JP2000144229A - Method for predicting slopping in converter and device therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To predict the timing developing slopping in a high accuracy with a simple adjustment without using many parameters. SOLUTION: Various informations which can catch over the whole stage of refining in converter equipment, are collected in a fixed period with a data collection treating part 2. A slopping-developing state value according to the waving state of exhaust gas flow rate, a slopping developing state value according to the waving state of decarburize-oxygen efficiency and a slopping- developing state value according to molten iron condition and operating condition, are calculated from the collected information. The slopping-developing probability and the slopping scale, are calculated with a neural network from the calculated each slopping-developing state value and auxiliary raw material charging quantity and blowing progressing degree to predict the slopping.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for predicting slopping in a converter, and more particularly to an improvement in the accuracy of slopping prediction.

[0002]

2. Description of the Related Art Decarburization reaction product CO in a converter
When the amount of generated gas is unstable, the decarburization reaction locally and abruptly proceeds in the converter, and slopping occurs in which molten steel and refining slag are ejected out of the converter. A method of predicting slopping in this converter is disclosed in, for example, Japanese Patent Application Laid-Open No. 5-222430.
No. in the official gazette. The method of predicting slopping disclosed in Japanese Patent Application Laid-Open No. 5-222430 discloses a method of classifying the flow rate waveform of exhaust gas generated by blowing into n types of waveform patterns according to the shape thereof, and thus classifying the waveform patterns. Is evaluated in n stages according to the degree of the possibility of slopping, the whole blowing period is divided into a plurality of stages, and the decarbonation efficiency at that time and the empirical reference value are compared for each stage, The possibility of slopping is evaluated for each stage in n stages, and the possibility of slopping is evaluated in n stages by knowledge processing from the type of hot metal, hot metal temperature, hot metal conditions such as hot metal components,
Weighting is applied to the exhaust gas flow rate waveform and the evaluation value of the possibility of slopping by decarbonation efficiency and knowledge processing in accordance with the progress of blowing, and the weighted evaluation values are added together to determine the possibility of slopping. To predict.

[0003]

However, in order to accurately predict the occurrence timing of the slopping by the above method, it is necessary to create a huge parameter table. Also, in the converter process, parameter adjustments were made multiple times during one furnace charge due to the adhesion of metal to the furnace opening, wear of the furnace body bricks, and changes in secondary combustion characteristics due to differences in lance height and acid supply flow rate. The adjustment work requires a great deal of labor each time. Furthermore, since the adjustment work is created based on the empirical reference value, adjustment experience is indispensable and is a complicated work.

SUMMARY OF THE INVENTION It is an object of the present invention to improve such disadvantages and to provide a method and an apparatus for predicting slopping which can accurately predict the timing of occurrence of slopping by simple adjustment without using many parameters. Is what you do.

[0005]

SUMMARY OF THE INVENTION A converter slopping prediction method according to the present invention is based on information available over the entire period of refining in a converter facility. Calculate the slopping occurrence state value based on the state value and the waveform state of the decarbonation efficiency, and the slopping occurrence state value based on the hot metal condition and the operating condition, and calculate the calculated slopping occurrence state value, auxiliary material input amount and blowing progress degree And a neural network is used to calculate the slopping occurrence probability and the slopping scale.

A converter slopping prediction apparatus according to the present invention comprises a data collection processing unit, a slopping possibility calculating unit based on a waveform state of an exhaust gas flow rate, and a slopping possibility calculating unit based on a decarbonation efficiency waveform. It has a unit for calculating the possibility of slopping due to hot metal conditions and operating conditions, and a unit for calculating the possibility of slopping and the estimated value of the slopping scale.The data collection and processing unit collects information that can be obtained over the entire period of refining in the converter equipment. Collecting and calculating the possibility of slopping based on the waveform state of the exhaust gas flow rate, the current data of the exhaust gas flow rate data collected by the data collection processing unit and the data up to a certain time before are input, and the waveform of the exhaust gas flow rate by the neural network Calculate the value of slopping occurrence by the state and calculate the possibility of slopping by the waveform of decarbonation efficiency The collected exhaust gas flow rate and the CO concentration and the exhaust gas in the exhaust gas CO ₂ concentration and the blown oxygen amount and the furnace opening blow coefficient data collection period turned auxiliary materials C content in the data collecting section and put auxiliary materials in oxygen content And the neural network calculates the slipping occurrence state value based on the waveform state of the decarbonation efficiency, and the slapping possibility calculation section based on the hot metal conditions and operating conditions calculates the hot metal temperature and hot metal collected by the data collection processing section. Input Si concentration, furnace order, miscellaneous raw materials, total CaO, bottom blown gas flow rate and molten metal level relative lance height, calculate the slipping occurrence state value by hot metal condition and operating condition by neural network,
The calculation unit of the possibility of slopping and the estimated value of the slopping scale inputs the above-mentioned slapping occurrence state value, the amount of the auxiliary material input, and the progress of blowing, and calculates the slopping occurrence probability and the slopping scale by the neural network. It is characterized in that it is calculated.

It is preferable to automatically learn and reset the weighting parameters of the neural network for calculating the above-mentioned slopping occurrence probability and the size of the slopping so that the estimation error satisfies the reference or less from the predicted slopping value and the actual slopping occurrence.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A converter slopping prediction apparatus according to the present invention comprises a data collection processing unit, a slopping possibility calculating unit based on a waveform state of an exhaust gas flow rate, and a slopping possibility calculating unit based on a decarbonation efficiency waveform. And a calculation unit for calculating the possibility of occurrence of slopping and the estimated value of the slopping scale based on hot metal conditions and operating conditions. The data collection processing unit collects various types of information that can be obtained over a whole period of refining in the converter equipment at a fixed period, for example, a period of 2 seconds. The slopping possibility calculation unit based on the waveform state of the exhaust gas flow rate calculates a predetermined time, for example, 6 hours from the current data in the exhaust gas flow rate data from the data collection processing unit.
Thirty data of the data up to 0 seconds before are input, and the slapping occurrence state value is calculated from the input thirty data based on the waveform state of the exhaust gas flow rate, and the slopping occurrence probability and the estimated value of the slopping scale are calculated. Output to the section. Slopping likelihood calculation unit by the waveform of the decarboxylation oxygen efficiency and exhaust gas flow rate and the exhaust gas in the CO concentration and the exhaust gas in CO ₂ concentration and blown oxygen amount and the furnace opening blow coefficient and data collection period from the data collecting section turned adjuncts Input the C content and the oxygen content in the input raw materials, calculate the decarbonation efficiency, calculate the slopping occurrence state value based on the waveform state of the decarbonation efficiency, and estimate the slopping occurrence possibility and the slopping scale. Is output to the calculation unit. The slopping possibility calculation part by the hot metal condition and the operating condition inputs the hot metal temperature, the hot metal Si concentration, the furnace order, the miscellaneous material, the total CaO, the bottom blown gas flow rate and the metal surface relative lance height from the data collection processing part, A slopping occurrence state value based on these conditions is calculated and output to a calculating unit for estimating a possibility of slopping and an estimated value of the slopping magnitude.
The calculation unit for the slopping probability and the estimated value of the slapping scale calculates the slopping occurrence probability and the slapping scale based on the input slapping occurrence state values, the input amount of the auxiliary material in the 2-second cycle, and the progress of the blowing. Then, the information is displayed on the display device and stored in the storage device. This process is performed at a constant cycle during blowing, for example, at a cycle of 2 seconds, and the occurrence of successive dropping is predicted during blowing.

[0009]

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention. As shown in the figure, a converter slopping prediction apparatus 1 includes a data collection processing unit 2, a slopping possibility calculating unit 3 based on a waveform state of an exhaust gas flow rate, and a slopping possibility calculating based on a decarbonation efficiency waveform. A calculating unit 5 for calculating a possibility of occurrence of slopping and a slapping scale based on hot metal conditions and operating conditions; a calculating unit 6 for estimating the estimated value of the slopping scale; a neural network learning unit 7; a display device 8; It has a storage device 10.

As shown in FIG. 2, the data collection processing unit 2 includes various kinds of information that can be obtained over the entire period of refining in a converter equipment 14 having a converter 11, a lance 12, an exhaust gas duct 13, and the like, for example, the weighing of auxiliary raw materials. The amount of auxiliary raw materials measured by the machine 15, the amount of injected oxygen measured by the lance oxygen flow meter 16, the CO concentration in the exhaust gas and the CO ₂ concentration in the exhaust gas detected by the exhaust gas component analyzer 17, Various data such as the bottom blown gas flow rate measured by the bottom blown gas flow meter 18, the exhaust gas flow rate measured by the exhaust gas flow meter 19, hot metal conditions, and operating conditions are collected and processed.

[0011] As shown in FIG.
And an intermediate layer 32 and an output layer 33. The neural network consists of a neural network in which learning progresses. The current data among the exhaust gas flow rate data generated in refining collected by the data collection processing unit 2 at a constant period, for example, a period of 2 seconds. After normalizing 30 data of the data (i-29) from i to 60 seconds before, the neural network adjusted so that the current data i becomes more important causes the dropping of the waveform due to the waveform state of the exhaust gas flow rate. Calculate and output status values. Here, the input layer 31, the intermediate layer 32, and the output layer 33 have a three-layer structure of 30 × 4 × 1.
2 may have any number of layers, and the number of units in each layer may be increased or decreased.

The unit 4 for calculating the possibility of occurrence of slopping based on the waveform of the decarbonation efficiency also comprises a neural network having an input layer, an intermediate layer and an output layer. However, parameters such as weight coefficients to be used are different. And exhaust gas discharge C
The amount Gc and, for example, the exhaust gas flow rate Vg collected in a cycle of 2 seconds.
And CO concentration in exhaust gas M1 and CO _{2 in} exhaust gas Concentration M2, injected oxygen amount V, furnace port injection coefficient K, data collection period T, input auxiliary raw material C content Qc, and input auxiliary raw material oxygen content Qo
Thus, the decarbonation efficiency is defined as follows. Exhaust gas emission C amount Gc = ｛Vg × (M1 + M2) / K｝ T Decarbonation efficiency = (Qc-Gc) / (V + Qo) Here, the furnace outlet coefficient is a coefficient obtained by mass balance calculation of nitrogen or argon, and input. The auxiliary material C content is obtained by totalizing the input amounts obtained by the auxiliary material input weighing machine 15 in, for example, a 2-second cycle for each brand, and totaling the weight of each C component.

The calculation of the decarboxylation efficiency is performed in a cycle of 2 seconds, and the data from the present time to 60 seconds before is normalized in the same manner as the slopping occurrence state value based on the waveform state of the exhaust gas flow rate Vg. Is calculated by the neural network adjusted so that is more important.

As shown in FIG. 4, a slapping occurrence possibility calculating section 5 based on hot metal conditions and operating conditions is also provided in the input layer 51.
, An intermediate layer 52 and an output layer 53. Although the likelihood of slipping can be related to some extent by the conditions of the hot metal temperature, the hot metal Si concentration, and the furnace age, other factors are also included in order to increase the accuracy of determining the possibility of the occurrence of slopping. As a result of examining the correlation with the frequency of occurrence of dropping, as shown in FIG. 5, as the total CaO, which is the content of CaO in the auxiliary material charged into the furnace, increased, the frequency of occurrence of dropping increased, and as shown in FIG. In addition, when the bottom gas flow rate was small, the frequency of occurrence of slopping increased. Therefore, hot metal temperature and hot metal S
The relative lance height, which is the distance between the molten steel surface and the lance, which affects the total CaO and bottom blown gas flow rate and the reaction rate in the furnace, which is highly correlated with the i concentration, furnace age and miscellaneous materials, as well as the frequency of slopping. The conditions are added, and these conditions are input to the input layer 51 at a cycle of 2 seconds, and the value of the slopping occurrence state is calculated by the learning rule. Here, the input layer 51 and the intermediate layer 52
And the output layer 53, the intermediate layer 52 may have any number of layers.

As shown in FIG. 7, the calculation unit 6 for calculating the possibility of occurrence of slapping and the estimated value of the slapping scale also includes an input layer 61.
, A neural network having an intermediate layer 62 and an output layer 63. The slapping probability calculation unit 3 based on the waveform state of the exhaust gas flow rate, the slopping probability calculation unit 4 based on the decarbonation efficiency waveform, and the slapping probability calculation unit 5 based on the hot metal conditions and operating conditions were calculated. For example, the input layer 6 indicates the slopping occurrence state value of the 2-second cycle, the auxiliary material input amount of the 2-second cycle, and the blowing progress degree which is the integrated amount of blown oxygen up to the present / the planned integrated amount of blowoff.
1 and calculates the dropping probability and the dropping magnitude by a neural network having an automatic learning function, and outputs the dropping probability in, for example, five steps from “1” to “5”. This output is displayed on the display device 8. Calculating section 6 for the possibility of occurrence of the slopping and the estimated value of the slopping scale
Uses a recurrent type that holds the output of the hidden layer 62 inside the neural network and facilitates learning with time-series data.

The neural network learning unit 7 calculates a slapping occurrence probability and a slopping scale estimation value.
The learning unit has the same structure as the neural network used in the calculating unit 6, but is configured independently of the other parts, and is necessary for calculating the neural network parameters without disturbing the online calculation. A huge amount of calculations is possible. The result data input unit 9 inputs the occurrence timing and its scale by the operation of the operator every time the slipping occurs during the actual operation. Storage device 10
Stores the predicted slopping value and the input actual slopping value, and also changes the acid flow rate, lance height, upper hood height, and anti-sloping material that are not in the blowing pattern collected after the blowing process. The timing at which an operation action such as input is performed is stored.

The data collection processing unit 2 of the slopping prediction device 1 configured as described above collects various information that can be obtained over a whole period of refining in the converter equipment 14 at a fixed period, for example, a period of 2 seconds. The slapping possibility calculation unit 3 based on the waveform state of the exhaust gas flow rate calculates the current data i from the exhaust gas flow rate data 60 from the data collection processing unit 2.
Thirty data of the data (i-29) up to two seconds before are input, and a slopping occurrence state value is calculated from the thirty data based on the waveform state of the exhaust gas flow rate, and a possibility of the slopping occurrence and an estimated value of the slopping scale are obtained. Is output to the calculation unit 6. Slopping likelihood calculator 4 exhaust gas flow rate Vg and CO concentrations M1 and in the exhaust gas in the exhaust gas CO ₂ concentration M2 and blown oxygen amount V from the data acquisition processing unit 2 by the waveform of the decarboxylation oxygen efficiency
, Furnace mouth blowing coefficient K, data collection period T, and input auxiliary material C
Inputting the content Qc and the oxygen content Qo in the input auxiliary material, calculating the decarbonation efficiency, calculating the slopping occurrence state value based on the waveform state of the decarbonation efficiency, and estimating the possibility of the slopping and the magnitude of the slopping. Is output to the calculation unit 6. The slapping possibility calculation unit 5 based on the hot metal conditions and operating conditions also inputs the hot metal temperature, hot metal Si concentration, furnace order, miscellaneous raw materials, total CaO, bottom blown gas flow rate, and metal surface relative lance height from the data collection processing unit 2. Then, a slopping occurrence state value based on these conditions is calculated and output to the calculating unit 6 for the slapping occurrence probability and the slopping scale estimated value. The calculator 6 for calculating the possibility of occurrence of the slapping and the estimated value of the amount of the slapping includes the slapping occurrence state values inputted from the respective slapping possibility calculating units 3, 4, and 5, the input amount of the auxiliary material in the 2-second cycle, and the blowing. Based on the degree of progress, a slopping occurrence probability and the magnitude of the slopping are calculated, output in five stages, displayed on the display device 8, and stored in the storage device 10. This process is performed at a constant cycle during blowing, for example, at a cycle of 2 seconds, and the occurrence of successive dropping is predicted during blowing.

If the accuracy of the slopping prediction is degraded while predicting the occurrence of the slopping as described above, the neural network learning section 7 reports the results of the detection omission, overdetection, and successful prediction to the teacher data. Then, the weight parameter of the neural network whose estimation error satisfies the reference or less is recalculated, and the calculated weight parameter is set in the calculation unit 6 for the estimation of the possibility of the occurrence of the slopping and the magnitude of the slapping. This learning method may be any of the back vacation method, the virtual impedance learning method, a combination method thereof, or another method.

FIG. 8 shows the slopping prediction estimation accuracy thus predicted and the result predicted by the conventional method. As shown in FIG. 8 (a), the prediction success rate according to this embodiment is about 97%, which is much higher than the prediction success rate 91% according to the conventional method shown in FIG. 8 (b). And the slopping could be predicted with high accuracy. Therefore, it is possible to reliably prevent the slopping and to perform the stable operation of the converter.

[0020]

As described above, the present invention collects various types of information that can be obtained over the entire period of refining in a converter facility, and, based on the collected information, the value of the state of occurrence of slopping due to the waveform state of the exhaust gas flow rate and the amount of decarbonation. Calculate the slapping occurrence state value due to the efficiency waveform state and the slopping occurrence state value due to the hot metal conditions and operating conditions, and use the neural network to perform the slopping based on the calculated slopping occurrence state values, the auxiliary material input amount, and the blowing progress. Since the occurrence probability and the magnitude of the slopping are calculated, the slapping can be predicted with high accuracy. Therefore, by performing the optimal control for suppressing the slopping using the prediction data, the prevention of the slopping can be surely prevented and the stable converter operation can be performed.

In addition, the weighting parameters of the neural network for calculating the slopping occurrence probability and the slopping scale are automatically learned and reset from the predicted slopping value and the actual slopping occurrence so that the estimation error satisfies the standard or less. Thereby, the slopping can be stably predicted with high accuracy.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of the present invention.

FIG. 2 is a configuration diagram of a converter facility.

FIG. 3 is a configuration diagram of a slopping possibility calculating unit based on a waveform state of an exhaust gas flow rate.

FIG. 4 is a configuration diagram of a slopping possibility calculation unit based on hot metal conditions and operating conditions.

FIG. 5 is a graph showing the frequency of occurrence of sloping of total CaO.

FIG. 6 is a graph showing the frequency of occurrence of slopping of the flow rate of bottom-blown gas.

FIG. 7 is a configuration diagram of a calculator for calculating a possibility of occurrence of a slopping and an estimated value of a slopping scale.

FIG. 8 is a distribution diagram of a prediction success rate of the embodiment and the conventional example.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 Slopping prediction apparatus 2 Data collection processing part 3 Sloping occurrence possibility calculation part by the waveform state of exhaust gas flow rate 4 Slopping occurrence possibility calculation part by decarbonation efficiency waveform 5 Slopping occurrence by hot metal conditions and operating conditions Calculating section 6 Calculating section for the possibility of occurrence of slopping and the estimated value of the dropping of the scale 7 Neural network learning section 8 Display device 9 Actual data input section 10 Storage device 11 Converter 12 Lance 13 Exhaust gas duct 14 Converter equipment 15 Secondary material input Weighing machine 16 Lance oxygen flow meter 17 Exhaust gas component analyzer 18 Bottom blown gas flow meter 19 Exhaust gas flow meter

Claims (4)

[Claims] 1. Based on information available over the entire period of refining in a converter facility, a neural network is used to determine a slipping state value based on a waveform state of an exhaust gas flow rate, a slipping state value based on a waveform state of decarbonation efficiency, and Calculate the slopping occurrence state value by the hot metal condition and the operating condition, and calculate the slopping occurrence probability and the slopping scale by the neural network from the calculated slopping occurrence state value, the input amount of the auxiliary material, and the degree of progress of blowing. Converter slopping prediction method characterized by the above-mentioned. 2. The neural network for calculating the above-mentioned slapping occurrence probability and the magnitude of the slopping is automatically learned and reset from the predicted slopping value and the actual slopping occurrence so that the estimation error satisfies the standard or less. The converter slopping prediction method according to claim 1. 3. A data collection processing unit, a slopping possibility calculating unit based on a waveform state of an exhaust gas flow rate, a slapping possibility calculating unit based on a decarbonation efficiency waveform, and a slopping possibility depending on hot metal conditions and operating conditions. It has a calculation section and a calculation section for the possibility of occurrence of slopping and the estimated value of the slopping scale.The data collection processing section collects information that can be obtained over the entire period of refining in the converter equipment, and performs slopping based on the waveform state of the exhaust gas flow rate. The likelihood calculation unit inputs the current data of the exhaust gas flow rate data collected by the data collection processing unit and the data up to a certain time before, calculates a slipping occurrence state value by the waveform state of the exhaust gas flow rate by a neural network, The calculation unit for the possibility of slopping based on the waveform of decarbonation efficiency calculates the flow rate of exhaust gas collected by the data collection processing unit. And the CO concentration in the exhaust gas, the CO ₂ concentration in the exhaust gas, the amount of oxygen injected, the blowing coefficient of the furnace port, the data collection cycle, the C content of the input raw material and the oxygen content of the input raw material, and the decarbonation efficiency by the neural network The slapping occurrence state value according to the hot metal condition and the operating condition is calculated by the hot metal temperature and hot metal S collected by the data collection processing unit.
Input i-concentration, furnace order, miscellaneous raw materials, total CaO, bottom blown gas flow rate and molten metal level relative lance height, calculate the value of slipping occurrence state by hot metal condition and operating condition by neural network, And the calculation unit of the estimated value of the slopping scale inputs the respective slapping occurrence state values, the input amounts of the auxiliary materials, and the degree of the progress of the blowing, and calculates the slopping occurrence probability and the slapping scale by a neural network. Converter slopping prediction device. 4. A neural network learning unit for resetting the weighting parameters of the slopping occurrence probability and the slopping magnitude estimation value calculating unit from the predicted slopping value and the actual slopping occurrence so that the estimation error satisfies the standard or less. The converter slopping prediction device according to claim 3, comprising: