Method and system for predicting material speed of blast furnace based on Stacking model

Abstract

The invention provides a method and a system for predicting the material speed of a blast furnace based on a Stacking model. The method comprises the following steps: acquiring process data of a blast furnace to be measured; based on the process data, preprocessing the process data to obtain target process data; based on the target process data, using a feature importance score model of a random forest to obtain target features; dividing the target features and the target actual material speed into a training set and a test set according to a preset proportion; using the training set to train a Stacking model, and obtaining a target Stacking model; and predicting the test set by using the target Stacking model, and generating a material speed predicted value, so as to solve the problems that the current blast furnace speed can only be manually calculated, the error of the calculation result of the furnace speed is large and unstable due to different experience levels of operators, and reliable guidance is difficult to provide for blast furnace operation.

Description

Technical Field

[0001] The present application relates to the technical field of blast furnace charge rate prediction, and in particular to a method and system for predicting blast furnace charge rate based on a Stacking model. Background Art

[0002] The blast furnace is a continuous reactor that reduces iron ore into pig iron. The main purpose of blast furnace operation is to ensure the smooth flow of rising coal gas and descending furnace charge, and to achieve the best technical and economic benefits as much as possible. In the daily blast furnace smelting process, the furnace charge is kept stable and balanced.

[0003] In daily production, the downward movement of the charge can be reflected by the charge rate, which generally refers to the batch of charge entering the furnace within a unit time (usually 1 hour). Accurately grasping the charge rate of the blast furnace has important guiding significance for the production operation of the blast furnace. The operator can make real-time adjustments to the blast furnace according to the size of the charge rate, such as adjusting the air supply temperature, adjusting the coal injection, adjusting the oxygen enrichment, etc., to maintain a stable and balanced downward movement of the charge and thus ensure the stable and smooth production of the blast furnace. At present, the blast furnace charge rate is mainly obtained by the operator by making simple calculations on some of the current blast furnace process parameters (such as oxygen enrichment, air volume entering the furnace, fuel ratio, wind consumption per ton of fuel, ore consumption, ore batch, material line, etc.) and combining them with experience.

[0004] The above adjustment method has many shortcomings: first, the data is not considered comprehensively and systematically enough, resulting in inaccurate calculated material speed; second, manual calculation increases the operator's workload, and the calculation frequency is limited by the operator's current work intensity, making it difficult to obtain calculation results and material speed change trends in a timely manner; finally, different operators have different experiences, ability levels, calculation habits, etc., resulting in large and unstable calculation errors, making it difficult to provide reliable guidance for blast furnace operations. Summary of the invention

[0005] The present application provides a method and system for predicting blast furnace charge rate based on the Stacking model, so as to solve the technical problem that the blast furnace charge rate can only be calculated manually at present, and the calculation results of the charge rate have large errors and instability due to the different experience levels of operators, making it difficult to provide reliable guidance for blast furnace operation.

[0006] The first aspect of the present application provides a method for predicting blast furnace charge rate based on a Stacking model, comprising:

[0007] Obtain the process data of the blast furnace to be tested;

[0008] Based on the process data, preprocessing operations are performed on the process data to obtain target process data; the target process data includes: process data, target furnace top temperature, target material line, and target actual material speed;

[0009] Based on the target process data, the target features are obtained by using the feature importance score model of random forest; the feature importance score model of random forest is:

[0010]

[0011] Where s is the reference score of the random forest model for the target process data; i j is the importance score of the jth feature in the target process data; K is the total number of times the order of a feature in the target process data is disrupted; k is any positive integer, k≤K;

[0012] According to a preset ratio, the target features and the target actual material speed are divided into a training set and a test set;

[0013] Using the training set to train the Stacking model to obtain a target Stacking model;

[0014] The target Stacking model is used to predict the test set and generate a material speed prediction value.

[0015] In some embodiments, obtaining process data of the blast furnace to be tested includes:

[0016] Obtaining the temperature, pressure and material feeding data of the blast furnace to be tested; the material feeding data includes: material feeding speed and whether stagnation occurs during material feeding;

[0017] If the temperature in the furnace is within the preset temperature range, the pressure in the furnace is within the preset pressure range, the material unloading speed is within the preset speed range and there is no stagnation during unloading, the process data of the blast furnace to be tested is obtained.

[0018] In some embodiments, the process data include: actual material speed, furnace air volume, oxygen enrichment, hot air pressure, hot air temperature, gas utilization rate, humidity, coal injection amount, coal ratio, fuel ratio, dry fuel ratio, wet fuel ratio, heat load, water temperature difference, furnace top pressure, wind pressure ratio, upper differential pressure, lower differential pressure, furnace top temperature, wind consumption per ton of fuel, oxygen enrichment rate, center temperature, furnace top temperature, actual wind speed, standard wind speed, edge temperature, furnace temperature, slag ratio, ore batch, material line;

[0019] The furnace top temperature includes: a first furnace top temperature, a second furnace top temperature, a third furnace top temperature, and a fourth furnace top temperature; the first furnace top temperature, the second furnace top temperature, the third furnace top temperature, and the fourth furnace top temperature are respectively collected at four positions of the blast furnace top;

[0020] The material line includes: a first orientation material line and a second orientation material line when the blast furnace is in the first process and the second process; the first orientation material line and the second orientation material line are material lines in two opposite directions of the blast furnace.

[0021] In some embodiments, performing a preprocessing operation on the process data based on the process data to obtain target process data includes:

[0022] Based on the process data, the target furnace top temperature is calculated; the average value of the target furnace top temperature is:

[0023]

[0024] Where, T K1 is the first furnace top temperature; T K2 is the second furnace top temperature; T K3 is the third furnace top temperature; T K4 is the fourth furnace top temperature;

[0025] Based on the process data, the target material line is calculated; the target material line is:

[0026]

[0027] Where, FL cl It is the first position material line when the blast furnace is in the first process; FL cr It is the second position material line when the blast furnace is in the first process; FL ol It is the first position material line when the blast furnace is in the second process; FL or It is the second position material line when the blast furnace is in the second process;

[0028] Based on the process data, the target furnace top temperature and the target material line, the target process data are obtained.

[0029] In some embodiments, performing a preprocessing operation on the process data based on the process data to obtain target process data further includes:

[0030] Based on the process data, an actual material speed reference value is obtained; the actual material speed reference value includes: a first actual material speed and a second actual material speed; the first actual material speed is less than the second actual material speed;

[0031] Based on the actual material speed reference value, the actual material speed upper limit is calculated; the actual material speed upper limit is:

[0032] UL=Q2+1.5×(Q2-Q1);

[0033] In the formula, Q2 is the second actual material speed; Q1 is the first actual material speed;

[0034] Based on the actual material speed reference value, the actual material speed lower limit is calculated; the actual material speed lower limit is:

[0035] LL = Q1 - 1.5 × (Q2 - Q1);

[0036] Based on the actual material speed upper limit and the actual material speed lower limit, an abnormal value is obtained; the abnormal value is a value of the actual material speed that is greater than the actual material speed upper limit and less than the actual material speed lower limit;

[0037] Deleting the abnormal value in the actual material speed to obtain the target actual material speed;

[0038] Based on the process data and the target actual material speed, target process data are obtained.

[0039] In some embodiments, obtaining the target feature based on the target process data using a feature importance score model of a random forest includes:

[0040] Based on the target process data, process calculation characteristics are obtained; the process calculation characteristics include: oxygen enrichment, furnace air volume, fuel ratio, air consumption per ton of fuel, ore batch, and material line;

[0041] Based on the target process characteristics, using a feature importance score model of a random forest, obtaining the importance score of each feature in the target process data;

[0042] Based on the importance score, obtaining an importance feature; the importance feature is a feature with a larger importance score in a preset number of the target process data;

[0043] Based on the process calculation features and importance features, target features are obtained.

[0044] In some embodiments, the Stacking model includes: a base learner, a meta learner;

[0045] The training of the Stacking model using the training set to obtain a target Stacking model includes:

[0046] Using the training set to train the base learner to obtain a data set;

[0047] The meta-learner is trained using the data set to obtain a target Stacking model.

[0048] In some embodiments, the base learner includes: a first base learner, a second base learner, and a third base learner;

[0049] The base learner is trained by using the training set to obtain a data set including:

[0050] Using the training set, training the first base learner to obtain a first prediction result;

[0051] Using the training set, training the second base learner to obtain a second prediction result;

[0052] Using the training set, training the third base learner to obtain a third prediction result;

[0053] A data set is obtained based on the first prediction result, the second prediction result, and the third prediction result.

[0054] In some embodiments, the using the training set to train the base learner to obtain the data set further comprises:

[0055] Splitting the training set according to a preset ratio to obtain a training subset and a validation set;

[0056] Performing a preset number of iterative training on the base learner using the training subset and the validation set, and obtaining an output result of the validation set in each iterative training;

[0057] Based on the output results, a data set is obtained.

[0058] The second aspect of the present application provides a system for predicting blast furnace material rate based on a Stacking model, comprising:

[0059] An acquisition module is used to acquire process data of the blast furnace to be tested;

[0060] The first obtaining module is configured to perform a preprocessing operation on the process data based on the process data to obtain target process data; the target process data includes: process data, target furnace top temperature, target material line, and target actual material speed;

[0061] The second obtaining module is configured to obtain the target feature based on the target process data using a feature importance score model of a random forest; the feature importance score model of the random forest is:

[0062]

[0063] Where s is the reference score of the random forest model for the target process data; i j is the importance score of the jth feature in the target process data; K is the total number of times the order of a feature in the target process data is disrupted; k is any positive integer, k≤K;

[0064] A splitting module, used to divide the target features and the target actual material speed into a training set and a test set according to a preset ratio;

[0065] A third obtaining module is configured to train the Stacking model using the training set to obtain a target Stacking model;

[0066] A generation module is configured to use the target Stacking model to predict the test set and generate a material speed prediction value.

[0067] The present application provides a method and system for predicting blast furnace material rate based on a Stacking model, the method comprising: obtaining process data of a blast furnace to be tested; based on the process data, performing a preprocessing operation on the process data to obtain target process data; the target process data comprises: process data, target furnace top temperature, target material line, and target actual material rate; based on the target process data, obtaining target features using a feature importance score model of a random forest; the feature importance score model of the random forest is:

[0068]

[0069] Where s is the reference score of the random forest model for the target process data; i j is the importance score of the jth feature in the target process data; K is the total number of times the order of a feature in the target process data is disrupted; k is an arbitrary positive integer, k≤K; according to a preset ratio, the target feature and the target actual material rate are divided into a training set and a test set; the Stacking model is trained using the training set to obtain a target Stacking model; the test set is predicted using the target Stacking model to generate a material rate prediction value, so as to obtain an accurate material rate prediction value, provide effective support for the operator's next blast furnace operation, and thus help ensure the stable operation of the blast furnace. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0071] Figure 1 This is a flow chart of a method for predicting blast furnace material rate based on the Stacking model in this application;

[0072] Figure 2 Box plots used to identify outliers in this application;

[0073] Figure 3 This is a schematic diagram of the Stacking ensemble learning model in this application;

[0074] Figure 4 It is a trend chart of the predicted value and actual value of the blast furnace material rate in this application. DETAILED DESCRIPTION

[0075] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0076] In some technologies, the blast furnace speed can only be calculated manually. Due to the different experience levels of operators, the calculation results of the furnace speed have large errors and are unstable, making it difficult to provide reliable guidance for blast furnace operation. In order to solve this technical problem, the present application provides a refrigerator door self-locking limit device. The structures of the various parts of the refrigerator door self-locking limit device are described below:

[0077] For example, a blast furnace is a continuous reactor that reduces iron ore to pig iron. Iron ore, coke, solvent and other furnace materials are fed into the blast furnace in batches from the top of the furnace by the furnace top charging device, forming an alternating layered structure inside the blast furnace. As the smelting proceeds, the furnace material continues to descend, contacting and reacting with the rising high-temperature coal gas flow, the ore and solvent are heated and reduced, gradually softening and melting in the soft melting zone, and finally forming molten iron and slag that are regularly discharged from the furnace. It can be seen that the blast furnace smelting process is a complex reaction process in which the descending furnace material and the rising coal gas flow move in opposite directions throughout.

[0078] Therefore, the main purpose of blast furnace operation is to ensure the smooth flow of rising coal gas and descending charge, and to achieve better technical and economic benefits as much as possible. In the daily blast furnace smelting process, it is extremely important to keep the charge flowing smoothly and evenly: (1) It can maintain the stability of the internal environment of the blast furnace and reduce the fluctuation of furnace conditions caused by uneven descent of charge, thus providing a basis for the continuous and stable production of the blast furnace, and at the same time, it helps to stabilize the control of parameters such as temperature and airflow in the furnace, thereby improving smelting efficiency; (2) Stable furnace temperature and airflow can reduce the impurity content in pig iron and improve the purity and quality of pig iron; (3) It reduces local wear caused by uneven descent of charge and prolongs the life of the blast furnace; (4) It reduces the energy waste caused by fluctuation of furnace conditions, reduces consumption and costs.

[0079] In daily production, the downward movement of the charge is reflected by the charge rate. The charge rate generally refers to the batch of charge entering the furnace within a unit time (usually 1 hour). Accurately mastering the charge rate of the blast furnace has important guiding significance for the production and operation of the blast furnace. The operator can make real-time adjustments to the blast furnace according to the size of the charge rate, such as adjusting the air supply temperature, adjusting the coal injection, adjusting the oxygen enrichment, etc., to maintain a stable and balanced downward movement of the charge and thus ensure the stable and smooth operation of the blast furnace production.

[0080] At present, steel mills mainly rely on operators to perform simple calculations on some of the current blast furnace process parameters (such as oxygen enrichment, air volume into the furnace, fuel ratio, ton fuel wind consumption, ore consumption, ore batch, material line, etc.) and combine them with experience to obtain the blast furnace material rate. However, the above simple calculation has shortcomings: (1) The data considered is not comprehensive and systematic enough, and it is difficult to reflect that the material rate is the result of the dynamic smelting of the blast furnace and the combined effect of multiple factors; (2) Manual calculation increases the workload of the operator, and the calculation frequency is limited by the current work intensity of the operator, making it difficult to obtain the calculation results and the material rate change trend in time; (3) The daily work of the blast furnace is carried out in three shifts, and the experience, ability level, calculation habits, etc. of each shift operator are different, resulting in large and unstable calculation errors, making it difficult to provide reliable guidance for blast furnace operation.

[0081] Depend on Figure 1 It can be seen that in view of the above problems, the first aspect of the present application provides a method for predicting blast furnace material rate based on the Stacking model, comprising the following steps:

[0082] S100: Obtain process data of the blast furnace to be tested; the process data include: actual material speed, air volume entering the furnace, oxygen enrichment, hot air pressure, hot air temperature, gas utilization rate, humidity, coal injection amount, coal ratio, fuel ratio, dry fuel ratio, wet fuel ratio, heat load, water temperature difference, furnace top pressure, wind pressure ratio, upper differential pressure, lower differential pressure, furnace top temperature, wind consumption per ton of fuel, oxygen enrichment rate, center temperature, furnace top temperature, actual wind speed, standard wind speed, edge temperature, furnace temperature, slag ratio, ore batch, material line; the furnace top temperature includes: first furnace top temperature, second furnace top temperature, third furnace top temperature The furnace top temperature, the fourth furnace top temperature; the first furnace top temperature, the second furnace top temperature, the third furnace top temperature, and the fourth furnace top temperature are collected from four positions of the blast furnace top respectively; the material line includes: the first position material line and the second position material line when the blast furnace is in the first process and the second process; the first position material line and the second position material line are material lines in two opposite positions of the blast furnace; for example: the first process is in the ore material line, and the second process is in the coke material line; the first position material line and the second position material line are material lines in two opposite positions of the blast furnace. Among them, the actual material speed is used as a label, and other parameters are used as features.

[0083] It is worth mentioning that there are several material rulers distributed in different positions in the blast furnace. It can be understood that the blast furnace is large in size, so the material surface in the blast furnace is not a flat state, but an uneven curved surface state. The material ruler is used to obtain the distance from one end of the material ruler close to the material surface to the material surface. When the distance is less than the preset value, the feeding is stopped. By obtaining two material lines in opposite directions, the material line parameters of the blast furnace to be tested can be more accurately represented.

[0084] The step of obtaining the process data of the blast furnace to be tested comprises the following sub-steps:

[0085] S110: Obtain the temperature, pressure and material feeding data of the blast furnace to be tested; the material feeding data includes: material feeding speed and whether stagnation occurs during material feeding; S120: If the temperature in the furnace is within the preset temperature range, the pressure in the furnace is within the preset pressure range, the material feeding speed is within the preset speed range and there is no stagnation during material feeding, then obtain the process data of the blast furnace to be tested. It can be understood that the process data are collected and obtained under the stable and smooth conditions of the blast furnace. The stable and smooth conditions of the blast furnace include: (1) the furnace cylinder works evenly and actively; (2) the material feeding is smooth, without stagnation, fast and slow phenomena, etc.; (3) the detected temperature and pressure fluctuations of various parts of the furnace body are small, and are basically stable within the normal range, that is, the temperature and pressure are within the preset temperature range and the preset pressure range, and the preset temperature range and the preset pressure range are set by the operator according to experience and the parameters of the normal operation of the blast furnace; (4) the slag and iron temperatures are sufficient, the fluidity is good, and the sulfur content of the molten iron is low; (5) the gas flow distribution is reasonable and stable, and the fuel consumption is low.

[0086] It is understandable that if the blast furnace is in an unstable, forward condition, it means that there are problems in the operation of the blast furnace, such as stagnation of material loading, etc. Then it is meaningless to calculate the material rate at this time, and the problems with the blast furnace need to be repaired. After the blast furnace is restored to a stable, forward condition, the process data of the blast furnace to be tested can be obtained again, and the blast furnace material rate can be obtained through the process data, which can provide effective support for the operator's next blast furnace operation, thereby ensuring the stable and forward operation of the blast furnace.

[0087] S200: Based on the process data, preprocessing the process data to obtain target process data; the target process data includes: process data, target furnace top temperature, target material line, target actual material rate; the preprocessing of the process data includes processing of the furnace top temperature, material line, and actual material rate.

[0088] The preprocessing operation on the process data based on the process data to obtain target process data includes the following sub-steps:

[0089] S210: Calculate the target furnace top temperature based on the process data; and obtain the furnace top temperature average value (T K ), the average value of the target furnace top temperature is:

[0090]

[0091] Where, T K1 is the first furnace top temperature; T K2 is the second furnace top temperature; T K3 is the third furnace top temperature; T K4 is the fourth furnace top temperature.

[0092] S220: Calculate the target material line based on the process data; for the material line, select the highest and lowest two material lines and take the average value, and set the target material line as FL, and the first position material line when the blast furnace is in the first process as FL cl ; The second position material line when the blast furnace is in the first process is FL cr ; The first position material line when the blast furnace is in the second process is FL ol ; The second position material line when the blast furnace is in the second process is FL or , the target material line is:

[0093]

[0094] Where, FL cl It is the first position material line when the blast furnace is in the first process; FL cr It is the second position material line when the blast furnace is in the first process; FL ol It is the first position material line when the blast furnace is in the second process; FL or It is the second azimuth material line when the blast furnace is in the second process. Max() takes the maximum value of the first azimuth material line and the second azimuth material line when the blast furnace is in the first process and the second process, and Min() takes the minimum value of the first azimuth material line and the second azimuth material line when the blast furnace is in the first process and the second process.

[0095] S230: Based on the process data, target furnace top temperature, and target material line, target process data is obtained. The above processing method is used to make the values ​​of the target furnace top temperature and target material line more accurately represent the furnace top temperature and material line of the blast furnace to be tested.

[0096] The step of performing a preprocessing operation on the process data based on the process data to obtain target process data further comprises the following sub-steps:

[0097] S240: Based on the process data, obtain an actual material speed reference value; the actual material speed reference value includes: a first actual material speed and a second actual material speed; the first actual material speed is less than the second actual material speed; the first actual material speed and the second actual material speed are obtained by sorting all the actual material speeds from small to large, and taking the actual material speed at the 25% position as the first actual material speed, and taking the actual material speed at the 75% position as the second actual material speed.

[0098] S250: Based on the actual material speed reference value, calculate the actual material speed upper limit; the upper limit is the maximum value within the non-abnormal range, and the actual material speed upper limit is:

[0099] UL=Q2+1.5×(Q2-Q1);

[0100] In the formula, Q2 is the second actual material speed; Q1 is the first actual material speed;

[0101] S260: Based on the actual material speed reference value, calculate the actual material speed lower limit; the lower limit is the minimum value within the non-abnormal range, and the actual material speed lower limit is:

[0102] LL = Q1 - 1.5 × (Q2 - Q1);

[0103] S270: Based on the actual material speed upper limit and the actual material speed lower limit, an abnormal value is obtained; the abnormal value is a value of the actual material speed that is greater than the actual material speed upper limit and less than the actual material speed lower limit; the abnormal value can be given by a box plot, and the box plot, such as Figure 2 S280: Delete the abnormal value in the actual material speed to obtain the target actual material speed; S290: Based on the process data and the target actual material speed, obtain the target process data. The above processing method enables the value of the target actual material speed to more accurately represent the actual material speed of the blast furnace to be tested.

[0104] S300: Based on the target process data, a feature importance score model of random forest is used to obtain a target feature; through the feature importance score model of random forest, a feature in the target process data that has a greater impact on the material speed, that is, the target feature, is obtained.

[0105] The feature importance score model of the random forest is:

[0106]

[0107] Where s is the reference score of the random forest model for the target process data; i j is the importance score of the jth feature in the target process data; K is the total number of times the order of a feature in the target process data is disrupted; k is an arbitrary positive integer, k≤K.

[0108] The step of obtaining the target feature based on the target process data by using the feature importance score model of random forest includes the following sub-steps:

[0109] S310: Based on the target process data, obtain process calculation features; the process calculation features include: oxygen enrichment, furnace air volume, fuel ratio (before tuyere), ton fuel wind consumption, ore batch, material line; the process calculation features refer to the blast furnace process parameters involved in the simple calculation formula of the blast furnace operator of a large steel plant: oxygen enrichment, furnace air volume, fuel ratio (before tuyere), ton fuel wind consumption, ore batch, material line. It can be understood that most of the characteristics in the target process data will be calculated by the blast furnace process parameters involved in the simple calculation formula, so the blast furnace process parameters involved in the simple calculation formula are more representative for each characteristic.

[0110] S320: Based on the target process feature, the importance score of each feature in the target process data is obtained using a feature importance score model of a random forest; the importance score is:

[0111]

[0112] It can be understood that the importance score of each feature is the influence of the calculated feature on the material rate. If the importance score of the feature is larger, it means that the influence of the feature on the material rate is greater.

[0113] S330: Based on the importance score, obtain the importance feature; the importance feature is the feature with a larger importance score in a preset number of the target process data; according to the importance score, arrange the features from large to small, for example, if the preset number is 13, then select the top 13 features as the importance features, for example, including: inlet air volume, oxygen enrichment, humidity, coal injection amount, coal ratio, dry fuel ratio, wet fuel ratio, fuel consumption per ton of air, ore batch, standard wind speed, actual wind speed, water temperature difference, and average furnace top temperature.

[0114] S340: Based on the process calculation features and importance features, target features are obtained. All features in the process calculation features and importance features are taken as target features by taking the union method. For example, they include: furnace air volume, oxygen enrichment, humidity, coal injection amount, coal ratio, fuel ratio (before the tuyere), dry fuel ratio, wet fuel ratio, ton wind fuel consumption, ore batch, standard wind speed, actual wind speed, water temperature difference, furnace top temperature average, material line, a total of 15 features.

[0115] S400: Divide the target feature and the target actual material speed into a training set and a test set according to a preset ratio; the target feature is used as a feature of the target process feature, and the target actual material speed is used as a label of the target process feature. In the embodiment of the present application, the target feature and the target actual material speed are combined into sample data. Divide the sample data into a training set and a test set according to a ratio of 7:3, that is, 70% of the samples are used as a training set and 30% of the samples are used as a test set.

[0116] S500: Using the training set to train the Stacking model to obtain a target Stacking model; Figure 3 As shown, the Stacking model is set with two layers, including: 3 base learners in the first layer: the base learners represent linear regression algorithm, LGBM regression algorithm, XGBoost regression algorithm, and the meta learner in the second layer: the meta learner represents linear regression algorithm. The Stacking model is an integrated machine learning algorithm that learns how to best combine the prediction results from two or more base machine learning algorithms by using a meta learner.

[0117] The Stacking model includes: a base learner and a meta learner; and the training of the Stacking model using the training set to obtain a target Stacking model includes the following sub-steps:

[0118] S510: Using the training set to train the base learner to obtain a data set; first, multiple different models, namely base learners, are trained using the training set; each base learner is trained independently and may use different algorithms or parameters.

[0119] The base learners include: a first base learner, a second base learner, and a third base learner; the base learners are trained using the training set to obtain a data set, including the following sub-steps: S511: the first base learner is trained using the training set to obtain a first prediction result; S512: the second base learner is trained using the training set to obtain a second prediction result; S513: the third base learner is trained using the training set to obtain a third prediction result; S514: a data set is obtained based on the first prediction result, the second prediction result, and the third prediction result. The training set is predicted using the above base learners to obtain the prediction results of each base learner, namely the first prediction result, the second prediction result, and the third prediction result; the prediction results of each base learner are combined in columns to form a new data set. Among them, the first base learner is a linear regression algorithm, the second base learner is an LGBM regression algorithm, and the third base learner is an XGBoost regression algorithm.

[0120] The method of using the training set to train the base learner to obtain a data set also includes the following sub-steps: using the training set as the original input of the base learner, obtaining the output through an 8-fold cross-validation method, and using the output result, i.e., the data set, as the input of the meta-learner to perform final model training, and finally obtaining a Stacking ensemble learning model for predicting blast furnace charge rate.

[0121] S515: Split the training set according to a preset ratio to obtain a training subset and a validation set; in the embodiment of the present application, the 8-fold cross-validation method refers to randomly shuffling the training set and dividing it into 8 equal parts, that is, the preset ratio is to split the training set into 8 equal parts, 7 of which are used as training sets for the base learner, and 1 is used as a validation set for the base learner.

[0122] S516: Perform a preset number of iterative training on the base learner using the training subset and the validation set, and obtain the output result of the validation set in each iterative training; in the embodiment of the present application, perform 8 rounds of iterative training on the base learner, and output and combine the prediction results of the corresponding validation set in each round of iterative training as a new data set. S517: Obtain a data set based on the output results.

[0123] S520: Based on the prediction results, the meta-learner is trained using a linear regression algorithm to obtain a target Stacking model. The new data set is input as a feature into a secondary learner, namely, a meta-learner, and the meta-learner is trained to obtain a trained blast furnace charge rate Stacking ensemble learning model, namely, a target Stacking model.

[0124] S600: Use the target Stacking model to predict the test set and generate a predicted material speed value. Use the trained blast furnace material speed Stacking ensemble learning model, i.e., the target Stacking model, to predict the test set and draw a trend chart of the predicted material speed value and the actual material speed value for the operator's reference. Figure 4As shown, if the predicted value of the material speed is in a state of continuous decline or increase, it indicates that the furnace condition is in an unstable state and the blast furnace is not running smoothly. For example, if the predicted value of the material speed is in a state of continuous decline, it means that the blast furnace will have a tendency of slow material speed, which may be caused by various factors such as poor permeability in the furnace, temperature fluctuations in the furnace, or changes in the properties of raw materials. In order to increase the material speed, the operator can take a series of positive adjustment measures. First, adjusting the air supply temperature is a direct and effective method. By increasing the air supply temperature, the heat in the furnace can be enhanced, and the softening and melting of the charge can be promoted, thereby increasing the material speed. Secondly, adjusting the amount of coal injection is also one of the important means. An appropriate increase in the amount of coal injection can increase the reducing atmosphere in the furnace, promote the reduction reaction of iron ore, and indirectly increase the material speed. In addition, adjusting the oxygen enrichment rate is also critical. Increasing the oxygen enrichment rate can improve combustion efficiency, increase the oxygen content in the hot air, and help speed up the combustion and descent rate of the charge. For example, if the predicted value of the material speed is in the above-mentioned state, it means that the blast furnace will have a tendency of too fast material speed, which may lead to uneven distribution of airflow in the furnace, large fluctuations in furnace temperature, and even affect the stable operation of the blast furnace. In order to reduce the material speed, the operator also needs to take corresponding adjustment measures. At this time, the air supply temperature can be appropriately lowered to slow down the heat supply in the furnace and slow down the rate of descent of the charge. At the same time, reduce the amount of coal injection or adjust the coal injection strategy to avoid excessive reducing atmosphere in the furnace, resulting in too fast a material speed. In addition, reducing the oxygen enrichment rate is also an effective means. By reducing the oxygen content in the hot air, the combustion rate of the charge can be slowed down, thereby controlling the material speed within a constant range. The adjustment methods include: adjusting the air supply temperature, adjusting coal injection, adjusting oxygen enrichment, etc., the purpose is to keep the charge stable and balanced downward and thus ensure the stable and smooth production of the blast furnace.

[0125] The present application provides a method for predicting blast furnace material rate based on the Stacking model, and the specific embodiments are as follows:

[0126] First, the process data of the blast furnace operation of a large steel plant were collected: actual material speed, air volume into the furnace, oxygen enrichment, hot air pressure, hot air temperature, gas utilization rate, humidity, coal injection amount, coal ratio, fuel ratio, dry fuel ratio, wet fuel ratio, heat load, water temperature difference, furnace top pressure, wind pressure ratio, upper differential pressure, lower differential pressure, furnace top temperature, wind consumption per ton of fuel, oxygen enrichment rate, center temperature, furnace top temperature (including temperature measurements at four locations, respectively T K1 , T K2 , T K3 , T K4 ), actual wind speed, standard wind speed, edge temperature, furnace temperature, slag ratio, material line (the first position material line and the second position material line when the blast furnace is in the first process and the second process). The actual material speed is used as a label, and other parameters are used as features.

[0127] Secondly, the process data is preprocessed to obtain target process data:

[0128] Get the average furnace top temperature (T K ):

[0129]

[0130] Set FL cl It is the first position material line when the blast furnace is in the first process; FL cr It is the second position material line when the blast furnace is in the first process; FL ol It is the first position material line when the blast furnace is in the second process; FL or The target material line FL is the second position material line when the blast furnace is in the second process.

[0131]

[0132] To calculate the actual material speed values ​​Q1 and Q2, first, arrange all the actual material speed values ​​from small to large, Q1 represents the actual material speed value at the 25th percentile, and Q2 represents the actual material speed value at the 75th percentile; the upper limit of the actual material speed is:

[0133] UL=Q2+1.5×(Q2-Q1);

[0134] The actual lower limit of material speed is:

[0135] LL = Q1 - 1.5 × (Q2 - Q1);

[0136] Based on the actual material speed upper limit and the actual material speed lower limit, an abnormal value is obtained; the abnormal value is a value in the actual material speed that is greater than the actual material speed upper limit and less than the actual material speed lower limit; the abnormal value in the actual material speed is deleted to obtain a target actual material speed; based on the process data and the target actual material speed, target process data is obtained.

[0137] The blast furnace process parameters involved in the simple calculation formula of the blast furnace operator of a large steel plant were selected: oxygen enrichment, furnace air volume, fuel ratio (before the tuyere), ton fuel wind consumption, ore batch, and material line; the first 13 important features were selected from large to small using the feature importance score of random forest: furnace air volume, oxygen enrichment, humidity, coal injection amount, coal ratio, dry fuel ratio, wet fuel ratio, ton wind fuel consumption, ore batch, standard wind speed, actual wind speed, water temperature difference, and furnace top temperature mean; the final 15 features, namely the target features, were determined: furnace air volume, oxygen enrichment, humidity, coal injection amount, coal ratio, fuel ratio, dry fuel ratio, wet fuel ratio, ton wind fuel consumption, ore batch, standard wind speed, actual wind speed, water temperature difference, furnace top temperature mean, and material line. The final selected features were combined with the actual material speed into sample data, and the sample data was divided into training set and test set according to the ratio of 7:3.

[0138] Finally, a Stacking ensemble learning model for predicting blast furnace material rate, namely the target Stacking model, is established: three base learners are set in the first layer: a linear regression algorithm, an LGBM regression algorithm, and an XGBoost regression algorithm are set. A meta-learner is set in the second layer: a linear regression algorithm is set; the set Stacking ensemble learning model is trained on the training set to obtain a Stacking ensemble learning model for predicting blast furnace material rate; the test set is input into the trained Stacking ensemble learning model for prediction to generate a predicted material rate value. A trend chart of the actual material rate value and the predicted material rate value is drawn, as shown in the figure below. Figure 4 As shown, it provides effective support for the operator's next blast furnace operation, which is beneficial to ensure the stable and smooth operation of the blast furnace.

[0139] The second aspect of the present application provides a system for predicting blast furnace material rate based on a Stacking model, comprising: an acquisition module for acquiring process data of a blast furnace to be tested; a first acquisition module configured to perform a preprocessing operation on the process data based on the process data to obtain target process data; the target process data includes: process data, target furnace top temperature, target material line, and target actual material rate; a second acquisition module configured to obtain target features based on the target process data using a feature importance score model of a random forest; the feature importance score model of the random forest is:

[0140]

[0141] Where s is the reference score of the random forest model for the target process data; i j is the importance score of the jth feature in the target process data; K is the total number of times the order of a feature in the target process data is disrupted; k is any positive integer, k≤K; a splitting module is used to divide the target feature and the target actual material speed into a training set and a test set according to a preset ratio; a third obtaining module is configured to use the training set to train the Stacking model to obtain a target Stacking model; a generating module is configured to use the target Stacking model to predict the test set and generate a predicted material speed value. The effects of the system embodiment when it is running can refer to the effects of the above-mentioned method embodiment, which will not be repeated here.

[0142] The present application provides a method and system for predicting blast furnace material speed based on the Stacking model, which can systematically and comprehensively consider the real-time process parameters of the blast furnace, reasonably and stably predict the current material speed, reduce the workload of the operator to a certain extent, and reduce the calculation error caused by human factors. At the same time, a real-time material speed trend curve is drawn to provide effective support for the operator's next blast furnace operation, which is conducive to ensuring the stable and smooth operation of the blast furnace.

[0143] The above specific implementation methods further explain in detail the purpose, technical solutions and beneficial effects of the embodiments of the present application. It should be understood that the above are only specific implementation methods of the embodiments of the present application and are not used to limit the protection scope of the embodiments of the present application. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solutions of the embodiments of the present application should be included in the protection scope of the embodiments of the present application.

Claims

1. A method for predicting blast furnace material rate based on Stacking model, characterized in that: include: Obtain the process data of the blast furnace to be tested; Based on the process data, preprocessing the process data to obtain target process data; The target process data include: process data, target furnace top temperature, target material line, target actual material speed; Based on the target process data, the target features are obtained by using the feature importance score model of random forest; the feature importance score model of random forest is: Where s is the reference score of the random forest model for the target process data; i j is the importance score of the jth feature in the target process data; K is the total number of times the order of a feature in the target process data is disrupted; k is any positive integer, k≤K; According to a preset ratio, the target features and the target actual material speed are divided into a training set and a test set; Using the training set to train the Stacking model to obtain a target Stacking model; The target Stacking model is used to predict the test set and generate a material speed prediction value.

2. The method for predicting blast furnace material rate based on the Stacking model according to claim 1, characterized in that: The process of obtaining the process data of the blast furnace to be tested includes: Obtaining the temperature, pressure and material feeding data of the blast furnace to be tested; the material feeding data includes: material feeding speed and whether stagnation occurs during material feeding; If the temperature in the furnace is within the preset temperature range, the pressure in the furnace is within the preset pressure range, the material unloading speed is within the preset speed range and there is no stagnation during unloading, the process data of the blast furnace to be tested is obtained.

3. The method for predicting blast furnace material rate based on the Stacking model according to claim 1, characterized in that: The process data include: actual material speed, air volume into the furnace, oxygen enrichment, hot air pressure, hot air temperature, gas utilization rate, humidity, coal injection amount, coal ratio, fuel ratio, dry fuel ratio, wet fuel ratio, heat load, water temperature difference, furnace top pressure, wind pressure ratio, upper differential pressure, lower differential pressure, furnace top temperature, wind consumption per ton of fuel, oxygen enrichment rate, center temperature, furnace top temperature, actual wind speed, standard wind speed, edge temperature, furnace temperature, slag ratio, ore batch, material line; The furnace top temperature includes: a first furnace top temperature, a second furnace top temperature, a third furnace top temperature, and a fourth furnace top temperature; the first furnace top temperature, the second furnace top temperature, the third furnace top temperature, and the fourth furnace top temperature are respectively collected at four positions of the blast furnace top; The material line includes: a first orientation material line and a second orientation material line when the blast furnace is in the first process and the second process; the first orientation material line and the second orientation material line are material lines in two opposite directions of the blast furnace.

4. The method for predicting blast furnace material rate based on the Stacking model according to claim 3, characterized in that: The preprocessing operation is performed on the process data based on the process data to obtain target process data, comprising: Based on the process data, the target furnace top temperature is calculated; the average value of the target furnace top temperature is: Where, T K1 is the first furnace top temperature; T K2 is the second furnace top temperature; T K3 is the third furnace top temperature; T K4 is the fourth furnace top temperature; Based on the process data, the target material line is calculated; the target material line is: Where FL cl It is the first position material line when the blast furnace is in the first process; FL cr It is the second position material line when the blast furnace is in the first process; FL ol It is the first position material line when the blast furnace is in the second process; FL or It is the second position material line when the blast furnace is in the second process; Based on the process data, the target furnace top temperature and the target material line, the target process data are obtained.

5. The method for predicting blast furnace material rate based on the Stacking model according to claim 4, characterized in that: The preprocessing operation on the process data based on the process data to obtain target process data further includes: Based on the process data, an actual material speed reference value is obtained; the actual material speed reference value includes: a first actual material speed and a second actual material speed; the first actual material speed is less than the second actual material speed; Based on the actual material speed reference value, the actual material speed upper limit is calculated; the actual material speed upper limit is: UL=Q2+1.5×(Q2-Q1); In the formula, Q2 is the second actual material speed; Q1 is the first actual material speed; Based on the actual material speed reference value, the actual material speed lower limit is calculated; the actual material speed lower limit is: LL = Q1 - 1.5 × (Q2 - Q1); Based on the actual material speed upper limit and the actual material speed lower limit, an abnormal value is obtained; the abnormal value is a value of the actual material speed that is greater than the actual material speed upper limit and less than the actual material speed lower limit; Deleting the abnormal value in the actual material speed to obtain the target actual material speed; Based on the process data and the target actual material speed, target process data are obtained.

6. The method for predicting blast furnace material rate based on Stacking model according to claim 1, characterized in that: Based on the target process data, the target features are obtained by using the feature importance score model of random forest, including: Based on the target process data, process calculation characteristics are obtained; the process calculation characteristics include: oxygen enrichment, furnace air volume, fuel ratio, air consumption per ton of fuel, ore batch, and material line; Based on the target process characteristics, using a feature importance score model of a random forest, obtaining an importance score of each feature in the target process data; Based on the importance score, obtaining an importance feature; the importance feature is a feature with a larger importance score in a preset number of the target process data; Based on the process calculation features and importance features, target features are obtained.

7. The method for predicting blast furnace material rate based on Stacking model according to claim 1, characterized in that: The Stacking model includes: a base learner and a meta learner; The training of the Stacking model using the training set to obtain a target Stacking model includes: Using the training set to train the base learner to obtain a data set; The meta-learner is trained using the data set to obtain a target Stacking model.

8. The method for predicting blast furnace material rate based on Stacking model according to claim 7, characterized in that: The base learners include: a first base learner, a second base learner, and a third base learner; The base learner is trained by using the training set to obtain a data set including: Using the training set, training the first base learner to obtain a first prediction result; Using the training set, training the second base learner to obtain a second prediction result; Using the training set, training the third base learner to obtain a third prediction result; A data set is obtained based on the first prediction result, the second prediction result, and the third prediction result.

9. The method for predicting blast furnace material rate based on Stacking model according to claim 7, characterized in that: The using the training set to train the base learner to obtain the data set also includes: Splitting the training set according to a preset ratio to obtain a training subset and a validation set; Performing a preset number of iterative training on the base learner using the training subset and the validation set, and obtaining an output result of the validation set in each iterative training; Based on the output results, a data set is obtained.

10. A system for predicting blast furnace material rate based on Stacking model, characterized in that: include: An acquisition module is used to acquire process data of the blast furnace to be tested; A first obtaining module is configured to perform a preprocessing operation on the process data based on the process data to obtain target process data; The target process data include: process data, target furnace top temperature, target material line, target actual material speed; The second obtaining module is configured to obtain the target feature based on the target process data using a feature importance score model of a random forest; the feature importance score model of the random forest is: Where s is the reference score of the random forest model for the target process data; i j is the importance score of the jth feature in the target process data; K is the total number of times the order of a feature in the target process data is disrupted; k is any positive integer, k≤K; A splitting module, used to divide the target features and the target actual material speed into a training set and a test set according to a preset ratio; A third obtaining module is configured to train the Stacking model using the training set to obtain a target Stacking model; A generation module is configured to use the target Stacking model to predict the test set and generate a material speed prediction value.

Images