CN106030424A  Rolling process learning control device  Google Patents
Rolling process learning control device Download PDFInfo
 Publication number
 CN106030424A CN106030424A CN201480075773.4A CN201480075773A CN106030424A CN 106030424 A CN106030424 A CN 106030424A CN 201480075773 A CN201480075773 A CN 201480075773A CN 106030424 A CN106030424 A CN 106030424A
 Authority
 CN
 China
 Prior art keywords
 value
 elementary cell
 learning coefficient
 rolling
 learning
 Prior art date
 Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
 Granted
Links
 238000005096 rolling process Methods 0.000 title claims abstract description 109
 238000011156 evaluation Methods 0.000 claims abstract description 46
 230000000875 corresponding Effects 0.000 claims abstract description 27
 238000004364 calculation method Methods 0.000 claims abstract description 8
 230000015572 biosynthetic process Effects 0.000 claims description 14
 238000005755 formation reaction Methods 0.000 claims description 14
 238000003860 storage Methods 0.000 claims description 10
 239000000463 material Substances 0.000 description 23
 238000000034 method Methods 0.000 description 10
 238000009499 grossing Methods 0.000 description 4
 239000000203 mixture Substances 0.000 description 4
 230000004069 differentiation Effects 0.000 description 3
 239000000284 extract Substances 0.000 description 3
 239000007795 chemical reaction product Substances 0.000 description 2
 230000000694 effects Effects 0.000 description 2
 238000005516 engineering process Methods 0.000 description 2
 238000005259 measurement Methods 0.000 description 2
 239000000047 product Substances 0.000 description 2
 239000000498 cooling water Substances 0.000 description 1
 238000005098 hot rolling Methods 0.000 description 1
 238000004519 manufacturing process Methods 0.000 description 1
 XLYOFNOQVPJJNPUHFFFAOYSAN water Substances data:image/svg+xml;base64,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 data:image/svg+xml;base64,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 O XLYOFNOQVPJJNPUHFFFAOYSAN 0.000 description 1
Classifications

 G—PHYSICS
 G05—CONTROLLING; REGULATING
 G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
 G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
 G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric

 B—PERFORMING OPERATIONS; TRANSPORTING
 B21—MECHANICAL METALWORKING WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
 B21B—ROLLING OF METAL
 B21B37/00—Control devices or methods specially adapted for metalrolling mills or the work produced thereby

 G—PHYSICS
 G05—CONTROLLING; REGULATING
 G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
 G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
 G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
 G05B13/0265—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion

 G—PHYSICS
 G06—COMPUTING; CALCULATING; COUNTING
 G06N—COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
 G06N20/00—Machine learning

 G—PHYSICS
 G06—COMPUTING; CALCULATING; COUNTING
 G06N—COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
 G06N3/00—Computer systems based on biological models
 G06N3/02—Computer systems based on biological models using neural network models
 G06N3/08—Learning methods
Abstract
An instantaneous value of a learning coefficient is calculated on the basis of a discrepancy between a result value and a result recalculated value, and an updated value of the learning coefficient is calculated from the instantaneous value. The calculated updated value is recorded in a cell corresponding to relevant rolling conditions. At the time of setting calculation, a predetermined number of adjacent cells are selected from cells satisfying references in terms of evaluation results of recency, saturation, and stability of the learning coefficient, the adjacent cells having a small spatial distance from an object cell corresponding to next rolling conditions in a space having rolling conditions as coordinate axes. By polynomial interpolation using a representative value of the learning coefficient in the selected adjacent cell, an estimated value of the learning coefficient in the object cell is calculated. The latest updated value of the learning coefficient in the object cell is corrected with the estimated value to determine a use value of the learning coefficient under the next rolling conditions.
Description
Technical field
The present invention relates to calculate based on real data the mould of prediction rolling phenomenon in the control of rolling mill practice
The learning coefficient of type formula are with the learning control device being managed each rolling condition.
Background technology
The control of rolling mill practice refers to, carries out rolling so that the rolling stock after having manufactured becomes desired
Size and temperature.Rolling mill practice controls, and is typically made up of setup control and sequencing contro.Setting control
In system, predict rolling phenomenon by model formation, thus determine mill speed, cooling water the water yield with
And the machine settings of the rolling equipment such as the roller gap of roll mill, so that rolling stock becomes desired chi
Very little and temperature.But, model formation can not show the physical phenomenon produced in rolling mill practice completely.Separately
Outward, for alleviating calculated load, adjusting the reasons such as convenience, and represent that the calculating formula of model obtains letter
Change.Therefore, send out between the actual value that gone out by sensor measurement and the predictive value calculated with model formation
Raw deviation.Then, in the setup control of existing rolling mill practice, in order to seek to roll the prediction of phenomenon
The raising of precision and stable, arranges learning coefficient in model formation, carries out automatically adjusting based on real data
The study of these learning coefficient whole controls.Comprising finishing temperature and rolling load in real data, these are actual
Data are collected for the impact point of setup algorithm.
Here, the summary of the study control of general rolling mill practice is illustrated.Study controls by multiple
Processing composition, one of them is that actual value calculating again processes.In actual value calculating again processes, use mould
Type formula calculates model predication value based on real data.Typically the model predication value calculated is referred to as reality
It is worth value of calculation again.Actual value included in actual value value of calculation again and real data compares, and calculates
Actual value value of calculation again is relative to the error of actual value, i.e. model error.Such as, with regard to the roll of roll mill
For the setup control in gap, by the actual value of the rolling load measured in load elementary cell with use mould
The actual value of the rolling load that type formula calculates according to real data value of calculation again compares, and calculates and rolls
The model error of load processed.
Then, learning coefficient are calculated based on model error.The learning coefficient now calculated are called learning coefficient
Instantaneous value.But, because model error is by the factor ignored when simplifying and build model formation
Being unknown caused by which kind of factor, the real data itself that additionally study is used also includes outside dry
Disturb or error, so the instantaneous value of the learning coefficient calculated according to model error can not be directly applied to down
One rolling stock.Then, the instantaneous value making learning coefficient passes through smoothing filter.Wink to learning coefficient
The value that duration obtains after being filtered is used as the updated value of learning coefficient.
Following formula is to calculate the concrete of the expression formula of the smoothing filter of updated value by the instantaneous value of learning coefficient
Example.By by between instantaneous value and the previous value (previous value of updated value) of learning coefficient of learning coefficient
Deviation be multiplied by more new gain, gained product is added with the previous value of learning coefficient, thus calculates study
The updated value of coefficient.
Znew=Zuse* (1α)+Zcur* α
Here, Znew: learning coefficient updated value
Zcur: learning coefficient (instantaneous value)
Zuse: learning coefficient (previous value)
α: learning coefficient more new gain (filter time constant)
The updated value of the learning coefficient calculated, is typically recorded to layered sheet.Layering is, rolls for differentiation
The concept of the rolling conditions such as the thickness of prepared material, width, strain, rate of straining, temperature.Such as, when
When distinguishing m thickness, n width of differentiation, layered sheet is made up of m × n elementary cell.This material every time
Calculate the updated value of learning coefficient at the end of the rolling of material, record consistent at the rolling condition with this material
In elementary cell.By layered sheet being used for the record of learning coefficient, it is possible to management is rolled according to each rightly
Condition processed and different learning coefficient, the precision of prediction of rolling phenomenon is improved.That is, layering is employed
The study of table controls, for guaranteeing the precision of prediction of the rolling phenomenon of model formation, guaranteeing the matter of finished product
The critical function of the stability of accuracy of measurement and rolling.
But, the study employing layered sheet controls to there is also problem.By management study in layered sheet
Coefficient, learning coefficient are managed in layered sheet in a corresponding elementary cell, at each corresponding base
This unit smooths and updates.Therefore, the learning coefficient saturated needs of change of an elementary cell are many
Individual roll mill meeting (rolling radical).As long as further, since rolling condition is slightly different will select difference
Elementary cell number, learning coefficient can be reupdated about different elementary cells.Therefore,
Constituting all of elementary cell of layered sheet, each elementary cell is required for multiple roll mill meeting.
As the scheme reducing required roll mill meeting, it is considered in the learning coefficient updating an elementary cell
Time also update the method for learning coefficient of adjacent elementary cell simultaneously.In this learning method, such as following formula
Shown in, for this elementary cell corresponding to this rolling condition, (i, j) by the method identical with preceding method
Calculate learning coefficient.Furthermore, (i j) shows the coordinate of this elementary cell in layered sheet.
Znew (i, j)=Zuse (i, j) * (1α)+Zcur (i, j) * α
Here, Znew (i, j): the learning coefficient updated value of this elementary cell
Zcur (i, j): the learning coefficient (instantaneous value) of this elementary cell
Zuse (i, j): the learning coefficient (previous value) of this elementary cell
The learning coefficient of α: this elementary cell more new gain (filter time constant)
For adjacent elementary cell, (p q), calculates learning coefficient with following formula.Furthermore, (p, q) table
Show the coordinate of elementary cell adjacent in layered sheet, wherein comprise (i1, j), (i, j1), (i+1, j)
And (i+1, j).
Znew (p, q)=Zuse (p, q) * (1α ')+Zcur (i, j) * α '
Here, Znew (p, q): the learning coefficient updated value of adjacent elementary cell
Zcur (i, j): the learning coefficient (instantaneous value) of this material elementary cell
Zuse (p, q): the learning coefficient (previous value) of adjacent elementary cell
α ': the learning coefficient of adjacent elementary cell more new gain (filter time constant)
According to this learning method, it is possible to the study of adjacent elementary cell can be made with the fewest roll mill
Coefficient is saturated.But, although adjacent elementary cell is learning, but from this elementary cell slightly away from
The learning coefficient of elementary cell can not be updated.That is, this learning method is used, about reducing roll mill meeting
Limited effect can only be obtained.It addition, when the learning coefficient of this elementary cell are unstable, it is big to update every time
When amplitude variation moves, the learning coefficient of adjacent elementary cell also can be affected by.
Further, employ layered sheet learns the other problem outside control there is also the problems referred to above.This is asked
Topic is, owing to elementary cell is segmented so being difficult to follow the change that rolling mill practice occurs through time change
Change.When under the rolling condition corresponding to certain elementary cell temporarily without rolling, this is temporarily without rolling during rolling
Technique is likely to occur change.The change of the rolling mill practice alleged at this, had both comprised the temperature etc. in hot rolling
The change etc. of level change actively, also comprises the passive change such as deterioration of equipment.If rolling mill practice becomes
Changing, the most real learning coefficient also produce change.Therefore, learning coefficient recorded in layered sheet
When updated value maintains old value constant, this value is the most incorrect.When inappropriate learning coefficient are applied to
During model formation, the error that model predication value is comprised can become big, can reduce the precision of machine settings.
As it has been described above, the study employing existing general layered sheet used controls there is variety of problems.Another
Aspect, the study employing layered sheet controls, and there is also following patent documentation 1,2 and is opened the scheme shown.
Scheme disclosed in patent documentation 1 is, by included in the deviation between model and phenomenon because of the time
The method that the deviation that sequentiality variation causes separates as time series learning coefficient.By this rolling condition
Corresponding time coefficient is divided into the coefficient and depend on time change of studying in groups depending on rolling phenomenon
Time series learning coefficient calculate, then carry out correction model predictive value based on both learning coefficient,
Thus improve the precision of model predication value.Specifically, based on the study system calculated according to model error
The instantaneous value of number and the use value of the learning coefficient relevant to the rolling phenomenon under this rolling condition, with smooth
Wave filter calculates the updated value of time series learning coefficient.Then, instantaneous values based on learning coefficient deduct
Value remaining after the updated value of time series learning coefficient, calculates and this rolling condition with smoothing filter
Under the updated value of the relevant learning coefficient of rolling phenomenon.
According to the learning method proposed in patent documentation 1, it is possible to by because of the time series of process production line
The deviation changed and produce extracts as time series learning coefficient.But, at this learning method
In, each elementary cell is also calculated by the coefficient that studies in groups depending on rolling phenomenon.Cause
This, even if the period rolling mill practice owing to wouldn't carry out rolling at certain rolling condition changes, and make this roll
Model under the conditions of system and the deviation between phenomenon create variation, and this variation is not the most reflected into this rolling condition
The learning coefficient that recorded of elementary cell.Thus, after rolling mill practice changes under this rolling condition the
Once in rolling, it is impossible to obtain the deviation between model and phenomenon and obtained the learning coefficient of appropriate correction.
Technical scheme disclosed in patent documentation 2 is, for each batch corresponding to the differentiation of rolling stock
The method of the study item of secondary (elementary cell) respectively record cast formula, in the method, batch whenever
Change, just the proficiency of the study item corresponding to next batch is judged, when proficiency is less than
During benchmark, the study item of other the batch that proficiency is high is used to correct the study item of next batch.Tool
For body, whether the study number of times of the most a batch of study item is more than benchmark number of times, and next group
On the basis of whether the standard deviation of nearest predetermined time of secondary corresponding study item is more than reference value, sentence
Fix the proficiency of a batch of study item.Then, when proficiency less than benchmark time, based on table with should
The study item of each batch that batch is adjacent and according to the correction coefficient depending on its proficiency, with smoothing filter
Ripple device carries out the correction of the study item of this batch.
According to the learning method proposed in patent documentation 2, though the study item greenly of next batch, also
The study item of the most skilled adjacent batch can be used to optimize the study item of the batch of greenly.But
It is that the study item of adjacent batch is not necessarily skilled than the study item of next batch.The most a batch of study
Greenly, adjacent batch study item also greenly time, must be based on greenly about next batch
Study item is set calculating.It addition, when the study item of adjacent batch is unstable, update it the most every time
When value the most significantly changes, the study item using its next batch corrected also can be unstable.Further, when by
Rolling mill practice in the period temporarily not carrying out rolling changes and makes the precision of the study item of adjacent batch
When there is deterioration in time, use the study item learning the next batch that item is corrected of adjacent batch
Precision also can stepdown.
Prior art literature
Patent documentation
Patent documentation 1: No. 2839746 publications of Japanese Patent No.
Patent documentation 2: Japanese Patent Publication 2000263110 publication
Summary of the invention
The technical problem to be solved
Even as it has been described above, the learning method proposed in patent documentation 1,2, can not solve to own
The problems referred to above.That is, the problem (problem of saturation) when the saturation of learning coefficient is low, learning coefficient
Unstable, problem (problem of degree of stability) when changing big, and owing to the period rolled wouldn't be carried out
Rolling mill practice change and the problem that makes learning coefficient wear (recency (recency) asks
Topic), can not say at present and obtain sufficient solution.
The present invention proposes in view of the above problems, it is intended that to setting under each rolling condition
Devise a stratagem calculate used by the learning coefficient of model formation carry out in the learning control device learning and managing, can be
Obtain saturation under each rolling condition, learning coefficient that degree of stability, recency are the most all met.
Solve the technical scheme that technical problem is used
Multiple elementary cells that learning control device involved in the present invention has by distinguishing rolling condition are constituted
Data base, utilize the study of the model formation that data base used to the setup algorithm managing rolling mill practice
Coefficient, in order to reach abovementioned purpose, its composition is as follows.
Learning control device involved in the present invention includes that instantaneous value calculates record unit and updated value calculates note
Record unit, is used as learning coefficient calculate and record the unit to data base.Instantaneous value calculates record
Unit is configured to, based on the actual value measured in rolling mill practice with the actual value calculated with model formation again
Deviation between value of calculation calculates the instantaneous value of learning coefficient, then the most single to this rolling condition
The instantaneous value calculated together is carried out record with the study date by unit.The upper of record unit is calculated by instantaneous value
State composition, for the historical information of each elementary cell record instantaneous value in data base.Updated value calculates
Record unit is configured to, the instantaneous value of the learning coefficient calculated based on instantaneous value computing unit and study system
Number previous value under this rolling condition calculates the updated value of learning coefficient, then to this rolling condition phase
The updated value calculated together is carried out record with the study date by the elementary cell answered.Note is calculated by updated value
The said structure of record unit, for the historical information of each elementary cell record updated value in data base.
Furthermore, when the rolling condition distinguished by elementary cell is two, it is also possible to by instantaneous value and more
New value respective historical information layered sheet is managed.I.e., it is also possible to be configured to data base include altogether
With first layer table and second layered sheet of elementary cell, first layer table record instantaneous value is calculated record
The instantaneous value of unit, calculates the updated value of record unit to the second layered sheet record updated value.
Learning control device involved in the present invention includes: the recency of the recency evaluating learning coefficient is commented
Valency unit, evaluate learning coefficient saturation saturation evaluation unit and evaluate learning coefficient steady
The stability evaluation unit of fixed degree.Recency evaluation unit is configured to, based on the renewal of record in data base
The historical information of value, for the recency each respectively evaluating learning coefficient of multiple elementary cells.Saturated
Degree evaluation unit is configured to, based on the historical information of updated value of record in data base, for multiple substantially
Each of unit evaluates the saturation of learning coefficient respectively.Stability evaluation unit is configured to, based on number
According to the historical information of the instantaneous value of record in storehouse, each for multiple elementary cells evaluates study respectively
The degree of stability of coefficient.
Furthermore, as each historical information of instantaneous value and updated value, it is also possible to recency, saturation
And degree of stability respective rating result layered sheet is managed.In this case, data base is permissible
It is configured to, also includes sharing the 3rd to the 5th layering of elementary cell with first layer table and the second layered sheet
Table.Then, the recency evaluation result to the 3rd layered sheet record recency evaluation unit, to the 4th point
The saturation evaluation result of layer table record saturation evaluation unit, evaluates the 5th layered sheet recording stability
The stability evaluation result of unit.
Learning control device involved in the present invention also includes: neighbouring elementary cell selectes unit, presumed value
Computing unit, and use value determining means.Neighbouring elementary cell is selected unit and is configured to, from recently
Degree, saturation and the respective evaluation result of degree of stability meet in the elementary cell of benchmark, are selected in roll
Condition processed is that the space middlerange downstream roll condition corresponding target elementary cell space length of coordinate axes is near
The neighbouring elementary cell of predetermined number.Presumed value computing unit is configured to, and determines the predetermined number selected
Neighbouring elementary cell in the representative value of learning coefficient, and coordinate based on object elementary cell and selected
The coordinate of the neighbouring elementary cell of fixed predetermined number and representative value, calculate object by polynomial interopolation
The presumed value of the learning coefficient in elementary cell.Additionally, use value determining means is configured to, utilize presumption
The latest update value of the learning coefficient in object elementary cell is corrected by value, and is set to downstream roll bar
The use value of the learning coefficient under part.
Preferably, use value determining means is configured to, by the weighted average of latest update value Yu presumed value
Calculate use value.Evaluation according to recency, saturation and degree of stability to object elementary cell
The weight of the highest then latest update value of result is the biggest, and the weight of the lowest then presumed value of evaluation result is the biggest
Mode, changes average weighted weight coefficient according to evaluation result.
Invention effect
Learning control device according to the present invention configured as described above, it is possible to downstream roll material is rolled
The learning coefficient that the application of condition processed corresponding elementary cell is uptodate, saturation is high, degree of stability is high.By
This, it is possible to increase the precision of the model predication value in setup control, and then improve the precision of end product quality with
And the stability of operation.
Accompanying drawing explanation
Fig. 1 is structure and the place of the learning control device illustrating the rolling mill practice according to embodiment of the present invention
The figure of reason flow process.
Fig. 2 is the figure of the computational methods of the presumed value that learning coefficient are described.
Detailed description of the invention
Hereafter, referring to the drawings, embodiments of the present invention are described.
The structure of the learning control device of rolling mill practice involved by present embodiment shown in Fig. 1 and process
Flow process.This learning control device 1 includes learning device 10, storage device 20 and setup algorithm device
30.Learning device 10 is that real data based on rolling mill practice learns model public affairs used in setup algorithm
The device of the learning coefficient of formula.Storage device 20 is by the learning coefficient of model formation and associated there
Information stores the device to data base.This data base is made up of the multiple elementary cells distinguishing rolling condition.
More specifically, it is made up of five layered sheets 21,22,23,24,25 of shared elementary cell.Again
Person, shared elementary cell refers to, if the coordinate of elementary cell is identical between table, corresponding rolling condition is also
Identical, the elementary cell that coordinate is identical is interrelated between table.In the present embodiment, rolling condition is
The thickness of rolling stock and width these two, each layered sheet uses with thickness and width the table as axle.Work as thickness
Degree number be " i ", width number be " j " time, (i j) is the seat of elementary cell in each layered sheet
Mark.Setup algorithm device 30 is to come computation model predictive value, base by the learning coefficient stored in data base
The device of the machine settings of rolling device is determined in model predication value.
First, learning device 10 is described in detail.Processing unit 11 real data based on rolling mill practice is entered
The actual value of row model predication value calculates again.The reality that processing unit 12 is calculated based on processing unit 11
It is worth again value of calculation and the model error that is included between the actual value in real data to calculate learning coefficient
Instantaneous value.The instantaneous value of the learning coefficient that process portion 13 is calculated by making processing unit 12 is by smooth
Wave filter calculates the updated value of learning coefficient.The instantaneous value of learning coefficient and the computational methods of updated value with
The content that the existing study of explanation controls in " background technology " is identical.
This learning control device is after the rolling of a rolling stock completes, by the instantaneous value of learning coefficient
The corresponding elementary cell of rolling condition to this material is recorded respectively with updated value.Specifically, study system
The instantaneous value of number is recorded to first layer table 21.Processing unit 12 is from constituting first layer table more than 21
Individual elementary cell selects the corresponding elementary cell of this rolling condition.Then, by selected basic
The instantaneous value record in chronological order that unit will calculate, makes the historical information of instantaneous value store to first layer
Table 21.More specifically, in first layer table 21, each elementary cell is rolled about past N root
Prepared material is by stored along for historical information with the study date for the instantaneous value of learning coefficient.Storage information from
New information open numbering.That is, using the information that newly stores as the 1st article, by ith during each supplement information
Bar information renumbers as i+1 bar.If it addition, having stored N root rolling stock during supplement information
Information, then abandon the N article the oldest information, the N1 article information is renumberd and is the N article.
The updated value of learning coefficient is recorded to the second layered sheet 22.Processing unit 13 is from constituting the second layering
Multiple elementary cells of table 22 select the corresponding elementary cell of this rolling condition.Then, by selected
The updated value record in chronological order that the elementary cell selected will calculate, make the historical information of updated value store to
Second layered sheet 22.More specifically, in the second layered sheet 22, to each elementary cell about mistake
Go M rolling stock by stored along for historical information with the study date for the updated value of learning coefficient.Storage
Information in the same manner as the historical information about the instantaneous value of learning coefficient, from new information open numbering.
That is, using the information that newly stores as the 1st article, ith information is renumberd into i+1 bar, if
Store the information of M rolling stock, then abandon the M article the oldest information, by the M1 article information weight
The most numbered the M article.
This learning control device the updated value of renewal learning coefficient and the respective historical information of instantaneous value it
After, carry out three evaluations about learning coefficient.First evaluation is the evaluation of the recency of learning coefficient,
All of elementary cell is carried out by this evaluation.Second evaluation is the evaluation of the saturation of learning coefficient, this
Evaluate and carry out historical record in this elementary cell corresponding to the rolling condition of this material, i.e. this process
The elementary cell updated is carried out.3rd evaluation is the evaluation of the degree of stability of learning coefficient, and this evaluates this
Elementary cell is carried out.
The recency evaluation of learning coefficient is carried out by processing unit 14.Processing unit 14 is based on the second layering
The historical information of the updated value of the learning coefficient that table 22 is stored, evaluates each base of the second layered sheet 22
The updated value of the learning coefficient of this unit is new.If the renewal of each elementary cell is after one fixes the date
Carry out, then can determine whether that the updated value of these learning coefficient is new.Following formula judges for this.
ε (k, l) ＜ ε time
Here, (k, l): the coordinate (wherein, k, l are the combinations of all rolling conditions) of elementary cell
ε (k, l): the update date of the learning coefficient of this material and elementary cell (k, update date l)
Between time difference
The determinating reference value of ε time: recency
When meeting above formula, processing unit 14 judges elementary cell (k, the renewal of learning coefficient l) stored
Value is uptodate, and (k is l) 1 to make FCZtime.When being unsatisfactory for above formula, processing unit 14 judges base
(k, the updated value of learning coefficient l) stored is not uptodate to this unit, and (k is l) 0 to make FCZtime.
Furthermore, FCZtime (k, l) is the numerical value of recency evaluation result of the updated value representing learning coefficient, if
Value is 1 and is meant to be uptodate, if value is 0, means it is not uptodate.Processing unit 14 is to institute
Elementary cell is had to carry out judgement based on above formula.
By recency evaluation of estimate FCZtime that processing unit 14 is the decision of each elementary cell, (k l) is recorded
Corresponding elementary cell to the 3rd layered sheet 23.
The saturation evaluation of learning coefficient is carried out by processing unit 15.Processing unit 15 is based on the second layering
The historical information of the updated value of the learning coefficient that table 22 is stored, evaluates its study system for this elementary cell
The updated value of number is saturated.About the 1st article of updated value of learning coefficient to the m article (m≤M)
Value, the value of the m article is deducted the value of the m1 article, the value of ith article deduct the ith1 article value, the 2nd
Article value deduct the value of the 1st article to calculate the variable quantity of the updated value of each learning coefficient, if updated value
Variable quantity to 0 convergence, then can determine whether that the updated value of the learning coefficient of this elementary cell is saturated.Under
Formula judges for this.
CZmdf (r, s) ＜ ε mdf
Here, (r, s): the coordinate of this elementary cell
CZmdf (r, s): the meansigma methods of the variable quantity of the updated value of the learning coefficient of this elementary cell
The determinating reference value of ε mdf: saturation
(r s) is calculated meansigma methods CZmdf of the variable quantity of updated value by following formula.
Here, Zmdf (the storage value of ith of the updated value of r, s) (i): learning coefficient
When meeting above formula, processing unit 15 judges this elementary cell, and (r, learning coefficient s) stored are more
New value is saturated, and (r is s) 1 to make FCZmdf.When being unsatisfactory for above formula, processing unit 15 judges
(r, the updated value of learning coefficient s) stored is unsaturated, and (r is s) 0 to make FCZmdf for this elementary cell.
It addition, as M ＜ m, processing unit 15 makes FCZmdf, and (r is s) 0.Furthermore, (r s) is FCZmdf
Represent the numerical value of the saturation evaluation result of the updated value of learning coefficient, if value is 1, be meant to be saturated
, if value is 0, mean unsaturation.
(r s) is recorded to the 4th layered sheet 24 saturation evaluation of estimate FCZmdf determined by processing unit 15
This elementary cell (r, s).
The stability evaluation of learning coefficient is carried out by processing unit 16.Processing unit 16 is based on first layer
The historical information of the instantaneous value of the learning coefficient that table 21 is stored, evaluates its study system for this elementary cell
The instantaneous value of number is stable.About the 1st article of instantaneous value of learning coefficient to nth bar (n≤N)
Value, if its amount of change is little, then can determine whether that the instantaneous value of the learning coefficient of this unit is stable.Following formula is used
Judge in this.
CZcur (r, s) ＜ ε cur
Here, (r, s): the coordinate of this elementary cell
CZcur (r, s): the standard deviation of the variable quantity of the instantaneous value of the learning coefficient of this elementary cell
The determinating reference value of ε cur: degree of stability
(r s) is calculated standard deviation CZcur of the variable quantity of instantaneous value by following formula.
Here, Zcur (the storage value of ith of the instantaneous value of r, s) (i): learning coefficient
When meeting above formula, processing unit 16 judges this elementary cell (r, the wink of learning coefficient s) stored
Duration is stable, and (r is s) 1 to make FCZcur.When being unsatisfactory for above formula, processing unit 16 judges
(r, the instantaneous value of learning coefficient s) stored is unstable to this elementary cell, and (r s) is to make FCZcur
0.It addition, as N ＜ n, processing unit 16 makes FCZcur, and (r is s) 0.Furthermore, FCZcur (r, s)
It is the numerical value of the stability evaluation result of the instantaneous value representing learning coefficient, if value is 1, is meant to be steady
Fixed, if value is 0, it is meant to be instability.
The stability evaluating determined by processing unit 16 is worth FCZcur, and (r s) is recorded to the 5th layered sheet 25
This elementary cell (r, s).
Processed, about the corresponding elementary cell of rolling condition of this material, the wink of learning coefficient by above
Duration and updated value are stored.Further, about all elementary cells, the recency of learning coefficient obtains
Evaluate, it addition, about this elementary cell, saturation and the degree of stability of learning coefficient are evaluated, these
Evaluation of estimate is stored to the elementary cell of correspondence the most again.
Then, setup algorithm device 30 is described in detail.The setup algorithm device 30 rolling bar to next material
The corresponding elementary cell of part, i.e. target as the model prediction target employing model formation are the most single
Unit, utilizes two kinds of methods to try to achieve learning coefficient, and answers when calculating the rolling of next material according to these learning coefficient
The use value of learning coefficient.A kind of method is, the learning coefficient stored from the second layered sheet 22
The historical information of updated value retrieve and read out the corresponding elementary cell of rolling condition of next material
The method of new updated value.Place's reason processing unit 33 according to the method is carried out.Here, by next material
The coordinate of elementary cell of material be designated as (t, u), the updated value of learning coefficient that will be read by processing unit 33
Be designated as Zmod (t, u).
Another kind of method is, uses the learning coefficient of the neighbouring elementary cell of the elementary cell of next material to calculate
The method going out the presumed value of learning coefficient in the elementary cell of next material.Place's reason according to the method
Processing unit 31,32 is carried out.Specifically, first, the most single from next material of processing unit 31
The selected elementary cell meeting predetermined condition in elementary cell near unit.Predetermined condition is three below bar
Part:
1) use newer data to obtain study to update.
2) updated value of learning coefficient is saturated.
3) instantaneous value of learning coefficient is stable.
Processing unit 31 selects in the 3rd layered sheet the 23, the 4th layered sheet 24 and the 5th layered sheet 25
Evaluation result is all judged as the elementary cell of 1.Then, from selected elementary cell, in order
Extract three on space length away from the corresponding elementary cell of next material (t, u) nearer elementary cell.
Here, three selected elementary cells are denoted as elementary cell A (ta, ua), elementary cell B
(tb, ub), elementary cell C (tc, uc).
Processing unit 32 is to elementary cell A (ta, ua) selected by processing unit 31, elementary cell B
(tb, ub), elementary cell C (tc, uc), history based on the instantaneous value stored in first layer table 21
Information, calculates and represents the representative value of instantaneous value uptodate for K (such as, K uptodate instantaneous value is average
Value or intermediate value or K2 instantaneous value after instantaneous value uptodate for K removes maximum and minima
Meansigma methods or intermediate value etc.).Here, the representative value of elementary cell A (ta, ua) is denoted as
Zestm (ta, ua), the representative value of elementary cell B (tb, ub) is denoted as Zestm (tb, ub), elementary cell C
The representative value of (tc, uc) is denoted as Zestm (tc, uc).
Processing unit 32 representative value of these three learning coefficient determines selected elementary cell A
(ta, ua), elementary cell B (tb, ub), elementary cell C (tc, uc) each coordinate representative value each with it it
Between set up an order polynomial.I.e., as shown in Figure 2, using as the width of rolling condition and thickness and
Learning coefficient are in the three dimensions of each coordinate axes, ask and are respectively by coordinate
(ta, ua, Zestm (ta, ua)), (tb, ub, Zestm (tb, ub)), (tc, uc, Zestm (tc, uc))
The plane of three points.This plane can be tried to achieve by meeting following formula.
Zestm (t, u)=a*t+b*u+c
Here, coefficient a, b, c of the expression formula of abovementioned plane can calculate by meeting following condition.
Zestm (ta, ua)=a*ta+b*ua+c
Zestm (tb, ub)=a*tb+b*ub+c
Zestm (tc, uc)=a*tc+b*uc+c
In the plane determined by above formula, each benchmark of the recency of learning coefficient, saturation and degree of stability obtains
To meeting, this point can be confirmed by experience.Processing unit 32 is the most single to next material
(t u), by using above formula to carry out polynomial interopolation, thus calculates learning coefficient, thus, calculates in unit
Go out to meet the presumed value of the learning coefficient of recency, saturation and degree of stability.Here, will be by processing unit
The presumed value of 34 learning coefficient calculated be designated as Zestm (t, u).
By above step, about the corresponding elementary cell of next material (t, u), it is thus achieved that learning coefficient
Updated value Zmod (t, u) and learning coefficient presumed value Zestm (t, u).Processing unit 35 is based on the two
Learning coefficient calculate the use value of the learning coefficient under the rolling condition of next material.More specifically,
By presumed value Zestm according to following average weighted expression formula learning coefficient, (t u) corrects
(t u), thus determines the use value of learning coefficient to updated value Zmod of habit coefficient.
Zuse (t, u)=β * Zmod (t, u)+(1β) * Zestm (t, u)
Here, Zuse (t, u): the use value of learning coefficient.
β: weight coefficient (0≤β≤1)
If (t, recency u), saturation, degree of stability are good, then process for updated value Zmod of learning coefficient
Unit 35 carrys out updated value Zmod to learning coefficient by the value increasing weight coefficient β, and (t u) adds
Power.On the contrary, if (t, recency u), saturation, degree of stability are good for estimated value Zestm of learning coefficient
Good, then (t u) adds to carry out estimated value Zestm to learning coefficient by the value of reduction weight coefficient β
Power.The example of the determining method of weight coefficient β is as follows.
β=1.0 are (when updated value Zmod (t, when u) meeting recency, saturation, degree of stability) simultaneously
β=0.6 is (as updated value Zmod (t, when u) meeting two in recency, saturation, degree of stability)
β=0.4 is (as updated value Zmod (t, when u) meeting in recency, saturation, degree of stability)
(when updated value Zmod, (t u) does not meets any one in recency, saturation, degree of stability in β=0.0
Time)
It is public that processing unit 36 carrys out calibration model according to the use value of the learning coefficient calculated by processing unit 35
Formula.Thus, try hard to improve prediction computational accuracy based on model formation.Then, the mould after correction is used
Type formula is set calculating, to determine the machine settings of rolling device.By according to depending on thus
Setting value scans each machine, it is possible to improve the precision of end product quality and the stability of operation.
Above embodiments of the present invention are illustrated.But, the present invention is not limited to abovementioned reality
Execute mode, can be variously modified on the premise of without departing from present inventive concept.Such as, by the most single
The rolling condition that unit is distinguished can be more than two.Although in the abovedescribed embodiment rolling condition be width and
Thickness these two but it also may make rolling condition be more than three, pass through the multidimensional database more than threedimensional
Manage learning coefficient.If rolling condition is three, then can be by the threedimensional with rolling condition as axle
Space extracts from the elementary cell meeting recency, saturation and degree of stability in order four
Away from the elementary cell that object elementary cell is nearer on space length, then calculate object base by polynomial interopolation
The presumed value of the learning coefficient in this unit.
Furthermore, in the abovedescribed embodiment, processing unit 12 is equivalent to instantaneous value and calculates record unit, place
Reason unit 13 is equivalent to updated value and calculates record unit, and processing unit 14 is equivalent to recency and evaluates single
Unit, processing unit 15 is equivalent to saturation evaluation unit, and processing unit 16 is equivalent to stability evaluation list
Unit.It addition, processing unit 31 is equivalent to neighbouring elementary cell selectes unit, processing unit 32 is equivalent to
Presumed value computing unit, processing unit 34,35 is equivalent to use value determining means.Constituting this study control
In the memorizer of the computer of device 1 processed, the routine corresponding to each processing unit is entered as a program
Row storage.By being performed this program by the processor of computer, thus computer plays each processing unit
Function.
Label declaration
1... learning control device
10... learning device
20... device (data base) is stored
21,22,23,24,25... layered sheet
30... setup algorithm device
Claims (4)
1. a learning control device for rolling process, has by the multiple elementary cells distinguishing rolling condition
The data base constituted, utilizes described data base to manage the model used in the setup algorithm of rolling mill practice
The learning coefficient of formula, the learning control device of this rolling process is characterised by, including:
Instantaneous value calculate record unit, its based on the actual value measured in described rolling mill practice with utilize institute
State the deviation between the actual value value of calculation again that model formation calculates to calculate the instantaneous of described learning coefficient
Value, described instantaneous value is together carried out record with the study date by elementary cell corresponding to this rolling condition；
Updated value calculates record unit, its based on described instantaneous value and described learning coefficient at this rolling condition
Under previous value calculate the updated value of described learning coefficient, elementary cell corresponding to this rolling condition
Described updated value is together carried out record with the study date；
Recency evaluation unit, it is based on the historical information of the described updated value of storage in described data base,
The recency each respectively evaluating described learning coefficient to multiple described elementary cells；
Saturation evaluation unit, it is based on the historical information of the described updated value of storage in described data base,
The saturation each respectively evaluating described learning coefficient to multiple described elementary cells；
Stability evaluation unit, it is based on the historical information of the described instantaneous value of storage in described data base,
The degree of stability each respectively evaluating described learning coefficient to multiple described elementary cells；
Neighbouring elementary cell selectes unit, and it is from described recency, described saturation and described degree of stability
Each evaluation result meets in the elementary cell of benchmark, is selected in the space middlerange with rolling condition as coordinate axes
The neighbouring elementary cell of the predetermined number that downstream roll condition corresponding object elementary cell space length is near；
Presumed value computing unit, the institute in the neighbouring elementary cell of the described predetermined number that its decision is selected
State the respective representative value of learning coefficient, and coordinate based on described object elementary cell and described in being selected
The coordinate of the neighbouring elementary cell of predetermined number and described representative value, calculate described by polynomial interopolation
The presumed value of the described learning coefficient in object elementary cell；And
Use value determining means, it utilizes described presumed value to the described study in described object elementary cell
The latest update value of coefficient is corrected, and is set to the use of described learning coefficient under the conditions of downstream roll
Value.
2. the learning control device of rolling process as claimed in claim 1, it is characterised in that
Described use value determining means includes:
The list of described use value is calculated by the weighted average of described latest update value Yu described presumed value
Unit；And
According to commenting the described recency of described object elementary cell, described saturation and described degree of stability
The weight of valency result the most described the highest latest update value is the biggest, the most described the lowest presumption of described evaluation result
The biggest mode of weight of value, to change described average weighted weight coefficient according to described evaluation result
Unit.
3. the learning control device of rolling process as claimed in claim 1 or 2, it is characterised in that
Described data base includes first layer table and the second layered sheet sharing elementary cell,
Described instantaneous value calculates record unit and is configured to, by described instantaneous value record to described first layer
Table,
Described updated value calculates record unit and is configured to, by described updated value record to described second layering
Table.
4. the learning control device of rolling process as claimed in claim 3, it is characterised in that
Described data base also includes sharing elementary cell with described first layer table and described second layered sheet
3rd layered sheet, the 4th layered sheet and the 5th layered sheet,
Described recency evaluation unit is configured to, by the evaluation result record of described recency to the described 3rd
Layered sheet；
Described saturation evaluation unit is configured to, by the evaluation result record of described saturation to the described 4th
Layered sheet；And
Described stability evaluation unit is configured to, by the evaluation result record of described degree of stability to the described 5th
Layered sheet.
Priority Applications (1)
Application Number  Priority Date  Filing Date  Title 

PCT/JP2014/053622 WO2015122010A1 (en)  20140217  20140217  Rolling process learning control device 
Publications (2)
Publication Number  Publication Date 

CN106030424A true CN106030424A (en)  20161012 
CN106030424B CN106030424B (en)  20190315 
Family
ID=53799772
Family Applications (1)
Application Number  Title  Priority Date  Filing Date 

CN201480075773.4A Active CN106030424B (en)  20140217  20140217  The learning control device of rolling mill practice 
Country Status (5)
Country  Link 

US (1)  US10124381B2 (en) 
JP (1)  JP6233423B2 (en) 
KR (1)  KR101832653B1 (en) 
CN (1)  CN106030424B (en) 
WO (1)  WO2015122010A1 (en) 
Cited By (4)
Publication number  Priority date  Publication date  Assignee  Title 

CN106777594A (en) *  20161201  20170531  武汉科技大学  A kind of selflearning method of the operation of rolling Mathematical Modeling based on continuous curve surface 
CN108687137A (en) *  20170410  20181023  株式会社日立制作所  Complete equipment control device, rolling mill control apparatus, control method and storage medium 
CN110918651A (en) *  20191206  20200327  东北大学  Steel cold rolling finishing production raw material distribution method based on multiclassification support vector machine 
CN111587156A (en) *  20181219  20200825  东芝三菱电机产业系统株式会社  Learning control device for rolling process 
Citations (8)
Publication number  Priority date  Publication date  Assignee  Title 

JP2000263110A (en) *  19990310  20000926  Toshiba Corp  Learning apparatus for rolling control model 
CN101648208A (en) *  20090826  20100217  苏州有色金属研究院有限公司  Rolling procedure making method of aluminium singlestand coldrolling machine 
US20100292825A1 (en) *  20060809  20101118  Auckland Uniservices Limited  Process control of an industrial plant 
CN102371279A (en) *  20100826  20120314  宝山钢铁股份有限公司  Adaptive control method for increasing thickness precision of finishrolled band steel by utilizing roll gap 
CN102513351A (en) *  20111224  20120627  东北大学  Rolling method and device for strip steel tandem cold rolling 
CN102654749A (en) *  20110301  20120905  东芝三菱电机产业系统株式会社  Learning coefficient control device 
CN102728624A (en) *  20110413  20121017  宝山钢铁股份有限公司  Method for setting load distribution of finish rolling band steel 
CN103018604A (en) *  20121207  20130403  浙江省电力公司电力科学研究院  Assessment method of electricity grid lightning strike risk and device thereof 
Family Cites Families (9)
Publication number  Priority date  Publication date  Assignee  Title 

JP2839746B2 (en) *  19910617  19981216  株式会社神戸製鋼所  Learning control method in process line 
JPH07200005A (en) *  19931228  19950804  Mitsubishi Electric Corp  Learning control method 
JP3508245B2 (en) *  19941104  20040322  Ｊｆｅスチール株式会社  Estimation method of rolling conditions using rolling condition estimation model 
JPH1031505A (en) *  19960716  19980203  Mitsubishi Electric Corp  Learning control method for process line 
JP4179709B2 (en) *  19990715  20081112  東芝三菱電機産業システム株式会社  Learning control device for reverse rolling mill 
JP2009116759A (en)  20071109  20090528  Jfe Steel Corp  Method and device for learning control model in process line, and production method of steel plate 
CN104303114B (en) *  20120516  20170616  东芝三菱电机产业系统株式会社  The learning device and learning method of setup algorithm system 
JP5835483B2 (en) *  20120702  20151224  東芝三菱電機産業システム株式会社  Temperature control device 
JP5939175B2 (en)  20130219  20160622  東芝三菱電機産業システム株式会社  Learning control device for rolling process 

2014
 20140217 CN CN201480075773.4A patent/CN106030424B/en active Active
 20140217 KR KR1020167025045A patent/KR101832653B1/en active IP Right Grant
 20140217 JP JP2015562674A patent/JP6233423B2/en active Active
 20140217 WO PCT/JP2014/053622 patent/WO2015122010A1/en active Application Filing
 20140217 US US15/119,313 patent/US10124381B2/en active Active
Patent Citations (8)
Publication number  Priority date  Publication date  Assignee  Title 

JP2000263110A (en) *  19990310  20000926  Toshiba Corp  Learning apparatus for rolling control model 
US20100292825A1 (en) *  20060809  20101118  Auckland Uniservices Limited  Process control of an industrial plant 
CN101648208A (en) *  20090826  20100217  苏州有色金属研究院有限公司  Rolling procedure making method of aluminium singlestand coldrolling machine 
CN102371279A (en) *  20100826  20120314  宝山钢铁股份有限公司  Adaptive control method for increasing thickness precision of finishrolled band steel by utilizing roll gap 
CN102654749A (en) *  20110301  20120905  东芝三菱电机产业系统株式会社  Learning coefficient control device 
CN102728624A (en) *  20110413  20121017  宝山钢铁股份有限公司  Method for setting load distribution of finish rolling band steel 
CN102513351A (en) *  20111224  20120627  东北大学  Rolling method and device for strip steel tandem cold rolling 
CN103018604A (en) *  20121207  20130403  浙江省电力公司电力科学研究院  Assessment method of electricity grid lightning strike risk and device thereof 
Cited By (5)
Publication number  Priority date  Publication date  Assignee  Title 

CN106777594A (en) *  20161201  20170531  武汉科技大学  A kind of selflearning method of the operation of rolling Mathematical Modeling based on continuous curve surface 
CN106777594B (en) *  20161201  20200117  武汉科技大学  Selflearning method of rolling process mathematical model based on continuous curved surface 
CN108687137A (en) *  20170410  20181023  株式会社日立制作所  Complete equipment control device, rolling mill control apparatus, control method and storage medium 
CN111587156A (en) *  20181219  20200825  东芝三菱电机产业系统株式会社  Learning control device for rolling process 
CN110918651A (en) *  20191206  20200327  东北大学  Steel cold rolling finishing production raw material distribution method based on multiclassification support vector machine 
Also Published As
Publication number  Publication date 

JP6233423B2 (en)  20171122 
US10124381B2 (en)  20181113 
KR101832653B1 (en)  20180226 
KR20160122208A (en)  20161021 
US20170008054A1 (en)  20170112 
JPWO2015122010A1 (en)  20170330 
CN106030424B (en)  20190315 
WO2015122010A1 (en)  20150820 
Similar Documents
Publication  Publication Date  Title 

CN106030424A (en)  Rolling process learning control device  
CN104415972B (en)  A kind of comprehensive selfadaptation control method of rough rolling strip steel width  
KR101011546B1 (en)  Prediction formula making device and prediction formula making method  
CN104511481B (en)  A kind of Long travel control method of roughing width multiple target rolling  
CN102654749B (en)  Learning coefficient control device  
KR100186903B1 (en)  Method of controlling form of strip in rolling mill  
CN102636624A (en)  Method for soft measurement of alumina concentration in electrolyzer during aluminum electrolysis process  
JP5939175B2 (en)  Learning control device for rolling process  
TW200806984A (en)  Rolling line material quality prediction and control apparatus  
KR101889668B1 (en)  Rolling simulation device  
JP3223856B2 (en)  Rolling mill control method and rolling mill control device  
Havinga et al.  Estimating producttoproduct variations in metal forming using force measurements  
CN105478490B (en)  The control device and control method of tandem mill  
JP4179709B2 (en)  Learning control device for reverse rolling mill  
CN103962388A (en)  Rough rolled band steel width control method of hot continuous rolling mill based on multiple heating furnaces  
JP4696775B2 (en)  Plate width control method and apparatus  
CN100512988C (en)  Method and apparatus for controlling rolling operation  
CN105618490B (en)  A kind of side drop autocontrol method of cold rolling electric  
JP2020157327A (en)  Control method for outlet side temperature of finished steel sheet, control device therefor and production method for steel sheet  
CN103909096B (en)  A kind of determination method and system of operational factor of hotrolling mill  
KR100841888B1 (en)  Rolling line material quality prediction and control apparatus  
CN108067506B (en)  Medium and Heavy Plate Rolling passage dynamic becomes setting control method  
JPH0667706A (en)  Initializing method for control system  
CN113102516A (en)  Hot continuous rolling strip steel head width prediction method integrating rolling mechanism and deep learning  
JP5415872B2 (en)  Stratified table numerical arithmetic device and method, control device and control method and rolling system provided with the device and method 
Legal Events
Date  Code  Title  Description 

PB01  Publication  
C06  Publication  
SE01  Entry into force of request for substantive examination  
C10  Entry into substantive examination  
GR01  Patent grant  
GR01  Patent grant 