CN106503413B - A method of accurately calculating desulfurizing iron magnesium powder amount - Google Patents
A method of accurately calculating desulfurizing iron magnesium powder amount Download PDFInfo
- Publication number
- CN106503413B CN106503413B CN201510542806.8A CN201510542806A CN106503413B CN 106503413 B CN106503413 B CN 106503413B CN 201510542806 A CN201510542806 A CN 201510542806A CN 106503413 B CN106503413 B CN 106503413B
- Authority
- CN
- China
- Prior art keywords
- tank switching
- magnesium powder
- sulfur content
- molten iron
- powder amount
- 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.)
- Active
Links
Abstract
The invention discloses a kind of methods for accurately calculating desulfurizing iron magnesium powder amount, mainly solve the problems such as existing in the prior art cumbersome, magnesium powder dosage is higher, calculating is abnormal.The present invention combines BP neural network model and regression model, wherein regression model is divided into four regression equations according to tank switching molten steel sulfur content and desulfurization target sulphur content size, study group is formed by choosing qualified historical data, training BP neural network model and the parameter for determining each regression equation, then current heat information is acquired, it is input to neural network model and corresponding regression equation, neural network model magnesium powder amount Y is calculated1With regression model magnesium powder amount Y2, compare Y1And Y2, work as 0.7Y2< Y1< 1.2Y2When, final output result is Y1, it is otherwise Y2.The present invention combines the advantage of BP neural network model and regression model, realizes the accurate control of magnesium powder dosage accurately calculated with sulfur content after desulfurization, reduces ton iron magnesium powder and is consumed to 0.33kg/t.
Description
Technical field
The present invention relates to a kind of methods for accurately calculating desulfurizing iron magnesium powder amount, belong to technical field of hot metal pretreatment.
Background technique
In general, sulphur is the harmful element in steel, molten iron must carry out desulfurization pretreatment before entering converter, and composite blowing method is
One of main Desulphuring Process of Molten Iron, common desulfurizing agent are magnesium powder, dosage by molten iron pretreatment desulfurizing target sulphur content and
Molten iron condition is determined.Traditional magnesium powder dosage determines that method is obtained by artificial enquiry doctor solution consumption static table, the table root
According to the analysis of historical data, magnesium corresponding to different molten steel sulfur contents, different molten iron pretreatment desulfurizing target sulphur is listed
Powder dosage.This mode is not only cumbersome, large labor intensity, be easy inquiry error, and without system consider weight of molten iron,
The influence of the factors such as molten iron temperature, magnesium powder dosage are bigger than normal.
To improve automatization level, mistake is avoided, desulfurizing agent computation model, such as Chinese patent application are developed
Number " a kind of system of Desulfurization Calculation method and its application " is disclosed for 201310268262.1, this method is according to tank switching molten iron temperature
The data such as degree, tank switching weight of molten iron, tank switching molten steel sulfur content, desulfurization target sulphur content, using the method meter of multiple linear regression
Doctor solution consumption is calculated, accuracy and desulfurization success rate are improved.But this method uses equation of linear regression, it cannot be faithfully
Reflect the influence of tank switching molten steel sulfur content and desulfurization target sulphur content to doctor solution consumption, and more importantly its parameter is
It is changeless, the variation of molten iron pretreatment desulfurizing ability can not be adapted to, to avoid sulphur exceeded, it is inclined to equally exist doctor solution consumption
The problem of height, sulphur mass excess.In the prior art, also there is relevant introduction, choose tank switching weight of molten iron, tank switching molten iron sulphur contains
As input parameter, doctor solution consumption establishes the desulfurizing agent based on improved Back Propagation as output parameter for amount, desulfurization target sulphur
Additional amount forecasting model makes desulfurization hit rate reach 96.8%.But work as the molten iron encountered in actual iron water bar part and study group
When condition difference is larger, the calculated result of neural network will appear jump, cause doctor solution consumption too high or too low.Above method
Desulfurization magnesium powder amount cannot all be efficiently solved accurately calculates problem, show as cumbersome, magnesium powder dosage is higher, calculate it is different
Often etc..Therefore, a kind of new technical solution of urgent need solves the technical problem.
Summary of the invention
To solve the above problems, the invention discloses a kind of sides for accurately calculating desulfurizing iron magnesium powder amount
Method, this combines BP neural network model and regression model, by the training and self study of historical data, determines that model is joined
Number, realize to magnesium powder amount it is automatic, accurately calculate, solve it is existing in the prior art it is cumbersome, magnesium powder dosage is higher, calculate
The problems such as abnormal, reduces company's production cost to reduce magnesium powder consumption.
To achieve the goals above, technical scheme is as follows, a kind of desulfurizing iron magnesium powder amount of accurately calculating
Method plants the method for accurately calculating desulfurizing iron magnesium powder amount, which is characterized in that the described method comprises the following steps:
A, the historical data nearest apart from current heat is acquired, five sample set U are established1、U2、U3、U4、U5As study
Group, U1~U4It is the condition sample set of m for sample size, 70≤m≤140, decision condition is respectively tank switching molten steel sulfur content <
0.03% and desulfurization target sulphur content≤0.002%, tank switching molten steel sulfur content < 0.03% and desulfurization target sulphur content >
0.002%, tank switching molten steel sulfur content >=0.03% and desulfurization target sulphur content≤0.002%, tank switching molten steel sulfur content >=
0.03% and desulfurization target sulphur content > 0.002%；U5The bulk sample sheet for being n for sample size, 40≤n≤100 are used for neural network
Model learning；
B, to sample set U5In data pre-processed, with eliminate sulfur content, temperature, weight data because the order of magnitude difference
And neural network is impacted, there is plateau phenomenon when reducing neural network learning, specific processing method is as follows:
In formula, x1k~ykFor the tank switching molten iron temperature of kth furnace in sample set, tank switching weight of molten iron, tank switching molten steel sulfur content,
Desulfurization target sulphur content, steel grade upper limit sulfur content, magnesium powder amount, x1,min~yminFor tank switching molten iron temperature minimum value in sample set, fall
Tank weight of molten iron minimum value, tank switching molten steel sulfur content minimum value, desulfurization target sulphur content minimum value, steel grade upper limit sulfur content are minimum
Value, magnesium powder amount minimum value, x1,max~ymaxFor tank switching molten iron temperature maximum value, tank switching weight of molten iron maximum value, tank switching in sample set
Molten steel sulfur content maximum value, desulfurization target sulphur content maximum value, steel grade upper limit sulfur content maximum value, magnesium powder amount maximum value,Contain for tank switching molten iron temperature, tank switching weight of molten iron, tank switching molten steel sulfur content, the desulfurization target sulphur of treated kth furnace
Amount, steel grade upper limit sulfur content, magnesium powder amount；
C, BP neural network model is established, BP neural network model is three layers comprising input layer, hidden layer and output layer
Network, wherein input layer is tank switching molten iron temperature x for receiving input signal, variable1, tank switching weight of molten iron x2, tank switching molten iron
Sulfur content x3, desulfurization target sulphur content x4, steel grade upper limit sulfur content x5；Output layer is used for output signal, that is, calculated result, variable
For magnesium powder amount Y1；The neuromere points f of hidden layer is according to formulaIt determines, p is that the neuromere of input layer is counted, q
For output layer neuromere points, r is the constant between 1~10；P=5, q=1 of the present invention, therefore the neuron node of hidden layer
The range of number f is 4~12, if the too small network of f cannot identify the sample not occurred, poor fault tolerance, f is too big to make network complicated
Change, increase the learning time of network, " over-fitting " phenomenon occurs, will increase error instead；
D, training BP neural network model adjusts nerve net according to the error of network reality output result and expected result
The weight w of network input layer and hidden layerih, hidden layer and output layer weight who, each neuron of hidden layer threshold θh, output layer
The threshold θ of each neurono, reduce error；Weight is identical with the method for adjustment of threshold value, specific as follows:
(1) network is initialized.W is given respectivelyih、who、θh、θoAssign the random number in a section (- 1,1), setup algorithm essence
Angle value α, maximum study number T, learning rate η, wherein can learning rate affect the speed of network convergence and network and restrain.
Learning rate setting is less than normal can to guarantee network convergence, but convergence rate is slack-off, and the training time increases.On the contrary, learning rate is arranged
It is bigger than normal, there is a possibility that network training is not restrained, influence recognition effect；
(2) sample set U is chosen5In kth furnace input sample after pretreatmentAnd it is right
The desired output answered
(3) outputting and inputting for hidden layer and each neuron of output layer is calculated, wherein the output of hidden layer uses function
Each neuron of hidden layer inputs hih(k) and output hoh(k),H=1,2,9；
The defeated people yi (k) of each neuron of output layer and output yo (k),
(4) global error is calculatedJudge whether E meets the requirements.If E > study essence
When spending α, (4) and step (5) are entered step, adjust weight and threshold value, adjustment terminates as E≤α or study number t > T；
(5) error function is calculated to the partial derivative of weight and threshold value
Wherein, ekFor network output error function
δh(k)=δo(k)whof'(hih(k))=δo(k)whohih(k)(1-hih(k))
(6) weight w is adjusted using iterative methodho(k)、wih(k) and threshold θo、θh, t is study number.
E, the recurrence computation model for establishing magnesium powder amount, using segmentation nonlinear multivariate regression equations, independent variable is tank switching iron
Coolant-temperature gage x1, tank switching weight of molten iron x2, tank switching molten steel sulfur content x3, desulfurization target sulphur content x4, specific calculation formula are as follows:
As tank switching molten steel sulfur content x3< 0.03%, desulfurization target sulphur content x4When≤0.002%:
As tank switching molten steel sulfur content x3< 0.03%, desulfurization target sulphur content x4When > 0.002%:
As tank switching molten steel sulfur content x3>=0.03%, desulfurization target sulphur content x4When≤0.002%:
Y2=c0+c1×x1+c2×x2+c3×x3+c4×x4
As tank switching molten steel sulfur content x3>=0.03%, desulfurization target sulphur content x4When > 0.002%:
F, sample set U is utilized1~U4In data calculate separately parameter a0~a5、b0~b6、c0~c4、d0~d5, calculating
When desulfurization target sulphur content x4With sulfur content x after desulfurization6Instead of；
Calculate a0~a5Formula it is as follows, wherein x1i、x2i、x3i、x6iFor sample set U1In the i-th furnace x1、x2、x3、x6:
Calculate b0~b6Formula it is as follows, wherein x1i、x2i、x3i、x6iFor sample set U2In the i-th furnace x1、x2、x3、x6:
Calculate c0~c4Formula it is as follows, wherein x1i、x2i、x3i、x6iFor sample set U3In the i-th furnace x1、x2、x3、x6:
Calculate d0~d5Formula it is as follows, wherein x1i、x2i、x3i、x6iFor sample set U4In the i-th furnace x1、x2、x3、x6:
G, the magnesium powder amount Y of neural network model is calculated1, after receiving molten iron pre-treatment job commencing signal, acquisition is current
The tank switching molten iron temperature of heat, tank switching weight of molten iron, tank switching molten steel sulfur content, desulfurization target sulphur content, steel grade upper limit sulfur content
Etc. data, be input in trained neural network model, magnesium powder amount Y obtained after the output valve of output layer is transformed1,
H, the magnesium powder amount Y of regression model is calculated2, judge whether < 0.03%, desulfurization target sulphur contain tank switching molten steel sulfur content
Amount whether≤0.002%, according to determining that result selection selects corresponding regression equation, and by tank switching molten iron temperature, tank switching molten iron
Weight, tank switching molten steel sulfur content, desulfurization target sulphur content substitute into equation, and magnesium powder amount Y is calculated2；
I, the final result for exporting magnesium powder amount determines the magnesium that neural network model calculates with the magnesium powder amount that regression model calculates
Whether powder amount is suitable, and to prevent neural network from local minimum or maximum occur, specific decision logic is to work as 0.7Y2< Y1<
1.2Y2When, take the magnesium powder amount calculated value Y of neural network model1, otherwise take the magnesium powder amount calculated value Y of regression model2；J, it updates and learns
Habit group after current heat, judges whether current heat meets the condition into study group, decision condition are as follows: tank switching molten iron
Sulfur content is in 0.005~0.1%, tank switching molten iron temperature in 1201~1480 DEG C, tank switching weight of molten iron in (converter nominal tonnage-
15)~(converter nominal tonnage+25) after t, desulfurization sulfur content in 0.0005~0.015%, magnesium powder utilization rate 15~65%, if
It is eligible, replace regression model learning sample collection U1、U2、U3、U4With the learning sample collection U of neural network model5In it is oldest
Heat, keep sample number constant, wherein current heat enters U1、U2、U3、U4Which of sample set, according to tank switching iron
Depending on the size of water sulfur content and desulfurization target sulphur content；
K, model restarting study, as regression model learning sample collection U1、U2、U3、U41 furnace of every update, return step f,
Recalculate parameter a0~a5、b0~b6、c0~c4, d0~d5.As neural network model sample set U55 furnaces of every update, return step
D, re -training network adapt to the variation of molten iron pretreatment desulfurizing ability with the magnesium powder dosage for ensuring to be calculated automatically.
As an improvement of the present invention, in the step a, to prevent the heat containing abnormal data from entering study group U1
~U5, when acquiring the historical data nearest apart from current heat, heat is screened, screening conditions include: tank switching molten iron sulphur
Content 0.005~0.1%, tank switching molten iron temperature exist in 1201~1480 DEG C, weight of molten iron are as follows: and -15 tons of converter nominal tonnage~
Sulfur content is in 0.0005~0.015%, magnesium powder utilization rate 15%~65% after+25 tons of converter nominal tonnage, desulfurization.
As an improvement of the present invention, in the step a, magnesium powder utilization rate is total for the magnesium powder amount Zhan for desulphurization reaction
The ratio of magnesium powder amount, calculation formula are as follows: (sulfur content after tank switching molten steel sulfur content-desulfurization) × tank switching iron water amount × 0.75 ÷ is total
Magnesium powder amount.
As an improvement of the present invention, in the step g, when tank switching molten iron temperature, tank switching weight of molten iron and tank switching
When molten steel sulfur content lacks, sample set U is calculated5Tank switching molten iron temperature, the tank switching of 20 nearest furnace data of the middle current heat of distance
The average value of weight of molten iron, tank switching molten steel sulfur content, using tank switching molten iron temperature average value and tank switching weight of molten iron average value as
The tank switching molten iron temperature and tank switching weight of molten iron of current heat, tank switching molten steel sulfur content average value using 1.1 times is as working as forehearth
Secondary tank switching molten steel sulfur content
Compared with the existing technology, advantages of the present invention is as follows, and the technical solution is by the recurrence of neural network model and segmentation
Model combines, and combines the advantage of the two, establishes magnesium powder dosage and molten iron condition, the relationship of target sulphur content, realize
The automatic calculating of magnesium powder dosage, solves the problems, such as artificial enquiry, efficiently avoids because neural network learning group covering surface is not wide
Caused calculated result abnormal problem.Simultaneously by the self study of neural network model and regression model, it can be achieved that magnesium powder amount
Adjust automatically, adapt to molten iron pretreatment desulfurizing ability variation, it is ensured that magnesium powder amount is more accurate, reasonable, control desulfurization after sulphur
Content is near target sulphur content；By the utilization of this technology, the ton iron magnesium powder of composite blowing sulfur removal technology is consumed by 0.41kg/
T is reduced to 0.33kg/t or less.
Detailed description of the invention
Fig. 1 is that magnesium powder amount of the invention calculates process schematic；
Fig. 2 is that BP nerve used in the present invention is over structure chart.
Specific embodiment
In order to further enhance the appreciation and understanding of the invention, with reference to the accompanying drawings and detailed description, this is furtherd elucidate
Invention.
Embodiment:
A method of accurately calculating desulfurizing iron magnesium powder amount, comprising the following steps:
A, the historical data nearest apart from current heat is acquired, five sample set U are established1、U2、U3、U4、U5As study
Group, the data content of sample include the molten iron pretreatment end time, molten iron processing number, tapping mark, steel grade upper limit sulfur content, take off
Sulphur after sulphur target sulphur content, tank switching molten iron temperature, tank switching weight of molten iron, tank switching molten steel sulfur content, desulfurization target sulphur content, desulfurization
Content, magnesium powder amount, sample are ranked up by the descending of molten iron pretreatment end time.U1~U4It is the condition sample of m for sample size
This collection, decision condition are respectively tank switching molten steel sulfur content < 0.03% and desulfurization target sulphur content≤0.002%, tank switching molten iron sulphur
Content < 0.03% and desulfurization target sulphur content > 0.002%, tank switching molten steel sulfur content >=0.03% and desulfurization target sulphur content
≤ 0.002%, tank switching molten steel sulfur content >=0.03% and desulfurization target sulphur content > 0.002%.U5The bulk sample for being n for sample size
This.To prevent the heat containing abnormal data from entering study group, heat should be screened, screening conditions include: tank switching molten iron
Sulfur content 0.005~0.1%, tank switching molten iron temperature 1201~1480 DEG C, weight of molten iron (- 15 tons of converter nominal tonnage)~
Sulfur content is in 0.0005~0.015%, magnesium powder utilization rate 15%~65% after (+25 tons of converter nominal tonnage), desulfurization.This reality
Apply a m=100, n=60.
B, to sample set U5In data pre-processed, to eliminate the data such as sulfur content, temperature, weight because of the order of magnitude not
Neural network is impacted together, plateau phenomenon occurs when reducing neural network learning, specific processing method is as follows:
In formula, x1k~ykFor the tank switching molten iron temperature of kth furnace in sample set, tank switching weight of molten iron, tank switching molten steel sulfur content,
Desulfurization target sulphur content, steel grade upper limit sulfur content, magnesium powder amount, x1,min~yminFor tank switching molten iron temperature minimum value in sample set, fall
Tank weight of molten iron minimum value, tank switching molten steel sulfur content minimum value, desulfurization target sulphur content minimum value, steel grade upper limit sulfur content are minimum
Value, magnesium powder amount minimum value, x1,max~ymaxFor tank switching molten iron temperature maximum value, tank switching weight of molten iron maximum value, tank switching in sample set
Molten steel sulfur content maximum value, desulfurization target sulphur content maximum value, steel grade upper limit sulfur content maximum value, magnesium powder amount maximum value,Contain for tank switching molten iron temperature, tank switching weight of molten iron, tank switching molten steel sulfur content, the desulfurization target sulphur of treated kth furnace
Amount, steel grade upper limit sulfur content, magnesium powder amount；
C, BP neural network model is established.BP neural network model is three layers comprising input layer, hidden layer and output layer
Network, wherein input layer is tank switching molten iron temperature x for receiving input signal, variable1, tank switching weight of molten iron x2, tank switching molten iron
Sulfur content x3, desulfurization target sulphur content x4, steel grade upper limit sulfur content x5；Output layer is used for output signal, that is, calculated result, variable
For magnesium powder amount Y1；The neuromere points f of hidden layer is according to formulaIt determines, p is that the neuromere of input layer is counted, q
For output layer neuromere points, r is the constant between 1~10.P=5, q=1 of the present invention, therefore the neuron node of hidden layer
The range of number f is 4~12, if the too small network of f cannot identify the sample not occurred, poor fault tolerance, f is too big to make network complicated
Change, increase the learning time of network, " over-fitting " phenomenon occurs, will increase error instead.The present embodiment determines optimal nerve
First number of nodes is 9；
D, training BP neural network model adjusts nerve net according to the error of network reality output result and expected result
The weight w of network input layer and hidden layerih, hidden layer and output layer weight who, each neuron of hidden layer threshold θh, output layer
The threshold θ of each neurono, reduce error.Weight is identical with the method for adjustment of threshold value, specific as follows:
(1) network is initialized.W is given respectivelyih、who、θh、θoAssign the random number in a section (- 1,1), setup algorithm essence
Angle value α, maximum study number T, learning rate η.The present embodiment takes α=0.0001, T=10000, η=0.1；
(2) sample set U is chosen5In kth furnace input sample after pretreatmentAnd it is right
The desired output answered
(3) outputting and inputting for hidden layer and each neuron of output layer is calculated, wherein the output of hidden layer uses function
Each neuron of hidden layer inputs hih(k) and output hoh(k),H=1,2,9；
The defeated people yi (k) of each neuron of output layer and output yo (k),
(4) global error is calculatedJudge whether E meets the requirements.If E > 0.0001
When, (4) and step (5) are entered step, weight and threshold value are readjusted, are weighed when E≤0.0001 or study number t > 10000
Value and adjusting thresholds terminate；
(5) error function is calculated to the partial derivative of weight and threshold value
Wherein, ekFor network output error function
δh(k)=δo(k)whof'(hih(k))=δo(k)whohih(k)(1-hih(k))；
(6) weight w is adjusted using iterative methodho(k)、wih(k) and threshold θo、θh, t is study number.
E, the recurrence computation model of magnesium powder amount is established.For formula using segmentation nonlinear multivariate regression equations, independent variable is to fall
Tank molten iron temperature x1, tank switching weight of molten iron x2, tank switching molten steel sulfur content x3, desulfurization target sulphur content x4, wherein tank switching molten iron sulphur contains
Amount and desulfurization target sulphur content, which consume magnesium powder, influences maximum, and different in different section influence degrees.It is specific to calculate public affairs
Formula are as follows:
As tank switching molten steel sulfur content x3< 0.03%, desulfurization target sulphur content x4When≤0.002%:
As tank switching molten steel sulfur content x3< 0.03%, desulfurization target sulphur content x4When > 0.002%:
As tank switching molten steel sulfur content x3>=0.03%, desulfurization target sulphur content x4When≤0.002%:
Y=c0+c1×x1+c2×x2+c3×x3+c4×x4；
As tank switching molten steel sulfur content x3>=0.03%, desulfurization target sulphur content x4When > 0.002%:
F, sample set U is utilized1~U4In data calculate separately parameter a0~a5、b0~b6、c0~c4、d0~d5, calculating
When desulfurization target sulphur content x4With sulfur content x after desulfurization6Instead of；
Calculate a0~a5Formula it is as follows, wherein x1i、x2i、x3i、x6iFor sample set U1In the i-th furnace x1、x2、x3、x6:
Calculate b0~b6Formula it is as follows, wherein x1i、x2i、x3i、x6iFor sample set U2In the i-th furnace x1、x2、x3、x6:
Calculate c0~c4Formula it is as follows, wherein x1i、x2i、x3i、x6iFor sample set U3In the i-th furnace x1、x2、x3、x6:
Calculate d0~d5Formula it is as follows, wherein x1i、x2i、x3i、x6iFor sample set U4In the i-th furnace x1、x2、x3、x6:
G, the magnesium powder amount Y of neural network model is calculated1.After receiving molten iron pre-treatment job commencing signal, acquisition is current
The tank switching molten iron temperature of heat, tank switching weight of molten iron, tank switching molten steel sulfur content, desulfurization target sulphur content, steel grade upper limit sulfur content
Etc. data, be input in trained neural network model, magnesium powder amount Y obtained after the output valve of output layer is transformed1,
H, the magnesium powder amount Y of regression model is calculated2.Judge whether < 0.03%, desulfurization target sulphur contain tank switching molten steel sulfur content
Amount whether≤0.002%, according to determining that result selection selects corresponding regression equation, and by tank switching molten iron temperature, tank switching molten iron
Weight, tank switching molten steel sulfur content, desulfurization target sulphur content substitute into equation, and magnesium powder amount Y is calculated2；
I, the final result of magnesium powder amount is exported.The magnesium that neural network model calculates is determined with the magnesium powder amount that regression model calculates
Whether powder amount is suitable, and to prevent neural network from local minimum or maximum occur, specific decision logic is to work as 0.7Y2< Y1<
1.2Y2When, take the magnesium powder amount calculated value Y of neural network model1, otherwise take the magnesium powder amount calculated value Y of regression model2；
J, renewal learning group.After current heat, judge whether current heat meets the condition into study group, determines
Condition are as follows: tank switching molten steel sulfur content is in 0.005~0.1%, tank switching molten iron temperature in 1201~1480 DEG C, tank switching weight of molten iron
Sulfur content is in 0.0005~0.015%, magnesium powder benefit after (converter nominal tonnage -15)~(converter nominal tonnage+25) t, desulfurization
With rate 15~65%.If eligible, regression model learning sample collection U is replaced1、U2、U3、U4With neural network model
Practise sample set U5In oldest heat, keep sample number it is constant.Wherein, current heat enters U1、U2、U3、U4Which of
Sample set, depending on the size of tank switching molten steel sulfur content and desulfurization target sulphur content；
K, model restarting study.As regression model learning sample collection U1、U2、U3、U41 furnace of every update, return step f,
Recalculate parameter a0~a5、b0~b6、c0~c4、d0~d5.As neural network model sample set U55 furnaces of every update, return step
D, re -training network adapt to the variation of molten iron pretreatment desulfurizing ability with the magnesium powder dosage for ensuring to be calculated automatically；
Table 1 shows the target sulphur content of embodiment, steel grade upper limit sulfur content and molten iron condition, and table 2 shows embodiment
Magnesium powder amount and desulfurization after sulfur content.
The steel grade and molten iron condition of 1 embodiment of the present invention of table
Sulphur after the magnesium powder amount and desulfurization of 2 embodiment of the present invention of table
A kind of method accurately calculating desulfurizing iron magnesium powder amount disclosed by the invention, using this method embodiment 1~14,
The ton more former technology of iron magnesium powder dosage averagely reduces 0.088kg/t, and desulfurized molten iron sulfur content is controlled in desulfurization target sulphur content
Zone of reasonableness in.
The inventive method realizes the automatic calculating of magnesium powder dosage and according to the variation adjust automatically of desulphurizing ability, makes magnesium powder
Dosage is more rationally and accurate.
Claims (5)
1. a kind of method for accurately calculating desulfurizing iron magnesium powder amount, which is characterized in that the described method comprises the following steps:
A, the historical data nearest apart from current heat is acquired, five sample set U are established1、U2、U3、U4、U5As study group, U1
~U4It is the condition sample set of m for sample size, 70≤m≤140, decision condition is respectively tank switching molten steel sulfur content < 0.03%
And desulfurization target sulphur content≤0.002%, tank switching molten steel sulfur content < 0.03% and desulfurization target sulphur content > 0.002%, fall
Tank molten steel sulfur content >=0.03% and desulfurization target sulphur content≤0.002%, tank switching molten steel sulfur content >=0.03% and desulfurization mesh
Mark sulfur content > 0.002%；U5The bulk sample sheet for being n for sample size, 40≤n≤100 learn for neural network model；
B, to sample set U5In data pre-processed, with eliminate sulfur content, temperature, weight data because the order of magnitude difference due to it is right
Neural network impacts, and plateau phenomenon occurs when reducing neural network learning, specific processing method is as follows:
In formula, x1k~x5k, ykFor the tank switching molten iron temperature of kth furnace in sample set, tank switching weight of molten iron, tank switching molten steel sulfur content,
Desulfurization target sulphur content, steel grade upper limit sulfur content, magnesium powder amount, x1,min~x5, min,yminIt is minimum for tank switching molten iron temperature in sample set
Value, tank switching weight of molten iron minimum value, tank switching molten steel sulfur content minimum value, desulfurization target sulphur content minimum value, steel grade upper limit sulphur contain
Measure minimum value, magnesium powder amount minimum value, x1,max~x5,max, ymaxFor tank switching molten iron temperature maximum value, tank switching molten iron weight in sample set
Measure maximum value, tank switching molten steel sulfur content maximum value, desulfurization target sulphur content maximum value, steel grade upper limit sulfur content maximum value, magnesium powder
Maximum value is measured, Contain for tank switching molten iron temperature, tank switching weight of molten iron, the tank switching molten iron sulphur of treated kth furnace
Amount, desulfurization target sulphur content, steel grade upper limit sulfur content, magnesium powder amount；
C, BP neural network model is established, BP neural network model is the three-layer network comprising input layer, hidden layer and output layer,
Wherein input layer is tank switching molten iron temperature x for receiving input signal, variable1, tank switching weight of molten iron x2, tank switching molten iron sulphur contains
Measure x3, desulfurization target sulphur content x4, steel grade upper limit sulfur content x5；Output layer is used for output signal, that is, calculated result, and variable is magnesium
Powder amount Y1；The neuromere points f of hidden layer is according to formulaIt determines, p is that the neuromere of input layer is counted, and q is defeated
Layer neuromere is counted out, and r is the constant between 1~10；
D, it is defeated to adjust neural network according to the error of network reality output result and expected result for training BP neural network model
Enter the weight w of layer and hidden layerih, hidden layer and output layer weight who, each neuron of hidden layer threshold θh, each mind of output layer
Threshold θ through membero, reduce error；
E, the recurrence computation model for establishing magnesium powder amount, using segmentation nonlinear multivariate regression equations, independent variable is tank switching molten iron temperature
Spend x1, tank switching weight of molten iron x2, tank switching molten steel sulfur content x3, desulfurization target sulphur content x4, specific calculation formula are as follows:
As tank switching molten steel sulfur content x3< 0.03%, desulfurization target sulphur content x4When≤0.002%:
As tank switching molten steel sulfur content x3< 0.03%, desulfurization target sulphur content x4When > 0.002%:
As tank switching molten steel sulfur content x3>=0.03%, desulfurization target sulphur content x4When≤0.002%:
Y2=c0+c1×x1+c2×x2+c3×x3+c4×x4
As tank switching molten steel sulfur content x3>=0.03%, desulfurization target sulphur content x4When > 0.002%:
F, sample set U is utilized1~U4In data calculate separately parameter a0~a5、b0~b6、c0~c4、d0~d5, de- when calculating
Sulphur target sulphur content x4With sulfur content x after desulfurization6Instead of；
Calculate a0~a5Formula it is as follows, wherein x1i、x2i、x3i、x6iFor sample set U1In the i-th furnace x1、x2、x3、x6:
Calculate b0~b6Formula it is as follows, wherein x1i、x2i、x3i、x6iFor sample set U2In the i-th furnace x1、x2、x3、x6:
Calculate c0~c4Formula it is as follows, wherein x1i、x2i、x3i、x6iFor sample set U3In the i-th furnace x1、x2、x3、x6:
Calculate d0~d5Formula it is as follows, wherein x1i、x2i、x3i、x6iFor sample set U4In the i-th furnace x1、x2、x3、x6:
G, the magnesium powder amount Y of neural network model is calculated1, after receiving molten iron pre-treatment job commencing signal, acquire current heat
Tank switching molten iron temperature, tank switching weight of molten iron, tank switching molten steel sulfur content, desulfurization target sulphur content, steel grade upper limit sulfur content data, it is defeated
Enter into trained neural network model, magnesium powder amount Y is obtained after the output valve of output layer is transformed1,
H, the magnesium powder amount Y of regression model is calculated2, judge tank switching molten steel sulfur content whether < 0.03%, desulfurization target sulphur content whether
≤ 0.002%, according to determining that result selection selects corresponding regression equation, and by tank switching molten iron temperature, tank switching weight of molten iron, fall
Tank molten steel sulfur content, desulfurization target sulphur content substitute into equation, and magnesium powder amount Y is calculated2；
I, the final result for exporting magnesium powder amount determines the magnesium powder amount that neural network model calculates with the magnesium powder amount that regression model calculates
Whether suitable, to prevent neural network from local minimum or maximum occur, specific decision logic is to work as 0.7Y2< Y1< 1.2Y2
When, take the magnesium powder amount calculated value Y of neural network model1, otherwise take the magnesium powder amount calculated value Y of regression model2；
J, renewal learning group after current heat, judges whether current heat meets the condition into study group, decision condition
Are as follows: tank switching molten steel sulfur content (is turning in 0.005~0.1%, tank switching molten iron temperature in 1201~1480 DEG C, tank switching weight of molten iron
Furnace nominal tonnage -15)~(converter nominal tonnage+25) t, sulfur content exists in 0.0005~0.015%, magnesium powder utilization rate after desulfurization
15~65%, if eligible, replace regression model learning sample collection U1、U2、U3、U4With the learning sample of neural network model
Collect U5In oldest heat, keep sample number constant, wherein current heat enters U1、U2、U3、U4Which of sample set,
Depending on the size of tank switching molten steel sulfur content and desulfurization target sulphur content；
K, model restarting study, as regression model learning sample collection U1、U2、U3、U41 furnace of every update, return step f, again
Calculating parameter a0~a5、b0~b6、c0~c4、d0~d5, as neural network model sample set U55 furnaces of every update, return step d,
Re -training network adapts to the variation of molten iron pretreatment desulfurizing ability with the magnesium powder dosage for ensuring to be calculated automatically.
2. the method according to claim 1 for accurately calculating desulfurizing iron magnesium powder amount, which is characterized in that the step a
In, to prevent the heat containing abnormal data from entering study group U1~U5, when acquiring the historical data nearest apart from current heat,
Heat is screened, screening conditions include: tank switching molten steel sulfur content 0.005~0.1%, tank switching molten iron temperature 1201~
1480 DEG C, weight of molten iron exists are as follows: sulfur content is 0.0005 after+25 tons of converter nominal tonnage -15 tons~converter nominal tonnage, desulfurization
~0.015%, magnesium powder utilization rate is 15%~65%.
3. the method according to claim 2 for accurately calculating desulfurizing iron magnesium powder amount, which is characterized in that the step a
In, magnesium powder utilization rate is the ratio of the total magnesium powder amount of magnesium powder amount Zhan for desulphurization reaction, calculation formula are as follows: (tank switching molten iron sulphur contains
Sulfur content after amount-desulfurization) the total magnesium powder amount of × tank switching iron water amount × 0.75 ÷.
4. the method according to claim 3 for accurately calculating desulfurizing iron magnesium powder amount, which is characterized in that the step d
In, weight is identical with the method for adjustment of threshold value, specific as follows:
(1) network is initialized, gives w respectivelyih、who、θh、θoAssign the random number in a section (- 1,1), setup algorithm accuracy value
α, maximum study number T, learning rate η, wherein can learning rate affect the speed of network convergence and network and restrain, study
Rate setting is less than normal can to guarantee network convergence, but convergence rate is slack-off, and the training time increases, on the contrary, learning rate setting is bigger than normal
Then there is a possibility that network training is not restrained, recognition effect is influenced；
(2) sample set U is chosen5In kth furnace input sample after pretreatmentAnd the corresponding phase
Hope output
(3) outputting and inputting for hidden layer and each neuron of output layer is calculated, wherein the output of hidden layer uses function
Each neuron of hidden layer inputs hih(k) and output hoh(k),
Each neuron input yi (k) of output layer and output yo (k),
(4) global error is calculatedJudge whether E meets the requirements, if E > learns precision α
When, (5) and step (6) are entered step, weight and threshold value are adjusted, adjustment terminates as E≤α or study number t > T；
(5) error function is calculated to the partial derivative of weight and threshold value
Wherein, ekFor network output error function
δh(k)=δo(k)whof'(hih(k))=δo(k)whohih(k)(1-hih(k))
(6) weight w is adjusted using iterative methodho(k)、wih(k) and threshold θo、θh, t is study number,
5. the method according to claim 4 for accurately calculating desulfurizing iron magnesium powder amount, which is characterized in that the step g
In, when tank switching molten iron temperature, tank switching weight of molten iron and tank switching molten steel sulfur content missing, calculate sample set U5Middle distance is current
The tank switching molten iron temperatures of 20 nearest furnace data of heat, tank switching weight of molten iron, tank switching molten steel sulfur content average value, by tank switching iron
The tank switching molten iron temperature and tank switching weight of molten iron of coolant-temperature gage average value and tank switching weight of molten iron average value as current heat, will
Tank switching molten steel sulfur content of 1.1 times of the tank switching molten steel sulfur content average value as current heat.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510542806.8A CN106503413B (en) | 2015-08-31 | 2015-08-31 | A method of accurately calculating desulfurizing iron magnesium powder amount |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510542806.8A CN106503413B (en) | 2015-08-31 | 2015-08-31 | A method of accurately calculating desulfurizing iron magnesium powder amount |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106503413A CN106503413A (en) | 2017-03-15 |
| CN106503413B true CN106503413B (en) | 2019-06-18 |
Family
ID=58286265
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510542806.8A Active CN106503413B (en) | 2015-08-31 | 2015-08-31 | A method of accurately calculating desulfurizing iron magnesium powder amount |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106503413B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109811107B (en) * | 2017-11-21 | 2021-08-13 | 上海梅山钢铁股份有限公司 | Method for controlling sulfur content in molten iron pre-desulfurization process |
| CN110760642B (en) * | 2018-07-27 | 2021-08-13 | 宝山钢铁股份有限公司 | Control method of desulfurizer input amount in molten iron desulfurization process |
| CN109840309B (en) * | 2018-11-01 | 2023-01-31 | 河钢股份有限公司 | Method for calculating molten iron desulfurizer consumption |
| CN111382393B (en) * | 2020-04-29 | 2024-01-09 | 鞍钢股份有限公司 | Molten iron pretreatment desulfurization powder computing method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101463407A (en) * | 2008-11-22 | 2009-06-24 | 大连理工大学 | Method for calculating converter steel melting lime adding amount |
| 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 |
| CN103074461A (en) * | 2012-10-17 | 2013-05-01 | 南京梅山冶金发展有限公司 | Scheduling device for steelmaking molten iron desulphurization model |
| CN103390098A (en) * | 2013-06-28 | 2013-11-13 | 武汉钢铁(集团)公司 | Desulfurization computing method and system applying same |
-
2015
- 2015-08-31 CN CN201510542806.8A patent/CN106503413B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101463407A (en) * | 2008-11-22 | 2009-06-24 | 大连理工大学 | Method for calculating converter steel melting lime adding amount |
| 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 |
| CN103074461A (en) * | 2012-10-17 | 2013-05-01 | 南京梅山冶金发展有限公司 | Scheduling device for steelmaking molten iron desulphurization model |
| CN103390098A (en) * | 2013-06-28 | 2013-11-13 | 武汉钢铁(集团)公司 | Desulfurization computing method and system applying same |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106503413A (en) | 2017-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106503413B (en) | A method of accurately calculating desulfurizing iron magnesium powder amount | |
| CN108676955B (en) | A kind of BOF Steelmaking Endpoint carbon content and temprature control method | |
| CN105224772B (en) | A kind of Nenjiang representative region Ecology methods of risk assessment based on system dynamics decision model | |
| CN105925750A (en) | Steelmaking end point prediction method based on neural networks | |
| CN101825870B (en) | Method and system for controlling supply quantity of water-treatment flocculating agent | |
| CN106873379A (en) | A kind of sewage disposal method for optimally controlling based on iteration ADP algorithms | |
| CN102937784A (en) | Artificial neural network based method for controlling online prediction of casting billet quality | |
| CN109447346A (en) | Based on gray prediction and neural network ensemble model converter oxygen consumption prediction technique | |
| CN106345823B (en) | The method of online real-time estimate mechanical performance based on coils of hot-rolled steel production procedure | |
| CN106503792B (en) | A kind of gas density prediction technique based on adaptation module neural network | |
| CN102540879A (en) | Multi-target evaluation optimization method based on group decision making retrieval strategy | |
| CN106119458A (en) | Converter steelmaking process cost control method based on BP neutral net and system | |
| CN104680012A (en) | Calculating model for sintering and burdening | |
| Tejaswini et al. | Multi-objective optimization based controller design for improved wastewater treatment plant operation | |
| CN106096724B (en) | Converter steelmaking process cost control method and system based on ELM neural network | |
| CN102033978A (en) | Method for forecasting and producing narrow hardenability strip steel by hardenability | |
| CN107368125A (en) | A kind of blast furnace temperature control system and method based on CBR Yu the parallel mixed inferences of RBR | |
| CN107256439A (en) | Joint EEMD and neutral net short-term load forecasting method and system | |
| CN106874568B (en) | A kind of material particular diameter distribution forecasting method of mechanical milling process | |
| Yan et al. | Multi-objective evaluation-based hybrid intelligent control optimization for shaft furnace roasting process | |
| CN107274007A (en) | Finance data Forecasting Methodology based on artificial neural network | |
| CN107703760A (en) | Absorbing natural gas tower sweetening process control method based on RBF and GDHP | |
| CN106529816A (en) | Power line channel adjusting method and system | |
| CN107977742A (en) | A kind of construction method of long-medium term power load forecasting model | |
| Janli et al. | Self-related process residual control chart based on neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |