A method for judging heat absorption capacity of a ladle

Abstract

The present application belongs to the technical field of casting control, and discloses a judgment method for the heat absorption capacity of a ladle, which comprises the following steps: collecting and dividing historical data; calculating process parameters related to HAC; constructing an initial model of the relationship between HAC and the temperature drop of molten steel in the ladle; constructing a final model of the relationship between HAC and the temperature drop of molten steel in the ladle; constructing a relationship model of HAC_1, HAC_2, HAC_3 and the HAC of the ladle; establishing a HAC_4 regression model to calculate the HAC value of the ladle; and screening to obtain an optimal combination model. The judgment method for the heat absorption capacity of the ladle by using characteristic quantities deeply mines and effectively integrates information data of the molten steel smelting and the ladle transportation process, comprehensively considers various factors through professional analysis and interpretation, realizes digital representation and identification of the heat absorption capacity of the ladle, and is easy to realize in field application, thereby providing strong support for fine management of the tapping temperature.

Description

Technical Field

[0001] This invention belongs to the field of casting control technology, and in particular relates to a method for judging the heat absorption capacity of a steel ladle. Background Technology

[0002] With the development of high-efficiency continuous casting technology, more stringent requirements have been placed on the temperature control of molten steel in the steelmaking-continuous casting process. The ladle, as the container holding the molten steel, has a significant heat absorption capacity that affects the steel temperature and directly restricts the tapping temperature. Accurately determining the ladle's heat absorption capacity is crucial for selecting appropriate ladles and optimizing ladle scheduling.

[0003] Currently, there are two main methods for determining the heat absorption capacity of steel ladles. The commonly used method in production is to classify the ladles, primarily considering factors such as new ladles, ladles under repair, empty ladle time, and baking time, classifying them into grades A, B, C, and D. However, the classification standards are roughly estimated manually based on experience, and vary from steel mill to steel mill, making the results susceptible to subjective influence. Furthermore, the number of grades is limited, and the factors considered are not comprehensive, failing to meet the requirements of high-efficiency continuous casting production.

[0004] Another approach is to use mathematical simulation to solve for the temperature distribution inside the ladle lining and determine the ladle's heat absorption capacity through its thermal saturation state. Mathematical simulations are generally based on heat transfer mathematical models under specific process conditions. However, the practicality of these models is affected by factors such as calculation errors due to assumptions in theoretical derivation, overly idealized treatment of ladle material properties within the model, and differences in production conditions among different steel mills. Therefore, there is an urgent need to design a new method for determining the heat absorption capacity of a ladle.

[0005] Based on the above analysis, the problems and shortcomings of the existing technology are as follows:

[0006] (1) The steel ladle grading standard is roughly estimated by human experience. The judgment result is easily affected by human subjectivity. Moreover, the number of grades is limited and the factors considered are not comprehensive, which cannot meet the requirements of efficient continuous casting production.

[0007] (2) The practicality of the judgment model is affected by the calculation error caused by the assumptions in the theoretical derivation, the overly ideal treatment of the physical property parameters of the steel ladle material in the model, and the differences in production conditions of different steel plants. Summary of the Invention

[0008] To address the problems existing in the prior art, the present invention provides a method for determining the heat absorption capacity of a steel ladle.

[0009] This invention is implemented as follows: a method for determining the heat absorption capacity of a steel ladle. The method includes: defining a characteristic quantity HAC to characterize the heat absorption capacity of the steel ladle; collecting and classifying historical data; calculating process parameters related to HAC; constructing an initial model of the relationship between HAC and the temperature drop of molten steel in the ladle; constructing a final model of the relationship between HAC and the temperature drop of molten steel in the ladle; constructing relationship models between HAC_1, HAC_2, HAC_3 and the HAC of the steel ladle; establishing a regression model of HAC_4 to calculate the HAC value of the steel ladle; and selecting the optimal combination model.

[0010] Furthermore, the method for determining the heat absorption capacity of a steel ladle includes the following steps:

[0011] Step 1: Define the characteristic quantity HAC to characterize the heat absorption capacity of the ladle;

[0012] Step 2: Divide the dataset and calculate the model parameters related to HAC;

[0013] Step 3: Construct the initial and final models of the relationship between HAC and the temperature drop of molten steel in the ladle;

[0014] Step 4: Construct a relationship model between HAC_1, HAC_2, HAC_3 and the ladle HAC;

[0015] Step 5: Establish the HAC_4 regression model and calculate the HAC value of the ladle;

[0016] Step 6: Select the optimal combination model.

[0017] Furthermore, in step one, HAC is positively correlated with the temperature drop of molten steel caused by heat absorption in the ladle.

[0018] HAC consists of four components: Environmental Component HAC_1, Ladle Temperature Component HAC_2, Ladle Physical Component HAC_3, and Ladle Genetic Component HAC_4. Among them, Environmental Component HAC_1, Ladle Temperature Component HAC_2, and Ladle Physical Component HAC_3 are related to production conditions and process parameters. Environmental Component HAC_1 includes two parameters: tapping molten steel temperature and workshop temperature. Ladle Temperature Component HAC_2 includes four parameters: the full ladle time (ladle filled with molten steel) in the previous cycle, the empty ladle time (ladle without molten steel) before loading, the baking intensity before loading, and the baking time before loading. Ladle Physical Component HAC_3 includes four parameters: ladle turnover times, nozzle usage times, permeable brick usage times, and slide plate usage times.

[0019] Furthermore, step two, which involves partitioning the dataset and calculating the model parameters related to HAC, includes:

[0020] Collect steel smelting test data from more than 6 months of normal production, with one heat as one sample; take 80% of the continuous samples as the model training set and 20% of the continuous samples as the algorithm screening set.

[0021] Process analysis and calculations were performed on the sample data to obtain the relevant parameters of HAC_1, HAC_2, and HAC_3. Among them, the full ladle time in the previous week's turnaround period = tapping time × 0.5 + time interval from the end of tapping molten steel to the start of casting + casting time × 0.5; the empty ladle time before loading steel = tapping start time - the stop casting time in the previous week's turnaround period; the baking intensity was marked as no baking 0, low heat 1, medium heat 2, and high heat 3.

[0022] Furthermore, the initial model for constructing the relationship between HAC and the temperature drop of molten steel in the ladle in step three includes:

[0023] (1) Setting the initial value of HAC: The initial value of HAC is calculated from the parameters of the environmental item HAC_1. The initial value of HAC = normalized value of tapping temperature × tapping temperature weight - normalized value of workshop temperature × workshop temperature weight. The normalization process adopts the linear interpolation method. The upper and lower limits of tapping temperature are 1500~1700℃, and the upper and lower limits of workshop temperature are -10~50℃. The actual values ​​of tapping temperature and workshop temperature are mapped to decimals between (0,1). The value range of tapping temperature weight is 0.8~1.0, and the value range of workshop temperature weight is 0~0.2.

[0024] (2) Calculate the temperature drop of molten steel in the ladle: Temperature drop of molten steel in the ladle = actual temperature drop of molten steel - temperature drop of molten steel caused by non-ladle influences; the temperature drop of molten steel caused by non-ladle influences includes the temperature drop during tapping, the temperature drop due to alloy melting and heat absorption, the temperature drop due to argon blowing and heat absorption, and the temperature drop due to the melting and heat absorption of cold steel at the bottom of the ladle; among them, the temperature drop during tapping is taken as 0.2~0.4℃ / min, the temperature drop due to alloy melting and heat absorption is obtained by thermodynamic calculation based on the type and amount of alloy added, the temperature drop due to argon blowing and heat absorption is taken as 0.1~0.2℃ / min, and the temperature drop due to the melting and heat absorption of cold steel at the bottom of the ladle is taken as 4~6℃ / t;

[0025] (3) The initial HAC value, steel tapping amount, and ladle holding time before temperature measurement of the training sample are used as input values, and the ladle molten steel temperature drop is used as the target value; where, ladle holding time before temperature measurement = tapping time × 0.5 + interval from tapping end to temperature measurement time;

[0026] (4) Different machine learning algorithms are used respectively, denoted as algorithm ①, algorithm ②, etc., and the algorithm parameters are set to construct an initial model A_Initial for the relationship between HAC and the temperature drop of molten steel in the ladle; at least two machine learning algorithms are selected, including but not limited to random forest, neural network, support vector machine, linear regression, gradient boosting tree, etc.

[0027] Furthermore, the final model for constructing the relationship between HAC and the temperature drop of molten steel in the ladle, as described in step three, includes:

[0028] (1) Compare the predicted temperature drop of molten steel in the ladle obtained by model A_Initial with the measured temperature drop of molten steel in the ladle to obtain the prediction deviation of the temperature drop of molten steel in the ladle, and calculate the mean absolute error (MAE) of the model.

[0029] (2) If the mean absolute error (MAE) is greater than the set threshold, which is 0.5 to 1, change the input value of HAC in model A_Initial. The input value of HAC = original value of HAC - prediction deviation / 100. Repeat step (1).

[0030] (3) If the mean absolute error (MAE) is less than the set threshold, the input value of HAC is specified as the actual HAC value of the ladle, and the corresponding training model is the final model A_Final, which is the relationship between HAC and the temperature drop of molten steel in the ladle.

[0031] Furthermore, step four involves constructing the relationship model between HAC_1, HAC_2, HAC_3 and the ladle HAC. This includes using training samples, taking the two parameters of the environmental term HAC_1, the four parameters of the ladle body temperature term HAC_2, and the four parameters of the ladle physical condition term HAC_3 as input values, and the actual HAC value of the ladle as the target value; and employing different machine learning algorithms, denoted as algorithms. ,algorithm ...Set the algorithm parameters and construct a model relating the environmental item HAC_1, the ladle body temperature item HAC_2, and the ladle physical condition item HAC_3 to the ladle HAC value, denoted as Model B; wherein, at least two machine learning algorithms are selected, including but not limited to random forest, neural network, support vector machine, linear regression, or gradient boosting tree, etc.

[0032] Furthermore, step five, establishing the HAC_4 regression model and calculating the HAC value of the ladle, includes:

[0033] (1) Based on model A_Final, the actual HAC value of the ladle in the first three turnover periods is obtained by back-calculating the measured temperature drop of molten steel; based on model B, the HAC value of the ladle in the first three turnover periods is obtained by HAC_1, HAC_2 and HAC_3.

[0034] (2) Set HAC_4 = actual HAC value - predicted HAC value, and use the linear fitting method to analyze the variation law of steel ladle gene item HAC_4 with turnover number to obtain the regression equation: HAC_4 = A + B × turnover number, and the correlation coefficient K of the regression equation; where A is a constant term and B is a regression coefficient;

[0035] (3) Based on model B, calculate the predicted HAC value of the ladle in this turnover period, where the input value of the molten steel temperature is the process standard value specified by the steel plant.

[0036] (4) If the correlation coefficient K is greater than the set threshold, the threshold range is 0.5 to 1.0, then the HAC value of the ladle = the predicted HAC value + HAC_4; otherwise, the HAC value of the ladle = the predicted HAC value.

[0037] Furthermore, step six involves selecting the optimal combination model, which includes:

[0038] (1) The different models A_Final obtained from Algorithm ①, Algorithm ②, etc., are compared. ,algorithm ...different models B obtained are combined in pairs to obtain a variety of combined models;

[0039] (2) The sample data of the selected furnaces are used as the input of the combined model. First, the ladle HAC value is calculated by the method in steps four and five. When using model B, the temperature of the molten steel is the actual production value. Then, the predicted value of the temperature drop of the molten steel in the ladle is obtained by model A_Final. Finally, the deviation between the predicted value of the temperature drop of the molten steel in the ladle and the actual value of the temperature drop of the molten steel in the ladle is calculated.

[0040] (3) Evaluate different combination models and obtain the combination model with the smallest mean absolute error (MAE) of temperature drop of molten steel in ladle;

[0041] (4) The selected optimal combination model is used for all furnace samples in the model training set and the algorithm selection set to obtain the final model A_Initial, model A_Final and model B.

[0042] Another object of the present invention is to provide a system for judging the heat absorption capacity of a ladle using the aforementioned method, the system comprising:

[0043] The data acquisition module is used to collect and classify historical data, define the characteristic quantity HAC to characterize the heat absorption capacity of the ladle, and calculate the process parameters related to HAC.

[0044] The model building module is used to build the initial and final models of the relationship between HAC and the temperature drop of molten steel in the ladle, and to build the relationship model B between HAC_1, HAC_2, HAC_3 and ladle HAC, and to establish the HAC_4 regression model to calculate the ladle HAC value;

[0045] The optimal combination model filtering module is used to filter and obtain the optimal combination model.

[0046] Another object of the present invention is to provide a computer device, the computer device including a memory and a processor, the memory storing a computer program, and when the computer program is executed by the processor, causing the processor to perform the steps of the method for determining the heat absorption capacity of the ladle.

[0047] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0048] First, this invention uses the RF-GBDT combined model to predict HAC and calculate the temperature drop of molten steel in 3196 consecutive heats on site. The experimental results show that the error between the predicted temperature drop and the actual temperature drop is between -3 and +3℃, and the average absolute error of the temperature drop of molten steel in the ladle is 1.34. Therefore, it is accurate and feasible to use HAC to judge the heat absorption capacity of the ladle, and it can serve as an important basis for guiding the selection of ladle and the temperature control of molten steel.

[0049] Secondly, in response to the defects and shortcomings of existing technologies, the method for judging the heat absorption capacity of a ladle using characteristic quantities provided by this invention deeply mines and effectively integrates information data from the steel smelting and ladle transportation processes. Through professional analysis and interpretation, and by comprehensively considering multiple factors, it can achieve digital characterization and identification of the heat absorption capacity of the ladle. It is also easy to implement in the field, providing strong support for the refined management of steel tapping temperature. Attached Figure Description

[0050] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a flowchart of the method for determining the heat absorption capacity of a steel ladle provided in an embodiment of the present invention;

[0052] Figure 2 This is a schematic diagram of the method for determining the heat absorption capacity of a steel ladle provided in an embodiment of the present invention;

[0053] Figure 3 This is a schematic diagram of the mean absolute error (MAE) of the temperature drop of molten steel in a ladle, provided in an embodiment of the present invention.

[0054] Figure 4 This is a distribution diagram of the predicted temperature drop error of molten steel in some heats provided in the embodiments of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0056] To address the problems existing in the prior art, the present invention provides a method for determining the heat absorption capacity of a steel ladle. The present invention will be described in detail below with reference to the accompanying drawings.

[0057] like Figure 1 As shown, the method for determining the heat absorption capacity of a steel ladle provided in this embodiment of the invention includes the following steps:

[0058] S101 defines the characteristic quantity HAC to characterize the heat absorption capacity of the ladle;

[0059] S102, collect and classify historical data, and calculate process parameters related to HAC;

[0060] S103, Construct an initial model for the relationship between HAC and the temperature drop of molten steel in the ladle;

[0061] S104, Construct the final model of the relationship between HAC and the temperature drop of molten steel in the ladle;

[0062] S105, Construct a relationship model between HAC_1, HAC_2, HAC_3 and HAC in the ladle;

[0063] S106, Establish the HAC_4 regression model and calculate the HAC value of the ladle;

[0064] S107, Select the optimal combination model.

[0065] As a preferred embodiment, such as Figure 2 As shown, the method for determining the heat absorption capacity of a steel ladle provided in this embodiment of the invention specifically includes the following steps:

[0066] Step 1: Define the characteristic quantity HAC (heat absorption capacity) to characterize the heat absorption capacity of the ladle. HAC is positively correlated with the temperature drop of the molten steel caused by heat absorption in the ladle (hereinafter referred to as ladle molten steel temperature drop).

[0067] HAC consists of four components: environmental component HAC_1, body temperature component HAC_2, physical constitution component HAC_3, and genetic component HAC_4.

[0068] The environmental item HAC_1, ladle temperature item HAC_2, and ladle physical condition item HAC_3 are related to production conditions and process parameters. The environmental item HAC_1 includes two parameters: the temperature of the molten steel at tapping and the workshop temperature. The ladle temperature item HAC_2 includes four parameters: the full ladle time (ladle filled with molten steel) in the previous cycle, the empty ladle time (ladle without molten steel) before loading steel, the baking intensity before loading steel, and the baking time before loading steel. The ladle physical condition item HAC_3 includes four parameters: the number of times the ladle is turned over, the number of times the nozzle is used, the number of times the permeable brick is used, and the number of times the slide plate is used.

[0069] Step 2: Divide the dataset and calculate the model parameters related to HAC.

[0070] 2.1 Collect steel smelting test data from more than 6 months of normal production, with one heat as one sample; take about 80% of the continuous samples as the model training set and about 20% of the continuous samples as the algorithm screening set;

[0071] 2.2 Process analysis and calculation were performed on the sample data to obtain the relevant parameters of HAC_1, HAC_2, and HAC_3; among them, the full ladle time in the previous week's turnaround period = tapping time × 0.5 + time interval from the end of tapping molten steel to the start of casting + casting time × 0.5; the empty ladle time before loading steel = tapping start time - the stop casting time in the previous week's turnaround period; the baking intensity was marked as 0 (no baking), 1 (low heat), 2 (medium heat), and 3 (high heat).

[0072] Step 3: Construct an initial model of the relationship between HAC and the temperature drop of molten steel in the ladle.

[0073] 3.1 Setting the initial HAC value: The initial HAC value is calculated from the parameters of the environmental item HAC_1. The initial HAC value = normalized tapping temperature × tapping temperature weight - normalized workshop temperature × workshop temperature weight. The normalization process uses linear interpolation. The upper and lower limits of the tapping temperature are 1500℃~1700℃, and the upper and lower limits of the workshop temperature are -10℃~50℃. The actual values ​​of the tapping temperature and the workshop temperature are mapped to decimals between (0, 1). The range of the tapping temperature weight is 0.8~1.0, and the range of the workshop temperature weight is 0~0.2.

[0074] 3.2 Calculate the temperature drop of molten steel in the ladle: Temperature drop of molten steel in the ladle = Actual temperature drop of molten steel - Temperature drop of molten steel caused by factors other than ladle influence;

[0075] The temperature drop of molten steel caused by factors other than the ladle includes the temperature drop during tapping (0.2-0.4℃ / min), the temperature drop due to the endothermic melting of the alloy (obtained by thermodynamic calculation based on the type and amount of alloy added), the temperature drop due to the endothermic blowing of argon (0.1-0.2℃ / min), and the temperature drop due to the endothermic melting of the cold steel at the bottom of the ladle (4-6℃ / t).

[0076] 3.3 The initial HAC value, tapping rate, and ladle holding time before temperature measurement of the training samples are used as input values, and the ladle molten steel temperature drop is used as the target value; where, ladle holding time before temperature measurement = tapping time × 0.5 + interval from tapping end to temperature measurement time;

[0077] 3.4 Different machine learning algorithms are used respectively, denoted as Algorithm ①, Algorithm ②, etc., and the algorithm parameters are set to construct an initial model of the relationship between HAC and the temperature drop of molten steel in the ladle (hereinafter referred to as "Model A_Initial"). At least two machine learning algorithms are used, including but not limited to: random forest, neural network, support vector machine, linear regression, gradient boosting tree, etc.

[0078] Step 4: Construct the final model of the relationship between HAC and the temperature drop of molten steel in the ladle.

[0079] 4.1 Compare the predicted temperature drop of molten steel in the ladle obtained by model A_Initial with the measured temperature drop of molten steel in the ladle to obtain the prediction deviation of the temperature drop of molten steel in the ladle, and calculate the mean absolute error (MAE) of the model;

[0080] 4.2 If the mean absolute error (MAE) is greater than the set threshold (the threshold range is 0.5 to 1), change the input value of HAC in model A_Initial. The input value of HAC = original value of HAC - prediction deviation / 100. Repeat step 4.1.

[0081] 4.3 If the mean absolute error (MAE) is less than the set threshold, the input value of HAC is specified as the actual HAC value of the ladle, and the corresponding training model is the final model of the relationship between HAC and the temperature drop of molten steel in the ladle (hereinafter referred to as "Model A_Final").

[0082] Step 5: Construct a relationship model between HAC_1, HAC_2, HAC_3 and the ladle HAC.

[0083] Using training samples, the two parameters of the environmental term HAC_1, the four parameters of the ladle body temperature term HAC_2, and the four parameters of the ladle physical condition term HAC_3 are used as input values, and the actual HAC value of the ladle is used as the target value. Different machine learning algorithms are employed, denoted as algorithms. ,algorithm ...Set the algorithm parameters and construct a model (hereinafter referred to as "Model B") relating the environmental item HAC_1, the ladle body temperature item HAC_2, and the ladle physical condition item HAC_3 to the ladle HAC value. At least two machine learning algorithms are used, including but not limited to: random forest, neural network, support vector machine, linear regression, gradient boosting tree, etc.

[0084] Step 6: Establish the HAC_4 regression model and calculate the HAC value of the ladle.

[0085] 6.1 Based on model A_Final, the actual HAC values ​​of the ladle in the first three turnover periods are obtained by back-calculating the measured temperature drop of molten steel; based on model B, the HAC values ​​of the ladle in the first three turnover periods are obtained by HAC_1, HAC_2, and HAC_3.

[0086] 6.2 Set HAC_4 = actual HAC value - predicted HAC value, and use the linear fitting method to analyze the variation law of steel ladle gene item HAC_4 with turnover number to obtain the regression equation: HAC_4 = A + B × turnover number, and the correlation coefficient K of the regression equation, where A is a constant term and B is a regression coefficient;

[0087] 6.3 Based on model B, calculate the predicted HAC value of the ladle in this turnover period, where the input value of the molten steel temperature is the process standard value specified by the steel plant;

[0088] 6.4 If the correlation coefficient K is greater than the set threshold (the threshold range is 0.5 to 1.0), then the HAC value of the ladle = the predicted HAC value + HAC_4; otherwise, the HAC value of the ladle = the predicted HAC value.

[0089] Step 7: Select the optimal combination model.

[0090] 7.1 The different models A_Final obtained from Algorithm ①, Algorithm ②, ... are compared. ,algorithm ...different models B obtained are combined in pairs to obtain a variety of combined models;

[0091] 7.2 The sample data of the selected furnace batches are used as input to the combined model. The HAC value of the ladle is calculated using the method in step 6 (when using model B, the temperature of the molten steel at tapping is the actual production value). Then, the predicted value of the temperature drop of the molten steel in the ladle is obtained through model A_Final. The deviation between the predicted value of the temperature drop of the molten steel in the ladle and the value of the temperature drop of the molten steel in the ladle obtained in step 3.2 is further calculated.

[0092] 7.3 Different combined models were evaluated to obtain the combined model with the smallest mean absolute error (MAE) of the temperature drop of molten steel in the ladle;

[0093] 7.4 The optimal combination model selected is used for all furnace samples in the model training set and the algorithm selection set to obtain the final model A_Initial, model A_Final and model B.

[0094] The ladle heat absorption capacity determination system provided in this embodiment of the invention includes:

[0095] The data acquisition module is used to collect and classify historical data, define the characteristic quantity HAC to characterize the heat absorption capacity of the ladle, and calculate the process parameters related to HAC.

[0096] The model building module is used to build the initial and final models of the relationship between HAC and the temperature drop of molten steel in the ladle, and to build the relationship model B between HAC_1, HAC_2, HAC_3 and ladle HAC, and to establish the HAC_4 regression model to calculate the ladle HAC value;

[0097] The optimal combination model filtering module is used to filter and obtain the optimal combination model.

[0098] The embodiments of the present invention have achieved some positive results during the research and development or use process, and have indeed great advantages compared with the prior art. The following content describes them in conjunction with the data, charts and other information of the experimental process.

[0099] As a preferred embodiment, such as Figure 2 As shown, the method for determining the heat absorption capacity of a steel ladle provided in this embodiment of the invention specifically includes the following steps:

[0100] 1. The data in this embodiment of the invention comes from 8 months of historical steel smelting data from a steel plant, totaling 14,479 heats. 11,041 consecutive heats from the first 6 months are used as the model training set samples, and 3,438 heats from the last 2 months are used as the model selection set samples.

[0101] The data for each sample were organized and calculated to obtain parameters related to HAC. Examples are shown in Table 1.

[0102] Table 1 Parameters related to HAC

[0103]

[0104] 2. Model Building

[0105] 2.1 Set the initial HAC value. Set the weight of tapping temperature to 0.95 and the weight of workshop temperature to 0.05.

[0106] Taking HRB400 steel with heat number 8606597 as an example, the calculation method is as follows:

[0107] Normalized tapping temperature = (1632 - 1500) / (1700 - 1500) = 0.66;

[0108] Normalized value of workshop temperature = (27+10) / (50+10) = 0.617;

[0109] HAC initial value = 0.66 × 0.95 - 0.617 × 0.05 = 0.5962.

[0110] 2.2 Calculate the temperature drop of molten steel in the ladle.

[0111] The actual temperature drop of molten steel = tapping temperature - the temperature of the first temperature measurement of molten steel after tapping. Because different steel grades undergo different processes during ladle turnover, the location of the first temperature measurement after tapping also varies. There are two main scenarios: one is temperature measurement when the molten steel leaves the argon station, and the other is temperature measurement before entering the LF station. For example, the HRB400 steel grade with furnace number 8606597 has a temperature of 1540℃ when leaving the argon station, and its actual temperature drop is 92℃.

[0112] Temperature drops not caused by the ladle include the temperature drop during tapping, the temperature drop due to argon blowing heat absorption, the temperature drop due to the melting of cold steel at the bottom of the ladle, and the temperature drop due to the addition of alloys. The temperature drop coefficients for the tapping process are set at 0.3℃ / min, for argon blowing heat absorption at 0.1℃ / min, and for the melting of cold steel at the bottom of the ladle at 5℃ / t. The values ​​for the alloy temperature drop coefficients are shown in Table 2.

[0113] Table 2 Values ​​of Alloy Temperature Drop Coefficient

[0114]

[0115] 2.3 In constructing the initial model of the relationship between HAC and the temperature drop of molten steel in the ladle, two algorithms, random forest and LASSO, were used.

[0116] In the Random Forest algorithm, the number of decision trees n_estimators = [100, 200, 300, 400, 500], the maximum depth of each decision tree max_depth = [18, 19, 20, 21, 22], the minimum number of samples that the root node of each decision tree can split min_samples_split = [2, 4, 6, 8], and the minimum number of samples that the leaf node of each decision tree min_samples_leaf = [2, 4, 8, 10, 12]. After a grid search with 10-fold cross-validation, the optimal combination of parameters is obtained as [300, 20, 2, 4]. At this time, the mean absolute error (MAE) of the temperature drop of molten steel in the ladle is 6.47. The obtained model is denoted as "Model A_Initial-RF".

[0117] In the LASSO algorithm, alpha (regularization coefficient) = 0.1, max_iter (maximum number of iterations) = 1000, tol (model convergence threshold) = 0.01, cv (cross-validation multiplicity) = 10, and selection (regression coefficients selected in each iteration) = 'cyclic'. After 10-fold cross-validation, a reasonable λ value of 0.0629 is obtained. At this time, the mean absolute error (MAE) of the temperature drop of molten steel in the ladle is 8.13. The obtained model is denoted as "Model A_Initial-LASSO".

[0118] 3. Based on "Model A_Initial-RF", the initial value of HAC is corrected for the prediction deviation of molten steel temperature drop for each training sample. For example, for the sample with furnace number 8606597, the initial HAC value is 0.5323, and the prediction deviation is 5.3. The corrected initial HAC value is 0.5323 - 5.3 / 100 = 0.4793. The corrected initial HAC values ​​for all samples are then input into Model A_Initial-RF to obtain a new "Model A_Initial-RF" and the corresponding MAE value. If the MAE value is greater than 0.8, the initial HAC value is corrected again, and the above steps are repeated. In this embodiment of the invention, after 11 iterations, the final MAE = 0.68, and the resulting model is denoted as "Model A_Final-RF", as shown in Table 3.

[0119] Table 3 shows the HAC values ​​and MAE values ​​for the loop in model A_Initial-RF.

[0120]

[0121] Based on "Model A_Initial-LASSO", the initial value of HAC is corrected for the prediction deviation of molten steel temperature drop for each training sample, resulting in a new "Model A_Initial-LASSO" and the corresponding MAE value. If the MAE value is greater than 0.8, the initial value of HAC is corrected again, and the above steps are repeated. In this embodiment of the invention, after 16 iterations, the final MAE = 0.75, and the resulting model is denoted as "Model A_Final-LASSO", as shown in Table 4.

[0122] Table 4. HAC values ​​and MAE values ​​for the corresponding cycles in model A_Initial-LASSO.

[0123]

[0124] 4. The relationship model between HAC_1, HAC_2, HAC_3 and HAC uses two algorithms: GBDT and nonlinear SVM regression.

[0125] In the GBDT algorithm, the number of base models n_estimators = [100, 200, 300, 400, 500], the maximum depth of each base model max_depth = [18, 19, 20, 21, 22], and the learning rate learning_rate = [0.01, 0.05, 0.1, 0.2].

[0126] The HAC value obtained through “Model A_Final-RF” is used as the target value of the GBDT model. After 5-fold cross-validation, the optimal combination of parameters is [300, 21, 0.1]. At this time, the mean absolute error of the HAC value is 0.0497, which is denoted as “Model B-GBDT_RF”.

[0127] The HAC value obtained through “Model A_Final-LASSO” is used as the target value of the GBDT model. After 5-fold cross-validation, the optimal combination of parameters is [400, 19, 0.05]. At this time, the mean absolute error of the HAC value is 0.0658, which is denoted as “Model B-GBDT_LASSO”.

[0128] In the nonlinear SVM algorithm, the penalty coefficient C is set to range from 100 to 1000 with a step size of 200, the epsilon value ranges from 0.1 to 1.5 with a step size of 0.2, and the gamma value ranges from 0.001 to 0.01 with a step size of 0.002.

[0129] Using the HAC value obtained through “Model A_Final-RF” as the target value of the SVM model, after 5-fold cross-validation, the optimal penalty coefficient C for nonlinear SVM regression is 300, the optimal epsilon value is 1.1, and the optimal gamma value is 0.001. At this time, the mean absolute error of the HAC value is 0.0791, which is denoted as “Model B-SVM_RF”.

[0130] Using the HAC value obtained through “Model A_Final-LASSO” as the target value of the SVM model, after 5-fold cross-validation, the optimal penalty coefficient C for nonlinear SVM regression is 500, the optimal epsilon value is 1.0, and the optimal gamma value is 0.004. At this time, the mean absolute error of the HAC value MAE = 0.0876, which is denoted as “Model B-SVM_LASSO”.

[0131] 5. Establish the HAC_4 regression model. Taking furnace number 8607633 as an example, the corresponding ladle number and turnover period are the 26th turnover period of ladle No. 2. See Table 5 for an example.

[0132] Table 5. Regression equations and correlation coefficients of HAC4 for steel ladles under different combination models.

[0133]

[0134] With a correlation coefficient threshold set at 0.6, when using model B-GBDT_RF, the HAC_4 for the 26th ladle turnover period is 0.0075 × 4 - 0.0116 = 0.0184, and the predicted HAC value is 0.5697. Therefore, the HAC value for the ladle is 0.5697 + 0.0184 = 0.5881. When using model B-GBDT_LASSO, the HAC value for the 26th ladle turnover period is not corrected, and the predicted HAC value is 0.5823. Therefore, the HAC value for the ladle is... 0.5823; When using model B-SVM_RF, the HAC value of the ladle in the 26th turnover period is not corrected, and the predicted HAC value is 0.4986, so the HAC value of the ladle is 0.4986; When using model B-SVM_LASSO, the HAC_4 of the ladle in the 26th turnover period is -0.0281×4+0.0778=-0.0358, and the predicted HAC value is 0.5256, so the HAC value of the ladle is 0.5256-0.0358=0.4898.

[0135] 6. The above model was evaluated using 3438 data points from the selected set. First, for each heat sample, the HAC value of the ladle was obtained by using models B-GBDT_RF, B-GBDT_LASSO, B-SVM_RF, and B-SVM_LASSO, combined with the HAC regression model for that heat. Then, using HAC as the input value, the predicted temperature drop of the molten steel in the ladle was obtained through the corresponding models A_Final-RF and A_Final-LASSO.

[0136] The mean absolute error (MAE) of the temperature drop of molten steel in the ladle, obtained by combining four models, is as follows: Figure 3 As shown.

[0137] The results from the MAE show that the RF-GBDT model combination is superior to the other three model combinations. Therefore, in this embodiment, RF-GBDT is selected as the optimal combination.

[0138] 7. Based on 14,479 furnace data points from the training and screening sets, the model structure and parameters were readjusted using a combination of RF-GBDT.

[0139] When constructing the initial model of the relationship between HAC and ladle molten steel temperature drop using Random Forest (RF), the number of decision trees n_estimators = [100, 200, 300, 400, 500], the maximum depth of each decision tree max_depth = [18, 19, 20, 21, 22], the minimum number of samples that the root node of each decision tree can split min_samples_split = [2, 4, 6, 8], and the minimum number of samples that the leaf node of each decision tree min_samples_leaf = [2, 4, 8, 10, 12]. After a grid search with 10-fold cross-validation, the optimal combination of parameters was obtained as [400, 21, 2, 4]. At this time, the mean absolute error (MAE) of the ladle molten steel temperature drop was 6.85.

[0140] When constructing the final model of the relationship between HAC and the temperature drop of molten steel in the ladle, the HAC value was revised 13 times, and the mean absolute error of the temperature drop of molten steel in the ladle was MAE = 0.73.

[0141] When constructing the relationship model between HAC_1, HAC_2, HAC_3 and HAC using GBDT, the number of base models n_estimators = [100, 200, 300, 400, 500], the maximum depth of each base model max_depth = [18, 19, 20, 21, 22], and the learning rate learning_rate = [0.01, 0.05, 0.1, 0.2]. After 5-fold cross-validation, the optimal combination of parameters is obtained as [400, 19, 0.05]. At this time, the mean absolute error (MAE) of the HAC values ​​is 0.0597.

[0142] 8. Application Results: The RF-GBDT combined model was used to predict the HAC (Heat Change) and calculate the temperature drop of molten steel in 3196 consecutive heats. The results show that the error between the predicted and actual temperature drops is between -3 and +3℃. The average absolute error of the temperature drop in the molten steel in the ladle is 1.34. The distribution of the predicted temperature drop error for some heats is shown below. Figure 4 As shown, using HAC to determine the heat absorption capacity of a ladle is accurate and feasible, and can serve as an important basis for guiding ladle selection and molten steel temperature control.

[0143] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0144] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for determining the heat absorption capacity of a steel ladle, characterized in that, Includes the following steps: Step 1: Define the characteristic quantity HAC to characterize the heat absorption capacity of the ladle; Step 2: Divide the dataset and calculate the model parameters related to HAC; Step 3: Construct the initial and final models of the relationship between HAC and the temperature drop of molten steel in the ladle; Step 4: Construct a relationship model between HAC_1, HAC_2, HAC_3 and the ladle HAC; Step 5: Establish the HAC_4 regression model and calculate the HAC value of the ladle; Step 6: Select the optimal combination model; In step one, HAC is positively correlated with the temperature drop of molten steel caused by heat absorption in the ladle; HAC consists of four components: Environmental Component HAC_1, Ladle Temperature Component HAC_2, Ladle Physical Component HAC_3, and Ladle Genetic Component HAC_4. Among these, Environmental Component HAC_1, Ladle Temperature Component HAC_2, and Ladle Physical Component HAC_3 are related to production conditions and process parameters. Environmental Component HAC_1 includes two parameters: tapping molten steel temperature and workshop temperature. Ladle Temperature Component HAC_2 includes four parameters: full ladle time in the previous cycle, empty ladle time before loading steel, baking intensity before loading steel, and baking time before loading steel. Ladle Physical Component HAC_3 includes four parameters: ladle turnover times, nozzle usage times, permeable brick usage times, and slide plate usage times. Step three involves constructing an initial model relating HAC to the temperature drop of molten steel in the ladle, which includes: (1) Set the initial value of HAC; (2) Calculate the temperature drop of molten steel in the ladle; (3) The initial HAC value, steel tapping amount, and ladle holding time before temperature measurement of the training sample are used as input values, and the ladle molten steel temperature drop is used as the target value; where, ladle holding time before temperature measurement = tapping time × 0.5 + interval from tapping end to temperature measurement time; (4) Different machine learning algorithms are used respectively, denoted as algorithm ①, algorithm ②, etc., and the algorithm parameters are set to construct the initial model A_Initial of the relationship between HAC and the temperature drop of molten steel in the ladle; Step three involves constructing the final model relating HAC to the temperature drop of molten steel in the ladle, which includes: (11) Compare the predicted temperature drop of molten steel in the ladle obtained by model A_Initial with the measured temperature drop of molten steel in the ladle to obtain the prediction deviation of the temperature drop of molten steel in the ladle, and calculate the mean absolute error (MAE) of the model. (22) If the mean absolute error (MAE) is greater than the set threshold, the threshold range is 0.5~1, then change the input value of HAC in model A_Initial. The input value of HAC = original value of HAC - prediction deviation / 100, and repeat step (11). (33) If the mean absolute error (MAE) is less than the set threshold, the input value of HAC is specified as the actual HAC value of the ladle, and the corresponding training model is the final model A_Final, which is the relationship between HAC and the temperature drop of molten steel in the ladle. Step four, constructing the relationship model between HAC_1, HAC_2, HAC_3 and the ladle HAC, includes: using training samples, taking two parameters of the environmental term HAC_1, four parameters of the ladle body temperature term HAC_2, and four parameters of the ladle physical condition term HAC_3 as input values, and the actual HAC value of the ladle as the target value; and employing different machine learning algorithms, denoted as algorithms. ,algorithm ...Set the algorithm parameters and construct a model relating the environmental item HAC_1, the ladle body temperature item HAC_2, and the ladle body quality item HAC_3 to the ladle HAC value, denoted as Model B; Step five, establishing the HAC_4 regression model and calculating the HAC value of the ladle, includes: (1) Based on model A_Final, the actual HAC value of the ladle in the first three turnover periods is obtained by back-calculating the measured temperature drop of molten steel; based on model B, the HAC values ​​of the ladle in the first three turnover periods are obtained by HAC_1, HAC_2 and HAC_3. (2) Set HAC_4 = actual HAC value - predicted HAC value, and use the linear fitting method to analyze the variation law of steel ladle gene item HAC_4 with turnover number to obtain the regression equation: HAC_4 = A + B × turnover number, and the correlation coefficient K of the regression equation; where A is the constant term and B is the regression coefficient; (3) Based on model B, calculate the predicted HAC value of the ladle in this turnover period, wherein the input value of the molten steel temperature is the process standard value specified by the steel plant; (4) If the correlation coefficient K is greater than the set threshold, the threshold range is 0.5~1.0, then the HAC value of the ladle = the predicted HAC value + HAC_4; otherwise, the HAC value of the ladle = the predicted HAC value.

2. The method for determining the heat absorption capacity of a steel ladle as described in claim 1, characterized in that, Step two, which involves splitting the dataset and calculating the HAC-related model parameters, includes: Collect steel smelting test data from more than 6 months of normal production, with one heat as one sample; take 80% of the continuous samples as the model training set and 20% of the continuous samples as the algorithm screening set. Process analysis and calculations were performed on the sample data to obtain the relevant parameters of HAC_1, HAC_2, and HAC_3. Among them, the full ladle time in the previous week's turnaround period = tapping time × 0.5 + time interval from the end of tapping molten steel to the start of casting + casting time × 0.5; the empty ladle time before loading steel = tapping start time - previous week's turnaround period stop casting time; the baking intensity was marked as no baking 0, low heat 1, medium heat 2, and high heat 3.

3. The method for determining the heat absorption capacity of a steel ladle as described in claim 1, characterized in that, The initial value of HAC is calculated from the parameters of the environmental item HAC_1. The initial value of HAC = normalized value of tapping temperature × tapping temperature weight - normalized value of workshop temperature × workshop temperature weight. The normalization process uses linear interpolation. The upper and lower limits of the tapping temperature are 1500~1700℃, and the upper and lower limits of the workshop temperature are -10~50℃. The actual values ​​of the tapping temperature and the workshop temperature are mapped to decimals between (0, 1). The weight of the tapping temperature ranges from 0.8 to 1.0, and the weight of the workshop temperature ranges from 0 to 0.

2. The process for calculating the temperature drop of molten steel in the ladle is as follows: Temperature drop of molten steel in the ladle = Actual temperature drop of molten steel - Temperature drop of molten steel caused by factors other than ladle influence; The temperature drop of molten steel caused by factors other than ladle influence includes the temperature drop during tapping, the temperature drop due to alloy melting and heat absorption, the temperature drop due to argon blowing and heat absorption, and the temperature drop due to the melting and heat absorption of cold steel at the bottom of the ladle; Among them, the temperature drop during tapping is taken as 0.2~0.4℃ / min, the temperature drop due to alloy melting and heat absorption is obtained by thermodynamic calculation based on the type and amount of alloy added, the temperature drop due to argon blowing and heat absorption is taken as 0.1~0.2℃ / min, and the temperature drop due to the melting and heat absorption of cold steel at the bottom of the ladle is taken as 4~6℃ / t.

4. The method for determining the heat absorption capacity of a steel ladle as described in claim 1, characterized in that, Step six involves selecting the optimal combination model, which includes: (1) The different models A_Final obtained by Algorithm ①, Algorithm ②, etc., are compared. ,algorithm ...different models B obtained are combined in pairs to obtain a variety of combined models; (2) Use the sample data of the selected furnace batches as the input of the combined model to calculate the ladle HAC value; use the actual production temperature of the molten steel when using model B; obtain the predicted temperature drop of the molten steel in the ladle through model A_Final, and calculate the deviation between the predicted temperature drop of the molten steel in the ladle and the actual temperature drop of the molten steel in the ladle. (3) Evaluate different combination models and obtain the combination model with the smallest mean absolute error (MAE) of temperature drop of molten steel in ladle; (4) The optimal combination model selected is used for all furnace samples in the model training set and the algorithm selection set to obtain the final model A_Initial, model A_Final and model B.

5. A system for determining the heat absorption capacity of a ladle using the method for determining the heat absorption capacity of a ladle as described in any one of claims 1 to 4, characterized in that, The system for determining the heat absorption capacity of a steel ladle includes: The data acquisition module is used to collect and classify historical data, define the characteristic quantity HAC to characterize the heat absorption capacity of the ladle, and calculate the process parameters related to HAC. The model building module is used to build the initial and final models of the relationship between HAC and the temperature drop of molten steel in the ladle, and to build the relationship model B between HAC_1, HAC_2, HAC_3 and ladle HAC, and to establish the HAC_4 regression model to calculate the ladle HAC value; The optimal combination model filtering module is used to filter and obtain the optimal combination model.

6. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, it causes the processor to perform the steps of the method for determining the heat absorption capacity of a ladle as described in any one of claims 1 to 4.

Images