CN1588346A - Method for predicting converter terminal point using artificial nurve network technology - Google Patents
Method for predicting converter terminal point using artificial nurve network technology Download PDFInfo
- Publication number
- CN1588346A CN1588346A CN 200410056892 CN200410056892A CN1588346A CN 1588346 A CN1588346 A CN 1588346A CN 200410056892 CN200410056892 CN 200410056892 CN 200410056892 A CN200410056892 A CN 200410056892A CN 1588346 A CN1588346 A CN 1588346A
- Authority
- CN
- China
- Prior art keywords
- carbon content
- network
- terminal temperature
- layers
- endpoint carbon
- 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.)
- Pending
Links
Images
Landscapes
- Carbon Steel Or Casting Steel Manufacturing (AREA)
Abstract
The present invention discloses artificial neural network technological method of forecasting the terminal of medium-sized and small converter. For steel-making converter system, there are many factors, such as furnace life, gun position, sputtering, etc. to affect the terminal carbon content and temperature in non-linear relation hard to describe mathematically. The present invention applies neural network technology in the control system, and can monitor and forecast the non-linearity, non-determinacy and complexity effectively to forecast the terminal temperature and terminal carbon content of converter accurately.
Description
Technical field:
The present invention relates to a kind of method of predicting middle-size and small-size converter terminal.
Background technology:
People just begin one's study and attempt the converter terminal that computerizeds control from the sixties in last century, have succeeded in developing the static model and the dynamic model of converter smelting.Because static model have only been considered the attitude condition at the whole story of smelting, according to the good track blowing of calculated in advance, can not consider that for the converting process factor of complexity correction is not done in the centre, therefore, the terminal point hit rate is lower.And dynamic model needs the complicated in-furnace information detection system of a cover, and majority is to use thermometric to decide the carbon sublance, constantly gives computing machine the information feedback in the stove, revise the blowing track until blowing end point, so hit rate is higher.But in order to obtain molten bath information accurately, sublance and oxygen rifle keep the certain intervals distance, if furnace port diameter is too little, then sublance is inserted and is not advanced in the stove, therefore dynamically control generally is applicable to mammoth conveter, and middle-size and small-size converter is difficult to adopt, and the sublance system is complicated, cost is higher, and practical effect is not satisfactory.In view of above reason, the automatic control of domestic middle-size and small-size converter terminal is difficult to always realize that the traditional flame of seeing of basic dependence is judged the method for smelting endpoint by rule of thumb.This method of judging terminal point by rule of thumb, it is more influenced by human factor, can not reflect the actual state of converter smelting endpoint accurately, objectively.
Artificial neural network (Artificial Neural Networks) is meant the neural network of constructing to biological learning process in fields such as information and computer science, at present, be utilized a plurality of fields.It is the function of simulation human brain, and is extensively interconnected and become a kind of complex networks system by a large amount of simple neurons, is the simplification to the human brain sophisticated functions, abstract and simulation, and its essential characteristic is as follows:
1) shows the huge complication system that as cerebral cortex, has extensively to connect;
2) computation process is based upon on the basis of large-scale parallel structure and parallel processing;
3) have stronger fault-tolerance and associative ability, be good at analogy, reasoning, summary and comprehensive;
4) to the uniting two into one property of information processing and storage;
5) have stronger self-organization and adaptivity.
Summary of the invention:
The object of the present invention is to provide a kind of method of utilizing the middle-size and small-size converter terminal of neural network prediction.
For achieving the above object, the technical solution adopted in the present invention is: selecting three layers of BP network for use is original shape, sets up two three layers of BP networks and respectively terminal temperature and endpoint carbon content is forecast that the principle of two networks is identical with algorithm.At first carry out the learning and memory of neural network: selecting some heat production datas of having produced in the recent period is sample, the a plurality of concrete manufacturing parameter of every heat is comprised terminal temperature and three layers of BP network of endpoint carbon content input, three layers of network based mathematic(al) mode of BP are calculated, calculate terminal temperature and endpoint carbon content, simultaneously, the terminal temperature that calculates and endpoint carbon content and actual terminal temperature and endpoint carbon content are compared, until calculated value and actual value differs enough little or be zero, determine three-layer network middle layer weights (wij) and threshold value (θ j); When predicting aborning, import the manufacturing parameter of this heat, in conjunction with middle layer weights (wij) and the threshold value (θ j) determined, three layers of BP network will calculate the terminal temperature and the endpoint carbon content of this heat, also just dope the terminal point of this heat again.
The BP network is the multilayer feedforward network that non-linear differentiable function is carried out weights training, is the core content of feedforward network in the artificial neural network, has embodied the elite part of artificial neural network.For the pneumatic steelmaking system, the factor that influences endpoint carbon content and temperature is a lot, and exist nonlinear relationship between each factor, as furnace life, rifle position, spraying state etc. the influence of aim carbon, temperature is difficult to be described with math equation accurately, therefore with suitable this control system that is applied to of neural network, can effectively monitor and forecast non-linear, the uncertainty and the complicacy of system.Because of beginning the time must be learnt neural network, memory training, make neural network possess predictive ability accurately, so should select the production data of abundant heat for use is sample, generally select nearest 100-150 heat for use, the manufacturing parameter of every heat also should be abundant, generally select 29 manufacturing parameters for use, these manufacturing parameters be basically all parameters during converter current is produced as: the addition of empty stove time, various furnace charges, go into stove molten iron, iron block and go up stove molten steel composition, temperature, this heat terminal temperature and endpoint carbon content etc.Neural network has possessed accurately just can instruct converter production after the predictive ability: in the production, every heat all has the terminal temperature and the endpoint carbon content of design, just terminal temperature that need reach and endpoint carbon content, if neural network is calculated terminal temperature and the endpoint carbon content that dopes to this heat manufacturing parameter, be not inconsistent with the design load of this heat, will send prompting, can be higher than design load as terminal temperature, just should in converter, increase the cooling number of poles, otherwise just reduce; Can be higher than design load as endpoint carbon content, just should increase oxygen blow duration, otherwise just reduce.In addition, the algorithm that computing machine adopts will predict the outcome automatically and legitimate reading compares, if met accuracy requirement computing machine write down automatically this group sample value, like this, computing machine can be preserved the data of the high heat of those precision automatically, deposit e-learning sample storehouse in, generate learning sample automatically, guarantee the real-time of learning sample.When proceeding to a certain degree (after having smelted the hundreds of stove) or precision of prediction, smelting can train network again when not high enough.Neural network model is constantly learnt, and constantly shakes down, and has guaranteed the forecast precision of model.
Description of drawings:
Fig. 1 is three layers of BP network diagram
Fig. 2 is the network learning procedure process flow diagram
Fig. 3 is a network forecasting process process flow diagram
Embodiment:
Fig. 1 is three layers of BP network diagram of program design, is that example is described its learning process first in conjunction with Fig. 2 with it:
Make that LA node layer ai is Wij to the connection weight between LB node layer bj, LB node layer bj is Vjt to the connection weight between LC node layer ct, and θ j is the threshold value of LB node layer, and γ t is the threshold value of LC node layer
(1) gives W
Ij, θ
j, V
Jt, γ
tCompose a less value at random.
(2) to each pattern to (A
(K), C
(K)) (k=1 ..., u), carry out following operation.
1. with A
(K)Value a
i (K)) input LA node layer, according to LA node layer activation value a
i, one time forward calculates:
2. calculate LC node layer output c
tWith desired output y
t (K)Error, the order
3. oppositely distribute error to the LB node layer, order
4. adjust connection weight V between LB layer and LC node layer
JtAnd LC node layer threshold gamma
t: V
Jt=V
Jt+ η b
jD
t
γ
t=γ
t+η·d
t (0<η<1)
5. adjust connection weight W between LA layer and LB node layer
IrAnd LB node layer threshold value θ
j:
W
ij=W
ij+η·a
i·e
j
θ
j=θ
j+η·e
j
(3) repeating step (2), until for j=1 ..., n, k=1 ..., u, error dt become enough little or vanishing.
The forecasting process of network is described in conjunction with Fig. 1, Fig. 3 again:
At first determine the parameter of input layer, according to the actual conditions of producing, the input layer of each network is provided with 29 parameters as input variable, as the addition of empty stove time, various furnace charges, go into stove molten iron, iron block and go up stove molten steel composition, temperature, this stove terminal point composition, temperature etc.
At the input port that program begins, after the operator imported these set 29 parameters, the middle layer neuron of network began according to input value input, the output valve in middle layer to be calculated, and its calculating process is described below:
With ai (i=1,2 ... 29) parameter value of expression input layer is with sj (j=1,2,30) input value in expression middle layer is with bk (k=1,2,30) each neuronic output valve of expression middle layer, according to the algorithmic formula of neural network, to bring known input layer parameter value into middle layer weights (wij) that train and threshold value (θ j) and can calculate sj and bk, algorithm is as follows:
b
k=f(s
j) k=1,2,…p
The output valve in network middle layer becomes the input value of network output layer, and the output valve of network output layer is the terminal temperature and the endpoint carbon content of prediction.The neuron of network output layer transmits the data of coming according to the middle layer input, the output valve of output layer is calculated, and its calculating process is identical with the middle layer, promptly
With c
p(p=1,2 ... 30) input value of expression output layer is with y
wAnd y
cTerminal temperature value and forecast endpoint carbon content value are forecast in expression respectively, for the temperature forecast network, then
y
w=f(c
p) p=1,2,…30
For carbon content forecast network, then
y
c=f(c
p) p=1,2,…30
In addition, in program, can add neural Network Model for Forecasting and the code that diffusing shape material charging system software is connected, realize by the diffusing shape material auto feed process of forecasting model control.Neural Network Model for Forecasting is connected with molten iron and steel scrap Weighing system software, the shape material charging system software that looses, according to the determined various raw-material additions of neural Network Model for Forecasting, and the addition data are stored with the excel form, by charging system software reading of data from forecasting model automatically, by the reinforced process of computer controlled automatic.
Claims (1)
1, a kind of method of utilizing artificial neural network technology prediction converter terminal, it is characterized in that: selecting three layers of BP network for use is original shape, sets up two three layers of BP networks and respectively terminal temperature and endpoint carbon content is forecast that the principle of two networks is identical with algorithm; 1) at first carry out the learning and memory of neural network: selecting some heat production datas of having produced in the recent period is sample, several concrete manufacturing parameters of every heat are comprised terminal temperature and three layers of BP network of endpoint carbon content input, three layers of network based mathematic(al) mode of BP are calculated, calculate terminal temperature and endpoint carbon content, simultaneously, the terminal temperature that calculates and endpoint carbon content and actual terminal temperature and endpoint carbon content are compared, until calculated value and actual value differs enough little or be zero, determine three-layer network middle layer weights (wij) and threshold value (θ j); When 2) predicting aborning, import the manufacturing parameter of this heat, in conjunction with middle layer weights (wij) and the threshold value (θ j) determined, three layers of BP network will calculate terminal temperature and the endpoint carbon content that dopes this heat again.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 200410056892 CN1588346A (en) | 2004-08-30 | 2004-08-30 | Method for predicting converter terminal point using artificial nurve network technology |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 200410056892 CN1588346A (en) | 2004-08-30 | 2004-08-30 | Method for predicting converter terminal point using artificial nurve network technology |
Publications (1)
Publication Number | Publication Date |
---|---|
CN1588346A true CN1588346A (en) | 2005-03-02 |
Family
ID=34603244
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN 200410056892 Pending CN1588346A (en) | 2004-08-30 | 2004-08-30 | Method for predicting converter terminal point using artificial nurve network technology |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN1588346A (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN100554982C (en) * | 2006-06-29 | 2009-10-28 | 南京南瑞继保电气有限公司 | Utilize voltage difference to judge the method that the hvdc transmission line open-circuit test is unusual |
CN102033978A (en) * | 2010-09-19 | 2011-04-27 | 首钢总公司 | Method for forecasting and producing narrow hardenability strip steel by hardenability |
CN102373310A (en) * | 2011-10-19 | 2012-03-14 | 北京科技大学 | Method for guiding converter reblowing process operation |
CN102392095A (en) * | 2011-10-21 | 2012-03-28 | 湖南镭目科技有限公司 | Termination point prediction method and system for converter steelmaking |
CN102393884A (en) * | 2011-10-12 | 2012-03-28 | 杭州电子科技大学 | Hot continuous rolling electromagnetic induction heating temperature prediction method based on BP (back-propagation) neural network |
CN101698896B (en) * | 2009-09-28 | 2013-01-30 | 南京理工大学 | System and method for steel-making online end-point control through furnace mouth radiation information fusion |
CN103645633A (en) * | 2013-12-25 | 2014-03-19 | 中国科学院自动化研究所 | Furnace temperature self-learning control method of conversion furnace system |
CN104864984A (en) * | 2015-05-21 | 2015-08-26 | 青岛大学 | Method for measuring minimal reactor temperature based on nerve network |
CN105925750A (en) * | 2016-05-13 | 2016-09-07 | 南阳理工学院 | Steelmaking end point prediction method based on neural networks |
CN108251591A (en) * | 2018-01-15 | 2018-07-06 | 上海大学 | Utilize the top bottom blowing converter producing process control method of LSTM systems |
CN109234491A (en) * | 2018-11-20 | 2019-01-18 | 北京科技大学 | A kind of BOF Steelmaking Endpoint manganese content prediction technique based on extreme learning machine |
CN109797291A (en) * | 2018-12-06 | 2019-05-24 | 江西力沃德科技有限公司 | A kind of converter copper making period terminal intelligent decision system |
CN113061683A (en) * | 2021-03-16 | 2021-07-02 | 马鞍山钢铁股份有限公司 | Automatic matching method for converter end point oxygen and converter end point reblowing times quality factor |
CN113661257A (en) * | 2019-04-02 | 2021-11-16 | 杰富意钢铁株式会社 | Method for predicting splash in converter, method for operating converter, and system for predicting splash in converter |
-
2004
- 2004-08-30 CN CN 200410056892 patent/CN1588346A/en active Pending
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN100554982C (en) * | 2006-06-29 | 2009-10-28 | 南京南瑞继保电气有限公司 | Utilize voltage difference to judge the method that the hvdc transmission line open-circuit test is unusual |
CN101698896B (en) * | 2009-09-28 | 2013-01-30 | 南京理工大学 | System and method for steel-making online end-point control through furnace mouth radiation information fusion |
CN102033978B (en) * | 2010-09-19 | 2012-07-25 | 首钢总公司 | Method for forecasting and producing narrow hardenability strip steel by hardenability |
CN102033978A (en) * | 2010-09-19 | 2011-04-27 | 首钢总公司 | Method for forecasting and producing narrow hardenability strip steel by hardenability |
CN102393884B (en) * | 2011-10-12 | 2015-04-15 | 杭州电子科技大学 | Hot continuous rolling electromagnetic induction heating temperature prediction method based on BP (back-propagation) neural network |
CN102393884A (en) * | 2011-10-12 | 2012-03-28 | 杭州电子科技大学 | Hot continuous rolling electromagnetic induction heating temperature prediction method based on BP (back-propagation) neural network |
CN102373310A (en) * | 2011-10-19 | 2012-03-14 | 北京科技大学 | Method for guiding converter reblowing process operation |
CN102373310B (en) * | 2011-10-19 | 2013-06-12 | 北京科技大学 | Method for guiding converter reblowing process operation |
CN102392095A (en) * | 2011-10-21 | 2012-03-28 | 湖南镭目科技有限公司 | Termination point prediction method and system for converter steelmaking |
CN102392095B (en) * | 2011-10-21 | 2013-09-11 | 湖南镭目科技有限公司 | Termination point prediction method and system for converter steelmaking |
CN103645633B (en) * | 2013-12-25 | 2017-01-18 | 中国科学院自动化研究所 | Furnace temperature self-learning control method of conversion furnace system |
CN103645633A (en) * | 2013-12-25 | 2014-03-19 | 中国科学院自动化研究所 | Furnace temperature self-learning control method of conversion furnace system |
CN104864984A (en) * | 2015-05-21 | 2015-08-26 | 青岛大学 | Method for measuring minimal reactor temperature based on nerve network |
CN104864984B (en) * | 2015-05-21 | 2017-04-05 | 青岛大学 | Microresponse device thermometry based on neutral net |
CN105925750A (en) * | 2016-05-13 | 2016-09-07 | 南阳理工学院 | Steelmaking end point prediction method based on neural networks |
CN108251591A (en) * | 2018-01-15 | 2018-07-06 | 上海大学 | Utilize the top bottom blowing converter producing process control method of LSTM systems |
CN109234491A (en) * | 2018-11-20 | 2019-01-18 | 北京科技大学 | A kind of BOF Steelmaking Endpoint manganese content prediction technique based on extreme learning machine |
CN109797291A (en) * | 2018-12-06 | 2019-05-24 | 江西力沃德科技有限公司 | A kind of converter copper making period terminal intelligent decision system |
CN109797291B (en) * | 2018-12-06 | 2020-11-20 | 江西力沃德科技有限公司 | Intelligent determination system for converter copper making period end point |
CN113661257A (en) * | 2019-04-02 | 2021-11-16 | 杰富意钢铁株式会社 | Method for predicting splash in converter, method for operating converter, and system for predicting splash in converter |
CN113061683A (en) * | 2021-03-16 | 2021-07-02 | 马鞍山钢铁股份有限公司 | Automatic matching method for converter end point oxygen and converter end point reblowing times quality factor |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN1588346A (en) | Method for predicting converter terminal point using artificial nurve network technology | |
WO2021189739A1 (en) | Production plan-based iron and steel plant oxygen load prediction method | |
Anh et al. | Hybrid control of a pneumatic artificial muscle (PAM) robot arm using an inverse NARX fuzzy model | |
CN110362084A (en) | A kind of vehicle route tracing control method based on ant group algorithm | |
CN103631140B (en) | Based on the coke oven heating-combustion process fire path temperature Automatic adjustment method of Performance Evaluation | |
CN110386768B (en) | Dynamic real-time control method for energy consumption in cement sintering process | |
WO2022121944A1 (en) | Method and apparatus for intelligent forecasting of industrial process operation indicators, device, and storage medium | |
WO2023070293A1 (en) | Long-term scheduling method for industrial byproduct gas system | |
CN114678080B (en) | Converter end point phosphorus content prediction model, construction method and phosphorus content prediction method | |
CN111562744B (en) | Boiler combustion implicit generalized predictive control method based on PSO algorithm | |
CN112132379B (en) | Economic-considered new energy cross-region consumption evaluation method and storage medium | |
CN113962050A (en) | Oxygen scheduling calculation method combining production consumption prediction and pipe network calculation | |
CN110674965A (en) | Multi-time step wind power prediction method based on dynamic feature selection | |
CN105447567A (en) | BP neural network and MPSO algorithm-based aluminium electrolysis energy-saving and emission-reduction control method | |
CN111650834A (en) | Sewage treatment process prediction control method based on Extreme Learning Machine (ELM) | |
CN109581864A (en) | The inclined format non-model control method of the different factor of the MIMO of parameter self-tuning | |
CN111367181A (en) | Hybrid drive intelligent judgment control method for sewage treatment system | |
Fu et al. | Dissolved oxygen control system based on the TS fuzzy neural network | |
CN114202063A (en) | Fuzzy neural network greenhouse temperature prediction method based on genetic algorithm optimization | |
CN107831666B (en) | Natural gas absorption tower desulfurization process control method based on RBF and ADDHP | |
CN105426960A (en) | BP neural network and MBFO algorithm-based aluminum electrolysis energy conservation and emission reduction control method | |
Li et al. | Variable structure control for 6-6 parallel manipulators based on cascaded CMAC | |
CN114036819B (en) | Sewage total phosphorus soft measurement method and system based on self-organizing random configuration network | |
CN107918282A (en) | Variable domain MPC methods for interconnected network Automatic Generation Control | |
CN107885083B (en) | Natural gas absorption tower desulfurization process control method based on UKF and ADHDP |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |