CN102896156A - Optimization method for hot rolled strip steel roll gap model - Google Patents

Abstract

The invention discloses an optimization method for a hot rolled strip steel roll gap model. According to the optimization method, a roll gap setting and calculating model structure is optimized, a roll gap zero drift compensation is increased, and a roll gap zero drift offset existent in a process of calculating a roll gap setting value is reduced; and meanwhile, a data processing method for roll gap self-learning is improved as follows: oscillation of roll gap self-learning is reduced by utilizing a roll gap self-learning data screening method for removing an extremal rolling force and an extremal roll gap at the same time and a combination method of short-time genetic learning and long-time genetic learning mechanisms. Due to the adoption of the optimization method, so that accuracy of the roll gap setting value and effectiveness of roll gap self-learning are guaranteed, calculation precision of the roll gap model is greatly improved, and thickness precision of a finished-product hot rolled strip steel is further improved.

Description

A kind of optimization method of hot-strip roll gap model

Technical field

The invention belongs to steel rolling automatic control technology field, related in particular to a kind of optimization method of hot-strip roll gap model.

Background technology

Thickness and precision becomes the focus of domestic and international metallurgy industry common concern as a kind of important quality index of weighing Strip.One of basic reason of belt steel thickness fluctuation is because fluctuation has occured mill roll-gap.So be the problem of roller gap with the main cause of steel generation thickness deviation, the setting accuracy of roll gap will directly have influence on the delivery gauge precision of product.

The mill spring curve is intermesh determination core foundation.Spring curve corresponding roll gap numerical value under zero pressure is exactly the roll gap at zero point of spring curve.Be not unalterable this zero point, can in the operation of rolling, drift about gradually, be difficult to realize accurate modeling and in real time calculating, the error of bringing can make roll gap null offset, existing gap Set Model is not considered roll gap null offset, so that there is deviation in fixed value of roller slit, reduced the setting accuracy of roll gap, can only lean on its error of self-learning method correction, can't be compensated by presetting in advance accurately, can occur the situation that thickness and precision descends after when the rolling working conditions change in front and back is larger, changing specification or roll change.

Model Self-Learning also is the important means that improves the model specification precision, and study will according to the variation of system mode, constantly utilize real time information, carry out the correction of model coefficient, learning process compares predicted value and actual value, calculates learning coefficient, and the setting that is used for next piece steel is calculated.Self study makes the continuous convergence actual value of setting value by the continuous correction of learning coefficient, thereby improves the setting accuracy of model.The roll gap self learning model that uses at present is to adopt exponential smoothing; but two problems of having chosen of coefficent of exponential smoothing exist; when coefficient is large, can cause the learning process vibration; this is to cause again study unusually slow a few hours; so be necessary on this basis roll gap self learning model learning method to be improved, thereby precision of assurance roll gap self study.

It is to have done a few thing for this respect that following several patent is arranged at present:

Provide a kind of control method with steel finishing mill roll gap such as patent publication No. for the invention of CN1483526A, it is by the accurate Calculation to finish rolling inlet temperature (being the temperature of intermediate blank head), solved and used Coil Box to batch the low problem of process control model head hit rate in the production process of intermediate blank, control thus the precision of finishing mill roll gap, shortcoming is the impact of only having considered the temperature pair roller slit die type in the operation of rolling, do not relate to other influence factor, but also just for the special circumstances that adopt Coil Box; Patent publication No. is that the invention of CN101829687A has then related to a kind of band steel finishing mill roll gap control method of eliminating influence of specification, and it just eliminates the rolling impact of specification by having increased the roll gap offset of dividing according to specification.By this method, can improve and change the thickness and precision of specification when rolling, shortcoming is the compensation that it just increases for changing the rolling caused error of specification; Also having patent publication No. is control method and the device that CN101927267A has mentioned a kind of cleaning between rolls of finish rolling strip steel, it is to adjust roll gap by calculating roll-force and roll torque, shortcoming is that list is started with from rolling force and moment, only considered rolling force and moment to the impact of roll gap, do not mentioned that other parameter is on the impact of its roll gap.

Summary of the invention

The technical problem to be solved in the present invention provides a kind of hot-strip roll gap model optimization method, solve roll gap Zero drift in main amplifier and the low problem of roll gap self learning model precision of bounce mold type in the operation of rolling, improve the setting accuracy of cleaning between rolls of finish rolling strip steel model with this, thereby reach the purpose that improves hot-strip finished product thickness precision.

The present invention realizes by following technical scheme:

A kind of optimization method of hot-strip roll gap model, comprise intermesh determination computation model and roll gap self study data processing method, in the roll gap computation model, added the roll gap drift compensating, reduced the caused roll gap calculation deviation of roll gap null offset, optimize the gap Set Model structure, thereby improved the setting computational accuracy of hot-strip roll gap model;

Intermesh determination computation model after the optimization is following form:

$\mathrm{Gap}=\mathrm{Fh}-\frac{(S-S_j)*\mathrm{Gwid}}{100*\mathrm{WCX}}+\frac{(S_0-S_j)*\mathrm{Gwid}}{100}+\frac{(F-F0)*\mathrm{<figure-callout\; id="100"\; label="Goil"\; filenames="2012103763532100003DEST\_PATH\_IMAGE001.png,2012103763532100003DEST\_PATH\_IMAGE002.png"\; state="\{\{state\}\}">Goil</figure-callout>}}{100*M}-\mathrm{Lcs}---\left(10\right)$

In the formula: Gap intermesh determination calculated value;

Fh strip exit thickness is set calculated value;

The gain of Gwid width compensation, the other data of layer;

The WCX width compensation;

The gain of Goil Oil Film Compensation, the other data of layer;

M milling train constant;

Lcs roll gap learning coefficient, the other data of layer;

WCX width compensation coefficient;

F, F0, Fj is respectively roll-force, throw-on pressure, initial throw-on pressure;

S, S0, Sj are respectively rolling roll gap, press roll gap, initial roll gap;

M, M ' mill stiffness.

The data processing method of described roll gap self study is: adopt and to remove simultaneously the roll gap self study data screening method of extreme value roll-force and extreme value roll gap, and adopt the genetic learning method when long of genetic learning method in short-term to alleviate the concussion of roll gap self study; Described when long the genetic learning method refer to the thickness of model memory tape steel head, press steel grade, specification memory, the operation of rolling is upgraded memory, heredity combines with in short-term heredity when long, in the calculating of input model specification.

The data processing method of described roll gap self study is: take to remove roll-force maximum or minimum of a value, remove simultaneously the gap values between rollers of roll-force maximum or minimum of a value corresponding points; The corresponding roll-force value of also removing corresponding points when removing roll gap maximum or minimum of a value simultaneously asks the mean value of remaining data to do the poor roll gap self study value that is used for calculating again.

Compared with prior art, the invention has the beneficial effects as follows:

Increase the roll gap drift compensating of spring model, optimize intermesh determination computation model structure with this, and the roll gap self-learning method is improved, can effectively improve the setting accuracy of hot-strip roll gap model.

Description of drawings

Fig. 1 is roll gap formula proving schematic diagram.

Fig. 2 is the band steel head thickness deviation before optimizing.

Fig. 3 is the band steel head thickness deviation after optimizing.

Fig. 4 is roll gap self study coefficient trend graph.

Fig. 5 is band steel head thickness deviation moon statistic curve.

The specific embodiment

The invention will be further described below in conjunction with accompanying drawing.

A kind of hot-strip roll gap model optimization method that is applied in the hot continuous rolling production process, the present invention optimizes intermesh determination computation model structure; Improve roll gap self study data processing method, optimize the roll gap self-learning method, can effectively improve the setting accuracy of hot-strip roll gap model.

The below is described in detail the concrete steps of intermesh determination computation model structure of the present invention.

(1) intermesh determination computation model structure is optimized in the impact of consideration roll gap null offset.

Former intermesh determination computation model is as follows:

$\mathrm{Gap}=\mathrm{Fh}-\frac{(S-S0)*\mathrm{Gwid}}{100*\mathrm{WCX}}+\frac{(F-F0)*\mathrm{Goil}}{100*M}-\mathrm{Lcs}---\left(1\right)$

In the actual operation of rolling, along with the strain of milling train working stand, the change of rolled piece width etc. factor, distortion in various degree will occur in roll, and actual roll gap also will change thereupon.Meanwhile, can drift about gradually in the operation of rolling zero point of spring curve, although this roll gap model has been considered the impact of width of rolling stock on mill stiffness, but have problems in the correction that changes roll gap null offset for width of rolling stock, so that fixed value of roller slit is devious all the time, so that the thickness and precision deviation of former blocks of steel is larger in the actual operation of rolling, and follow-up band steel also must roller rest seam self study guarantee target thickness.

Consider in the master mould that width of rolling stock is following formula (WCX is width impact compensation in the formula) on the roll gap spring amount of mill stiffness impact:

$\mathrm{\&Delta;S}=\frac{S-S_0}{\mathrm{WCX}}---\left(2\right)$

In roll gap spring model calculates, fail breaker roll to press null offset and make compensation, the spring computation model of again deriving of starting with thus now, as follows according to the derivation of the spring curve derivation principle figure of milling train such as accompanying drawing 1:

The calculating formula that can be got mill stiffness by Fig. 1 is:

$M=\frac{F-F_0}{S-S_0}=\frac{F_0-F_j}{S_0-S_j};$ $M^{\&prime;}=\frac{F-F_0}{S_1-S_0^{\&prime;}}=\frac{F_0-F_j}{S_0^{\&prime;}-S_j}---\left(3\right)$

Because: $\mathrm{WCX}=\frac{M^{\&prime;}}{M}=\frac{S-S_0}{S_1-S_0^{\&prime;}}=\frac{S_0-S_j}{S_0^{\&prime;}-S_j}---\left(4\right)$

So: ${\mathrm{<figure-callout\; id="1"\; label="S"\; filenames="HDA00002222167900011.png"\; state="\{\{state\}\}">S</figure-callout>}}_1=\frac{S-S_0}{\mathrm{WCX}}+S_0^{\&prime;}---\left(5\right)$

$S_0=(S_0^{\&prime;}-S_j)\&times;\mathrm{WCX}+S_j---\left(6\right)$

Therefore formula (7) is:

$\mathrm{\&Delta;S}=S_1-S_0=\frac{S-S_0}{\mathrm{WCX}}+(S_0^{\&prime;}-S_j)(1-\mathrm{WCX})---\left(7\right)$

Getting formula (8) by formula (6) is:

$S_0^{\&prime;}=\frac{S_0-S_j}{\mathrm{WCX}}+S_j---\left(8\right)$

Formula (8) substitution formula (7) is obtained formula (9):

$\mathrm{\&Delta;S}=\frac{S-S_0}{\mathrm{WCX}}+(S_0-S_j)(\frac{1}{\mathrm{WCX}}-1)---\left(9\right)$

According to derivation noted earlier, in the former intermesh determination computing formula of mill spring formula substitution, the intermesh determination computation model after finally being optimized is following form:

$\mathrm{Gap}=\mathrm{Fh}-\frac{(S-S_j)*\mathrm{Gwid}}{100*\mathrm{WCX}}+\frac{(S_0-S_j)*\mathrm{Gwid}}{100}+\frac{(F-F0)*\mathrm{<figure-callout\; id="100"\; label="Goil"\; filenames="2012103763532100003DEST\_PATH\_IMAGE001.png,2012103763532100003DEST\_PATH\_IMAGE002.png"\; state="\{\{state\}\}">Goil</figure-callout>}}{100*M}-\mathrm{Lcs}---\left(10\right)$

In the formula: Gap intermesh determination calculated value;

Fh strip exit thickness is set calculated value;

The gain of Gwid width compensation, the other data of layer;

The WCX width compensation;

The gain of Goil Oil Film Compensation, the other data of layer;

M milling train constant;

S, S0, Sj are respectively rolling roll gap, press roll gap, initial roll gap;

F, F0, Fj is respectively roll-force, throw-on pressure, initial throw-on pressure;

Lcs roll gap learning coefficient, the other data of layer;

WCX width compensation coefficient;

M, the M` mill stiffness.

By in the intermesh determination computation model, increasing the roll gap drift compensating, optimize the intermesh determination computation model, eliminate the impact that roll gap null offset amount deviation is calculated roll gap, thereby improve the setting computational accuracy of cleaning between rolls of finish rolling strip steel model.

(2) improve roll gap self study data processing method, optimize the roll gap self-learning method.

By the observation analysis to sampled data, data processing method to the roll gap self study is improved: adopt the roll gap self study data screening method remove simultaneously extreme value roll-force and gap values between rollers, and adopt genetic learning mechanism when long to alleviate the concussion of roll gap self study.

Roll gap self study formula is as follows:

LCS1N=β _S*LCS1I+(1-β _S)*LCS1 （11）

In the formula: LCS1N roll gap self study currency;

LCS1I roll gap self study instantaneous value;

The roll gap self study value of piece steel before the LCS1;

β _SSmoothing factor.

The self-learning method that above-mentioned roll gap self learning model adopts is the index exponential smoothing; two problems of having chosen of coefficent of exponential smoothing exist; may cause the learning process vibration when this coefficient is large, this is to cause again study unusually slow a few hours, and we improve the model learning method for this reason:

1) the real data processing method of improvement self study:

Here the roll gap self learning model that proposes is 0.75 second 10 point afterwards getting continuously roll-force, roll gap sampled value to the processing of real data, removing maximin is averaging again, then by the mean value calculation of roll-force, roll gap spring thickness (GM), using the mean value of finish rolling exit thickness (FH7) actual value to deduct GM thickness, namely is roll gap self study value again.

Here taked distinct methods to carry out self study data sampling and Data Management Analysis:

Consider that (suddenly bigger than normal or less than normal) when unusual appears in sampled data, to have mutual interaction relation between roll-force and the roll gap, therefore remove simultaneously the gap values between rollers of its corresponding points when taking to remove roll-force maximum (little) value, also remove simultaneously the roll-force value of corresponding points when removing accordingly roll gap maximum (little) value, ask again the mean value of remaining data to do the poor roll gap self study value of calculating.

Genetic learning method when 2) increasing length:

The roll gap self learning model that adopts at present is to divide according to roll material to the mode of learning of thickness, and a steel one is learned, implementation be the short-term learning method, not ideal to the precision control effect of thickness.For this reason, in the roll gap self learning model, increase genetic learning method when long, described when long heredity refer to the thickness of model memory tape steel head, press steel grade, specification memory, the operation of rolling is upgraded memory, and heredity combines with in short-term heredity when long, in the calculating of input model specification.

The steel grade that the present embodiment adopts is Q235, is rolled at the 1780mm hot tandem rolling mill, and finished product thickness is 3.544mm, finished width 1042mm, 880 ℃ of finishing temperatures.Implementation step is as follows:

(1) the gap Set Model structure behind definite the optimization

According to roll gap formula proving schematic diagram such as accompanying drawing 1, and the intermesh determination computation model of the derivation of mentioning in the technical scheme after being optimized is formula (10):

$\mathrm{Gap}=\mathrm{Fh}-\frac{(S-S_j)*\mathrm{<figure-callout\; id="100"\; label="Gwid"\; filenames="2012103763532100003DEST\_PATH\_IMAGE001.png,2012103763532100003DEST\_PATH\_IMAGE002.png"\; state="\{\{state\}\}">Gwid</figure-callout>}}{100*\mathrm{WCX}}+\frac{(S_0-S_j)*\mathrm{<figure-callout\; id="100"\; label="Gwid"\; filenames="2012103763532100003DEST\_PATH\_IMAGE001.png,2012103763532100003DEST\_PATH\_IMAGE002.png"\; state="\{\{state\}\}">Gwid</figure-callout>}}{100}+\frac{(F-F0)*\mathrm{<figure-callout\; id="100"\; label="Goil"\; filenames="2012103763532100003DEST\_PATH\_IMAGE001.png,2012103763532100003DEST\_PATH\_IMAGE002.png"\; state="\{\{state\}\}">Goil</figure-callout>}}{100*M}-\mathrm{Lcs}$

(2) initial data such as collection site actual process parameter, layer other parameter and second-level model parameter are carried out the roll gap variable quantity that mill spring amount and oil film cause and are calculated, and calculation procedure is as follows:

At first, whether judgement symbol position IFB is 0, if be 0, and the formula that is calculated as follows of bending roller force FB then:

FB.cal1=2*Arf*BP.lay1 （12）

The Arf here is the roller influence coefficient, is worth to be normal value 0.2, and BP.lay1 is roller pressure, by providing in the other data of secondary layer.If flag bit is not 0, then the formula that is calculated as follows of bending roller force FB:

FB.cal1=2*Arf*BP.mod （13）

Difference is that the BP.mod here extracts in the second-level model setting data.

Secondly, the calculating of bouncing, formula is as follows: $S=\mathrm{SP}0.\mathrm{mod}\left(i\right)+\frac{\mathrm{SP}0.\mathrm{mod}(i+1)-\mathrm{SP}0.\mathrm{mod}\left(i\right)}{\mathrm{RF}0.\mathrm{mod}(i+1)-\mathrm{RF}0.\mathrm{mod}\left(i\right)}*[\mathrm{TF}-\mathrm{RF}0.\mathrm{mod}\left(i\right)]---\left(14\right)$

In the formula, RF.cal1 is the roll-force calculated value.

TF= RF.cal1+FB.cal1 （15）

The method that need determine herein the i value is: if TF 〉=RF0.mod (i) then determine current i value (span of i from 1 to 9).

$S0=\mathrm{SP}0.\mathrm{mod}\left(i\right)+\frac{\mathrm{SP}0.\mathrm{mod}(i+1)-\mathrm{SP}0.\mathrm{mod}\left(i\right)}{\mathrm{RF}0.\mathrm{mod}(i+1)-\mathrm{RF}0.\mathrm{mod}\left(i\right)}*[\mathrm{RF}0.\mathrm{etc}-\mathrm{RF}0.\mathrm{mod}\left(i\right)]---\left(16\right)$

In the formula

Roll-force when returning to zero for roll gap, SP0.mod, RF0.mod are respectively field by using and press a series of gap values between rollers and the throw-on pressure value of surveying when method is surveyed mill stiffness.Here the i value is determined the same formula of method.

Then, carry out width impact compensation and calculate, computing formula is as follows:

$\mathrm{WCX}=\mathrm{WCC}\left(i\right)+\frac{\mathrm{WCC}(i+1)-\mathrm{WCC}\left(i\right)}{\mathrm{WCDIV}(i+1)-\mathrm{WCDIV}\left(i\right)}*[\mathrm{<figure-callout\; id="1"\; label="F"\; filenames="HDA00002222167900011.png"\; state="\{\{state\}\}">F</figure-callout>}1\mathrm{EW}.\mathrm{cal}1-\mathrm{WCDIV}\left(i\right)]---\left(17\right)$

The method of determining i value herein is at this moment i value (i span from 1 to 5) more than or equal to WCDIV (i) time for judging as F1EW.cal1 (2).

In the formula, WCC width compensation coefficient 1, normal value;

The wide calculated value of F1EW.cal1 F1E exit plate;

WCDIV width compensation coefficient 2 is provided by subprogram.

Then, carry out the Oil Film Compensation relevant calculation:

$F=\mathrm{OLFF}\left(i\right)+\frac{\mathrm{OLFF}(i+1)-\mathrm{OLFF}\left(i\right)}{\mathrm{REVF}.\mathrm{con}(i+1)-\mathrm{REVF}.\mathrm{con}\left(i\right)}*[\mathrm{REV}.\mathrm{cal}1-\mathrm{REVF}.\mathrm{con}\left(i\right)]---\left(18\right)$

$F0=\mathrm{OLFF}\left(i\right)+\frac{\mathrm{OLFF}(i+1)-\mathrm{OLFF}\left(i\right)}{\mathrm{REVF}.\mathrm{con}(i+1)-\mathrm{REVF}.\mathrm{con}\left(i\right)}*[\mathrm{REV}0.\mathrm{etc}-\mathrm{REVF}.\mathrm{con}\left(i\right)]---\left(19\right)$

In the formula: OLFF oil-film force (constant);

REV.cal1 roll revolution calculated value;

Roll revolution (constant) when REVF.con surveys oil-film force;

The roll revolution of REV0.etc timing signal.

Adopt at the scene at last the roller pressing method to carry out the milling train constant calculations:

$M=\frac{\mathrm{RF}0.\mathrm{mod}(i+1)-\mathrm{RF}0.\mathrm{mod}\left(i\right)}{\mathrm{SP}0.\mathrm{mod}(i+1)-\mathrm{SP}0.\mathrm{mod}\left(i\right)}---\left(20\right)$

Finally draw fixed value of roller slit by computational process noted earlier, in order to investigate the practical application effect of the model formation after the optimization, band steel head thickness deviation before and after statistical model is optimized, aberration curve figure such as Fig. 2, Fig. 3, as can be seen from the figure come, the precision of gap Set Model is improved, and the thick difference of band steel head is significantly less than before the optimization.

(3) gather the initial data such as actual process parameter and model parameter, the data processing method of the roll gap self study of application enhancements is calculated roll gap self study coefficient, and calculation procedure is as follows:

Roll gap learning coefficient computing formula is at first proposed:

LCS1I(I) =Gofset*(MFH(I)-GMH(I)) （21）

In the formula: LCS1I (I) roll gap learning coefficient instantaneous value;

Gofset thickness deviation penalty coefficient;

GMH (I) thickness gauge THICKNESS CALCULATION;

MFH (I) second flow THICKNESS CALCULATION;

The I shelf number.

Thickness deviation penalty coefficient in the formula is got normal value here: 0.90,0.95,0.95,1.00,1.00,1.00,0.85.The roll gap self study mainly divides two parts: the one, and the calibrator THICKNESS CALCULATION, the 2nd, the second flow THICKNESS CALCULATION, the below calculates for these two parts, and calculation procedure is as follows:

1) second flow thickness (MFH) calculates

1. second flow thickness calculates for the first time

At first calculate each frame outlet belt speed by advancing slip calculated value and roller speed measured value:

The 1st～5: V (I)=(the 6th of 1+FS-AFS (I+1,2)-LCF) * VR (22): V (I)=(1+FS-AFS (I+1,2)) * VR (23)

The 7th: V (I)=(1+FS) * VR (24)

In the formula: FS is advancing slip setting calculated value;

The advancing slip computation layer of AFS (I, 2) is not worth;

LCF is advancing slip, and the learning coefficient layer is not worth;

VR roller speed measured value.

Calculate the second flow thickness of each frame:

According to the second flow equal principle, by the finish rolling exit thickness of actual measurement, the exit thickness of each frame of inverse front:

The 7th: MFH (I)=FDH_T (FDH_T finish rolling exit thickness measured value) (25)

The 1st～6: MFH (I)=MFH (I+1) * V (I+1)/V (I); (26)

Or MFH (I)=MFH (I+1) (I+1 is built on stilts out-of-date); (27)

Or MFH (I)=MFH (I+1) * V (I+1)/V (I-1) (I is built on stilts out-of-date).（28）

2. second flow thickness calculates again

Recomputate reduction ratio, advancing slip and each frame outlet belt speed according to above second flow thickness:

$\mathrm{DR}=\frac{\mathrm{MFH}(I-1)-\mathrm{MFH}\left(I\right)}{\mathrm{MFH}(I-1)}---\left(29\right)$ FSR(I)=AFS(I,1)*DR （30）

V(I)=(1+ FSR(I) )*VR （31）

In the formula: the DR reduction ratio is calculated value again;

FSR is advancing slip calculated value again;

Each frame outlet belt speed of V (I) is calculated value again;

The front sliding parameter layer of AFS (I, 1) is not worth.

According to the calculating principle of step in 1., adopt above every again calculated value, recomputate second flow thickness, obtain the again calculated value of MFH (I).

Here, advancing slip technological parameter get the layer not value be:

Each frame value of MFS1F.LAY1.AFS (I, 1) is 0.25; MFS1F.LAY1.AFS (I, 2)=2.6000001E-02,2.6000001E-02,2.2000000E-02,1.5000000E-02,1.2000000E-02,8.9999996E-03,0.0000000E+00;

MFS1F.LAY1.AFS (I, 3) and each frame value of MFS1F.LAY1.AFS (I, 4) are 0.

2) thickness gauge thickness (GMH) calculates

$\mathrm{GMh}\left(I\right)=S\_T\left(I\right)+\frac{(S-S0)*\mathrm{Gwid}.\mathrm{lay}1}{100*\mathrm{WCX}}-\frac{(F-F0)*\mathrm{<figure-callout\; id="1"\; label="Goillay"\; filenames="HDA00002222167900011.png"\; state="\{\{state\}\}">Goillay</figure-callout>}1}{100*M}---\left(32\right)$

Three on equation the right is respectively the roll gap measured value, the roll gap bouncing value, and oil film effect is to the compensation of roll gap.This computational process is consistent with the setting computational process that the front is narrated, therefore repeated description no longer.

In the formula: S_T (I) roll gap measured value;

The rolling roll gap of S;

WCC width compensation coefficient 1;

F1EW (2) F1E exit plate is wide;

WCDIV width compensation coefficient 2;

Gwid.lay1 width compensation coefficient (being 100% entirely)

The WCX width compensation;

Goil.lay1 oil film rigidity penalty coefficient (being 100% entirely);

M mill stiffness (rigidity that method is measured milling train is pressed in employing);

RF_T surveys roll-force;

REV_T surveys roll rotational speed.

3) roll gap learning coefficient instantaneous value calculates:

Calculating is divided into two kinds of situations, a kind of being used in the calculating of finishing stand setup model specification, and another kind is used in the AGC setting.

LCS1I (I)=Gofset* (MFH (I)-GMH (I)) (FSU sets calculating) (33)

LCS2I (I)=FS1.ACT.GME (I) (AGC setting) (34)

4) the roll gap learning coefficient upgrades and calculates:

LCS1N(I) = B1(I) * LCS1I(I) +(1- B1(I) )* LCS1(I) （35）

FL2.LCS1(I) = LCS1N(I) （36）

In the formula: this calculated value of LCS1N (I) roll gap learning coefficient;

B1 (I) roll gap study smoothing factor;

LCS1I (I) roll gap learning coefficient instantaneous value;

The upper piece steel rider seam of LCS1 (I) learning coefficient.

Here FL2.LCS1 (I) is the roll gap learning coefficient of this piece steel, deposits in after having calculated in the other document data bank of layer, sets for lower steel rider seam and calculates.The heredity self-learning algorithm deposits its learning coefficient current band steel in the other data file of self study layer in by steel grade, specification classification when using length, carry out the other index of layer by steel grade and specification, when next piece band steel steel grade or specification not simultaneously, during the setting of extracting and be used in this piece band steel from the other data file of layer is calculated.

According to above-mentioned steps, after adopting new roll gap seam self study data sampling and data processing method, obtain roll gap self study coefficient trend graph as shown in Figure 4.As seen from Figure 4, the roll gap self study coefficient before optimizing is increasing always, and this explanation intermesh determination error is very large, owing to the existence that smoothing factor is arranged, rolls continuously and also fails this error in the full remuneration behind several the steel.Roll gap self study coefficient fluctuation after the optimization is very little, proves that the formula validity after optimizing is very good, so that each frame roll gap deviation has obtained reducing largely.At last the head thickness deviation in each month has been carried out statistics such as Fig. 5, can find out, the July behind the roll gap model optimization is to December, be significantly less than optimization with the head thickness deviation of steel before, thickness and precision has had significant raising.

Claims (3)

1. the optimization method of a hot-strip roll gap model, comprise intermesh determination computation model and roll gap self study data processing method, it is characterized in that, in the roll gap computation model, added the roll gap drift compensating, reduced the caused roll gap calculation deviation of roll gap null offset, optimize the gap Set Model structure, thereby improved the setting computational accuracy of hot-strip roll gap model;

Intermesh determination computation model after the optimization is following form:

$\mathrm{Gap}=\mathrm{Fh}-\frac{(S-S_j)*\mathrm{Gwid}}{100*\mathrm{WCX}}+\frac{(S_0-S_j)*\mathrm{Gwid}}{100}+\frac{(F-F0)*\mathrm{Goil}}{100*M}-\mathrm{Lcs}$

In the formula: Gap intermesh determination calculated value;

Fh strip exit thickness is set calculated value;

The gain of Gwid width compensation, the other data of layer;

The WCX width compensation;

The gain of Goil Oil Film Compensation, the other data of layer;

M milling train constant;

Lcs roll gap learning coefficient, the other data of layer;

WCX width compensation coefficient;

F, F0, Fj is respectively roll-force, throw-on pressure, initial throw-on pressure;

S, S0, Sj are respectively rolling roll gap, press roll gap, initial roll gap;

M, M ' mill stiffness.

2. a kind of hot-strip roll gap model optimization method according to claim 1, it is characterized in that, the data processing method of described roll gap self study is: adopt and to remove simultaneously the roll gap self study data screening method of extreme value roll-force and extreme value roll gap, and adopt the genetic learning method when long of genetic learning method in short-term to alleviate the concussion of roll gap self study; Described when long the genetic learning method refer to the thickness of model memory tape steel head, press steel grade, specification memory, the operation of rolling is upgraded memory, heredity combines with in short-term heredity when long, in the calculating of input model specification.

3. a kind of hot-strip roll gap model optimization method according to claim 1, it is characterized in that, the data processing method of described roll gap self study is: take to remove roll-force maximum or minimum of a value, remove simultaneously the gap values between rollers of roll-force maximum or minimum of a value corresponding points; The corresponding roll-force value of also removing corresponding points when removing roll gap maximum or minimum of a value simultaneously asks the mean value of remaining data to do the poor roll gap self study value that is used for calculating again.

Images