CN101403890B - Method for improving model prediction precision by utilizing neuroid classification modeling method - Google Patents
Method for improving model prediction precision by utilizing neuroid classification modeling method Download PDFInfo
- Publication number
- CN101403890B CN101403890B CN2008100797691A CN200810079769A CN101403890B CN 101403890 B CN101403890 B CN 101403890B CN 2008100797691 A CN2008100797691 A CN 2008100797691A CN 200810079769 A CN200810079769 A CN 200810079769A CN 101403890 B CN101403890 B CN 101403890B
- Authority
- CN
- China
- Prior art keywords
- net
- roll
- pressure correction
- neuroid
- draught pressure
- 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.)
- Expired - Fee Related
Links
Landscapes
- Control Of Metal Rolling (AREA)
Abstract
The invention provides a method for improving model prediction precision by a neuron network classification and modeling method, is applicable to strip steel hot rolling production line which adopts aneuron network to perform model correction, and aims at improving the model prediction precision. In the method, different neuron networks are used according to roll types, that is, the original networks are maintained for the production of high chromium-nickel cast iron rolls; new networks are added to high-speed steel rolls, and the form of the networks are the same as the form of the originalneuron networks; the high-speed steel rolls and the high chromium-nickel cast iron rolls use the respective pressure correction, convexity and flatness to control the neuron networks; the new networkconfiguration files are the same as the original networks files; and models call different network files according to the types of different roll materials; and the networks of different roll materials are relatively independent and non-interfering.
Description
Technical field
The present invention relates to a kind of method of utilizing neuroid classification modeling method to improve model prediction precision.
Background technology
Hot strip rolling production is a most ripe field of present appliance computer control, and its range of control has comprised whole process of production.The hot strip rolling computer system is divided into basic automatization control system and process automation control system basically; Basic automatization is towards unit and equipment, and extensively adopts digital drive technology.The basic automatization control system is carried out device control according to the setup parameter of process automation system, simultaneously relevant detection data in the operation of rolling is passed to the process automation control system.Process automation is towards whole production line, and its central task is that each unit on the production line and each equipment are carried out set-up and calculated, makes rolled band steel obtain indexs such as the thickness of target call, width, temperature, convexity and flatness.Process computer is provided with raw data input, on-line data acquisition and model self study etc. and is setting model service and supporting function.The major function of hot continuous rolling process automation control is according to raw data and rolling strategy, use a series of mathematical models to carry out set-up and calculated, under send controlled variable to arrive basic automatization, therefore, model prediction precision to the band steel output and quality play crucial effects.Improve traditional form leveling factor method or the neuroid leveling factor method of the main at present employing of model specification precision.Because neuroid has the ability of self study, make oneself to adapt to external environment more, so neuroid has obtained application more and more widely.But neuroid needs long exploitation and debugging cycle and expensive soft, hardware development funds.
Summary of the invention
The present invention seeks to overcoming the deficiency of above-mentioned prior art, provide a kind of when input states such as equipment, technology, environment change, low-cost, utilize the method for neuroid classification modeling method raising model prediction precision to fast speed.
On the hot-rolled strip production line of mm finishing mill unit, be used for mm finishing mill unit with high-speed steel roll alternate standard grain roll with 7 tandem mills (F0-F6) composition.According to the roll classification, use different neuroids, promptly keep legacy network to be used for the production of high chromium nickel iron roll; For high-speed steel roll increases new network in addition, latticed form is identical with original neuroid.High-speed steel roll and high chromium nickel iron roll use pressure correction, convexity and flatness control neuroid separately; New network profile and legacy network file are identical, only are that model calls different network files according to different roll roller matter types; The network of various differential roller matter is relatively independent, does not disturb mutually.
The concrete steps of the inventive method:
(1) increase by 8 neuroids and use for high-speed steel roll, legacy network supplies the production control of high chromium nickel iron roll:
Fm_prof.net target convexity and target flatness forecast net
Fm_rf0.net F0 frame draught pressure correction net
Fm_rf1.net F1 frame draught pressure correction net
Fm_rf2.net F2 frame draught pressure correction net
Fm_rf3.net F3 frame draught pressure correction net
Fm_rf4.net F4 frame draught pressure correction net
Fm_rf5.net F5 frame draught pressure correction net
Fm_rf6.net F6 frame draught pressure correction net
Copy above-mentioned 8 networks, rename following 8 networks respectively as and supply the high-speed steel roll production control:
Fm_hsprof.net target convexity and target flatness forecast net
Fm_hsrf0.net F0 frame draught pressure correction net
Fm_hsrf1.net F1 frame draught pressure correction net
Fm_hsrf2.net F2 frame draught pressure correction net
Fm_hsrf3.net F3 frame draught pressure correction net
Fm_hsrf4.net F4 frame draught pressure correction net
Fm_hsrf5.net F5 frame draught pressure correction net
Fm_hsrf6.net F6 frame draught pressure correction net
This method keeps the legacy network structure, only increases the neuroid that uses for high-speed steel roll, makes high-speed steel roll and high chromium nickel iron roll use pressure correction, convexity and flatness control neuroid separately.The network of various differential roller matter is relatively independent, does not disturb mutually.
(2) revise model Controlling Source code, according to the differential roller class, call different neuroids, thereby improve the setting accuracy of follow-up band steel, operation steps is as follows:
1) reads raw data, process system and roller class sign;
2) do you judge it is high-speed steel roll?, do not enter the 4th) step;
3) read high chromium nickel iron roll neuroid data, jump to the 5th) step;
4) read high-speed steel roll neuroid data;
5) call the temperature computation model, obtain the outlet temperature of band steel at each milling train by chilled water, the hot spoke of air cooling etc. between rolling deformation heat, roll heat conduction, frame;
6) call the calculation of pressure model, use its corresponding roll neuroid data, calculate each mill milling pressure;
7) call plate shape computation model, use its corresponding roll neuroid data, calculate each mill roll bending, string roller amount;
8) transmit setting value; The process that presets finishes.
Comprise process system between de-scaling mode, mill load partition coefficient, frame cooling water flow, working roll cooling water flow, the technological lubrication sign etc. that comes into operation.
Call different neuroids by improving source code according to roll roller class, realized the control of high chromium nickel iron roll and high-speed steel roll pressure prediction category of model, make the high chromium nickel iron roll operation of rolling not be subjected to the influence of the system reform fully; High-speed steel roll constantly improves its model prediction precision by online training process in the follow-up operation of rolling of band steel.
Use on the hot strip rolling production line that the present invention is suitable for adopting neuroid to carry out the model correction.The present invention's control thought of will classifying is applied to neuroid, has avoided the neuroid structure is developed once more, and is when keeping the neuroid advantage, low-cost, finish improvement apace.When input states such as equipment, technology, environment changed, not only operating process was comparatively simple to adopt the inventive method, and when guaranteeing ordinary production, can improve the forecast precision of model.
Embodiment
Hot continuous rolling factory of Taiyuan Iron ﹠ Steel Corp's (hereinafter to be referred as Taiyuan Iron and Steel Co.) Process Control System has 11 neuroids, is used for that model spreads, the forecast of pressure, temperature, convexity and flatness model.Wherein: 7 of F0～F7 frame draught pressure predicted value correction factor, 1 of band steel Target Board convexity and the forecast of target flatness, 1 of finishing temperature correction factor, 1 of finish rolling district absolute spread forecast, 1 of band steel matter hardness correction factor, every network all has strict separately division of duty, plays a part extremely crucial to improving model prediction precision.
Taiyuan Iron and Steel Co. hot continuous rolling factory buys high-speed steel roll and is used for mm finishing mill unit.Compare with the standard grain roll, high-speed steel roll has higher wearing quality and preferable surface quality.But high-speed steel roll also has significant disadvantages, and promptly friction factor is big, and the roll surface convexity is difficult to control.Friction factor is big, and the also corresponding increase of roll-force under similarity condition, can make roll-force increase by 10%～20%.Because the characteristic of high-speed steel roll and standard grain roll has bigger difference, pressure model can't correctly forecast.The pressure model forecast precision is low excessively, causes the mill roll-gap specification error big, and the second flow amount can't keep balance between the mm finishing mill unit frame, very easily causes the unit steel scrap, and the finished product thickness control accuracy is low, template is poor, has a strong impact on the output and the quality of finish to gauge product.In the single chassis high-speed steel roll trial period, the necessary online training function of manual-lock neuroid of finish rolling platform operative employee avoids the training of network modified value unusual, influences the ordinary production control of follow-up band steel.In the operation of rolling, the operative employee need constantly manually intervene simultaneously, as adjust the anti-milling train excess current power down of frame load distribution rate according to the actual measurement rolling load; Manual adjustment frame depressing position guarantees that second flow amount balance has prevented cover or drawn steel or the like, operation of rolling instability between frame in the crossing process.After Taiyuan Iron and Steel Co. 1# hot rolling system three electric control systems are transformed, process control level adopts SIEMENS company development product under the SIROLL2 environment, used neuroid in the Model Calculation, through long-term online and off-line training process, model prediction precision improves constantly.
With the F0 milling train is example, and its draught pressure forecast network input item and output item are defined as follows:
name：fm_rf0
Number of inputs:22//input item number
Number of outputs:1//output item number
Inputs: // input item
no. descr minimum maximum accuracy
1 HE, 1 70 0.05 // inlet thickness
2 TE, 600 1,200 1 // exit thickness
3 BTENS, 0 100 0.1 // backward pull
4 FTENS, 0 100 0.1 // forward pull
5 EPS0,0.8 0.001 // reduction ratio
6 WRR, 100 600 1 // roller footpath
7 VU, 0.01 50 0.05 // speed
8 C, 0 100 0.001 // carbon content
9 SI, 0 100 0.01 // silicone content
10 MN, 0 100 0.01 // manganese content
11 P, 0 100 0.01 // phosphorus content
12 S, 0 100 0.001 // sulfur content
13 AL, 0 100 0.001 // aluminium content
14 CR, 0 100 0.01 // chromium content
15 CU, 0 100 0.01 // copper content
16 MO, 0 100 0.01 // molybdenum content
17 TI, 0 100 0.01 // Ti content
18 NI, 0 100 0.01 // nickel content
19 V, 0 100 0.01 // content of vanadium
20 NB, 0 100 0.01 // content of niobium
21 N, 0 100 0.001 // nitrogen content
22 B, 0 100 0.001 // boron content
Outputs: // output item
no.descr minimum maximum
1?RFC 0 0.5 1.6
According to conventional way, need change the structure of neuroid, promptly in above-mentioned original input item, increase roll roller category, readjust the weight matrix of input layer and hiding interlayer in the neuroid then.This way disadvantage is more, big as the retrofit work amount, existing rolling data are destroyed, may influence the network precision after the structural change, need carry out long-time, jumbo off-line data analysis and improvement, the on-line testing cycle is long, needs to follow the tracks of the production of different cultivars, specification band steel, to confirm the reliability of data.This not only needs the long construction cycle, adds the great development funds, and may have influence on the stable of production run.
In that to give full play to neuroid flexible, on the basis of characteristics such as stable, consider the influence of high-speed steel roll to the pressure prediction model, and improve all difficulties that neuroid may bring, employing is similar to the sorting technique of conventional art solution, according to production status and demand neuroid is carried out classification model construction, automatically, realize the online training process of heterogeneous networks in real time, solve because many on-the-spot factors such as equipment and performance are disturbed problems such as neuroid study disorder, improve model prediction precision effectively, improve the hot-strip end product quality.
Increase by 8 neuroids in mm finishing mill unit, be respectively applied for pressure prediction value correction neuron net and a convexity and the flatness control neuroid of each frame F0--F6 when using high-speed steel roll.After model structure, process data and steering logic are analyzed, increase by 8 neuroids, use for high-speed steel roll, legacy network supplies the production control of high chromium nickel iron roll:
Fm_prof.net target convexity and target flatness forecast net
Fm_rf0.net F0 frame draught pressure correction net
Fm_rf1.net F1 frame draught pressure correction net
Fm_rf2.net F2 frame draught pressure correction net
Fm_rf3.net F3 frame draught pressure correction net
Fm_rf4.net F4 frame draught pressure correction net
Fm_rf5.net F5 frame draught pressure correction net
Fm_rf6.net F6 frame draught pressure correction net
Copy above-mentioned 8 networks, rename following 8 networks respectively as and supply the high-speed steel roll production control:
Fm_hsprof.net target convexity and target flatness forecast net
Fm_hsrf0.net F0 frame draught pressure correction net
Fm_hsrf1.net F1 frame draught pressure correction net
Fm_hsrf2.net F2 frame draught pressure correction net
Fm_hsrf3.net F3 frame draught pressure correction net
Fm_hsrf4.net F4 frame draught pressure correction net
Fm_hsrf5.net F5 frame draught pressure correction net
Fm_hsrf6.net F6 frame draught pressure correction net
(2) revise model Controlling Source code,, call different neuroids according to the differential roller class:
1) reads raw data, process system and roller class sign;
2) do you judge it is high-speed steel roll? be to enter the 4th) step;
3) read high chromium nickel iron roll neuroid data, jump to the 5th) step;
4) read high-speed steel roll neuroid data;
5) call the temperature computation model, obtain the outlet temperature of band steel at each milling train by chilled water, the hot spoke of air cooling etc. between rolling deformation heat, roll heat conduction, frame;
6) call the calculation of pressure model, use its corresponding roll neuroid data, calculate each mill milling pressure;
7) call plate shape computation model, use its corresponding roll neuroid data, calculate each mill roll bending, string roller amount;
8) transmit setting value.
After the online input of new system according to above-mentioned embodiment, produced on-site control is steady, and control accuracy promotes steadily.Amount to rolling 16 unit stainless steels in the test, 1056 blocks of band steel, rolling result adds up as follows:
Table 1, rolling load forecast deviation (stainless steel unit)
| Rolling load forecast deviation % | Frame F1 | Frame F2 | Frame F3 |
| Use high-speed steel roll | 1.15 | 1.13 | 1.12 |
| Use standard grain roll | 1.13 | 1.12 | 1.09 |
Amount to 247 of rolling general steel, rolling result adds up as follows:
Table 2, rolling load forecast deviation (general steel unit)
| Rolling load forecast deviation % | Frame F1 | Frame F2 |
| Use high-speed steel roll | 1.15 | 1.14 |
| Use standard grain roll | 1.13 | 1.11 |
After the project reconstruction, the rolling load forecast deviation during with the high-speed steel roll on probation of single chassis before transforming is reduced to present maximum 1.15% by 24.9%, is in same level with the rolling load forecast deviation maximal value 1.13% of standard grain roll.
When realizing that high-speed steel roll is produced stable control, guaranteed that finished strip thickness and template reach former design control accuracy, statistics sees Table 3.
Table 3, belt steel thickness and template control accuracy statistics
The inventive method also extend to produce some kinds more rambunctious rolling on.As the silicone content difference in the different grade silicon steels, but the output of pressure neuroid is basic identical, causes the pressure prediction deviation bigger, and the steel scrap phenomenon frequently takes place.For this reason, in the production run of silicon steel, use neuroid classification modeling method,, can obviously improve model prediction precision, significantly improve thickness control accuracy, effectively control the failure rate that silicon steel is produced according to the different neuroid of different silicon steel trade mark definition.
Claims (2)
1. a method of utilizing neuroid classification modeling method to improve model prediction precision on the hot-rolled strip production line of the mm finishing mill unit with 7 tandem mills (F0-F6) composition, is used for mm finishing mill unit with high-speed steel roll alternate standard grain roll; It is characterized in that according to the roll classification, use different neuroids, promptly keep legacy network to be used for the production of high chromium nickel iron roll; For high-speed steel roll increases new network in addition, latticed form is identical with original neuroid; High-speed steel roll and high chromium nickel iron roll use pressure correction, convexity and flatness control neuroid separately; New network profile and legacy network file are identical, only are that model calls different network files according to different roll roller matter types; The network of various differential roller matter is relatively independent, does not disturb mutually.
2. the method for utilizing neuroid classification modeling method to improve model prediction precision as claimed in claim 1 is characterized in that:
(1) increase by 8 neuroids and use for high-speed steel roll, legacy network supplies the production control of high chromium nickel iron roll:
Fm_prof.net target convexity and target flatness forecast net
Fm_rf0.net F0 frame draught pressure correction net
Fm_rf1.net F1 frame draught pressure correction net
Fm_rf2.net F2 frame draught pressure correction net
Fm_rf3.net F3 frame draught pressure correction net
Fm_rf4.net F4 frame draught pressure correction net
Fm_rf5.net F5 frame draught pressure correction net
Fm_rf6.net F6 frame draught pressure correction net
Copy above-mentioned 8 networks, rename following 8 networks respectively as and supply the high-speed steel roll production control:
Fm_hsprof.net target convexity and target flatness forecast net
Fm_hsrf0.net F0 frame draught pressure correction net
Fm_hsrf1.net F1 frame draught pressure correction net
Fm_hsrf2.net F2 frame draught pressure correction net
Fm_hsrf3.net F3 frame draught pressure correction net
Fm_hsrf4.net F4 frame draught pressure correction net
Fm_hsrf5.net F5 frame draught pressure correction net
Fm_hsrf6.net F6 frame draught pressure correction net
(2) revise model Controlling Source code, according to the differential roller class, call different neuroids, thereby improve the setting accuracy of follow-up band steel, operation steps is as follows:
1) reads raw data, process system and roller class sign;
2) do you judge it is high-speed steel roll? be to enter the 4th) step;
3) read high chromium nickel iron roll neuroid data, jump to the 5th) step;
4) read high-speed steel roll neuroid data;
5) call the temperature computation model, obtain the outlet temperature of band steel at each milling train by chilled water, the hot spoke of air cooling etc. between rolling deformation heat, roll heat conduction, frame;
6) call the calculation of pressure model, use its corresponding roll neuroid data, calculate each mill milling pressure;
7) call plate shape computation model, use its corresponding roll neuroid data, calculate each mill roll bending, string roller amount;
8) transmit setting value; The process that presets finishes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008100797691A CN101403890B (en) | 2008-11-08 | 2008-11-08 | Method for improving model prediction precision by utilizing neuroid classification modeling method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008100797691A CN101403890B (en) | 2008-11-08 | 2008-11-08 | Method for improving model prediction precision by utilizing neuroid classification modeling method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101403890A CN101403890A (en) | 2009-04-08 |
| CN101403890B true CN101403890B (en) | 2010-06-09 |
Family
ID=40537928
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2008100797691A Expired - Fee Related CN101403890B (en) | 2008-11-08 | 2008-11-08 | Method for improving model prediction precision by utilizing neuroid classification modeling method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101403890B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101901280B (en) * | 2009-06-01 | 2014-03-05 | 上海宝信软件股份有限公司 | Method for obtaining broadening by using computer program |
| CN101972780B (en) * | 2010-11-11 | 2013-06-26 | 攀钢集团钢铁钒钛股份有限公司 | Hot rolling titanium casting blank temperature control method |
| CN103678893B (en) * | 2013-12-03 | 2016-08-17 | 太原理工大学 | A kind of regular modeling method for special steel grade |
| CN112170501B (en) * | 2020-09-16 | 2022-05-27 | 太原理工大学 | Prediction method for wear crown and thermal crown of roller |
| CN115608793B (en) * | 2022-12-20 | 2023-04-07 | 太原科技大学 | Finish rolling temperature regulation and control method for mechanism fusion data |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5740686A (en) * | 1994-07-07 | 1998-04-21 | Siemens Aktiengesellschaft | Method and apparatus for rolling a metal strip |
| CN1234755A (en) * | 1996-10-23 | 1999-11-10 | 西门子公司 | Method for optimizing band-width distribution of steel band end passing rolling-mill |
-
2008
- 2008-11-08 CN CN2008100797691A patent/CN101403890B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5740686A (en) * | 1994-07-07 | 1998-04-21 | Siemens Aktiengesellschaft | Method and apparatus for rolling a metal strip |
| CN1234755A (en) * | 1996-10-23 | 1999-11-10 | 西门子公司 | Method for optimizing band-width distribution of steel band end passing rolling-mill |
Non-Patent Citations (1)
| Title |
|---|
| JP特开平11-707A 1999.01.06 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101403890A (en) | 2009-04-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101890435B (en) | Method for automatically controlling convexity and/or wedge of hot rolled in series type rolling machine and system thereof | |
| CN101403890B (en) | Method for improving model prediction precision by utilizing neuroid classification modeling method | |
| CN101797588B (en) | Secondary control rolling force forecasting method of hot mill | |
| CN101733289B (en) | Method for dynamically setting tandem rolling schedule of hot rolling strip | |
| CN101890433B (en) | On-line coordination control method for plate shape and plate thickness of strip tandem hot continuous rolling mill | |
| CN102513351B (en) | Rolling method and device for strip steel tandem cold rolling | |
| CN101537431B (en) | Method for online predicting and controlling of roughness of surface of cold-rolled thin steel strip | |
| CN102654749B (en) | Learning coefficient control device | |
| CN108655186B (en) | Roll-force presetting method based on artificial neural network and mathematical model | |
| CN103357669B (en) | Plate model prediction control method | |
| CN101745549B (en) | Method for controlling steel feeding temperature of band steel of hot strip mill | |
| CN102921743B (en) | Method for determining five stand tandem cold mill depressing distribution | |
| CN101587347A (en) | Design method of cold rolling mill band steel shape control target model | |
| CN105268748A (en) | Hot rolling load distribution method based on product classification | |
| CN107442574A (en) | A kind of five Stands Cold Tandem Mills group automated system gets rid of the control method of frame rolling | |
| CN101829687A (en) | Method for controlling roll gap of strip steel finishing mill to eliminate influence of specification change | |
| CN101648208A (en) | Rolling procedure making method of aluminium single-stand cold-rolling machine | |
| CN108405625A (en) | A kind of fair current roll-changing method for realizing the online roll change of ESP mm finishing mill units | |
| CN103962391A (en) | Rolling load optimization method for hot continuous finishing mill group | |
| CN102161054B (en) | Plate shape closed-loop control method based on influencing self learning of matrix | |
| CN103357670A (en) | Reduction schedule optimizing method applicable to five-stand UCM (universal crown mill) type cold continuous rolling unit | |
| CN110434172A (en) | A kind of sharing of load calculation method of furnace volume and mm finishing mill unit tandem rolling | |
| CN201720260U (en) | Convexity and/or wedge automatic control system of hot rolling series-connection type rolling mill | |
| CN106269911A (en) | A kind of roughing pressure Control System of Load Distribution method and roughing control system | |
| CN105032945A (en) | Method for evaluating strip shape and strip crown comprehensive control capacity of hot continuous rolling mill |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20100609 Termination date: 20161108 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |