Alloy addition amount control method and system in steelmaking process

Abstract

The invention belongs to the technical field of metallurgical process control, and discloses a method and system for controlling the alloy addition amount in the steelmaking process. The method comprises the steps that state information such as molten steel components, temperature and weight is collected in real time; alloy components are obtained, and molten steel target components are set; forecasting the yield of the current heat in real time by using an alloy element yield prediction model constructed based on a neural network; and then, by taking the lowest total cost of alloy addition as an optimization target, calculating an optimal alloy addition amount scheme meeting all component constraints by adopting an optimization algorithm. The corresponding system comprises a data acquisition module, an alloy component management module, a target setting module, a yield prediction module, an optimization calculation module and an output module. According to the method, intelligent prediction and cost optimization are combined, the defect that a conventional method depends on a fixed yield empirical value is overcome, precise control over alloy adding and cost minimization are achieved, and the method is suitable for various steelmaking processes such as a converter, an electric furnace and refining.

Description

Technical Field

[0001] This invention belongs to the field of metallurgical process control technology, specifically relating to a method and system for controlling the amount of alloy added during steelmaking. Background Technology

[0002] With the increasing demands for product quality and production efficiency in the modern steel industry, precise control of alloying elements during steelmaking has become a crucial step in ensuring the mechanical properties, corrosion resistance, and process stability of steel. In actual smelting operations, to meet the composition design requirements of different steel grades, various alloying elements such as carbon, silicon, manganese, chromium, and niobium need to be added to the molten steel. However, traditional alloying methods heavily rely on the operator's experience and judgment, lacking a systematic quantitative analysis of the current state of the molten steel, alloy yield, and cost constraints. This results in large fluctuations in composition control, low first-time success rates, and frequent need for secondary alloy additions, which not only prolongs the smelting cycle but also significantly increases raw material consumption and energy waste.

[0003] The accurate calculation of alloy addition depends on a comprehensive consideration of multiple parameters, including the initial composition of the molten steel, the target composition, the amount of molten steel, the alloy grade, and the elemental yield. While existing technologies have attempted to incorporate basic material balance formulas for estimation, they generally treat the alloy yield as a fixed empirical value, neglecting its dynamic changes under different temperatures, oxygen potentials, molten steel compositions, and operating conditions. This leads to significant deviations between theoretical calculations and actual results. Furthermore, there are compositional coupling relationships among various alloys; adjusting a single element can affect the final content of other elements. Traditional methods lack the ability to perform synergistic optimization of multiple variables, making it difficult to minimize alloy costs while satisfying all compositional constraints.

[0004] Therefore, there is an urgent need for an intelligent control method and system for alloy addition in the steelmaking process that integrates real-time data perception, dynamic yield prediction, and multi-objective optimization decision-making, in order to break through the bottleneck of experience dependence and improve the accuracy of composition control and production economy. Summary of the Invention The purpose of this invention is to provide a method and system for controlling the amount of alloy added during the steelmaking process, which can effectively solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is a method for controlling the amount of alloy added in the steelmaking process, comprising the following steps: Step 1: Collect the molten steel state information of each smelting furnace in real time. The molten steel state information includes the molten steel composition, temperature and weight. Step 2: Obtain the composition information of each alloy category; Step 3: Set the target content of each alloying element in the molten steel; the target content can be automatically collected from the system or manually entered; when the automatic collection method is used, the system retrieves the corresponding steel grade composition requirements from the preset steel grade standard composition database; when the manual input method is used, the operator inputs the specific composition control target value through the human-machine interface.

[0006] Step 4: Based on the molten steel state information, the recovery rate of each alloy element in the current heat is predicted in real time using a pre-trained neural network model; the alloy element recovery rate refers to the ratio of the mass of alloy in the molten steel during the production process to the mass of alloy actually added to the molten steel.

[0007] (1) In the formula, m represents the mass of alloying element i added, in kg; m g This indicates the mass of the alloy in the molten steel after its addition, expressed in kg. m d This indicates the mass of alloying element i in the molten steel before the addition of the alloy, in kg; This indicates the content of alloying element i in the ferroalloy, in % %. Y This represents the yield when an iron alloy containing element i is added to molten steel.

[0008] The yield of alloying elements can be calculated based on a neural network model, or it can be manually entered based on actual production experience. Step 5: Based on the molten steel state information, the alloy composition information, the target content, and the predicted alloy element yield, with the goal of minimizing the total cost of alloy addition, use an optimization algorithm to calculate the alloy addition amount scheme that satisfies all composition constraints.

[0009] Preferably, the production process database in step 1 is a primary or secondary database used on-site. The database type includes Oracle, SQL Server, DB2, etc. These database systems can provide stable and reliable data storage and access services to ensure the integrity and real-time performance of steelmaking process data.

[0010] Preferably, the steel composition information in step 1 includes carbon, silicon, manganese, phosphorus, chromium, niobium, vanadium, titanium and free oxygen. Real-time monitoring data of these elements provides accurate initial state parameters for subsequent alloy addition calculation. Preferably, the alloy types in step 2 include ferrosilicon, ferromanganese, ferrotitanium, and ferrochrome. The composition information of these alloys is updated in real time through the enterprise's material management system to ensure that the alloy grade data used for calculation is accurate and reliable.

[0011] Preferably, the method for predicting the yield of alloying elements in step 4 is to calculate the yield of ferroalloys based on neural network technology, or the yield can be manually input based on actual production experience. When a neural network model is used, the system can adaptively learn the variation law of yield under different process conditions. When experience input is used, the system allows operators to input the experience yield value based on actual production conditions.

[0012] More preferably, in step 4, the training process of the neural network model includes: collecting historical production data to construct a training dataset, wherein the historical production data includes steel molten state information and the corresponding actual alloy element yield; The training dataset is normalized. The neural network is trained and tested using the processed dataset until the model's prediction accuracy reaches the predetermined requirements.

[0013] Real-time data collection for the current furnace batch is used to input the collected data into the established alloy yield prediction model, and the yield of ferroalloys under the current furnace conditions is calculated.

[0014] A further preferred formula for normalizing the data is: (2); In the formula: y is the normalized value; min To find the minimum value of y, we take the value 1; y max Let y be the maximum value, set to 1, and let X be the original input variable. min and X max X represents the minimum and maximum values ​​of the input variable determined based on historical data statistics; min and X max Based on historical data statistics, normalization can eliminate differences between data of different dimensions, thereby improving the stability and convergence speed of neural network training.

[0015] More preferably, in the neural network model, the transfer function between neurons adopts a combination of sigmoid function and linear function. The sigmoid function is used for the intermediate layer; the linear function is used for numerical prediction of the output layer. The weights of the connections between neuron nodes and the thresholds of neuron nodes are continuously adjusted and optimized through the training process.

[0016] More preferably, the number of intermediate layer nodes, num, in the neural network model is determined based on the number of input layer nodes and the number of output layer nodes, ensuring that the neural network has sufficient expressive power without overfitting. The calculation formula is as follows: (3); In the formula, m is the number of input layer nodes, n is the number of output layer nodes, and a ranges from 1 to 15.

[0017] More preferably, the transfer function formula between neurons in the neural network model is as follows: (4); (5); In equations (3) and (4), w represents the weight of the connection between neuron nodes, i represents the i-th node of the current layer of the neural network, j represents the j-th node of the previous layer of the neural network, x represents the normalized variable, and θ represents the threshold of the neuron node.

[0018] Furthermore, the root mean square error (RMSE) value is used to evaluate the effectiveness of the yield prediction model. The formula for calculating the RMSE value is as follows: (6); In the formula: Y Cal The alloy yield, Y, calculated by the alloy element yield prediction model. Exp The actual alloy element yield is represented by n, which represents the number of test data selected. The root mean square error (RMSE) value is controlled between 3% and 10%. When the RMSE value is less than 5%, the model training effect is considered to have met the requirements.

[0019] In a further preferred embodiment, the target content of alloying elements, the content under current conditions, and the content parameters of element i in ferroalloy in formula (5) are all collected from the actual production process database; the alloying element recovery rate is calculated by the alloying element recovery rate prediction model, or it can be input into the system database based on actual production experience for data collection.

[0020] Preferably, the calculation of the alloy addition amount in step 5 is based on the principle of material balance, taking into account multiple factors such as the initial composition of molten steel, target composition, molten steel quantity, alloy grade and element recovery rate. Through accurate calculation, it is ensured that the alloy addition amount meets the composition control requirements while avoiding waste caused by excessive addition.

[0021] Furthermore, the formula for calculating the amount of single alloying element i added in step 5 is as follows: (7); In the formula, m i Indicates the mass of alloying element i added, in kg; w ( i ) g This indicates the target content of alloying element i in steel, % . w ( i ) dThis indicates the content of alloying element i in the steel under the current conditions, in % . m steel The mass of molten steel is expressed in kg. This indicates the content of alloying element i in the ferroalloy, in % %. Y ( i () represents the yield when an iron alloy containing element i is added to molten steel.

[0022] Furthermore, the optimization algorithm applied to the calculation of alloy addition can ensure that the alloy usage cost is minimized while meeting the target composition conditions of the molten steel. The constraint conditions between alloy elements ensure that the content of each element reaches the target range at the same time, and the control effect of other elements will not be affected by the adjustment of a single element.

[0023] The optimization formula for calculating the minimum usage cost of the alloy is as follows: (8); In the formula, Cost is the cost of the total amount of alloy added. i Here is the unit price of alloying element i, in yuan / kg; m i Let be the mass of alloying element i added, in kg.

[0024] In the formula for calculating the cost of using the alloy, the unit price of the alloy element is updated in real time according to market conditions, and the mass of alloy added is obtained through optimized calculation. The system can automatically select the alloy ratio scheme with the best cost performance based on the principle of optimal cost.

[0025] Further preferably, the cost of the total alloy addition is calculated using the aforementioned optimization algorithm. Available algorithms include interior-point methods, quadratic programming, and penalty function methods. These algorithms can effectively handle multivariate constrained optimization problems, achieving cost optimization while ensuring the accuracy of composition control.

[0026] Preferably, the control method of the present invention is applicable to various steelmaking processes, including converter steelmaking, electric furnace steelmaking and refining processes, and can handle the problem of controlling the amount of alloy added for different steel grades such as carbon steel, alloy steel and stainless steel.

[0027] This invention enables fully automated operation, requiring no human intervention from data acquisition and yield prediction to alloy addition calculation and optimization decision-making, thus greatly improving production efficiency and control accuracy.

[0028] This invention also discloses an alloy addition control system for steelmaking processes, comprising: 1) Data Acquisition Module: Used to collect real-time information on the composition, temperature, and weight of molten steel; responsible for obtaining the initial state parameters of molten steel and alloy material information from the enterprise's on-site database in real time; 2) Alloy Composition Management Module: Stores and manages composition data for various alloys; 3) Target setting module: Used to set the target content of each alloying element in molten steel; 4) Yield prediction module: Real-time prediction of alloy yield based on a neural network model; 5) Optimization Calculation Module: Used to calculate the optimal alloy addition amount scheme with the goal of minimizing the total alloy cost; 6) Output module: Outputs the optimal alloy addition scheme.

[0029] The system has a self-learning function, which can continuously correct the yield prediction model based on actual production results. As production data accumulates, the prediction accuracy of the model will continue to improve, forming a virtuous cycle of self-improvement.

[0030] Preferably, the system provides a complete human-computer interaction interface, allowing operators to monitor the control process in real time, view detailed explanations of the calculation results, and make manual interventions and parameter adjustments when necessary.

[0031] Preferably, the system establishes a comprehensive data management mechanism, where all production process data, model parameters, and operation records are stored in the database, facilitating subsequent data analysis and process optimization.

[0032] Compared with the prior art, the beneficial effects of the present invention are: This invention improves the precision control of molten steel composition by collecting real-time production data from the steelmaking process and predicting the current ferroalloy yield based on neural network technology. This determines the optimal amount of ferroalloy to be added under the current conditions, reduces the need for secondary alloy addition during production, lowers production costs, and increases enterprise efficiency. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a comparison chart of the actual amount of ferrosilicon alloy added in a certain casting process of low carbon steel production and the calculated results. Detailed Implementation To better understand the present invention, the following embodiments and accompanying drawings further illustrate the content of the present invention, but the content of the present invention is not limited to the following embodiments.

[0034] Example 1: This embodiment uses the steelmaking process of a certain steel grade as an example to explain in detail the method for controlling the optimal amount of alloy addition. The specific steps are as follows: Step 1: Real-time acquisition of smelting process data Key process parameters for the target steel grade before alloy addition were read in real time from the on-site secondary Oracle database and the enterprise production process management system. Specific data collected are as follows: The chemical composition of the molten steel, by mass content, is as follows: C: 0.12%, Si: 0.05%, Mn: 0.50%, P: 0.01%, S: 0.004%, Al: 0.02%, Ca: 0.0003%; Weight of molten steel: 254t; Molten steel temperature: 1610℃.

[0035] Step 2: Obtain alloy raw material composition information Retrieve the composition data of the alloy used in the current batch from the alloy raw material database: Ferrosilicon alloy: Si content is 75% by mass; Medium carbon ferromanganese alloy: Mn content is 80%.

[0036] Step 3: Set the target composition for the steel grade According to the product standard for this steel grade, the target composition after alloying is set as follows, based on mass content: C: 0.14%, Si: 0.07%, Mn: 0.60%, P: 0.01%, S: 0.004%, Al: 0.02%, Ca: 0.0003%, Free oxygen: 0.0003%.

[0037] Step 4: Establish a model for predicting the recovery rate of alloying elements and make predictions in real time. 4.1 Data Acquisition and Preprocessing: Historical production data for this steel grade was collected, including the content of elements C, Si, Mn, P, S, Al, Ca, and O in the molten steel, as well as the temperature of the molten steel. The collected data was normalized using the following formula: (2); In the formula: y is the normalized value; min To find the minimum value of y, we take the value 1; y max Let y be the maximum value, set to 1, and let X be the input variable. min Let X be the minimum value of X. max Let X be the maximum value. 4.2 The neural network model is calculated using the following formula: (3); In the formula, m is the number of input layer nodes, n is the number of output layer nodes, and a ranges from 1 to 15.

[0038] In this embodiment: the number of input layer nodes m=9; the number of intermediate layer nodes: 8; the number of output layer nodes n=1. Maximum number of iterations: 15000.

[0039] 4.3 Neuron Transfer Function: The following function is used as the transfer function between neurons: (4); (5); In the formula, x represents the normalized variable, w represents the weight of the connection between neuron nodes, and θ represents the threshold of the neuron node.

[0040] 4.4. Model Training and Evaluation: The root mean square error (RMSE) is used to evaluate the model's predictive performance. The calculation formula is as follows: (6); In the formula: Y Cal The alloy yield, Y, calculated by the alloy element yield prediction model. Exp The actual yield of alloying elements is represented by n, which represents the number of test data selected. When the RMSE value is less than 5%, the model training is considered complete.

[0041] 4.5. Real-time yield prediction: The trained model is used to predict the yield of the current batch, resulting in: Ferrosilicon alloy yield: 81%; medium-carbon ferromanganese alloy yield: 86%.

[0042] Step 5: Calculate alloy addition amount and optimize cost 5.1. Calculation of Alloy Addition Amount: The mass of each alloying element added is calculated based on the following formula: (7); In the formula, m i Indicates the mass of alloying element i added, in kg; w ( i ) g This indicates the target content of alloying element i in steel, % . w ( i ) d This indicates the content of alloying element i in the steel under the current conditions, in % . m steel The mass of molten steel is expressed in kg. This indicates the content of alloying element i in the ferroalloy, in % %. Y ( i () represents the yield when an iron alloy containing element i is added to molten steel.

[0043] 5.2. Cost Optimization Calculation: With the goal of minimizing the total cost of the alloy, a quadratic programming method is used for optimization. The objective function is: (8); In the formula, Cost is the cost of the total amount of alloy added. i Here is the unit price of alloying element i, in yuan / kg; m i Let be the mass of alloying element i added, in kg.

[0044] Optimization results: After calculation and optimization using the above method, the optimal addition amount of the alloy is obtained as follows: Ferrosilicon: 83.6 kg; Medium carbon ferromanganese: 369.2 kg.

[0045] This embodiment achieves precise control of alloy addition amount through real-time data acquisition, accurate yield prediction and cost optimization, which significantly reduces the cost of alloy use while ensuring that the composition of molten steel meets the standards.

[0046] Example 2: This embodiment discloses an alloy addition control system for steelmaking processes, the system comprising: 1) Data acquisition module: used to collect real-time information on the composition, temperature, and weight of molten steel; 2) Alloy composition management module: Stores composition data for various alloys; 3) Target setting module: Used to set the target composition of the steel grade; 4) Yield prediction module: Real-time prediction of alloy yield based on a neural network model; 5) Optimized calculation module: Employs an optimization algorithm to calculate the alloy addition amount with the lowest cost; 6) Output module: Outputs the optimal alloy addition scheme.

[0047] The system achieves intelligent and optimized control of the alloy addition process through the coordinated work of its various modules.

[0048] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for controlling the amount of alloy added during steelmaking, characterized in that, Includes the following steps: Step 1: Collect the molten steel state information of each smelting furnace in real time. The molten steel state information includes the molten steel composition, temperature and weight. Step 2: Obtain the composition information of the various alloys currently in use; Step 3: Set the target content of each alloying element in the molten steel; Step 4: Based on the molten steel state information, predict the yield of each alloying element in the current furnace in real time using a pre-trained neural network model; Step 5: Based on the molten steel state information, the alloy composition information, the target content, and the predicted yield, with the goal of minimizing the total cost of alloy addition, use an optimization algorithm to calculate the alloy addition amount scheme that satisfies all composition constraints.

2. The method according to claim 1, characterized in that, In step 4, the training process of the neural network model includes: A training dataset is constructed by collecting historical production data, which includes information on the state of molten steel and the corresponding actual yield of alloying elements. The training dataset is normalized. The neural network is trained and tested using the processed dataset until the model's prediction accuracy reaches the predetermined requirements.

3. The method according to claim 2, characterized in that, The formula used to normalize the data is: (2) In the formula: y is the normalized value; min To find the minimum value of y, we take the value 1; y max Let y be the maximum value, set to 1, and let X be the original input variable. min and X max These are the minimum and maximum values ​​of the input variables determined based on historical data statistics.

4. The method according to claim 2, characterized in that, In the neural network model, the transfer function between neurons is a combination of sigmoid and linear functions; the sigmoid function is used in the intermediate layer, and the linear function is used in the output layer.

5. The method according to claim 4, characterized in that, The number of intermediate layer nodes, num, in the neural network model is determined by the following formula, based on the number of input layer nodes and the number of output layer nodes: (3)； In the formula, m is the number of input layer nodes, n is the number of output layer nodes, and a ranges from 1 to 15.

6. The method according to claim 4, characterized in that, The transfer function formula between neurons in the neural network model is as follows: (4) (5) In equations (4) and (5), w represents the weight of the connection between neuron nodes, i represents the i-th node of the current layer of the neural network, j represents the j-th node of the previous layer of the neural network, x represents the normalized variable, and θ represents the threshold of the neuron node.

7. The method according to claim 1, characterized in that, In step 5, the optimization algorithm is any one of the interior point method, quadratic programming method, or penalty function method.

8. The method according to claim 1, characterized in that, The formula for calculating the amount of a single alloying element i added is as follows: (7)； In the formula, m i Indicates the mass of alloying element i added, in kg; w ( i ) g This indicates the target content of alloying element i in steel, % . w ( i ) d This indicates the content of alloying element i in the steel under the current conditions, in % . m steel The mass of molten steel is expressed in kg. This indicates the content of alloying element i in the ferroalloy, in % %. Y ( i () represents the yield when an iron alloy containing element i is added to molten steel.

9. The method according to claim 1, characterized in that, The objective function for the optimization objective is: (8)； In the formula, Cost is the cost of the total amount of alloy added. i Here is the unit price of alloying element i, in yuan / kg; m i Let be the mass of alloying element i added, in kg.

10. A control system for alloy addition in a steelmaking process, used to implement the method as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to collect real-time information on the state of molten steel and alloy materials. The alloy composition management module is used to store and manage the composition data of various alloys; The target setting module is used to set the target content of each alloying element in molten steel; The yield prediction module has a built-in neural network model as described in claim 1, which is used to predict the yield of each alloy element in real time. The optimization calculation module is used to calculate the optimal alloy addition scheme with the goal of minimizing the total alloy cost; The output module is used to output the optimal alloy addition scheme.

Images