JP2737139B2 - Blast furnace heat control system - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a blast furnace heat control device for controlling the temperature of hot metal discharged from a blast furnace.

[Prior art] Conventionally, the temperature of hot metal in a blast furnace is estimated, and this is managed and controlled.
As a control method, a method is generally adopted in which a blast furnace operator evaluates the blast furnace by qualitatively judging information from various sensors installed in the blast furnace and performs optimal adjustment of the operation factor. . However, there are individual differences in the results of the evaluation due to the capabilities and experience of operators, which makes it difficult to standardize the operation actions, and it is difficult to estimate the hot metal temperature because the evaluation is not quantitative. there were.

For this reason, for example, a blast furnace process control method as disclosed in Japanese Patent Publication No. 51-30007 has been proposed. This process control method keeps the blowing temperature constant, calculates the amount of direct reduction in the furnace from the measurement values that can be obtained continuously during operation, and calculates the target value of the Si content in the pig and the exponential smoothed value that represents the actual value. The blast moisture is determined by an equation with a correction term added to prevent long-term fluctuations of the Si content in the pig, and the heat balance in the furnace is controlled by the blast moisture determination value. ing.

For this reason, a large wave change (change in a long cycle) that occurs when the blast furnace condition is calculated and controlled is corrected to realize an accurate operation.

[Problems to be Solved by the Invention] In the conventional process control method disclosed in Japanese Patent Publication No. 51-30007, information from a sensor is analyzed and input to a model to perform a predetermined calculation. ing. For this reason, a computer that executes the computation uses, for example, Fortran as a language, but the computation capacity is extremely large. Further, since the blast furnace changes over time, maintenance must be performed by changing the analysis model itself. However, since the analysis model itself is complicated, changing the conditions of the analysis model is extremely troublesome.

As means for solving the above problems, the maintainability can be improved by a computer system using an artificial intelligence language, for example, LISP. Here, sensor information (true / false data, various sensor data) and ,
In inferring the furnace heat situation using the knowledge base,
When the production rule is used, for example, IF (the temperature of the sensor i is in the range of T1 to T2) THEN for all the sensors concerned.
(The CF value that is the high heat level is C1)... (The CF value that is the low heat level is Cn); IF (The temperature of the sensor i is in the range of T1 to T3.) THEN
(CF value at high heat level C′1),... (CF value at low heat level is C′n);... In addition, the adjustment of the CF value is extremely complicated.

For this reason, there is a problem that the amount of numerical values indicating the degree of certainty described in the creation of the rule program or in the rules becomes enormous, and the input operation becomes complicated.

Furthermore, the effects of actions taken in the past are not taken into account, and thus the actual condition of the furnace is not always grasped, and from this point, the accuracy of controlling the furnace temperature is insufficient. there were.

The present invention has been made in order to solve such a problem, it is possible to control the furnace heat of the blast furnace with high precision, and when implemented by a computer, to improve its arithmetic capacity and arithmetic speed, It is another object of the present invention to provide a blast furnace heat control apparatus that can easily add and modify rules even in a new situation such as aging of the blast furnace and that takes into account the effects of past actions.

[Means for Solving the Problems] FIG. 1 is a block diagram showing the concept of a blast furnace heat control device according to the present invention. The blast furnace heat control device according to the present invention is a data input unit that captures data from various sensors installed in the blast furnace at a predetermined timing, and based on the data from the sensors, tuyere embedding temperature, unloading speed,
Create various data indicating the blast furnace status, such as pressure loss, furnace top temperature, gas utilization rate, solution loss amount, etc.
Processing data creating means for comparing the various data with the reference data to create the difference data, further comprising a storage means for temporarily storing the various data and the difference data (hereinafter referred to as processing data); Knowledge base storage means in which various knowledge bases based on operation experience, results, mathematical models and the like are stored, and furnace heat levels and furnaces based on the processing data in the storage means and the knowledge base in the knowledge base storage means. It has inference means for inferring a heat transition and determining an action amount for the blast furnace, and means for outputting the action amount.

The knowledge base storage means includes a "furnace heat level determination knowledge base" used for estimating a furnace heat level, a "furnace heat transition determination knowledge base" used for estimating a furnace heat transition, "Action determination knowledge base" used to determine the action amount from the heat level and furnace heat transition, "action correction knowledge base" to determine the correction value of the action amount based on past actions and disturbances, and action It includes an “overall judgment knowledge base” for determining an actual action amount from the amount and the action correction amount.

The "furnace heat level determination knowledge base" and the "furnace heat transition determination knowledge base" use the processing data and the furnace heat level as independent variables, and the likelihood of occurrence of a combination thereof (hereinafter, referred to as a certainty factor) as a dependent variable. A confidence function, and a rule group for determining a method of applying the confidence function, respectively, wherein the `` action determination knowledge base '' has a rule for determining an action amount from a furnace heat level and a furnace heat transition, The “comprehensive judgment knowledge base” has a rule for determining an actual action amount from an action amount and an action correction amount.

Further, the "action correction knowledge base" has an action-in correction rule used to determine an action taken in the past and its influence amount to obtain an action correction amount.

[Operation] In the present invention, various data indicating the state of the blast furnace is created by the machining data creation means based on the blast furnace data from the data input means, and then machining data is created based on the data. An inference operation as artificial intelligence is performed based on the processed data and the knowledge base stored in the knowledge base storage means to determine an action amount for the blast furnace.

In the inference calculation, the furnace heat level is estimated using the furnace heat level determination knowledge base, and the furnace heat transition is estimated using the furnace heat transition determination knowledge base. When inferring these furnace heat levels and furnace heat transitions, a certainty function and a group of rules that determine how to apply the function are used to obtain certainty for the inference result.

Next, the action amount is determined by applying the furnace heat level and the furnace heat transition to the action determination knowledge base. Further, the influence of the past action is determined using the action correction rule of the action correction knowledge base, and the action correction amount is determined based on the influence.

Next, the actual action amount is determined from the action amount and the action correction amount using the comprehensive judgment knowledge base.
Then, the blast furnace heat is controlled by adjusting the operation factor based on the action amount.

Example An example of the present invention will be described below with reference to the drawings. Second
FIG. 1 is a block diagram showing a blast furnace heat control apparatus and related equipment according to an embodiment of the present invention.

In the figure, (1) is a blast furnace to be controlled, and (10) is a blast furnace heat control device according to the present invention, which is a data scanner (11), a sensor data collection unit (12), a file unit (14), and a calculation unit. (16), output means (18), knowledge base (22), inference engine (26), and common data buffer (24).

The sensor data collection means (12) inputs data from various sensors, for example, a temperature sensor, a pressure sensor, a gas sensor, etc., in a time-series manner through data scanners (11a), (11b), and (11c).

The file means (14) fulfills a file data banking function. The arithmetic means (16) performs exponential smoothing processing on the data from the sensor data collecting means (12) stored in the file means (14), and then re-executes the file means (14).
To be stored. The average value and the difference data between the average value and the reference value are set as processing data every predetermined time, for example, every 20 minutes, and the common data buffer (2
Send out to 4). The calculation means (16) also performs a part of calculation processing of an action correction rule described later in addition to the above.

The inference engine (26) has its data and knowledge base (22)
Based on the knowledge of the above, a predetermined inference operation is performed, and the amount of action to be taken next is obtained and stored in the common data buffer (24) again and also stored in the file means (14).

(30) is a CRT, and the inference result of the inference engine (26)
The information is transmitted and displayed via the file means (14).

(32) is a digital instrumentation device, which is a blast furnace heat control device (1
It controls the temperature of the blast furnace based on the command 0), and adjusts the operation amount of the operation type that can adjust the furnace heat such as the blast moisture, the blast temperature, and the liquid fuel such as heavy oil. At this time, the command of the blast furnace heat control device (10) is sent out to the digital instrumentation device (32) via the output means (18).

(40) is a hot blast stove, and (42), (44) and (46) are control valves respectively.

An outline of the operation of the present embodiment having the above configuration will be described.

(1) First, data of various sensors are sequentially read at predetermined timings, for example, at one-minute intervals by a sensor data collecting means (12) via a data scanner (11) and stored in a file means (14). .

. The data stored in the file means (14) is subjected to exponential smoothing processing by the arithmetic means (16). Here, data relating to unloading, temperature, gas utilization rate, tapping slag, and the like are processed. The various data subjected to the arithmetic processing are stored again in the file means (14). Next, an average value of these various data is determined at predetermined time intervals, for example, every 20 minutes, and a difference between the average value and a predetermined reference value is obtained and transferred to the common data buffer (24) as processed data. .

(3) The inference engine means (26) infers the situation in the blast furnace based on the knowledge data stored in the knowledge base (22) in advance and the processing data of the common data buffer (24).

Here, the knowledge base includes a furnace heat level determination KS group, a furnace heat transition determination KS group (KS; knowledge source), an action determination KS, an action correction determination KS, a comprehensive determination KS group, and an operation state determination as shown in FIG. Formed from each knowledge base unit of KS.

Furnace heat level determination The KS group is a knowledge base used by the inference engine (26) to determine the furnace heat level of the blast furnace, and determines the furnace heat level using the hot metal temperature as the main determining factor. "Hot metal temperature-furnace heat level KS", and "sensor-furnace heat level KS" for determining the furnace heat level based on the measured amount of other sensors.

Each of these KSs is a confidence function (hereinafter referred to as CF function) with each measured quantity and furnace heat level as an independent variable, and the probability (confidence) of the occurrence of the combination as a dependent variable.
And the rules that determine the procedure for using the CF function.

The KS Group is a knowledge base used by the inference engine (26) to determine the level of the furnace heat transition of the blast furnace. Includes "hot metal temperature-furnace heat transition KS", which determines the transition of furnace heat, and "sensor-furnace heat transition KS", which determines the transition of furnace heat mainly based on the transition of measured values of other sensors. .

These KS also have a CF function that makes each measured quantity and furnace heat transition level an independent variable, a dependent variable of the probability (confidence) of the combination of them, and a set of rules that determine the procedure for using the CF function. It consists of.

The action determination KS is made up of a rule group for determining an action amount based on a combination of a furnace heat level and a furnace heat transition level. The action correction amount KS group is made up of a rule group for correcting the current action amount based on information on actions taken in the past and disturbances occurring in the past. The comprehensive judgment KS is made up of a group of rules for determining the final action amount based on the result of the action judgment and the result of the action correction amount judgment.

Then, the operation state determination KS determines whether or not the operation of the furnace is normally performed based on, for example, the furnace heat level,
If it is normal, the above-mentioned action amount is sent to the control system as it is, and if it is abnormal, a rule group is displayed to indicate the fact and guide the operator.

The inference engine means (26) executes each knowledge base. First, as shown in the flowchart of FIG. 3, the furnace heat level and the furnace heat transition are determined, and then the action amount is determined based on these determination results. Is determined. This action amount is subjected to a predetermined correction, and the result is generally stored in a common data buffer (24) and then sent to a digital instrumentation device (32) via a file means (14) and an output means (18). Sent.

Then, the control valve (4
2), the opening degree of (44) and (46) is appropriately controlled, an action operation is performed, the temperature of the blast furnace (1) is controlled, and as a result, the hot metal temperature is controlled to a desired value.

Next, the configuration of the knowledge base and the outline of the specific inference will be described with reference to FIG.

(A) Furnace heat level determination KS (KS (Knowlege Source) knowledge source (group; this furnace heat level determination KS group) is a knowledge base that determines the state of furnace heat at the inference start time, as described above. -Furnace heat level KS "," sensor-furnace heat level
As shown below, for each furnace heat level divided into seven stages from high to low level for each KS group, the CF value distribution The confidence level is the furnace heat level at the current time.

Furnace heat level evaluation 7 Large heat 6 Medium heat 5 Normal 4 Small cold 3 Medium cold 2 Large cold 1 Extra large cold Here, an example of hot metal temperature-furnace heat level KS will be described.
The KS is composed of an IF section in which a condition is set and a THEN section in which the content of an instruction when the condition is satisfied is set. Examples are as follows.

Rule No.1 [IF section] Pot order = 1 NOT (many residues) NOT (elapsed time after wind reduction ≤ 180 minutes) Judgment of Si and S is "low" [THEN section] Melting by ordinary three-dimensional function Calculate CF value of pig temperature-furnace heat level.

This rule No. 1 indicates that when the operating state of the blast furnace is in a steady state, the CF value of the hot metal temperature-furnace heat level is obtained by a normal three-dimensional function.

That is, the hot metal temperature-furnace heat level CF value function is usually ~ f
There are three types: _MTN , slightly higher ~ f _MTH , and higher ~ f _MTEH , and the inference engine (26) selects which function to use based on the above logic.

In other words, the hot metal temperature-furnace heat level KS includes the order of the ladle (the number indicating the ladle in which the ladle for which the hot metal temperature was measured hits the ladle used since the start of tapping) and the amount of residue A rule for determining which of the three types of functions is to be used is stored in accordance with the elapsed time from the end. Another example of this rule is as follows: Rule No.2 [IF section] Pot order = 1 NOT (many residues) NOT (elapsed time after the end of wind reduction ≤ 180 minutes) Judgment of Si and S is "somewhat high" [THEN part] The CF value of the hot metal level is determined by a three-dimensional function when the temperature is slightly high.

Rule No.3 [IF section] Pot order = 1 NOT (many residues) NOT (elapsed time after wind reduction ≤ 180 minutes) Judgment of Si and S is "high" [THEN section] 3D when "high" The CF value of the hot metal level is determined by the function.

These rules No. 1 to No. 3 are that when there is not much residue and enough time has passed since the end of wind reduction, Si and S
Indicates that the type of function is selected.

The three CF functions of hot metal temperature-furnace heat level are all functions of ladle order LN, hot metal temperature MT, and furnace heat level FHL.

That is, f _MTN = f _MTN (LN, MT, FHL) f _MTH = f _MTH (LN, MT, FHL) f _MTEH = f _MTEH (LN, MT, FHL) The inference engine (26) was selected according to the rule.
With respect to the CF value function, LN and MT are applied to actual measured values, and for FHL, the numerical values from 1 to 7 are applied to obtain CF values corresponding to respective furnace heat levels.

FIG. 4 shows a CF value function for one pot order. Since the pot order is fixed, the CF value is a function of the hot metal temperature and the furnace heat level.

For example, if f _MTN is selected by the rules, when the hot metal temperature is 1400 ° C, the CF value is highest when the furnace heat level is 4
When CF = 0.2 and the melting point temperature is 1480 ° C., the furnace value of 7 indicates that the CF value is the highest and the CF value is 0.4.

Next, the sensor furnace heat level (tuyere embedded temperature-furnace heat level
KS, solution loss C amount-furnace heat level KS) will be described.

This KS is a rule for determining whether to use tuyere filling temperature-furnace heat level KS and solution loss C amount-furnace heat level KS, and tuyere filling temperature and furnace heat level, respectively.
It consists of CF value functions f _HT and f _{SL using the} solution loss C amount and the furnace heat level as two independent variables.

f _HT = f _HT (HT, FHL) f _SL = f _SL (SL, FHL) Tuyere embedding temperature HT used to calculate this function
As the solution loss C amount SL, an actually measured value (the exponential smoothed value is a moving average value) is used.

The following are examples of rules that determine the use and non-use of these functions.

Rule No.1 [IF section] NOT (many residues) [THEN section] (1) Tuyere embedding temperature-furnace heat level according to tuyere embedding temperature
Find the CF value.

(2) The CF value of the solution loss C amount-furnace heat level is obtained from the solution loss C amount.

In this case, the inference engine (26) uses both functions.

Rule No.2 [IF section] (many residues) [THEN section] (1) Set the CF value of tuyere embedding temperature-furnace heat level to "0".

(2) Set the CF value of the solution loss C amount-furnace heat level to “0”.

In this case, the tuyere embedding temperature-furnace heat level CF value and the solution loss C amount-furnace heat level CF value are constant regardless of the furnace heat level.

This means that these factors do not contribute to the determination of the furnace heat level described later, and that the furnace heat level is determined only by the aforementioned hot metal temperature-furnace heat level KS.

FIG. 5 is an explanatory diagram showing the operation of the inference engine (26). The CF value for each furnace heat level is obtained based on the hot metal temperature-furnace heat level KS. The CF value for each furnace heat level is determined based on the level KS and the solution loss C amount-furnace heat level KS. And the CF value of each level and the solution loss C amount by the tuyere embedding temperature KS
Add the CF value of each level by KS. The CF value of the sensor level obtained in this way and the CF value of each level based on the hot metal level KS are added. Thus, the CF value of each furnace heat level (7-1) is obtained.

(B) Furnace heat transition judgment KS group; This furnace heat transition judgment KS group
KS and sensor-furnace heat transition KS are included, and the furnace heat transition is divided into five stages from steep rise to constant to steep descent according to the degree of change from the past to the present, as shown below. The CF value is obtained, and the step position of the maximum value is set as the furnace heat transition state at the current time.

Level Contents 5 Rapid rise 4 Up rise 3 Sideways 2 Down 1 Drop suddenly Here, the hot metal temperature-furnace heat transition KS will be described.

The KS is a CF function _fΔMT = _fΔMT (ΔMT, VFHL) with the difference ΔMT of the hot metal temperature between the preceding and following taps (one tapping) and the furnace heat transition level VFHL as two independent numbers.
Pre-processing and post-processing rules for the F function are stored.

 Hereinafter, the method of using the KS will be described.

Rule No. 0 [IF section] (Initial setting) [THEN section] The following values are set as CF values of hot metal transition.

Rule No.1 [IF section] (1) NOT (Judgment of Si and S is "slightly high") (2) NOT (Judgment of Si and S is "high") (3) Stability flag is ON (reactor condition is ON) (Stable state) [THEN part] (1) Determine the CF value for each furnace heat transition level as ΔMT = (hot metal temperature of current tap-hot metal temperature of previous tap). CF value ”.

(2) Next, assuming that ΔMT = (hot metal temperature of front tap−hot metal temperature of front tap), a CF value is obtained for each furnace heat transition level,
It is assumed to be "last hot metal temperature-furnace heat transition CF value".

(3) “Current hot metal temperature-furnace heat transition CF value” and “previous hot metal temperature-furnace heat transition CF value” are each multiplied by a weighting factor and added to “hot metal temperature-furnace heat transition CF value”.

(These calculations are performed separately for each level of the furnace heat transition.) In other words, in the stable state of the furnace condition, up to the latest data is used for estimating the furnace heat transition.

Rule No.2 [IF section] (1) (Si, S judgment is "slightly high") (2) NOT (Si, S judgment is "high") (3) Stability flag is OFF [THEN section] ( 1) Assuming that ΔMT = (pre-tap hot metal temperature-pre-tap hot metal temperature), a CF value is determined for each furnace heat transition level, and is defined as "previous hot metal temperature-furnace heat transition CF value".

(2) Multiply the “last hot metal temperature – furnace heat transition” by a weighting factor and add it to “hot metal temperature – smoothing furnace heat transition CF value” and set it as “hot metal temperature – smoothing furnace heat transition CF value” again. I do.

In this case, since the furnace condition is not stable, the “current hot metal temperature-furnace heat transition” is not considered.

 Next, the sensor furnace heat transition KS will be described.

The sensor furnace heat transition KS stores the CF value function that uses the transition of measured values of many sensors and the furnace heat transition as two independent variables, and rules that determine how to use it. Is provided.

That is, f _ΔSi = f _ΔSi (ΔSi, VFHL) (i = 1 to n: for each corresponding sensor) An example in which the sensor has the tuyere embedded temperature will be described.

Rule No.1 [IF part] NOT (many residues) [THEN part] Assuming that ΔSi = (tuyere embedding temperature-60 min before tuyere embedding temperature), the tuyere embedding temperature-CF value of furnace heat transition Ask.

Rule No.2 [IF part] A lot of residue CF value at tuyere transition level 4> 0 [THEN part] Overwrite CF value at tuyere transition level 4 with "0".

(* Indicates that the value is the same as before.) Rule No.3 Many residues CF value of tuyere transition level 5> 0 [THEN part] "0" for the CF value of tuyere transition level 5 Set overwrite.

FIG. 6 shows the tuyere embedding temperature-furnace heat CF function using the tuyere embedding temperature (difference from the reference value) and the furnace heat transition level as independent variables.

In addition, it is also possible to make rules for the hot metal temperature-furnace heat transition KS by dividing it into “short term transition” and “long term transition”.
Further, in addition to the tuyere embedded KS, other KSs such as unloading, blast pressure, gas utilization rate, solution loss amount, etc. are added to the sensor transition KS, and such information is also taken into consideration.

The inference engine (26) obtains the CF value for each transition based on each rule of the hot metal temperature-furnace heat transition KS for each furnace heat transition level, and detects each sensor based on each sensor-furnace heat transition KS rule. Find the CF value for each transition. Then, the CF values of these KSs are added for each furnace heat transition level to determine the CF value of each furnace heat transition level.

(C) Action determination KS; This action determination KS is a knowledge base for determining the furnace heat state at the current time on a matrix based on the furnace heat transition and the furnace heat level, and determining an action to be taken.

The inference engine (26) calculates the product of the CF value of the furnace heat level and the CF value of the furnace heat transition based on the action determination KS, and writes the product in a matrix.

FIG. 7 shows an example of this. In this example, the peak (maximum value) of the CF value indicates that the furnace heat level = 4 and the furnace heat transition = 3. Note that the action type and the action amount at each position on the matrix are stored in the frame in advance as knowledge.

FIG. 8 is a diagram showing an example of an action type, and FIG. 9 is a diagram showing an example of an action amount.

When the action type action at a predetermined position is adopted, it is necessary that the CF value has reached a predetermined size. Further, the action amount is basically controlled automatically, but a part of the action amount can be controlled manually (for example, the action amount G in FIG. 9).

(D) Action correction amount determination KS group: In the action correction amount determination KS group, a past action or disturbance is determined, and a correction action amount is determined in consideration of the influence amount at the current time. Various KS to do is included. The contents include a change in manipulated variables such as ventilation humidity, ventilation temperature, liquid fuel, coke ratio, and the like, and rules for detecting and responding to disturbances such as coke moisture and adhering matter falling off.

For example, when the ventilation humidity is changed, the change time and the change amount are automatically detected by the “operation amount change detection” rule,
Subsequent effects are considered as a function of time according to the "blast humidity" rule.
Based on the "wall drop" rule, the location of dropout, the amount of influence on the furnace heat, and the tuyere descent time are automatically determined, and the operation time and amount of the preliminary action are determined and incorporated into the correction calculation.

The inference engine (26) executes the above-described rules to obtain a necessary correction action amount and operation time.

(E) Comprehensive judgment KS; This comprehensive judgment KS is a knowledge base for comprehensively judging an action amount to be taken based on the judgment results of the above (C) and (D). When the KS is inferred by the inference engine (26) and a determination result is obtained, the determination result is input to the operation state determination KS, and the operation state is determined,
The operator is instructed on the amount of action to be taken by displaying it on the CRT (30) and gives guidance, and at the same time, feeds back to the digital instrumentation device (32) to perform predetermined automatic control (see FIG. 2).

Next, a description will be given of a method of creating a multidimensional function used for obtaining a CF value in each of the above knowledge bases (A) and (B), for example, a CF function of hot metal temperature-furnace heat level.

FIG. 10 is a characteristic diagram showing transition of hot metal temperature in a tap (result of continuous measurement). As shown in the figure, even if the thermal state inside the furnace is stable,
Cooling in a tapping gutter, which is a flow path of the hot metal, causes the hot metal having a relatively low temperature to be discharged in the initial stage of tapping. With the passage of time, the tapping gutter rises in temperature due to the sensible heat of the hot metal, and since the effect of stagnation in the furnace decreases, the temperature of the discharged hot metal gradually increases and stabilizes, The information is well representative of the furnace heat that is the control target of the system.

Furthermore, since the tendency of the temperature rise is not always constant and changes depending on the operating conditions, the information of the hot metal temperature is different from the information waiting for the ambiguity that the eligibility as the information representing the control target of the system changes every moment. Become. Therefore, in order to observe the hot metal temperature and link it to the estimation of the furnace heat state, keep in mind the elapsed time from the start of tapping, and further consider the process-specific ambiguity. It is necessary.

When the hot metal temperature during tapping is grasped from the above viewpoint, the temperature distribution is represented as shown in FIG.
In the figure, each axis represents X axis; dimensionless time Y axis; hot metal temperature Z axis; appearance frequency (occurrence frequency ratio).

Fig. 12 shows the hot metal temperature at tapping time X = Xi in Fig. 11.
It shows the relationship between Ti and the maximum tapping metal temperature Tmi. By adjusting the relationship between the tap maximum hot metal temperature and the furnace heat and the relationship between the appearance frequency and the CF value based on this figure, the three-dimensional function shown in FIG. 13 is obtained. In addition, since the tapping temperature is measurement information depending on the elapsed time from the start of tapping and operating conditions, more than 30 types are prepared so that they can be used properly, and are automatically selected according to the conditions.

When the characteristic diagram of FIG. 13 is illustrated by replacing the X-axis with the sensor data, it is expressed as shown in FIG. In the figure, the dots
The straight line connecting A1 and A4 and the straight line connecting points A3 and A6 each have a CF value of “0”, and the CF value increases toward the middle of both straight lines. Then, the straight line connecting points A2 and A5 indicates that the CF value has the maximum value "1", indicating that the reliability is the highest.

As a result of controlling the furnace temperature based on such a three-dimensional function, as shown in FIG. 15, it can be seen that the frequency of occurrence of errors is reduced as compared with the conventional operator method.

The three-dimensional function in FIG. 14 is formed by connecting the minimum value and the maximum value of a predetermined range of the sensor data with a straight line.
FIG. 16 expands the area of the sensor data and divides the area into a plurality according to the size of the sensor data (three divisions in this example)
A three-dimensional function is formed by connecting polygonal lines so that each divided area is continuous. Such 3 in FIG.
The dimensional function indicates the relationship between the actual sensor data, the furnace heat level, and the CF value (certainty factor) as compared with the three-dimensional function in FIG. The maximum value of the CF value indicates “1” in each of the divided areas, but it goes without saying that the maximum value may be different for each area.

The reason why the sensor data of the three-dimensional function is difference data from the reference value (= measured value−target value) is as follows.

B) The target value of the hot metal temperature changes due to component adjustment and the like.

B) The standard of the sensor data is changed daily depending on the operation policy, for example, whether or not the fuel data is low fuel ratio.

C) Further, depending on the degree of wear of the installation position of the brick or the like, the detected temperature of the thermometer or the like is different even if the conditions in the furnace are the same.

The three-dimensional function is adapted to cope with various changes based on the above difference data for the above reasons.

FIG. 17 shows the relationship between the action instruction, the furnace heat level and the hot metal temperature. The action instruction by the three-dimensional function in FIG. 14 is made at the timing indicated by the solid line in the tap at the tap, and the action instruction (dashed line) by the three-dimensional function in FIG. 16 is performed at the timing shown by A and B in the figure. . As a result, in the case of the three-dimensional function in FIG. 16, the hot metal temperature becomes the target value at the next tap. On the other hand, in the case of the three-dimensional function in FIG. 14, the value slightly deviates from the target value.

FIG. 18 shows the results for the hot metal temperature, Si and S when using the three-dimensional functions of FIGS. 14 and 16, and the results when using the three-dimensional functions of FIG. 16 are better. The characteristics superior to those in the case of FIG. 14 are obtained.

As a result, in the case of the three-dimensional function in FIG. 16, the hot metal temperature becomes the target value at the next tap. Compared to this
In the case of the three-dimensional function shown in FIG. 14, the value slightly deviates from the target value.

FIG. 18 shows the results for the hot metal temperature, Si and S when using the three-dimensional functions of FIGS. 14 and 16, and the results when using the three-dimensional functions of FIG. 16 are better. The characteristics superior to those in the case of FIG. 14 are obtained.

By the way, since there is a certain relationship between the action instruction, its response and the effect, a rule called “action correction rule” is introduced in the above-mentioned action correction determination KS loop D to correct the amount of action. . This action correction rule is a rule used to determine an action taken in the past and its influence amount, and to consider it when determining the next action amount. Hereinafter, the action correction rule will be described.

FIG. 19 shows the results of an action response test performed during the furnace heat stabilization period. The tuyere embedding temperature (Ttu) immediately responds to the action of the blowing temperature of 30 ° C, but the gas utilization rate (ηco) and the blowing pressure (Pb) have a response delay (dead time). However, all of them are stable in 80 to 90 minutes. On the other hand, the sensor behavior in actual operation is different from that in this example. In other words, when the tuyere implantation temperature and gas utilization rate decrease and the furnace heat can clearly be predicted to decrease, it is customary to perform an operation to return to the reference level, so the behavior of the sensor is also based on a certain reference value. It is controlled within a certain range.

Next, the processing of the “action correction rule” will be described more specifically with reference to FIGS. 20 and 21.

Fig. 20 shows the change over time of the humidified air.
FIG. 21 is a flowchart showing the operation of the “action correction rule”. Note that the blast moisture in FIG. 20 is data taken in through the data scanner (11c), and the calculation in the flowchart in FIG.
Done by

When obtaining the manipulated variable at the current time point Xn in FIG. 20, first, the parameter i is set to 2 and the change flag SW is set to 0,
(S1), the maximum value of the difference between the three points (X1, X2, X3) around i = 2 is determined (S2). The maximum value is compared with the allowable value XfM (S3), and when the maximum value is smaller than the allowable value number XfM,
An average value of the three points is obtained (S5).

Next, it is checked whether or not the change flag SW = 0 (S6). Since the change flag SW = 0 here, the above average value is stored as the operation value Xb (S7). Then, the content of the parameter i is increased by "1" and advanced by one minute (i = i + 1
= 3) After (S8), the process returns to the step of obtaining the maximum value of the difference between the three points (S2).

As is clear from FIG. 20, since the change is small from X1 to X6, the above steps are repeated. Therefore, the average value Xt6 is stored in the operation value Xb.

In the same manner, the above calculation operation is repeated, but in a section where the humidity is changing, the maximum value of the three points is larger than the allowable value, and it is detected that the state is changing. The time is advanced again as 1 (S4), and the maximum value of three points is obtained.

Then, go into the next section where there is no change, find the maximum value of the difference between the three points (Xi−1, Xi, Xi + 1) in the same manner as above,
If it is determined that the maximum value is smaller than the allowable value, an average value of the three points is obtained, it is determined that the change flag SW is not 0 (S6), and the average value Xti and the operation value already obtained are determined. Xb
It is checked whether or not the absolute value of the difference from (= Xt6) is larger than the minimum action value Xd (S9). If it is, the manipulated variable Xz is obtained by Xz = Xti−Xb (S10). . Next, next, the current value Xti and the change flag SW at this time are set as Xb = Xti,
SW = 0 is set (S11), and the time is advanced by one minute again (S11).
8) Return to the first calculation (S2) As described above, the past operation amount of the blast humidity is grasped, and is grasped as a correction amount when calculating the next operation amount to be taken. In this calculation, the past manipulated variables are similarly obtained not only for the humid air content but also for the blast temperature, coke ratio, blast flow rate, and the like.

The manipulated variable obtained as described above is determined by the following overall judgment KS
Is used in the arithmetic processing using, and the action amount is obtained. FIG. 22 shows a flowchart of the calculation process, and FIG. 23 is an explanatory diagram showing various cases of the correction order.

 First, the following equation is calculated.

Hv1 = No. Of correction order + current action amount Here, No. 1 of correction order is the past operation amount of the operating factor to be controlled first, and this is the blast humidity in the example of case A in FIG. If it corresponds to the minute, the sum Hv1 of the operation amount Xz and the action amount (output of the action determination KS) is obtained.
If this Hv1 is smaller than the first upper limit value of the preset correction order, it is calculated as the current action amount A.
Output as

If Hv1 exceeds the first upper limit of the correction order or the upper limit, Hv1 = the first upper limit of the correction order Hv2 = the second of the correction order + the remaining B of the current action amount is set. In the case A, the second correction order is a past operation amount (a value corresponding to the above Xz) of the blowing temperature, and the remaining B of the current action amount is the remaining B of the current action amount = the current B of the current action amount. Action amount A—the amount indicated by the first upper limit of the correction order.

If Hv2 obtained here is smaller than the second upper limit of the correction order, Hv1 and Hv2 are output as the current action amount.

If Hv2 exceeds the second upper limit of the correction order or the upper limit, Hv2 = the second upper limit of the correction order Hv3 = third of the correction order + the remaining C of the current action amount is set. The second correction order is the past operation amount of the coke ratio (the value corresponding to the above Xz) in case A.
The remaining C of the current action amount is an amount indicated by the remaining C of the current action amount = the remaining B of the current action amount−the second upper limit of the correction order.

If Hv3 obtained here is smaller than the third upper limit of the correction order, Hv1 to Hv3 are output as the current action.

If Hv3 is equal to or exceeds the third upper limit of the correction order, Hv3 = the third upper limit of the correction order Hv4 = the fourth of the correction order + the remaining D of the current action amount is set. The fourth correction order is the past operation amount of the air flow rate (the value corresponding to the above Xz) in case A, and the remaining D of the current action amount is the remaining D of the current action amount = The remaining amount of the current action is the amount indicated by the third C-correction order. Then, the above Hv1 to Hv4 are output as the current action amount.

The flowchart of FIG. 22 is based on the upper limit value of each correction order, but the lower limit is also processed by the same flowchart. The action amount obtained in this way is processed based on the operating state KS as described above, and then each control valve is appropriately controlled via the digital instrumentation device (32).

FIG. 24 is a characteristic diagram showing operation results when fully automatic control was performed by the blow water ratio control. For example, action D1 (water ratio −3 g / Nm3) is due to a decrease in furnace heat level. The action at 15:00 (water ratio -2g / Nm3) is C type (no action) at the current time, but was taken in consideration of the coke ratio -2kg / T falling to the tuyere tip later. It is.

Comparing the operation results according to the present embodiment with the operation results obtained by the conventional operator, it can be seen that the accuracy of controlling the hot metal temperature is improved as shown in FIG. This means that operation management has been standardized.

[Effects of the Invention] As described above, according to the present invention, processing data is created from various data set in a blast furnace, and a predetermined inference as artificial intelligence is made based on the processing data and a knowledge base based on experience and the like. To make full use of previous experience, standardize operation management, prevent human misjudgment,
Stable supply of high-quality hot metal with little component fluctuation to the next process, avoidance of furnace cooling, labor saving, etc. have been realized.

In addition, when the control device is implemented by computer, the knowledge base is used as a reference, so the calculation capacity and calculation speed are improved, and rules can be easily added and modified even in new situations such as aging of the blast furnace. It has become.

Since the confidence function is used to obtain each furnace heat level and each furnace heat transition CF value, the number of rules, the work of inputting the numerical values, and the like are reduced.

Further, since the next action amount is determined in consideration of the past operation results, control corresponding to an actual furnace can be performed, and therefore, highly accurate control can be performed. Especially,
Since the operation results are taken into consideration, it is effective, for example, when controlling offline (manual control) and then switching to online control.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of the present invention, FIG. 2 is a block diagram of a blast furnace heat control device according to one embodiment of the present invention,
FIG. 3 is a flowchart showing a knowledge base and its inference order, FIG. 4 is a characteristic diagram showing a three-dimensional function of a furnace heat level, and FIG. 5 is an explanatory diagram for explaining an inference operation. FIG. 6 is a characteristic diagram showing a three-dimensional function of the tuyere transition level,
FIG. 7 is an explanatory diagram showing an example of the action matrix, FIG. 8 is an explanatory diagram showing an example of the type of the action matrix, and FIG. 9 is an explanatory diagram showing an example of the action amount. Fig. 10 is a characteristic diagram showing the transition of hot metal temperature in the tap, and Fig. 11
FIG. 13 to FIG. 13 are explanatory diagrams showing a method of creating a three-dimensional function of the furnace heat level. FIG. 14 is an explanatory diagram showing an example of a three-dimensional function, and FIG.
FIG. 14 is a characteristic diagram showing the results of the three-dimensional function, FIG. 16 is an explanatory diagram showing another example of the three-dimensional function, and FIG. 17 is a three-dimensional function of FIG. 14 and a three-dimensional function of FIG. 18 is a characteristic diagram showing the relationship between the action instruction of the control, the furnace heat level and the hot metal temperature, FIG. 18 is a three-dimensional function of FIG. 14 and a temperature difference with respect to a target temperature indicating a control result by the three-dimensional function of FIG. FIG. 4 is a characteristic diagram showing distributions of Si and S. FIG. 19 is a characteristic diagram showing a result of an action response test performed when the furnace is stabilized, FIG. 20 is a characteristic diagram showing a change in humidified air, FIG. 21 is a flowchart showing an operation of a correction operation,
FIG. 22 is a flowchart showing the operation of calculating the action output rule, and FIG. 23 is an explanatory diagram showing the priorities at the time of correction. FIG. 24 is a characteristic diagram showing the results of an operation example of the fully automatic control based on the injection water proportional control, and FIG. 25 is a characteristic diagram showing the error rate of hot metal temperature. (11), (12); data input means (16), (14); processed data creation means (19), (24); storage means (22); knowledge base storage means (26); inference means

Claims (1)

(57) [Claims] 1. A data input means for taking in data from various sensors installed in a blast furnace at a predetermined timing; and a tuyere embedding temperature, based on data from the sensors.
Create various data indicating the condition of the blast furnace, such as unloading speed, pressure loss, furnace top temperature, gas utilization rate, solution loss amount, etc., compare the various data with the reference data, and create the difference data. Processing data creating means; the various data and difference data (hereinafter referred to as processing data)
Storage means for temporarily storing information, knowledge base storage means for storing various knowledge bases based on experiences, results, mathematical models, and the like regarding blast furnace operation; processing data of the storage means and knowledge of the knowledge base storage means. Inference means for inferring a furnace heat level and a furnace heat transition based on a base and determining an action amount for the blast furnace; and means for outputting an action amount, wherein the knowledge base storage means estimates the furnace heat level. A furnace heat level judgment knowledge base used for estimating a furnace heat transition judgment base used for estimating a furnace heat transition,
An action determination knowledge base used to determine the action amount from the furnace heat level and furnace heat transition, an action correction knowledge base that determines a correction value of the action amount based on past actions and disturbances, an action amount and an action Comprehensive judgment knowledge base that determines the actual action amount from the correction amount, the furnace heat level judgment knowledge base and the furnace heat transition judgment knowledge base, the machining data and the furnace heat level as independent variables,
A confidence function having the likelihood of occurrence of these combinations (hereinafter referred to as confidence) as a dependent variable, and a rule group for determining a method of applying the confidence function; A rule for determining an action amount from a level and a furnace heat transition, the comprehensive judgment knowledge base has a rule for determining an actual action amount from the action amount and the action correction amount, and the action correction knowledge base further includes: A blast furnace heat control apparatus, comprising: an action correction rule used to determine an action taken in the past and an influence amount thereof to obtain an action correction amount.