are the adjustable weight vector of decision plane, and b is decision plane
Biasing, i.e. offset of the decision plane relative to origin；
C is penalty coefficient, C>0, ε is f (x) and yiBetween maximum deviation, lεFor insensitive loss function,
Step 6.2：Introduce slack variable ξi,The problem of exporting Crown Prediction of Media model to support vector machines strip expression formula carries out
It rewrites, obtains revised problem expression formula：
Step 6.3：Introduce Lagrange multiplier α, α*,μ,μ*, obtain Lagrangian such as following formula：
Step 6.4：Enable L (w, b, α, α*,ξ,ξ*,μ,μ*) to w, b, ξ, ξ*Partial derivative is zero
Step 6.5：Lagrangian is substituted into revised problem expression formula, obtains dual problem expression formula：
Wherein, Q=φ (xi)Tφ(xj)；
It solves dual problem and obtains w, the solution of b
Step 6.6：It obtains support vector machines strip and exports Crown Prediction of Media model：
Wherein,σ is kernel functional parameter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710588439.4A CN107377634B (en) | 2017-07-19 | 2017-07-19 | A kind of hot-strip outlet Crown Prediction of Media method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710588439.4A CN107377634B (en) | 2017-07-19 | 2017-07-19 | A kind of hot-strip outlet Crown Prediction of Media method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107377634A CN107377634A (en) | 2017-11-24 |
| CN107377634B true CN107377634B (en) | 2018-10-16 |
Family
ID=60340921
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710588439.4A Active CN107377634B (en) | 2017-07-19 | 2017-07-19 | A kind of hot-strip outlet Crown Prediction of Media method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107377634B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108229072A (en) * | 2018-02-10 | 2018-06-29 | 东北大学 | Continuous annealing soaking pit strip tension On-line Measuring Method based on data parsing |
| CN108637020B (en) * | 2018-05-09 | 2020-04-10 | 北京科技大学 | Self-adaptive variation PSO-BP neural network strip steel convexity prediction method |
| CN109513750B (en) * | 2018-11-06 | 2021-07-23 | 首钢集团有限公司 | Convexity feedback method considering waveform adjustment among racks |
| CN110428876B (en) * | 2019-07-31 | 2022-11-29 | 东北大学 | Steel material design method based on machine learning algorithm of physical guidance |
| CN110428175B (en) * | 2019-08-05 | 2022-05-24 | 东北大学秦皇岛分校 | Hot rolled strip steel plate convexity prediction method based on deep learning |
| CN110929347A (en) * | 2019-10-25 | 2020-03-27 | 东北大学 | Hot continuous rolling strip steel convexity prediction method based on gradient lifting tree model |
| CN111209967A (en) * | 2020-01-08 | 2020-05-29 | 太原科技大学 | Rolling process plate convexity prediction method based on support vector machine |
| US11769426B2 (en) | 2020-02-18 | 2023-09-26 | BlueOwl, LLC | Systems and methods for creating driving challenges |
| CN112170502A (en) * | 2020-09-16 | 2021-01-05 | 太原理工大学 | Method for predicting roll bending force in rolling process |
| CN112170501B (en) * | 2020-09-16 | 2022-05-27 | 太原理工大学 | Prediction method for wear crown and thermal crown of roller |
| CN112749505B (en) * | 2020-12-16 | 2023-10-24 | 太原科技大学 | Hot rolled strip steel cross-sectional shape prediction method based on mechanism fusion data |
| CN114021290B (en) * | 2022-01-04 | 2022-04-05 | 东北大学 | Plate strip steel convexity prediction method based on data driving and mechanism model fusion |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPS5477261A (en) * | 1977-12-02 | 1979-06-20 | Kawasaki Steel Co | Production of high crown rolled plate |
| CN1150068A (en) * | 1995-11-10 | 1997-05-21 | 东北重型机械学院南校 | Roller shape of axial movement capable of changing roll pass concavity and shape |
| JP3587180B2 (en) * | 2001-06-07 | 2004-11-10 | 住友金属工業株式会社 | Stainless steel plate with excellent stain resistance and corrosion resistance and its manufacturing method. |
| CN101214501A (en) * | 2007-12-27 | 2008-07-09 | 武汉钢铁(集团)公司 | Working roll consideration to strip steel convexity, edge drop control and abrasion control and using method |
| JP2009006373A (en) * | 2007-06-28 | 2009-01-15 | Hitachi Ltd | Crown controller and crown control method in hot-rolling mill |
| CN201720260U (en) * | 2010-07-20 | 2011-01-26 | 江苏省沙钢钢铁研究院有限公司 | Convexity and/or wedge automatic control system of hot rolling series-connection type rolling mill |
| CN102847721A (en) * | 2011-06-29 | 2013-01-02 | 宝山钢铁股份有限公司 | Method for determining thermal crown of hot rolled strip roller |
-
2017
- 2017-07-19 CN CN201710588439.4A patent/CN107377634B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPS5477261A (en) * | 1977-12-02 | 1979-06-20 | Kawasaki Steel Co | Production of high crown rolled plate |
| CN1150068A (en) * | 1995-11-10 | 1997-05-21 | 东北重型机械学院南校 | Roller shape of axial movement capable of changing roll pass concavity and shape |
| JP3587180B2 (en) * | 2001-06-07 | 2004-11-10 | 住友金属工業株式会社 | Stainless steel plate with excellent stain resistance and corrosion resistance and its manufacturing method. |
| JP2009006373A (en) * | 2007-06-28 | 2009-01-15 | Hitachi Ltd | Crown controller and crown control method in hot-rolling mill |
| CN101214501A (en) * | 2007-12-27 | 2008-07-09 | 武汉钢铁(集团)公司 | Working roll consideration to strip steel convexity, edge drop control and abrasion control and using method |
| CN201720260U (en) * | 2010-07-20 | 2011-01-26 | 江苏省沙钢钢铁研究院有限公司 | Convexity and/or wedge automatic control system of hot rolling series-connection type rolling mill |
| CN102847721A (en) * | 2011-06-29 | 2013-01-02 | 宝山钢铁股份有限公司 | Method for determining thermal crown of hot rolled strip roller |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107377634A (en) | 2017-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107377634B (en) | A kind of hot-strip outlet Crown Prediction of Media method | |
| CN108817103B (en) | Steel rolling model steel family layer classification optimization method | |
| CN104517162B (en) | A kind of continuous annealing product hardness Online integration learns forecasting procedure | |
| CN108655186B (en) | Roll-force presetting method based on artificial neural network and mathematical model | |
| JP5604945B2 (en) | Quality prediction apparatus, quality prediction method, computer program, and computer-readable recording medium | |
| CN104375478B (en) | A kind of method and device of Rolling production process product quality on-line prediction and optimization | |
| CN102601127A (en) | High-precision strip shape control prediction method for CVC (continuously variable crown) four-roll cold rolling mill | |
| CN110929347A (en) | Hot continuous rolling strip steel convexity prediction method based on gradient lifting tree model | |
| CN103942422B (en) | Granular-computation-based long-term prediction method for converter gas holder positions in metallurgy industry | |
| CN108637020A (en) | A kind of TSP question PSO-BP neural networks strip profile prediction technique | |
| WO2023130666A1 (en) | Strip steel plate convexity prediction method based on data-driving and mechanism model fusion | |
| CN106779384B (en) | Iron and steel industry blast furnace gas long-term interval prediction method based on optimal information granularity distribution | |
| CN104899425A (en) | Variable selection and forecast method of silicon content in molten iron of blast furnace | |
| Ji et al. | Application of SVM and PCA-CS algorithms for prediction of strip crown in hot strip rolling | |
| CN104657602A (en) | Method and system for forecasting thickness of band steel in hot continuous rolling production process | |
| CN107358363A (en) | Coal work incidence of disease Forecasting Methodology based on radial basis function neural network built-up pattern | |
| CN107716560B (en) | A kind of Hot Strip Rolling load distribution method | |
| Hu et al. | A novel forecast model based on CF-PSO-SVM approach for predicting the roll gap in acceleration and deceleration process | |
| Zhao et al. | Prediction of mechanical properties of cold rolled strip based on improved extreme random tree | |
| Song et al. | The prediction model for transverse thickness difference of electric steel in 6‐high cold rolling mills based on GA‐PSO‐SVR approach | |
| CN106825068A (en) | A kind of Forecasting Methodology of operation of rolling belt steel surface roughness | |
| Bordonaro et al. | Design of Experiment as a powerful tool when applying Finite Element Method: a case study on prediction of hot rolling process parameters | |
| CN102886384A (en) | Flatness defect identification method of 20-roller Sendzimir rolling mill based on support vector machine | |
| CN102277477A (en) | Establishment method for prediction model of tensile strength of cold rolled strip | |
| JP2011085970A (en) | Quality prediction device, quality prediction method, program and computer-readable recording medium |
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 | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20171124 Assignee: SHANDONG PROVINCE METALLURGICAL ENGINEERING Co.,Ltd. Assignor: Northeastern University Contract record no.: X2021210000025 Denomination of invention: A prediction method of exit crown of hot rolled strip Granted publication date: 20181016 License type: Common License Record date: 20210520 |
|
| EE01 | Entry into force of recordation of patent licensing contract |