CN103729571A - Modeling method for utilization rate of carbon monoxide in iron-making process of blast furnace - Google Patents

Abstract

The invention discloses a modeling method for the utilization rate of carbon monoxide in the iron-making process of a blast furnace. The modeling method comprises the following steps: (1) data acquisition and calculation, namely acquiring operating parameters, including wind temperature, wind pressure, wind speed, wind volume, coal-injection rate, jacking pressure and volume percentages of CO and CO2 in coal gas of the blast furnace, and calculating out the utilization rate EtaCO of the carbon monoxide; (2) time-lag registration for data, namely carrying out correlation analysis on an operation parameter time sequence and a carbon-monoxide utilization rate time sequence with different time-lag degrees by using a grey-correlation-degree analysis method, thus respectively determining the time-lag time of each operating parameter, finishing time-lag registration of data and forming a sample set; and (3) establishment of a model, namely based on the sample set in the step (2), establishing a real-time predication model for the utilization rate of carbon monoxide of the blast furnace based on SVM. The modeling method disclosed by the invention has the advantage that the established model can implement accurate prediction on the utilization rate of carbon monoxide in the iron-making process of the blast furnace.

Description

A kind of modeling method of blast furnace ironmaking process carbon monoxide utilization factor

Technical field

The modeling method that the present invention relates to a kind of blast furnace ironmaking process carbon monoxide utilization factor, belongs to blast furnace ironmaking field.

Background technology

Steel and iron industry is the basic industry of national economy, is again high flow rate simultaneously, high pollution " rich and influential family ".In the face of current shortage of resources, the phenomenon that environmental pollution is day by day serious, steel and iron industry is needed badly and is born energy-saving and emission-reduction, green important task of producing.

Blast furnace ironmaking is the main power consumption operation of iron and steel flow process, indirect reduction process in carbon monoxide utilization factor reaction blast furnace, directly affect the energy consumption of ton iron, to evaluate high capacity of furnace to make good use of bad important indicator, and closely related with blast furnace stable operation, in the short-term regulation and control of blast furnace ironmaking process, bringing into play important directive function.

Yet, due to the leakproofness of blast furnace production, the reasons such as complicacy of process mechanism, make the relation that affects between carbon monoxide utilization factor and blast furnace operating be difficult to determine, cannot realize accurate quantitatively adjusting, energy consumption model is difficult to accurate foundation.

Meanwhile, because blast furnace response exists hysteresis quality, making the blast furnace carbon monoxide utilization factor that " in real time " detects is not the result of current time operation.After current operation parameter change, can response change once oxidation utilization factor, but can just can reflect that operation changes the impact bringing through after a while, i.e. the carbon monoxide utilization factor value of current detection, can only react a period of time operational circumstances before.This just greatly reduces carbon monoxide utilization factor reference value at the scene, also cannot provide effective guidance to blast furnace stable operation.

At present, also do not have carbon monoxide utilization factor model accurately, blast furnace production scene can only, by observing the real-time detection numerical value of carbon monoxide utilization factor in present and the past period, judge according to knowhow the variation tendency that it is following possible.This judgement has subjectivity, easily occurs careless mistake and error, can not guarantee the stable smooth operation that blast furnace is produced, and has also strengthened on-the-spot labour intensity, has reduced information automation level.

Due to the shortage of model, the adjusting of execute-in-place also can only by virtue of experience come to optimize as far as possible, to improve carbon monoxide utilization factor, can not reach the effect of quantitative fine adjustment.

The present invention proposes a kind of blast furnace carbon monoxide utilization factor modeling method, for determining the relation between blast furnace operating parameter and carbon monoxide utilization factor, and can, according to model, realize the blast furnace carbon monoxide utilization factor real-time estimate based on operating parameter.

Summary of the invention

Technical matters to be solved by this invention is the modeling method that proposes carbon monoxide utilization factor in a kind of blast furnace ironmaking process, the method is based on SVM algorithm, can realize the real-time accurately predicting of blast furnace ironmaking process carbon monoxide utilization factor, effectively solve at present in blast furnace ironmaking process, leakproofness and the hysteresis quality of execute-in-place on blast furnace impact due to blast furnace production, cause the relation between blast furnace operating parameter and carbon monoxide not clear, can only rely on subjective experience to judge the variation tendency of carbon monoxide utilization factor and the problem of carrying out blast furnace operating adjusting.

The technical solution of invention is as follows:

A modeling method for blast furnace ironmaking process carbon monoxide utilization factor, comprises the following steps:

Step 1: data acquisition and calculating:

Acquisition operations supplemental characteristic in the local data base [as oracle database] of blast furnace industrial computer; Described operating parameter comprises: CO and CO in wind-warm syndrome, blast, wind speed, air quantity, coal powder injection speed, top pressure, blast furnace gas ₂percent by volume; And calculate carbon monoxide utilization factor η by following formula _cO:

wherein, (CO ₂) be CO in blast furnace gas ₂percent by volume; (CO) be the percent by volume of CO in blast furnace gas;

Step 2: data are carried out to time lag registration;

After the data that gather are carried out to data pre-service, utilize grey relational grade analysis method, respectively the time series of operating parameter Unequal time lag degree and carbon monoxide utilization factor time series are carried out to correlation analysis, determine the Slack time of each operating parameter, complete the time lag registration of data, and form sample set;

Step 3: the foundation of model:

Sample set based on described in step 2, using air quantity, wind-warm syndrome, wind speed, blast, coal powder injection speed, top six supplemental characteristics of pressure as input, carbon monoxide utilization factor is output, sets up the blast furnace carbon monoxide usage forecast model based on SVM (support vector machine).Described SVM adopts RBF kernel function, and adopts cross-validation method to be in optimized selection the penalty parameter c of SVM and kernel functional parameter g.After modelling verification is good, the input parameter data of current time lag registration of take are input, utilize model can obtain the carbon monoxide utilization factor under current operation parameter;

In step 3, the step of described cross-validation method is as follows:

1. determine the span of penalty parameter c and kernel functional parameter g;

2. determine test set grouping number V;

3. test set is carried out to cross validation:, when c and g change from small to large with step-length, take each group test set is training set, and rear one group of test set, for checking collection, calculates the Average Accuracy of checking; Corresponding c and g value when retaining Average Accuracy maximum, as final parameter value.

In step 2, the data that gather are carried out to data pre-service, data pre-service comprises rejecting abnormal data, and utilizes wavelet-decomposing method denoising, the interference noise detecting to eliminate industry spot, and concrete grammar is as follows.

The method of rejecting abnormal data is: maximum 3 data and 3 minimum data in each data sequence of air quantity, wind-warm syndrome, blast, wind speed, coal powder injection speed, top pressure, these 7 data sequences of carbon monoxide utilization factor are replaced, replace with a previous moment and rear mean value of data constantly, to reduce the interference of spike data; Such as x1 (i) is in air quantity data, in 3 maximum data one, replaces x1 (i) and is: $x1\left(i\right)=\frac{x1(i-1)+x1(i+1)}{2}$

Denoising method is: each data sequence of 7 data sequences is done to following operation;

1. choosing demy is wavelet basis function, and this data sequence is carried out to 5 layers of wavelet decomposition;

2. according to soft-threshold method, select the threshold value of each layer; Two layer signals of highest frequency are shielded completely, retain low frequency signal; [ shield two layer signals after remaining signal]

3. low frequency signal is reconstructed, covers original data sequence, complete the Wavelet Denoising Method of data sample.[Wavelet Denoising Method is existing mature technology.】

In step 2, determine that the process of the Slack time of certain operating parameter is:

1. for current operating parameter, take 10 minutes as interval, extract air quantity, wind-warm syndrome, blast, wind speed, coal powder injection speed, the top of hysteresis different time and press data sample, be expressed as Z _fl, Z _fw, Z _fy, Z _fs, Z _ps, Z _dy, with air quantity under for example:

Lag behind 0 minute: z0 _fl(k)=x1 (k), k=1,2 ..., 480)

Lag behind 10 minutes: z1 _fl(k)=x1 (k), k=11,12 ..., 490;

Lag behind 20 minutes: z2 _fl(k)=x1 (k), k=21,22 ..., 500;

……

Lag behind 120 minutes: z12 _fl(k)=x1 (k), k=121,122 ..., 600;

Form matrix $Z_{\mathrm{fl}}=\left[\begin{array}{ccccc}{\mathrm{<figure-callout\; id="0"\; label="z"\; filenames="HDA0000461150970000012.png"\; state="\{\{state\}\}">z</figure-callout>}0}_{\mathrm{fl}}\left(1\right)& {z1}_{\mathrm{fl}}\left(11\right)& {z2}_{\mathrm{fl}}\left(21\right)& ...& {z12}_{\mathrm{fl}}\left(121\right)\\ {z0}_{\mathrm{fl}}\left(2\right)& {z1}_{\mathrm{fl}}\left(12\right)& {z2}_{\mathrm{fl}}\left(22\right)& ...& {z12}_{\mathrm{fl}}\left(122\right)\\ z{0}_{\mathrm{fl}}\left(3\right)& {z1}_{\mathrm{fl}}\left(13\right)& {z2}_{\mathrm{fl}}\left(23\right)& ...& {z12}_{\mathrm{fl}}\left(123\right)\\ ...& ...& ...& ...& ...\\ {z0}_{\mathrm{fl}}\left(480\right)& {z1}_{\mathrm{fl}}\left(490\right)& {z2}_{\mathrm{fl}}\left(500\right)& ...& {z12}_{\mathrm{fl}}\left(600\right)\end{array}\right],$

And extract carbon monoxide utilization factor sample sequence X7=[x7 (1) x7 (2) x7 (3) ... x7 (480)] ^t;

2. by grey relevant degree method, find out the column vector in the Z with X7 degree of association maximum, to determine the retardation time of corresponding factor, concrete steps are as follows:

A. take X7 sequence as characteristic sequence, in Z matrix, each column vector is comparative sequences;

B. characteristic sequence and each comparative sequences are done to dimensionless processing, by first data of same sequence, remove all data below, obtain each data with respect to the multiple of first data, make it to be converted to the close dimensionless number certificate of order of magnitude cardinal principle, the characteristic sequence Q7 obtaining after processing and each comparative sequences Pi (i=1,2 ..., 12) as follows:

$Q7=\frac{X7}{x7\left(1\right)}=[q_7\left(1\right),q_7\left(2\right),...,q_7\left(k\right)],k=\mathrm{1,2},...,480;$ Q wherein ₇(1), q ₇(2) ..., q ₇(k) be each element in Q7; $\mathrm{Pi}=\frac{\mathrm{Zi}}{\mathrm{zi}(1+i*10)}=[p_i\left(1\right),p_i\left(2\right),...,p_i\left(k\right)],k=\mathrm{1,2},...,480;$ P wherein _i(1), p _i(2) ..., p _i(k) be each element in Pi.

[Zi (1+i*10) (i=1,2 ..., 12), represent the first row data in Z matrix, i.e. first data in each comparative sequences, i gets and finally equals 12, and z12 (121) is first data in the data rows lagging behind 120 minutes.】

C. calculate Q7 and Pi (i=1,2 ..., 12) correlation coefficient: first calculate the correlation coefficient ζ i (k) of each point,

$\mathrm{\&zeta;i}\left(k\right)=\frac{\underset{i}{\min }\underset{k}{\min }|q_7\left(k\right)-p_i\left(k\right)|+\mathrm{\&xi;}\underset{i}{\max }\underset{k}{\max }|q_7\left(k\right)-p_i\left(k\right)|}{|q_7\left(k\right)-p_i\left(k\right)|+\mathrm{\&xi;}\underset{i}{\max }\underset{k}{\max }|q_7\left(k\right)-p_i\left(k\right)|}$

In formula, ξ is resolution ratio, ξ=0.5; The degree of association of comprehensive each point, obtains Q7 and Pi (i=1,2 again ..., 12) degree of association γ (i),

$\mathrm{\&gamma;}\left(i\right)=\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&zeta;}}_i\left(k\right),n=480$

D. compare γ (i) (i=1,2 ..., 12) size, suppose γ (t) maximum, in matrix Z, the degree of association of t column data Zt and X7 is maximum; Thereby be t*10min the retardation time of determining wind-warm syndrome operation; Zt is as the sample data of this new operating parameter;

E. the sample data of each operating parameter is combined, finally determines sample set S=[x1 x2 x3 x4 x5 x6 x7], deposit in Computer Database;

In step 2, the Slack time 90min of air quantity, the Slack time 70min of wind-warm syndrome, the Slack time 20min of wind speed, the Slack time 40min of blast, the Slack time 60min that press on top, the Slack time of coal powder injection speed is 70min; In step 3, the final parameter value of c and p is respectively 2 and 1.867.

Beneficial effect:

The modeling method of blast furnace ironmaking process carbon monoxide utilization factor of the present invention, comprises following step: step 1: according to blast furnace ironmaking Principle of Process, by the analysis to blast furnace internal feature, choose the operating parameter that carbon monoxide utilization factor is had to considerable influence; Step 2: image data in the on-the-spot local data base of blast furnace, and after processing by analysis, set up matching database; Step 3: taking into account under the prerequisite of high precision and real-time, utilizing the database in step 2, setting up the blast furnace carbon monoxide usage forecast model based on SVM, realizing the real-time accurately predicting of carbon monoxide utilization factor.The relation that the present invention is clear and definite between blast furnace carbon monoxide utilization factor and operating parameter, realize the real-time accurately predicting of blast furnace carbon monoxide utilization factor, solved in the past the problem because blast furnace operating exists hysteresis quality to cause carbon monoxide utilization factor and operating parameter relation is not clear and cannot real-time estimate.

At present, also there is not carbon monoxide utilization factor model accurately.Blast furnace ironmaking production scene can obtain current carbon monoxide utilization factor value by detecting analysis of blast furnace gas ingredient.Relation between carbon monoxide utilization factor size and blast furnace operating is indefinite; And operation exists serious hysteresis to the impact of blast furnace, making the blast furnace carbon monoxide utilization factor that " in real time " detects is not the result of current time operation.Be the carbon monoxide utilization factor value of current detection, can only react a period of time operational circumstances before, and the change situation of carbon monoxide utilization factor afterwards determined in current operation in fact.

At present, blast furnace production scene can only, according to knowhow, qualitatively judge the relation between carbon monoxide utilization factor and blast furnace operating; Also can only obtain former and current carbon monoxide utilization factor detected value, and by virtue of experience, the following possible variation tendency of judgement carbon monoxide utilization factor.

The innovative point of maximum of the present invention is: proposed a kind of blast furnace carbon monoxide usage forecast method, realized the real-time estimate of the blast furnace carbon monoxide utilization factor based on operating parameter.

The present invention can set up the corresponding relation model between blast furnace carbon monoxide utilization factor and blast furnace operating parameter, and its beneficial effect is:

1, on the basis in conjunction with blast furnace production process mechanism, choose air quantity, wind-warm syndrome, wind speed, blast, coal powder injection speed, 6 parameters of top pressure for inputting, thereby select and less independent variable with rational parameter, simplify the complicacy of model structure, improved the forecast accuracy of blast furnace carbon monoxide utilization factor.

2, utilize Wavelet Denoising Method, reduced interference and the noise of industry spot, make utilized detection data more reliable, make to utilize the model of data-driven foundation more accurate.

3, taking into full account each operating parameter in the situation that of retardation time, setting up the quantitative corresponding relation model between blast furnace ironmaking process carbon monoxide utilization factor and main operating parameters, for the adjusting of execute-in-place later provides strong foundation.

4, utilize institute's established model can carry out the blast furnace carbon monoxide usage forecast based on true-time operation.

5, taken into full account the retardation time of each operation, made model more reasonable and accurate.

6, utilize support vector machine, reduced the operation time of modeling, and utilize cross validation to choose the optimized parameter of support vector machine, improved precision of prediction.

Accompanying drawing explanation

Fig. 1 is forecast model structural drawing;

Fig. 2 cross validation optimization of parameter choice procedure chart;

Fig. 3 is the comparison diagram of an oxidation usage forecast result and actual value;

Fig. 4 is an oxidation usage forecast relative error result figure;

Fig. 5 is process flow diagram of the present invention.

Embodiment

Below with reference to the drawings and specific embodiments, the present invention is described in further details:

Embodiment 1:

A kind of blast furnace ironmaking process carbon monoxide usage forecast method, by wind-warm syndrome, blast, wind speed, air quantity, coal powder injection speed, top presses data and carbon monoxide utilization factor to carry out time lag registration, utilize support vector machine to set up forecast model, and by the parameter of cross validation Support Vector Machines Optimized, to improve model accuracy.Concrete steps are as follows:

1) image data sample in computer server at the scene.

Take 1 minute as the sampling time, in the local oracle database on the industrial computer of blast furnace operating chamber, collect the wind-warm syndrome of first 10 hours, blast, wind speed, air quantity, coal powder injection speed, press on top, the historical data of carbon monoxide and carbon dioxide content in coal gas,, forms initial sample set by totally 600 groups.Sample is represented to have by data rows sequence:

Wind-warm syndrome (℃) be: X1=[x1 (1) x1 (2) ... x1 (600)] ';

Wind speed (m/s) is: X2=[x2 (1) x2 (2) ... x2 (600)] ';

Air quantity (Nm ³/ min) be: X3=[x3 (1) x3 (2) ... x3 (600)] ';

Blast (kPa) is: X4=[x4 (1) x4 (2) ... x4 (600)] ';

Coal powder injection speed (t/h) is: X5=[x5 (1) x5 (2) ... x5 (600)] ';

Top presses (kPa) to be: X6=[x6 (1) x6 (2) ... x6 (600)] ';

In blast furnace gas, carbon monoxide content number percent (%) is: X71=[x71 (1) x71 (2) ... x71 (600)] ';

In coal gas, carbon dioxide volume content number percent (%) is: X72=[x72 (1) x72 (2) ... X72 (600)] '.

Computing formula by blast furnace carbon monoxide utilization factor:

${\mathrm{\&eta;}}_{\mathrm{CO}}=\frac{\left({\mathrm{CO}}_2\right)}{\left({\mathrm{CO}}_2\right)+\left(\mathrm{CO}\right)}$

Wherein,

η _cO: blast furnace carbon monoxide utilization factor;

(CO ₂): the volume content number percent of carbon dioxide in blast furnace gas;

(CO): the volume content number percent of carbon monoxide in blast furnace gas,

Calculate the sample of blast furnace carbon monoxide utilization factor: X7=[x7 (1) x7 (2) ... x7 (600)] ^t;

2) sample data being carried out to abnormal data kicks out of and replaces.

Operation change due to industry spot, such as changing-over stove, or the detection of pick-up unit error, can make the abnormal data that in data sequence, appearance is too fluctuateed, when carrying out data analysis modeling, these data can affect normal training process, so these data need to be rejected and replaced.

To air quantity, wind-warm syndrome, blast, wind speed, coal powder injection speed, top pressure, carbon monoxide utilization factor, in 7 data sequences, maximum 3 data and 3 minimum data, replace.Replace with a previous moment and rear mean value of data constantly.To reduce as much as possible the interference of spike data.

3) sample data is carried out to Wavelet Denoising Method processing.

Industrial environment is complicated, noise is many, detects data and unavoidably can be subject to the interference of uncertain noise, and this can affect training and the precision of model, therefore need to carry out denoising to sample data.

Utilize Wavelet noise-eliminating method, 7 sample sequences are carried out respectively to denoising.With wind-warm syndrome x1 (k), k=(1,2 ..., 600) be example, concrete steps are as follows:

1. choosing demy is wavelet basis function, and x1 is carried out to 5 layers of wavelet decomposition;

2. according to the soft value of cutting off from method, select the threshold value of each layer.Two layer signals of highest frequency are shielded completely, retain low frequency signal;

3. the small echo signal after processing is reconstructed, covers original x1 (k), complete the Wavelet Denoising Method of data sample.

4) field data is carried out to time lag registration.

In blast furnace production process, operating parameter changes the impact of blast furnace state and carbon monoxide utilization factor is existed to serious hysteresis, current operation, and blast furnace was wanted time of one and just can be showed response.Therefore, in to the modeling of blast furnace carbon monoxide utilization factor, need to take into full account the different retardation times of each different operating parameter, carry out time lag registration.The design adopts grey relational grade analysis method, finds out operating parameter sequence retardation time with carbon monoxide utilization factor correlativity maximum, determines the retardation time of each operating parameter.Take air quantity data sample as example, and concrete steps are as follows:

1. take 10 minutes as interval, extract the data sample of hysteresis different time, be expressed as Z _fl, Z _fw, Z _fy, Z _fs, Z _ps, Z _dy, with air quantity under for example,

Lag behind 0 minute: z0 _fl(k)=x1 (k), k=(1,2 ..., 480);

Lag behind 10 minutes: z1 _fl(k)=x1 (k), k=(11,12 ..., 490);

Lag behind 20 minutes: z2 _fl(k)=x1 (k), k=(21,22 ..., 500);

……

Lag behind 120 minutes: z1 _fl2 (k)=x1 (k), k=(121,122 ..., 600);

Form matrix $Z_{\mathrm{fl}}=\left[\begin{array}{ccccc}{\mathrm{<figure-callout\; id="0"\; label="z"\; filenames="HDA0000461150970000012.png"\; state="\{\{state\}\}">z</figure-callout>}0}_{\mathrm{fl}}\left(1\right)& {\mathrm{<figure-callout\; id="1"\; label="z"\; filenames="HDA0000461150970000012.png"\; state="\{\{state\}\}">z</figure-callout>}1}_{\mathrm{fl}}\left(11\right)& {\mathrm{<figure-callout\; id="2"\; label="z"\; filenames="HDA0000461150970000012.png,HDA0000461150970000022.png"\; state="\{\{state\}\}">z</figure-callout>}2}_{\mathrm{fl}}\left(21\right)& ...& {z12}_{\mathrm{fl}}\left(121\right)\\ {\mathrm{<figure-callout\; id="0"\; label="z"\; filenames="HDA0000461150970000012.png"\; state="\{\{state\}\}">z</figure-callout>}0}_{\mathrm{fl}}\left(2\right)& {\mathrm{<figure-callout\; id="1"\; label="z"\; filenames="HDA0000461150970000012.png"\; state="\{\{state\}\}">z</figure-callout>}1}_{\mathrm{fl}}\left(12\right)& {\mathrm{<figure-callout\; id="2"\; label="z"\; filenames="HDA0000461150970000012.png,HDA0000461150970000022.png"\; state="\{\{state\}\}">z</figure-callout>}2}_{\mathrm{fl}}\left(22\right)& ...& {z12}_{\mathrm{fl}}\left(122\right)\\ \mathrm{<figure-callout\; id="0"\; label="z"\; filenames="HDA0000461150970000012.png"\; state="\{\{state\}\}">z</figure-callout>}{0}_{\mathrm{fl}}\left(3\right)& {\mathrm{<figure-callout\; id="1"\; label="z"\; filenames="HDA0000461150970000012.png"\; state="\{\{state\}\}">z</figure-callout>}1}_{\mathrm{fl}}\left(13\right)& {\mathrm{<figure-callout\; id="2"\; label="z"\; filenames="HDA0000461150970000012.png,HDA0000461150970000022.png"\; state="\{\{state\}\}">z</figure-callout>}2}_{\mathrm{fl}}\left(23\right)& ...& {z12}_{\mathrm{fl}}\left(123\right)\\ ...& ...& ...& ...& ...\\ {\mathrm{<figure-callout\; id="0"\; label="z"\; filenames="HDA0000461150970000012.png"\; state="\{\{state\}\}">z</figure-callout>}0}_{\mathrm{fl}}\left(480\right)& {\mathrm{<figure-callout\; id="1"\; label="z"\; filenames="HDA0000461150970000012.png"\; state="\{\{state\}\}">z</figure-callout>}1}_{\mathrm{fl}}\left(490\right)& {\mathrm{<figure-callout\; id="2"\; label="z"\; filenames="HDA0000461150970000012.png,HDA0000461150970000022.png"\; state="\{\{state\}\}">z</figure-callout>}2}_{\mathrm{fl}}\left(500\right)& ...& {z12}_{\mathrm{fl}}\left(600\right)\end{array}\right],$

And extract carbon monoxide utilization factor sample sequence X7=[x7 (1) x7 (2) x7 (3) ... x7 (480)] ^t.

2. by grey relevant degree method, find out the column vector in the Z with X7 degree of association maximum, to determine the retardation time of corresponding factor, concrete steps are as follows:

A be take X7 sequence as characteristic sequence, Z _flin matrix, each column vector is comparative sequences.

B does dimensionless processing to characteristic sequence and each comparative sequences, by first data of same ordered series of numbers, remove all data below, obtain each data with respect to the multiple of first data, make it to be converted to the close dimensionless number certificate of order of magnitude cardinal principle, the characteristic sequence Q7 obtaining after processing and each comparative sequences Pi (i=1,2 ..., 12) as follows:

$Q7=\frac{X7}{x7\left(1\right)}=[q_7\left(1\right),q_7\left(2\right),...,q_7\left(k\right)],k=(\mathrm{1,2},...,480);$

$\mathrm{Pi}=\frac{{\mathrm{Zi}}_{\mathrm{fl}}}{\mathrm{zi}{(1+i*10)}_{\mathrm{fl}}}=[p_i\left(1\right),p_i\left(2\right),...,p_i\left(k\right)],k=(\mathrm{1,2},...,480);$

C calculates Q7 and Pi (i=1,2 ..., 12) correlation coefficient.First calculate the correlation coefficient ζ i (k) of each point,

$\mathrm{\&zeta;i}\left(k\right)=\frac{\underset{i}{\min }\underset{k}{\min }|q_7\left(k\right)-p_i\left(k\right)|+\mathrm{\&xi;}\underset{i}{\max }\underset{k}{\max }|q_7\left(k\right)-p_i\left(k\right)|}{|q_7\left(k\right)-p_i\left(k\right)|+\mathrm{\&xi;}\underset{i}{\max }\underset{k}{\max }|q_7\left(k\right)-p_i\left(k\right)|}$

In formula, ξ is resolution ratio, and ξ ∈ [0,1], gets ξ=0.5 here.The degree of association of comprehensive each point, obtains Q7 and Pi (i=1,2 again ..., 12) degree of association γ (i),

$\mathrm{\&gamma;}\left(i\right)=\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&zeta;}}_i\left(k\right),n=480.$

D is γ (i) (i=1,2 relatively ..., 12) size, suppose γ (t) maximum, in matrix Z, the degree of association of t column data Zt and X7 is maximum.Thereby be (t*10) min the retardation time of determining wind-warm syndrome operation.Zt is as new wind-warm syndrome sample data.

Other operating parameters of e are determined retardation time by same steps, finally determine sample set S=[x1 x2 x3 x4 x5 x6 x7], deposit in Computer Database.

In this example, the Slack time of each operating parameter determines that result is as shown in table 1.As seen from table, the Slack time 90min of air quantity, the Slack time 70min of wind-warm syndrome, the Slack time 20min of wind speed, the Slack time 40min of blast, the Slack time 60min that press on top, the Slack time of coal powder injection speed is 70min.

5) sample data is normalized.Each column data of matrix S is normalized by following mapping:

$y=(y_{\max }-y_{\min })\&times;\frac{x-x_{\min }}{x_{\max }-x_{\min }}+y_{\min }$

Wherein, x is the input value of a column data in matrix, and y is the output valve through normalized, y _maxand y _minthe maximal value and the minimum value that are respectively normalization scope, be made as [1,1].X _maxand x _maxrespectively maximal value and the minimum value in matrix one column data.Sample matrix after normalization is S '.

6) parameter c of cross validation method Support Vector Machines Optimized and g.Concrete steps are as follows:

1. determine the span of parameter c and g.This example is chosen cmin=-4, cmax=4, gmin=-4, gmax=4.Step size cstep=0.1, gstep=0.1.

2. determine test set grouping number V.This example is defined as 3.

3. test set is carried out to cross validation.That is, when c and g change from small to large with step-length, take each group test set 1 is training set, and rear one group of test set, for checking collection, calculates the Average Accuracy of checking.

C when 4. retaining Average Accuracy minimum and g value, the most best parameter value.Optimization of parameter choice process as shown in Figure 2.In this example, optimal parameter c is 2, and optimal parameter g is 1.867.

7) utilize optimal parameter c and g, sample data is carried out to support vector machine training, to obtain carbon monoxide utilization factor model.In example, using that front 200 groups of data are as training set in sample, rear 49 groups of data are for testing model precision of prediction.The Selection of kernel function of support vector machine is RBF function: exp (r|u-v|^2).

Predict the outcome and the contrast of actual result as shown in Figure 3, (owing to being output as ratio, so without unit), Relative Error as shown in Figure 4.As seen from the figure, relative error, within 0.2%, can meet the requirement of industry spot.

Each parameter Unequal time lag serial correlation degree table of table 1

Claims (4)

1. a modeling method for blast furnace ironmaking process carbon monoxide utilization factor, is characterized in that, comprises the following steps:

Step 1: data acquisition and calculating;

Acquisition operations supplemental characteristic in local data base from blast furnace industrial computer; Described operating parameter comprises: CO and CO in wind-warm syndrome, blast, wind speed, air quantity, coal powder injection speed, top pressure, blast furnace gas ₂percent by volume, and calculate carbon monoxide utilization factor η by following formula _cO:

${\mathrm{\&eta;}}_{\mathrm{CO}}=\frac{\left({\mathrm{CO}}_2\right)}{\left({\mathrm{CO}}_2\right)+\left(\mathrm{CO}\right)}$

Wherein, (CO ₂) be CO in blast furnace gas ₂percent by volume; (CO) be the percent by volume of CO in blast furnace gas;

Step 2: data are carried out to time lag registration;

After the data that gather are carried out to data pre-service, utilize grey relational grade analysis method, respectively the operating parameter time series of Unequal time lag degree and carbon monoxide utilization factor time series are carried out to correlation analysis, determine respectively the Slack time of each operating parameter, complete the time lag registration of data, and form sample set;

Step 3: the foundation of model;

Sample set based on described in step 2, usings air quantity, wind-warm syndrome, wind speed, blast, coal powder injection speed, top six supplemental characteristics of pressure as input, and carbon monoxide utilization factor is output, sets up the blast furnace carbon monoxide usage forecast model based on SVM.Described SVM adopts RBF kernel function, and adopts cross-validation method to be in optimized selection the penalty parameter c of SVM and kernel functional parameter g; After modelling verification completes, the input parameter data of current time lag registration of take are input, by model, obtain the carbon monoxide utilization factor under current operation parameter;

The step of described cross-validation method is as follows:

1. determine the span of penalty parameter c and kernel functional parameter g;

2. determine test set grouping number V;

3. test set is carried out to cross validation:, when c and g change from small to large with step-length, take each group test set is training set, and rear one group of test set, for checking collection, calculates the Average Accuracy of checking;

Corresponding c and g value when 4. retaining Average Accuracy maximum, as final parameter value.

2. blast furnace ironmaking process carbon monoxide utilization factor modeling method according to claim 1, it is characterized in that, in step 2, the data that gather are carried out to data pre-service, data pre-service comprises rejecting abnormal data, and utilize wavelet-decomposing method denoising, to eliminate the interference noise of industry spot detection.

The method of rejecting abnormal data is: in each data sequence of air quantity, wind-warm syndrome, blast, wind speed, coal powder injection speed, top pressure, these 7 data sequences of carbon monoxide utilization factor, maximum 3 data and 3 minimum data are replaced, replace with a previous moment and rear mean value of data constantly, to reduce the interference of spike data;

Denoising method is: each data sequence of 7 data sequences is done to following operation:

1. choosing demy is wavelet basis function, and this data sequence is carried out to 5 layers of wavelet decomposition;

2. according to soft-threshold method, select the threshold value of each layer, two layer signals of highest frequency are shielded completely, retain low frequency signal;

3. low frequency signal is reconstructed, covers original data sequence, complete the Wavelet Denoising Method of data sample.

3. the modeling method of blast furnace ironmaking process carbon monoxide utilization factor according to claim 2, is characterized in that, in step 2, determines that the process of certain operating parameter Slack time is:

1. for current operating parameter, take 10 minutes as interval, extract air quantity, wind-warm syndrome, blast, wind speed, coal powder injection speed, the top of hysteresis different time and press data sample, be expressed as Z _fl, Z _fw, Z _fy, Z _fs, Z _ps, Z _dy, with air quantity under for example,

Lag behind 0 minute: z0 _fl(k)=x1 (k), k=1,2 ..., 480)

Lag behind 10 minutes: z1 _fl(k)=x1 (k), k=11,12 ..., 490;

Lag behind 20 minutes: z2 _fl(k)=x1 (k), k=21,22 ..., 500;

……

Lag behind 120 minutes: z12 _fl(k)=x1 (k), k=121,122 ..., 600;

Form matrix $Z_{\mathrm{fl}}=\left[\begin{array}{ccccc}{z0}_{\mathrm{fl}}\left(1\right)& {z1}_{\mathrm{fl}}\left(11\right)& {z2}_{\mathrm{fl}}\left(21\right)& ...& {z12}_{\mathrm{fl}}\left(121\right)\\ {z0}_{\mathrm{fl}}\left(2\right)& {z1}_{\mathrm{fl}}\left(12\right)& {z2}_{\mathrm{fl}}\left(22\right)& ...& {z12}_{\mathrm{fl}}\left(122\right)\\ z{0}_{\mathrm{fl}}\left(3\right)& {z1}_{\mathrm{fl}}\left(13\right)& {z2}_{\mathrm{fl}}\left(23\right)& ...& {z12}_{\mathrm{fl}}\left(123\right)\\ ...& ...& ...& ...& ...\\ {z0}_{\mathrm{fl}}\left(480\right)& {z1}_{\mathrm{fl}}\left(490\right)& {z2}_{\mathrm{fl}}\left(500\right)& ...& {z12}_{\mathrm{fl}}\left(600\right)\end{array}\right],$

And extract carbon monoxide utilization factor sample sequence X7=[x7 (1) x7 (2) x7 (3) ... x7 (480)] ^t;

2. by grey relevant degree method, find out the column vector in the Z with X7 degree of association maximum, to determine the retardation time of corresponding factor, concrete steps are as follows:

A. take X7 sequence as characteristic sequence, in z matrix, each column vector is comparative sequences;

B. characteristic sequence and each comparative sequences are done to dimensionless processing, by first data of same sequence, remove all data below, obtain each data with respect to the multiple of first data, make it to be converted to the close dimensionless number certificate of order of magnitude cardinal principle, the characteristic sequence Q7 obtaining after processing and each comparative sequences Pi (i=1,2 ..., 12) as follows:

$Q7=\frac{X7}{x7\left(1\right)}=[q_7\left(1\right),q_7\left(2\right),...,q_7\left(k\right)],k=(\mathrm{1,2},...,480);$ Q wherein ₇(1), q ₇(2) ..., q ₇(k) be each element in Q7;

$\mathrm{Pi}=\frac{\mathrm{Zi}}{\mathrm{zi}(1+i*10)}=[p_i\left(1\right),p_i\left(2\right),...,p_i\left(k\right)],k=(\mathrm{1,2},...,480);$ P wherein _i(1), p _i(2) ..., p _i(k) be each element in Pi;

C. calculate Q7 and Pi (i=1,2 ..., 12) correlation coefficient.First calculate the correlation coefficient ζ i (k) of each point,

$\mathrm{\&zeta;i}\left(k\right)=\frac{\underset{i}{\min }\underset{k}{\min }|q_7\left(k\right)-p_i\left(k\right)|+\mathrm{\&xi;}\underset{i}{\max }\underset{k}{\max }|q_7\left(k\right)-p_i\left(k\right)|}{|q_7\left(k\right)-p_i\left(k\right)|+\mathrm{\&xi;}\underset{i}{\max }\underset{k}{\max }|q_7\left(k\right)-p_i\left(k\right)|}$

In formula, ξ=0.5 is resolution ratio; The degree of association of comprehensive each point, obtains Q7 and Pi (i=1,2 again ..., 12) degree of association γ (i),

$\mathrm{\&gamma;}\left(i\right)=\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&zeta;}}_i\left(k\right),n=480;$

D. compare γ (i) (i=1,2 ..., 12) size, suppose γ (t) maximum, in matrix Z, the degree of association of t column data Zt and X7 is maximum; Thereby be t*10min the retardation time of determining wind-warm syndrome operation; Zt is as the sample data of this new operating parameter; In addition, the sample data of each operating parameter is combined, finally determine sample set S=[x1 x2 x3 x4 x5 x6 x7].

4. the modeling method of blast furnace ironmaking process carbon monoxide utilization factor according to claim 3, it is characterized in that, in step 2, the Slack time 90min of air quantity, the Slack time 70min of wind-warm syndrome, the Slack time 20min of wind speed, the Slack time 40min of blast, the Slack time 60min that press on top, the Slack time of coal powder injection speed is 70min; In step 3, the final parameter value of c and p is respectively 2 and 1.867.

Images