Raw material charging control device for blast furnace, method for generating opening degree command value, and program

Abstract

A raw material charging control device 90 comprises: a data collection unit 118 that accumulates, in a data storage unit, training records that include an opening degree of a gate 40 for charging a raw material from a bunker 30 in which the raw material is accommodated above a blast furnace 2, a charging speed of the raw material from the bunker 30, and charging conditions including characteristics of the raw material; an offline learning unit 211 that generates a prediction model that expresses, by a multi-stage input / output relationship, the relationship between the opening degree, the charging speed, and the charging conditions, through machine learning based on the plurality of training records accumulated in the data storage unit; a command generation unit 220 that generates, on the basis of the prediction model and charging conditions and a charging target speed of a raw material to be newly inserted, an opening degree command value for the gate 40, the opening degree command value corresponding to the charging target speed; and a gate 40 controller that sets the opening degree so as to correspond to the opening degree command value.

Description

Blast furnace raw material charging control device, opening command value generation method, and program

[0001] The present disclosure relates to a raw material charging control device for a blast furnace, a method for generating an opening command value, and a program.

[0002] Patent Document 1 discloses a control method in which a model formula is prepared that represents the relationship between the number of revolutions of the rotating chute and the opening degree of the flow adjustment gate, the opening degree of the flow adjustment gate is controlled based on the model formula, and the actual values ​​of the opening degree of the flow adjustment gate, the number of revolutions of the rotating chute, the revolution speed, and the weighing value obtained from the results after the loading of the raw materials is completed are taken in to learn the parameters of the model formula.

[0003] Japanese Patent Application Publication No. 10-251719

[0004] The present disclosure provides an apparatus that is effective for controlling the charging rate of raw materials from a bunker into a blast furnace with high precision.

[0005] A raw material charging control device for a blast furnace according to one aspect of the present disclosure includes: a data collection unit that accumulates learning records in a data storage unit, the learning records including the opening of a gate through which raw materials are charged from a bunker that stores the raw materials at the top of the blast furnace, the charging rate of the raw materials from the bunker, and charging conditions including characteristics of the raw materials; an offline learning unit that generates a prediction model that expresses the relationship between the opening, the charging rate, and the charging conditions using a multi-stage input / output relationship through machine learning based on the plurality of learning records accumulated in the data storage unit; a command generation unit that generates a gate opening command value that corresponds to the target charging rate based on the charging conditions and target charging rate of newly added raw materials and the prediction model; and a furnace top controller that corresponds the opening to the opening command value.

[0006] A method for generating an opening command value according to another aspect of the present disclosure includes accumulating learning records in a data storage unit, the learning records including the opening of a gate through which raw materials are charged from a bunker that stores raw materials at the top of a blast furnace, the charging rate of the raw materials from the bunker, and charging conditions including characteristics of the raw materials; generating a prediction model that represents the relationship between the opening, the charging rate, and the charging conditions using a multi-stage input / output relationship based on the plurality of learning records stored in the data storage unit; and generating a gate opening command value corresponding to the target charging rate based on the charging conditions and target charging rate of newly added raw materials and the prediction model.

[0007] A program according to yet another aspect of the present disclosure causes an apparatus to execute the following operations: accumulate learning records in a data storage unit, including the opening of a gate through which raw materials are charged from a bunker that stores the raw materials in the upper part of a blast furnace, the charging rate of the raw materials from the bunker, and charging conditions including the characteristics of the raw materials; generate a prediction model that represents the relationship between the opening, the charging rate, and the charging conditions using a multi-stage input / output relationship based on the multiple learning records accumulated in the data storage unit; and generate a gate opening command value corresponding to the target charging rate based on the charging conditions and target charging rate of newly added raw materials and the prediction model.

[0008] According to the present disclosure, it is possible to provide an apparatus that is effective in controlling the charging speed of raw materials from a bunker into a blast furnace with high precision.

[0009] FIG. 1 is a schematic diagram illustrating the configuration of a raw material charging device. FIG. 2 is a block diagram illustrating the functional configuration of a server device and a learning calculation device. FIG. 3 is a schematic diagram illustrating a prediction model generated by deep learning. FIG. 4 is a schematic diagram illustrating a prediction model generated by a random forest method or a gradient boosting method. FIG. 5 is a block diagram illustrating the configuration of a command generation unit. FIG. 6 is a block diagram illustrating the hardware configuration of a server device and a learning calculation device. FIG. 7 is a flowchart illustrating a record accumulation procedure. FIG. 8 is a flowchart illustrating an offline learning procedure. FIG. 9 is a flowchart illustrating an opening control procedure. FIG. 10 is a flowchart illustrating a procedure for acquiring learning records.

[0010] Hereinafter, the embodiments will be described in detail with reference to the drawings. In the description, the same elements or elements having the same functions are denoted by the same reference numerals, and redundant description will be omitted.

[0011] 1 is a device that charges raw materials into a blast furnace 2 used for steel production, etc. The raw material charging device 1 includes a bunker 30, a gate 40, a collection hopper 50, a distribution chute 60, a hopper 10, and a conveyor 20. The bunker 30 stores raw materials in the upper part of the blast furnace 2. The raw materials charged into the blast furnace 2 may include multiple types of raw materials.

[0012] The raw material charging device 1 may be equipped with a plurality of bunkers 30, and each of the plurality of bunkers 30 may store raw materials in the upper part of the blast furnace 2. In the illustrated example, the raw material charging device 1 has two bunkers 30A and 30B. Above the bunkers 30A and 30B, a distributing device 31 is provided that distributes raw materials charged from above to either the bunker 30A or 30B.

[0013] The plurality of bunkers 30 are selectively used based on a predetermined order, etc. When the raw materials include a plurality of types of raw materials, each of the plurality of bunkers 30 stores the plurality of types of raw materials.

[0014] Each of the multiple bunkers 30 may be provided with a pressure sensor 32, and the blast furnace 2 may be provided with a pressure sensor 3. The pressure sensor 32 detects the internal pressure of the bunker 30, and the pressure sensor 3 detects the internal pressure of the blast furnace 2.

[0015] The gate 40 sends out the raw material from the bunker 30. The opening degree of the gate 40 can be changed by an electric actuator. The opening degree is expressed, for example, as a ratio to the fully open opening degree.

[0016] When the raw material charging device 1 is equipped with a plurality of bunkers 30, it is equipped with a plurality of gates 40 corresponding to the plurality of bunkers 30. Each of the plurality of gates 40 is provided below the corresponding bunker 30. In the illustrated example, the raw material charging device 1 has two gates 40A, 40B corresponding to the two bunkers 30A, 30B, respectively.

[0017] The collecting hopper 50 temporarily stores the raw materials sent out from the multiple gates 40 and collects them at a single charging port 51. The distribution chute 60 distributes the raw materials that drop from the charging port 51 into the blast furnace 2. For example, the distribution chute 60 tilts and rotates using an electric actuator, distributing the raw materials in a spiral pattern inside the blast furnace 2.

[0018] The multiple hoppers 10 are provided outside the blast furnace and each accommodate a different type of raw material. For example, the multiple hoppers 10 include a hopper 10A that accommodates relatively small coke lumps, a hopper 10B that accommodates main raw materials such as a plurality of types of iron ore, a hopper 10C that accommodates auxiliary raw materials, and a hopper 10D that accommodates relatively large coke lumps. Each of the multiple hoppers 10 has a discharge device 11 at its bottom. The discharge device 11 discharges the raw materials accommodated in the hopper 10 onto a conveyor 20. Hereinafter, as necessary in the description, the raw materials discharged by the discharge device 11 will be referred to as "discharged raw materials" to distinguish them from the raw materials accommodated in the hopper 10.

[0019] The multiple hoppers 10 may be provided with weigh scales 12 that detect the weight of each of the multiple types of discharged raw materials. For example, each of the multiple hoppers 10 further includes a weigh scale 12. For example, the weigh scale 12 detects the weight of the discharged raw materials based on the difference between the weight of the hopper 10 before the materials are discharged by the discharge device 11 and the weight of the hopper 10 after the materials are discharged by the discharge device 11. The multiple weigh scales 12 provided in the multiple hoppers 10 respectively detect the weight of each of the multiple types of discharged raw materials before the multiple types of discharged raw materials are placed in a bunker 30 described below.

[0020] The multiple hoppers 10 may be provided with moisture meters 13 that detect the moisture content of the multiple types of raw materials. For example, each of the multiple hoppers 10 further includes a moisture meter 13. For example, the moisture meter 13 detects the moisture content of the raw materials in the hopper 10 using a non-contact method such as infrared. The moisture content is expressed, for example, as the ratio of the weight of the moisture content to the total weight of the raw materials. The moisture content of the raw materials is substantially the same when they are stored in the hopper 10 and when they are discharged from the hopper 10. Therefore, the moisture content detected by the moisture meter 13 represents the moisture content of the discharged raw materials.

[0021] The degree of influence of the moisture content on the flow characteristics, etc., may vary depending on the type of raw material. Since there may be types of raw material for which the influence of the moisture content can be ignored, it is not necessary for each of the multiple hoppers 10 to have a moisture meter 13.

[0022] The conveyor 20 is, for example, a belt conveyor, and transports the multiple types of discharged raw materials discharged from the multiple hoppers 10 to the bunker 30. The multiple types of discharged raw materials transported by the conveyor 20 are dumped into the bunker 30. When the raw material charging device 1 is equipped with multiple bunkers 30, the multiple types of discharged raw materials transported by the conveyor 20 are dumped into one of the multiple bunkers 30 by the sorting device 31.

[0023] In the above, an example has been given in which a plurality of gates 40 are provided for a plurality of bunkers 30 on the inlet side of the collecting hopper 50, but the gates 40 may also be provided on the outlet side of the collecting hopper 50. If only one bunker 30 is required, the bunker 30 and the collecting hopper 50 may be integrated into one unit.

[0024] The raw material charging control device 90 controls at least the multiple hoppers 10 and the gate 40. The raw material charging control device 90 may include multiple controllers 901. The multiple controllers 901 are capable of communicating with each other. The multiple controllers 901 sequentially perform multiple-stage control, from discharging multiple types of raw materials from the multiple hoppers 10 to charging the multiple types of raw materials from the bunker 30 into the blast furnace 2. For example, the multiple controllers 901 include a raw material controller 92 and a furnace top controller 93. The raw material charging control device 90 may further include a schedule controller 91. The raw material controller 92 controls each of the multiple hoppers 10 so that the discharge device 11 discharges raw materials by a weight corresponding to a predetermined target weight. The raw material controller 92 may retain the elapsed time from the time the discharge device 11 starts discharging raw materials to the time the discharge device 11 stops discharging raw materials, as information representing the discharge rate of the raw materials.

[0025] The furnace top controller 93 adjusts the opening of the gate 40 so that the raw materials are charged (dumped) from the bunker 30 at a charging speed corresponding to a predetermined charging target speed. The charging speed and the target charging speed are expressed, for example, in terms of the weight of raw materials charged per unit time. The charging speed and the target charging speed may be expressed by the charging period of the raw materials from the bunker 30 (the length of time from the start of charging to the completion of charging), assuming that information on the weight of the raw materials fed into the bunker 30 is separately acquired. The raw material charging period may also be expressed by the number of rotations and rotation speed of the distribution chute 60. The furnace top controller 93 may retain the elapsed time from the time when charging of the raw materials from the bunker 30 starts to the time when charging is completed, and use this information as information indicating the charging period.

[0026] When the raw material charging device 1 is equipped with a plurality of bunkers 30, the furnace top controller 93 may control the allocating device 31 to charge the raw materials transported by the conveyor 20 into one of the plurality of bunkers 30. The furnace top controller 93 may retain identification information of the bunker 30 into which the raw materials are to be charged.

[0027] The schedule controller 91 communicates with the material controller 92 and the furnace top controller 93 via a network line 94. The network line 94 may be a local area network for control or a wide area network. For example, the schedule controller 91 receives material status information of the multiple hoppers 10 from the material controller 92 and receives furnace top status information of the bunker 30 and the gate 40 from the furnace top controller 93 via the network line 94. The material status information of the multiple hoppers 10 may include the status of the discharge device 11, the detection results of the weight scale 12, the detection results of the moisture meter 13, etc. The furnace top status information of the bunker 30 and the gate 40 may include identification information of the bunker 30 in use, the remaining weight of material in the bunker 30, the opening degree of the gate 40, etc. The schedule controller 91 calculates the above-mentioned target weight and target charging speed based on a predetermined production plan, the material status information of the multiple hoppers 10, and the furnace top status information of the bunker 30 and the gate 40. The schedule controller 91 transmits the target weight to the raw material controller 92 and the target charging rate to the furnace top controller 93 via a network line 94 .

[0028] The actual charging rate at the opening adjusted by the furnace top controller 93 may have an error relative to the target charging rate. To distribute the raw materials evenly within the blast furnace 2, it is desirable for this error to be small. However, since the relationship between the opening adjusted by the furnace top controller 93 and the actual charging rate varies significantly depending on various conditions, such as the characteristics of the raw materials, it is difficult to reduce this error. Therefore, the raw material charging control device 90 may be configured to: accumulate learning records in a data storage unit, including the opening of the gate 40, the charging rate of the raw materials from the bunker 30, and charging conditions including the characteristics of the raw materials; generate a prediction model that represents the relationship between the opening, the charging rate, and the charging conditions using a multi-stage input-output relationship based on the multiple learning records accumulated in the data storage unit; generate a gate opening command value corresponding to the target charging rate based on the charging conditions and target charging rate of the newly charged raw materials and the prediction model; and correspond the opening of the gate 40 to the generated opening command value.

[0029] By using a prediction model that expresses the relationship between the opening, the charging rate, and the charging conditions as a multi-stage input / output relationship, it is possible to generate an opening command value corresponding to the target charging rate with higher accuracy. By making the opening correspond to the opening command value generated with high accuracy, it is possible to suppress the error of the charging rate relative to the target charging rate. Therefore, it is effective in controlling the charging rate of raw materials from a bunker into a blast furnace with high accuracy.

[0030] For example, the raw material charging control device 90 further includes a server device 100 and a learning calculation device 200. The server device 100 and the learning calculation device 200 communicate with each other via a network line 94 and with the raw material controller 92 and the furnace top controller 93 via the network line 94. The server device 100 collects the learning records from the raw material controller 92 and the furnace top controller 93 via the network line 94 and stores them in a data storage unit. The learning calculation device 200 acquires the learning records stored in the data storage unit via the network line 94, generates the prediction model based on the acquired learning records, and generates a gate opening command value corresponding to the target charging rate based on the charging conditions and target charging rate of newly added raw materials and the prediction model. The learning calculation device 200 transmits the opening command value to the furnace top controller 93 via the network line 94.

[0031] FIG. 2 is a block diagram illustrating the functional configuration of the server device 100 and the learning calculation device 200. As shown in FIG. 2, the server device 100 has, as its functional configuration (hereinafter referred to as "functional blocks"), an opening information acquisition unit 111, a speed information acquisition unit 112, a weight information acquisition unit 113, a flow characteristic information acquisition unit 114, a moisture information acquisition unit 115, an environmental information acquisition unit 116, a bunker information acquisition unit 117, a data collection unit 118, and a data storage unit 119. The opening information acquisition unit 111 acquires opening information indicating the opening of the gate 40 from the furnace top controller 93. The speed information acquisition unit 112 acquires speed information indicating the actual charging speed at the opening indicated by the opening information from the furnace top controller 93. For example, the speed information acquisition unit 112 acquires information on the actual charging period at the opening indicated by the opening information from the furnace top controller 93.

[0032] The weight information acquisition unit 113 acquires type information indicating the type of each of the multiple types of charged raw materials and weight information indicating the weight of each of the multiple types of charged raw materials. For example, the weight information acquisition unit 113 acquires detection results from the multiple weighing scales 12 from the raw material controller 92 and acquires type information and weight information based on the acquired detection results. The flow characteristic information acquisition unit 114 acquires flow characteristic information indicating the flow characteristic of each of the multiple types of raw materials. For example, the flow characteristic information acquisition unit 114 acquires information indicating the discharge speed of each of the multiple types of raw materials when it is discharged from the corresponding hopper from the raw material controller 92 as flow characteristic information. The moisture content information acquisition unit 115 acquires moisture content information indicating the moisture content of the multiple types of raw materials. For example, the moisture content information acquisition unit 115 acquires moisture content information from the raw material controller 92 based on detection results from the multiple moisture meters 13 included in the multiple hoppers 10, respectively.

[0033] The environmental information acquisition unit 116 acquires environmental information about the inside of the bunker 30 and the inside of the blast furnace 2 from the furnace-top controller 93. For example, the environmental information acquisition unit 116 acquires pressure information including the detection results by the pressure sensor 32 and the detection results by the pressure sensor 3 from the furnace-top controller 93, and acquires differential pressure information indicating the differential pressure between the inside of the bunker 30 and the inside of the blast furnace 2 based on the acquired pressure information. The bunker information acquisition unit 117 acquires bunker identification information indicating which of the multiple bunkers 30 was used to accommodate and charge multiple types of raw materials from the furnace-top controller 93. The bunker identification information may be represented by, for example, a categorical variable.

[0034] The data collection unit 118 accumulates in the data storage unit 119 a learning record including the opening degree of the gate 40, the charging speed of the raw material from the bunker 30, and the charging conditions including the characteristics of the raw material.

[0035] The raw material characteristics may include the type information and weight information, the flow property information, and the moisture content information. The charging conditions may further include the environmental information and the bunker identification information.

[0036] For example, the data collection unit 118 acquires opening information from the opening information acquisition unit 111, speed information from the speed information acquisition unit 112, type information and weight information from the weight information acquisition unit 113, flow property information from the flow property information acquisition unit 114, moisture information from the moisture information acquisition unit 115, environmental information from the environmental information acquisition unit 116, and bunker identification information from the bunker information acquisition unit 117, associates all the acquired information to generate a learning record, and stores the generated learning record in the data storage unit 119. In this way, when the raw material characteristics include weight information, the speed information may be represented by the charging time, assuming the weight of the raw material specified based on the weight information.

[0037] When machine learning is performed based on training records containing multiple types of numerical values, numerical values ​​with a large absolute value variation range tend to have a greater impact on the training results than numerical values ​​with a small absolute value variation range. However, the magnitude of the absolute value variation range does not necessarily correlate with importance in machine learning. Therefore, the data collection unit 118 may be configured to generate training records by performing a standardization process on at least one type of information. For example, one example of standardization process for one type of information performed by the data collection unit 118 is to divide a numerical value representing one type of information by a predetermined variation range to make it dimensionless.

[0038] The learning calculation device 200 has, as functional blocks, an offline learning unit 211, a prediction model storage unit 212, and a command generation unit 220. The offline learning unit 211 generates a prediction model that represents the relationship between the aperture, the charging rate, and the charging conditions by a multi-stage input-output relationship through machine learning based on a plurality of learning records stored in the data storage unit 119. The input-output relationship may be represented by a case classification process that classifies inputs and generates outputs according to the case classification results, or a probabilistic case classification process.

[0039] The offline learning unit 211 may generate a prediction model using a neural network based on deep learning. Fig. 3 is a schematic diagram illustrating a prediction model based on deep learning. The prediction model 310 shown in Fig. 3 is a neural network and includes an input layer 311, one or more intermediate layers 313, and an output layer 312.

[0040] The input layer 311 outputs an input vector to the next intermediate layer 313. The intermediate layer 313 transforms the input from the previous layer using an activation function and outputs the result to the next layer. The output layer 312 transforms the input from the intermediate layer 313, which is the farthest from the input layer 311, using an activation function and outputs the transformation result as an output vector. In the prediction model 310, the activation functions in each intermediate layer 313 and the activation function in the output layer 312 are an example of the multi-stage input-output relationship described above. When the input is transformed using an activation function, the transformation result changes depending on the input. In other words, since the transformation result changes depending on the input, transformation using an activation function is an example of the case classification process or probabilistic case classification process described above.

[0041] The offline learning unit 211 may generate a prediction model using a random forest method or a gradient boosting method. FIG. 4 is a schematic diagram illustrating a prediction model using a random forest method or a gradient boosting method. The prediction model 320 shown in FIG. 4 includes multiple decision trees 321 and an output selection unit 322. The decision trees 321 represent the relationship between input and output using conditional branching in multiple stages of branches 323. The output selection unit 322 selects one of the outputs of the multiple decision trees 321 by majority voting. In the random forest method, the offline learning unit 211 generates the multiple decision trees 321 by bagging. In the gradient boosting method, the offline learning unit 211 generates the multiple decision trees 321 by boosting.

[0042] In the prediction model 320, the decision tree 321 and the output selection unit 322 are an example of the above-mentioned multi-stage input / output relationship. According to the decision tree 321, the output changes depending on the input, and therefore the input / output conversion by the decision tree 321 is an example of case classification processing or probabilistic case classification processing. Furthermore, the multi-stage branches 323 in the decision tree 321 are also an example of the above-mentioned multi-stage input / output relationship. According to the branches 323, the branch to which the process moves to in the subsequent stage changes depending on the input, and therefore the conditional branch by the branches 323 is an example of case classification processing or probabilistic case classification processing.

[0043] According to the multiple regression model, the relationship between the opening, the charging rate, and the charging conditions is expressed as a single function. Therefore, the multiple regression model does not correspond to a prediction model that expresses the relationship between the opening, the charging rate, and the charging conditions as a multi-stage input-output relationship.

[0044] The offline learning unit 211 stores the generated prediction model in the prediction model storage unit 212. The command generation unit 220 generates an opening command value for the gate 40 corresponding to the target charging rate, based on the charging conditions and target charging rate of the newly charged raw materials and the prediction model stored in the prediction model storage unit 212. The newly charged raw materials refer to the raw materials that are charged into the bunker 30 after the prediction model is generated. The command generation unit 220 acquires the charging conditions of the newly charged raw materials from, for example, the data collection unit 118.

[0045] The command generating unit 220 transmits the generated opening command value to the furnace top controller 93. The furnace top controller 93 makes the opening of the gate 40 correspond to the opening command value.

[0046] As described above, the offline learning unit 211 may generate a prediction model based on the plurality of learning records, each of which includes type information and weight information as charging conditions, and the command generating unit 220 may generate an opening command value based on the charging conditions including type information and weight information of the newly charged raw material, the target charging speed, and the prediction model.

[0047] As described above, the offline learning unit 211 may generate a prediction model based on a plurality of learning records each including weight information acquired from the weighing scale 12 as the loading conditions, and the command generating unit 220 may generate an opening command value based on the loading conditions including weight information acquired from the weighing scale 12 for newly added raw materials, the target loading speed, and the prediction model.

[0048] As described above, the offline learning unit 211 may generate a prediction model based on a plurality of learning records each including flow characteristic information as a charging condition, and the command generating unit 220 may generate an opening command value based on the charging condition including the flow characteristic information of the newly charged raw material, the target charging rate, and the prediction model.

[0049] As described above, the offline learning unit 211 may generate a prediction model based on a plurality of learning records each including flow characteristic information acquired based on the discharge rate as the charging condition, and the command generating unit 220 may generate an opening command value for a newly input raw material based on the charging condition including the flow characteristic information acquired based on the discharge rate, the target charging rate, and the prediction model.

[0050] As described above, the offline learning unit 211 may generate a prediction model based on a plurality of learning records each including moisture information as a charging condition, and the command generating unit 220 may generate an opening command value based on the charging conditions including moisture information of the newly charged raw material, the target charging rate, and the prediction model.

[0051] As described above, the offline learning unit 211 may generate a prediction model based on a plurality of learning records each including environmental information as charging conditions, and the command generating unit 220 may generate an opening command value based on the charging conditions including environmental information when the newly charged raw material is stored in the bunker 30, the target charging speed, and the prediction model.

[0052] As described above, the offline learning unit 211 may generate a prediction model based on multiple learning records, each of which includes information on differential pressure as a loading condition, and the command generating unit 220 may generate an opening command value based on the loading conditions including information on differential pressure when the newly charged raw material is stored in the bunker 30, the target loading speed, and the prediction model.

[0053] As described above, the offline learning unit 211 may generate a predictive model based on multiple learning records, each of which includes bunker identification information as a loading condition, and the command generating unit 220 may generate an opening command value based on loading conditions including bunker identification information indicating which of multiple bunkers 30 will be used as the bunker 30 for the newly added raw material, the loading target speed, and the predictive model.

[0054] For example, the offline learning unit 211 may generate a prediction model based on multiple learning records, each of which includes all of the type information, weight information, flow characteristics information, moisture content information, environmental information, and bunker identification information, and the command generating unit 220 may generate an opening command value based on the charging conditions, which include all of the type information, weight information, flow characteristics information, moisture content information, environmental information, and bunker identification information of the newly charged raw material, the target charging rate, and the prediction model.

[0055] The offline learning unit 211 may generate a prediction model to output an opening degree in response to inputs of the charging conditions and the charging rate. The command generating unit 220 may input the charging conditions and the charging rate of newly added raw materials to the prediction model to calculate an opening degree command value.

[0056] The offline learning unit 211 may generate a prediction model to output a charging rate in response to input of the charging conditions and the aperture. The command generating unit 220 may input the charging conditions of newly charged raw materials and the aperture command value into the prediction model to calculate a charging rate and change the aperture command value, repeatedly calculating an aperture command value corresponding to the charging target rate. For example, as shown in FIG. 5 , the command generating unit 220 may include an initial value generating unit 221, a simulation unit 222, and a command value changing unit 223.

[0057] The initial value generating unit 221 generates an initial value of the opening command value. The simulation unit 222 inputs the charging conditions of the newly charged raw material and the opening command value into a prediction model to calculate the charging speed. Hereinafter, the charging speed calculated by the simulation unit 222 is referred to as the "estimated speed." The command value changing unit 223 changes the opening command value.

[0058] For example, the command value changing unit 223 calculates the deviation between the charging target speed and the estimated speed, calculates a correction value for the opening command value so as to reduce the deviation, and adds the correction value to the opening command value to change the opening command value. The command value changing unit 223 may calculate the correction value by assigning a sign in the direction of reducing the deviation to the absolute value of a predetermined one step. For example, the command value changing unit 223 may assign a positive sign to the absolute value when the deviation is a positive value (when the estimated speed is smaller than the charging target speed), and may assign a negative sign to the absolute value when the deviation is a negative value. The command value changing unit 223 may calculate the correction value by performing a proportional operation, a proportional-integral operation, or a proportional-integral-differential operation on the deviation.

[0059] The simulation unit 222 and the command value modification unit 223 calculate the estimated speed, calculate the deviation, and modify the opening command value so as to reduce the deviation, starting with the opening command value as an initial value, and repeating this process until the deviation becomes equal to or less than a predetermined threshold. The command value modification unit 223 transmits the opening command value at the time when the deviation becomes equal to or less than the threshold to the furnace top controller 93 as a generated result of the opening command value.

[0060] By setting the initial value to a value close to the opening command value that is ultimately transmitted to the furnace top controller 93, it is possible to reduce the number of repetitions of calculations by the simulation unit 222 and the command value modification unit 223. Therefore, the learning calculation device 200 may further include a profile generation unit 231 and a profile storage unit 232.

[0061] The profile generating unit 231 repeatedly inputs the charging conditions and the opening command value into the prediction model to calculate the charging rate and changes at least one of the charging conditions and the opening command value to generate a profile representing the relationship between the charging conditions, the charging rate, and the opening command value, and stores the profile in the profile storage unit 232. For example, the profile generating unit 231 generates a profile representing the relationship between the charging conditions, the charging rate, and the opening command value as a one-stage input / output relationship. For example, the profile generating unit 231 may generate the profile as a function or as a discrete lookup table. The profile generating unit 231 may generate a profile that outputs an opening command value according to inputs of the charging conditions and the charging rate.

[0062] In a configuration in which the learning calculation device 200 further includes a profile generation unit 231 and a profile storage unit 232, the initial value generation unit 221 may calculate an initial value of the opening command value based on the profile stored in the profile storage unit 232 and the charging target rate. This makes the initial value close to the opening command value that is finally transmitted to the furnace top controller 93, thereby reducing the number of repetitions of calculations by the simulation unit 222 and the command value change unit 223.

[0063] FIG. 6 is a block diagram illustrating an example of the hardware configuration of the server device 100 and the learning calculation device 200. As shown in FIG. 6, the server device 100 includes a circuit 190. The circuit 190 includes a processor 191, a memory 192, a storage 193, and a communication port 194. The processor 191 includes one or more processing elements, and the memory 192 includes one or more memory elements such as a random access memory. The storage 193 is a non-volatile storage device. Specific examples of the storage 193 include a hard disk, a flash memory, a read-only memory, etc. The storage 193 may also be a portable storage medium such as a USB memory, an optical disk, or a magnetic disk.

[0064] The storage 193 stores a program for causing the server device 100 to accumulate the learning records. For example, the storage 193 stores a program for causing the server device 100 to configure each of the above-described functional blocks.

[0065] The memory 192 temporarily stores programs and the like loaded from the storage 193. The processor 191 executes the programs loaded into the memory 192 while temporarily storing the results of calculations in the memory 192, and configures each of the above-mentioned functional blocks in the server device 100. The communication port 194 performs communication via the network line 94 in response to a request from the processor 191.

[0066] The learning calculation device 200 includes a circuit 290. The circuit 290 includes a processor 291, a memory 292, a storage 293, and a communication port 294. The processor 291 includes one or more calculation elements, and the memory 292 includes one or more memory elements such as a random access memory. The storage 293 is a non-volatile storage device. Specific examples of the storage 293 include a hard disk, a flash memory, and a read-only memory. The storage 293 may also be a portable storage medium such as a USB memory, an optical disk, or a magnetic disk.

[0067] The storage 293 stores a program for causing the learning calculation device 200 to generate the above-mentioned prediction model based on a plurality of learning records accumulated in the storage 193 of the server device 100, and to generate an opening command value corresponding to the target charging rate based on the charging conditions and target charging rate of newly charged raw materials and the prediction model. For example, the storage 293 stores a program for causing the learning calculation device 200 to configure each of the above-mentioned functional blocks.

[0068] The memory 292 temporarily stores programs and the like loaded from the storage 293. The processor 291 executes the programs loaded into the memory 292 while temporarily storing the calculation results in the memory 292, and configures each of the above-mentioned functional blocks in the learning calculation device 200. The communication port 294 performs communication via the network line 94 in response to a request from the processor 291. The above hardware configuration is merely an example and can be modified as appropriate. For example, at least a part of the server device 100 may be incorporated into the learning calculation device 200. Furthermore, at least a part of the server device 100 and the learning calculation device 200 may be incorporated into the furnace top controller 93 or the raw material controller 92.

[0069] [Control Procedure] The control procedure executed by the raw material charging control device 90 will be exemplified below. This control procedure includes a procedure for generating an opening command value, which includes storing the learning records in the data storage unit 119, generating the prediction model based on the multiple learning records stored in the data storage unit 119, and generating an opening command value for the gate 40 corresponding to the target charging rate based on the charging conditions and target charging rate of the newly charged raw materials and the prediction model, and associating the opening of the gate 40 with the opening command value. Below, this control procedure will be exemplified by dividing it into a record storage procedure, an offline learning procedure, and an opening control procedure. Furthermore, an example of a procedure for acquiring the learning records will be given below.

[0070] 7 , the server device 100 executes steps S01, S02, S03, and S04. In step S01, raw materials are loaded from the bunker 30, and the data collection unit 118 waits for various information to be acquired by the opening information acquisition unit 111, the speed information acquisition unit 112, the weight information acquisition unit 113, the flowability information acquisition unit 114, the moisture content information acquisition unit 115, the environmental information acquisition unit 116, and the bunker information acquisition unit 117. In step S02, the data collection unit 118 acquires opening information from the opening information acquisition unit 111, speed information from the speed information acquisition unit 112, type information and weight information from the weight information acquisition unit 113, flowability information from the flowability information acquisition unit 114, moisture content information from the moisture content information acquisition unit 115, environmental information from the environmental information acquisition unit 116, and bunker identification information from the bunker information acquisition unit 117. In step S03, the data collection unit 118 performs standardization processing on at least one type of information. In step S04, the data collection unit 118 associates all information, including the standardized information, to generate a learning record, and stores the generated learning record in the data storage unit 119. Thereafter, the server device 100 returns the processing to step S01. The server device 100 repeatedly executes the above processing.

[0071] (Offline Learning Procedure) This procedure is executed after a sufficient number of learning records for machine learning have been accumulated in the data storage unit 119. As shown in Fig. 8 , the learning calculation device 200 first executes steps S11 and S12. In step S11, the offline learning unit 211 generates a prediction model that represents the relationship between the aperture, the charging rate, and the charging conditions by a multi-stage input-output relationship through machine learning based on the multiple learning records accumulated in the data storage unit 119. In step S12, the offline learning unit 211 stores the generated prediction model in the prediction model storage unit 212.

[0072] Next, the learning calculation device 200 executes steps S13 and S14. In step S13, the profile generation unit 231 repeatedly inputs the charging conditions and the aperture command value into the prediction model to calculate the charging rate and changes at least one of the charging conditions and the aperture command value to generate a profile representing the relationship between the charging conditions, the charging rate, and the aperture command value. In step S14, the profile storage unit 232 stores the generated profile in the profile storage unit 232. This completes the offline learning procedure.

[0073] (Opening degree control procedure) As shown in Fig. 9, the learning calculation device 200 first executes steps S21, S22, and S23. In step S21, the data collection unit 118 acquires type information and weight information for a newly charged raw material from the weight information acquisition unit 113, acquires flowability information from the flowability information acquisition unit 114, acquires moisture content information from the moisture content information acquisition unit 115, acquires environmental information from the environmental information acquisition unit 116, and acquires bunker identification information from the bunker information acquisition unit 117. In step S22, the data collection unit 118 performs the same standardization process as in step S03 on the information acquired in step S21. In step S23, the initial value generation unit 221 calculates an initial value of the opening degree command value based on the charging target rate and the profile stored in the profile storage unit 232.

[0074] Next, the learning calculation device 200 executes steps S24, S25, and S26. In step S24, the simulation unit 222 inputs the charging conditions of the newly charged raw materials and the opening command value into the prediction model to calculate the charging speed (the above-mentioned estimated speed). In step S25, the command value modification unit 223 calculates the deviation between the charging target speed and the estimated speed. In step S26, the command value modification unit 223 checks whether the deviation is equal to or smaller than a threshold value.

[0075] If it is determined in step S26 that the deviation exceeds the threshold, the learning calculation device 200 executes step S27. In step S27, the command value change unit 223 changes the opening command value so as to reduce the deviation. The learning calculation device 200 then returns the process to step S24. Thereafter, the calculation of the estimated speed, the calculation of the deviation, and the change of the opening command value so as to reduce the deviation are repeated until the deviation becomes equal to or less than the threshold.

[0076] If it is determined in step S26 that the deviation is equal to or less than the threshold, the learning calculation device 200 executes steps S31 and S32. In step S31, the command value change unit 223 transmits the opening command value at the time when the deviation becomes equal to or less than the threshold to the furnace top controller 93 as the opening command value generation result. In step S32, the furnace top controller 93 makes the opening of the gate 40 correspond to the opening command value. This completes the opening control procedure.

[0077] (Procedure for acquiring learning records) The procedure for acquiring learning records includes the multiple stages of controllers 901 sequentially transferring tracking data from the first stage controller 901 to the last stage controller 901 while adding actual data of the executed control to the tracking data, and the data collection unit acquiring the learning records from the tracking data to which actual data has been added by the last stage controller 901.

[0078] The raw material controller 92 may add performance data including the weight and discharge rate of each of the multiple types of raw materials to the tracking data, and the furnace top controller 93 may add performance data including the opening degree of the gate 40 and the charging rate to the tracking data.

[0079] For example, as shown in FIG. 10 , the raw material charging control device 90 first executes step S41. In step S41, the schedule controller 91 generates weighing / discharge setting data including the target weight and dump setting data including the target charging speed for one cycle of the above-mentioned multi-stage control. The schedule controller 91 associates the weighing / discharge setting data with cycle identification information and transmits them to the raw material controller 92. The cycle identification information is the identification information for one cycle described above. The schedule controller 91 associates the dump setting data including the target charging speed with the cycle identification information and transmits them to the furnace top controller 93.

[0080] Next, the raw material charging control device 90 executes steps S42, S43, S44, and S45. In step S42, the raw material controller 92 waits for the timing to start weighing. The weighing start timing is, for example, a timing after the timing at which discharge in the multiple-stage control of the previous cycle is completed.

[0081] In step S43, the raw material controller 92 adds the weighing / discharge setting data and the cycle identification information associated with the weighing / discharge setting data to the tracking data. In step S44, the raw material controller 92 causes the multiple hoppers 10 to respectively weigh the multiple types of raw materials to be discharged onto the conveyor 20 based on the weighing / discharge setting data. In step S45, the raw material controller 92 adds weighing record data including the weights and moisture content information of each of the multiple types of raw materials that have been weighed to the tracking data including the cycle identification information associated with the weighing / discharge setting data.

[0082] Next, the raw material charging control device 90 executes steps S46, S47, and S48. In step S46, the raw material controller 92 causes each of the hoppers 10 to discharge the weighed raw materials onto the conveyor 20 based on the weighing and discharge setting data. In step S47, the raw material controller 92 adds discharge history data including the discharge speeds of each of the raw materials to tracking data including cycle identification information associated with the weighing and discharge setting data. In step S48, the raw material controller 92 transfers the tracking data to which the discharge history data has been added to the furnace top controller 93.

[0083] Next, the raw material charging control device 90 executes steps S51, S52, and S53. In step S51, the furnace top controller 93 waits for multiple types of raw materials to be charged into the bunker 30 by the conveyor 20. In step S52, the furnace top controller 93 adds dump setting data to tracking data including cycle identification information associated with the dump setting data. In step S53, the furnace top controller 93 adds environmental data including the above-mentioned environmental information to the tracking data to which the dump setting data has been added.

[0084] Next, the raw material charging control device 90 executes steps S54, S55, and S56. In step S54, the furnace top controller 93 causes the gate 40 to charge the raw materials input into the bunker 30 into the blast furnace 2 based on the dump setting data. In step S55, the furnace top controller 93 adds charging history data including the identification information of the bunker 30, the opening degree of the gate 40, and the raw material charging speed to tracking data including the cycle identification information associated with the dump setting data. In step S56, the furnace top controller 93 transfers the tracking data to which the charging history data has been added to the server device 100.

[0085] Next, the server device 100 executes step S57. In step S57, the data collection unit 118 acquires a learning record from the tracking data to which performance data has been added by the furnace top controller 93, which is the final-stage controller 901. For example, the opening information acquisition unit 111 acquires the opening information of the gate 40 from the tracking data. The speed information acquisition unit 112 acquires the above-mentioned speed information from the tracking data. The weight information acquisition unit 113 acquires the above-mentioned type information and weight information from the tracking data. The flowability information acquisition unit 114 acquires the above-mentioned flowability information from the tracking data. The moisture information acquisition unit 115 acquires the above-mentioned moisture information from the tracking data. The environmental information acquisition unit 116 acquires the above-mentioned environmental information from the tracking data. The bunker information acquisition unit 117 acquires the above-mentioned bunker identification information from the tracking data. The data collection unit 118 acquires, as a learning record, the information acquired from the tracking data by the opening information acquisition unit 111, the speed information acquisition unit 112, the weight information acquisition unit 113, the flow characteristics information acquisition unit 114, the moisture information acquisition unit 115, the environmental information acquisition unit 116, and the bunker information acquisition unit 117 as described above. This completes the procedure for acquiring the learning record. After the learning model is generated, the charging conditions and target charging speed of the newly added raw material may be acquired from the tracking data by the procedure described above.

[0086] The procedure for acquiring the learning record shown above is an example and can be changed. For example, in step S41, the schedule controller 91 may send the dump setting data associated with the cycle identification information to the raw material controller 92 instead of the furnace top controller 93. In this case, in step S43, the raw material controller 92 may add the dump setting data to the tracking data in addition to the weighing discharge setting data. In step S52, the furnace top controller 93 may read the dump setting data from the tracking data.

[0087] [Summary] The above-described embodiment includes the following configuration: (1) A raw material charging control device 90 for a blast furnace 2, including: a data collection unit 118 that accumulates learning records in a data storage unit, the learning records including the opening of a gate 40 through which raw materials are charged from a bunker 30 that accommodates raw materials in the upper part of the blast furnace 2, the charging rate of the raw materials from the bunker 30, and charging conditions including raw material characteristics; an offline learning unit 211 that generates a prediction model that expresses the relationship between the opening, the charging rate, and the charging conditions using a multi-stage input-output relationship through machine learning based on the multiple learning records accumulated in the data storage unit; a command generation unit 220 that generates an opening command value for the gate 40 corresponding to the target charging rate based on the charging conditions and target charging rate of newly charged raw materials and the prediction model; and a gate 40 controller that corresponds the opening to the opening command value. By using the prediction model that expresses the relationship between the opening, the charging rate, and the charging conditions using a multi-stage input-output relationship, the opening command value corresponding to the target charging rate can be generated with higher accuracy. By making the opening correspond to the opening command value generated with high accuracy, it is possible to suppress the error of the charging speed with respect to the charging target speed, which is therefore effective in controlling the charging speed of the raw materials from the bunker 30 into the blast furnace 2 with high accuracy.

[0088] (2) In the raw material charging control device 90 for a blast furnace 2 described in (1), the raw materials include multiple types of raw materials, the offline learning unit 211 generates a prediction model based on multiple learning records, each of which includes, as charging conditions, type information indicating each type of the multiple types of raw materials and weight information indicating each weight of the multiple types of raw materials, and the command generating unit 220 generates an opening command value based on the charging conditions including the type information and weight information of newly charged raw materials. The influence of the raw material composition on the relationship between the opening and the charging rate can be reflected in the prediction model. Therefore, the reliability of the prediction model can be further improved.

[0089] (3) The raw material charging control device 90 for the blast furnace 2 described in (2) is provided with a weighing scale 12 that detects the weight of each of multiple types of raw materials before the raw materials are charged into the bunker 30, the raw material charging control device 90 further includes a weight information acquisition unit that acquires weight information based on the detection results by the weighing scale 12, the offline learning unit 211 generates a prediction model based on multiple learning records each including weight information acquired from the weighing scale 12 as charging conditions, and the command generation unit 220 generates an opening command value for newly charged raw materials based on the charging conditions including the weight information acquired from the weighing scale 12. The weight information of the raw materials before they are charged into the bunker 30 does not include errors caused by the environment within the bunker 30. This improves the reliability of the weight information in each learning record. This further improves the reliability of the prediction model.

[0090] (4) The raw material charging control device 90 for the blast furnace 2 according to (2) or (3), wherein the offline learning unit 211 generates a prediction model based on a plurality of learning records, each including the flow characteristics of a plurality of types of raw materials, as charging conditions, and the command generating unit 220 generates an opening command value based on the charging conditions including the flow characteristics of the newly charged raw materials. By including the flow characteristics of each constituent material in the charging conditions in addition to information representing the raw material composition, the reliability of the prediction model can be further improved.

[0091] (5) The raw material charging control device 90 for the blast furnace 2 according to (4), further comprising: a plurality of hoppers 10 each containing a plurality of types of raw materials; and a conveyor 20 for transporting the plurality of types of raw materials discharged from the plurality of hoppers 10 to the bunker 30, the raw material charging control device 90 further comprising a flow characteristic acquisition unit that acquires flow characteristics based on the discharge speed of each of the plurality of types of raw materials when the plurality of types of raw materials are discharged from the corresponding hopper 10, the offline learning unit 211 generates a prediction model based on a plurality of learning records each including the flow characteristics acquired based on the discharge speed as a charging condition, and the command generation unit 220 generates an opening command value for newly charged raw materials based on the charging conditions including the flow characteristics acquired based on the discharge speed. By including the flow characteristics actually measured immediately before charging into the bunker 30 in the charging conditions, the reliability of the prediction model can be further improved.

[0092] (6) The raw material charging control device 90 for the blast furnace 2 according to any one of (2) to (5), wherein the offline learning unit 211 generates a prediction model based on a plurality of learning records, each including moisture content information representing the moisture content of each of a plurality of types of raw materials, as charging conditions, and the command generating unit 220 generates an opening command value based on the charging conditions including the moisture content information of newly charged raw materials. By including information on the moisture content of each constituent material in addition to information representing the composition of the raw materials in the charging conditions, the reliability of the prediction model can be further improved.

[0093] (7) The raw material charging control device 90 for the blast furnace 2 according to any one of (1) to (6), wherein the offline learning unit 211 generates a prediction model based on a plurality of learning records each including environmental information about the inside of the bunker 30 and the inside of the blast furnace 2 as charging conditions, and the command generating unit 220 generates an opening command value based on the charging conditions including environmental information about the state in which newly charged raw materials are accommodated in the bunker 30. By including environmental information about the inside of the bunker 30 and the inside of the blast furnace 2 in the extraction conditions, the reliability of the prediction model can be further improved.

[0094] (8) The raw material charging control device 90 for the blast furnace 2 described in (7) above, wherein the offline learning unit 211 generates a prediction model based on a plurality of learning records, each of which includes, as a charging condition, information on the differential pressure between the inside of the bunker 30 and the inside of the blast furnace 2, and the command generating unit 220 generates an opening command value based on the charging condition including information on the differential pressure when newly charged raw materials are accommodated in the bunker 30. By including information on the differential pressure in the extraction conditions, the reliability of the prediction model can be further improved.

[0095] (9) A raw material charging control device 90 for a blast furnace 2 according to any one of (1) to (8), wherein a plurality of bunkers 30 are provided in the upper part of the blast furnace 2, the offline learning unit 211 generates a prediction model based on a plurality of learning records, each of which includes bunker 30 identification information indicating which of the plurality of bunkers 30 has been used as the bunker 30 as a charging condition, and the command generating unit 220 generates an opening command value based on the charging condition including the bunker 30 identification information indicating which of the plurality of bunkers 30 will be used as the bunker 30 for newly charged raw material. The influence of individual differences between the bunkers 30 can be reflected in the prediction model. Therefore, the reliability of the prediction model can be further improved.

[0096] (10) The raw material charging control device 90 for a blast furnace 2 according to any one of (1) to (9), wherein the offline learning unit 211 generates a prediction model to output a charging rate in response to inputs of charging conditions and an aperture, and the command generation unit 220 inputs the charging conditions of newly charged raw materials and an aperture command value into the prediction model to calculate a charging rate and repeatedly changes the aperture command value to calculate an aperture command value corresponding to a target charging rate. The search range for the aperture command value corresponding to the target charging rate can be freely adjusted. This prevents the prediction model from generating a low-reliability aperture command value in a range where the reliability of the aperture command value is insufficient, thereby enabling more reliable control of the aperture of the gate 40.

[0097] (11) The raw material charging control device 90 for a blast furnace 2 according to (10), further comprising: a profile generating unit 231 configured to generate a profile representing the relationship between the charging conditions, the charging rate, and the opening command value by repeatedly inputting the charging conditions and the opening command value into a prediction model to calculate a charging rate and changing at least one of the charging conditions and the opening command value; and an initial value calculating unit configured to calculate an initial value of the opening command value based on the profile and the target charging rate, wherein the command generating unit 220 starts the repetition of inputting the charging conditions of newly charged raw materials and the opening command value into the prediction model to calculate a charging rate and changing the opening command value of the gate 40, using the opening command value as an initial value. This allows for a reduction in the time required to search for the opening command value corresponding to the target charging rate.

[0098] (12) The raw material charging control device 90 for the blast furnace 2 according to any one of (1) to (11), wherein the prediction model is a neural network, and the offline learning unit 211 generates the prediction model by deep learning. This can further improve the reliability of the prediction model.

[0099] (13) The raw material charging control device 90 for the blast furnace 2 according to any one of (1) to (11), wherein the offline learning unit 211 generates a prediction model by a gradient boosting method. This can further improve the reliability of the prediction model.

[0100] (14) The raw material charging control device 90 for the blast furnace 2 according to any one of (1) to (11), wherein the offline learning unit 211 generates a prediction model by a random forest method. This can further improve the reliability of the prediction model.

[0101] (15) A raw material charging control device 90 according to (5) or (6), which includes a plurality of controllers 901 including a furnace top controller 93, and which sequentially executes a plurality of stages of control, from discharging a plurality of types of raw materials from a plurality of hoppers 10 to charging the plurality of types of raw materials from a bunker 30 into a blast furnace 2, and sequentially transfers the tracking data from the first-stage controller 901 to the highest-stage controller 901 while adding performance data of the executed control to the tracking data, and the data collection unit 118 acquires a learning record from the tracking data to which performance data has been added by the last-stage controller 901. Performance data is accumulated in the tracking data in stages as the plurality of stages of control are sequentially executed, and a learning record can be easily acquired from the tracking data in which all performance data has been collected.

[0102] (16) The raw material charging control device 90 according to (15), wherein the multi-stage controller 901 further includes a raw material controller 92 that controls the multiple hoppers 10, the raw material controller 92 adds performance data including the weight and discharge speed of each of the multiple types of raw materials to the tracking data, and the furnace top controller 93 adds performance data including the gate opening and charging speed to the tracking data. By including the performance data required for the learning record in the tracking data, the learning record can be acquired more easily.

[0103] (17) A method for generating an opening command value, the method comprising: storing learning records in a data storage unit, the learning records including the opening of a gate 40 through which raw materials are charged from a bunker 30 that stores raw materials in the upper part of the blast furnace 2, the charging rate of the raw materials from the bunker 30, and charging conditions including characteristics of the raw materials; generating a prediction model that represents the relationship between the opening, the charging rate, and the charging conditions using a multi-stage input-output relationship based on the plurality of learning records stored in the data storage unit; and generating an opening command value for the gate 40 that corresponds to the charging target rate based on the charging conditions and target charging rate of the newly charged raw materials and the prediction model.

[0104] (18) A program for causing an apparatus to execute the following operations: storing learning records in a data storage unit, the learning records including the opening of a gate 40 through which raw materials are charged from a bunker 30 that stores raw materials in the upper part of a blast furnace 2, the charging rate of the raw materials from the bunker 30, and charging conditions including the characteristics of the raw materials; generating a prediction model that represents the relationship between the opening, the charging rate, and the charging conditions using a multi-stage input-output relationship based on the multiple learning records stored in the data storage unit; and generating an opening command value for the gate 40 that corresponds to the charging target rate based on the charging conditions and target charging rate of newly added raw materials and the prediction model.

[0105] 2...blast furnace, 90...raw material charging control device, 30...bunker, 40...gate, 10...hopper, 12...weigher, 20...conveyor, 118...data collection unit, 211...offline learning unit, 220...command generation unit, 221...initial value generation unit, 231...profile generation unit.

Claims

1. The opening degree of a gate for charging raw materials from a bunker that stores raw materials at the top of the blast furnace; The charging rate of the raw material from the bunker; and Charging conditions including characteristics of the raw materials; A data collection unit that accumulates learning records including the above in a data storage unit; an offline learning unit that generates a prediction model that represents a relationship between the opening degree, the charging rate, and the charging conditions by a multi-stage input / output relationship through machine learning based on a plurality of learning records stored in the data storage unit; A command generating unit that generates an opening command value of the gate corresponding to the target charging speed based on the charging conditions and the target charging speed of the newly charged raw material and the prediction model; a furnace top controller that causes the opening to correspond to the opening command value; A raw material charging control device for a blast furnace.

2. The raw material includes a plurality of raw materials, The offline learning unit generates the prediction model based on the plurality of learning records, each of which includes type information representing the type of each of the plurality of types of raw materials and weight information representing the weight of each of the plurality of types of raw materials, as the charging conditions; The command generating unit generates the opening command value based on the charging conditions including the type information and the weight information of the newly charged raw material. The raw material charging control device for a blast furnace according to claim 1.

3. a weight scale is provided to detect the weight of each of the plurality of types of raw materials before the raw materials are put into the bunker; The raw material charging control device further includes a weight information acquisition unit that acquires the weight information based on the detection result by the weighing scale, The offline learning unit generates the prediction model based on the plurality of learning records each including the weight information acquired from the weighing scale as the charging condition, The command generating unit generates the opening command value based on the charging conditions including the weight information acquired from the weighing scale for the newly charged raw material. The raw material charging control device for a blast furnace according to claim 2.

4. The offline learning unit generates the prediction model based on the plurality of learning records, each of which includes flow characteristics of the plurality of types of raw materials as the charging conditions, and The command generating unit generates the opening command value based on the charging conditions including the flow characteristics of the newly charged raw material. The raw material charging control device for a blast furnace according to claim 2.

5. A plurality of hoppers for accommodating the plurality of types of raw materials, respectively, and a conveyor for transporting the plurality of types of raw materials discharged from the plurality of hoppers to the bunker are provided around the blast furnace, The raw material charging control device further includes a flow characteristic acquisition unit that acquires the flow characteristics based on the discharge speed when each of the plurality of types of raw materials is discharged from the corresponding hopper, The offline learning unit generates the prediction model based on the plurality of learning records, each of which includes the flow characteristics acquired based on the discharge rate as the charging condition; The command generating unit generates the opening command value based on the charging conditions including the flow characteristics acquired based on the discharge rate for the newly charged raw material. The raw material charging control device for a blast furnace according to claim 4.

6. the offline learning unit generates the prediction model based on the plurality of learning records, each of which includes moisture content information representing a moisture content of each of the plurality of types of raw materials as the charging conditions; The command generating unit generates the opening command value based on the charging conditions including the moisture content information of the newly charged raw material. The raw material charging control device for a blast furnace according to claim 2.

7. The offline learning unit generates the prediction model based on the plurality of learning records each including environmental information in the bunker and in the blast furnace as the charging conditions, The command generating unit generates the opening command value based on the charging conditions including the environmental information in a state in which the newly charged raw material is stored in the bunker. The raw material charging control device for a blast furnace according to any one of claims 1 to 6.

8. The offline learning unit generates the prediction model based on the plurality of learning records each including information on a differential pressure between the inside of the bunker and the inside of the blast furnace as the charging condition, The command generating unit generates the opening command value based on the charging conditions including information on the differential pressure in a state in which the newly charged raw material is stored in the bunker. The raw material charging control device for a blast furnace according to claim 7.

9. A plurality of bunkers are provided at the upper portion of the blast furnace, The offline learning unit generates the prediction model based on the plurality of learning records, each of which includes bunker identification information indicating which of the plurality of bunkers was used as the bunker as the charging condition, The command generating unit generates the opening command value based on the charging conditions including the bunker identification information indicating which of the plurality of bunkers is to be used as the bunker for the newly charged raw material. The raw material charging control device for a blast furnace according to any one of claims 1 to 6.

10. The offline learning unit generates the prediction model so as to output the charging rate in response to inputs of the charging conditions and the opening degree, The command generating unit calculates the charging speed by inputting the charging conditions of the newly charged raw material and the opening command value into the prediction model, and repeats changing the opening command value to calculate the opening command value corresponding to the charging target speed. The raw material charging control device for a blast furnace according to any one of claims 1 to 6.

11. A profile generating unit that generates a profile representing the relationship between the charging condition, the charging rate, and the opening command value by repeatedly inputting the charging condition and the opening command value into the prediction model to calculate the charging rate and changing at least one of the charging condition and the opening command value; Further provided is an initial value calculation unit that calculates an initial value of the opening command value based on the profile and the charging target speed; The command generating unit inputs the charging conditions of the newly charged raw material and the opening command value into the prediction model to calculate the charging rate and changes the opening command value, and starts repeating the process of changing the opening command value with the opening command value as an initial value. The raw material charging control device for a blast furnace according to claim 10.

12. the predictive model is a neural network; The offline learning unit generates the prediction model by deep learning. The raw material charging control device for a blast furnace according to any one of claims 1 to 6.

13. The offline learning unit generates the prediction model by a gradient boosting method. The raw material charging control device for a blast furnace according to any one of claims 1 to 6.

14. The offline learning unit generates the prediction model by a random forest method. The raw material charging control device for a blast furnace according to any one of claims 1 to 6.

15. A multi-stage controller including the furnace top controller is provided, The multi-stage controller comprises: Sequentially executing a plurality of stages of control from discharging the plurality of types of raw materials from the plurality of hoppers to charging the plurality of types of raw materials from the bunker into a blast furnace; adding performance data of the executed control to the tracking data, and sequentially transferring the tracking data from the first controller to the last controller; The raw material charging control device according to claim 5 , wherein the data collection unit acquires the learning record from the tracking data to which the performance data has been added by the last stage controller.

16. The multiple stage controller further includes a raw material controller that controls the multiple hoppers, the raw material controller adds the performance data including the weight and the discharge rate of each of the plurality of types of raw materials to the tracking data; The material charging control device according to claim 15, wherein the furnace top controller adds the performance data including the opening degree and the charging rate to the tracking data.

17. The opening degree of a gate for charging raw materials from a bunker that stores raw materials at the top of the blast furnace; The charging rate of the raw material from the bunker; and Charging conditions including raw material characteristics; storing a learning record including the learning record in a data storage unit; Generating a prediction model that represents a relationship between the opening, the charging speed, and the charging conditions by a multi-stage input / output relationship based on a plurality of learning records stored in the data storage unit; Generating an opening command value of the gate corresponding to the target charging speed based on the charging conditions and the target charging speed of the newly charged raw material and the prediction model; A method for generating an opening command value, comprising:

18. The opening degree of a gate for charging raw materials from a bunker that stores raw materials at the top of the blast furnace; The charging rate of the raw material from the bunker; and Charging conditions including raw material characteristics; storing a learning record including the learning record in a data storage unit; Generating a prediction model that represents a relationship between the opening, the charging speed, and the charging conditions by a multi-stage input / output relationship based on a plurality of learning records stored in the data storage unit; Generating an opening command value of the gate corresponding to the target charging speed based on the charging conditions and the target charging speed of the newly charged raw material and the prediction model; A program for causing a device to execute the above.