JPH05195035A - Device for controlling blowing converter - Google Patents

Abstract

PURPOSE:To provide a blowing control device of molten steel in a converter. CONSTITUTION:In the converter blowing control device for securing the prescribed molten steel components and molten steel temp. based on informations directly and indirectly obtd. over the whole term of the refining in converter equipment, the optimum input data is automatically selected in order to execute the setting control to various kinds of the aimed characteristics, and a reaction characteristic value outputting device 7 constituted of a neural net operating the reaction characteristic value during blowing is set in each reaction characteristic of each aimed characteristic. Further, a control operational quantity outputting device 8 for outputting the control operational quantity by the fuzzy inference constituted with the neural net, based on a membership function formulating the blowing control deciding function of the skilled operation with the fuzzy rule from the reaction characteristic value outputted through the reaction characteristic value outputting device 7, and a factor learning device 9 for calculating and correcting a combining factor between units in each neural net of the reaction characteristic value outputting device 7 and the control operational quantity outputting device 8 from the blowing result, are provided.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a molten steel blowing control apparatus for a converter.

[0002]

2. Description of the Related Art Various blowing control methods have been proposed to improve the quality of molten steel by controlling the components and temperatures of molten steel and slag produced by converter blowing.

The methods of controlling the composition and temperature of the molten steel are roughly classified according to the estimation and prediction method of the in-furnace condition. A method of predicting the composition and temperature of the molten steel from the reaction model during the blowing,
Directly or indirectly measure the component and temperature from various sensor information such as gas flow rate, component and vibration, and acoustics in the furnace during blowing, and based on the measurement result, oxygen supply to the furnace and quick lime, iron ore There is a method for controlling so as to obtain a desired molten steel composition and temperature at the end of blowing.

The former method is the information such as the weight, composition and blowing lance of the main raw materials such as hot metal and scrap charged into the converter, auxiliary raw materials such as quick lime, and the amount of oxygen supplied from the bottom tuyeres of the furnace. Based on the above, the composition and temperature of molten steel are predicted momentarily from the blowing reaction model, and the weight of auxiliary raw materials such as quick lime, etc., so that the composition and temperature of the molten steel at the end of the blowing reach predetermined target values, It controls the composition, blowing lance, amount of oxygen supplied from the tuyere of the furnace bottom, and the like.
No. 6 publication.

However, in the prediction by the above-mentioned blowing reaction model, it is not possible to accurately grasp the fluctuations of the components and temperature in the molten steel with the progress of the reaction in the furnace, especially the short-term fluctuations of the reaction at the end of blowing, The composition and temperature at the end of blowing may fall outside the target range.

In the latter method, the reaction condition is calculated from the acoustic pattern during blowing by a device for collecting the in-furnace sound installed in the furnace mouth or in the vicinity of the furnace mouth, and the blowing condition is appropriately changed. The composition of molten steel at the end of blowing is controlled by the method described in JP-A-51-075610.

A method of controlling the molten steel component at the end of the blowing by changing the blowing condition according to the value of each spectrum of the molten steel component at the blowing fire point of the blowing lance measured through an optical device ( For example, Japanese Patent Laid-Open No. 02-22561
No. 1) is also known.

However, in any of the above methods, oxidation and
With the progress of the reduction reaction, the composition and properties of molten steel and slag,
Due to differences in the blowing conditions such as the temperature of hot metal at the start of blowing, the components, and the manipulated variables during blowing, the temperature and other conditions are intricately entangled with each other, and have a complicated entangled relationship. However, since the measurement accuracy of the measurement situation is insufficient, the control accuracy by the accurate calculation of the composition and properties of molten steel and slag and the calculation of the control operation amount was not sufficient.

[0009]

As described above, conventionally, it has been impossible to accurately grasp the short-term fluctuation of the reaction at the final stage of blowing, and the situation in the reactor accompanying the progress of the reaction in the reactor is complicated, and the measurement accuracy is low. Since insufficient calculation and control of molten steel and slag are not possible due to insufficientness, proper blowing control was not performed. For this reason, the composition and temperature of molten steel at the end of blowing will fall outside the target range, and as a result, it will take time to readjust at the end of blowing, which hinders logistics in the factory and prevents fire There were drawbacks such as the progress of melting of objects.

An object of the present invention is to provide a converter blowing control device that solves the above drawbacks.

[0011]

The converter blowing control apparatus of the present invention secures a predetermined molten steel composition and molten steel temperature based on information that can be obtained directly or indirectly over the entire refining period in the converter equipment. A converter blowing control device for
A reaction characteristic value output device composed of a neural network that automatically selects optimum input data for setting control for various target characteristics and calculates reaction characteristic values during blowing is provided for each reaction characteristic of each target characteristic. Installed, control operation by fuzzy reasoning constructed by neural network based on membership function that formulates the blowing control judgment function of refining operator by fuzzy rule from reaction characteristic value output from said reaction characteristic value output device A control operation amount output device for outputting a quantity,
A coefficient learning device is provided for calculating and correcting the coupling coefficient between the units in each neural network of the reaction characteristic value output device and the control operation amount output device from the blowing result.

In one embodiment, each of the reaction characteristic value output devices has main raw material information, auxiliary raw material information, which can be directly or indirectly obtained over the entire period of refining in the converter equipment.
The reaction characteristic correlation judgment that selects the data with a high correlation between each input data and the reaction characteristic value record, using the input data of the calculated manipulated variable information before blowing, the measured value information indicating the progress of blowing, and the converter characteristic information. Means for extracting the data selected by the reaction characteristic correlation determining means, normalizing the data for each update, and determining the number of the data, and outputting from the data number determining means. Each time the output data is changed, a combination creating means including a unit for inputting a combination of the respective data and data having a high correlation between the output data value of the combination creating means and the actual result of the reaction characteristic value are calculated. Then, the neural network is composed of selecting means for selecting and reaction characteristic value calculating means for calculating and outputting the reaction characteristic value based on the output data from the selecting means.

In one embodiment, the control operation amount output device is a blowing control which is performed during blowing, which is empirically obtained from a past reaction record by a refining operator based on a reaction characteristic value. In order to formulate the judgment function by the fuzzy rule, the antecedent part of the fuzzy inference uses the reaction characteristic value which is the output from each reaction characteristic value output device as the input data, and the respective input data and the control operation amount result. Operation amount correlation determining means for selecting data having a high correlation, and data number determining means for extracting the data selected by the operation amount correlation determining means, normalizing the data for each update, and determining the number thereof. Of the correlation between the output data value of the combination creating means and the actual result of the control operation amount. Control data, which is a target characteristic based on the output data from the selecting means, and the fuzzy inference consequent part is configured by selecting means for calculating and selecting the desired data every time the output data is changed. Is configured to perform fuzzy inference by a neural network composed of control operation amount computing means.

In one embodiment, the coefficient learning device calculates the unit coupling coefficient between layers in each neural network of each reaction characteristic value output device and control manipulated variable output device as a reaction characteristic after completion of blowing. Of the actual value and the calculated value, and a correcting means for correcting each unit coupling coefficient based on the calculation result.

[0015]

As described above, the present invention is based on empirical knowledge that a blowing expert estimates and controls the composition, properties, and temperature of molten steel and slag from past performances to control changes in various sensor information during blowing. Focusing on the judgment ability, the empirical knowledge and the judgment ability are realized by a fuzzy control device using a neural network, and the weight, composition and blowing lance of auxiliary materials such as quick lime, oxygen supplied from the tuyere of the furnace bottom. It is a converter blowing control device for controlling the amount and the like.

[0016]

Embodiments of the present invention will be described below in detail with reference to the drawings.

FIG. 1 shows an outline of a converter equipment and a converter blowing control apparatus according to an embodiment of the present invention. As shown in the figure, for example, main raw materials such as hot metal and scrap charged in the converter 1 and auxiliary raw materials such as quick lime react with oxygen supplied from the blowing lance 2 and the bottom tuyere 3 of the furnace. With refining, the internal conditions of the molten steel and slag in the converter, such as composition, properties, and temperature, change.

This change in the inside of the furnace, for example, the change in the slag characteristics, is calculated based on the time series change pattern of the sound inside the furnace during blowing, which is measured by the sound collector 4 installed near the furnace mouth. The carbon concentration and temperature in the molten steel are observed by sampling the molten steel directly with the sublance 5. Information obtained in this way, such as sensor signals, which can be directly or indirectly obtained in relation to the converter equipment, is input to the reaction characteristic value output device 7 forming a part of the blowing control device 6. It

This reaction characteristic value output device 7 is installed independently for each target characteristic reaction such as decarburization, deMn removal, and temperature rise. In the example of FIG. 4, 7-1 to 7- Four
4 sets of reaction characteristic value output devices are provided.

The types of information input to the reaction characteristic value output devices 7 are, for example, as main raw material information,
The weight of molten steel, the weight of scrap, the weight of pig iron, the weight ratio of molten pig iron, etc., and the auxiliary raw material information include the amount of quicklime, iron ore, graphites, fluorite, etc. Indicates the amount of oxygen supplied from the lance during blowing and the tuyere of the bottom of the furnace, the oxygen supply rate, the height of the blowing lance, the timing of feeding the auxiliary materials, etc. The direct information measured during blowing is the molten steel by the sublance. There are carbon concentration, molten steel temperature, molten steel composition by spectrum analysis at oxygen hot point, etc., and indirect information measured during blowing includes exhaust gas flow rate, exhaust gas concentration or in-reactor acoustic value, furnace vibration value, lance vibration value, There are alternative measurement values such as the amount of the furnace mouth frame (flame), and as the converter characteristic information, there are characteristic values representative of the converter characteristics such as the progress of melting of refractory at the end of the previous blowing.

A specific structure of the reaction characteristic value output device 7,
That is, the neural network is as shown in FIG. Each unit 15-28 consists of a linear or non-linear function allowing a plurality of input data, and the input / output coupling of these units is a linear or non-linear coupling multiplied by a constant coefficient.

First, the above-mentioned main raw material information, auxiliary raw material information,
All input data such as manipulated variable calculation value information before blowing, direct measurement information during blowing, indirect measurement information during blowing, and converter characteristic information are input to the correlation determining means 10 and the above-mentioned respective targets For each characteristic reaction to be performed, for example, data having a high correlation with the past actual value of the reaction characteristic value is selected for each of temperature rise, decarburization, and deMn removal. For example, when the reaction characteristics are Mn-free, the in-reactor acoustics, lance height, and oxygen supply amount that are highly correlated with the intermediate and blow-off Mn are selected from the input data.

The input data selected by the correlation determining means 10 are the units 15 to 15 of the data number determining means 11.
It is output to 17, and the variation range of the input data is automatically recognized and normalized, and the number of the data is determined. For example, the selected data are normalized in consideration of the variation in the input data due to the difference in the blowing process.

Each unit 1 of the data number determining means 11
The output data of 5 to 17 are the respective units 18 to 2 of the combination creating means 12 which executes all combinations of data.
It becomes the input data of 4. The combination creating means 12 executes a combination of input data having a high correlation with a target characteristic, for example, a de-Mn reaction characteristic, and when there are three input data, seven kinds of combinations from one to all combinations are executed.

The units 25 to 27 of the selecting means 13 receive the output data of the combination creating means 12 as an input, and the unit of the number of combination data of the input data in the data number determining means 11, for example, the unit in the data number determining means 11. Number 3
If the number is 3, the combination of input data has 1 unit, 2 units, and 3 units, and for each unit of the number of each combination, the data having the highest correlation with the reaction characteristic value is selected, and the reaction characteristic is selected. It is output to the value calculation means 14.

The unit 28 of the reaction characteristic value calculating means 14
Is the output of the selecting means 13, that is, the combination data is input data, and calculates the reaction characteristic value of, for example, de-Mn by linear combination thereof.

The reaction characteristic values calculated and output by the reaction characteristic value output device 7 are as follows:
Composition, temperature of slag, composition, properties, total Fe (total iron oxide in slag), etc.

As shown in FIG. 4, the control operation amount output device 8 which constitutes a part of the converter blowing control device 6 has a secondary component such as quick lime based on the target molten steel composition, temperature, etc. at the end of the blowing. The control operation amount such as the weight and composition of the raw materials and the blowing lance 2 and the amount of oxygen supplied from the furnace tuyere 3 is calculated and controlled.

As the control manipulated variable output value calculated and output by the control manipulated variable output device 8, the oxygen supply amount from the lance during blowing and the tuyere of the bottom of the furnace, the oxygen supply rate, the lance height, the auxiliary raw material Input timing, auxiliary raw material input amount, alloy iron input time, alloy iron input amount, etc.

The algorithm for calculating the control manipulated variable is fuzzy inference, and its membership function is a neural network composed of a combination of units of a linear function such as a multiple regression function and a non-linear function such as a sigmoid function. Is built by.

The specific configuration of the control operation amount output device 8, that is, the neural network is empirically obtained from a past reaction characteristic value record by a skilled operator, as shown in FIG.
In order to formulate the blowing control judgment function performed during blowing with fuzzy rules, it is composed of antecedent part and consequent part of fuzzy reasoning.

The antecedent part of the fuzzy inference is that the reaction in the converter is divided into molten steel and slag, and the reaction characteristic value which is the output from each reaction characteristic value output device 7 is used as input data. From the correlation with the actual operation amount, the operation amount correlation determining unit 29 that automatically selects a highly correlated one, and the data selected by the operation amount correlation determining unit 29 are extracted and the normalization of the input data is input. Data number determining means 30 that is automatically performed each time data is updated, and the data number determining means 3
0, a combination creating means 31 including units 37 to 43 each receiving a combination of outputs from the respective units 34 to 36, an output of the combination creating means 31, and a control as a teacher signal in a so-called neural network. The selection means 32 is configured to automatically calculate the correlation with the actual operation amount each time the output data from the combination creating means 31 is changed, and increase the operation speed in the neural network.

The consequent part of the fuzzy inference uses the output data from the selecting means 32 as an input, and based on the target components of the molten steel at the end of the blowing, the temperature, etc., the control operation amount such as the amount of acid supply, the amount of quick lime input, etc. Is constituted by the control operation amount calculation means 33 for calculating and outputting

The neural network is composed of a linear or non-linear function in which each unit permits a plurality of input data, and the input / output coupling of these units is a linear or non-linear coupling multiplied by a constant coefficient.

First, using the output data of the reaction characteristic value output device 7 as input data, all of them are manipulated variable correlation determining means 8
, And select the data that has a high correlation with the past actual control operation amount. For example, when the manipulated variable is the Mn ore input amount, the in-reactor sound, the blowing lance height, and the oxygen supply amount, which have a high correlation with the intermediate and blow-stop Mn, are selected from the input data.

The selected input data is output to each of the units 34 to 36 of the data number determining means 30, and the variation range of the input data is automatically recognized and normalized. Here, the selected data are normalized in consideration of the variation in the input data due to the difference in the blowing process.

Each unit 34 of the data number determining means 30
The output data of 36 becomes the input data of each unit 37 to 43 of the combination creating means 31 which creates all combinations of data. The combination creating means 31 executes a combination of input data having a high correlation with the target characteristic, and in the case of three input data, seven kinds of combinations from one to all combinations are executed.

The units 44 to 46 of the selecting means 32 receive the output data of the combination creating means 31 as an input, and have a unit of the number of combination data of the input data in the combination creating means 31,
For example, if the number of units in the data number determination means 30 is 3, the input data combination has three units of 1, 2, and 3, and the past control operation amount record for each unit of the number of combinations. Control operation amount calculation means 3 by selecting data having a high correlation with, for example, a combination having a high correlation with the Mn ore input amount.
Output to 3.

The unit 47 of the control operation amount calculation means 33
Calculates the control operation amount by using the output of the selection means 32 as input data. For example, based on the combination data having a high correlation with the above-mentioned Mn ore input amount, de-M
The operation amount is calculated in consideration of the reaction characteristic of n.

As shown in FIG. 4, the coefficient learning device 9 forming a part of the converter blowing control device 6 includes the reaction characteristic value output device 7 and the control manipulated variable output device 8 for each interlayer in each neural network. Based on the difference between the molten steel after smelting, the composition of slag, the actual temperature and the calculated value of the reaction characteristic value, the unit coupling coefficient of is reduced by a learning method such as a back propagation method. The calculation is performed to correct the coupling coefficient of each unit of the reaction characteristic value output device 7 and the control operation amount output device 8.

Thus, the converter blowing control device 6 of the present invention
Calculates the weight and composition of secondary raw materials such as quicklime, and the amount of oxygen supplied from the blowing lance 2 and furnace tuyere 3 based on various sensor information, and calculates the composition, properties, temperature, etc. of molten steel and slag. It controls the situation inside the furnace.

By such control, the judgment based on the empirical knowledge for estimating and controlling the composition, properties, and temperature of the molten steel and slag, which had been performed by the blowing expert, was constructed by the fuzzy neural network. By automating by the smelting control device 6, variations due to the operator can be eliminated, and the composition and temperature of the molten steel can be set to the target values with high accuracy at the end of the smelting.

The control method according to the present invention and the conventional control method and sublance using indirect information such as in-reactor acoustics,
You may combine with the control method using direct information, such as a spectrum analyzer. That is, the conventional control method is used in the range where pattern recognition by the acoustic in the furnace is possible, the sublance, and the range where the spectrum analysis accuracy is sufficient. A method that implements the control method according to the invention can be expected to have favorable effects.

In the fuzzy inference, in addition to the method described above, a method in which the membership function in the antecedent part is replaced by a linear function, or a method in which the membership function in the antecedent part is replaced by a linear function, the antecedent part and There is a method in which the membership function in the consequent part is replaced with a linear function, and any fuzzy reasoning may be used.

In the neural network, in addition to the method described above, a method of increasing or decreasing the number of intermediate layers, a method of increasing or decreasing the number of units of each layer, and a nonlinear function of each unit other than the sigmoid function are used. There is a method of replacing with a function, and any neural network may be used.

[0046]

[Effect] FIG. 5 shows M in molten steel at the end of blowing in the embodiment.
Although the correspondence between the actual value and the calculated value of the n concentration is shown, the actual value and the calculated value of the Mn concentration in the molten steel at the end of the blowing are calculated by the blowing control by the device of the present invention during the blowing of the converter. Has been obtained with high accuracy.

[Brief description of drawings]

FIG. 1 is a block diagram showing an outline of a converter equipment and a converter blowing control apparatus according to an embodiment of the present invention.

FIG. 2 is a block diagram showing details of a reaction characteristic output device.

FIG. 3 is a block diagram showing details of a control operation amount output device.

FIG. 4 is a block diagram showing an outline of a converter blowing control device.

FIG. 5 is a graph showing the correspondence between actual values and calculated values of Mn concentration in molten steel at the end of blowing.

[Explanation of symbols]

1: Converter 2: Blowing lance 3: Blowing bottom tuyer 4: Sound collecting device 5: Sublance 6: Converter blowing control device 7: Reaction characteristic output device 8: Control operation amount output device 9: Coefficient learning device 10 : Reaction characteristic correlation judging means 11: Data number determining means 12: Combination creating means 13: Selection means 14: Reaction characteristic value calculating means 15-28: Each unit in neural network 29: Manipulation amount correlation judging means 30: Data number determining means Means 31: Combination creating means 32: Selection means 33: Control operation amount computing means 34 to 47: Each unit in the neural network

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Yoshiyuki Fukuda 5-3 Tokai-cho, Tokai City Nippon Steel Corporation Nagoya Steel Works (72) Inventor Yoshiki Uchikawa 468 Miyukiyama, Tenpaku-ku, Nagoya ( 72) Inventor Takeshi Furuhashi 301-Yokomachi, Tenpaku-ku, Nagoya 6-202 Tenpaku House

Claims (4)

[Claims] 1. Various target characteristics in a converter blowing control device for ensuring a predetermined molten steel composition and molten steel temperature based on information that can be obtained directly or indirectly throughout the refining process in the converter equipment. Automatically select the optimum input data to perform the setting control to calculate the reaction characteristic value during blowing, install the reaction characteristic value output device configured by the neural network for each reaction characteristic of each target characteristic, Control that outputs control operation amount by fuzzy reasoning constructed by neural network based on membership function that formulates blowing control judgment function of skilled operator by fuzzy rule from reaction characteristic value output from reaction characteristic value output device The coupling coefficient between the units in the neural network of the operation amount output device, the reaction characteristic value output device and the control operation amount output device is calculated from the blowing result. A converter blowing control device having a coefficient learning device for correction. 2. A main raw material information, auxiliary raw material information, operation amount calculation value information before blowing, which is directly or indirectly obtainable by the above-mentioned reaction characteristic value output device over the entire period of refining in a converter equipment, blowing Using the measured value information indicating the progress and the converter characteristic information as input data, the reaction characteristic correlation determining means for selecting data having a high correlation between each input data and the reaction characteristic value actual results, and the reaction characteristic correlation determining means are selected by the reaction characteristic correlation determining means. From the data number determining means including a unit for extracting the data, normalizing the data for each update and determining the number, and a unit for inputting the mutual combinations of the data output from the data number determining means. And a selecting means for calculating and selecting data having a high correlation between the output data value of the combination creating means and the actual result of the reaction characteristic value each time the output data is changed, and 2. The converter blowing control apparatus according to claim 1, comprising a neural network comprising a reaction characteristic value calculating means for calculating and outputting a reaction characteristic value based on output data from the selecting means. 3. The control operation amount output device formulates a blowing control judgment function, which is performed by a skilled operator empirically from past reaction records during blowing, based on a reaction characteristic value by a fuzzy rule. In order to do so, the antecedent part of the fuzzy inference uses the reaction characteristic value which is the output from each of the reaction characteristic value output devices as input data, and selects the data having a high correlation between each input data and the actual control operation amount. Output from the quantity correlation determining means, the data quantity determining means for extracting the data selected by the operation quantity correlation determining means, normalizing the data for each update, and determining the number thereof, and the data quantity determining means A combination creating means consisting of units each of which inputs a combination of the respective data, and data having a high correlation between the output data value of the combination creating means and the actual result of the control operation amount,
The fuzzy reasoning consequent unit calculates and outputs a control operation amount as a target characteristic based on the output data from the selecting unit. 3. The converter blowing control apparatus according to claim 1 or 2, which is configured to perform fuzzy inference by a neural network composed of control operation amount computing means. 4. The coefficient learning device calculates a unit coupling coefficient between each layer in each neural network of each reaction characteristic value output device and control operation amount output device as an actual value of reaction characteristic after completion of blowing. 2. The method according to claim 1, comprising arithmetic means for performing an operation to bring a difference from the value closer to zero and correction means for correcting each unit coupling coefficient based on the operation result.
The converter blowing control device according to 2 or 3.