CN106153550A - Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection method - Google Patents
Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection method Download PDFInfo
- Publication number
- CN106153550A CN106153550A CN201510170782.8A CN201510170782A CN106153550A CN 106153550 A CN106153550 A CN 106153550A CN 201510170782 A CN201510170782 A CN 201510170782A CN 106153550 A CN106153550 A CN 106153550A
- Authority
- CN
- China
- Prior art keywords
- carbon content
- svm
- model
- flame
- molten steel
- 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
Landscapes
- Carbon Steel Or Casting Steel Manufacturing (AREA)
Abstract
The present invention provides a kind of converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection method, including: according to the field stop in pneumatic steelmaking environment regulation telescope optical system, thus the visual field regulating fire door flame detecting make the predeterminated position of fire door flame by this telescope optical system imaging, then via Transmission Fibers by fire door flame image information transmit to a spectrogrph;Spectrogrph receives as afterwards, carrying out spectrum analysis acquisition flame spectrum distributed intelligence utilizes end-point control method based on SVM to carry out the real-time detection of pneumatic steelmaking carbon content according to flame spectrum information, and wherein carbon content is detected by the dynamic detection model of SVM carbon content.Detection method proposed by the invention, on-the-spot test precision is high, and is affected little by outside environmental elements, and capacity of resisting disturbance is strong.
Description
Technical field
Various aspects of the invention relate to the real-time prison of carbon content of molten steel in converter steeling technology field, especially convertor steelmaking process
Survey, in particular to converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection method.
Background technology
Main flow steel-smelting technology the most in the world is exactly pneumatic steelmaking, and its yield accounts for more than the 70% of iron and steel total output.And in converter
In steelmaking process, a most important ring is exactly the terminal point control in latter stage, is directly connected to the quality of last molten steel.Since pneumatic steelmaking
Since method occurs, the terminal point control of pneumatic steelmaking mainly experienced by artificial experience control, static model control, dynamic model control
System and optical information control four developmental stage.
Artificial experience controls, i.e. experience steel-making, utilizes thermocouple temperature measurement to determine carbon and the means of stokehold sampling quickly analysis, to normally
Converter terminal under blowing condition carries out artificial experience and judges to control.Reaction between carbon and oxygen speed is the important evidence of divided into three stages,
And the severe degree of reaction between carbon and oxygen and the temperature of molten steel, can be reflected by fire door flame.Steel making operation workman is by observing
Fire door flame, spark and oxygen supply time carry out comprehensive descision steel-making terminal.But, rely solely on the perusal of operator, deposit
The problem such as low at Target hit rate, labor strength is big.
Static model control is exactly according to statistical principle, and the primary data in the past bessemerized carries out statistical analysis, calculates
Go out the initial condition required for blowing, carry out converting process with this condition.In general, static model control compares artificial warp
Testing control can more efficiently utilize the initial condition of converting process to carry out quantitative Analysis and control.Static model control can depend on
Find optimum material proportion according to material condition, and determine smelting scheme according to actual dispensing, overcome randomness in experience control with
Discordance.Existing static models include mechanism model, statistical model and incremental model three kinds.And in actual applications, often
Often combine to improve the precision of terminal point control with these three model.But do not consider in converting process due to static model control
Multidate information, it is impossible to carrying out on-line tracing and revise in real time, therefore accuracy is very restricted.
Model controlling is mainly sublance dynamic control method, utilizes sublance to the molten steel in converter on the basis of static models
Detect, the result obtained according to detection, initial parameter is revised, obtains accurate terminal.The most in recent years,
Along with the research application in model controlling method of artificial neural network, overcome traditional static model cootrol and ignore blowing
During the problem of multidate information, further increase the accuracy of detection, make the hit rate of end point determination result obtain entering one
The lifting of step, makes the automaticity of steel-making be greatly improved simultaneously.But its cost is higher, need converter is entered
Row transformation, thus the most applicable to general mini-medium BOF plants.
Traditional method or inaccurate to endpoint, or cost high-adaptability is limited, therefore along with the development of steel-smelting technology and relevant
The progress of technology, people continuously attempt to apply more efficient and method accurately in terminal point control technology.80 years 20th century
In generation, occur in that the novel end-point control method utilizing converter mouth optical information that BOF Steelmaking Endpoint is judged.Such as, utilize red
The situation of change occurred during outer laser penetration furnace gas measures composition of fumes to judge the optical detector of terminal, and this detector passes through
Detecting the situation of change occurred through furnace gas laser and judge terminal, its cardinal principle is the content of the carbon monoxide in detection furnace gas,
Composition transfer according to the carbon monoxide in furnace gas carries out terminal point control.In experience or model controlling, all the time can not
Ignore is exactly that operator to obtain information in various degree from the change of flame, these information be exactly in fact flame aperture,
Spectral distribution and the image information of flame.Along with the development of photoelectric device, the continuous maturation of optical processing method, optics is visited
Survey technology has obtained great development, and optical control method has been also applied in the terminal point control of pneumatic steelmaking.Such as Zhang Jinjin,
The fire door flame intensity signal that the molten steel radiation spectral information probe technique of Shi Yanjie et al. proposition, Bethlehem Steel Company of the U.S. propose
The flame image information detection method etc. that probe technique, Wei Chengye, tight Jian Hua et al. propose.
Although the research of steel-making terminal point control theory deepens continuously, but the cost needed for these methods is high, detection and analytical equipment
Cost be all extremely expensive, and install and maintenance be inconvenient for, only in some powerful iron and steel enterprises apply.
In most of medium or small sized steel company, or control or static model control based on single experience.And up-to-date optical information
Although control method provides some valuable thinking and application directions, but owing to being limited by production scale, working condition,
The most complicated, severe STEELMAKING PRODUCTION environment, in terms of optical information collection, capacity of resisting disturbance is weak, it is impossible to continuous print carries rapidly
Take required parameter information, thus be difficult to some medium or small sized steel company and accepted.
Therefore, a kind of accurate in the urgent need to developing, it is adaptable to medium or small sized steel company, the online terminal of making steel in real time of middle primary converter
Control program.
Summary of the invention
Present invention aim at providing a kind of converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection method, should
Method designs based on algorithm of support vector machine, has carbon content accuracy of detection height, noncontact, capacity of resisting disturbance strong, easily operated
Etc. advantage, thus solve the problem in terms of current pneumatic steelmaking carbon content dynamic on-line monitoring.
The above-mentioned purpose of the present invention is realized by the technical characteristic of independent claims, and dependent claims is with alternative or favourable side
The technical characteristic of formula development independent claims.
For reaching above-mentioned purpose, the present invention proposes a kind of converter steel-smelting molten steel online Real-time and Dynamic Detection of carbon content based on SVM
Method, the realization of this detection method includes:
According to the field stop in pneumatic steelmaking environment regulation telescope optical system, thus the visual field regulating fire door flame detecting makes
The predeterminated position of fire door flame is by this telescope optical system imaging, then fire door flame image information is transmitted via a Transmission Fibers
To a spectrogrph;
Spectrogrph receives as afterwards, carries out spectrum analysis and obtains flame spectrum distributed intelligence;
Utilize the real-time detection carrying out pneumatic steelmaking carbon content based on SVM end-point control method according to flame spectrum distributed intelligence,
Wherein carbon content is detected by the dynamic detection model of SVM carbon content.
In further example, described field stop is variable field of view light hurdle.
In further example, the detection in real time of described pneumatic steelmaking carbon content stores program in computer systems by operation
Realize, including procedure below:
The characteristic parameter of carbon content change in receiving flame spectrum information and building sign stove;
The characteristic parameter of carbon content change in sign stove is inputted a dynamic detection model of SVM carbon content detect;And
The testing result of output carbon content.
In further example, the described dynamic detection model of SVM carbon content is the model that training in advance is good, and its training process is such as
Under:
Using actual carbon content of molten steel as standard, by repetition training, optimized choice, determine involved by SVM learning algorithm
Each parameter, it specifically includes:
The characteristic parameter of carbon content change in stove can be characterized by flame spectrum information architecture;
The kernel function of selected SVM learning algorithm;
Optimal control parameter kernel function width δ and penalty factor;
Selection Model training sample, utilizes SVM learning algorithm that characteristic parameter is carried out classification model construction;
Input the model set up with test sample, and whether analytical error and generalization meet design and require: if it is satisfied, then
Output model, if be unsatisfactory for, then returns described step and re-starts the selection of kernel function width δ and penalty factor with again
Modeling, until meeting requirement.
In further example, during described model training, construction feature parameter in the following manner:
At wavelength 600nm, spectral shape is protruding spike, characteristic parameter a1For light intensity normalized value herein;
Spectral shape spike of projection at 770nm is bimodal, characteristic parameter a2For the light intensity at wavelength 770nm and 772nm
Normalization average value;
This section of spectral line acutely, is divided into three sections, to each section of light intensity normalizing by the continuous spectrum change in the middle of said two spike
Average after change and obtain three characteristic parameter a3, a4, a5;And
Using the ratio of the peak wavelength λ of spectrum in spectral distribution and investigative range maximum of T max of described spectrogrph as the 6th
Individual parameter: a6。
In further example, during described model training, the kernel function of described SVM learning algorithm is selected from linear kernel letter
Number, Polynomial kernel function, the one in RBF kernel function and S type kernel function.
In further example, during the detection of carbon content, receiving flame spectrum information the structure of online real time collecting
After building characteristic parameter, first pass through described SVM carbon content dynamic prediction model and determine the classification of aim carbon, and based on aim carbon
Classification use corresponding terminal fitting function to determine the carbon content of molten steel corresponding to current gathered flame spectrum information.
In further example, described terminal fitting function include the classification of different aim carbon each belonging to terminal fitting function,
Wherein:
Described terminal fitting function is expressed as:
Y=f (X),
This formula have expressed the mapping relations of X Yu Y, and the characteristic variable extracted during wherein X is terminal moment flame spectrum, Y is
Terminal carbon value, this terminal fitting function uses MATLAB to provide a polynomial fit function to be fitted data, thus
Obtain fitting function.
As long as should be appreciated that all combinations of described design and the extra design described in greater detail below are at such structure
Think the most conflicting in the case of can be viewed as the part of subject matter of the disclosure.It addition, theme required for protection
All combinations be considered as the part of subject matter of the disclosure.
Described and other aspects that can be more fully appreciated with from the following description in conjunction with accompanying drawing present invention teach that, embodiment and
Feature.Feature and/or the beneficial effect of other additional aspect such as illustrative embodiments of the present invention will show in the following description
See, or by the practice according to the detailed description of the invention that present invention teach that is learnt.
Accompanying drawing explanation
Accompanying drawing is not intended to drawn to scale.In the accompanying drawings, each identical or approximately uniform ingredient illustrated in each figure
Can be indicated by the same numeral.For clarity, in each figure, the most each ingredient is the most labeled.Now,
By by example embodiment that various aspects of the invention are described in reference to the drawings, wherein:
Fig. 1 is the signal of converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection system proposed by the invention
Figure.
Fig. 2 be Fig. 1 detection method in the schematic diagram of telescope optical system.
Fig. 3 is the flow process of the converter steel-smelting molten steel carbon content online Real-time and Dynamic Detection method realized based on detecting system shown in Fig. 1
Schematic diagram.
Fig. 4 is the schematic flow sheet of the svm classifier modeling that the present invention proposes.
Fig. 5 is the data decimation scope schematic diagram that the present invention proposes.
When Fig. 6 a-6b is penalty factor=30 and kernel function width δ=0.2, the classification hit results of training sample and test sample is shown
It is intended to.
When Fig. 7 a-8b is penalty factor=20 and kernel function width δ=0.8, the classification hit results of training sample and test sample is shown
It is intended to.
When Fig. 8 a-8b is penalty factor=20 and kernel function width δ=3, the classification hit results of training sample and test sample is shown
It is intended to.
Training sample and the classification hit results of test sample when Fig. 9 a-9b is penalty factor=100 and kernel function width δ=0.06
Schematic diagram.
Figure 10 is the disaggregated model result schematic diagram of 30 carbon training.
Figure 11 is the disaggregated model result schematic diagram of final training.
Figure 12 is the workflow schematic diagram after reality starts of the detection method shown in Fig. 1 of the present invention.
Figure 13 is the carbon content testing process schematic diagram that the present invention proposes.
Detailed description of the invention
In order to know more about the technology contents of the present invention, especially exemplified by specific embodiment and coordinate institute's accompanying drawings to be described as follows.
Each side the most with reference to the accompanying drawings to describe the present invention, the embodiment illustrated shown in the drawings of many.The disclosure
Embodiment must not be intended to include all aspects of the invention.Should be appreciated that multiple design presented hereinbefore and embodiment, with
And describe in more detail below those design and embodiment can in many ways in any one is implemented, this is because
Design disclosed in this invention and embodiment are not limited to any embodiment.It addition, aspects more disclosed by the invention can be single
Solely use, or otherwise any appropriately combined use with disclosed by the invention.
Converter steel-smelting molten steel carbon content online Real-time and Dynamic Detection method proposed by the invention, on the whole, is to be by looking in the distance
System obtains fire door flame image information at a distance, and utilizes spectrogrph to be analyzed obtaining flame spectrum information to it, accordingly,
Therefrom extract the useful information that i.e. can express carbon content and carry out construction feature parameter, and the training sample chosen is carried out SVM instruction
Get detection model, finally utilize detection model to detect the sample data that on-line checking obtains, thus be finally reached in stove
The on-line real-time measuremen of carbon content of molten steel.
Operation, as requested institute is bessemerized accordingly according to the difference of institute's steelmaking kind owing to present pneumatic steelmaking is usually
The steel grade of blowing is divided into mild steel (in molten steel the content of carbon be less than 1.5%), medium carbon steel (in molten steel the content of carbon 1.5%~3.0%)
With three kinds of high-carbon steel (in molten steel, the content of carbon is more than 3.0%), 15 carbon reality mentioned in present disclosure
On be 15/1000ths, i.e. carbon content is 1.5%, and same 30 carbon are also so to represent, 30 carbon models described below are exactly
The model obtained by classification based training is carried out near some converters that carbon content is 3.0%.
In conjunction with converter steel-smelting molten steel carbon content based on the SVM online Real-time and Dynamic Detection system of Fig. 1, including optical system of looking in the distance
System 1, spectrogrph 2 and terminal point control device 3 based on SVM.
It is connected by optical fiber 4 between telescope optical system 1 with spectrogrph 2.
Telescope optical system 1, is arranged to the flame image information of Real-time Collection steel-making fire door.
Spectrogrph 2, is configured to receive the flame image information from telescope optical system 1 by optical fiber 4, and obtains flame figure
Flame spectrum information as information.
Spectrogrph 2, has selected grating spectrograph, the miniature CCD of the USB4000-VIS-NIR of such as marine optics in this example
Grating spectrograph, its volume is little, failure rate is low, and easy for installation, and the telescope optical system designed with this example coordinates and can stably obtain
Obtain the stable spectra of fire door flame.
Terminal point control device 3 based on SVM, this device has an arithmetic element 31 and controls central authorities' control that arithmetic element is run
Unit 32 processed, the flame spectrum information that this arithmetic element is arranged for according to described real-time acquisition is dynamic by SVM carbon content
Detection model carries out the real-time detection of carbon content in steel-making molten steel.
In this example, terminal point control device based on SVM is configured to a circuit board.It is integrated with as arithmetic element on circuit board
Fpga chip and as the microprocessor of CPU, certainly, circuit board also includes for providing burning voltage to supply
The interfaces such as the power module answered, serial line interface, RS232 interface.
Described SVM carbon content dynamic detection model burning is in described fpga chip, and is receiving flame spectrum letter
Automatically the detection of carbon content is carried out after breath.
Certainly, in the embodiment of alternative, described arithmetic element can also be realized by CPLD chip.
In conjunction with Fig. 2, described telescope optical system 1 includes the object lens of common optical axis, eyepiece and independent of object lens and the visual field of eyepiece
Light hurdle, in the optical imagery path that this field stop is arranged in described object lens, eyepiece is formed, is used for regulating fire door flame detecting
Visual field.
As shown in Figure 2, it is preferable that field stop is arranged on object lens between eyepiece, is such as positioned on the focal plane of described object lens.
In other example, described field stop is positioned at described eyepiece rear and presses close to described optical fiber.
As optional example, described object lens are air-spaced doublet, by one piece of plus lens and the distribution of one piece of minus lens common optical axis
Constitute.
Described eyepiece is Kellner eyepiece, is made up of one piece of simple lens and the distribution of one piece of cemented doublet common optical axis.
Preferably, described field stop is variable field of view light hurdle.
As in figure 2 it is shown, label l represents the optical axis of object lens, eyepiece, f '1Represent the focal length of object lens, f '2Represent the focal length of eyepiece.
As it was previously stated, described terminal point control device 3 based on SVM, arithmetic element 31 therein is by FPGA, CPLD
A kind of realization, the dynamic detection model burning of SVM carbon content is in FPGA or CPLD, and is receiving flame spectrum letter
Controlled automatically to carry out the detection of carbon content by central control unit 32 after breath.
In this example, the described dynamic detection model of SVM carbon content includes:
In in the flame spectrum information according to input, structure characterizes stove, the parameter of the characteristic parameter of carbon content change builds module;
The carbon content dynamic detection module of carbon content detection is carried out for characteristic parameter based on described structure;And
Output module for testing result output.
In conjunction with the flow chart shown in Fig. 1, Fig. 2 and Fig. 3, according to the present invention, a kind of converter steel-smelting molten steel based on SVM is proposed
Carbon content online Real-time and Dynamic Detection method, the realization of this detection method includes:
According to the field stop in pneumatic steelmaking environment regulation telescope optical system 1, thus the visual field regulating fire door flame detecting makes
The predeterminated position of fire door flame by this telescope optical system 1 imaging, then via Transmission Fibers, fire door flame image information is passed
Transport to a spectrogrph 2;
Spectrogrph 2 receives as afterwards, carries out spectrum analysis and obtains flame spectrum distributed intelligence
Utilize the real-time detection carrying out pneumatic steelmaking carbon content based on SVM end-point control method according to flame spectrum information, wherein
Carbon content is detected by the dynamic detection model of SVM carbon content.
Preferably, the field stop being previously used for regulating fire door flame detecting is variable field of view light hurdle.
In this example, shown in Fig. 3, the detection in real time of pneumatic steelmaking carbon content stores program in a computer by operation
Realize, including procedure below:
The characteristic parameter of carbon content change in receiving flame spectrum information and building sign stove;
The characteristic parameter of carbon content change in sign stove is inputted a dynamic detection model of SVM carbon content detect;And
The testing result of output carbon content.
The dynamic detection model of SVM carbon content is the model that training in advance is good, and its training process is as follows:
Using actual carbon content of molten steel as standard, by repetition training, optimized choice, determine involved by SVM learning algorithm
Each parameter, it specifically includes:
The characteristic parameter of carbon content change in stove can be characterized by flame spectrum information architecture;
The kernel function of selected SVM learning algorithm;
Optimal control parameter kernel function width δ and penalty factor;
Selection Model training sample, utilizes SVM learning algorithm that characteristic parameter is carried out classification model construction;
Input the model set up with test sample, and whether analytical error and generalization meet design and require: if it is satisfied, then
Output model, if be unsatisfactory for, then returns described step and re-starts the selection of kernel function width δ and penalty factor with again
Modeling, until meeting requirement.
And during model training, construction feature parameter in the following manner:
At wavelength 600nm, spectral shape is protruding spike, characteristic parameter a1For light intensity normalized value herein;
Spectral shape spike of projection at 770nm is bimodal, characteristic parameter a2For the light intensity at wavelength 770nm and 772nm
Normalization average value;
This section of spectral line acutely, is divided into three sections, to each section of light intensity normalizing by the continuous spectrum change in the middle of said two spike
Average after change and obtain three characteristic parameter a3, a4, a5;And
Using the ratio of the peak wavelength λ of spectrum in spectral distribution and investigative range maximum of T max of described spectrogrph as the 6th
Individual parameter: a6。
The kernel function of described SVM learning algorithm is selected from linear kernel function, Polynomial kernel function, RBF kernel function and S type
One in kernel function.
When actual model generation, by training to be tested is divided into three classes so that the carbon value span scope of data reduces.
As it was previously stated, when carrying out model training and setting up, i.e. use the mode of classification model construction to carry out.
In this example, shown in Fig. 4, classification model construction implements and includes:
1) training and the choosing of test sample
Collect 100 flame spectrum data of converter mouth flamew from steel mill, these 100 spectroscopic datas classified,
<15,15≤C<30, C>=30 three major types, learns from svm classifier algorithm, asks for many classification to be divided into carbon content C
Topic, can convert it into several two classification problems, this simplifies the difficulty of algorithm, reduce the polytypic time simultaneously
Complexity.For purposes of the invention, it is evident that be many classification problems, therefore can solve with two two disaggregated models, in training
Time only two models need to be trained near 30 carbon and 15 carbon to can reach the purpose that data are divided into three classes.
Exemplarily, it is assumed that carbon in the range of between 0~50, choosing as shown in Figure 5 of concrete training data.
Therefore, for the training of 30 carbon disaggregated models, the scope of its two classes respectively: between 0~30 carbon and 30
More than carbon (maximum recorded is near 50 carbon).After determining two classes, then determine for training and the sample of test,
The number of training sample and test sample selects generally according to the ratio of 2:1 or 3:1, had better not number maintain an equal level, the most unfavorable
In the generalization of model, the number of training sample is more preferably greater than the number of test sample.
Similarly for the training of 15 carbon models, its two classes are respectively as follows: between 15~30 carbon and below 15 carbon, choosing
It is the same for selecting training with the number of test sample and 30 carbon Model Selection rules.
After here sample determines, be not unalterable, it is possible to choose sample when disaggregated model training, nothing
How opinion is done all does not reaches requirement.At this moment it is accomplished by changing sample, the heat not having hit in test sample is put into training
Re-training in sample, needs exist for illustrating a bit, if changing training sample at this moment, then the model of subsequent time
Training, sample also changes, say, that training sample the most in the same time is consistent.
2) structure of characteristic variable
Fire door light intensity can characterize the brightness of flame, and the spectral information of fire door flame can characterize the state in converter.In converter
In steelmaking process, in stove, the state of molten steel always changes to next state from a state, therefore can be by a value quantified
Represent it, i.e. can represent with Yd that the status information in converter, this value contain the temperature in stove and composition information.?
Here Yd can be discrete, it is also possible to is continuous print.When observing the variation tendency of flame spectrum, do not find spectrum
Abnormal sudden change, however, it was found that spectrum can occur slight change on some wave band, this change latter stage can be brighter in blowing
Aobvious, the mark of endpoint may be become.
Therefore, this example utilizes M (λ) represent the spectral distribution of fire door flame, and thinks that Yd is a continually varying value,
Then with a mapping, independent variable (i.e. the value of M (λ)) can be converted into dependent variable (i.e. Yd value), obtain following functional relationship
Formula: f (M (λ))=Yd.
Molten steel state in converting process is relevant with the spectrum of fire door flame, i.e. can detect steel in stove by the distribution of spectrum
The state of water.Therefore spectral distribution is done pretreatment, ignore the absolute value impact of spectral intensity, only consider spectral distribution
The change impact on molten steel state.
For the spectroscopic data in a certain moment, can process by equation below:
In above formula, M (λi) represent in wavelength XiThe absolute value of place's spectrum intensity, its effect is to be normalized spectrum intensity
Processing, after processing, the shape of spectral distribution does not change, but in the whole converting process of converter its value 0~1 it
Between.
Wherein wavelength XiSpan depend on the investigative range of spectrogrph 2, in this example, the detection model of the spectrogrph used
Enclose is [350,1000].
In research process, having obvious two protruding spikes in flame spectrum distribution, in repeatedly research, they are whole
Converting process changes the most regular, especially blow latter stage near terminal time, their shape can represent to a certain extent
The state of molten steel in stove.The wavelength of its correspondence is respectively 600nm and 770nm, 772nm, therefore in the present invention this two
The spectral normalization value at place is as characteristic variable a1, a2.I.e. a1=M'(600), owing to the spike at 700nm is bimodal,
So it is averaged, i.e. light intensity normalization average value at calculating 770nm and 772nm:
Also finding during repeatedly research, the continuous spectrum change in the middle of two spikes also the most acutely, among these can be anti-
Reflecting the change of flame brightness, in order to obtain characterizing the characteristic variable of this section of spectral line, we are in this example by this section of spectral line
It is divided into three sections, each section of light intensity normalized value is averaged, such that it is able to obtain three characteristic parameter: a3, a4,
a5。
The peak value of spectral distribution can also reflect certain information, but the light intensity normalized value of peak value is at parameter a4In be able to
Reflection, when choosing parameter, first have to guarantee is to be separate between parameter, it is impossible to exist in which that two parameters are permissible
Reflect the situation of mutually reflection between another parameter, or parameter, so need not repeat to choose.
For this situation, in this example, by the investigative range maximum of the peak wavelength of spectrum in spectral distribution Yu described spectrogrph
Ratio as the 6th parameter: a6。
Therefore, in the present embodiment, described parameter builds module and is configured to construction feature parameter in the following manner:
At wavelength 600nm, spectral shape is protruding spike, characteristic parameter a1For light intensity normalized value herein;
Spectral shape spike of projection at 770nm is bimodal, characteristic parameter a2For the light intensity at wavelength 770nm and 772nm
Normalization average value;
This section of spectral line acutely, is divided into three sections, to each section of light intensity normalizing by the continuous spectrum change in the middle of said two spike
Average after change and obtain three characteristic parameter a3, a4, a5;
Using the ratio of the peak wavelength of spectrum in spectral distribution and the investigative range maximum of described spectrogrph as the 6th parameter:
a6。
By analysis introduction above, the value scope of 6 parameters, all between [0,1], this is done to keep the sane of algorithm
Property.Having chosen complete to spectral parameter here, parameter can be calculated by spectrum and obtain, for N number of training sample,
Each training sample comprises many data points xi=(a1,a2,a3,a4,a5,a6), i=1 ..., N, each data point contains 6
Next parameter, be then just introduced into sample parameter, utilize SVM to carry out the training of disaggregated model.
3) the choosing of training parameter
When here sample being carried out classification based training, choose sample and to follow a condition: sample is independent identically distributed.Through dividing
Analysing, the information of each moment spectrum is the state in reflection stove, and for the process of bessemerizing, each state is phase
It is the most independent, say, that each moment flame spectrum information is also separate, is the relation of a kind of correspondence between them,
And then be independent between available each sample point.Again because the spectral distribution in each moment is all to be obtained by same agent burns
Come, identical physical characteristic must be met, so they meet same probability distribution.In sum, it is believed that choose
Training sample is independent identically distributed.
The corresponding different network of kernel functions different during SVM training, in this example, the kernel function of SVM learning algorithm is selected from line
Property kernel function, Polynomial kernel function, the one in RBF kernel function and S type kernel function.
Linear kernel function: K (x, x')=<x, x'>;
Heterogeneous formula kernel function: K (x, x')=(<x, x'>+1)dWherein d is arithmetic number;
S type (sigmoid) kernel function: K (x, x')=thanh (v<x, x'>+r) wherein v and r is normal number.
Gauss (RBF, RBF) kernel function: K (x, x')=exp (-| | x-x'| |2/2σ2) wherein σ be core width, and be positive integer.
(reference: A Practical Guide to Support Vector Classi fication page 2, Hsu C W, Chang C C, Lin C J.
A practical guide to support vector classification[J].2003.)
In actual applications, generally suitable kernel function and parameter are selected according to the concrete condition of problem.For finding different converter
Corresponding most suitable SVM model kernel function, can respectively to the sample spectrum data acquisition of different converters with above 4 kinds of core letters
Number training, the kernel function that then selection training result is best is as this converter SVM model kernel function
Svm classifier is exactly after selected kernel function, finds the hyperplane having largest interval in kernel function space.
In this example, utilization is the training that carries out disaggregated model of LS_SVM algorithm, and the optimization obtaining former problem is defined as:
Subject to:yi(<ω·xi>+b)≥1-ξi, i=1 ..., l [remarks: support vector machine introduction page 91]
In above formula, C is to need specified value, and it is the penalty factor of sample error.
Ensuing work is exactly training pattern, and concrete step is: import the spectral information of training sample in chronological order, also
It it is exactly the spectral parameter of each stove.Training rules selected here is to train forward from terminal, i.e. chooses end 100 frame training
One model, then proceedes to push away forward one model of 100 frame retraining.
First the heat in a period of time is trained, such as training sample is chosen they 200 frames reciprocal between 300 frames
100 frame data training patterns, the extraction of 100 frame data is consistent to all training samples here.
As a example by the disaggregated model of 30 carbon, after determining kernel function, the parameter next controlled is mainly: kernel function width δ
And penalty factor.
Parameter adjustment method specifically includes that intelligence Particle Swarm (PSO) algorithm is (such as Eberhart R C, Kennedy J.A new
optimizer using particle swarm theory[C]//Proceedings of the sixth international symposium
On micro machine and human science.1995,1:39-43.), heredity (GA) algorithm (thunderous sword. based on
The modeling of SVM and genetic algorithm and global optimizing method [J]. Science Plaza, 2008 (5): 120-122.), grid data service (as
Wang Jianfeng, Zhang Lei, Chen Guoxing, etc. SVM parameter optimization [J] based on the grid data service improved. applicating technology, 2012,
39 (3): 28-31.) etc., can carry out adjusting ginseng with above distinct methods respectively, in conjunction with hit rate and the generalization of model of sample,
Select optimum kernel function width δ and penalty factor.
The most still simple ginseng process of adjusting of introducing as a example by the disaggregated model of 30 carbon:
First, choose at random a pair penalty factor=30 and the parameter of kernel function width δ=0.2, utilize SVM training the most permissible
Obtaining an i.e. hyperplane of grader, the training precision of its own and test sample precision are as shown in Fig. 6 a, 6b, and asterisk represents
It is actual carbon value, the miss heat that circle represents, the heat of namely classification error.
As shown in Fig. 6 a, 6b, for the classification hit situation of training sample and test sample, Fig. 6 a, 6b can be seen that model pair
The classifying quality of training sample is the best, it is possible to all classification is correct, and under this parameter, training sample is entirely and supports vector.
But for test sample, there is the carbon of 4 heats the most correctly to classify, if other parameter situation, it is believed that
This situation is acceptable, but actually needs to attempt substantial amounts of parameter to find suitable model.Said
Primary concern is that the hit rate of sample and the generalization of model when preference pattern, generalization is then by the classification feelings of test data
Condition is found out.
Next the classification situation under the conditions of parameter C=20 and δ=0.8 is considered, training and hit rate such as Fig. 7 a of test sample,
Shown in 7b.
From Fig. 6 a, 6b, Fig. 7 a, 7b are it can be seen that under these two groups of parameters, the hit rate of training sample and test sample is one
Sample, but in the case of parameter C=20 and δ=0.8, the nicety of grading of training sample reduces, and the number of vector supported by sample
Mesh reduces, and the generalization of model is better than the former, and new test data accuracy is better than the former.So in this case,
Comparatively speaking the model selecting one rear pair parameter to be trained is the most suitable.
Already explained above, the hyperplane of largest interval error in N number of random sample S is to be not more than with probability 1-δ:
In above formula, d=#sv means that the number supporting vector, and above formula shows to support that the number of vector is the fewest, and it is general
Change ability is the strongest.Being obtained by analysis, the value adjusting δ can change the number supporting vector, and then changes disaggregated model
Generalization, in any case, the value of δ is the biggest, supports that the number of vector is the fewest, and generalization is the best, and this point is returned at SVM
When returning matching, can show becomes apparent from.But it is not that δ is the bigger the better, when δ increases to a certain degree, model general
Change property do not improved, it is possible to also can degenerate, the even hit rate of training sample is also possible to be deteriorated, as Fig. 8 a,
8b show parameter C=20, the classification hit results of training sample and test sample during δ=3.
By Fig. 8 a, 8b it can be seen that be at this moment accomplished by regulating the value of penalty factor, the value of C represents exceeding target
Penalty factor.Although the number above supporting vector is the reduction of, but the restriction ability of parameter C is for training vector
Reduce, cause the training result not being to support vector beyond extensive error bounds certain limit, thus have impact on its hit rate.
So the two parameter is all to need regulation, it is not that a parameter is fixed, only changes another parameter and find suitable point
Class model, but two parameters will change and carry out selection sort model.By attempting substantial amounts of parameter, have finally chosen
Disaggregated model under the parameter combination of C=100 and δ=0.06, it is used for the hit rate to sample classification as shown in Fig. 9 a, 9b.
So far, for the disaggregated model of 30 carbon, the disaggregated model between 100 to 200 frames reciprocal has just been trained,
According to same optimal searching principle, can find until 15 disaggregated models of 1500 frames reciprocal, whole model training situations.
When actually detected flame spectrum data, from spectrometer collection to fire door flame spectrum data initially enter the 15th model, so
After sequentially enter the 14th, 13 ..., 2,1 disaggregated model, as shown in Figure 10.Can draw during to each disaggregated model
One label value, this end value means that the classification situation of model, i.e. belongs to which class.
Here need clear and definite any exactly: the testing result that the spectroscopic data of Primary period collection obtains after entering model is not
Accurately, train, after only bessemerizing entrance because the disaggregated model of the present invention is the spectroscopic data according to the blowing later stage
After phase, disaggregated model just starts really to play a role, and the experience of this and steel mill workman controls to be consistent with, workman be also
Just start simply are carrying out according to experience for a long time in real terminal point control, early stage and mid-term during to blowing latter stage.
Equally, the disaggregated model for 15 carbon is also such, is as the criterion with 100 frames and trains successively, waits 30 carbon disaggregated models
After all having trained with the disaggregated model of 15 carbon, the model inspection flow process of actual spectrum can be expressed as shown in Figure 11,
All can have two disaggregated models in work in each moment section, can definite know after model inspection carbon content be belonging to height,
In, which big class low, and then utilize known matched curve to detect the carbon value of reality.
Shown in Figure 12, carbon content proposed by the invention online Real-time and Dynamic Detection method upon actuation, carries out on-the-spot test
Time, the disaggregated model of training proceeds by classification the most therewith, and what it can be real-time provides a classification results, and this result shows this
Time carbon be belonging to high, medium and low in which class.Obviously, after the classification determining aim carbon, ensuing worked
Journey is exactly to draw now actual carbon value, so the help being accomplished by terminal matched curve just can realize.
As shown in figure 12, in detection method proposed by the invention, described terminal point control device 3 based on SVM, such as examine
Drafting boards etc., can be connected by data wire and a upper industrial computer, to receive and to send data message, it is achieved to whole detection method
The uploading of debugging, control and data, show, store and subsequent analysis etc., such as by the spectral information of online real time collecting
Display on the display screen of industrial computer, and/or, the carbon content result detected is shown in real time or with the form of curve representation
Characterized to operator by display screen.
During the detection of carbon content, after the flame spectrum information receiving online real time collecting construction feature parameter, first
First pass through described SVM carbon content dynamic prediction model and determine the classification of aim carbon, and classification based on aim carbon uses correspondence
Terminal fitting function determines the carbon content of the currently molten steel that gathered flame spectrum information is corresponding.
The classification (such as mild steel, medium carbon steel, high-carbon steel) that described terminal fitting function includes different aim carbon is each affiliated
Terminal fitting function, wherein:
Described terminal fitting function is expressed as:
Y=f (X),
This formula have expressed the mapping relations of X Yu Y, and the characteristic variable extracted during wherein X is terminal moment flame spectrum, Y is
Terminal carbon value, this terminal fitting function uses MATLAB to provide a polynomial fit function to be fitted data, thus
Obtain fitting function.
In this example, say, that combine shown in Fig. 1, arithmetic element is receiving the flame spectrum information of online real time collecting also
After construction feature parameter, first pass through the described dynamic detection model of SVM carbon content and determine the classification of carbon, i.e. determine and blown
Steel grade belongs to mild steel, medium carbon steel or high-carbon steel, and classification based on carbon uses corresponding terminal fitting function to determine current institute
Gather the carbon content of molten steel corresponding to flame spectrum information.
In this example, terminal fitting function includes the terminal fitting function belonging to respective difference of mild steel, medium carbon steel, high-carbon steel.
That is, heat is divided into high, medium and low three classes according to carbon content, the sample of each class is fitted respectively, draws one
Fitting function, namely will eventually get high, medium and low three matched curves.
Described terminal fitting function Y=f (X), actually have expressed the mapping relations of X Yu Y, and wherein X is terminal moment fire
The characteristic variable extracted in flame spectrum, Y is terminal carbon value.Such as, MATLAB is used to provide a polynomial fit function
Data are fitted, obtain fitting function.
The example that with polynomial fit function carry out data matching is given below.
Polynomial fit function equation below:
[p, s]=polyfit (X, Y, N)
The criterion of matching is method of least square, i.e. finds and makesMinimum f (x).N in formula represents the rank of matching
Number, the coefficient vector of p representative polynomial, s represents the error estimation generating detected value.
For choosing the precision of the evaluation criterion of fitting function, mainly matching, i.e. match value and actual value within range of error
Difference.In order to consider matched curve identity in actual applications and observation property, it is also contemplated that the downward trend of matched curve
Situation.
After successfully training required disaggregated model and aim carbon fitting function, propose as mentioned above, when converter is refined
Steel blow latter stage time, disaggregated model starts to play its real effect, the carbon content in molten steel is carried out real-time classification and Detection,
As shown in figure 13, utilize fitting function to obtain definite carbon value, the regulation and control foundation of real-time carbon content is provided for steel mill.
Through overtesting, by the fire door flame spectrum data of 40 converters of Site Detection, obtain the testing result of aim carbon, analyze
Its hit situation and error distribution, draw aim carbon classification and Detection method hit rate that the present invention studies up to more than 85%, completely
Meet the actual demand of steel mill.
Although the present invention is disclosed above with preferred embodiment, so it is not limited to the present invention.The technical field of the invention
Middle tool usually intellectual, without departing from the spirit and scope of the present invention, when being used for a variety of modifications and variations.Therefore, originally
The protection domain of invention is when being as the criterion depending on those as defined in claim.
Claims (8)
1. converter steel-smelting molten steel carbon content based on a SVM online Real-time and Dynamic Detection method, it is characterised in that this detection
The realization of method includes:
According to the field stop in pneumatic steelmaking environment regulation telescope optical system, thus the visual field regulating fire door flame detecting makes
The predeterminated position of fire door flame is by this telescope optical system imaging, then fire door flame image information is transmitted via a Transmission Fibers
To a spectrogrph;
Spectrogrph receives as afterwards, carries out spectrum analysis and obtains flame spectrum distributed intelligence;
Utilize the real-time detection carrying out pneumatic steelmaking carbon content based on SVM end-point control method according to flame spectrum distributed intelligence,
Wherein carbon content is detected by the dynamic detection model of SVM carbon content.
Converter steel-smelting molten steel carbon content based on SVM the most according to claim 1 online Real-time and Dynamic Detection method, its
Being characterised by, described field stop is variable field of view light hurdle.
Converter steel-smelting molten steel carbon content based on SVM the most according to claim 1 online Real-time and Dynamic Detection method, its
Being characterised by, the program that detection is stored in computer systems by operation in real time of described pneumatic steelmaking carbon content realizes, bag
Include procedure below:
The characteristic parameter of carbon content change in receiving flame spectrum information and building sign stove;
The characteristic parameter of carbon content change in sign stove is inputted a dynamic detection model of SVM carbon content detect;And
The testing result of output carbon content.
Converter steel-smelting molten steel carbon content based on SVM the most according to claim 3 online Real-time and Dynamic Detection method, its
Being characterised by, the described dynamic detection model of SVM carbon content is the model that training in advance is good, and its training process is as follows:
Using actual carbon content of molten steel as standard, by repetition training, optimized choice, determine involved by SVM learning algorithm
Each parameter, it specifically includes:
The characteristic parameter of carbon content change in stove can be characterized by flame spectrum information architecture;
The kernel function of selected SVM learning algorithm;
Optimal control parameter kernel function width δ and penalty factor;
Selection Model training sample, utilizes SVM learning algorithm that characteristic parameter is carried out classification model construction;
Input the model set up with test sample, and whether analytical error and generalization meet design and require: if it is satisfied, then
Output model, if be unsatisfactory for, then returns described step and re-starts the selection of kernel function width δ and penalty factor with again
Modeling, until meeting requirement.
Converter steel-smelting molten steel carbon content based on SVM the most according to claim 4 online Real-time and Dynamic Detection method, its
It is characterised by, during described model training, construction feature parameter in the following manner:
At wavelength 600nm, spectral shape is protruding spike, characteristic parameter a1For light intensity normalized value herein;
Spectral shape spike of projection at 770nm is bimodal, characteristic parameter a2For the light intensity at wavelength 770nm and 772nm
Normalization average value;
This section of spectral line acutely, is divided into three sections, to each section of light intensity normalizing by the continuous spectrum change in the middle of said two spike
Average after change and obtain three characteristic parameter a3, a4, a5;And
Using the ratio of the peak wavelength λ of spectrum in spectral distribution and investigative range maximum of T max of described spectrogrph as the 6th
Individual parameter: a6。
Converter steel-smelting molten steel carbon content based on SVM the most according to claim 4 online Real-time and Dynamic Detection method, its
Being characterised by, during described model training, the kernel function of described SVM learning algorithm is selected from linear kernel function, multinomial
One in kernel function, RBF kernel function and S type kernel function.
Converter steel-smelting molten steel carbon content based on SVM the most according to claim 4 online Real-time and Dynamic Detection method, its
It is characterised by, during the detection of carbon content, is receiving flame spectrum information the construction feature parameter of online real time collecting
After, first pass through described SVM carbon content dynamic prediction model and determine the classification of aim carbon, and classification based on aim carbon uses
Corresponding terminal fitting function determines the carbon content of the currently molten steel that gathered flame spectrum information is corresponding.
Converter steel-smelting molten steel carbon content based on SVM the most according to claim 7 online Real-time and Dynamic Detection method, its
Be characterised by, described terminal fitting function include the classification of different aim carbon each belonging to terminal fitting function, wherein:
Described terminal fitting function is expressed as:
Y=f (X),
This formula have expressed the mapping relations of X Yu Y, and the characteristic variable extracted during wherein X is terminal moment flame spectrum, Y is
Terminal carbon value, this terminal fitting function uses MATLAB to provide a polynomial fit function to be fitted data, thus
Obtain fitting function.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510170782.8A CN106153550A (en) | 2015-04-10 | 2015-04-10 | Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510170782.8A CN106153550A (en) | 2015-04-10 | 2015-04-10 | Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106153550A true CN106153550A (en) | 2016-11-23 |
Family
ID=57336795
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510170782.8A Pending CN106153550A (en) | 2015-04-10 | 2015-04-10 | Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106153550A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112418161A (en) * | 2020-12-08 | 2021-02-26 | 浙江大学 | Finished tea type and grade identification method based on image color and texture features |
CN112734722A (en) * | 2021-01-08 | 2021-04-30 | 昆明理工大学 | Flame endpoint carbon content prediction method based on improved complete local binary pattern |
CN113113088A (en) * | 2021-04-08 | 2021-07-13 | 西安道法数器信息科技有限公司 | Converter carbon content index evaluation and temperature analysis method based on artificial intelligence |
CN113221827A (en) * | 2021-05-31 | 2021-08-06 | 中冶赛迪重庆信息技术有限公司 | Intelligent converter flame state identification and rechecking method, system, medium and terminal |
CN113420412A (en) * | 2021-05-26 | 2021-09-21 | 南京信息工程大学 | Imaging spectrum-based continuous depth distribution extraction method for organic carbon content in soil |
CN113832283A (en) * | 2020-06-24 | 2021-12-24 | 上海梅山钢铁股份有限公司 | Molten steel temperature prediction method based on fire hole flame |
EP3795702A4 (en) * | 2018-05-14 | 2021-12-29 | JFE Steel Corporation | Melt component estimation device, melt component estimation method, and method for producing melt |
CN116067900A (en) * | 2023-03-29 | 2023-05-05 | 北京科技大学 | Method and device for analyzing temperature and composition data of autonomous luminous monitoring molten pool in fire spot area |
CN116463469A (en) * | 2023-03-29 | 2023-07-21 | 北京科技大学 | Automatic control system and operation method for converter steelmaking based on fire spot area luminescence detection |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH10287478A (en) * | 1997-04-11 | 1998-10-27 | Nippon Steel Corp | Carbon-containing flame spray material and its production |
USH1925H (en) * | 1998-08-18 | 2000-12-05 | The United States Of America As Represented By The Secretary Of The Navy | Apparatus and method for monitoring steel decarburization by remote flame emission spatial imaging spectroscopy |
CN101698896A (en) * | 2009-09-28 | 2010-04-28 | 南京理工大学 | System and method for steel-making online end-point control through furnace mouth radiation information fusion |
CN102206727A (en) * | 2011-05-31 | 2011-10-05 | 湖南镭目科技有限公司 | Converter steelmaking endpoint determination method and system, control method and control system |
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 |
CN102876838A (en) * | 2012-10-30 | 2013-01-16 | 湖南镭目科技有限公司 | System for detecting carbon content and temperature in converter |
-
2015
- 2015-04-10 CN CN201510170782.8A patent/CN106153550A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH10287478A (en) * | 1997-04-11 | 1998-10-27 | Nippon Steel Corp | Carbon-containing flame spray material and its production |
USH1925H (en) * | 1998-08-18 | 2000-12-05 | The United States Of America As Represented By The Secretary Of The Navy | Apparatus and method for monitoring steel decarburization by remote flame emission spatial imaging spectroscopy |
CN101698896A (en) * | 2009-09-28 | 2010-04-28 | 南京理工大学 | System and method for steel-making online end-point control through furnace mouth radiation information fusion |
CN102206727A (en) * | 2011-05-31 | 2011-10-05 | 湖南镭目科技有限公司 | Converter steelmaking endpoint determination method and system, control method and control system |
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 |
CN102876838A (en) * | 2012-10-30 | 2013-01-16 | 湖南镭目科技有限公司 | System for detecting carbon content and temperature in converter |
Non-Patent Citations (3)
Title |
---|
吴大江等编: "《测绘仪器使用与检测》", 31 August 2012 * |
许凌飞: "基于炉口火焰光谱信息的转炉炼钢终点在线碳含量测量方法研究", 《中国博士学位论文全文数据库工程科技Ⅰ辑》 * |
郭达志等编: "《大地测量仪器学》", 31 October 1986 * |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP3795702A4 (en) * | 2018-05-14 | 2021-12-29 | JFE Steel Corporation | Melt component estimation device, melt component estimation method, and method for producing melt |
US11966669B2 (en) | 2018-05-14 | 2024-04-23 | Jfe Steel Corporation | Molten metal component estimation device, method of estimating molten metal component, and method of manufacturing molten metal |
CN113832283A (en) * | 2020-06-24 | 2021-12-24 | 上海梅山钢铁股份有限公司 | Molten steel temperature prediction method based on fire hole flame |
CN113832283B (en) * | 2020-06-24 | 2023-02-07 | 上海梅山钢铁股份有限公司 | Molten steel temperature prediction method based on fire hole flame |
CN112418161A (en) * | 2020-12-08 | 2021-02-26 | 浙江大学 | Finished tea type and grade identification method based on image color and texture features |
CN112734722A (en) * | 2021-01-08 | 2021-04-30 | 昆明理工大学 | Flame endpoint carbon content prediction method based on improved complete local binary pattern |
CN112734722B (en) * | 2021-01-08 | 2022-09-13 | 昆明理工大学 | Flame endpoint carbon content prediction method based on improved complete local binary pattern |
CN113113088B (en) * | 2021-04-08 | 2024-03-08 | 山西晋钢智造科技实业有限公司 | Converter carbon content index evaluation and temperature analysis method based on artificial intelligence |
CN113113088A (en) * | 2021-04-08 | 2021-07-13 | 西安道法数器信息科技有限公司 | Converter carbon content index evaluation and temperature analysis method based on artificial intelligence |
CN113420412A (en) * | 2021-05-26 | 2021-09-21 | 南京信息工程大学 | Imaging spectrum-based continuous depth distribution extraction method for organic carbon content in soil |
CN113221827A (en) * | 2021-05-31 | 2021-08-06 | 中冶赛迪重庆信息技术有限公司 | Intelligent converter flame state identification and rechecking method, system, medium and terminal |
CN116463469A (en) * | 2023-03-29 | 2023-07-21 | 北京科技大学 | Automatic control system and operation method for converter steelmaking based on fire spot area luminescence detection |
CN116463469B (en) * | 2023-03-29 | 2024-03-26 | 北京科技大学 | Automatic control system and operation method for converter steelmaking based on fire spot area luminescence detection |
CN116067900A (en) * | 2023-03-29 | 2023-05-05 | 北京科技大学 | Method and device for analyzing temperature and composition data of autonomous luminous monitoring molten pool in fire spot area |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106153550A (en) | Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection method | |
CN106153551A (en) | Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection system | |
CN111254243B (en) | Method and system for intelligently determining iron notch blocking time in blast furnace tapping process | |
CN107704806A (en) | A kind of method that quality of human face image prediction is carried out based on depth convolutional neural networks | |
CN112070393B (en) | Wastewater quality detection system that industry and agriculture production discharged | |
CN105678332A (en) | Converter steel-making endpoint determination method and system based on flame image CNN recognizing and modeling process | |
CN109934447A (en) | A kind of Fuzzy Synthetic Evaluation of secondary equipment of intelligent converting station efficiency | |
CN107133460A (en) | A kind of online dynamic prediction method of boiler flyash carbon content | |
JP7172400B2 (en) | Geological evaluation system, geological evaluation method and geological evaluation program | |
CN107817223A (en) | The construction method of quick nondestructive real-time estimate oil property model and its application | |
Carr et al. | Road crack detection using a single stage detector based deep neural network | |
CN102456142A (en) | Analysis method for smoke blackness based on computer vision | |
CN106649919A (en) | Method and system for predicting carbon content of fly ash in coal-fired power plant boiler | |
CN105868556A (en) | Bridge safety assessment method based on Beidou bridge safety monitoring system | |
CN104483292A (en) | Multi-spectral-line calibration method for improving analysis precision of laser probe | |
CN109799245A (en) | A kind of insulator contamination degree non-contact detection method and device | |
CN104503406A (en) | Optimized operation evaluation system and method for thermal power plant | |
CN105956591B (en) | A kind of online hot parts infrared image spectrum sampling Detection method | |
RU2713850C1 (en) | Fuel combustion modes monitoring system by means of torch images analysis using classifier based on convolutional neural network | |
CN105868115A (en) | Building method and system for software test model of software intensive system | |
CN106148637B (en) | The pneumatic steelmaking carbon content dynamic detection system of high stable | |
CN109858707A (en) | A kind of Ethylene Cracking Furnace Tubes intelligence coking diagnostic method based on ABC-ANFIS-CTF | |
Zhang et al. | Grouped time series networks for grade monitoring of zinc tailings with multisource features | |
CN108627500A (en) | A kind of solid fuel measurement of caloric value method | |
CN115829324A (en) | Personnel safety risk silent monitoring method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20161123 |
|
RJ01 | Rejection of invention patent application after publication |