SCADA (Supervisory Control and Data Acquisition) system for converter bottom blowing and control method thereof
By introducing a neural network optimization module with an attention mechanism into the converter bottom blowing system, the gas supply parameters are adjusted in real time, solving the problem of mismatch between process parameters in the converter blowing system and achieving higher control precision and resource saving.
Patent Information
- Application Number
- CN202511326107.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-17
- Publication Date
- 2025-12-16
AI Technical Summary
Traditional converter blowing systems suffer from inconsistent process parameters due to differences in actual raw material composition and fluctuations in molten iron state parameters, resulting in reduced molten iron purity and increased raw material and auxiliary material losses.
A neural network optimization module with an attention mechanism is used to collect data on molten iron temperature and composition in real time. The module then optimizes the gas supply parameters online, adjusting the gas supply location, type, flow rate, and temperature.
The control precision of converter bottom blowing process parameters was improved, the carbon content of ultra-low carbon steel was reduced, and the amount of bottom blowing argon gas used was reduced.
Smart Images
Figure CN121137293A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of industrial automation technology, and in particular to a SCADA system and control method for bottom blowing in converters. Background Technology
[0002] Bottom blowing is a process that supplies gas from the bottom of a converter to smelt steel products. It involves multiple parameters, including the type of gas, flow rate, temperature, and location of the gas supply. Traditional converter blowing systems control the bottom blowing gas supply using process parameters stored in a database for different steel products. However, due to differences in actual raw material composition and fluctuations in the state parameters of molten iron within the converter during the blowing process, the preset process parameters cannot perfectly match the current molten iron requirements, leading to problems such as reduced molten iron purity and increased raw material losses. Summary of the Invention
[0003] The technical problem to be solved by the present invention is to provide a SCADA system and control method for converter bottom blowing, which improves the control accuracy of converter bottom blowing process parameters.
[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows.
[0005] A SCADA system for converter bottom blowing includes, Temperature acquisition module, used to acquire the temperature of molten iron detected by the secondary lance detector; Gas flow sensor, used to detect the gas supply flow of different spray guns; The molten iron composition content acquisition module is used to acquire the molten iron composition content detected by the secondary lance detector. The air supply adjustment module is used to adjust the air supply position, air type, air flow rate and air temperature of the bottom blowing nozzle; The converter bottom blowing gas supply parameter database is used to store preset bottom blowing gas supply parameters under different raw materials and blowing process requirements; The neural network optimization module has its input terminals connected to the temperature acquisition module, gas flow sensor, molten iron composition acquisition module, and converter bottom blowing gas supply parameter database, respectively. It is used to acquire molten iron temperature, gas supply flow rate, and molten iron composition content, as well as read the preset bottom blowing gas supply parameters. The output terminal of the neural network optimization module is connected to the gas supply regulation module. It is used to send the optimized converter bottom blowing gas supply parameters to the gas supply regulation module based on the preset bottom blowing gas supply parameters and the current converter blowing status data.
[0006] A control method for the above-mentioned SCADA system for converter bottom blowing includes the following steps: The neural network optimization module reads the corresponding preset bottom blowing gas supply parameters from the converter bottom blowing gas supply parameter database according to the raw material and blowing process requirements, and sends them directly to the gas supply regulation module. As the converter blowing process proceeds, the neural network optimization module reads the molten iron temperature, gas supply flow rate and molten iron composition content data in real time, and then obtains the optimized converter bottom blowing gas supply parameters through data analysis, and sends them to the gas supply regulation module. The gas supply regulation module adjusts the gas supply position, gas type, gas supply flow rate and gas supply temperature of the bottom blowing lance according to the received optimized converter bottom blowing gas supply parameters.
[0007] Preferably, the neural network optimization module includes an input layer, several hidden layers, and an output layer arranged sequentially. The input layer is used to receive preset bottom blowing gas supply parameters, as well as data on molten iron temperature, gas supply flow rate, and molten iron composition content; Hidden layers are used to extract and learn features from the input data; The output layer is used to transform the results output by the hidden layer to obtain optimized converter bottom blowing gas supply parameters.
[0008] As a preferred method, the input layer normalizes the received preset bottom blowing gas supply parameters, as well as the data on molten iron temperature, gas supply flow rate, and molten iron composition content, and then expands them into a one-dimensional vector.
[0009] Preferably, the hidden layer includes several neurons, which are divided into several clusters. The neurons in each cluster are set with the same activation function. Different neurons in the same cluster are set with corresponding attention weights. Feedback neurons are set between different clusters. All feedback neurons are interconnected to form a feedback link. After the data is input into the hidden layer, each neuron obtains the output result according to its corresponding activation function. The output result is weighted by attention weights and used as the output of the hidden layer.
[0010] Preferably, before the first optimization calculation in the neural network optimization module (6), the preset attention weight matrix for each cluster in the hidden layer is W. Q W K W V Let X be the data vector input to the hidden layer, and calculate the query vector Q = XW. Q The bond vector K = XW K Value vector V=XW V Then calculate the attention weights. d k The dimension of the key vector.
[0011] As a preferred approach, the attention weight matrix is updated after the neuron outputs the current result; The cross-entropy loss function is used as the loss function in the attention weight matrix update process. The average gradient of the cross-entropy loss function with respect to all attention weights in each cluster is calculated. The average gradients of all clusters are then weighted to obtain the weighted gradient. The magnitude of the weight value is related to whether the cluster contains the neuron to be updated. If the cluster contains the neuron to be updated, a high weight is used for weighting; otherwise, a low weight is used. The high and low weights are obtained by preset empirical values. W Q’ =W Q -η▽ W , W K’ =W K -η▽ W , W V’ =W V -η▽ W , Among them, W Q’ W K’ W V’ Here is the updated attention weight matrix, where η is the learning rate, ▽ W This is a weighted gradient.
[0012] Preferably, the attention-weighted outputs of neurons in the same cluster first enter the feedback neurons connected to them. The feedback neurons perform Gaussian clustering on the input results and input all cluster centers into the Mean function. The output of the Mean function is used as the output of the hidden layer.
[0013] Preferably, the output layer performs a linear transformation on the received output from the hidden layer, maps it to the target space, and then selects an activation function according to the task type, using the activation function to transform the output in the target space into the corresponding output format.
[0014] The beneficial effects of adopting the above technical solution are as follows: This invention introduces neural network control logic with attention mechanism into the converter bottom blowing control system, which can realize the analysis of the real-time status of molten iron and optimize the current bottom blowing gas supply parameters online based on the analysis results, thereby improving the control accuracy of converter bottom blowing process parameters. Attached Figure Description
[0015] Figure 1 This is a system schematic diagram of the present invention. Detailed Implementation
[0016] In the following description of the embodiments, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of this application with unnecessary detail.
[0017] See Figure 1 A SCADA system for converter bottom blowing includes, Temperature acquisition module, used to acquire the temperature of molten iron detected by the secondary lance detector; Gas flow sensor, used to detect the gas supply flow of different spray guns; The molten iron composition content acquisition module is used to acquire the molten iron composition content detected by the secondary lance detector. The air supply adjustment module is used to adjust the air supply position, air type, air flow rate and air temperature of the bottom blowing nozzle; The converter bottom blowing gas supply parameter database is used to store preset bottom blowing gas supply parameters under different raw materials and blowing process requirements; The neural network optimization module has its input terminals connected to the temperature acquisition module, gas flow sensor, molten iron composition acquisition module, and converter bottom blowing gas supply parameter database, respectively. It is used to acquire molten iron temperature, gas supply flow rate, and molten iron composition content, as well as read the preset bottom blowing gas supply parameters. The output terminal of the neural network optimization module is connected to the gas supply regulation module. It is used to send the optimized converter bottom blowing gas supply parameters to the gas supply regulation module based on the preset bottom blowing gas supply parameters and the current converter blowing status data.
[0018] This invention was tested in Converter No. 2 of Tangshan Donghai Iron and Steel Group. The bottom blowing system of Converter No. 2 is equipped with three sets of blowing lances at different heights. The blowing of ultra-low carbon steel 022Cr11Ti was carried out using Converter No. 2. The auxiliary lance detector performed a sampling operation every 1 minute.
[0019] A control method for the above-mentioned SCADA system for converter bottom blowing includes the following steps: The neural network optimization module reads the corresponding preset bottom blowing gas supply parameters from the converter bottom blowing gas supply parameter database according to the raw material and blowing process requirements, and sends them directly to the gas supply regulation module.
[0020] The bottom blowing gas supply parameters for the refining of ultra-low carbon steel 022Cr11Ti are as follows: (1) Furnace lining protection: use a low-position blowing gun to purge with room temperature argon at a flow rate of 0.08 m³ / (min·t) for 30 min.
[0021] (2) Add molten iron and use low- and medium-position blowing guns to supply argon gas preheated to 250°C at a flow rate of 0.04 m³ / (min·t) for 3 minutes.
[0022] (3) Initial slag formation: Argon gas preheated to 250°C is supplied for 5 minutes using a high-level blowing gun at a flow rate of 0.06 m³ / (min·t).
[0023] (4) Mid-term enhanced decarburization: use argon gas preheated to 250°C to supply gas for 10 min at a flow rate of 0.1 m³ / (min·t) using a mid-level blowing gun.
[0024] (5) For the later stage of deep decarbonization and component homogenization, use a low-level blowing gun to supply room temperature argon gas at a flow rate of 0.15m³ / (min·t) for 15min.
[0025] As the converter blowing process proceeds, the neural network optimization module reads the data of molten iron temperature, gas flow rate, and molten iron composition in real time. Then, it obtains optimized converter bottom blowing gas supply parameters through data analysis and sends them to the gas supply regulation module. The gas supply regulation module adjusts the gas supply position, gas type, gas flow rate, and gas temperature of the bottom blowing lance according to the received optimized converter bottom blowing gas supply parameters.
[0026] The neural network optimization module consists of an input layer, several hidden layers, and an output layer, arranged sequentially.
[0027] The input layer receives preset bottom-blowing gas supply parameters, as well as data on molten iron temperature, gas flow rate, and molten iron composition content. The input layer normalizes the received preset bottom-blowing gas supply parameters and data on molten iron temperature, gas flow rate, and molten iron composition content, and then expands them into a one-dimensional vector.
[0028] The hidden layer is used for feature extraction and learning from the input data. The hidden layer consists of several neurons, which are divided into several clusters. Neurons within each cluster are assigned the same activation function. Different neurons within the same cluster are assigned corresponding attention weights. Feedback neurons are set between different clusters, and all feedback neurons are interconnected to form a feedback loop. After data is input into the hidden layer, each neuron obtains an output result based on its corresponding activation function. The output result is then weighted by the attention weights and used as the output of the hidden layer. Before the initial optimization calculation in the neural network optimization module (6), the preset attention weight matrix for each cluster in the hidden layer is W. Q W K W V Let X be the data vector input to the hidden layer, and calculate the query vector Q = XW. Q The bond vector K = XW K Value vector V=XW VThen calculate the attention weights. d k This represents the key vector dimension. After the neuron outputs the current result, it updates the attention weight matrix.
[0029] The cross-entropy loss function is used as the loss function in the attention weight matrix update process. The average gradient of the cross-entropy loss function with respect to all attention weights in each cluster is calculated. The average gradients of all clusters are then weighted to obtain the weighted gradient. The magnitude of the weight value is related to whether the cluster contains the neuron to be updated. If the cluster contains the neuron to be updated, a high weight is used for weighting; otherwise, a low weight is used. The high and low weights are obtained by preset empirical values. W Q’ =W Q -η▽ W , W K’ =W K -η▽ W , W V’ =W V -η▽ W , Among them, W Q’ W K’ W V’ Here is the updated attention weight matrix, where η is the learning rate, ▽ W This is a weighted gradient.
[0030] The results of the neurons in the same cluster, after being weighted by attention, first enter the feedback neurons connected to them. The feedback neurons perform Gaussian clustering on the input results and input all the cluster centers into the Mean function. The output of the Mean function is used as the output of the hidden layer.
[0031] The output layer transforms the output of the hidden layer to obtain optimized converter bottom blowing gas supply parameters. The output layer performs a linear transformation on the received hidden layer output, mapping it to the target space. Then, based on the task type, it selects an activation function and uses this function to transform the output in the target space into the corresponding output format.
[0032] Compared with directly using preset bottom blowing parameters, by optimizing the bottom blowing parameters online in real time through a neural network optimization module, the carbon content of the produced ultra-low carbon steel 022Cr11Ti is less than 0.008%, and the bottom blowing argon consumption is reduced by about 0.75 m³ per ton of molten iron.
[0033] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this invention, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.
Claims
1. A SCADA system for bottom blowing in a converter, characterized in that: include, Temperature acquisition module, used to acquire the temperature of molten iron detected by the secondary lance detector; Gas flow sensor, used to detect the gas supply flow of different spray guns; The molten iron composition content acquisition module is used to acquire the molten iron composition content detected by the secondary lance detector. The air supply adjustment module is used to adjust the air supply position, air type, air flow rate and air temperature of the bottom blowing nozzle; The converter bottom blowing gas supply parameter database is used to store preset bottom blowing gas supply parameters under different raw materials and blowing process requirements; The neural network optimization module has its input terminals connected to the temperature acquisition module, gas flow sensor, molten iron composition content acquisition module and converter bottom blowing gas supply parameter database, respectively. It is used to acquire molten iron temperature, gas supply flow rate and molten iron composition content, and read preset bottom blowing gas supply parameters. The output of the neural network optimization module is connected to the gas supply regulation module and is used to send optimized converter bottom blowing gas supply parameters to the gas supply regulation module based on the preset bottom blowing gas supply parameters and the current converter blowing status data.
2. A control method for a SCADA system for converter bottom blowing as described in claim 1, characterized in that... Includes the following steps: The neural network optimization module reads the corresponding preset bottom blowing gas supply parameters from the converter bottom blowing gas supply parameter database according to the raw material and blowing process requirements, and sends them directly to the gas supply regulation module. As the converter blowing process proceeds, the neural network optimization module reads the molten iron temperature, gas supply flow rate and molten iron composition content data in real time, and then obtains the optimized converter bottom blowing gas supply parameters through data analysis, and sends them to the gas supply regulation module. The gas supply regulation module adjusts the gas supply position, gas type, gas supply flow rate and gas supply temperature of the bottom blowing lance according to the received optimized converter bottom blowing gas supply parameters.
3. The control method for the SCADA system used in converter bottom blowing according to claim 2, characterized in that: The neural network optimization module consists of an input layer, several hidden layers, and an output layer, arranged sequentially. The input layer is used to receive preset bottom blowing gas supply parameters, as well as data on molten iron temperature, gas supply flow rate, and molten iron composition content; Hidden layers are used to extract and learn features from the input data; The output layer is used to transform the results output by the hidden layer to obtain optimized converter bottom blowing gas supply parameters.
4. The control method for the SCADA system used in converter bottom blowing according to claim 3, characterized in that: The input layer normalizes the received preset bottom blowing gas supply parameters, as well as the data on molten iron temperature, gas supply flow rate, and molten iron composition content, and then expands them into a one-dimensional vector.
5. The control method for a SCADA system for converter bottom blowing according to claim 3, characterized in that: The hidden layer consists of several neurons, which are divided into several clusters. The neurons in each cluster are set with the same activation function. Different neurons in the same cluster are set with corresponding attention weights. Feedback neurons are set between different clusters. All feedback neurons are interconnected to form a feedback chain. After data is input into the hidden layer, each neuron obtains an output result according to its corresponding activation function. The output result is then weighted by attention weights and used as the output of the hidden layer.
6. The control method for the SCADA system used in converter bottom blowing according to claim 5, characterized in that: Before the initial optimization calculation by the neural network optimization module, the preset attention weight matrix for each cluster in the hidden layer is W. Q W K W V Let X be the data vector input to the hidden layer, and calculate the query vector Q = XW. Q The bond vector K = XW K Value vector V=XW V Then calculate the attention weights. d k The dimension of the key vector.
7. The control method for a SCADA system for converter bottom blowing according to claim 6, characterized in that: After the neuron outputs the current result, it updates the attention weight matrix; The cross-entropy loss function is used as the loss function in the attention weight matrix update process. The average gradient of the cross-entropy loss function with respect to all attention weights in each cluster is calculated. The average gradients of all clusters are then weighted to obtain the weighted gradient. The magnitude of the weight value is related to whether the cluster contains the neuron to be updated. If the cluster contains the neuron to be updated, a high weight is used for weighting; otherwise, a low weight is used. The high and low weights are obtained by preset empirical values. IN Q’ =W Q -η▽ W , IN K’ =W K -η▽ W , IN V’ =W V -η▽ W , Among them, W Q’ W K’ W V’ Here is the updated attention weight matrix, where η is the learning rate, ▽ W This is a weighted gradient.
8. The control method for a SCADA system for converter bottom blowing according to claim 5, characterized in that: The results of the neurons in the same cluster, after being weighted by attention, first enter the feedback neurons connected to them. The feedback neurons perform Gaussian clustering on the input results and input all the cluster centers into the Mean function. The output of the Mean function is used as the output of the hidden layer.
9. The control method for a SCADA system for converter bottom blowing according to claim 8, characterized in that: The output layer performs a linear transformation on the received output from the hidden layer, mapping it to the target space. Then, it selects an activation function based on the task type and uses the activation function to transform the output in the target space into the corresponding output format.