Converter steelmaking flue gas analysis and carbon determination method

Through CFD modeling and AI algorithm optimization sampling points, combined with sensor real-time monitoring and neural network model, the sampling points are dynamically adjusted, which solves the problem that traditional sampling methods cannot represent the flue gas flow and component distribution in the flue, and achieves higher data accuracy and representativeness, improving the accuracy and response speed of converter steelmaking flue gas analysis.

CN118982359APending Publication Date: 2024-11-19QINGDAO SPECIAL STEEL CO LTD

Patent Information

Application Number
CN202410896582.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-05
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

In the traditional converter steelmaking flue gas analysis method, the fixed sampling method lacks flexibility and cannot represent the actual flow and component distribution of flue gas in the flue, resulting in low representativeness and accuracy of data, affecting the accuracy of carbon concentration. .

Method used

CFD software is used to simulate flue gas flow and component distribution, combine AI algorithm to determine the initial optimal sampling point, and monitor the flue gas changes in real time through sensors, dynamically adjust the sampling point using neural network model, and use automatic rotation sampler and multiple Point sampling is combined with the Kalman filtering model for data filtering processing and abnormal detection to ensure the accuracy and representativeness of the data.

Benefits of technology

By dynamically optimizing the location of the sampling point and data processing, the comprehensiveness and accuracy of the smoke analysis data are improved, human intervention and error are reduced, the high stability and accuracy of the data are ensured, and the response speed and decision-making accuracy are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982359A_ABST
    Figure CN118982359A_ABST
Patent Text Reader

Abstract

The invention discloses a converter steelmaking flue gas analysis and carbon determination method, and relates to the technical field of ferrous metallurgy, and the method comprises the following steps: carrying out three-dimensional modeling on a converter flue by adopting CFD software, simulating flue gas flow and component distribution, analyzing a simulation result in combination with an AI algorithm, and determining an initial optimal sampling point; establishing a flue model; acquiring a velocity field, a pressure field, a temperature field and component distribution of the flue gas; in the aspects of analyzing real-time data through a neural network model, dynamically adjusting sampling points and detecting and automatically correcting data anomaly, collected real-time flue gas data are input into a Kalman filtering model for filtering, noise and abnormal values are removed, smooth carbon content data are output, and the stability and accuracy of the data are improved; in addition, real-time data abnormity monitoring is carried out by using a Z score, after abnormal data is detected, abnormal values are corrected through regression analysis, the data are automatically corrected, abnormal conditions are fed back in real time, and an operator is prompted to adjust process parameters.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of iron and steel metallurgy, in particular to a method for determining carbon content by analyzing flue gas from converter steelmaking. Background Art

[0002] Converter flue gas is the thick brown smoke discharged during the converter blowing process. It is a mixture of furnace gas and smoke. Its main component is CO. In addition, it also contains a small amount of carbon dioxide and trace amounts of other high-temperature gases and smoke. There are two methods for collecting converter flue gas: combustion method and unburned method. The combustion method is to mix the furnace gas containing a large amount of CO with the inhaled air when it exits the furnace, burn it in the flue, generate high-temperature exhaust gas, use the heat contained in it to generate electricity, and then discharge it into the atmosphere after cooling and purification. The unburned method is to use a liftable smoke hood and an adjusting device to control the exhaust system so that the flue gas does not come into contact with the air when it exits the furnace. After cooling and purification, it is drawn into the recovery system for storage and utilization. Compared with the combustion method, the unburned method has a small amount of flue gas, a low temperature, a small purification equipment size, a small footprint, and a small investment. Since the furnace gas is unburned, the smoke particles are larger and easier to purify.

[0003] For example, the patent with publication number CN108647407B discloses a method for determining carbon by analyzing flue gas from converter steelmaking, which belongs to the field of iron and steel metallurgy. The method includes the following steps: first, based on the principle of conservation of mass, the decarburization rate of the molten pool is calculated in combination with the percentage of CO and CO2 in the flue gas and the flue gas flow rate obtained by the detection equipment during the converter smelting process; secondly, the concept of molten pool mixing degree is introduced into the terminal carbon curve fitting exponential model; then, based on the metallurgical slag molecular theory combined with the characteristics of the produced steel grade, the smelting final slag composition and the smelting terminal molten steel temperature are obtained to calculate the smelting pool limit carbon content W[C]0; finally, a converter steelmaking flue gas analysis and carbon determination model is obtained. The model in this method introduces the concept of molten pool mixing degree, fully considers the influence of the oxygen gun position, top blowing oxygen flow rate and bottom blowing gas flow rate operating process parameters on the converter molten pool decarburization process; and the molten pool limit carbon content w[C]0 is calculated based on the actual smelting steel grade using slag activity, which is more accurate.

[0004] Although the above scheme has the above advantages, the traditional converter steelmaking flue gas analysis and carbon determination method has many shortcomings in accuracy, efficiency and data processing capabilities. In the traditional sampling method, the sampling points are usually fixed and lack flexibility. The static sampling method cannot represent the actual flow and composition distribution of the flue gas in the flue, resulting in low representativeness and accuracy of the data. At the same time, fixed-point sampling is difficult to fully reflect the spatial changes of the flue gas in the flue, resulting in incomplete monitoring of key parameters, affecting the accuracy of carbon determination. Therefore, a converter steelmaking flue gas analysis and carbon determination method is urgently needed to solve such problems. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a method for analyzing and determining carbon in converter steelmaking flue gas, which solves the problems that the fixed sampling method in the prior art lacks flexibility, cannot represent the actual flow and component distribution of the flue gas in the flue, and is difficult to fully reflect the spatial changes of the flue gas in the flue, resulting in incomplete monitoring of key parameters and affecting the accuracy of carbon determination.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0007] The present invention provides a method for analyzing and determining carbon in converter steelmaking flue gas, comprising:

[0008] Step 1. Deploy sampling points, use CFD software to perform three-dimensional modeling of the converter flue, simulate flue gas flow and component distribution, analyze the simulation results with AI algorithm, and determine the initial optimal sampling point;

[0009] Step 2. Real-time sampling point adjustment: the sensor installed on the sampler monitors the flue gas flow and composition changes in real time, uses a neural network model to analyze real-time data, dynamically adjusts the sampling point, and automatically rotates the sampler to adjust its position at set time intervals and according to real-time data;

[0010] Step 3. Data anomaly detection: input the collected real-time flue gas data into the Kalman filter model for filtering, removing noise and outliers, and outputting smoothed carbon content data; and performing anomaly detection and automatic correction.

[0011] The present invention is further configured as follows: the sampling point sampling method includes:

[0012] Multi-point sampling, the samplers are installed at different positions of the flue, and the best sampling point is selected through model output;

[0013] The automatic rotary sampler is used for sampling, and the automatic rotary sampler rotates in the flue to perform uniform sampling;

[0014] The present invention is further configured as follows: the step of deploying sampling points comprises:

[0015] Obtain flue length L, diameter D, bending radius R, local expansion and contraction area dimensions, and converter parameters, including oxygen flow rate Q O2 , initial molten iron temperature T0, carbon content C0;

[0016] Use CAD software to draw the three-dimensional geometric model of the flue, divide the model into meshes, use CFD pre-processing software ANSYSMeshing to generate tetrahedral meshes, optimize the mesh quality, and ensure that the mesh skewness is less than 0.85;

[0017] Set boundary conditions and initial conditions. Boundary conditions include inlet boundary conditions, outlet boundary conditions and wall boundary conditions.

[0018] Inlet boundary condition, set the flow velocity v in , Temperature T in and smoke components, v in =v0,T in =T0,Y CO,in =Y CO,0 , Y CO2,in =Y CO2,0 ;

[0019] Among them, v in is the inlet flow rate, T in is the inlet temperature, Y CO,in is the mass fraction of CO at the inlet, Y CO2,in is the mass fraction of CO2 at the inlet;

[0020] Outlet boundary condition, set pressure P out is atmospheric pressure, P out =P atm , where P out is the outlet pressure, P atm Indicates atmospheric pressure;

[0021] The wall boundary condition is set to no slip and adiabatic, that is, v wall =0, where v wall is the wall velocity, represents the gradient of wall temperature along the normal direction;

[0022] Set initial flow rate, temperature and flue gas composition;

[0023] The present invention is further configured as follows: the step of deploying the sampling points further includes:

[0024] The turbulence model is used to describe the turbulence characteristics, the energy equation is used to describe the heat exchange between the flue gas and the flue wall, and the mass transfer model is used to simulate the diffusion and reaction of CO and CO2;

[0025] Use CFD solver to perform numerical solution and analyze the simulation results to obtain the velocity field, pressure field, temperature field and component distribution of the flue gas;

[0026] Extract key data from CFD simulation results, including velocity, pressure, temperature, and component concentration, and normalize the data;

[0027] Use convolutional neural network (CNN) to process 3D spatial data, capture spatial features, and label each sampling point based on historical data;

[0028] Then, the preprocessed data and labels are used to train the CNN model, adjust the model parameters, input the CFD simulation data into the trained CNN model, and the model outputs the representative score of each sampling point. The sampling point with the highest score is selected as the initial optimal sampling point;

[0029] The present invention is further configured to: monitor the smoke flow and composition changes based on the sensor, and construct a neural network model to analyze real-time data;

[0030] The sensors include flow rate, temperature, and gas composition sensors. The flow rate sensor is installed at the inlet of the sampler to measure the flue gas flow rate;

[0031] The temperature sensor is used to measure the flue gas temperature, and the gas composition sensor is installed inside the sampler to measure the concentration of CO and CO2;

[0032] Build a convolutional neural network model and use historical data for model training;

[0033] The preprocessed real-time data is input into the neural network model, and the model outputs the representative score and smoke flow characteristics of the current sampling point;

[0034] According to the output of the neural network model, determine whether the current sampling point needs to be adjusted;

[0035] If adjustment is required, calculate the new sampling point location;

[0036] The present invention is further configured as follows: the step of constructing a convolutional neural network model comprises:

[0037] Collect historical flue gas data from multiple samplings during the converter steelmaking process, including flow rate, temperature, CO and CO2 concentration parameters;

[0038] Normalize the data and divide the normalized data into training set, validation set and test set;

[0039] Build a CNN model with input data in the shape of a four-dimensional tensor, including batch size, time steps, number of sensors, and number of channels. The number of sensors includes flow rate, temperature, CO, and CO2 concentrations.

[0040] Add a convolution layer to extract local features of the input data through the convolution kernel. The features include smoke changes within the time step and the correlation between each sensor.

[0041] Add a ReLU activation layer after each convolution layer, and add a pooling layer after the activation layer. Use the maximum pooling operation to reduce the dimension of the data, retain important features, and reduce the amount of calculation;

[0042] The present invention is further configured as follows: the step of constructing a convolutional neural network model further includes:

[0043] Adding residual blocks to alleviate the vanishing gradient problem in deep networks and adding multi-head attention mechanisms to capture the temporal and spatial relationships in the input data;

[0044] Add a fully connected layer after the convolutional and pooling layers to flatten the multi-dimensional tensor into a vector to further extract and combine features;

[0045] The output layer outputs representative scores and smoke flow characteristics through a linear activation function;

[0046] According to the output of the CNN model, analyze the representative score of the current sampling point, and adjust the sampling point if the score is low;

[0047] Calculate the new sampling point position and dynamically adjust the position of the sampler;

[0048] The present invention is further configured as follows: the abnormality detection and automatic correction method includes:

[0049] The Z-score and isolation forest algorithms are used to monitor data anomalies in real time. After abnormal data is detected, the abnormal values ​​are automatically corrected through regression analysis, and abnormal conditions are fed back in real time to prompt operators to adjust process parameters.

[0050] The present invention is further configured as follows: the filtering process step in step 3 includes:

[0051] Set the initial state estimate based on the average of historical data and the initial error covariance matrix P0;

[0052] At each time step K, the current state and error covariance are predicted using the state transfer matrix F and the control input matrix B, P k|k-1 =FP k-1|k-1 F T +Q;

[0053] According to the observed value z k Update the state estimate and error covariance, first calculate the Kalman gain K k =P k|k-1 H T (HP k|k-1 H T +R) -1 , and then use the Kalman gain to adjust the predicted state to get an updated state estimate Finally, update the error covariance matrix, P k|k =(IK k H)P k|k-1 ,in represents the predicted state estimate at time k, represents the updated state estimate at time k, F represents the state transfer matrix, B is the control input matrix, uk represents the control input, P k|k-1 represents the prediction error covariance at time k, P k|k The updated error covariance at time k, Q represents the process noise covariance, K k represents the Kalman gain, H is the observation matrix, R represents the measurement noise covariance, z k represents the observation value at time k, and I is the unit matrix;

[0054] The present invention is further configured as follows: the abnormality detection and automatic correction method is:

[0055] Use Z-score to detect abnormal data, calculate the Z-score of each data point, and determine whether it is an outlier. Among them, X is the current data, μ is the mean value of the data, and σ is the standard deviation of the data;

[0056] The isolation forest algorithm is used to detect abnormal data, and regression analysis is used to automatically correct the detected abnormal data.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] The present invention dynamically optimizes the sampling point position through CFD modeling and AI algorithm. Multi-point sampling and automatic rotating sampler can flexibly select the most representative sampling point in the flue to ensure the accuracy and representativeness of the data.

[0059] The present invention adopts an automatic rotating sampler and multi-point sampling, which can uniformly sample in the flue, capture more comprehensive flue gas data by dynamically adjusting the sampling position, and improve the comprehensiveness and accuracy of the data;

[0060] The present invention introduces a Kalman filter model to perform real-time filtering processing, remove noise and outliers, output smooth carbon content data, and combines the Z score and isolation forest algorithm to perform real-time anomaly detection and automatic correction to ensure high stability and accuracy of the data.

[0061] The present invention adopts a convolutional neural network (CNN) model for intelligent data analysis and processing, and the automatic rotating sampler dynamically adjusts its position at a set time interval according to real-time data, thereby realizing the intelligence and automation of the sampling process and reducing human intervention and errors.

[0062] The present invention monitors the flue gas flow and composition changes based on sensors and analyzes real-time data in combination with a neural network model. It can dynamically adjust the sampling point position at set time intervals or according to real-time data to ensure that the data is synchronized with the actual production situation, thereby improving the response speed and decision-making accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 The present invention is a flow chart of the method for analyzing and determining carbon in converter steelmaking flue gas. DETAILED DESCRIPTION

[0064] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0065] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0066] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0067] Example 1

[0068] See also Figure 1 The present invention provides a method for analyzing and determining carbon in converter steelmaking flue gas, comprising:

[0069] Step 1. Deploy sampling points, use CFD software to perform three-dimensional modeling of the converter flue, simulate flue gas flow and component distribution, analyze the simulation results with AI algorithm, and determine the initial optimal sampling point;

[0070] Sampling methods include:

[0071] Multi-point sampling, the samplers are installed at different positions of the flue, and the best sampling point is selected through model output;

[0072] The automatic rotary sampler is used for sampling, and the automatic rotary sampler rotates in the flue to perform uniform sampling;

[0073] The steps for deploying sampling points include:

[0074] Obtain flue length L, diameter D, bending radius R, local expansion and contraction area dimensions, and converter parameters, including oxygen flow rate Q O2 , initial molten iron temperature T0, carbon content C0;

[0075] Use CAD software to draw the three-dimensional geometric model of the flue, divide the model into meshes, use CFD pre-processing software ANSYSMeshing to generate tetrahedral meshes, optimize the mesh quality, and ensure that the mesh skewness is less than 0.85;

[0076] Set boundary conditions and initial conditions. Boundary conditions include inlet boundary conditions, outlet boundary conditions and wall boundary conditions.

[0077] Inlet boundary condition, set the flow velocity v in , Temperature T in and smoke components, v in =v0,T in =T0,Y CO,in =Y CO,0 , Y CO2,in =Y CO2,0 ;

[0078] Among them, v in is the inlet flow rate, T in is the inlet temperature, Y CO,in is the mass fraction of CO at the inlet, Y CO2,in is the mass fraction of CO2 at the inlet;

[0079] Outlet boundary condition, set pressure P out is atmospheric pressure, P out =P atm , where P out is the outlet pressure, P atm Indicates atmospheric pressure;

[0080] The wall boundary condition is set to no slip and adiabatic, that is, v wall =0, where v wall is the wall velocity, represents the gradient of wall temperature along the normal direction;

[0081] Set initial flow rate, temperature and flue gas composition;

[0082] The turbulence model is used to describe the turbulence characteristics, the energy equation is used to describe the heat exchange between the flue gas and the flue wall, and the mass transfer model is used to simulate the diffusion and reaction of CO and CO2;

[0083] Use CFD solver to perform numerical solution and analyze the simulation results to obtain the velocity field, pressure field, temperature field and component distribution of the flue gas;

[0084] Extract key data from CFD simulation results, including velocity, pressure, temperature, and component concentration, and normalize the data;

[0085] Use convolutional neural network (CNN) to process 3D spatial data, capture spatial features, and label each sampling point based on historical data;

[0086] Then, the preprocessed data and labels are used to train the CNN model, adjust the model parameters, and input the CFD simulation data into the trained CNN model. The model outputs the representative score of each sampling point, and the sampling point with the highest score is selected as the initial optimal sampling point. Through the sampling point deployment steps and the use of CFD and AI algorithms, the converter flue is effectively three-dimensionally modeled, the flue gas flow and component distribution are simulated, and the initial optimal sampling point is optimized to ensure the sampling representativeness and measurement accuracy, and optimize the overall steelmaking process.

[0087] Step 2. Real-time sampling point adjustment: the sensor installed on the sampler monitors the flue gas flow and composition changes in real time, uses a neural network model to analyze real-time data, dynamically adjusts the sampling point, and automatically rotates the sampler to adjust its position at set time intervals and according to real-time data to ensure uniform sampling;

[0088] Based on the sensor-based monitoring of smoke flow and composition changes, a neural network model is constructed to analyze real-time data;

[0089] The sensors include flow rate, temperature, and gas composition sensors. The flow rate sensor is installed at the inlet of the sampler to measure the flue gas flow rate;

[0090] The temperature sensor is used to measure the flue gas temperature, and the gas composition sensor is installed inside the sampler to measure the concentration of CO and CO2;

[0091] Build a convolutional neural network model and use historical data for model training;

[0092] The preprocessed real-time data is input into the neural network model, and the model outputs the representative score and smoke flow characteristics of the current sampling point;

[0093] According to the output of the neural network model, determine whether the current sampling point needs to be adjusted;

[0094] If adjustment is required, calculate the new sampling point location;

[0095] The steps to build a convolutional neural network model include:

[0096] Collect historical flue gas data from multiple samplings during the converter steelmaking process, including flow rate, temperature, CO and CO2 concentration parameters;

[0097] Normalize the data and divide the normalized data into training set, validation set and test set;

[0098] Build a CNN model with input data in the shape of a four-dimensional tensor, including batch size, time steps, number of sensors, and number of channels. The number of sensors includes flow rate, temperature, CO, and CO2 concentrations.

[0099] Add a convolution layer to extract local features of the input data through the convolution kernel. The features include smoke changes within the time step and the correlation between each sensor.

[0100] Add a ReLU activation layer after each convolution layer, and add a pooling layer after the activation layer. Use the maximum pooling operation to reduce the dimension of the data, retain important features, and reduce the amount of calculation;

[0101] Adding residual blocks to alleviate the vanishing gradient problem in deep networks and adding multi-head attention mechanisms to capture the temporal and spatial relationships in the input data;

[0102] Add a fully connected layer after the convolutional and pooling layers to flatten the multi-dimensional tensor into a vector to further extract and combine features;

[0103] The output layer outputs representative scores and smoke flow characteristics through a linear activation function;

[0104] According to the output of the CNN model, analyze the representative score of the current sampling point, and adjust the sampling point if the score is low;

[0105] Calculate the new sampling point location and dynamically adjust the position of the sampler to ensure the representativeness and accuracy of the collected flue gas data;

[0106] In the model training process, weighted mean square error is used as the loss function, and the Adam optimization algorithm is used to update the model parameters. The gradient descent and momentum methods are combined to make the model training more efficient and stable.

[0107] Through the real-time sampling point adjustment steps, a convolutional neural network (CNN) model is constructed, and historical data is used for model training to achieve dynamic adjustment of real-time sampling points. Combined with the data monitored in real time by the sensor, the CNN model outputs the representative score and smoke flow characteristics of the current sampling point. Based on this information, the sampling point position is dynamically adjusted to optimize the sampling process and improve the representativeness and accuracy of the sampling.

[0108] Step 3. Data anomaly detection: input the collected real-time flue gas data into the Kalman filter model, perform filtering processing, remove noise and outliers, and output smooth carbon content data; improve data stability and accuracy; and perform anomaly detection and automatic correction;

[0109] The Z-score and isolation forest algorithms are used to monitor data anomalies in real time. After abnormal data is detected, the abnormal values ​​are automatically corrected through regression analysis, and abnormal conditions are fed back in real time to prompt operators to adjust process parameters.

[0110] Set the initial state estimate based on the average of historical data and the initial error covariance matrix P0;

[0111] At each time step K, the current state and error covariance are predicted using the state transfer matrix F and the control input matrix B, P k|k-1 =FP k-1|k-1 F T +Q;

[0112] According to the observed value z k Update the state estimate and error covariance, first calculate the Kalman gain K k =P k|k-1 H T (HP k|k-1 H T +R) -1 , and then use the Kalman gain to adjust the predicted state to get an updated state estimate Finally, update the error covariance matrix, P k|k =(IK k H)P k|k-1 ,in represents the predicted state estimate at time k, represents the updated state estimate at time k, F represents the state transfer matrix, B is the control input matrix, uk represents the control input, P k|k-1 represents the prediction error covariance at time k, P k|k The updated error covariance at time k, Q represents the process noise covariance, K k represents the Kalman gain, H is the observation matrix, R represents the measurement noise covariance, z k represents the observation value at time k, and I is the unit matrix;

[0113] The anomaly detection and automatic correction methods are:

[0114] Use Z-score to detect abnormal data, calculate the Z-score of each data point, and determine whether it is an outlier. Among them, X is the current data, μ is the mean value of the data, and σ is the standard deviation of the data;

[0115] The isolation forest algorithm is used to detect abnormal data. For the detected abnormal data, regression analysis is used for automatic correction. Through the data anomaly detection step, the real-time flue gas data is filtered using the Kalman filter model to remove noise and outliers, and smooth carbon content data is output. In addition, the Z score and the isolation forest algorithm are combined for anomaly detection. For the detected abnormal data, regression analysis is used for automatic correction, and abnormal conditions are fed back in real time to prompt operators to adjust process parameters, improve data stability and accuracy, and optimize the overall sampling and data processing process.

[0116] The present invention uses CFD software to perform three-dimensional modeling of the converter flue in the sampling point deployment stage, simulates the flue gas flow and component distribution, and analyzes the simulation results in combination with an AI algorithm to determine the initial optimal sampling point; by obtaining the length, diameter, bending radius, local expansion and contraction area size of the flue, as well as converter parameters, oxygen flow, initial molten iron temperature and carbon content, the three-dimensional geometric model of the flue is drawn using CAD software, and then the model is meshed, and ANSYS Meshing is used to generate a tetrahedral mesh; the flue model is established through boundary conditions and initial conditions;

[0117] The turbulence model is used to describe the turbulence characteristics, the energy equation is used to describe the heat exchange between the flue gas and the flue wall, the mass transfer model is used to simulate the diffusion and reaction of CO and CO2, the numerical solution is performed through the CFD solver, and the simulation results are analyzed to obtain the velocity field, pressure field, temperature field and composition distribution of the flue gas;

[0118] In terms of sampling point adjustment, the sensors installed on the sampler monitor the flue gas flow and composition changes in real time, and analyze the real-time data through the neural network model to dynamically adjust the sampling point. The flow rate sensor, temperature sensor and gas composition sensor installed in the sampler are used to construct a convolutional neural network model that is trained with historical data. The real-time data is pre-processed and then input into the model. The model outputs the representative score and flue gas flow characteristics of the current sampling point. Based on the output of the model, it is determined whether the sampling point needs to be adjusted, and the new sampling point position is calculated to dynamically adjust the position of the sampler.

[0119] In terms of data anomaly detection and automatic correction, the collected real-time flue gas data is input into the Kalman filter model for filtering, noise and outliers are removed, and smooth carbon content data is output to improve data stability and accuracy. In addition, real-time data anomaly monitoring is performed using the Z score. After abnormal data is detected, the outliers are corrected through regression analysis, the data is automatically corrected, and abnormal conditions are fed back in real time to prompt operators to adjust process parameters.

[0120] The above contents are only for explaining the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A method for determining carbon content in converter steelmaking flue gas. It is characterized in that include: Step 1. Deploy sampling points, use CFD software to perform three-dimensional modeling of the converter flue, simulate flue gas flow and component distribution, analyze the simulation results with AI algorithm, and determine the initial optimal sampling point; Step 2. Real-time sampling point adjustment: the sensor installed on the sampler monitors the flue gas flow and composition changes in real time, uses a neural network model to analyze real-time data, dynamically adjusts the sampling point, and automatically rotates the sampler to adjust its position at set time intervals and according to real-time data; Step 3. Data anomaly detection: input the collected real-time flue gas data into the Kalman filter model, perform filtering processing, remove noise and outliers, and output smooth carbon content data; And perform anomaly detection and automatic correction.

2. A method for determining carbon content in converter steelmaking flue gas according to claim 1, It is characterized in that Sampling methods include: Multi-point sampling, the samplers are installed at different positions of the flue, and the best sampling point is selected through model output; The sampling is carried out by using an automatic rotary sampler, which rotates in the flue to carry out uniform sampling.

3. A method for determining carbon content by analyzing converter steelmaking flue gas according to claim 2, It is characterized in that The steps for deploying sampling points include: Obtain flue length L, diameter D, bending radius R, local expansion and contraction area dimensions, and converter parameters, including oxygen flow rate Q O2 , initial molten iron temperature T 0 , Carbon content C 0 ; Use CAD software to draw the three-dimensional geometric model of the flue, divide the model into meshes, use CFD pre-processing software ANSYSMeshing to generate tetrahedral meshes, optimize the mesh quality, and ensure that the mesh skewness is less than 0.85; Set boundary conditions and initial conditions. Boundary conditions include inlet boundary conditions, outlet boundary conditions and wall boundary conditions. Inlet boundary condition, set the flow velocity v in , Temperature T in and smoke components, v in =v 0 , T in =T 0 , Y CO,in =Y CO,0 , Y CO2,in =Y CO2,0 ; Among them, v in is the inlet flow rate, T in is the inlet temperature, Y CO,in is the mass fraction of CO at the inlet, Y CO2,in is the mass fraction of CO2 at the inlet; Outlet boundary condition, set pressure P out is atmospheric pressure, P out =P atm , where P out is the outlet pressure, P atm Indicates atmospheric pressure; The wall boundary condition is set to no slip and adiabatic, that is, v wall =0, where v wall is the wall velocity, represents the gradient of wall temperature along the normal direction; Set the initial flow rate, temperature and flue gas composition.

4. A method for determining carbon content by analyzing converter steelmaking flue gas according to claim 3, It is characterized in that The steps for deploying sampling points also include: The turbulence model is used to describe the turbulence characteristics, the energy equation is used to describe the heat exchange between the flue gas and the flue wall, and the mass transfer model is used to simulate the diffusion and reaction of CO and CO2; Use CFD solver to perform numerical solution and analyze the simulation results to obtain the velocity field, pressure field, temperature field and component distribution of the flue gas; Extract key data from CFD simulation results, including velocity, pressure, temperature, and component concentration, and normalize the data; Use convolutional neural network (CNN) to process 3D spatial data, capture spatial features, and label each sampling point based on historical data; Then, the preprocessed data and labels are used to train the CNN model, adjust the model parameters, input the CFD simulation data into the trained CNN model, and the model outputs the representative score of each sampling point. The sampling point with the highest score is selected as the initial optimal sampling point.

5. A method for determining carbon content by analyzing converter steelmaking flue gas according to claim 4, It is characterized in that Based on the sensor-based monitoring of smoke flow and composition changes, a neural network model is constructed to analyze real-time data; The sensors include flow rate, temperature, and gas composition sensors. The flow rate sensor is installed at the inlet of the sampler to measure the flue gas flow rate; The temperature sensor is used to measure the flue gas temperature, and the gas composition sensor is installed inside the sampler to measure the concentration of CO and CO2; Build a convolutional neural network model and use historical data for model training; The preprocessed real-time data is input into the neural network model, and the model outputs the representative score and smoke flow characteristics of the current sampling point; According to the output of the neural network model, determine whether the current sampling point needs to be adjusted; If adjustment is required, the new sampling point locations are calculated.

6. A method for determining carbon content in converter steelmaking flue gas according to claim 5, It is characterized in that The steps to build a convolutional neural network model include: Collect historical flue gas data from multiple samplings during the converter steelmaking process, including flow rate, temperature, CO and CO2 concentration parameters; Normalize the data and divide the normalized data into training set, validation set and test set; Build a CNN model with input data in the shape of a four-dimensional tensor, including batch size, time steps, number of sensors, and number of channels. The number of sensors includes flow rate, temperature, CO, and CO2 concentrations. Add a convolution layer to extract local features of the input data through the convolution kernel. The features include smoke changes within the time step and the correlation between each sensor. A ReLU activation layer is added after each convolution layer, and a pooling layer is added after the activation layer. The maximum pooling operation is used to reduce the dimension of the data, retain important features, and reduce the amount of calculation.

7. A method for determining carbon content in converter steelmaking flue gas according to claim 6, It is characterized in that The steps to build a convolutional neural network model also include: Adding residual blocks to alleviate the vanishing gradient problem in deep networks and adding multi-head attention mechanisms to capture the temporal and spatial relationships in the input data; Add a fully connected layer after the convolutional and pooling layers to flatten the multi-dimensional tensor into a vector to further extract and combine features; The output layer outputs representative scores and smoke flow characteristics through a linear activation function; According to the output of the CNN model, analyze the representative score of the current sampling point, and adjust the sampling point if the score is low; Calculate the new sampling point position and dynamically adjust the position of the sampler.

8. A method for determining carbon content in converter steelmaking flue gas according to claim 7, It is characterized in that Anomaly detection and automatic correction methods include: The Z-score and isolation forest algorithms are used to monitor data anomalies in real time. After abnormal data is detected, the outliers are automatically corrected through regression analysis, and abnormal situations are fed back in real time to prompt operators to adjust process parameters.

9. A method for determining carbon content in converter steelmaking flue gas according to claim 8, It is characterized in that The filtering process in step 3 includes: Set the initial state estimate based on the average of historical data and the initial error covariance matrix P 0 ; At each time step K, the current state and error covariance are predicted using the state transfer matrix F and the control input matrix B, According to the observed value z k Update the state estimate and error covariance, first calculate the Kalman gain K k =P k|k-1 H T (HP k|k-1 H T +R) -1 , and then use the Kalman gain to adjust the predicted state to get an updated state estimate Finally, update the error covariance matrix, P k|k =(IK k H)P k|k-1 ,in represents the predicted state estimate at time k, represents the updated state estimate at time k, F represents the state transfer matrix, B is the control input matrix, and uk represents PP Control input, P k|k-1 represents the prediction error covariance at time k, F k|k The updated error covariance at time k, Q represents the process noise covariance, K k represents the Kalman gain, H is the observation matrix, R represents the measurement noise covariance, z k represents the observation value at time k, and I is the unit matrix.

10. A method for determining carbon content in converter steelmaking flue gas according to claim 9, It is characterized in that The anomaly detection and automatic correction methods are: Use Z-score to detect abnormal data, calculate the Z-score of each data point, and determine whether it is an outlier. Among them, X is the current data, μ is the mean value of the data, and σ is the standard deviation of the data; The isolation forest algorithm is used to detect abnormal data, and regression analysis is used to automatically correct the detected abnormal data.

Citation Information

Patent Citations

  • A method for carbon determination of converter steelmaking flue gas

    CN108647407B

Cited By

  • Monitoring system for converter gas recovery

    TWI925778B