JPH02166203A - Method for controlling blast temperature in blast furnace - Google Patents

Abstract

PURPOSE:To drastically reduce deviation of molten iron temps. by making processing data with data fetched from various kinds of sensors in a blast furnace, converting them into plural membership functions and deciding the related blast temp. variative rate. CONSTITUTION:The blast furnace operational data 1 from the various kinds of sensors set in the blast furnace are fetched into a system through a data inputting means 2 at the prescribed timing, and processed into the necessary observing event data with a data processing means 3. The processed data is converted with a membership functional means 4, and actual condition recognition 51 and predicting result 52 related to furnace heat are outputted. The actual condition recognition 51 and the prediction 52 hereafter is converted into the necessary action with an expert rule 6, and output processing means 7 decides the final action rate with synthetic rule in fuzzy theory from the action rate and the reliability thereof to output the blast temp. variative rate 8. This variative rate 8 is transmitted to a blast furnace control computer and the blast temp. in the blast furnace is controlled with the output of this computer.

Description

[Detailed description of the invention] [Industrial application field 1 The present invention relates to a blast furnace blast temperature control method.

[Prior Art] Estimating and managing and controlling the temperature inside the blast furnace, which is represented by the temperature of hot metal, has traditionally been done by blast furnace operators qualitatively determining information from various sensors installed in the blast furnace. This was done by estimating and evaluating the blast furnace heat level and furnace heat transition, and making optimal adjustments to operational factors.

However, the results of these evaluations and estimates vary from person to person depending on the ability and experience of the operator, making it difficult to standardize operation methods, and because the evaluation is not quantitative, appropriate actions were not always taken. . In particular, predicting future trends in furnace heat depends more on the operator's experience and judgment than the current level of awareness.
The problem is that it is more difficult to take accurate predictive actions.

In recent years, due to the difficulty of improving the accuracy of analysis models, technologies are being developed that use AI, incorporate the know-how of blast furnace operators, and improve the accuracy of predicting the furnace heat necessary for blast furnace operation. For example, Japanese Patent Application Laid-Open No. 62270708 , 62-2707
No. 12 is disclosed.

The former determines the furnace heat according to human judgment rules based on the observation of the molten iron temperature, tuyeres, and slag, estimates the furnace heat transition based on information obtained from various sensors, and calculates the furnace heat level and furnace heat transition. This is a system that takes actions to control furnace heat, and the latter is a system that enables diagnosis of slips and blow-throughs.

The furnace heat transition here is based on ■ Judgment rules based on the past to present transition of molten iron temperature, ■ Levels of tuyere embedding temperature, unloading, blowing pressure, gas utilization rate, solution loss amount, furnace top temperature, etc. Judgment is made based on a five-step transition evaluation level comprehensively determined from judgment rules based on changes in reactor conditions, and judgment rules based on changes in reactor conditions and their respective confidence levels.

Changes in the furnace heat level are caused by many factors and can be observed, so we are trying to solve this problem by expressing the empirical rules of regular operators in production rules.

However, even experienced operators do not fully understand the causal relationship between the above various data and the furnace heat transition. Therefore, the accuracy of the judgment rules and the reliability determined for each of the above ■ to ■ is insufficient, and there is a risk that the reaction amount taken at the timing of the furnace heat control action is inappropriate, resulting in a functional decline. was there.

Also, Japanese Patent Application Publication No. 60-248804, Publication No. 601872
1. Japanese Patent Publication No. 60-204813, Japanese Patent Application Publication No. 61-2548
53 and others are all techniques for predicting furnace heat using mathematical models such as statistical estimation, exponential averaging, and frequency analysis. These techniques depend on the accuracy of mathematical models, and it has been difficult to assemble model equations with high precision.

In addition, model output does not correspond one-to-one with inference results, and there is a large gap in matching the operator's senses, and tuning by each individual requires a large maintenance load. was.

[Problem to be Solved by the Invention 1] The present invention was developed in order to solve the following problems with the conventional method of predicting furnace heat transition using production rules and mathematical models in expert systems. It is something that

(1) The causal relationship between various observed data and changes in the furnace heat level is unclear, making it difficult to assemble a formalized logic.

(2) The amount of action cannot be determined uniquely,
It is difficult to correspond to the furnace heat level and prediction results.

(3) Maintenance load to maintain system performance is high. Parameter tuning etc. takes time. Its parameters may also change.

[Means for Solving the Problems] The present invention provides: (a) Monitoring of blast furnace conditions such as unloading of the blast furnace, gas composition, furnace heat level, and furnace body temperature based on various sensors installed in the blast furnace. (b) A process of processing these into various processed data that can show recent changes and fluctuations, for example,
(c) calculating the amount of change in ηco and the amount of change in N2 as data indicating the state of the gas components based on the measured gas analysis value; (c) converting these processed data into at least two or more data regarding the current furnace heat; (d) Current status (e) A step in which an action rule is determined based on a knowledge base based on judgments regarding the furnace heat and future predictions; It consists of the process of deciding.

Additionally, the system automatically learns data that should serve as a standard for making accurate judgments. That is, a method has been added in which various data output from membership functions for furnace heat level determination and furnace heat prediction determination are aggregated, and their distribution states are formulated and automatically replaced with the membership functions.

[Step 1] FIG. 1 is a block diagram illustrating the human output and processing steps for blast furnace heat according to an embodiment of the present invention.

Although this system is naturally implemented using a blast furnace process computer, it is not limited by the scale, number, configuration, etc. of the computers. However, when using manual input or prediction results of blast furnace operation data for furnace heat control, the output needs to have sufficient real-time performance.

Now, in FIG. 1, blast furnace operation data 1 is operation data based on measurement data by various sensors, and these are taken into the system through data input means 2, and are necessary for determining furnace heat by data processing means 3. The data is processed into observational event data. In the past, operators did not predict trends in furnace heat by looking only at the current values of blast furnace operation, but by looking at trends in measured data from the past few hours up to the present. Therefore, information is processed to include temporal factors such as the amount of change over a certain period of time from the past to the present.

The processed data is converted by membership function converting means 4 and outputs current recognition 51 and prediction results 52 regarding furnace heat.

These outputs are: (a) hot, (b) just right, (C) young, and (d) will get hot, (e) will stay the same, and (f)
) 52 on a three-level prediction that he would become younger. Since each item is converted to a membership function, the above 2(a)
A corresponding ratio (certainty) is calculated using a number between the judgment classification of ~(f) and O~1.

The current state recognition 51 and the future prediction 52 are converted into required actions by the expert rules 6, and the output processing means 7 determines the final action amount from the amount of action received from the expert rule 6 and its certainty level by a synthesis rule based on the fuzzy theory. , output to the air blowing temperature change amount 8. This change amount 8 is sent to the blast furnace control computer, and the blast furnace feed J! I Control temperature.

FIG. 2 shows the membership function converting means 4 shown in the first section.
This is an explanation of the following.

Figure 2 (a) is a membership function that determines the current furnace heat, and the furnace heat index AR regarding the hot metal temperature index AuMy and Si.
5i and furnace heat index AHO, each processed data is expressed with a certain degree of certainty. These are converted into intermediate outputs indicating whether the furnace heat is hot, young, or suitable by synthesizing the confidence level using the knowledge base of operator rules.

An example of this rule is shown in Table 1.

For the synthesis of CFx, CFYI, and CFZ, the minimum value may be selected according to the fuzzy theory, or it may be indexed using arithmetic processing unique to the expert system.

FIG. 2(b) shows the membership function that determines the action (the amount of change in air temperature), and also shows nine rules. Examples 2 and 3 will be explained as follows.

■ If it is determined that the current furnace heat level is "young" and the future prediction is "it will become younger", take action.
Select.

■ If it is determined that the current furnace heat level is "young" and the future prediction is "not likely to change," take action 4.

■ If the current furnace heat level is "hot" and the future prediction is "it will get hot", take action 9.
Take.

Figure 3(a) shows the action 1 in Figure 2 again.

Action 1 indicates the amount of change in air temperature between 20 and 30°C, and when the confidence level is 1, the peak value (marked with a square mark, 25°C)
It is a membership function that means .

For example, the final action amount, which is the conclusion of the action obtained in this way, is obtained by defuzzifying the fuzzy composite amount using the centroid method, as illustrated in FIG. 3(b).

Next, automatic creation of membership functions related to furnace heat determination will be explained. From the membership functions for each data shown in FIG. 2, membership functions related to the determination of furnace heat are automatically formed in the following manner.

The average value of the representative hot metal temperature during several tappings is plotted on the horizontal axis, and a membership function as shown in Fig. 4 is created. AHMT at the same timing when the representative temperature is suitable,
If data from processing processes such as AR9L and AHO are sampled, a set as shown in FIG. 5 will be obtained. Similarly, AHMT is applied at the appropriate timing when the representative temperature is hot.
, AR5j, and AHO can be sampled to obtain another set, and if these are subjected to normal distribution approximation or trapezoidal processing in a computer, membership functions related to furnace heat determination can be automatically created.

Next, automatic correction of the membership function regarding prediction of furnace heat is performed as follows. The horizontal axis is the amount of change in temperature of representative hot metal that has undergone the same treatment as in Figure 4 compared to the previous time, and the membership functions of hotter, younger, and unchanged are calculated. Create as shown in Figure 6.

When the amount of change in the representative temperature falls under the judgment that it has become hot, a scatter diagram of the predicted value of the furnace heat transition is drawn in the same way as in Figure 5, going back three hours corresponding to the time difference of the representative hot metal temperature. create.

If you go back for the past 3 hours and create data for the furnace heat transition prediction judgment when it belongs to the judgment that it has not changed or has become younger, and perform normal distribution processing on a computer, you can determine the membership regarding the judgment of furnace heat prediction. Functions can be obtained automatically.

Here, as shown in Figure 7, the time to go back in the past was set to 3 hours, but this indicates the future time difference in which operators predict changes in furnace heat by looking at various data, and is usually 2 to 4 hours. It is between. Although the time of change in the representative hot metal temperature is not constant, it is sufficient to process only data within 2 to 3 hours in either case.

The one-time ventilation control method having the above configuration will be explained in more detail below.

First, the blast furnace operating pig given to the manual means 2 in FIG. 1 is as follows.

(1) Judgment of the current furnace heat level ■ (Pig iron volume (grade) IMT The molten metal temperature determined for each volume indicates the blast furnace heat level.

■ Percentage ratio of the actual distribution ratio of R51 Si to the equilibrium distribution ratio R5L”
(Lsi/LSL) x 100 is the furnace heat index. LSi has a ratio of 9 outside the actual results, LSi"fsi' [%S t ] / a SS
10212o fsj=0. l 8 [%C]+0.
Calculated using 11[%Si].

The equilibrium distribution ratio LSi is calculated. (b) Here PCO and T are P co (
arm, ) (1 + 0.967X blow gauge pressure) + 0.192 + distance between feather tap taps 0, 75) T = (Hot pig iron quality + 273) ``C''.

■ HQ Indicates the amount of heat used to heat pig iron slag, and differentiates the amount of heat generated before the tuyere into the solution loss reaction, the reduction reaction by N2, and the endotherm due to the decomposition reaction of CaC0a.

(2) Judgment of furnace heat transition prediction ■ Unidirectional index Unidirectional index = Σ ((index finger interval, min/ch) l =
4 (Calculated charging time required)) This is an index indicating whether the index finger interval is becoming faster or slower.

■ Charging number The number that indicates the current charge of the day ■ Gas utilization rate (ηC0) Co, CO2: Top gas components ■ N2 component (N2) [N21 analysis value of furnace top gas ■ Furnace heat index (Ho ) It represents the amount of heat used to heat pig iron slag, and is used as a so-called furnace heat index. Solution loss reaction from the immediate calorific value, reduction reaction by N2,
This differentiates the endotherm caused by the decomposition reaction of CaCO3.

■ Solution-8 Amount of carbon (, C3OL) Amount of carbon gasified by direct reduction [phase] Stave heat load (S T L ) STL=Fx
(To Ti ) , there are also some DPLs.

Among the above data, ■ and ■ are data related to unloading, ■
, ■ is data on gas components, ■, ■ are data on the furnace heat itself, [phase], and ■ are data on the temperature and pressure of the furnace body. ing.

Next, an embodiment of the data processing means 3 shown in FIG. 1 will be described. The following data processing is performed in response to items ① to ① and ② to ② in O11.

■'"Topedo every HMTJ is the past 3 Tovido every HM
Let it be the latest moving average value of T.

■'rARsi is the latest moving average value of J past three tap representative R5i.

■'rAHoj is the latest moving average value obtained by sambling the 30-minute average value of the furnace heat index HQ for the past 3 hours as AHQ.

■′ “One-way index” is processed into the amount of change. Latest 5
The difference between the average charge and the average of the previous five charges is defined as the amount of change in the one-sided index.

■′ “Number of charges” is processed to calculate the difference in the number of charges. Latest 2
The difference between the number of charging times and the same value used in the previous estimation is the difference in the number of charging times.

■'ry+cOJ is processed into the amount of change in ηCO. The difference between the average value of ηCO for the most recent 30 minutes and the average value of 71CO for the most recent 2 hours is defined as the amount of change in ηCO.

■' [N2] is processed to the amount of change in N2. The processing method is the same as ■'.

■' rHoJ is processed into the amount of change in HQ. 30 minutes m
1 (moving average of the average of Q over the last 3 hours and 5 to 71L'f
The difference from the previous moving average is defined as the amount of change in HO.

■'"Solution loss C amount" is the amount of change in solution loss C amount. The processing method is the same as ■'.

[Phase]' rsTLJ is processed into the amount of change in STL.

The difference between the STL average value for the most recent 30 minutes and the STL average value 1 to 2 hours ago is defined as the amount of change in STL.

■'rDPLJ is processed into the amount of change in DPL. For processing method, turn ■′.

Next, a specific example will be described.

Now the values of ■' to ■' are 1487℃, 125.200, +
100, +5, +3.0, +2-0.50, -20, +
The determination output regarding the furnace heat when the temperature is 50, +0.03 is obtained using the membership function shown in FIG. 8, which is automatically created by a computer. Judgment of the current furnace heat level is shown in Figure 8 (a) A HMT - Just the right confidence level 1.0AR5i
Just right confidence 0.5, Young confidence 0.5 AHO: Young confidence 1,0 If these are combined, just right confidence 0.5 and Young confidence 0.5 are calculated by minimum value composition.

Similarly, from FIG. 8(b), Confidence that it will be hot: 0.5 Confidence that Tarou will not change: 0.5 is calculated.

From the membership functions for the actions in Figure 2(b), we can see that (a) the current furnace heat level is younger with a younger confidence of 20.5, and the same confidence of 20.5 is the action 4; (b) the current reactor heat level is Confidence that the heat level is just right.

0.5, and with confidence unchanged, 0.5, action 5: (c) Current reactor heat level young confidence: 2 0.5, and heat (with confidence that it will be: 0.5, Action 7 is selected, (d) Action 8 is selected when the current furnace heat level is just right 0.5 and the confidence that it will get hot is 20.5, and the distribution of the amount of action according to the corresponding ratio using the mountain cutting method. is shown in Figure 9.

FIG. 9 shows the distribution of the manipulated variables corresponding to these imperfections. The operation amount distribution defined in FIG. 9 is cut and peaked according to the corresponding ratio of the judgment function.

When combining multiple operation distributions as shown in Figure 9, the distribution consisting of the maximum value (shaded area) is re-evaluated, and this is calculated as follows (
1) Calculate using the formula. The manipulated variable output C is calculated using a method called the weighting method as follows (1).

Specifically, the above formula (1) is as follows: Molecule = (-20)xO+ (-1'7)xo, 5+
(-13)xo-5+(-10)xQ+ (-7) x
0.5 + (-〇, 3) x 0.5+OxO+
I OxO +13xO-5+l 7x0.5+20xO3,65 Denominator = 0.5 + 0.5 + 0.5 + 0.5
+ 0.5 + 0.5. Therefore, an output of C=-1,22 is obtained.

The results of implementing the present invention are not shown in FIG. Compared to the conventional case where operations relied on the experience of individual operators, the deviation in hot metal temperature, which is a typical target for furnace heat, was reduced by half.

[Effect of the invention 1] With the present invention, the point where the relationship between data and judgment criteria was unclear in the past can be filled in to the senses of the operator using an ambiguous judgment function, and the output of the action can be averaged by ambiguous judgment. Therefore, there is no gap between the operator's sense and the operator's sense, and maintenance can be done by simply converting the sense into a membership function, which greatly reduces the burden of tuning.

[Brief explanation of the drawing]

Figure 1 is a block diagram of the system of the present invention, Figures 2 and 3.
Figure 4 is a diagram showing an example of a membership function, Figure 5 is a diagram showing an example of a membership function.
The figure is a scatter diagram of predicted values, Figure 6 is a graph of membership function, Figure 7 is a graph of change in hot metal temperature, Figure 8 is a graph of membership function of the example, and Figure 9 is a graph of action amount synthesis. The graph shown in FIG. 10 is a graph showing the effects of the present invention. -Blast furnace operation data 2...Manual means 3--Processing means 4--Functioning means 6--Expert rules 7--Output processing means 8--Change amount Fig. 6 Diagram (a) Melt a, temperature] 530 Pig star diagram 10

Claims (1)

[Claims] 1. Data is taken in from various sensors installed in the blast furnace at a predetermined timing, processed data indicating the status of the blast furnace is created and indexed based on the data, and the processed data is individually Convert into two or more membership functions based on fuzzy theory regarding furnace heat level and furnace heat transition prediction, calculate corresponding proportions corresponding to the membership functions, and calculate by knowledge base means of operator rules or knowledge of operator rules. An air temperature control method, characterized in that an amount of air temperature change corresponding to the membership function is determined by a synthetic rule of a fuzzy theory based on a base. 2. Automatically determine the boundary values of membership functions for determining furnace heat levels and predicting furnace heat changes based on the distribution of blast furnace operation data collected at the same timing when the membership functions are in the same state. 2. The air blowing temperature control method according to claim 1, wherein the temperature is generated by learning.