CN101463407B - Method for calculating converter steel melting lime adding amount - Google Patents

Abstract

The invention belongs to the automatic control technical field and relates to the construction of a converter steelmaking static model, in particular to a calculating method of the adding amount of lime during a converter steelmaking process. In the aspect of constructing an alkalinity deviation estimation model, a proper variable is selected as the input of the alkalinity deviation estimation model through the analysis of reasons causing deviations between the input alkalinity and the slag terminal alkalinity. Then, qualified heat data in a historical database are used for constructing the alkalinity deviation estimation model of a support vector machine and forecasting the deviation between the input alkalinity and the slag terminal alkalinity of the current heat. In the aspect of the calculation of the lime adding amount, a predictive value of the alkalinity deviation estimation model is used for rectifying alkalinity parameters in an empirical formula and eliminating effects of the lime adding amount on the adding amount of dolomite. Finally, a calculation formula of the lime adding amount is obtained. The calculating method of the adding amount of lime during the converter steelmaking process has the beneficial effects of effectively improving the calculation accuracy of the lime adding amount and simultaneously ensuring the slag terminal alkalinity to satisfy technological requirements.

Description

A kind of method for calculating converter steel melting lime adding amount

Technical field

The invention belongs to technical field of automatic control, relate to the foundation that static model are produced in converter steelmaking, the method for calculation of lime adding amount in particularly a kind of converter steelmaking production process.

Background technology

It is with impurity contents such as molten iron carbon drop, intensification, reduction phosphorus sulphur that converter steelmaking is produced, and obtains the commercial run of qualified molten steel., add slag making materials impurity is removed from molten steel the impurity element oxidation in the molten iron by top blast oxygen.Slagging regime is one of process system important during converter steelmaking is produced, and the basicity of slag is to weigh the important indicator of slag making quality, also be influence phosphorus, sulphur removes one of principal element of ratio.Basicity is normally defined CaO content and SiO in the slag ₂The ratio of content: R=w (CaO)/w (SiO ₂).Control suitable lime adding amount, guarantee that the slag end-point alkalinity is most important for producing qualified molten steel.

The lime adding amount computation model belongs to the part of converter static model.At present, the modeling method of calculating at lime adding amount mainly contains:

Experimental formula method based on material balance, the experimental formula method with silicone content in the molten iron and process goal basicity as the reference (Chen Zhongwei that determines lime adding amount, Yuan keeps the static mathematical model and the realization [J] of modest .LD converter smelting. steel-making, 2000,16 (5): 31-34);

The increment Return Law based on statistical theory, the increment homing method is the lime adding amount (Zhu Guangjun of the current heat of basic calculation with the increment size between current heat molten iron condition and the calibration furnace time, this river of beam. Converter static control is optimized model [J]. steel-making, 1999,15 (4): 25-28).

Yet these two kinds of lime adding amount method of calculation can't guarantee well that the slag end-point alkalinity satisfies processing requirement, sometimes even make the basicity of slag fluctuation between heat very big.

Summary of the invention

The technical problem to be solved in the present invention provides a kind of method for calculating converter steel melting lime adding amount, this method is by the deviation between basicity estimation of deviation model prediction input basicity and the slag end-point alkalinity, revise the basicity parameter in the lime adding amount empirical Calculation formula, satisfy processing requirement to guarantee the slag end-point alkalinity.

Technical scheme of the present invention is:

Analysis is bessemerized influences lime dissolved factor in the process, select suitable input variable according to analytical results, and set up SVMs basicity estimation of deviation model, forecasts the deviation between current heat input basicity and the terminal point basicity of slag.Utilize the basicity value in the estimated value correction experimental formula of basicity deviation afterwards, the lime adding amount computation model that is improved.

The step of SVMs basicity estimation of deviation modelling is as follows:

Step 1: the principal element of analyzing influence slag end-point alkalinity comprises slag composition, bath temperature, lime quality and blowing oxygen quantity.

Step 2: determine the input variable of SVMs basicity estimation of deviation model, comprise molten iron silicon content, molten iron manganese content, molten iron phosphorus content, molten steel sulfur content, rhombspar add-on, Molten Steel End Point, lime activity and the total oxygen-consumption of heat.

Step 3 makes up the standard heat storehouse that is applicable to modeling:

(1) data logging that selection has the terminal point slag composition to chemically examine in historical data base;

(2) data logging of selecting endpoint carbon content and temperature to hit simultaneously;

(3) select molten iron silicon content between 0.2～0.6%, and the data logging of slag end-point alkalinity between 2.8～3.5, standard heat storehouse set up.

Step 4, set up SVMs basicity estimation of deviation model:

(1) the input basicity R of each heat in the base of calculation heat storehouse _Input:

$R_{\mathrm{input}}=\frac{(\mathrm{Ca}{O}_{\mathrm{lime}}+\mathrm{Ca}{O}_{\mathrm{dolomite}})}{(\mathrm{Si}{\mathrm{<figure-callout\; id="2"\; label="O"\; filenames="H2008102290126E00012.png,H2008102290126E00022.png"\; state="\{\{state\}\}">O</figure-callout>}}_{2\mathrm{lime}}+\mathrm{<figure-callout\; id="2"\; label="Si\; O"\; filenames="H2008102290126E00012.png,H2008102290126E00022.png"\; state="\{\{state\}\}">Si</figure-callout>}{O}_{}{}_{2\mathrm{dolomite}}+K\&times;w\left[\mathrm{Si}\right]\&times;W_{\mathrm{HM}}+\mathrm{<figure-callout\; id="2"\; label="Si\; O"\; filenames="H2008102290126E00012.png,H2008102290126E00022.png"\; state="\{\{state\}\}">Si</figure-callout>}{O}_{}{}_{2\mathrm{scrap}})}$

Wherein, CaO _LimeThe calcium oxide content of bringing into for lime; CaO _DolomiteThe calcium oxide content of bringing into for rhombspar; SiO _2limeThe silica volume of bringing into for lime; SiO _2dolomiteBe the silica volume of bringing into by rhombspar; K=2.14 is SiO ₂The ratio of the molecular mass of/Si; K * w[Si] * W _HMSilica volume for pasc reaction generation in the molten iron; SiO _2scrapBe silica volume of bringing into by steel scrap and the silica volume that reacts generation thereof;

(2) calculate deviation delta R between input basicity and the slag end-point alkalinity, as the output of SVMs basicity estimation of deviation model:

ΔR＝R _input-R _aim

Wherein, R _InputBe input basicity; R _AimBe target basicity;

(3) normalization method input/output variable, and the kernel function and the loss function of selection SVMs;

(4) use the crosscheck method to determine the parameter of SVMs basicity estimation of deviation model.

The step that current heat lime adding amount is calculated is as follows:

Step 1, forecast the deviate of current heat input basicity and slag end-point alkalinity:

(1) search and this heat molten iron silicon content, molten iron manganese content, molten iron phosphorus content, molten steel sulfur content, rhombspar add-on, lime activity and the most close heat of terminal point target temperature in standard heat storehouse, with the reference of total oxygen-consumption of this heat as the total oxygen-consumption of current heat, use the process goal temperature of current heat of molten steel to replace Molten Steel End Point simultaneously, as the input of this heat SVMs basicity estimation of deviation model;

(2) the current heat molten iron silicon content that will determine, molten iron manganese content, molten iron phosphorus content, molten steel sulfur content, rhombspar add-on, Molten Steel End Point, lime activity and the total oxygen-consumption normalization method of heat are brought normalized each input variable value into SVMs basicity estimation of deviation Model Calculation;

(3) with the anti-normalization method of calculation result of SVMs basicity estimation of deviation model, obtain predicted value to current heat basicity deviation.

Step 2: use the basicity deviate of forecast that the basicity parameter in the experimental formula is revised, and calculate lime adding amount:

$W_{\mathrm{lime}}=\frac{K\&times;w\left[\mathrm{Si}\right]\&times;W_{\mathrm{HM}}\&times;(R_{\mathrm{aim}}+\mathrm{\&Delta;R})}{w{\left(\mathrm{CaO}\right)}_{\mathrm{lime}\_\mathrm{eff}}}-\frac{W_{\mathrm{dolomite}}\&times;w\left(\mathrm{Ca}{O}_{\mathrm{dolomite}\_\mathrm{eff}}\right)}{w\left(\mathrm{Ca}{O}_{\mathrm{lime}\_\mathrm{eff}}\right)}$

Wherein, K=2.14 is SiO ₂The ratio of the molecular mass of/Si; W[Si] be the massfraction of silicon in the molten iron; W _HMBe molten steel quality; R _AimBe target basicity; Δ R is the estimated value of basicity deviation; W (CaO _{Lime_eff}) be the calcareous amount mark of the efficient oxidation in the lime; W _DolomiteBe the rhombspar add-on; W (CaO _{Dolomite_eff}) be the massfraction of the efficient oxidation calcium in the rhombspar.

After the current heat finishing blowing, if carry out slag chemical examination, and basicity of slag and endpoint carbon content and temperature satisfy processing requirement, then this heat information added standard heat storehouse, uses for setting up SVMs basicity estimation of deviation model.

The gordian technique of aforesaid method is to have set up SVMs basicity estimation of deviation model, forecasts the deviation between current heat input basicity and the slag end-point alkalinity, and basicity deviate is according to weather report adjusted lime adding amount.

Effect of the present invention and benefit are to remedy deviation between input basicity and the end-point alkalinity preferably, improve the accuracy that lime adding amount calculates, make the slag end-point alkalinity more near the target basicity value of processing requirement, help controlling the stable in properties of slag between heat.

Description of drawings

Fig. 1 is a method of calculation synoptic diagram of the present invention.

Wherein, R _AimBe terminal point target basicity; x ₁, x ₂... x _nInput for SVMs basicity estimation of deviation model; Δ R is input basicity and slag end-point alkalinity predicted value; R _InputBe input basicity; W _LimeBe lime adding amount; R _EndBe the slag end-point alkalinity; ε ₁Actual value for input basicity and slag end-point alkalinity deviation; ε ₂Poor for basicity deviation predicted value and actual value.

Fig. 2 is a SVMs basicity estimation of deviation model structure synoptic diagram.

Wherein, x ₁Be molten iron silicon content; x ₂Be molten iron manganese content; x ₃Be the molten iron phosphorus content; x ₄Be molten steel sulfur content; x ₅Be the rhombspar add-on; x ₆Be terminal temperature; x ₇Be lime activity; x ₈Be the stove oxygen-consumption; Y is the basicity deviation; S is final support vector number.

The extensive error curve diagram of test samples of correspondence when Fig. 3 gets different value for parameter beta and C.

Fig. 4 is the predict the outcome figure of SVMs basicity estimation of deviation model to test sample book.

Fig. 5 is the synoptic diagram that concerns between the calculated value of test sample book lime adding amount and the actual value.

Wherein, long and short dash line represents that calculated value equates with actual value, and two interior points of solid line all satisfy Error Absolute Value in 1 ton scope, and two interior points of dotted line all satisfy Error Absolute Value in 0.7 ton scope.

Embodiment

Be described in detail specific embodiments of the invention below in conjunction with technical scheme and accompanying drawing 1.

Analyze the influencing factor that produces the basicity deviation, determine the input of SVMs basicity estimation of deviation model.

In actual production, deviation is that incomplete dissolving by lime causes between input basicity and the blow end point basicity of slag.At the blowing initial stage, the lime surface that adds converter generates fusing point up to 2130 ℃ and fine and close hard Dicalcium Phosphate (Feed Grade) layer, hinders the dissolving of lime.The dissolving of lime continues whole converting process, in addition during finishing blowing lime still in dissolving.By the principal element of analyzing influence lime dissolution rate, determine the input of SVMs basicity estimation of deviation model:

A. the composition of slag: it has a significant impact the lime dissolution rate, and the content of compositions such as the calcium oxide in the slag, magnesium oxide, manganese oxide and ferric oxide all affects the dissolution rate of lime.Slag is mainly by the auxiliary material dissolving of the oxidation of elements such as the silicon in the molten iron, phosphorus, manganese, iron and adding and generate.

B. bath temperature: the bath temperature height helps the reduction of slag viscosity, quickens slag and permeates in lime block, impels the rapid fusion of the Dicalcium Phosphate (Feed Grade) formation slag that comes off, thereby improves the dissolution rate of lime.

C. the quality of lime: surface porosity, void content height, the quickened lime that response capacity is strong, help slag and enter in the lime block, enlarge reaction area, quicken the dissolution process of lime.The activity of lime is the physical quantity that characterizes the lime hydration speed of response, also is an important parameter that influences the lime dissolution rate.

D. blowing oxygen quantity: the dissolving of lime can continue whole process, and the carbon drop amount that how much depends primarily on of blowing oxygen quantity and the requirement of intensification amount increase the abundant dissolving that blowing oxygen quantity helps lime, improves basicity of slag.

To sum up, eight input variables determining SVMs basicity estimation of deviation model are respectively: molten iron silicon content, molten iron manganese content, molten iron phosphorus content, molten steel sulfur content, rhombspar add-on, Molten Steel End Point, lime activity and the total oxygen-consumption of heat.

After the input of determining SVMs basicity estimation of deviation model, need to select suitable sample to carry out modeling.In actual production, the slag of not all heat all can be chemically examined, and therefore selects to have the data of terminal point slag composition chemical examination; Endpoint carbon content and temperature are to weigh the leading indicator of bessemerizing quality, therefore select endpoint carbon content and temperature to hit the data logging of heat simultaneously; For guaranteeing not carry out the deslagging operation in the converting process, select the data logging of molten iron silicon content between 0.2～0.6%; According to the requirement of on-the-spot technology, the slag end-point alkalinity preferably is controlled between 2.8～3.5, thereby selects the data logging of slag end-point alkalinity between 2.8～3.5.So selecting data is in order to guarantee the accuracy of SVMs basicity estimation of deviation model.

Use the above-mentioned data of selecting to set up SVMs basicity estimation of deviation model, model is output as the difference of input basicity and slag end-point alkalinity.Input basicity R _InputBe defined as: contain in the calcium oxide content that contains in all auxiliary materials of the converter of packing into and the main auxiliary material and the ratio of the silica volume that may generate.Calculation formula is

$R_{\mathrm{input}}=\frac{(\mathrm{Ca}{O}_{\mathrm{lime}}+\mathrm{Ca}{O}_{\mathrm{dolomite}})}{(\mathrm{<figure-callout\; id="2"\; label="Si\; O"\; filenames="H2008102290126E00012.png,H2008102290126E00022.png"\; state="\{\{state\}\}">Si</figure-callout>}{O}_{}{}_{2\mathrm{lime}}+\mathrm{<figure-callout\; id="2"\; label="Si\; O"\; filenames="H2008102290126E00012.png,H2008102290126E00022.png"\; state="\{\{state\}\}">Si</figure-callout>}{O}_{}{}_{2\mathrm{dolomite}}+K\&times;w\left[\mathrm{Si}\right]\&times;W_{\mathrm{HM}}+\mathrm{<figure-callout\; id="2"\; label="Si\; O"\; filenames="H2008102290126E00012.png,H2008102290126E00022.png"\; state="\{\{state\}\}">Si</figure-callout>}{O}_{}{}_{2\mathrm{scrap}})}---\left(1\right)$

Wherein:

CaO _LimeThe calcium oxide content of bringing into for lime;

CaO _DolomiteThe calcium oxide content of bringing into for rhombspar;

SiO _2limeThe silica volume of bringing into for lime;

SiO _2dolomiteBe the silica volume of bringing into by rhombspar;

K=2.14 is SiO ₂The ratio of the molecular mass of/Si;

K * w[Si] * W _HMSilica volume for pasc reaction generation in the molten iron;

SiO _2scrapBe silica volume of bringing into by steel scrap and the silica volume that reacts generation thereof.

SVMs basicity estimation of deviation model is output as the deviation delta R of input basicity and blow end point basicity, has:

ΔR＝R _input-R _aim (2)

Wherein:

R _InputBe input basicity;

R _AimBe target basicity.

Because there is different dimensions in selected inputoutput data, and different dimensional data numerically differ bigger, some in addition differ several magnitude.For avoiding variable bigger on the numerical value to fall into oblivion less variable, each input and output variable is done normalized:

$x_i^{\&prime;}\left(k\right)=\frac{x_i\left(k\right)-x_{i\left(\min \right)}}{x_{i\left(\max \right)}-x_{i\left(\min \right)}}---\left(3\right)$

In the formula (4):

I=1,2, L, 8 corresponding i variablees;

K=1,2, L, k sample in the corresponding every group of variable of n, n is a total sample number;

x _{I (min)}It is the minimum value in the i group variable;

x _{I (max)}It is the maximum value in the i group variable;

x _i(k) be k original value in the i group variable;

It is the value after k normalization method in the i variable.

The computation process of SVMs is as follows:

If sampled data is { (x ₁, y ₁), (x ₂, y ₂) ..., (x _N, y _N), x wherein _i∈ R ^m, y _i∈ R, i=1 ..., N.SVMs at first passes through nonlinear mapping Φ () the importation { x of sample ₁, x ₂..., x _NBe mapped to high-dimensional feature space F, carry out linear regression then

f(x)＝w ^TΦ(x)+b (4)

Wherein, w ∈ F, b represent biasing, w ^TThe dot product of Φ (x) expression w and Φ (x) vector.(w, asking for b) can draw by the optimization to following formula unknown quantity

$\min C\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}{E}_{\mathrm{\&epsiv;}}(f\left(x_n\right)-y)+\frac{1}{2}{\left|\right|w\left|\right|}^2---\left(5\right)$

Wherein ‖ ‖ is an Euclidean distance, and C is the regularization coefficient, E _ε(g) be loss function.In order to obtain sparse solution, improve computing velocity to test sample book, the error loss function is chosen as and has the insensitive loss function of ε among the present invention:

For each sample point is introduced two relaxation factor ξ _nAnd ξ _n, optimization problem is converted into the convex quadratic programming shown in the following formula:

$\min C\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}({\mathrm{\&xi;}}_n+{\widehat{\mathrm{\&xi;}}}_n)+\frac{1}{2}{\left|\right|w\left|\right|}^2$

s.t.y _i-w ^TΦ(x _i)-b≤εt _i+ξ _i (7)

$w^T\mathrm{\&Phi;}\left(x_i\right)+b-y_i\&le;\mathrm{\&epsiv;}{t}_i+{\widehat{\mathrm{\&xi;}}}_i$

${\mathrm{\&xi;}}_i\&GreaterEqual;0,{\widehat{\mathrm{\&xi;}}}_i\&GreaterEqual;0$

For separating this optimization problem, introduce Lagrangian multiplier and this optimization problem be converted into lagrange duality problem:

$\max \{W(a,a\%)=-\frac{1}{2}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}\underset{m=1}{\overset{N}{\mathrm{\&Sigma;}}}(a_n-{\hat{a}}_n)(a_m-{\hat{a}}_m)k(x_n,x_m)$

$-\mathrm{\&epsiv;}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}(a_n+{\hat{a}}_n)+\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}(a_n-{\hat{a}}_n)y_n\}$

$s.t.\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}(a_n-{\hat{a}}_n)=0---\left(8\right)$

0≤a _n≤C

$0\&le;{\hat{a}}_n\&le;C$

Wherein, a _nWith Be the Lagrange multiplier, k (x _n, x _m) be the kernel function equation, kernel function is chosen as radially basic kernel function among the present invention:

$k(x_n,x_m)=e^{-\frac{{(x_n-x_m)}^T(x_n-x_m)}{2{\mathrm{\&beta;}}^2}}---\left(9\right)$

Wherein, β is the RBF width.

RBF width beta and regular terms coefficient C use the crosscheck method to determine.

Because the application of an insensitive loss function only has minority Value is not 0, and its pairing vector is a support vector.The value of biasing b can calculate by Karush-Kuhn-Tucker (KKT) condition.

Finally set up SVMs basicity estimation of deviation model, its structure iron as shown in Figure 2:

$y\left(x\right)=\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}(a_n-{\hat{a}}_n)k(x,x_n)+b---\left(10\right)$

In order to obtain actual basicity deviate, need do following anti-normalization method conversion to the output result of SVMs basicity estimation of deviation model:

ΔR＝y′(k)×(y _max-y _min)+y _min (11)

Wherein:

Δ R is the estimated value of basicity deviation;

The normalized value that y ' (k) exports for model;

y _MaxBe the maximum deflection difference value in the learning sample;

y _MinBe the minimum deviation value in the learning sample.

What use when setting up SVMs basicity estimation of deviation model is the data logging that the blowing result meets processing requirement in the historical data.When current heat was predicted, some input variables can't directly obtain, and need determine these variablees.The method of determining is as follows:

For Molten Steel End Point, only after finishing blowing, could obtain, and bessemerize terminal temperature is required relatively strictness, the target temperature of terminal temperature and processing requirement is more approaching, therefore, select the target temperature of current heat processing requirement to replace the input of Molten Steel End Point as SVMs basicity estimation of deviation model.

For rhombspar add-on and the total oxygen-consumption of heat, search and the historical heat of each smelting condition of current heat success the most similar and that target call is identical (the end point carbon temperature is hit and basicity satisfies processing requirement) are with the input as current heat SVMs basicity estimation of deviation model of the rhombspar add-on of this history heat and the total oxygen-consumption of heat.

Molten iron silicon content, molten iron manganese content, molten iron phosphorus content, molten steel sulfur content, rhombspar add-on, Molten Steel End Point, lime activity and the total oxygen-consumption of heat of definite current heat are carried out normalization method with setting up the employed data of SVMs basicity estimation of deviation model.

Bring each input variable value of the current heat after the normalization method into SVMs basicity estimation of deviation model and calculate,, obtain predicted value Δ R current heat basicity deviation again to the anti-normalization method of output valve of SVMs basicity estimation of deviation model.

The experimental formula of lime adding amount is relevant with the technology basicity requirement of slag, is expressed as:

$W_{\mathrm{lime}}=\frac{K\&times;w\left[\mathrm{Si}\right]\&times;W_{\mathrm{HM}}}{w\left(\mathrm{Ca}{O}_{\mathrm{lime}\_\mathrm{eff}}\right)}\&times;R_{\mathrm{aim}}---\left(12\right)$

Wherein:

R _AimTarget basicity for slag;

W (CaO _{Lime_eff}) be the calcareous amount mark of the efficient oxidation in the lime, and w (CaO _{Lime_eff})=w (CaO _Lime)-R _Aim* w (SiO _2lime);

K=2.14 is SiO ₂The ratio of the molecular mass of/Si;

W[Si] be the massfraction of silicon in the molten iron;

W _HMBe molten steel quality.

Owing to also contain CaO and SiO in the rhombspar ₂, can the basicity value of terminal point be exerted an influence, for eliminating the influence of rhombspar, experimental formula is revised as lime adding amount:

$W_{\mathrm{lime}}=\frac{K\&times;w\left[\mathrm{Si}\right]\&times;W_{\mathrm{HM}}\&times;\left(R_{\mathrm{aim}}\right)}{w{\left(\mathrm{CaO}\right)}_{\mathrm{lime}\_\mathrm{eff}}}-\frac{W_{\mathrm{dolomite}}\&times;w\left(\mathrm{Ca}{O}_{\mathrm{dolomite}\_\mathrm{eff}}\right)}{w\left(\mathrm{Ca}{O}_{\mathrm{lime}\_\mathrm{eff}}\right)}---\left(13\right)$

Wherein:

K=2.14 is SiO ₂The ratio of the molecular mass of/Si;

W _DolomiteBe the rhombspar add-on;

W (CaO _{Dolomite_eff}) be the massfraction of the efficient oxidation calcium in the rhombspar;

And w (CaO _{Dolomite_eff})=w (CaO _Dolomite)-R _Aim* w (SiO _2dolomite).

Further, use the basicity deviate of SVMs basicity estimation of deviation model prediction that the basicity parameter in the experimental formula is revised, the calculation formula that finally obtains lime adding amount is:

$W_{\mathrm{lime}}=\frac{K\&times;w\left[\mathrm{Si}\right]\&times;W_{\mathrm{HM}}\&times;(R_{\mathrm{aim}}+\mathrm{\&Delta;R})}{w{\left(\mathrm{CaO}\right)}_{\mathrm{lime}\_\mathrm{eff}}}-\frac{W_{\mathrm{dolomite}}\&times;w\left(\mathrm{Ca}{O}_{\mathrm{dolomite}\_\mathrm{eff}}\right)}{w\left(\mathrm{Ca}{O}_{\mathrm{lime}\_\mathrm{eff}}\right)}---\left(14\right)$

The lime adding amount that uses following formula to calculate has taken into full account influences the reason that basicity produces deviation, and it is compensated, and guarantees that at utmost the slag end-point alkalinity satisfies processing requirement.

After the current heat finishing blowing, if carry out slag chemical examination, and basicity of slag and end point carbon temperature satisfy processing requirement, then this heat information added standard heat storehouse, uses when later heat is set up SVMs basicity estimation of deviation model.

Example: the validity for checking institute of the present invention extracting method, adopt the actual production data of certain 150 tons of converter of steel mill to test.Select 230 groups of data according to the requirement of setting up SVMs basicity estimation of deviation model step 3, preceding 150 groups are used as modeling, and the back is used as test for 80 groups.When using definite radially sound stage width degree parameter beta of crosscheck method and regular terms coefficient C, use 150 preceding 100 groups of setting up in the modulus certificate to carry out modeling, test in back 50, definite process of parameter is used the grid search method, its Search Results as shown in Figure 3, selecting one group of optimum parameter is β=0.62, C=24000.

Use the SVMs basicity estimation of deviation model of setting up that 80 groups of test datas are carried out the basicity estimation of deviation, the relation between predicted value and the calculated value as shown in Figure 4.The accurate calculating that accurately is established as lime adding amount of basicity estimation of deviation model provides strong guarantee.With the predicted value substitution formula (14) of basicity deviation, the relation between lime adding amount that obtains and the actual add-on as shown in Figure 5.As can be seen, be that the data point of coordinate is evenly distributed on dashdotted both sides with calculated value and actual value, the absolute value major part of the error between calculated value and the actual value is all in 0.7 ton scope.

Use same data, method of the present invention and experimental formula method and increment homing method are compared, the result is as shown in table 1.

Lime adding amount Model Calculation value and the square error between the actual value based on SVMs basicity estimation of deviation are 0.2042 ton, Error Absolute Value is 96.34% less than the shared ratio of 1 ton test sample book point, and Error Absolute Value is 89.02% less than the shared ratio of 0.7 ton test sample book point.The method of this paper all is better than existing method on every index, and brings up to 0.7 ton hour when error precision by 1 ton, and the degree that accuracy rate descends is significantly less than existing method.

The comparison of table 1 and traditional method

Claims (4)

1. method for calculating converter steel melting lime adding amount, set up SVMs basicity estimation of deviation model, utilize the deviation between current heat input basicity of basicity estimation of deviation model prediction and slag end-point alkalinity, use the basicity parameter in this deviate compensation experimental formula; Described experimental formula is as follows:

$W_{\mathrm{lime}}=\frac{2.14\&times;w\left[\mathrm{Si}\right]\&times;W_{\mathrm{HM}}}{w\left({\mathrm{CaO}}_{\mathrm{lime}\_\mathrm{eff}}\right)}\&times;R_{\mathrm{aim}}$

Wherein, R _AimTarget basicity for slag; W (CaO _{Lime_eff}) be the massfraction of the efficient oxidation calcium in the lime; 2.14 be SiO ₂The ratio of the molecular mass of/Si; W[Si] be the massfraction of silicon in the molten iron; W _HMBe molten steel quality;

The massfraction calculation formula of the efficient oxidation calcium is as follows in the lime:

w(CaO _{lime_eff})＝w(CaO _lime)-R _aim×w(SiO _2lime)

Calculate lime adding amount after the compensation basicity parameter, its calculation formula is as follows;

$W_{\mathrm{lime}}=\frac{K\&times;w\left[\mathrm{Si}\right]\&times;W_{\mathrm{HM}}\&times;(R_{\mathrm{aim}}+\mathrm{\&Delta;R})}{w\left({\mathrm{CaO}}_{\mathrm{lime}\_\mathrm{eff}}\right)}-\frac{W_{\mathrm{dolomite}}\&times;w\left({\mathrm{CaO}}_{\mathrm{dolomite}\_\mathrm{eff}}\right)}{w\left({\mathrm{CaO}}_{\mathrm{lime}\_\mathrm{eff}}\right)}$

Wherein, K=2.14 is SiO ₂The ratio of the molecular mass of/Si; W[Si] be the massfraction of silicon in the molten iron; W _HMBe molten steel quality; R _AimBe target basicity; Δ R is the deviation between input basicity and the slag end-point alkalinity; W (CaO _{Lime_eff}) be the calcareous amount mark of the efficient oxidation in the lime; W _DolomiteBe the rhombspar add-on; W (CaO _{Dolomite_eff}) be the massfraction of the efficient oxidation calcium in the rhombspar, its calculation formula is as follows:

w(CaO _{dolomite_eff})＝w(CaO _dolomite)-R _aim×w(SiO _2dolomite)

It is characterized in that:

The step of SVMs basicity estimation of deviation modelling is as follows:

(1) principal element of analyzing influence slag end-point alkalinity comprises slag composition, bath temperature, lime quality and blowing oxygen quantity;

(2) determine the input variable of SVMs basicity estimation of deviation model, comprise molten iron silicon content, molten iron manganese content, molten iron phosphorus content, molten steel sulfur content, rhombspar add-on, Molten Steel End Point, lime activity and the total oxygen-consumption of heat;

(3) make up the standard heat storehouse that is applicable to modeling;

(4) set up SVMs basicity estimation of deviation model;

The step that current heat lime adding amount is calculated is as follows:

(1) deviate of current heat input basicity of forecast and slag end-point alkalinity;

(2) use the basicity deviate of forecast that the basicity parameter in the experimental formula is revised;

(3) calculate current heat lime adding amount;

After the current heat finishing blowing,, then this heat information is added standard heat storehouse, use for setting up SVMs basicity estimation of deviation model if basicity of slag, endpoint carbon content and temperature satisfy processing requirement.

2. a kind of method for calculating converter steel melting lime adding amount according to claim 1 is characterized in that: described step (3) makes up and is applicable to that the standard heat storehouse of modeling comprises:

(1) data logging that selection has the terminal point slag composition to chemically examine in historical data base;

(2) data logging of selecting endpoint carbon content and temperature to hit simultaneously;

(3) select molten iron silicon content between 0.2～0.6%, and the data logging of slag end-point alkalinity between 2.8～3.5, standard heat storehouse set up.

3. a kind of method for calculating converter steel melting lime adding amount according to claim 1 is characterized in that: described step (4) is set up SVMs basicity estimation of deviation model and is comprised:

(1) the input basicity R of each heat in the base of calculation heat storehouse _Input:

$R_{\mathrm{input}}=\frac{({\mathrm{CaO}}_{\mathrm{lime}}+{\mathrm{CaO}}_{\mathrm{dolomite}})}{({\mathrm{SiO}}_{2\mathrm{lime}}+{\mathrm{SiO}}_{2\mathrm{dolomite}}+K\&times;w[\mathrm{Si}]\&times;W_{\mathrm{HM}}+{\mathrm{SiO}}_{2\mathrm{scrap}})}$

Wherein, CaO _LimeThe calcium oxide content of bringing into for lime; CaO _DolomiteThe calcium oxide content of bringing into for rhombspar; SiO _2limeThe silica volume of bringing into for lime; SiO _2dolomiteBe the silica volume of bringing into by rhombspar; K=2.14 is SiO ₂The ratio of the molecular mass of/Si; K * w[Si] * W _HMSilica volume for pasc reaction generation in the molten iron; SiO _2scrapBe silica volume of bringing into by steel scrap and the silica volume that reacts generation thereof;

(2) calculate deviation delta R between input basicity and the slag end-point alkalinity, as the output of SVMs basicity estimation of deviation model:

ΔR＝R _input-R _aim

Wherein, R _InputBe input basicity; R _AimBe target basicity;

(3) normalization method input/output variable, and the kernel function and the loss function of selection SVMs;

(4) use the crosscheck method to determine the parameter of SVMs basicity estimation of deviation model.

4. a kind of method for calculating converter steel melting lime adding amount according to claim 1 is characterized in that: comprise in the step 1 to current heat lime adding amount calculating:

(1) search and this heat molten iron silicon content, molten iron manganese content, molten iron phosphorus content, molten steel sulfur content, rhombspar add-on, lime activity and the most close heat of terminal point target temperature in standard heat storehouse, with the reference of total oxygen-consumption of this heat as the total oxygen-consumption of current heat, use the process goal temperature of current heat of molten steel to replace Molten Steel End Point simultaneously, as the input of this heat SVMs basicity estimation of deviation model;

(2) the current heat molten iron silicon content that will determine, molten iron manganese content, molten iron phosphorus content, molten steel sulfur content, rhombspar add-on, Molten Steel End Point, lime activity and the total oxygen-consumption normalization method of heat are brought normalized each input variable value into SVMs basicity estimation of deviation Model Calculation;

(3) with the anti-normalization method of calculation result of SVMs basicity estimation of deviation model, obtain predicted value to current heat basicity deviation.

Images