JP7380604B2 - Learning model generation method, learning model generation device, blast furnace control guidance method, and hot metal manufacturing method - Google Patents
Learning model generation method, learning model generation device, blast furnace control guidance method, and hot metal manufacturing method Download PDFInfo
- Publication number
- JP7380604B2 JP7380604B2 JP2021002835A JP2021002835A JP7380604B2 JP 7380604 B2 JP7380604 B2 JP 7380604B2 JP 2021002835 A JP2021002835 A JP 2021002835A JP 2021002835 A JP2021002835 A JP 2021002835A JP 7380604 B2 JP7380604 B2 JP 7380604B2
- Authority
- JP
- Japan
- Prior art keywords
- blast furnace
- hot metal
- data
- operation amount
- operator
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims description 93
- 239000002184 metal Substances 0.000 title claims description 61
- 229910052751 metal Inorganic materials 0.000 title claims description 61
- 238000004519 manufacturing process Methods 0.000 title claims description 7
- 230000008569 process Effects 0.000 claims description 66
- 238000009423 ventilation Methods 0.000 claims description 34
- 238000010801 machine learning Methods 0.000 claims description 28
- 230000003542 behavioural effect Effects 0.000 claims description 20
- 230000006399 behavior Effects 0.000 claims description 17
- 239000000571 coke Substances 0.000 claims description 12
- 239000003245 coal Substances 0.000 claims description 7
- 238000013527 convolutional neural network Methods 0.000 claims description 5
- 230000009471 action Effects 0.000 description 9
- 238000010586 diagram Methods 0.000 description 7
- 230000000087 stabilizing effect Effects 0.000 description 6
- 230000008859 change Effects 0.000 description 3
- 230000000052 comparative effect Effects 0.000 description 3
- 239000002994 raw material Substances 0.000 description 3
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 229910000831 Steel Inorganic materials 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000004090 dissolution Methods 0.000 description 1
- 239000000446 fuel Substances 0.000 description 1
- 238000002309 gasification Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 229910052742 iron Inorganic materials 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000006263 metalation reaction Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
Images
Landscapes
- Testing And Monitoring For Control Systems (AREA)
- Manufacture Of Iron (AREA)
- Blast Furnaces (AREA)
Description
本発明は、学習モデル生成方法、学習モデル生成装置、高炉の制御ガイダンス方法、及び溶銑の製造方法に関する。 The present invention relates to a learning model generation method, a learning model generation device, a blast furnace control guidance method, and a method for producing hot metal.
製鉄業における高炉プロセスにおいて、溶銑温度は重要な管理指標である。特に近年の高炉操業は、原燃料コストの合理化を追及すべく、低コークス比及び高微粉炭比の条件下で行われており、炉況が不安定化しやすいために、溶銑温度のばらつき低減のニーズは大きい。ところが、高炉プロセスでは、コークスのガス化や鉱石の還元及び溶解等の種々の反応が生じている。また、炉内に固体が充填された状態で操業が行われるため、プロセス全体の熱容量が大きく、アクションに対する応答の時定数が長い。このため、溶銑温度のばらつき低減のためには、高炉の複雑なダイナミクスを適切に考慮した溶銑温度の制御則が求められる。 Hot metal temperature is an important control index in the blast furnace process in the steel industry. In particular, blast furnace operations in recent years have been carried out under conditions of low coke ratios and high pulverized coal ratios in order to rationalize raw material and fuel costs. The need is great. However, in the blast furnace process, various reactions occur such as gasification of coke and reduction and dissolution of ore. Additionally, since the furnace is operated with solids filled in it, the heat capacity of the entire process is large and the time constant of response to action is long. Therefore, in order to reduce variations in hot metal temperature, a control law for hot metal temperature that appropriately takes into account the complex dynamics of the blast furnace is required.
このような背景から、物理モデルに基づいた溶銑温度制御方法（特許文献１参照）や過去データに基づいた溶銑温度制御方法（特許文献２参照）が提案されている。具体的には、特許文献１には、実測の炉頂ガス組成に合致するように物理モデル中のガス還元速度パラメータを調整しつつ、現在の操作量が保持されたと仮定して物理モデルを用いて溶銑温度の予測値を算出し、算出された予測値に基づいて溶銑温度を制御する方法が記載されている。また、特許文献２には、過去の操業事例の中から現在の操業条件に類似する操業事例を抽出し、抽出された操業事例に基づいて溶銑温度の予測値を算出し、算出された予測値に基づいて溶銑温度を制御する方法が記載されている。 Against this background, a hot metal temperature control method based on a physical model (see Patent Document 1) and a hot metal temperature control method based on past data (see Patent Document 2) have been proposed. Specifically, Patent Document 1 discloses that a physical model is used on the assumption that the current manipulated variable is maintained while adjusting the gas reduction rate parameter in the physical model to match the actually measured furnace top gas composition. A method is described in which a predicted value of hot metal temperature is calculated based on the calculated predicted value, and the hot metal temperature is controlled based on the calculated predicted value. In addition, Patent Document 2 discloses that operation cases similar to current operating conditions are extracted from past operation cases, a predicted value of hot metal temperature is calculated based on the extracted operation cases, and the calculated predicted value is A method for controlling hot metal temperature based on the following is described.
しかしながら、特許文献１に記載の方法によれば、荷下り速度（原料降下速度）の変動や原料中の鉄分の変動等の物理モデルでは考慮することが困難、且つ、オンライン測定が困難な外乱が発生した場合、溶銑温度の予測精度が低下し、結果として、溶銑温度の制御精度が低下する可能性がある。一方、特許文献２に記載の方法では、溶銑温度の予測精度が担保されるのは現在の操業条件に類似する操業事例がある場合に限られ、今までに実績のない操業条件に対しては溶銑温度の予測精度が低下し、結果として、溶銑温度の制御精度が低下する可能性がある。 However, according to the method described in Patent Document 1, disturbances that are difficult to take into account in physical models, such as fluctuations in unloading speed (raw material descending speed) and fluctuations in iron content in raw materials, and which are difficult to measure online, can be detected. If this occurs, the prediction accuracy of the hot metal temperature may decrease, and as a result, the control accuracy of the hot metal temperature may decrease. On the other hand, in the method described in Patent Document 2, the prediction accuracy of hot metal temperature is guaranteed only when there is an operation example similar to the current operation conditions, and it is not possible to guarantee the prediction accuracy of the hot metal temperature only when there are operation cases similar to the current operation conditions. The prediction accuracy of the hot metal temperature may be reduced, and as a result, the control accuracy of the hot metal temperature may be reduced.
本発明は、上記課題に鑑みてなされたものであって、その目的は、高炉プロセスの操業状態を精度よく制御可能な学習モデルを生成可能な学習モデル生成方法及び学習モデル生成装置を提供することにある。また、本発明の他の目的は、高炉の溶銑温度及び通気状況を精度よく制御可能な高炉の制御ガイダンス方法を提供することにある。さらに、本発明の他の目的は、溶銑を歩留まりよく製造可能な溶銑の製造方法を提供することにある。 The present invention has been made in view of the above problems, and its purpose is to provide a learning model generation method and a learning model generation device that can generate a learning model that can accurately control the operational status of a blast furnace process. It is in. Another object of the present invention is to provide a blast furnace control guidance method that can accurately control the temperature of hot metal and ventilation conditions in the blast furnace. Furthermore, another object of the present invention is to provide a method for producing hot metal that can produce hot metal with a high yield.
本発明に係る学習モデル生成方法は、溶銑温度、羽口埋込温度、通気抵抗、炉熱指数、及びコークス比のうちの少なくとも一つを含む高炉プロセスの操業状態を示す観測量の履歴データを画像化した画像データを入力データ、前記履歴データに基づいてオペレータが決定した微粉炭比及び送風湿分の少なくとも一方を含む高炉プロセスの操作量を出力データとして、オペレータが高炉プロセスの操作量を決定する行動モデルを機械学習し、機械学習された行動モデルを出力するステップを含み、前記ステップは、溶銑温度に関係する観測量の履歴データを画像化した画像データを入力データ、溶銑温度に関係する観測量の履歴データに基づいてオペレータが決定した高炉プロセスの操作量を出力データとした第１機械学習と、高炉内の通気状況に関係する観測量の履歴データを画像化した画像データを入力データ、高炉内の通気状況に関係する観測量の履歴データに基づいてオペレータが決定した高炉プロセスの操作量を出力データとした第２機械学習と、を分離して実行することにより、オペレータが溶銑温度を制御する操作量を決定する第１行動モデル及びオペレータが高炉内の通気状況を制御する操作量を決定する第２行動モデルを出力するステップを含むことを特徴とする。 The learning model generation method according to the present invention collects historical data of observables indicating the operational status of the blast furnace process, including at least one of hot metal temperature, tuyere embedding temperature, ventilation resistance, furnace heat index, and coke ratio. The operator determines the operation amount of the blast furnace process by using the imaged image data as input data and the operation amount of the blast furnace process including at least one of a pulverized coal ratio and blast moisture determined by the operator based on the historical data as output data. The step includes a step of performing machine learning on a behavioral model and outputting the machine-learned behavioral model, and the step includes inputting image data obtained by imaging historical data of observed quantities related to hot metal temperature, and inputting image data related to hot metal temperature. The first machine learning uses the output data as the operation amount of the blast furnace process determined by the operator based on the historical data of the observed amount, and the input data uses image data that is an image of the historical data of the observed amount related to the ventilation status in the blast furnace. , and the second machine learning, which uses as output data the operation amount of the blast furnace process determined by the operator based on the historical data of the observed amount related to the ventilation status in the blast furnace. The method is characterized in that it includes the step of outputting a first behavior model for determining an operation amount for controlling the ventilation condition of the blast furnace and a second behavior model for determining an operation amount for the operator to control the ventilation situation in the blast furnace.
本発明に係る学習モデル生成方法は、上記発明において、前記行動モデルは、畳み込みニューラルネットワークであることを特徴とする。 The learning model generation method according to the present invention is characterized in that, in the above invention, the behavioral model is a convolutional neural network.
本発明に係る学習モデル生成装置は、溶銑温度、羽口埋込温度、通気抵抗、炉熱指数、及びコークス比のうちの少なくとも一つを含む高炉プロセスの操業状態を示す観測量の履歴データを画像化した画像データを入力データ、前記履歴データに基づいてオペレータが決定した微粉炭比及び送風湿分の少なくとも一方を含む高炉プロセスの操作量を出力データとして、オペレータが高炉プロセスの操作量を決定する行動モデルを機械学習し、機械学習された行動モデルを出力する手段を備え、前記手段は、溶銑温度に関係する観測量の履歴データを画像化した画像データを入力データ、溶銑温度に関係する観測量の履歴データに基づいてオペレータが決定した高炉プロセスの操作量を出力データとした第１機械学習と、高炉内の通気状況に関係する観測量の履歴データを画像化した画像データを入力データ、高炉内の通気状況に関係する観測量の履歴データに基づいてオペレータが決定した高炉プロセスの操作量を出力データとした第２機械学習と、を分離して実行することにより、オペレータが溶銑温度を制御する操作量を決定する第１行動モデル及びオペレータが高炉内の通気状況を制御する操作量を決定する第２行動モデルを出力することを特徴とする。 The learning model generation device according to the present invention generates historical data of observables indicating the operational status of the blast furnace process, including at least one of hot metal temperature, tuyere embedding temperature, ventilation resistance, furnace heat index, and coke ratio. The operator determines the operation amount of the blast furnace process by using the imaged image data as input data and the operation amount of the blast furnace process including at least one of a pulverized coal ratio and blast moisture determined by the operator based on the historical data as output data. means for machine learning a behavioral model to perform machine learning and outputting the machine learned behavioral model, the means for inputting image data obtained by converting historical data of observed quantities related to hot metal temperature into images, and inputting image data related to hot metal temperature. The first machine learning uses the output data as the operation amount of the blast furnace process determined by the operator based on the historical data of the observed amount, and the input data uses image data that is an image of the historical data of the observed amount related to the ventilation status in the blast furnace. , and the second machine learning, which uses as output data the operation amount of the blast furnace process determined by the operator based on the historical data of the observed amount related to the ventilation status in the blast furnace. The blast furnace is characterized by outputting a first behavior model that determines the amount of operation to control the operator, and a second behavior model that determines the amount of operation that the operator uses to control the ventilation situation in the blast furnace.
本発明に係る高炉の制御ガイダンス方法は、本発明に係る学習モデル生成方法によって生成された第１行動モデル及び第２行動モデルを用いて高炉の溶銑温度を目標温度に制御可能な前記操作量及び高炉内の通気状況を安定化させることが可能な前記操作量を決定し、決定した操作量に従って該操作量の変更ガイダンスを行うステップを含むことを特徴とする。 The blast furnace control guidance method according to the present invention includes the operation amount and the control amount that can control the temperature of hot metal in the blast furnace to a target temperature using the first behavior model and the second behavior model generated by the learning model generation method according to the present invention. The method is characterized by including the step of determining the amount of operation capable of stabilizing the ventilation situation in the blast furnace, and providing guidance for changing the amount of operation in accordance with the determined amount of operation.
本発明に係る溶銑の製造方法は、本発明に係る高炉の制御ガイダンス方法を用いて溶銑を製造するステップを含むことを特徴とする。 The method for producing hot metal according to the present invention is characterized by including the step of producing hot metal using the blast furnace control guidance method according to the present invention.
本発明に係る学習モデル生成方法及び学習モデル生成装置によれば、高炉プロセスの操業状態を精度よく制御可能な学習モデルを生成することができる。また、本発明に係る、高炉の制御ガイダンス方法によれば、高炉の溶銑温度及び通気状況を精度よく制御することができる。さらに、本発明に係る溶銑の製造方法によれば、溶銑を歩留まりよく製造することができる。 According to the learning model generation method and learning model generation device according to the present invention, it is possible to generate a learning model that can accurately control the operating state of a blast furnace process. Further, according to the blast furnace control guidance method according to the present invention, the temperature of hot metal and the ventilation status of the blast furnace can be controlled with high accuracy. Furthermore, according to the method for producing hot metal according to the present invention, hot metal can be produced with a high yield.
以下、図面を参照して、本発明の一実施形態である学習モデル生成装置の構成及びその動作について説明する。 Hereinafter, the configuration and operation of a learning model generation device that is an embodiment of the present invention will be described with reference to the drawings.
〔構成〕
まず、図１を参照して、本発明の一実施形態である学習モデル生成装置の構成について説明する。
〔composition〕
First, with reference to FIG. 1, the configuration of a learning model generation device that is an embodiment of the present invention will be described.
図１は、本発明の一実施形態である学習モデル生成装置の構成を示すブロック図である。図１に示すように、本発明の一実施形態である学習モデル生成装置１は、オペレータが高炉プロセスの操作量を決定する行動モデルを機械学習により生成する装置であり、プロセッサやメモリ等を備える周知の情報処理装置によって構成されている。本実施形態では、学習モデル生成装置１は、プロセッサがコンピュータプログラムを実行することにより、モデル学習部１１及びモデル出力部１２として機能する。各部の機能については後述する。 FIG. 1 is a block diagram showing the configuration of a learning model generation device that is an embodiment of the present invention. As shown in FIG. 1, a learning model generation device 1, which is an embodiment of the present invention, is a device that uses machine learning to generate a behavioral model for determining the operation amount of a blast furnace process by an operator, and is equipped with a processor, memory, etc. It is configured by a well-known information processing device. In this embodiment, the learning model generation device 1 functions as a model learning section 11 and a model output section 12 by a processor executing a computer program. The functions of each part will be described later.
学習モデル生成装置１には、機械学習の際に用いる学習データが格納されている履歴データデータベース（履歴データＤＢ）２がデータ読み取り可能な形態で接続されている。本実施形態では、履歴データＤＢ２には、溶銑温度、羽口埋込温度、通気抵抗、炉熱指数、及びコークス比のうちの少なくとも一つを含む高炉プロセスの操業状態を示す１以上の観測量の履歴データを画像化した画像データ、と、履歴データに基づいてオペレータが決定した微粉炭比及び送風湿分の少なくとも一方を含む高炉プロセスの操作量とが、関連付けされて学習データとして格納されている。 A history data database (history data DB) 2 in which learning data used in machine learning is stored is connected to the learning model generation device 1 in a data readable form. In the present embodiment, the historical data DB2 includes one or more observed quantities indicating the operational status of the blast furnace process, including at least one of hot metal temperature, tuyere embedding temperature, ventilation resistance, furnace heat index, and coke ratio. The image data obtained by converting the historical data of There is.
ここで、画像データの形態は、一方の軸を時間軸とし、他方の軸を観測量とする観測量の履歴データを、時間軸をあわせて時間軸とは異なる方向に１以上の観測量を配置させた２次元配置の画像データとしている。各観測量の履歴表示は通常のトレンドグラフのような線図でもよいし、各履歴データの数値に色及び／又は色の濃淡を対応づけて得られる画像とする、いわゆるコンター図としてもよい。また、画像データには、高炉プロセスの操作量の履歴データを含めるようにしてもよい。さらに、観測量や操作量の履歴データに対しては、値域を統一するために画像データとする前に正規化処理を施すことが望ましい。 Here, the format of the image data is historical data of observed quantities with one axis as the time axis and the other axis as the observed quantity, and one or more observed quantities in a direction different from the time axis together with the time axis. This is image data in a two-dimensional arrangement. The history display of each observation amount may be a line diagram such as a normal trend graph, or it may be a so-called contour diagram, which is an image obtained by associating colors and/or color shading with numerical values of each historical data. Further, the image data may include historical data of the manipulated variables of the blast furnace process. Furthermore, it is desirable to perform normalization processing on historical data of observed quantities and manipulated quantities before converting them into image data in order to unify the range.
このような構成を有する学習モデル生成装置１は、以下に示す学習モデル生成処理を実行することにより、オペレータが高炉プロセスの操作量を決定する行動モデルを機械学習により生成する。以下、図２を参照して、学習モデル生成処理を実行する際の学習モデル生成装置１の動作について説明する。 The learning model generation device 1 having such a configuration generates, by machine learning, a behavioral model by which an operator determines the operation amount of the blast furnace process by executing the learning model generation process described below. Hereinafter, with reference to FIG. 2, the operation of the learning model generation device 1 when executing the learning model generation process will be described.
〔学習モデル生成処理〕
図２は、本発明の一実施形態である学習モデル生成処理の流れを示すフローチャートである。図２に示すフローチャートは、学習モデル生成装置１に対して学習モデル生成処理の実行指令が入力されたタイミングで開始となり、学習モデル生成処理はステップＳ１の処理に進む。
[Learning model generation process]
FIG. 2 is a flowchart showing the flow of learning model generation processing according to an embodiment of the present invention. The flowchart shown in FIG. 2 starts at the timing when a command to execute the learning model generation process is input to the learning model generation device 1, and the learning model generation process proceeds to step S1.
ステップＳ１の処理では、モデル学習部１１が、履歴データＤＢ２から学習データを取得する。これにより、ステップＳ１の処理は完了し、学習モデル生成処理はステップＳ２の処理に進む。 In the process of step S1, the model learning unit 11 acquires learning data from the history data DB2. Thereby, the process of step S1 is completed, and the learning model generation process proceeds to the process of step S2.
ステップＳ２の処理では、モデル学習部１１が、ステップＳ１の処理において取得した学習データを用いて、制御対象である高炉プロセスの操業状態を示す１以上の観測量の履歴データを画像化した画像データを入力データ、履歴データに基づいてオペレータが決定した高炉プロセスの操作量を出力データとして、オペレータが高炉プロセスの操作量を決定する行動モデルを機械学習する。また、この際、モデル学習部１１は、溶銑温度に関係する観測量（例えば溶銑温度、羽口埋込温度、コークス比等）の履歴データを画像化した画像データを入力データ、溶銑温度に関係する観測量の履歴データに基づいてオペレータが決定した高炉プロセスの操作量（例えば微粉炭比や送風湿分等）を出力データとした第１機械学習と、高炉内の通気状況に関係する観測量（例えば通気抵抗や炉熱指数等）の履歴データを画像化した画像データを入力データ、高炉内の通気状況に関係する観測量の履歴データに基づいてオペレータが決定した高炉プロセスの操作量（例えば微粉炭比や送風湿分等）を出力データとした第２機械学習と、を分離して実行する。 In the process of step S2, the model learning unit 11 uses the learning data acquired in the process of step S1 to generate image data obtained by converting historical data of one or more observable quantities indicating the operational status of the blast furnace process to be controlled. The machine learns a behavioral model for the operator to determine the operating amount of the blast furnace process, using input data as input data and output data as the operating amount of the blast furnace process determined by the operator based on historical data. In addition, at this time, the model learning unit 11 inputs image data obtained by converting historical data of observed quantities related to the hot metal temperature (for example, hot metal temperature, tuyere embedded temperature, coke ratio, etc.) into images, and inputs image data related to the hot metal temperature. The first machine learning uses as output data the manipulated variables of the blast furnace process (e.g. pulverized coal ratio, blast moisture, etc.) determined by the operator based on the historical data of observed quantities, and the observed quantities related to the ventilation status in the blast furnace. The input data is image data of historical data (e.g. ventilation resistance, furnace heat index, etc.), and the operation amount of the blast furnace process (e.g. The second machine learning is executed separately from the second machine learning using output data such as pulverized coal ratio and air humidity.
ここで、機械学習によって生成する行動モデルは、畳み込みニューラルネットワーク（Convolutional neural network：CNN）であることが望ましい。また、画像データの形態が、一方の軸に時間軸、他方の軸に履歴データを配置した２次元画像である場合には、ＣＮＮの畳み込み演算は時間軸方向にのみに対して行うことが望ましい。また、入力データは、予め定めた連続した所定時間区間の画像データとし、出力データは所定時間区間の直後の操作量の値とするとよい。また、学習データとしては、操作量変更後、予め定めた時間経過後の制御量として定めた観測量の１つ又は複数の値が、それぞれの目標値を中心とした所定範囲内にある、画像データと操作量との組み合わせを用いることが望ましい。これは、オペレータの操作結果として制御量が適切な変化をし、目標値から所定の範囲に制御できたデータを学習データとして利用するためである。これにより、ステップＳ２の処理は完了し、学習モデル生成処理はステップＳ３の処理に進む。 Here, the behavior model generated by machine learning is preferably a convolutional neural network (CNN). Additionally, if the image data is a two-dimensional image with a time axis on one axis and historical data on the other axis, it is desirable to perform CNN convolution operations only in the time axis direction. . Further, the input data may be image data of a predetermined continuous time period, and the output data may be the value of the manipulated variable immediately after the predetermined time period. In addition, the learning data includes an image in which one or more values of the observed quantity determined as the controlled quantity after a predetermined time has elapsed after changing the manipulated variable is within a predetermined range centered on each target value. It is desirable to use a combination of data and manipulated variables. This is because the control amount changes appropriately as a result of the operator's operation, and data that can be controlled within a predetermined range from the target value is used as learning data. Thereby, the process of step S2 is completed, and the learning model generation process proceeds to the process of step S3.
ステップＳ３の処理では、モデル出力部１２が、ステップＳ２の処理において生成された、制御対象である高炉プロセスの操業状態を示す１以上の観測量の履歴データを画像化した画像データを入力データ、履歴データに基づいてオペレータが決定した高炉プロセスの操作量を出力データとする、オペレータが高炉プロセスの操作量を決定する行動モデルを出力する。また、モデル出力部１２は、オペレータが溶銑温度を制御する操作量を決定する第１行動モデル及びオペレータが高炉内の通気状況を制御する操作量を決定する第２行動モデルを出力する。これにより、ステップＳ３の処理は完了し、一連の学習モデル生成処理は終了する。 In the process of step S3, the model output unit 12 inputs image data generated in the process of step S2, which is an image of historical data of one or more observable quantities indicating the operating state of the blast furnace process to be controlled, as input data. Outputs a behavioral model in which the operator determines the operation amount of the blast furnace process, which uses the operation amount of the blast furnace process determined by the operator based on historical data as output data. Furthermore, the model output unit 12 outputs a first behavior model for determining the amount of operation by which the operator controls the hot metal temperature, and a second behavior model for determining the amount of operation by which the operator controls the ventilation situation in the blast furnace. As a result, the process in step S3 is completed, and the series of learning model generation processes ends.
以後、オペレータは、予め定めた所定区間の観測量履歴の画像データを行動モデルに入力することにより得られた操作量に従って高炉プロセスの操業状態を制御することができる。また、第１行動モデル及び第２行動モデルを用いて高炉の溶銑温度を目標温度に制御可能な操作量及び高炉内の通気状況を安定化させることが可能な操作量を決定し、決定した操作量に従って操作量の変更ガイダンスを行うことにより、高炉の溶銑温度及び通気状況を精度よく制御することができる。また、この変更ガイダンスに従って溶銑を製造することにより、溶銑を歩留まりよく製造することができる。 Thereafter, the operator can control the operating state of the blast furnace process according to the manipulated variable obtained by inputting the image data of the observed amount history in a predetermined section into the behavior model. In addition, using the first behavior model and the second behavior model, the operation amount that can control the temperature of hot metal in the blast furnace to the target temperature and the operation amount that can stabilize the ventilation situation in the blast furnace are determined, and the determined operation By providing guidance for changing the operation amount according to the amount, it is possible to accurately control the temperature of hot metal and the ventilation situation in the blast furnace. Moreover, by manufacturing hot metal according to this change guidance, hot metal can be manufactured with a high yield.
以上の説明から明らかなように、本発明の一実施形態である学習モデル生成装置１は、制御対象である高炉プロセスの操業状態を示す１以上の観測量の履歴データを画像化した画像データを入力データ、履歴データに基づいてオペレータが決定した高炉プロセスの操作量を出力データとして、オペレータが高炉プロセスの操作量を決定する行動モデルを機械学習し、機械学習した行動モデルを出力する。これにより、予測対象の画像データを行動モデルに入力することにより得られた操作量に従って高炉プロセスの状態を制御することにより、高炉プロセスの操業状態を精度よく制御することができる。 As is clear from the above description, the learning model generation device 1, which is an embodiment of the present invention, generates image data obtained by converting historical data of one or more observable quantities indicating the operational status of the blast furnace process that is the control target. Using the operating amount of the blast furnace process determined by the operator based on input data and historical data as output data, the operator performs machine learning on a behavioral model for determining the operating amount of the blast furnace process, and outputs the machine-learned behavioral model. Thereby, by controlling the state of the blast furnace process according to the manipulated variable obtained by inputting the image data to be predicted into the behavioral model, it is possible to accurately control the operating state of the blast furnace process.
また、一般に、溶銑温度を制御する操作と高炉内の通気状況を安定化させるための操作とは重複しているため、干渉が発生する。そのため、溶銑温度を制御する操作量と高炉内の通気状況を安定化するための操作量とを分離せずに機械学習した場合、オペレータの操作を適切に再現できない。これに対して、本実施形態では、溶銑温度に関係する観測量の履歴データを画像化した画像データを入力データ、溶銑温度に関係する観測量の履歴データに基づいてオペレータが決定した高炉プロセスの操作量を出力データとした第１機械学習と、高炉内の通気状況に関係する観測量の履歴データを画像化した画像データを入力データ、高炉内の通気状況に関係する観測量の履歴データに基づいてオペレータが決定した高炉プロセスの操作量を出力データとした第２機械学習と、を分離して実行する。これにより、溶銑温度の制御だけでなく高炉内の通気状況を安定化するためのガイダンスをすることができる。また、溶銑温度及び高炉内の通気状況どちら起因のガイダンスであるか明確にできるため、オペレータにガイダンスの理由を明確に説明できるようになる。 In addition, since the operations for controlling the hot metal temperature and the operations for stabilizing the ventilation inside the blast furnace generally overlap, interference occurs. Therefore, if machine learning is performed without separating the manipulated variable for controlling the hot metal temperature from the manipulated variable for stabilizing the ventilation situation in the blast furnace, the operator's operations cannot be appropriately reproduced. In contrast, in this embodiment, image data obtained by converting historical data of observed quantities related to hot metal temperature into images is used as input data, and blast furnace process determined by an operator based on historical data of observed quantities related to hot metal temperature is used as input data. The first machine learning uses the manipulated variable as output data, and the image data obtained by converting the historical data of the observed quantities related to the ventilation status in the blast furnace into the input data, and the historical data of the observed quantities related to the ventilation status in the blast furnace. and second machine learning using the operation amount of the blast furnace process determined by the operator as output data. This makes it possible to provide guidance for not only controlling the hot metal temperature but also for stabilizing the ventilation situation in the blast furnace. Furthermore, it is possible to clearly explain whether the guidance is due to the temperature of the hot metal or the ventilation condition in the blast furnace, so that the reason for the guidance can be clearly explained to the operator.
なお、履歴データＤＢ２に新たな学習データが追加された場合、モデル学習部１１は、新たに追加された学習データを追加学習用データとして行動モデルを追加学習させることにより、行動モデルを更新することが望ましい。 Note that when new learning data is added to the history data DB 2, the model learning unit 11 updates the behavior model by additionally learning the behavior model using the newly added learning data as additional learning data. is desirable.
図３，４は、オペレータの実際操作及びモデル出力の時間変化の一例を示す図である。ここで、図３は、本発明例として、溶銑温度を制御する操作量と高炉内の通気状況を安定化するための操作量とを分離して機械学習させたモデルで出力した結果を示す。また、図４は、比較例として、溶銑温度を制御する操作量と高炉内の通気状況を安定化するための操作量とを分離せずに機械学習させたモデルで出力した結果を示す。また、図中、丸印はオペレータの実際操作、実線はモデル出力、破線はガイダンス出力判定線を示す。 FIGS. 3 and 4 are diagrams showing examples of actual operations by an operator and changes over time in model output. Here, FIG. 3 shows, as an example of the present invention, the results output by a model in which the manipulated variable for controlling the hot metal temperature and the manipulated variable for stabilizing the ventilation situation in the blast furnace are separated and subjected to machine learning. Moreover, FIG. 4 shows, as a comparative example, the results output by a model in which machine learning was performed without separating the manipulated variable for controlling the hot metal temperature and the manipulated variable for stabilizing the ventilation situation in the blast furnace. Further, in the figure, circles indicate actual operations by the operator, solid lines indicate model output, and broken lines indicate guidance output determination lines.
図３，４に示す例では、オペレータは、モデル出力が炉熱上げアクション、炉熱維持、及び炉熱下げアクションの３つのアクションのうち、どのアクションに属するか判定線を用いて判定を行っている。図４に示す比較例では、点線で囲まれた部分のモデル出力は炉熱維持のアクションとなっているが、実際にはオペレータが炉熱変更の実アクションをとっており、乖離している。これに対して、図３に示す発明例では、モデル出力でも炉熱変更のアクションが指示されており、実アクションと適合している。このことから、本発明によれば、高炉プロセスの操業状態を精度よく制御可能な学習モデルを生成することができ、結果として、高炉の溶銑温度及び通気状況を精度よく制御できることが確認された。 In the examples shown in Figures 3 and 4, the operator uses the judgment line to determine which action the model output belongs to among the three actions: furnace heat raising action, furnace heat maintenance, and furnace heat lowering action. There is. In the comparative example shown in FIG. 4, the model output in the area surrounded by the dotted line indicates an action to maintain the furnace heat, but in reality, the operator takes the actual action to change the furnace heat, so there is a discrepancy. In contrast, in the example of the invention shown in FIG. 3, the model output also instructs the action of changing the furnace heat, which matches the actual action. From this, it was confirmed that according to the present invention, it is possible to generate a learning model that can accurately control the operational status of the blast furnace process, and as a result, it is possible to accurately control the hot metal temperature and ventilation status of the blast furnace.
以上、本発明者らによってなされた発明を適用した実施の形態について説明したが、本実施形態による本発明の開示の一部をなす記述及び図面により本発明は限定されることはない。すなわち、本実施形態に基づいて当業者等によりなされる他の実施の形態、実施例、及び運用技術等は全て本発明の範疇に含まれる。 Although the embodiments applying the invention made by the present inventors have been described above, the present invention is not limited by the description and drawings that form part of the disclosure of the present invention by the present embodiments. That is, all other embodiments, examples, operational techniques, etc. made by those skilled in the art based on this embodiment are included in the scope of the present invention.
１ 学習モデル生成装置
２ 履歴データデータベース（履歴データＤＢ）
１１ モデル学習部
１２ モデル出力部
1 Learning model generation device 2 Historical data database (historical data DB)
11 Model learning section 12 Model output section
Claims (5)
前記ステップは、溶銑温度、羽口埋込温度、及びコークス比の履歴データを画像化した画像データを入力データ、溶銑温度、羽口埋込温度、及びコークス比の履歴データに基づいてオペレータが決定した高炉プロセスの操作量を出力データとした第１機械学習と、通気抵抗及び炉熱指数の履歴データを画像化した画像データを入力データ、通気抵抗及び炉熱指数の履歴データに基づいてオペレータが決定した高炉プロセスの操作量を出力データとした第２機械学習と、を分離して実行することにより、オペレータが溶銑温度を制御する操作量を決定する第１行動モデル及びオペレータが高炉内の通気状況を制御する操作量を決定する第２行動モデルを出力するステップを含むことを特徴とする学習モデル生成方法。 Input data is image data obtained by converting historical data of observed quantities indicating the operational status of the blast furnace process, including at least one of hot metal temperature, tuyere embedding temperature, ventilation resistance, furnace heat index, and coke ratio; Machine learning is performed on a behavioral model in which the operator determines the operation amount of the blast furnace process, using the operation amount of the blast furnace process, including at least one of the pulverized coal ratio and the air humidity determined by the operator based on historical data, as output data. including a step of outputting a behavioral model obtained by
In the step, the operator inputs image data of the historical data of the hot metal temperature , tuyere filling temperature, and coke ratio, and determines the hot metal temperature , the tuyere filling temperature, and the coke ratio based on the historical data. The first machine learning uses the operation amount of the blast furnace process as output data, the input data is the image data of the historical data of ventilation resistance and the furnace heat index, and the operator uses the historical data of the ventilation resistance and the furnace heat index as input data. By separately executing the second machine learning using the determined operation amount of the blast furnace process as output data, the first behavioral model determines the operation amount for the operator to control the hot metal temperature, and the second machine learning uses the determined operation amount for the blast furnace process as output data. A learning model generation method comprising the step of outputting a second behavior model that determines an amount of operation for controlling a situation.
前記手段は、溶銑温度、羽口埋込温度、及びコークス比の履歴データを画像化した画像データを入力データ、溶銑温度、羽口埋込温度、及びコークス比の履歴データに基づいてオペレータが決定した高炉プロセスの操作量を出力データとした第１機械学習と、通気抵抗及び炉熱指数の履歴データを画像化した画像データを入力データ、通気抵抗及び炉熱指数の履歴データに基づいてオペレータが決定した高炉プロセスの操作量を出力データとした第２機械学習と、を分離して実行することにより、オペレータが溶銑温度を制御する操作量を決定する第１行動モデル及びオペレータが高炉内の通気状況を制御する操作量を決定する第２行動モデルを出力することを特徴とする学習モデル生成装置。 Input data is image data obtained by converting historical data of observed quantities indicating the operational status of the blast furnace process, including at least one of hot metal temperature, tuyere embedding temperature, ventilation resistance, furnace heat index, and coke ratio; Machine learning is performed on a behavioral model in which the operator determines the operation amount of the blast furnace process, using the operation amount of the blast furnace process, including at least one of the pulverized coal ratio and the air humidity determined by the operator based on historical data, as output data. and a means for outputting the behavioral model that has been created.
The means inputs image data obtained by converting historical data of hot metal temperature , tuyere filling temperature, and coke ratio, and determines the determination by an operator based on the historical data of hot metal temperature , tuyere filling temperature, and coke ratio. The first machine learning uses the operation amount of the blast furnace process as output data, the input data is the image data of the historical data of ventilation resistance and the furnace heat index, and the operator uses the historical data of the ventilation resistance and the furnace heat index as input data. By separately executing the second machine learning using the determined operation amount of the blast furnace process as output data, the first behavioral model determines the operation amount for the operator to control the hot metal temperature, and the second machine learning uses the determined operation amount for the blast furnace process as output data. A learning model generation device characterized by outputting a second behavior model that determines an amount of operation for controlling a situation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2021002835A JP7380604B2 (en) | 2021-01-12 | 2021-01-12 | Learning model generation method, learning model generation device, blast furnace control guidance method, and hot metal manufacturing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2021002835A JP7380604B2 (en) | 2021-01-12 | 2021-01-12 | Learning model generation method, learning model generation device, blast furnace control guidance method, and hot metal manufacturing method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| JP2022108041A JP2022108041A (en) | 2022-07-25 |
| JP7380604B2 true JP7380604B2 (en) | 2023-11-15 |
Family
ID=82556220
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| JP2021002835A Active JP7380604B2 (en) | 2021-01-12 | 2021-01-12 | Learning model generation method, learning model generation device, blast furnace control guidance method, and hot metal manufacturing method |
Country Status (1)
| Country | Link |
|---|---|
| JP (1) | JP7380604B2 (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2003328017A (en) | 2002-05-10 | 2003-11-19 | Sumitomo Metal Ind Ltd | Method for educating blast furnace operation |
| JP2008097458A (en) | 2006-10-13 | 2008-04-24 | Nippon Steel Corp | Plant operation support device and supporting method, computer program, and storage medium |
| JP2008144265A (en) | 2006-11-13 | 2008-06-26 | Kobe Steel Ltd | Method for predicting molten iron temperature in blast furnace |
| JP2019019385A (en) | 2017-07-19 | 2019-02-07 | Ｊｆｅスチール株式会社 | Method and device for predicting molten iron temperature, operation method of blast furnace, operation guidance device, and method and device for controlling molten iron temperature |
| JP2021503042A (en) | 2017-11-14 | 2021-02-04 | ポスコＰｏｓｃｏ | Blast furnace blower control device and its method |
| JP2021507115A (en) | 2017-12-19 | 2021-02-22 | ポスコＰｏｓｃｏ | Reactor condition management equipment and method |
| JP2021509440A (en) | 2017-12-26 | 2021-03-25 | ポスコＰｏｓｃｏ | Blast furnace operation status evaluation system and method |
| JP6897723B2 (en) | 2019-07-19 | 2021-07-07 | Ｊｆｅスチール株式会社 | Learning model generation method, learning model generation device, blast furnace hot metal temperature control method, blast furnace hot metal temperature control guidance method, and hot metal manufacturing method |
-
2021
- 2021-01-12 JP JP2021002835A patent/JP7380604B2/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2003328017A (en) | 2002-05-10 | 2003-11-19 | Sumitomo Metal Ind Ltd | Method for educating blast furnace operation |
| JP2008097458A (en) | 2006-10-13 | 2008-04-24 | Nippon Steel Corp | Plant operation support device and supporting method, computer program, and storage medium |
| JP2008144265A (en) | 2006-11-13 | 2008-06-26 | Kobe Steel Ltd | Method for predicting molten iron temperature in blast furnace |
| JP2019019385A (en) | 2017-07-19 | 2019-02-07 | Ｊｆｅスチール株式会社 | Method and device for predicting molten iron temperature, operation method of blast furnace, operation guidance device, and method and device for controlling molten iron temperature |
| JP2021503042A (en) | 2017-11-14 | 2021-02-04 | ポスコＰｏｓｃｏ | Blast furnace blower control device and its method |
| JP2021507115A (en) | 2017-12-19 | 2021-02-22 | ポスコＰｏｓｃｏ | Reactor condition management equipment and method |
| JP2021509440A (en) | 2017-12-26 | 2021-03-25 | ポスコＰｏｓｃｏ | Blast furnace operation status evaluation system and method |
| JP6897723B2 (en) | 2019-07-19 | 2021-07-07 | Ｊｆｅスチール株式会社 | Learning model generation method, learning model generation device, blast furnace hot metal temperature control method, blast furnace hot metal temperature control guidance method, and hot metal manufacturing method |
Also Published As
| Publication number | Publication date |
|---|---|
| JP2022108041A (en) | 2022-07-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6729514B2 (en) | Hot metal temperature prediction method, hot metal temperature prediction device, blast furnace operating method, operation guidance device, hot metal temperature control method, and hot metal temperature control device | |
| WO2021014782A1 (en) | Learning model generation method, learning model generation device, method for controlling molten iron temperature in blast furnace, method for guiding molten iron temperature control in blast furnace, and molten iron manufacturing method | |
| JP6930507B2 (en) | Hot metal temperature prediction method, hot metal temperature prediction device, blast furnace operation method, operation guidance device, hot metal temperature control method, and hot metal temperature control device | |
| JP2018024935A (en) | Molten iron temperature prediction method, molten iron temperature prediction device, operation method of blast furnace, operation guidance device, molten iron temperature control method and molten iron temperature control device | |
| JP7380604B2 (en) | Learning model generation method, learning model generation device, blast furnace control guidance method, and hot metal manufacturing method | |
| WO2022009617A1 (en) | Method for controlling hot metal temperature, operation guidance method, method for operating blast furnace, method for producing hot metal, device for controlling hot metal temperature, and operation guidance device | |
| EP3989013B1 (en) | Method for controlling process, operation guidance method, method for operating blast furnace, method for producing hot metal, and device for controlling process | |
| JP2021082139A (en) | Device, method and program for additional learning | |
| US11556111B2 (en) | Human-plausible automated control of an industrial process | |
| US20100082121A1 (en) | Validation of laboratory test data | |
| RU2251721C2 (en) | Intellectual control system | |
| JP7537636B2 (en) | Sintering process control method, operation guidance method, sinter ore manufacturing method, sintering process control device, operation guidance device, sintering operation guidance system, and terminal device | |
| JPH0598325A (en) | Device for controlling distribution of charging materials in blast furnace | |
| JP7384326B1 (en) | Blast furnace hot metal temperature prediction method, blast furnace hot metal temperature prediction model learning method, blast furnace operating method, blast furnace hot metal temperature prediction device, blast furnace hot metal temperature prediction system and terminal device | |
| JP6933196B2 (en) | Blast furnace unloading speed prediction model learning method, blast furnace unloading speed prediction method, blast furnace operation guidance method, blast furnace unloading speed control method, hot metal manufacturing method, blast furnace operation method, and blast furnace unloading speed prediction Model learning device | |
| JP2000129319A (en) | Method for controlling furnace heat in blast furnace and device therefor | |
| CN117556711B (en) | Blast furnace coal injection optimization method, system, terminal and medium based on neural network | |
| Kronberger et al. | Latest generation sinter process optimization systems | |
| JP2737136B2 (en) | Blast furnace heat control system | |
| Fan et al. | Real-time operation guide system for sintering process with artificial intelligence | |
| JPH01136912A (en) | Automatic control system of furnace heat in blast furnace | |
| JPH05156327A (en) | Device for controlling furnace heat in blast furnace | |
| JP4110780B2 (en) | Estimation method of hot metal temperature and disturbance in blast furnace | |
| CN118159797A (en) | Method and system for observing cement kiln process | |
| JPH01205008A (en) | Apparatus for controlling furnace heat in blast furnace |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| A621 | Written request for application examination |
Free format text: JAPANESE INTERMEDIATE CODE: A621 Effective date: 20220822 |
|
| A977 | Report on retrieval |
Free format text: JAPANESE INTERMEDIATE CODE: A971007 Effective date: 20230704 |
|
| A131 | Notification of reasons for refusal |
Free format text: JAPANESE INTERMEDIATE CODE: A131 Effective date: 20230905 |
|
| A521 | Request for written amendment filed |
Free format text: JAPANESE INTERMEDIATE CODE: A523 Effective date: 20230921 |
|
| TRDD | Decision of grant or rejection written | ||
| A01 | Written decision to grant a patent or to grant a registration (utility model) |
Free format text: JAPANESE INTERMEDIATE CODE: A01 Effective date: 20231003 |
|
| A61 | First payment of annual fees (during grant procedure) |
Free format text: JAPANESE INTERMEDIATE CODE: A61 Effective date: 20231016 |
|
| R150 | Certificate of patent or registration of utility model |
Ref document number: 7380604 Country of ref document: JP Free format text: JAPANESE INTERMEDIATE CODE: R150 |