JPH02259006A - Automatic deciding system for pattern data in blast furnace - Google Patents

Abstract

PURPOSE:To execute pattern recognition and pattern decision as same as an operator by fetching the measured pattern from various sensors set at various positions in a blast furnace and using neural net after processing data. CONSTITUTION:Signals 50 from stave temp. sensors in height direction, temp. sensors with a sonde in radius direction and gas analysis sensor, etc., in the blast furnace 10, are inputted to automatic deciding system 20 for pattern data and processed with there. After that, it is shown that any output signal in the kinds of preset patterns equal to the deciding results, which the operator executes, is transmitted to a display device 41 and/or the other system 40. The automatic deciding system 20 for pattern data is constituted with the neural net by using signals from a data input device 21 for fetching the inputted signal 50, a data processor 22 for processing the data outputted from the device 21, such as noise processing, abnormal value deleting processing or normalizing processing and a data storage device 23 for storing the outputted signal in a memory and adjusting the timing.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention provides a blast furnace that automates the pattern discrimination ability of an operator by making use of all the input signals of a series of pattern element data such as stave temperature and sonde signals in the blast furnace. This invention relates to an automatic pattern data determination system.

[Conventional technology]

During blast furnace operation, operators estimate the situation inside the furnace based on various data and take actions. In this process, the temperature and gas components at multiple points in the radial direction are measured using a sonde inserted into the furnace, and this is evaluated as pattern data to estimate the gas flow distribution in the furnace, and to estimate the temperature of hot metal over time. As seen in the example of estimating the state of the heat level in a furnace from the pattern of changes, there are cases where pattern data is directly evaluated and judged, and this data is extremely important. However, these evaluations and judgments are made directly by the operator, which poses problems such as individual differences and intuitive nature that cannot be systematized.

Recently, artificial intelligence
With the spread of knowledge technology, attempts have also begun to apply knowledge engineering to blast furnaces. The application of knowledge engineering to blast furnaces seems to be convenient because it allows operators to use conventional operating techniques, including intuitive judgment, as is, but pattern data cannot be included in knowledge engineering. This is the actual situation.

On the other hand, as a method of evaluating and judging pattern data, there is a method of determining two to three pieces of representative data from among a plurality of pieces of data that form a pattern based on past operational experience, and evaluating and making judgments using these representative data. Technical literature: Classification of process conditions (Measurement and Control, Vol. 24. No. 8), based on pattern data over a certain period of time (
Attempts have been made to use the discriminant function method, which finds a mapping matrix that maximizes the ratio (between-group variance)/(within-group variance) and uses this to index data.

[Problem that the invention seeks to improve]

The method of evaluating and making judgments using representative data has the advantage of requiring less calculations because the number of representative data is generally small (2 to 3), but on the other hand, the disadvantage is that it is often difficult to find representative data (, Even if representative data is found by chance, if the representative data is missing or noise is mixed into the data, it may become impossible to make a judgment or result in an incorrect judgment, making it vulnerable to disturbances and noise. Since the position of the representative data often changes with changes in conditions, this method has a limited scope of application and cannot be highly reliable.

On the other hand, the latter method using a discriminant function finds the mapping matrix using a statistical method, which requires a lot of effort and ingenuity.
There are cases where a mapping matrix cannot be found, and even when a mapping matrix is found, in most cases it is an indicator that does not fit the senses of conventional operators, and it is not used in operations because conventional knowledge cannot be utilized. Alternatively, there is a problem in that a control system using the discriminant function method must be newly developed. Furthermore, since this mapping matrix generally changes with changes in operating conditions, applying this method not only imposes a heavy burden on creation, but also on subsequent follow-up.
It used to be that it eventually stopped being used unless something very good was made.

In the blast furnaces used in the steel industry, extremely complex chemical reactions and phenomena such as heat exchange continue at the same time, so statistical methods such as those described above are difficult to establish except with certain preconditions. The reality is that it is complicated and we have no choice but to rely on human senses. Therefore, it was a well-established theory that automation and systemization were not possible without taking into account the unique characteristics of blast furnaces.

The problem to be solved by the present invention is to develop a method that uses all of a series of pattern data to obtain the same results as the intuitive pattern recognition or pattern judgment that operators perform during operation, and to The aim is to realize the method as an automatic judgment system using the following.

[Means to solve the problem]

Recently, neural networks that imitate biological nervous systems have emerged as a new method for determining patterns. Its advantages are that simultaneous parallel processing is possible and self-learning is possible. In the future, it is expected that its advantages will be utilized to expand its application to various fields, but at present it requires a lot of effort and ingenuity to apply, so the promotion of basic research is awaited. The present invention has been made in order to solve the above-mentioned problems, and is less susceptible to the effects of data loss and noise (in order to make it less susceptible to data loss, noise, etc., it uses all of the pattern element data forming the pattern data, and the The purpose of the present invention is to provide an automatic judgment system for blast furnace pattern data using a neural network method that systemizes the pattern judgment method that is consistent with the conventional evaluation and judgment.

The means for achieving the above object are as follows, and the basic means are in (1) to (2).

■A system equipped with a pattern element data input means that takes in measurement pattern element data from sensors installed in the blast furnace at a predetermined timing, and a hierarchical neural network means that determines the measurement pattern element data into a predetermined pattern. is the means.

■A pattern element data input means that takes in measurement pattern element data from sensors installed in the blast furnace at a predetermined timing, a signal processing means that performs signal processing such as noise processing on the measurement pattern element data, and a signal processing means that performs signal processing such as noise processing on the measurement pattern element data. The system includes a hierarchical neural network that determines pattern element data into a predetermined pattern.

■A pattern element data input means that takes in measurement pattern element data from a sensor installed in the blast furnace at a predetermined timing; a signal processing means that performs signal processing such as noise processing on the measurement pattern element data; The system includes a storage means for temporarily storing measurement pattern element data, and a hierarchical neural network means for determining the stored pattern element data into a predetermined pattern.

Furthermore, there are the following practical means.

■ Including normalization processing in the signal processing means 03-layer neural net means is the best, and 4-layer neural network means is the best. ■ Including normalization processing in the input layer of the hierarchical neural network means. The input signal level is a normalized signal ■The number of units in the input layer of the hierarchical neural network means is the number of measured pattern element data plus one data for the average value signal of the measured pattern element data. - The number of units in the output layer of the hierarchical neural network means is the same as the number of patterns to be determined, and the output signal is normalized to a range of 0 to 1. - The sensor is connected to the wall of the blast furnace. A stave thermometer [Phase] The sensor is a sonde equipped with a sensor that measures at least one of the gas components or the temperature in the blast furnace ■Measurement pattern element data from the sensor is a pressure signal in the blast furnace [Operation] In order to automatically determine a series of pattern data such as stave temperature and sonde signals in a blast furnace, a neural network method may be applied in addition to the above-mentioned conventional techniques. This is because the neural network method has the same self-learning function as the human brain, so by applying it, it is possible to use all of a series of pattern data to output the same judgment results as the operator's judgment results and create an automatic judgment system. It is from.

Neural networks are classified into mutually connected types and hierarchical types based on the network structure, and the hierarchical type shown in Figure 6 is said to be suitable for pattern determination, and we chose this type as well. The hierarchical neural network shown in Figure 6a) is
It is composed of one input layer and output layer, and zero layers and multiple intermediate layers, and each unit of the intermediate layer and output layer is connected to each unit of the previous layer. Each unit has a detailed structure as shown in Figure 6b), and the output of each unit is the sum of the outputs of each unit in the previous layer multiplied by a weighting function ωij and the sum of the values for all units in the previous layer. is obtained and converted by a function having the relationship of the output characteristics shown in Fig. 6C). Therefore, when a value based on measurement data is input to each unit of the input layer, the above calculation is performed, and an output value is output from each unit of the output layer.In general, it is possible to determine which predetermined pattern the output value corresponds to. It will be decided.

Looking at recent research trends in neural network methods, it has come to be said that hierarchical methods are suitable for pattern determination, and pack propagation methods are effective for self-learning methods.

Therefore, an attempt was made to apply the above method to automatic determination of pattern data in a blast furnace as shown in FIGS. 1 and 2.

FIG. 1 is a system configuration diagram, in which signals 50 from a stave temperature sensor in the height direction in the blast furnace 10, a temperature sensor using a radial sonde, a gas analysis sensor, etc. are inputted to the pattern data automatic judgment system 20, and the pattern data is After being processed by the automatic judgment system 20, an output signal of one of the predetermined pattern types (for example, pattern B) that is equivalent to the judgment result conventionally performed by the operator is displayed on the display device 41 and/or other systems 40. (for example, an automatic control system for a blast furnace). The pattern data automatic determination system 20 includes a data input device 21 that takes in an input signal 50, a data processing device 22 that performs data processing such as noise processing, abnormal value deletion processing, or normalization processing on the output signal, and the output signal It is comprised of a data storage device 23 that stores the data in a memory and adjusts the timing, and a neural network 24 that performs hierarchical neural network processing on signals from the data storage device 23.

FIG. 2 is a hardware configuration diagram corresponding to the system configuration diagram shown in FIG. 1, in which neural network processing is realized by computer software. Reference numeral 61 in FIG. 2 is an interface device, which is comprised of a sampling device, an A/D converter, etc. for taking in data at a predetermined cycle. 62 is a CPU 1.63.6 for executing data processing and pattern judgment using a neural network.
4 is a ROM in which data processing and neural network programs and necessary parameters are written in advance.

65.66 is RA for temporarily storing pattern data
M, 67 is a CRT for displaying the determination results, and 68 is an interface device for connecting with an automatic control system that uses the results of the pattern determination device to determine the amount of action to be taken to the blast furnace.

The challenges when applying neural networks to automatic determination of blast furnace pattern data are summarized as follows.

■Structural design of neural network (number of layers, number of units in each layer) ■Signal processing of input data ■How to create data used for learning ■Means for realizing neural network Gas generated by blast furnace sonde, which will be explained in detail in the example below The above-mentioned problem was investigated through trial and error, using an example in which the utilization rate was applied to measurement data at eight points in the radial direction of the blast furnace.

First, for the first task, we selected a hierarchical neural network. Regarding the number of layers, the input layer is 1 layer and the output layer is 1 layer.
The layers and the intermediate layers can have 0 to 1.2, . . . layers. Although the number of units in the input layer depends on the signal processing method of input data, we discovered that by adopting the signal processing method described later, it is possible to make it the same as the number of data forming the pattern, making it possible to realize a simple network. Ta.

Furthermore, we focused on the characteristics of blast furnace pattern data, where the difference in absolute value level is also useful information for pattern determination, and calculated the number of units in the input layer as the number of data forming the pattern data (n).
It has been found that adding one value to (n-1) for n average value data not only can provide even better accuracy but is also convenient for learning the weighting coefficients between units in each layer. Further, the number of units in the output layer may be any number as long as the number of patterns to be determined can be identified, so various numbers can be considered. We adopted a format that seemed to be the simplest, assigning one output to each judged pattern, that is, the number of output layer units was the same as the number of judged patterns.

The number of intermediate layers is determined, in principle, by the characteristics of the data to be determined, and is considered to be at the stage of trial and error. The gas utilization rate data from this sonde shows that
Without an intermediate layer, it is not possible to make good judgments, so having one intermediate layer is appropriate and simple, and having two intermediate layers has the same accuracy as one layer, but it is simple and can be achieved using computer software. In this case, the calculation time will be longer. We found that when increasing the number of stages to 3 or more, there is no particular change in accuracy and sometimes it worsens, and when viewed individually, the number of dead routes (those with weight coefficients close to zero) only increases. Furthermore, we have confirmed that if this is achieved using computer software, the calculation time would be too long and learning the weighting coefficients between each unit would be too complex, making it practically difficult to implement. Therefore, we found that 3 layers with 1 layer in each layer is best, and 4 layers with 2 layers in the middle is also possible. Note that the number of units in the intermediate layer is complicated and will be described later.

Next, for the second task, we attempted to input eight points of data into each unit of the input layer using the following two methods.

■ Normalize the data and input it as an analog signal in the range of 0 to l. ■ Binarize the data and input it as a digital signal of 0.1. It took a long time to learn, and sometimes learning was difficult. It was confirmed that method (■) does not have the problem (■), so it was adopted. In the input signal processing method (method ①), the i-th numerical data forming the pattern data is
1) Use a method of normalizing within the range of 0 to 1 using formula.

In the equation (1), Input-i is the input to the neural network, the i-th data after normalization Yi is the i-th numerical data forming the pattern, Ymax-i and Ya+1
ni is the i-th maximum value or minimum value, and these are set in advance, for example, as the maximum value or minimum value determined from past data. Using this normalization method provides the following advantages.

■As it is only necessary to allocate one input layer unit to each piece of data that forms a pattern, it can be handled with a small-scale neural network.

■As the operating conditions of the blast furnace change and the sensor conditions change, the area of the appropriate pattern changes, so it becomes necessary to retrain the neural network.
By adjusting x-i and Ymin-i by an amount corresponding to the change in the appropriate pattern, the input to the neural network In
Since put-i can be kept the same, there is no need for relearning.

■Since normalization is performed within a preset range, sensitivity becomes large and stable.

On the other hand, the following can be said about how to create data used for learning.

Typical data corresponding to each of the 0 judgment patterns (data whose output is 0 for those other than output kaninit corresponding to each judgment pattern) is essential.

The more supplementary data corresponding to each of the 0 judgment patterns (the data in which the output value of the output kaninit corresponding to each judgment pattern is large and the other items have outputs other than zero), the better, but the more the more the better. It takes time to learn.

Therefore, first, after learning using typical data, we judge the actual pattern data, and then perform learning based on re-learning data in which the results that do not match the operator's judgment are added as supplementary learning data. repeated.

Further, the means for realizing the neural network may be hardware or software. Originally, in order to take advantage of the simultaneous parallel processing of neural networks, it should be implemented in hardware, but when it comes to almost the same parallel processing even if it is implemented in software, it is desirable to use software that can be implemented very easily. In this case, even if realized by software, the calculation time for the determination is on the order of 1/100 seconds or less, so it can be ignored in view of the time constant of the controlled object, so this method was selected. Furthermore, the advantage of implementing it with software is that it requires no adjustment, is easy to maintain, and is easy to change.

On the other hand, the number of units in the omitted middle layer explained above was varied from 5 to 12. Note that the number of units in the input layer is the data forming the pattern plus their average value, and the intermediate layer is one stage.
This was realized using software and simulated using actual data.

First, the convergence of learning was investigated by defining convergence as when the learning error in determining the weighting coefficients between units in each layer using the learning data was 0.05 or less, and the results were as shown in Table 1.

Table 1 When the number of units was 5, the calculation was discontinued because it did not converge even after 5000 calculations. Number of units is 7-12
It should be seen that there is no difference in the number of calculations until convergence. Next, assuming that the operator's judgment was correct, simulations were performed for cases in which the number of units in the intermediate layer was 7 to 12. As a result, the correct answer rate reached 98%, which was the highest in the case of 7 units. Furthermore, an accuracy rate of 98% is considered to be at a level that can be automated. However, there are signs in the data that the number of units in this intermediate layer may change depending on the target signal, the number of units in the input layer, the learning data, etc. Also,
Considering that some of the seven units are not working very well (the weight coefficients between units are small), it would be desirable to reflect the reliability and maintenance status of each measurement data.

The present invention has been made based on the above various findings, and is based on spatial patterns such as gas analysis values and temperature data of various sondes, which are typical pattern data in blast furnaces, and furnace wall stave temperature data, as well as furnace pressure. The above-mentioned means realize an automatic judgment system that targets time-series patterns such as.

The inventors have applied the present invention to energy management within the steel industry,
When I applied it to pattern data in the rolling process, I found that it could be applied as is, except for special cases such as those with a large number of measured data and short fluctuation cycles.

In addition, the method of determining the predetermined patterns and the number of types are determined based on many years of experience in blast furnace operation, know-how, furnace condition judgment data, etc., and are unique to each sensor data, and are particularly important in relation to blast furnace operation. It is also related to the method of action, and the basic pattern is based on typical examples of good, normal, and bad furnace conditions.

In the processing flow, the processing from 5TARTII to END17 shown in FIG. 3a) is repeatedly performed. When a start signal is received from a periodic signal or an interrupt signal,
5TART II, and then each measurement signal forming the pattern data is read. After inputting all data in step 12, data processing such as noise processing, abnormal value deletion processing, average value processing, and normalization processing is performed in step 13, and the results are stored in a file in step 14. Pull out the necessary data from those 14 files, give signals to each unit of the input layer, and use a hierarchical neural network to judge the pattern.
Do step 5. The result is output 16 and the subsequent processing (
For example, it can be displayed to notify operators, or it can be used in automatic control systems for blast furnaces.

FIG. 3b) shows the processing flow in pattern determination 15 in more detail. First, in step 151, necessary data is extracted from the file 14 and set in each unit in the input layer of the hierarchical neural network. Next, in step 152, as shown in FIG. 6 a) and b), the value of each unit in the input layer is multiplied by the weighting coefficient ωij, and the value of each unit in the intermediate layer is calculated as the cumulative value of the combined values. It is output via a function with input/output characteristics as shown in FIG. 6C). Then, in 153, the value of each unit of the output layer is calculated and converted using the same method, and is outputted from 154.

〔Example〕

(Example 1) The present invention will be described in detail below. FIG. 1 is a block diagram showing the system configuration of an embodiment of the present invention. In the figure, 10 is a blast furnace, 20 is an automatic pattern determination device,
A data input device 21 that takes in signals from sensors installed in each part of the blast furnace at regular intervals, and a data processing device 22 that processes the taken data into a normalized signal suitable for noise processing or input to a neural network. , a data storage bag W23 for temporarily storing processed data, and a hierarchical neural network 24 for determining data. Hierarchical neural networks use the operator's qualitative judgment method as a learning algorithm, for example, literature: Neural networks are used for pattern recognition, signal processing, and knowledge processing (Nikkei Electronics 1987.8. IQ (No.
, 427)), and the results are stored. Further, in order to learn online, a learning algorithm may be provided in a neural network. 41 is a CRT for displaying the results of the automatic pattern determination device, and 40 is an automatic control system that uses the results of the pattern determination device to determine the amount of action to be applied to the blast furnace.

FIG. 2 is a hardware configuration diagram of a computer corresponding to the system configuration diagram shown in FIG. be done. 62 is a CPU for executing data processing and pattern judgment using a neural network, 63.64 is a ROM in which data processing and neural network programs and necessary parameters are written in advance, and 65.66 is for temporarily storing pattern data. 67 is a CRT for displaying the determination results, and 68 is an interface device for connecting with an automatic control system that uses the results of the pattern determination device to determine the amount of action to be taken to the blast furnace.

The first example is a concrete example of the operation of this embodiment having the above configuration.
Gas utilization rate in the furnace measured with sonde 71 installed at the position shown in Figure 0 (100X CO□/(CO x C0z
)) Since it has been applied to a pattern, it will be explained with reference to the flowchart in FIG. 3 and the explanatory diagrams in FIGS. 4 to 7. FIG. 4 shows a gas utilization rate pattern in the radial direction inside the furnace measured with a sonde, which is a typical example of blast furnace pattern data. This pattern of gas utilization rate is one of the important data for blast furnace operators to estimate the gas flow distribution in the furnace. Based on past operational experience, this can be broadly classified into eight types (A-H) as shown in Figure 5, and the operator determines which of these eight patterns the measured pattern data is similar to and then controls the gas flow. Actions are being taken to make the distribution appropriate.

First, 8 points of data measured by the ship are taken in at regular intervals by the data input device shown in Fig. 1, 21, and after the specified noise processing is performed by the data processing device shown in 22, the data becomes suitable as input to the neural network. It is normalized using the above equation (1) as follows.

Next, the pattern data processed in the above method is temporarily stored in the data storage device shown in FIG. 1 and used as input data for the neural network. The adopted neural network is a three-layer neural network, and in FIG. 6, 30 input layers, 31, 32 intermediate layers, and 33 output layers each have one layer. Each layer is composed of a plurality of units 34, and the units between each layer are coupled to each other via weighting coefficients 35. Also, its output characteristics are shown in Figure 6C)
As shown in Figure 2, when the total human power of a unit exceeds a certain threshold, it is a non-linear characteristic that outputs to the next layer unit. A signal is input to each unit 34 of the input layer, and when it is input to each unit of the intermediate layer via a weighting coefficient 35, the signal is strengthened or weakened by the weighting coefficient 35. Inside the unit, the sum of all input signals is calculated, and if the sum exceeds a predetermined threshold, the signal is output to the next unit.

While repeating such operations, signals are propagated from the input layer to the intermediate layer and from the intermediate layer to the output layer, and pattern determination is performed.

Furthermore, the above-mentioned learning algorithm is for learning the weighting coefficients and thresholds of the network.

Regarding the structure of the neural network, first, the number of units in the input layer is determined by using the data of each measurement point and the average value of the data of all points as input data, and performing the normalization process described above. And so. Next, the number of intermediate layers or the number of units within the intermediate layer was determined to be seven while verifying the pattern determination ability. The number of units in the output layer is determined by the number of determined patterns, so in this example, since the number of patterns to be determined is eight, the number of units in the output layer is also set to eight. A neural network with this structure is given multiple results for each pattern as learning data, and the result is the same as the Habo operator. Judgment became possible. An example of automatic determination using this system and the determination results are shown in FIG. 7 and Table 2. From Table 2, it can be seen that this system has almost the same ability as the operator's judgment.

Table 2 (Example 2) The three-layer neural network realized by the computer shown in FIGS. 1 and 2 was applied to the stave thermometer 72 shown in FIG. 10. Representative data of the stave thermometer is shown in FIG. Table 3 shows the conditions and results at that time. As a result, it was determined that automatic discrimination is possible. Note that the results are organized assuming that the operator's judgment is correct.

Table 3 [Effects of the Invention] As described above, the present invention captures measurement patterns from various sensors installed at each part of the blast furnace at a predetermined timing, processes the data, and then judges the pattern using a neural network. This enables automatic pattern determination that is less susceptible to data loss, noise, etc., and provides determination results that match the operator's past experience. This not only reduces the burden on the operator and allows them to concentrate on operations, but also dramatically expands the range of automation by combining it with the blast furnace automatic operation system and automatic control, as well as improving accuracy and responsiveness. This is useful for preventing mistakes and reducing individual differences among operators.

[Brief explanation of drawings]

FIG. 1 is a system configuration diagram in a typical embodiment of the present invention, and FIG. 2 is a hardware configuration diagram when realized by a computer in a typical embodiment of the present invention. Third
Figure a) is a processing flow in a system configuration diagram that is a typical embodiment of the present invention (Embodiment 3) The four-layer neural network realized by the computer in Figures 1 and 2 is shown in Figure 10. It was applied to computer 730 time series data. Figure 9 shows the representative data. Table 4 shows the conditions and results at that time. As a result, it was determined that automatic discrimination is possible. Note that the results are organized assuming that the operator's judgment is correct. Table 4 is a diagram, and b) in the same figure is a detailed processing flow diagram of pattern determination in a) in the same figure. FIG. 4 shows an example of a pattern of gas utilization rate measurement data at eight points by a sonde measuring device which is a typical embodiment of the present invention. FIG. 5 is an example of a typical pattern determined in an 8-point gas utilization rate pattern by a sonde measuring device which is a typical embodiment of the present invention. Figure 6a) shows the structure of a hierarchical neural network,
b) shows the detailed structure of the unit, and C) is an input/output relationship diagram of the unit. FIG. 7 is an example of pattern determination results in Example 1 in which the present invention is applied to eight points of gas utilization rate measurement data by a sonde measuring device. FIG. 8 is an example of measurement pattern element data in Example 2 in which the present invention is applied to data from a stave thermometer on the wall of a blast furnace. FIG. 9 is an example of measurement pattern element data in Example 3 in which the present invention is applied to pressure gauge data on the wall of a blast furnace. FIG. 10 is a layout diagram of a sensor in an embodiment in which the present invention is applied to a blast furnace. 10 is a blast furnace, 20 is an automatic pattern determination device, 21 is a data input device, 22 is a data processing device, 23 is a data storage device, 24 is a neural network, 30 is an input layer, 31.
32 is an intermediate layer, 33 is an output layer, 34 is a unit, 35 is a weighting coefficient between units, 40 is another system, 41 is a display device, 50 is measurement pattern element data, 61 is an interface device, 62 is a CPU, 63 .64 is ROM, 6
5.66 is a RAM, 67 is a CRT, 6B is an interface device, 71 is a sonde, 72 is a stave temperature needle on the blast furnace wall, and 73 is a pressure gauge on the blast furnace wall. Applicant Nippon Steel Corporation Patent Attorney Minoru Aoyagi R <R1 +00 200 Temperature (°C) Stave thermometer measurement butter/element data side time (seconds) Pressure gauge time series pattern

Claims (1)

[Claims] 1. A pattern element data input means that takes in measured pattern element data from a sensor installed in a blast furnace at a predetermined timing, and a hierarchical type that determines the measured pattern element data into a predetermined pattern. An automatic determination system for blast furnace pattern data characterized by being equipped with a neural network means. 2. pattern element data input means for taking in measurement pattern element data from a sensor installed in the blast furnace at a predetermined timing; signal processing means for performing signal processing such as noise processing on the measurement pattern element data; What is claimed is: 1. An automatic determination system for blast furnace pattern data, characterized in that it is equipped with a hierarchical neural network means for determining a predetermined pattern from pattern element data. 3. Pattern element data input means that takes in measurement pattern element data from a sensor installed in the blast furnace at a predetermined timing, signal processing means that performs signal processing such as noise processing on the measurement pattern element data, and the signal-processed measurement. storage means for temporarily storing pattern element data;
An automatic determination system for blast furnace pattern data, comprising hierarchical neural network means for determining the stored pattern element data into a predetermined pattern. 4. The automatic determination system for blast furnace pattern data according to claim 2, wherein the signal processing means includes normalization processing. 5. The blast furnace pattern data according to any one of claims 1 to 4, wherein the hierarchical neural network means is a three-layer neural network means consisting of one layer each of an input layer, an intermediate layer, and an output layer. automatic judgment system. 6. Claims 1 to 4 characterized in that said hierarchical neural network means is a four-layer neural network means each comprising one input layer, one output layer, and two intermediate layers.
An automatic determination system for blast furnace pattern data according to any one of the above. 7. The automatic determination system for blast furnace pattern data according to claim 5 or 6, characterized in that the signal level to the input layer of the hierarchical neural network means is a normalized signal. 8. The number of units in the input layer of the hierarchical neural network means is equal to the number of measurement pattern element data from various sensors and one for the average value signal of the measurement pattern data.
8. The automatic determination system for blast furnace pattern data according to claim 5 or 7, wherein the quantity is the sum of the data. 9. The number of terminals in the output layer of the hierarchical neural network means is the same as the number of patterns to be determined, and the output signal from the terminal is a signal normalized to a range of 0 to 1. The automatic determination system for blast furnace pattern data according to claim 5 or 6. 10. The automatic determination system for blast furnace pattern data according to any one of claims 1 to 9, wherein the sensor is a stave thermometer or a pressure gauge on a blast furnace wall. 11. The automatic determination system for blast furnace pattern data according to any one of claims 1 to 9, wherein the sensor is a sonde equipped with a sensor that measures at least one of gas components and temperature within the blast furnace. 12. The automatic determination system for blast furnace pattern data according to any one of claims 1 to 9, wherein the measured pattern element data from the sensor is time-series pattern data of a blast furnace internal pressure signal.