JPS63210214A - Prediction of furnace heat in blast furnace - Google Patents

Abstract

PURPOSE:To accurately predict lowering of furnace heat without any delaying the prediction, by outputting the predicting value by ARMA model adopted as a model predicting means at the time of executing alarm to the lowering of furnace heat in an alarming means. CONSTITUTION:The alarming means for the lowering of the furnace heat, alarming at the time of reaching to the prescribed condition by using operational result having high correlation and fore-sightedness to the lowering of the furnace heat in a blast furnace and the model predicting means, predicting the variation with time of the furnace heat by the ARMA model using molten iron temp. tapping from the blast furnace, are arranged. And, at the time of alarming the lowering of the furnace heat by the alarming means, the predicting value of the furnace heat by the model predicting means is outputted. By this method, the accurate furnace heat without any delay can be predicted.

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial Application Field) The present invention relates to a blast furnace furnace heat prediction method for stable operation of a blast furnace.

(Prior art and its problems) It has been known for a long time that in order to maintain stable operation of a blast furnace, it is necessary to keep the temperature of hot metal constant. For this reason, blast furnace operators have always needed to predict changes in blast furnace heat.

Prediction of temperature drop is extremely important in blast furnace furnace thermal changes, especially since hot metal may solidify due to temperature drop and may no longer flow out of the blast furnace.

A method for predicting blast furnace heat is to consider the blast furnace as a black box and obtain a model through statistical analysis.
There is one based on the AR (Auto-Reoressive) model.

Although the AR model described above has excellent characteristics as a model for predicting temporal fluctuations in the actual furnace heat level of a blast furnace, it has the following problems.

FIG. 20 is a diagram showing changes over time in actual values and predicted values of hot metal temperature. In the figure, 11 is a curve based on the actual value x and 12 is a curve based on the predicted value. As shown in the figure, the change in the predicted value tends to lag behind the change in the actual value by about τ sampling time (that is, τ×Δ is time, where the sampling interval is Δt), and its amplitude tends to be small.

For this reason, there was a problem that the prediction function could not be fully fulfilled, and the predicted value was not accurate enough. To solve this problem, ARM A (Auto-
Reoressive Hovina Averaae
) (autoregressive moving average) model is published in JP-A-60-248804, Tetsu-to-Hagane J 1984, No. 1.

Although it is disclosed on page 54, etc., the current situation is that no definitive variable to be introduced into the MA term has yet been found.

(Objective of the Invention) The object of the present invention is to provide a blast furnace furnace heat prediction method that eliminates the problems of the above-mentioned prior art and can accurately predict hot metal temperature changes without delay in prediction. .

(Means for Achieving the Object) In order to achieve the above object, the blast furnace furnace heat prediction method according to the present invention uses the temperature of hot metal tapped from the blast furnace to calculate temporal fluctuations in the blast furnace furnace heat level using an ARMA model. and a furnace heat drop warning means that warns of a furnace heat drop when the operation results meet a predetermined condition, using operation results that have a strong correlation and foresight with respect to the blast furnace furnace heat drop. In addition, when the furnace heat decrease warning means warns of a decrease in furnace heat, the predicted value of the blast furnace furnace heat by the model prediction means is output.

(Example) It has been known that an increase or decrease in the amount of solution loss carbon (hereinafter referred to as "Tsuru loss CIJ"), which indicates the quality of the reduction state of a blast furnace, has a strong correlation with the heat of the blast furnace. In other words, an increase in the amount of tsuru loss C indicates that the so-called tsuru loss reaction shown below is promoted. The fever will drop.

Therefore, it can be said that the increase or decrease in the amount of truss loss C can be predicted with respect to the blast furnace furnace heat.

Therefore, AR that introduces the average value of Truloss CI into the MA term
A first prediction method using an MA model is considered.

FIGS. 1(a) and 1(b) respectively show the temporal changes in the instantaneous value of the hot metal melt and the average value of the pot loss Cff1. Same figure (
In a), Ll is the actual value and L2 is the predicted value. The instantaneous value of truss loss CM (Ng/1-D) is calculated based on the proportions of Co, CO2, N2, etc. in the top gas components, feeding conditions, raw material charging conditions, etc., and is shown in Figure 1 (b). Tsururos Cf
The average value u(t) of f1 is obtained by calculating the instantaneous value of the amount of truss C at each sampling time of gas chromatography, for example, and averaging this at each time width Δt1. On the other hand, the instantaneous value y(t) of the hot metal temperature is determined at every sampling time Δt2. Note that t is the current time.

Using such variables, the predicted hot metal temperature value V1 (t:+Δt2) at one point in time can be estimated using the following equation using the ARMA model.

Vl (to 10 t2) ...(1) The first term on the right side is the AR term, and the second term is the MA term.

Note that ab is a coefficient.

1, p Ii FIG. 2 is a flowchart showing the processing procedure of the first model prediction means based on equation (1). First, in step S1, the average value u (
At t-, N pieces of Δt1) (k=1 to N) are extracted. Note that this process is performed every sampling time Δt2. Next, in step S2, the actual value of the hot metal temperature i at the current time t y(
t), by comparing this with the prediction result V1 (t), the coefficients a, j, and i of equation (1) are determined.

Then, in step S3, the predicted value V (t+Δt2) of the hot metal temperature of the light at time 11j.11 is determined from equation (1) using the coefficient ab determined as described above.
In the next step S4, this predicted value y1 (t+Δt) is compared with the control temperature 11, and if 2CLIC]1 t ≧V (t+Δt2)≧tcL, an abnormal H
1 It is assumed that there is no result, and the process returns to step S1 to continue prediction.

On the other hand, 'S/ (t+Δt2) tC4 or y1
(10Δt)>tCH, it is assumed that a decrease or increase in the furnace heat will occur sooner or later, and a predicted value V (10Δ12) is output in step S5 to warn of a decrease or increase in the furnace heat. . Thereafter, the process returns to step S1 and steps 81 to S5 are repeated to predict a decrease in furnace heat.

I-8, second model prediction means ARM that also introduces the maximum value of the amount of truss loss C into the MA term
Model A is also possible. This is to introduce the maximum value A of the average value u (t- to Δt1) (k-1 to N) of the true loss C amount in FIG. 1 into the TMA term as x(t).

Using such variables, the predicted hot metal temperature value y2 (t+Δt2) at one point in time can be estimated using the following equation using the ARMA model. Note that a and b are coefficients.

i 2 y2 (1 Δt2) 1,) m a2. y(tiΔt2)+b2x(t)1
=6 (2) FIG. 3 is a flowchart 1 showing the processing procedure of the second model prediction means based on equation (2). First, in step S11, the average value u(
N pieces of Δtt > (k=1 to N) are calculated for t-.

Note that this process is performed every sampling time Δt2. Next, in step S12, the maximum value of the average value u (Δ11 in t-) obtained in step 811 is obtained, and this is calculated as x (
t).

X (t) −max (u (−Δt1))k=
1. ..., N ... (3) Then, in step 813, the actual value of the hot metal temperature y(t) at the current time t is determined.
Therefore, by comparing this with the past prediction result y2(t), the coefficients a21 and b2 of equation (2) are determined. Next, in step 814, the predicted temperature of hot metal V2 at one point in time is
(+Δt2) is the coefficient a21. It is obtained from equation (2) using b2, and in the next step 815, this predicted value V (with 10Δt2) is compared with the control temperature tC, and if y2 (with 10Δt2)≧tc, it is considered that there is no abnormality. The process returns to step S11 to continue prediction. On the other hand, if V2 (t+Δt2)<t, it is assumed that a decrease in furnace heat will occur sooner or later, and a predicted value y2 (t+Δt2) is output in step 816 to warn of a decrease in furnace heat.

After that, the process returns to step S11 and steps 811 to 816
The furnace heat is predicted by repeating the steps.

I-C, Third Model Prediction Means By the way, normally in a blast furnace, iron is tapped alternately through two tap holes, so the change in hot metal temperature over time for each tap shown in Figure 4 at different tap holes is As shown, deviations often occur in the hot metal temperature, and in such cases, prediction accuracy is affected. Therefore, a third model prediction means can be considered that predicts the hot metal temperature by setting up an ARMA model in which a variable for correcting the temperature deviation between tap holes A and B is introduced as an MA term.

When tapping is performed every sampling time Δt3,
Predict the hot metal temperature using the following formula.

V3 (t+Δt3) =, Σa3HV(i tΔt3) +c3z (t
)...(4) Note that a and C are each coefficients, and z (t)i
3 is the taphole correction variable.

The taphole correction variable z (t) is determined as follows. Focusing on M hot metal temperatures that have been reversed in the past, the average value Y is determined, and the average value YA of M for taphole A and M for taphole B (M
-1-M,=M), find the average A value YB. Here, time t. If the tap hole in is A, then z (to)=YA-Y ・(5) If B, then Z (t, ) −y, −Y ・・
- Find the taphole correction variable z (to) from (6).

Furthermore, when tapping from the same taphole 8 times (S≧2) in succession, the taphole correction variable z (to) determined by equation (5) or equation (6) is changed according to the following equation.

z (t) − (1+Σ(k/i))z (tn
)...(1) Here, k is lkl<1 and if z (to)>O, then k>O. If z(t,)<O, then k<Q. This is done in consideration of the heat retention effect of continuously tapping from the same tap hole. In other words, the continuous tapping warms the tap flute, so the actual hot metal temperature value rises higher than normal. In consideration of this temperature rise, the taphole correction variable z (to) includes a temperature rise corresponding to (,Σ(lkl/1))l Z (to)l compared to normal. Furthermore, the control temperature, which is the standard for the hot metal temperature actual value, is changed in consideration of the temperature increase due to the continuous tapping described above.

FIG. 5 is a flowchart showing the processing procedure of the third model prediction means based on equation (4). First, step 8
In step 821, the taphole at the predicted time (t + Δt2) is determined, and if the taphole is A, then in step 822, the taphole correction variable z (is 10Δt2) at the predicted time (t + Δt2) is determined using equation (5). If the taphole is B, the taphole correction variable z is determined in step 823 using equation (6).
Find (t+Δt2). Subsequently, in step 824, it is checked whether the tap hole at the current time t and the predicted time (t+Δ12) are the same. Therefore, when tapping is performed continuously from the same taphole, the taphole correction coefficient z(t+Δt2) is changed in step S25 using equation (7) in consideration of the heat retention effect due to continuous tapping.

Next, in step S26, actual hot metal temperature value y(t>
The coefficients a and C of equation (4) are determined based on the comparison between the prediction result V3(t) and the prediction result V3(t). Furthermore, in step 3i 3 tube 827, the predicted value of the hot metal temperature y3 (
t+Δt2) is the coefficient a obtained in step S26.
31. Obtained from equation (4) using C3, step 828
Move to. In step 828, the control temperature t is compared with y3 (t+Δt2)<t.

If so, it is assumed that the furnace heat has decreased abnormally, and step 8 is performed.
At step 29, the predicted value y3 of the hot metal temperature (= Δt2) is output to warn of a decrease in furnace heat.

On the other hand, if 'Sl' (t+Δt2)≧tC, it is assumed that there is no abnormality, and the process returns to step 821 to continue predicting the furnace heat. It goes without saying that the control temperature tc described here takes into consideration the temperature rise during tapping due to the same continuous taphole mentioned above.

In this way, the taphole correction variable z(
The predicted value obtained by the ARMA model in which t) is introduced into the MA term is quite accurate.

ID, Fourth Model Prediction Means A fourth model based on the ARMA model, which introduces both the maximum value of the trundle loss C amount and the taphole correction variable described in I-8, I-C, into the MA term. Prediction methods are possible.

(2) If the current time t and the predicted time (t+Δt2) are within the same tapping time, I-B is similarly predicted using the formula below.

V (t+Δt2) -Σa-y(t-iΔt,,)+b4X (t)I 6゜...(8) ■If the current time t and the predicted time (t+Δt2) are not within the same tapping time, lower (9 ) Make predictions according to the formula.

V (t+Δt2) = -ja , y (t −iΔt2) i2,51 +b x(t)+c5z(t) ・(9) Note that a
4. , b4. a, b, c are coefficients, x (t) is the maximum value of the amount of truss loss C mentioned in I-B,
z(t) is the taphole correction variable described in I-C.

FIG. 6 is a flowchart showing the processing procedure of the fourth model prediction means in intermittent tapping (for example, one tapping takes about 2 hours, and about 30 minutes are left between each tapping). In the same figure, in step S31, a time period N from the current time t is
Going backwards to the past Δt1, calculate the average value u (Δt1 at t-) of the amount of trailing loss C for each section Δt1 (k=1...N). Next, in step 832, according to the following formula (3) (reprinted),
The maximum value x(
Find t).

x (t)=max (u (Δt1 to t-)) (k=
1 to N) ... (3) Then, in step 833, it is checked whether the current time t and the predicted time (t+Δt2) are within the same tapping time, and if they are within the same tapping time, in step S34, the hot metal temperature actual Based on the comparison between the value y(t), , and the prediction result V4(t), the coefficients a4i and b4 of equation (8) are determined. Next, in step 835, predicted value of hot metal temperature y4 (
Δt2) is obtained from equation (8) using the coefficients a41 and b4 obtained in step 834, and the process proceeds to step 844.

On the other hand, in step 333, the current time t and the predicted time (
t+Δt2) is not within the same tapping time, step 8
In step 36, we look back at the M hot metal temperatures in the past, and calculate the average value Y of the M hot metal temperatures, and the MA and MB (MA
+MB = M ) hot metal temperature average values YA and Y8 are used for sieving. Next, in step 837, the predicted time (+Δt2
), and if it is taphole A, then in step 838 the predicted time (1+Δt
Find the taphole correction variable z (t + Δt2) in 2),
If it is the taphole B, then in step 839, the taphole correction variable z(t+Δt2) is calculated using equation (6). Furthermore, in step 840, the current time t and the predicted time (t+Δt2
) is checked to see if the tap ports are the same.

Therefore, when tapping is performed continuously at the same taphole, in step 341, the taphole correction variable Z (t+Δt2) is changed using equation (7) in consideration of the heat retention effect due to continuous tapping. Next, in step 842, the hot metal temperature actual value y
(t) and the prediction result V4 (t), the coefficient a51 of equation (9). Determine b5°C5. Furthermore, in step 843, the predicted temperature of hot metal y4 (t
+Δt2) is the coefficient a51 obtained in step 842.
．． Obtained from equation (9) using b5°C5, step S4
Move on to 4.

The predicted hot metal temperature value y4 (t+Δt2) obtained in steps 834 to S35 or S36 to S4.3 is the control temperature t in step S44. compared to 'Sl'
If (t+Δ1.,)<1c, it is assumed that the furnace heat has decreased abnormally, and a predicted value of hot metal temperature y (t+Δt2) is output in step 845 to warn of the decrease in furnace heat. On the other hand, if V4 (t+Δt2)≧tC, it is assumed that there is no abnormality, and the process returns to step 831 to continue predicting the furnace heat. Note that, like the third model prediction means, the control temperature tC described here also takes into account the temperature rise during tapping due to the same continuous taphole.

Using the first to fourth model prediction means described in I-E and 1-A and ~ID using the optimal ARMA model, error evaluation J (j
The first model prediction means with the smallest value (J!=1 to 4) is adopted as the comprehensive model prediction means.

First, in the first to fourth model prediction means, the predicted hot metal temperature y − which has increased backward from the current time t to the past LΔt2
Find the average values y1 to 4 and standard width differences σ1 to σ4 of y4, and calculate the average value y and standard deviation σ of the actual hot metal temperature value y.
Then, according to the following equation (10), the average error ε(j (j!=1 to 4) is determined.

−V(t−iΔt2))l (10) Furthermore, the prediction error J1 is calculated by equations (11) and (12).
Find (It>(l=1-4).

−tsu 〇 − εIllax=max (e (jri) =
・(11) Jl(i)-ε(1)/εl1lax
(12) Then, according to equations (13) to (16), the degree of correlation between the predicted values v1 to y4 of the first to fourth model prediction means and the actual measured values X (V-1/ (t-iΔt) ))・
...(13)φ, =SR/(σ・C1) ...
(14)...(15) φlax=max(φl) J (1)-φ,/φmax...(16
) (10) to (12) Prediction error J1 (A)
and the correlation degree J (j), the error evaluation J (J) is sieved out using the following formula.

J (jri=J (1-J2(jri...(17
) (11) Any one of the first to fourth model prediction means with the minimum J (J) (the highest prediction accuracy) is adopted as the comprehensive model prediction means to predict the furnace heat.

In this way, the prediction tU with the highest prediction accuracy (the lowest error evaluation J(1)) of the first to fourth model prediction means
- By always automatically employing the sub-means as a comprehensive model prediction means, it is possible to always make predictions using the optimal ARMA model.

If the comprehensive model prediction method described in 2. is used, considerable prediction accuracy can be expected. By the way, the most important information for blast furnace operators is an early warning of a decrease in furnace heat and the degree of decrease in furnace heat at that time.

For the latter, the above-mentioned comprehensive model prediction means is sufficient, but for the former, the ARMA model that outputs predicted values every predetermined sampling time Δt2 is used, so there is room for improvement as described below.

■If the sampling time Δt2 is long enough, for example, the first
．． Third. If the time t when the amount of true loss C introduced into the MA term of the fourth model prediction means suddenly increases is immediately after the predicted value output time t (tc=p t +Δε, Q<Δε 〈<Δt2), then p
It is only at the next predicted value output time (1, 10Δt2) that it becomes possible to make a prediction that takes into account the rapid increase in the amount of trailing loss C. Therefore, in the worst case, the predicted value output time is delayed by a time Δ↑2. It is possible to put it away.

■If the sampling time Δt2 is sufficiently short, the delay in predicted value output time as in ■ will be eliminated, but there is a high possibility that the data obtained at each short sampling time Δt2 will contain a considerable error component. , there is a risk that the accuracy of the predicted value itself will decrease.

In order to improve the above-mentioned points (1) and (2), it is conceivable to provide a separate means for early warning of a decrease in furnace heat.

The following can be considered as operational results that have a strong correlation with the decrease in furnace heat of the furnace 11 of I-A, 1, and the reduction blast furnace heat and have a foresight.

High-temperature air (gas flow) blown up from the blast furnace lining to adjust the temperature and amount of hot metal is usually blown into the center of the furnace. However, due to reasons such as raw material charging conditions and charge distribution, a large amount of gas may suddenly flow to the periphery of the furnace.

As a result, Fe　10G−+Fe　10CO -Riri- The endothermic reaction of is promoted and the furnace heat decreases.

By the way, when a large amount of gas flows around the furnace, Na
, when the deposits in the furnace such as Pb and the stagnation layer peel off and fall off the wall, the temperature of the furnace wall in that part increases rapidly. If this rapid temperature rise is detected, it can be used to warn of a decrease in furnace heat.

II-B. Second Reason for Decrease in Furnace Heat The following may also be considered as contributing factors to the decrease in furnace heat in the blast furnace.

When the unloading speed in the blast furnace increases for the same reason as I-A, the level of the cohesive zone in the blast furnace decreases due to so-called raw ore unloading, resulting in a decrease in furnace heat.

By the way, when the cohesive zone level decreases, the furnace wall temperature at the corresponding portion also decreases rapidly. If this rapid temperature drop is detected, it can be used to warn of a decrease in furnace heat.

n-c, Third Reason for Decrease in Furnace Heat Furthermore, the following may be considered as a contributing factor to the decrease in furnace heat in the blast furnace.

High-temperature air (gas flow) blown up from the blast furnace tuyeres to adjust the temperature and amount of hot metal is usually blown into the center of the furnace. However, for the same reason as If-A, B1, a part of the gas flow may flow to the periphery of the furnace. If this state continues for a long time, heat dissipation from the gas flow from the blast furnace wall becomes greater than during normal operation, resulting in a decrease in furnace heat.

By the way, when a part of the gas flow steadily flows into the periphery of the furnace, the temperature of the furnace wall gradually increases. By detecting such a gradual temperature rise over a relatively long period of time, it is possible to warn of a decrease in temperature.

II-D, first to third furnace heat drop warning means Fig. 7(a)
, (b) are a side sectional view and a plan sectional view showing the arrangement of an inner wall thermometer used in an embodiment of the present invention, respectively.

As shown in Figure (a), there are seven inner wall temperatures 43 in the height direction of the blast furnace 1 (three at the back).

(2 abdominal parts, 2 morning glory parts), and 4 pieces are installed in the circumferential direction of the blast furnace 1 as shown in FIG. In other words, a total of 28 inner wall temperatures 3 are installed at 7 levels in 4 directions.

The inner wall thermometer is, for example, disclosed in Japanese Utility Model Application No. 57-81 by the present applicant.
531. The one disclosed in Japanese Utility Model Publication No. 59-16816 may be used, and FIG. 8 is a conceptual diagram showing the inner wall thermometer (hereinafter referred to as "rFM sensor") disclosed in the latter.

In the figure, reference numeral 4 denotes a sheath type thermometer having two conductive wires 5 buried in parallel insulatively and having a temperature sensing part 6 on the front end side. The moisture sensing parts 6 are arranged in parallel so that they are arranged at different lengthwise positions, and a sheath type dummy rod 7 is connected to the tips of the temperature sensing parts 6 to align the leading ends. The sheath type dummy rod 7 has two conductive wires 5 buried in parallel insulating manner, and has substantially the same thermal conductivity as the sheath type temperature measuring element 4. The FM sensor 3 is formed by keeping the sheath type temperature measuring element 4 in a non-contact manner with an insulating material 8 and housing it in a sheath tube 9.

FIG. 9 is an explanatory diagram of the installation of the FM sensor 3. In the figure, 10 to 13 are the walls of the blast furnace, 10 is a brick, and 1
1 is a stave, 12 is a stamp, and 13 is an iron skin. F
As shown in the figure, the M sensor 3 is installed inside the furnace wall by welding to a packing 14 and a welded portion 15. Note that 16 is a filler, and 17 is a milk injection port into which the filler 16 is poured.

Note that due to the installation and structure of the FM sensor 3 described here, the FM sensor 3 itself is eroded along with the erosion of the furnace wall, and the sheath type side heating body 4 may be exposed inside the furnace near the furnace wall. In addition to "furnace wall temperature", "furnace temperature near the furnace wall"
is being measured. Hereinafter, a concept including both will be described as "furnace wall temperature."

As mentioned above, compared to conventional thermometers such as sheathed thermocouples, the FM sensor 3 has a large number of measurement points, satisfies rapid temperature measurement response, is capable of long-term continuous temperature measurement, and is highly reliable. Improvements in performance, durability, and workability are being aimed at.

Each FM sensor 3 measures the inner wall temperature of the blast furnace 1 at every predetermined sampling time Δt, as shown in FIG.

Here, the inner wall temperature of the i-th FM sensor 3 at time j is expressed as 3. Let J-1 be 1 sun J at time j, 1 pull time Δ, and the previous inner wall temperature J-1
,1 and the inner wall temperature difference (difference value) J,l between T... and T.
J-1,1 ΔT0. is J, 1 ΔT,', = T, - Dj-1, i ... (1
8) J, l J, 1 becomes. This state is shown in FIG.

This difference value ΔTj,i is multiplied by a weight W· in consideration of the height of each FM sensor 3 in one circumferential direction, etc. For Zarapo, the difference value ΔTj, , is negative, V.

-01 For other values, v is also multiplied by the positive/negative parameter V, which indicates -1, to obtain the corrected difference value (positive difference value) CT at time j.

J, I CT・,=W,・■,・6-cho1. River (19) J, I
I I J, first order, corrected difference value CT
The sum of J,1 for all FM sensors 3 of , , is taken, and this is defined as STI.

ST1. = Σ CT, ... (2
0) J t = 7 J, 1 Then, according to the following equation (21), this difference value sum STI is calculated.

If the value becomes larger than a predetermined threshold value ε1, it is assumed that there has been a sudden temperature rise, and an evaluation point of 1 is given for a predetermined period of time to warn of a decrease in furnace heat.

STI, ≧ε...(21)
The above is the first furnace heat drop warning means based on the reason I-A.

The second furnace heat drop warning means based on the reason of IF-B is
It is as shown below.

In equation (19), the positive/negative parameter ■ is set to V+ = O for those with a positive difference value of 4, and for those other than 1, it is set to ■ = 1, and then , correction ■ For all the FM sensors 3 of the absolute value of the positive difference value CTj, , and according to the following equation (21)', the value of the difference value sum ST2 based on equation (20)' is determined from the predetermined threshold value ε2. If the temperature increases, it is assumed that a rapid temperature drop due to raw ore descent has occurred, and an evaluation point of 1 is given for a predetermined period of time to warn of a decrease in furnace heat. .

ST2. ≧ε2...(21)'I
The third furnace heat drop warning means based on the reason -C is as shown below.

As with the first prediction means, the positive/negative parameter V in equation (19) is J for negative differential values, 1 v-=O1, and %J for other values. =1. Also, at time j, before the sampling time (i.e., Δtx
k hours ago) corrected difference value '+ CT j-,.

Then, for all FM sensors 3 of the value at time j of the moving average of the predetermined time width nΔt of this corrected difference value...(20
)” Then, according to the following formula (21), this moving average sum ST3
If the value of .epsilon. exceeds a predetermined threshold value ε3, it is assumed that there has been a gradual temperature rise for a long period of time, and an evaluation point of 1 is given for a predetermined period of time to warn of a decrease in furnace heat.

ST3・≧ε3...(21)"
Since the above-mentioned first to third furnace heat drop warning means each use the furnace wall temperature difference (difference value), accurate warning can be given regardless of whether the absolute value of the furnace wall temperature is above or below. . Furthermore, due to its ease of installation and good temperature measurement response, the FM sensor 3 can be placed to cover the entire circumference of the blast furnace, and by being able to grasp continuous inner wall temperature differences, more accurate predictions can be made. .

Moreover, the above-mentioned first to third furnace heat drop warning means can be realized by a combination coater. FIG. 12 is a flowchart showing the processing flow of the first furnace heat drop warning means. In the figure, in step 851, the furnace wall temperature Tj, , of each FM sensor 3 is measured at every sampling time Δ. Next, in step S52, the difference value of each FM sensor 3 is calculated based on equation (18).

Then, in step 853, (19), (20
) A total positive difference value 5T1j is calculated based on the equation. Furthermore, in step 854, this positive difference value sum 5T1
Compare j with a predetermined threshold ε1, and (21)
If the formula is satisfied as 1, in step 855 it is assumed that a decrease in furnace heat will occur due to rapid flow of the gas flow to the periphery of the furnace, and a predetermined Jil]lil evaluation point of 1 is given to warn of a decrease in furnace heat. On the other hand, if the formula (21) is not satisfied, it is assumed that there is no abnormality and the process returns to step S51.
Furnace heat reduction evaluation is performed by repeating steps S51 to S54.

FIG. 13 is a flowchart that does not show the process flow of the second furnace heat drop warning means. In the same figure, step 861
and the furnace wall temperature T of each FM set.

J,1 is measured every sampling time Δ. Next, in step 862, the difference value of each FM sensor 3 is calculated based on equation (18).

Then, in step 863, (19)', (20
) Determine the total negative difference value 5T2j based on the formula.

Furthermore, in step S64, this negative difference value sum 5T2j is compared with a predetermined threshold value ε2, and (
21) If the formula 'is satisfied, it is assumed in step 865 that a decrease in furnace heat will occur due to the increase in the unloading speed, and an evaluation point of 1 is given for the predetermined period. On the other hand, if the formula (21)' is not satisfied, it is assumed that there is no abnormality, and step S61
Returning to step 861 and repeating steps 864, the furnace heat drop is evaluated and a warning of the furnace heat drop is issued.

FIG. 14 is a flowchart showing the processing flow of the third furnace heat drop warning means. In the figure, step 871
and the furnace wall aIjtTj of each FM sensor 3.

1 is measured every sampling time Δ. Next, in step 872, the difference value of each FM sensor 3 is calculated based on equation (18).

Then, in step 873, (19)'', (20
) The moving average sum 5T3j of the positive difference values at the same width Δt between IIs is determined based on the equation 5T3j. Furthermore, step S74
In step 875, this positive difference value moving average sum ST3 is compared with a predetermined threshold value ε3, and if the formula (21) is satisfied, a decrease in furnace heat will occur due to furnace body heat dissipation in step 875. If the formula (21) is not satisfied, it is assumed that there is no abnormality and the process returns to step 871, and steps 871 to 874 are repeated. Evaluate furnace heat drop.

II-E, 4th. The fifth furnace heat drop warning means also includes I-A
For the reasons stated in , the Truloss Cff1 is used as a furnace heat drop warning means according to the following procedure. Depending on top gas component analysis by gas chromatography, ventilation conditions, raw material charging conditions, etc.
) is calculated for each sampling time Δt. here,
If the amount of trail loss C at time j is XJ, and the amount of trail loss C at a sampling time before time j (that is, Δtxk hours before) is X J −k, then the moving average xH of the predetermined time width nΔt at the current time j is , can be calculated by .

×−8 based on formula (22) is calculated for each sampling time Δt, and according to formula (23) below,
An evaluation point i is given based on the maximum threshold value εxi when exceeding εXn)×1,
Gives a warning of low furnace heat.

xH〉εxi(i-1~n)...(23) The above is the fourth
This is a means of reducing furnace heat.

In addition, the amount of nitrogen (%) in the furnace top gas detected by gas chromatography (hereinafter referred to as "gas chromatography N2 amount") has a strong negative correlation with trundle loss Chi, and instead of increasing trundle loss CI, gas chromatography A decrease in the amount of N2 can warn of a decrease in blast furnace heat.

As a result, if the gas chronograph N2 period at the current time j is Vj, and the gas chronograph Nl at the sampling time before time j (that is, ΔtXk hours before) is yj-, then the predetermined time width nΔt at the current time j is The moving average yH can be calculated as follows.

Calculate yH based on equation (24) for each sampling time Δ, and use equation (25) below to calculate y8 to a predetermined threshold εyj (J...1 to m) (εy1>εy2>...
・Evaluation point j is given based on the minimum threshold value ε, j when the value falls below εym), and a furnace heat drop warning is issued.

yHkuεyj(j・1~m)...(2
5) The above is the fifth furnace heat drop warning means.

Furthermore, when calculating the moving average of the amount of vine loss C, which is the fourth furnace heat drop warning means, the instantaneous value of the amount of vine loss C is
As shown in Figure 5 (a), the abnormal value E1.
E2 may occur.

Here, if the amount of vine loss C at time j is X, and the amount of vine loss C before one sampling time Δ is X, then the absolute value ΔX of the difference value of the amount of vine loss C is Δx-=lx, -x=
l - (26) J J J-
It becomes 1. By comparing this ΔX with the threshold value ε7 as shown in the same figure (b), the abnormal value E1. Find E2 and draw the same figure (C
), a method of smoothing by replacing it with the immediately previous measured value can be considered. By applying this method, a more accurate moving average of the true loss Cff1 can be obtained, and as a result, prediction with considerably high accuracy is possible.

A method for predicting a decrease in furnace heat by using a moving average of the amount of truss loss C, which includes such abnormal value correction, can be realized using a computer. FIG. 16 is a flowchart showing the flow of the process. In the figure, first, in step S81, threshold values εx1 to xl x2 xn εxn are set in n stages with a magnitude of ε <ε . . . ε . Then, in step 882, the instantaneous value Xj of the amount of trailing loss C is determined at every sampling time Δ.

Then, in step 883, the absolute value ΔXJ of the difference value of the amount of trailing loss C is calculated, and then in Stella 1884, if the absolute value ΔXJ of the difference value is larger than the threshold value ε2,
In step 885, this instantaneous value xj is regarded as an abnormal value and replaced with the immediately previous measured value X.
to move to. On the other hand, if it is smaller than the threshold value ε2 in step 884, the process moves to step 886 without changing the instantaneous value xj. In step 886, the time width nΔ
The moving average XH of t is determined, and the evaluation point i is initially set to 0 in the next Stella 1887.

Then, in step 888, a comparison is made between the moving average value X of the amount of trailing loss C and a threshold value εx1 (from i=0), and if XH≧εx1, the value of i is increased by 1 from 0 to 1 in step S89, and step In step 890, it is determined that i=n, or in step 888, xH〈εx(i
The evaluation point i is calculated by repeating steps 888 to S90 while increasing the threshold value ε stepwise until it is determined that the threshold value ε is +1), and the process proceeds to step 891. If XH<εx1, step S89. S90 is never executed and the evaluation point i is 0.
As a result, the process moves to step 391.

Finally, in step 891, steps 887-89
Output the evaluation point i obtained from 0.

Note that, of course, the above-mentioned abnormal value processing.

Application to a computer can be similarly realized in the case of the fifth furnace heat drop warning means based on the moving average value y8 of the amount of gas chromatin N2.

4. mentioned above. Since the fifth furnace heat drop warning means is based on a moving average for each sampling time Δ, it is possible to quickly obtain the evaluation score for the furnace heat drop warning, and the accuracy can be said to be sufficiently reliable.

Comprehensive prediction is made as described below by using the evaluation points of the first to fifth furnace heat decrease warning means described in II-F, comprehensive reactor heat decrease warning means If-D, If-E. .

FIG. 17 is a flowchart showing the flow of the process, and the following description will be made with reference to the same figure.

In step 5101, evaluation points f to f of the first to third furnace heat drop warning means are determined. Evaluation points 1 to f3 are 0 once →
If it becomes 1, that value is held during a data hold period, which will be described later. Therefore, if the hold time (approximately 2 hours) is set at time t1, the evaluation points f1 to f3 from time t to time t1+hr (i.e., data hold period) will change once from O to 1. , again 1
→It never changes to 0. During this data hold period, the FM sensor total value F (-f1+f2+f3) is
It is set at the same time as initializing the fluctuation time t when it changes from =O to F>0, and is set from here on.

As mentioned above, the time defined by is the data hold period.

Next, in step 8102, by determining whether the FM sensor total value F is O, it is determined whether the data hold time is set, and if F=O,
There is no setting for the data hold time, so step 5
Moving on to 105. On the other hand, if it is 110, the data bold time has already been set, so step 5103
By comparing the fluctuation time t and the hold time hr,
It is checked whether or not the current time is during the data hold period. If t≦h, it is during the data hold period and there is no need to initialize the FM sensor total value F, so the process moves to step 8105. However, step 5103
If t>h, it is determined that the data hold period has ended, and the FM sensor total value F is initialized to "0".

In this way, the FM sensor total value F is determined while taking the data hold period into consideration. This is because the first to third furnace heat drop warning means are the fourth. Since it has higher foresight than the fifth furnace heat drop warning means (predictions can be obtained quickly), it is necessary to use the first to third furnace heat drop warning means to comprehensively evaluate the prediction results for the same point in the future. This is because it is necessary to hold the results for a certain period of time.

Then, in step 5105, the evaluation point i by the fourth furnace heat drop warning means is calculated, and further in step 810
6, the evaluation score j by the fifth prediction means is calculated. Next, in step 8107, the overall evaluation C is determined according to the following formula.

C=w F+w i+w j...(
27) AB Nicode, W, We, Wc are 1st to 3rd. 4th゜5th
is the weight for each of the furnace heat drop warning means. This comprehensive evaluation C is examined in step 8108, and C-0
If so, it is assumed that there is no tendency for furnace heat to decrease at all, and step 5 is performed.
Returning to step 101, the overall evaluation C is determined again through steps 8101 to 5107. On the other hand, if C>0, an alarm is output as a furnace heat drop warning in step 8109, and a predicted value is output at alarm output time tAL by the integrated model prediction means described in I-E.

Figure 18 is a graph showing an operation example of comprehensive prediction based on comprehensive evaluation C, in which (a) is the representative value of hot metal temperature, (b) is the sum of positive difference values (first prediction means), and (C) is negative (d) is the sum of the moving average of positive difference values (third prediction means), (e) is the gas chromatography N2fA
(f) shows the moving average value (fourth prediction means), (f) shows the moving average value of the crane loss C (fifth prediction sub-means), and (g) shows the change over time in the comprehensive evaluation C. The representative value of the hot metal temperature in (a) is the average value of the humidity actually measured every 30 minutes during tapping into the hot metal pot (about 2 hours per tap). For the comprehensive evaluation C, W=w=Wc=1 in equation (27), n=B m=3 in equations (23) and (25), and the domain of the comprehensive evaluation C is 0 to 9.

In the figure, the maximum value of the comprehensive evaluation C is 9, and the actual control temperature is T as shown in (a) (■) in the figure. It can be seen that the typical hot metal temperature values measured were considerably lower than .

(2) Comprehensive model prediction at the time of warning of decrease in furnace heat At time tAL when an alarm is generated by the comprehensive furnace heat drop warning means described in section 2, the comprehensive model prediction means described in section Wang is activated as described below. FIG. 19 is a flowchart showing the flow of the process.

This will be explained below with reference to the same figure.

First, in step 5111, the predicted values by the first to fourth model prediction means described in Section 1 are calculated for each sampling time Δt2, regardless of whether they are adopted as comprehensive model prediction means or not. And step 5
112, the period NΔt starts from the latest predicted value calculation time t.
2 Backward to the past, actual hot metal temperature value y and predicted value y1 to y
4 average value V, V1 ~ y4 and standard deviation σ, σ1 ~ σ4
Calculate. Next, in step 5113, t=t
The prediction error J1 (1) is calculated by equations (10) to (12)
is calculated, and further in step 5114 1=1.
Correlation degree J2 (1
) is calculated.

Then, in step 5115, step 5113
．． 5114, the prediction error J1 (j and correlation degree J
2 Calculate the error evaluation J (1) from equation (17) by calculating the error evaluation J (k-th (k = 1...4) where j has the minimum value). The model prediction means is employed as a comprehensive prediction means.

Thereafter, in step 5117, the predicted value y (t+Δt2) of the hot metal temperature one point ahead of the latest predicted value calculation time t by the first prediction means is calculated again, and in step 8118, the predicted value? ((t, +Δt2)
Output. In this case, the predicted value yk (t.

+Δt2) has already been calculated, but the value differs in the following points. For example, when = 1, the predicted value V according to the following equation (1a)
1 (t, +Δt2).

V (t + Δt2) 1 11.

-Σ ay(t-jΔt2) ;, a1J11 (1a) In formula (1a), the starting point time of the MA term is changed to t, → tAL, so that the predicted value V1 (t, +Δt2) is It may show a different value than before. In other words, at the alarm output time tA, the trail loss C, which is a variable introduced into the MA term, is
Since it is highly likely that the amount has increased, the data at alarm output time tA ((
In the case of formula la), incorporating u(t.L iΔt1)) allows the predicted value V (t
+Δt2) can be output p, and the prediction accuracy does not decrease.

In the case of k=2.3.4, the predicted value yk(t +
Of course, the output of Δt2) can be performed at an early stage, and the prediction accuracy does not deteriorate. This was mentioned in ■−〇■
It is clear that this will lead to improvement of the problems in ■.

IV. Supplementary note: In this example, the model prediction means with the highest prediction accuracy among the first to fourth model prediction means is used, but at least one
Of course, this can be realized by using two means.

Also, 1st. Second. When calculating the average of the amount of trail loss C in the fourth model prediction means, E8. By performing abnormal value processing as shown in FIG. 13, a more accurate average value of the true loss Cff1 can be determined, and as a result, the first. Second. The fourth model prediction means can make even more accurate predictions.

Also, since Gask0N2ffi has a strong negative correlation with Truross CI, this is the first. Second. A similar effect can be expected by introducing the MA term 144- instead of the truss loss C amount used in the fourth prediction means. Furthermore, it is also conceivable to introduce variables related to the furnace wall temperature difference used in the first to third furnace heat drop warning means into the MA term.

Although an FM sensor was used as the inner wall thermometer in the first to third furnace heat drop warning means in this embodiment, a normal temperature sensor (for example, a sheathed thermocouple) may also be used as a substitute, although it has problems in terms of service life. It is possible. Furthermore, a stave thermometer or a brick-embedded thermometer can be used as a substitute, although the prediction accuracy will be slightly lower due to their reliability and low temperature response.

Further, in the first to third furnace heat drop warning means in this embodiment, 28 FM sensors 3 are installed in 7 levels and in 4 directions, but it goes without saying that they may be installed appropriately depending on the characteristics of the blast furnace.

Furthermore, as described in II-F, it is desirable to use all of the first to fifth furnace heat drop warning means as the comprehensive furnace heat drop warning means, but at least one of the first to fifth prediction means is used.
It can be substituted by using one.

(Effects of the Invention) As explained above, accurate furnace heat is delayed by outputting a predicted value by the ARMA model adopted as a model prediction means at the time when the furnace heat decrease warning means issues a furnace heat decrease warning. It can be predicted without any problem.

[Brief explanation of the drawing]

FIGS. 1(a) and 3(b) are graphs showing the changes over time in the moving average value of the hot metal temperature and the amount of tulle loss C, respectively, and FIGS. A flowchart showing the processing procedure of each prediction method of the second model prediction means, the fourth
The figure is a graph showing the change in hot metal temperature over time when the hot metal temperature differs due to the difference in the tap hole. Noro chart showing the processing procedure of each prediction method of the fourth model prediction means, No. 7
Figures (a) and (b) are a side sectional view and a plane sectional view showing the arrangement of the FM sensor used in the embodiment of the present invention in the blast furnace wall, respectively, and Figures 8 and 9 are respectively the FM sensor. Conceptual diagram,
Installation explanatory diagram, Fig. 10 is a graph showing changes over time in the furnace wall temperature measured by the FM sensor, and Fig. 11 is a graph showing changes over time in the difference value of the furnace wall temperature measured by the FM sensor.
Figure 2 is a flowchart showing the process flow of the first furnace heat drop warning means, Figure 13 is a flowchart showing the process flow of the second furnace heat drop warning unit, and Figure 14 is the third furnace heat drop warning unit. Flowchart showing the flow of processing of means, No. 15
Figures (a), (b), and (c) are graphs showing the instantaneous value of the amount of vine loss C including abnormal values, the absolute value of the difference value of the amount of vine loss C, and the instantaneous value of the amount of vine loss C after removing abnormal values, respectively. ,
FIG. 16 is a flowchart showing the processing flow of the fourth furnace heat drop warning means, FIG. 17 is a flowchart showing the processing flow of the comprehensive furnace heat drop warning means, and FIG. 18(a),
b), (c), (d), (e), m, (a
) are the representative value of hot metal temperature, the sum of positive difference values, the sum of absolute values of negative difference values, and the moving average sum of positive difference values. Figure 19 is a graph showing the moving average value of the gas chromatin N2 amount, the moving average value of the tsuru loss C amount, and the changes over time in the overall evaluation C.
FIG. 20 is a flowchart showing the flow of the integrated model prediction means at the time of a furnace heat drop warning, and is a graph showing prediction results based on actual hot metal temperature values based on the conventional AR method.

Claims (3)

[Claims] (1) A model prediction method that uses the ARMA model to predict temporal fluctuations in the blast furnace heat level using the temperature of hot metal tapped from the blast furnace, and operations that have a strong correlation and foresight against the decline in blast furnace heat. Furnace heat drop warning means is provided which uses the results to warn of a furnace heat drop when the operation results satisfy a predetermined condition, and when the furnace heat drop warning means warns of a furnace heat drop, the model prediction means A blast furnace furnace heat prediction method that outputs a predicted value of blast furnace furnace heat. (2) The model prediction means is a first model prediction means that predicts the furnace heat using an ARMA model in which an average value of the amount of solution loss carbon or the amount of nitrogen in the furnace top gas component over a predetermined time width is introduced as the MA term. a second model prediction means for predicting furnace heat using an ARMA model that introduces the maximum value of the amount of solution loss carbon in a predetermined time width or the minimum value of the nitrogen amount in the furnace top gas component in a predetermined time width as an MA term; A in which a variable to compensate for differences in pigtails is introduced as an MA term.
A third model prediction means for predicting furnace heat using an RMA model, the maximum value of the amount of solution loss carbon in a predetermined time range or the minimum value of the amount of nitrogen in the furnace top gas component in a predetermined time range, and the difference in the tap port. At least two of the fourth model prediction means in which a variable for correcting is introduced as an MA term
Compare the past predicted values by the first to fourth model prediction means with the actual measured hot metal temperature, and make a model prediction with the highest prediction accuracy among the first to fourth prediction means. A blast furnace furnace heat prediction method according to claim 1, wherein the blast furnace furnace heat is predicted by means. (3) The furnace heat drop warning means installs an inner wall thermometer at a predetermined location of the blast furnace, measures the inner wall temperature difference at predetermined time intervals with the inner wall thermometer, and measures the inner wall temperature difference at a certain time. The first one gives an evaluation point for a predetermined period when the total value of the parts showing a positive value exceeds a threshold value.
a furnace heat drop warning means, and a second means for giving an evaluation point for a predetermined period when the total value of the portion indicating a negative value of the inner wall temperature difference at a certain time exceeds a threshold value.
a furnace heat drop warning means, and a third furnace heat drop that gives an evaluation point for a predetermined period when the sum of moving average values over a predetermined time width of the portion showing a positive value of the inner wall temperature difference at a certain time exceeds a threshold value. a fourth furnace which is provided with at least one warning means and which calculates the amount of solution loss carbon at each predetermined time interval and gives an evaluation point when the moving average value of the calculated value over a predetermined time width exceeds a threshold value; a fifth furnace heat reduction unit that calculates the amount of nitrogen in the furnace top gas component at predetermined time intervals and gives an evaluation point when the moving average value of the calculated values over a predetermined time width falls below a threshold value; further comprising at least one of warning means, an evaluation score of at least one of the first to third furnace heat drop warning means and an evaluation of at least one of the fourth and fifth prediction means. A blast furnace furnace heat prediction method according to claim 1 or 2, which warns of a decrease in furnace heat based on a comprehensive evaluation of points.