Method and system for on-line determination of carbon content of molten steel in converter steelmaking

Abstract

The invention provides a method and system for on-line determination of molten steel carbon content in converter steelmaking, and relates to the technical field of converter steelmaking, the method comprises the following steps: obtaining multi-source process data, the multi-source process data comprising exhaust gas components, furnace mouth flame multispectrum and process parameters; performing time synchronization and feature extraction on the multi-source process data to generate a feature vector; inputting the feature vector into a carbon content prediction model to obtain a carbon content prediction value and an uncertainty estimation value thereof; the molten pool temperature is obtained, and based on the molten pool temperature, the waste gas components and the technological parameters, a theoretical carbon content value is calculated through a thermodynamic carbon content calculation model; and carrying out weighted fusion on the carbon content predicted value and the theoretical carbon content value to generate a fused carbon content as a target carbon content. According to the method and system for online determination of the carbon content of the molten steel in converter steelmaking, the precision and reliability of online determination of the carbon content of the molten steel can be effectively improved, and powerful support is provided for intelligent production of converter steelmaking.

Description

Technical Field

[0001] This application relates to the field of converter steelmaking technology, and in particular to a method and system for online determination of carbon content in molten steel during converter steelmaking. Background Technology

[0002] In converter steelmaking, the carbon content of molten steel is a key indicator determining the quality of the steel grade and the final smelting result. Its precise control directly impacts production efficiency, energy consumption, and product qualification rate. Existing online measurement technologies, such as waste gas analysis, are limited in accuracy due to interference from furnace gas circulation; flame spectroscopy is susceptible to factors like smoke and dust contamination of the temperature measurement window, resulting in insufficient stability. These methods struggle to balance accuracy and real-time performance, leading to significant fluctuations in carbon content control during smelting and increasing the risk of over- or under-blowing, which not only increases production costs but also introduces quality risks.

[0003] Therefore, there is an urgent need for a method and system for online determination of carbon content in molten steel during converter steelmaking, in order to improve the accuracy and reliability of online determination of carbon content in molten steel. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a method and system for online determination of carbon content in molten steel during converter steelmaking.

[0005] A first aspect of this application provides a method for online determination of carbon content in molten steel during converter steelmaking, comprising: Acquire multi-source process data, which includes: waste gas composition, furnace flame multispectral data, and process parameters; The multi-source process data is time-synchronized and feature-extracted to generate feature vectors; The feature vector is input into the carbon content prediction model to obtain the predicted carbon content and its uncertainty estimate. The molten pool temperature is obtained, and based on the molten pool temperature, the waste gas composition, and the process parameters, the theoretical carbon content value is calculated using a thermodynamic carbon content calculation model. The predicted carbon content value and the theoretical carbon content value are weighted and fused to generate a fused carbon content as the target carbon content.

[0006] A second aspect of this application provides a system for online determination of carbon content in molten steel during converter steelmaking, comprising: The data acquisition module is used to acquire multi-source process data, which includes: exhaust gas composition, furnace flame multispectral data, and process parameters; The feature extraction module is used to perform time synchronization and feature extraction on the multi-source process data to generate feature vectors; The data prediction module is used to input the feature vector into the carbon content prediction model to obtain the predicted carbon content value and its uncertainty estimate. The theoretical calculation module is used to obtain the molten pool temperature and, based on the molten pool temperature, the waste gas composition, and the process parameters, calculate the theoretical carbon content value through a thermodynamic carbon content calculation model. The data fusion module is used to perform weighted fusion of the predicted carbon content value and the theoretical carbon content value to generate a fused carbon content as the target carbon content.

[0007] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for online determination of carbon content in molten steel during converter steelmaking.

[0008] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the system for online determination of carbon content in molten steel during converter steelmaking described above.

[0009] The beneficial effects of the method and system for online determination of carbon content in molten steel during converter steelmaking provided in this application are as follows: Firstly, this application utilizes multi-dimensional information such as waste gas composition, flame multispectral analysis, and process parameters to overcome the limitations of a single data source, reducing the impact of external interference on the measurement results and making carbon content determination more stable and reliable. Secondly, it achieves rapid online determination through a carbon content prediction model, incorporating theoretical carbon content values ​​calculated by thermodynamics, thus combining the real-time nature of machine learning with the theoretical rigor of thermodynamic models to improve the reliability of the results. Finally, it dynamically weights the predicted and theoretical values ​​based on their characteristics, adaptively adjusting the weights according to operating conditions to reduce errors caused by single model biases, obtaining a more accurate actual carbon content in the molten steel, and providing a reliable basis for smelting endpoint control. This application achieves real-time online carbon content determination, avoiding the lag of manual sampling, reducing over-blowing and under-blowing phenomena, lowering production costs, improving production efficiency, and promoting the intelligent upgrading of converter steelmaking. Attached Figure Description

[0010] Figure 1 This is a schematic flowchart of a method for online determination of carbon content in molten steel during converter steelmaking, provided in an embodiment of this application. Figure 2 This is a structural block diagram of a system for online determination of carbon content in molten steel during converter steelmaking, provided in an embodiment of this application. Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0011] In the following description, 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 have been omitted so as not to obscure the description of this application with unnecessary detail.

[0012] To make the purpose, technical solution, and advantages of this application clearer, the following will be described in conjunction with the appendix. Figure 1-3 The following is an explanation using specific examples.

[0013] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a method for online determination of carbon content in molten steel during converter steelmaking, as provided in an embodiment of this application. The method includes: S101: Acquire multi-source process data, including: exhaust gas composition, furnace flame multispectral data, and process parameters; In this embodiment, the waste gas composition can be collected in real time by a gas analyzer installed on the converter waste gas emission pipe, which can characterize the process of the reaction between molten steel and oxygen and the oxidation of carbon. The waste gas composition includes various gas components and their concentrations generated during the converter steelmaking process, such as carbon monoxide, carbon dioxide, oxygen, and nitrogen. The multispectral data of the furnace mouth flame is obtained by capturing images of the flame at the converter furnace mouth using a multispectral camera. The spectral characteristics of the flame are closely related to the temperature and carbon content of the molten steel; different carbon contents will result in significant differences in the color, brightness, and spectral distribution of the flame. The multispectral data of the furnace mouth flame includes spectral information of the flame in different wavelength ranges, such as the intensity and wavelength distribution characteristics of visible light and near-infrared bands. Process parameters include the basic parameters and operating parameters of the converter. The basic parameters include the nominal capacity and furnace dimensions of the converter; the operating parameters include oxygen flow rate, oxygen supply time, lance height, scrap steel addition amount, initial composition and temperature of molten iron, and type and amount of slagging agent. These process parameters are recorded and transmitted in real time through the converter's control system.

[0014] S102: Perform time synchronization and feature extraction on multi-source process data to generate feature vectors; In this embodiment, due to the different acquisition devices for waste gas components, furnace flame multispectral data, and process parameters, the data acquisition time intervals and start times differ, thus requiring time synchronization. This embodiment uses a key operational moment of the converter, such as the start of oxygen supply, as the time reference, aligning data from different sources along the time axis to ensure that data at the same point in time correspond to each other.

[0015] In terms of feature extraction, this embodiment can calculate features such as the concentration ratio and concentration change rate of different gas components for exhaust gas component data. These features can characterize the rate and extent of carbon oxidation. For multispectral data of furnace flames, characteristic wavelengths, peak spectral intensity, and slope of spectral curves can be extracted using spectral analysis techniques. These characteristics are strongly correlated with the temperature and carbon content of molten steel. For process parameter data, key parameter values ​​can be directly selected as features, such as oxygen flow rate and gun height, or new features can be obtained by calculating the changes and average values ​​of the parameters.

[0016] In this embodiment, the extracted features are combined to form a feature vector for subsequent carbon content prediction, and the influence of dimensions is eliminated through standardization.

[0017] S103: Input the feature vector into the carbon content prediction model to obtain the predicted carbon content and its uncertainty estimate; In this embodiment, the carbon content prediction model can be constructed using machine learning algorithms, such as neural networks, support vector machines, and random forests. During the model training phase, this embodiment utilizes historical multi-source process data and corresponding actual molten steel carbon content data to train the model. By adjusting the model parameters, the model can accurately predict the carbon content of molten steel based on the input feature vector. Simultaneously with obtaining the predicted carbon content, it is also necessary to estimate the uncertainty of the prediction result. Uncertainty estimation characterizes the reliability of the prediction result; methods for uncertainty estimation include model-based variance estimation and Bayesian inference. For example, for neural network models, the uncertainty of the prediction result can be quantified by incorporating Monte Carlo (dropout) or Bayesian neural network techniques during training, or by using ensemble learning methods to estimate the confidence interval of the prediction result, i.e., the uncertainty estimate.

[0018] S104: Obtain the molten pool temperature, and calculate the theoretical carbon content value based on the molten pool temperature, waste gas composition and process parameters using a thermodynamic carbon content calculation model; In this embodiment, the molten pool temperature can be measured in real time using temperature sensors or infrared thermometers installed on the converter. The thermodynamic carbon content calculation model in this embodiment is based on the thermodynamic principles of the steelmaking process. This model is designed based on factors such as the chemical reaction between molten steel and oxygen, and the distribution balance of various elements in the molten steel and slag. Based on the molten pool temperature, waste gas composition, and process parameters, this embodiment uses thermodynamic laws and related chemical reaction equations to calculate the theoretical value of the carbon content in the molten steel. This theoretical value represents the carbon content of the molten steel under ideal thermodynamic equilibrium conditions.

[0019] S105: Weighted fusion of predicted and theoretical carbon content values ​​to generate a fused carbon content as the target carbon content.

[0020] In this embodiment, the data-driven carbon content prediction model and the mechanism-driven thermodynamic carbon content calculation model each have different characteristics and advantages. Fusing the results of the two can improve the accuracy and reliability of carbon content determination.

[0021] In this embodiment, the weights for weighted fusion can be dynamically adjusted based on the prediction accuracy of the carbon content prediction model and the thermodynamic carbon content calculation model. For example, if the historical prediction error of the carbon content prediction model is small while the error of the thermodynamic calculation model is large at a certain steelmaking stage, the predicted carbon content value can be assigned a larger weight; conversely, the theoretical carbon content value can be assigned a larger weight. Secondly, the weights can be determined through analysis of historical data and model evaluation, or an adaptive weight adjustment algorithm can be used to dynamically update the weights based on real-time prediction errors. The predicted carbon content value and the theoretical carbon content value are weighted and summed according to their respective weights to obtain the final target carbon content. This target carbon content can more accurately characterize the actual carbon content of molten steel during converter steelmaking, providing accurate and reliable carbon content measurement results for the control and adjustment of the steelmaking process. As can be seen from the above, this application, based on multi-dimensional information such as exhaust gas composition, flame multispectral analysis, and process parameters, overcomes the limitations of a single data source, reduces the impact of external interference on the measurement results, and makes carbon content measurement more stable and reliable. Secondly, it achieves rapid online measurement through a carbon content prediction model, incorporating theoretical carbon content values ​​calculated by thermodynamics, thus combining the real-time nature of machine learning with the theoretical rigor of thermodynamic models to improve the reliability of the results. Finally, it dynamically weights the predicted and theoretical values ​​based on their characteristics, adaptively adjusting the weights according to operating conditions to reduce errors caused by single model biases, obtaining a more accurate actual carbon content in molten steel, and providing a reliable basis for smelting endpoint control. This application achieves online real-time carbon content measurement, avoiding the lag of manual sampling, reducing over-blowing and under-blowing phenomena, lowering production costs, improving production efficiency, and promoting the intelligent upgrading of converter steelmaking.

[0022] In one embodiment of this application, a weighted fusion of predicted carbon content and theoretical carbon content values ​​is performed to generate a fused carbon content, including: The confidence weighting factor is calculated based on the uncertainty estimate. Based on the current blowing stage and the estimated error values ​​of the molten pool temperature and the estimated error values ​​of the waste gas component measurement, the applicability weighting factor is calculated. The carbon content is generated by weighting and fusing the confidence weight factor, applicability weight factor, predicted carbon content value and theoretical carbon content value. Among them, the confidence weight factor is the weight corresponding to the predicted carbon content value, and the applicability weight factor is the weight corresponding to the theoretical carbon content value.

[0023] In this embodiment, the confidence weight factor is the weight corresponding to the predicted carbon content value. The uncertainty estimate represents the reliability of the output result of the carbon content prediction model. The smaller the uncertainty, the more reliable the prediction result, and the larger the corresponding confidence weight factor. For example, if the uncertainty estimate of the predicted carbon content at a certain moment is low obtained through Bayesian inference, it indicates that the deviation between the predicted value and the actual carbon content is small. In this case, the calculated confidence weight factor will be relatively large to highlight the role of the predicted value in the fusion result; conversely, if the uncertainty estimate is high, the confidence weight factor will be small.

[0024] Specifically, the confidence interval width in the uncertainty estimate is used as the core parameter, and it is converted into a weight value through a preset mapping function. For example, when the 95% confidence interval width is 0, the confidence weight factor is 1; when the confidence interval width increases to 0.1%, the confidence weight factor linearly decreases to 0.3. This mapping function can be adjusted according to the statistical analysis of actual smelting data to ensure that the confidence weight factor can accurately represent the reliability of the carbon content prediction value, and its value range is 0-1.

[0025] In this embodiment, the applicability weighting factor corresponds to the theoretical carbon content value. The reaction characteristics and thermodynamic equilibrium states of the steelmaking process differ at different blowing stages, and the applicability of the theoretical carbon content calculation model varies at different stages. In the early stages of blowing, the carbon content in the molten steel is high, the reaction is more vigorous, and the deviation between the thermodynamic equilibrium assumption and the actual situation is larger; at this time, the applicability weighting factor is relatively small. However, in the later stages of blowing, the reaction gradually stabilizes, the applicability of the thermodynamic model increases, and the applicability weighting factor becomes relatively larger.

[0026] In this embodiment, the molten pool temperature is a crucial parameter in the thermodynamic carbon content calculation model. The larger the measurement error, the lower the reliability of the theoretical carbon content value calculated based on this temperature, and the smaller the applicability weighting factor becomes. Similarly, the larger the measurement error of the waste gas components, the lower the reliability of the theoretical carbon content value, and the smaller the applicability weighting factor becomes. For example, if the accuracy analysis of the temperature sensor shows a large estimated error in the molten pool temperature, and the gas analyzer's measurement data also shows a high estimated error in the waste gas component measurement, it indicates a significant deviation in the theoretical carbon content value. The calculated applicability weighting factor will then be smaller, reducing the weight of this theoretical value in the fusion result.

[0027] For example, the estimated error value of the molten pool temperature is determined based on the accuracy parameters of the molten pool temperature measuring equipment and historical error data. When the estimated error value is less than 5℃, there is no correction to the basic weighting coefficient; when the estimated error value is between 5-10℃, the basic coefficient is multiplied by a correction factor of 0.8; when the estimated error value is greater than 10℃, the basic coefficient is multiplied by a correction factor of 0.5.

[0028] The estimated error values ​​for waste gas component measurement are obtained based on the measurement accuracy of the gas analyzer. For example, the estimated error values ​​for each component in the waste gas are CO, etc. The system is divided into equal components; when the error estimate of each component is less than 1%, there is no correction to the basic weight coefficient; when the error estimate of a component is between 1% and 3%, the basic coefficient is multiplied by a correction factor of 0.9; when the error estimate of a component is greater than 3%, the basic coefficient is multiplied by a correction factor of 0.7.

[0029] The formula for calculating the fused carbon content in this embodiment can be expressed as: Fusion carbon content = predicted carbon content × confidence weighting factor + theoretical carbon content × applicability weighting factor.

[0030] In this embodiment, a weighted fusion is performed based on the uncertainty of the predicted carbon content value and the applicability of the theoretical carbon content value under the current blowing stage and measurement error conditions. This combines the characteristics and reliability of the two carbon content values, making the weighted fusion more scientific and reasonable. It can dynamically adjust the weights of the carbon content prediction model results and the thermodynamic carbon content calculation model results according to the actual situation, so that the fused carbon content is closer to the actual carbon content of molten steel, further improving the accuracy of the fused carbon content and providing a stronger basis for the precise control of the converter steelmaking process.

[0031] In one embodiment of this application, the confidence weighting factor is calculated based on the uncertainty estimate, including: The uncertainty estimate is input into the exponential decay function to obtain the initial confidence weight, wherein the exponential decay function satisfies that the initial confidence weight decreases monotonically as the uncertainty estimate increases. The initial confidence weights are normalized to generate confidence weight factors.

[0032] In this embodiment, the core characteristic of the exponential decay function is that as the uncertainty estimate increases, the initial confidence weight exhibits a monotonically decreasing trend. This perfectly aligns with the logic that the lower the uncertainty, the more reliable the prediction result, and the larger the weight should be. Its functional form can be expressed as: Initial confidence weight = exp(-k × uncertainty estimate), Where k is an adjustment parameter greater than 0, used to control the decay rate of the function.

[0033] In this embodiment, the value of k needs to be determined based on the correlation between the uncertainty estimate and the actual prediction error in historical data. If a small change in the uncertainty estimate is required to cause a significant adjustment in the weight, a larger k value can be selected; if a relatively gradual change in the weight is required, a smaller k value can be selected. For example, when the uncertainty estimate of the carbon content prediction is 0.02% at a certain moment, and k is 50, the initial confidence weight is approximately exp(-50×0.02)=exp(-1)≈0.3679; if the uncertainty estimate increases to 0.05%, the initial confidence weight becomes exp(-50×0.05)=exp(-2.5)≈0.0821, reflecting a significant decrease in the initial weight when the uncertainty increases. The purpose of normalization in this embodiment is to ensure that the weighting factors are within a reasonable numerical range and can form an effective complementary relationship with the applicability weighting factors, so that their sum is 1, thereby ensuring the stability of the fusion result. The normalization calculation formula is as follows: Confidence weight factor = Initial confidence weight / (Initial confidence weight + Initial applicability weight); The initial applicability weight is a weight base calculated based on the correlation coefficient of the theoretical carbon content value. Its calculation logic is similar to that of the initial confidence weight, and it needs to reflect the basic reliability of the theoretical value.

[0034] For example, if the initial confidence weight is 0.3679 and the initial applicability weight is 0.5, then the confidence weight factor = 0.3679 / (0.3679+0.5)≈0.425, and the applicability weight factor is 1-0.425=0.575.

[0035] In this embodiment, the uncertainty estimate is converted into an initial confidence weight using an exponential decay function, which intuitively reflects the relationship between uncertainty and weight: the greater the uncertainty, the lower the weight. Secondly, the confidence weight factor is obtained through normalization, ensuring that the weight factor value is within a reasonable range. This makes the carbon content prediction weights comparable under different conditions and provides an accurate calculation basis for subsequent weighted fusion.

[0036] In one embodiment of this application, an applicability weighting factor is calculated based on the estimated error values ​​of the current blowing stage and molten pool temperature, as well as the estimated error values ​​of the exhaust gas component measurement, including: The blowing process is divided into early stage, middle stage and late stage based on the proportion of cumulative oxygen supply to planned total oxygen supply. If the blowing process is in the early or middle stage, and the estimated value of the waste gas component measurement error is greater than the preset component error threshold or the estimated value of the molten pool temperature error is greater than the preset temperature error threshold, then the applicability weight factor is 0; otherwise, the applicability weight factor is the preset base value. If the blowing process is in the later stage, obtain the error estimate of the molten pool temperature sensor and the error estimate of the waste gas component measurement; The applicability attenuation coefficient is calculated by inputting the estimated values ​​of the molten pool temperature error and the estimated values ​​of the waste gas component measurement error into a preset function. Among them, the preset function satisfies the condition that the attenuation coefficient decreases monotonically as the error estimate increases; Based on the preset basic weights, error decay coefficients, and later-stage enhancement factors, the applicability weight factor is calculated; among them, the later-stage enhancement factor is greater than 1, which is used to reflect the improvement in the applicability of the thermodynamic model in the later stage.

[0037] In this embodiment, the stages of the blowing process are divided based on the ratio of the cumulative oxygen supply to the planned total oxygen supply. When the ratio of the cumulative oxygen supply to the planned total oxygen supply is between 0% and 30%, it is classified as the early stage; when the ratio is between 30% and 70%, it is classified as the middle stage; and when the ratio is between 70% and 100%, it is classified as the late stage. This method of division in this embodiment can better represent the reaction characteristics and thermodynamic state of different stages in the blowing process.

[0038] If the blowing process is in the early or middle stage, the applicability weighting factor needs to be determined based on the estimated error values ​​of the exhaust gas component measurement and the molten pool temperature. When the estimated error value of the exhaust gas component measurement is greater than the preset component error threshold, or the estimated error value of the molten pool temperature is greater than the preset temperature error threshold, it indicates that the data on which the theoretical carbon content calculation is based has low reliability, and the applicability of the thermodynamic model at this stage is poor. Therefore, the applicability weighting factor is set to 0. Conversely, if the estimated error value of the exhaust gas component measurement is less than or equal to the preset component error threshold, and the estimated error value of the molten pool temperature is less than or equal to the preset temperature error threshold, then the applicability weighting factor is the preset base value. The preset base value in this embodiment is set according to the general applicability level of the thermodynamic model in the early and middle stages, for example, it can be set to 0.3. If the blowing process is in its later stages, calculating the applicability weighting factor becomes relatively complex. First, the error estimates of the molten pool temperature sensor and the waste gas component measurement errors must be obtained. Then, these two error estimates are input into a preset function to calculate the applicability attenuation coefficient. The preset function must satisfy the characteristic that the attenuation coefficient monotonically decreases as the error estimate increases. For example, an exponential decay function similar to that used to calculate the initial confidence weight can be used. Specifically: The applicability attenuation coefficient = exp(-m×(estimated value of molten pool temperature error + estimated value of exhaust gas component measurement error)), where m is an adjustment parameter greater than 0, used to control the rate of attenuation. The applicability weight factor is calculated based on the preset base weights, error decay coefficient, and later-stage enhancement factor. The calculation formula is: Applicability weight factor = Preset base weights × Error decay coefficient × Later-stage enhancement factor.

[0039] The preset base weight is a benchmark value for the applicability of the thermodynamic model in the later stage, for example, set to 0.4. The enhancement factor for the later stage is greater than 1, for example, 1.5. Its function is to reflect that as the blowing process enters the later stage, the reaction gradually stabilizes, and the applicability of the thermodynamic model is significantly improved, thereby increasing the weight of the theoretical carbon content value in the fusion result. For example, if the preset base weight is 0.4, the error attenuation coefficient is calculated to be 0.8, and the enhancement factor for the later stage is 1.5, then the applicability weight factor = 0.4 × 0.8 × 1.5 = 0.48.

[0040] In this embodiment, the blowing stages are divided according to the cumulative oxygen supply ratio, and different applicability weighting factor calculation rules are set for different stages, fully considering the characteristics of different stages in the blowing process and the impact of measurement errors. In the early and middle stages, when the measurement error is large, the applicability weighting factor is directly set to 0 to avoid the influence of unreliable theoretical carbon content values ​​on the results. In the later stage, by introducing a later-stage enhancement factor, the applicability of the later-stage thermodynamic model is improved, making the weight calculation more consistent with the actual situation.

[0041] In one embodiment of this application, the method for online determination of carbon content in molten steel during converter steelmaking further includes: At the preset measurement time of the secondary gun, the measured value of the carbon content of the secondary gun is obtained; Calculate the relative deviation between the predicted carbon content and the measured carbon content of the secondary gun at the time of measurement. The calibration method for online calibration of the carbon content prediction model is selected based on the relative deviation. The carbon content prediction model is calibrated online based on the selected calibration method and calibration objective to obtain the calibrated carbon content prediction model; the calibration objective is to minimize the deviation between the predicted carbon content value and the measured carbon content value of the secondary gun. The calibrated carbon content prediction model will be used as the carbon content prediction model for subsequent predictions.

[0042] The calculation process for relative deviation includes: Calculate the absolute difference between the predicted carbon content at the time of measurement of the secondary gun and the measured carbon content of the secondary gun; The relative deviation is the ratio of the absolute difference to the measured value of the carbon content in the secondary gun.

[0043] In this embodiment, secondary lance measurement is a commonly used method in converter steelmaking for directly determining the composition of molten steel, and it has high accuracy. The preset timing of secondary lance measurement is usually determined based on the characteristics and experience of the steelmaking process, such as taking measurements at key points or near the end of the blowing process to obtain data on the true carbon content of the molten steel.

[0044] In this embodiment, the calculation of the relative deviation between the predicted carbon content value and the measured carbon content value at the time of secondary gun measurement is divided into two steps. The first step is to calculate the absolute difference between the predicted carbon content value and the measured carbon content value at the time of secondary gun measurement. That is, the absolute difference = |predicted carbon content - measured carbon content of the secondary gun|.

[0045] The second step is to calculate the relative deviation, which is the ratio of the absolute difference to the measured value of the carbon content in the secondary gun. The relative deviation is calculated as: relative deviation = absolute difference / measured carbon content of the secondary gun. The relative deviation more intuitively represents the degree of deviation between the predicted and measured values, is not affected by the absolute value of the carbon content, and facilitates comparison of deviations at different measurement times. In this embodiment, the calibration method for online calibration of the carbon content prediction model is selected based on the relative deviation. If the relative deviation is small, i.e., less than the first deviation threshold (e.g., less than 1%), it indicates that the model's current prediction accuracy is high, and a light calibration method can be selected, which only fine-tunes the output layer parameters of the model. If the relative deviation is large, i.e., greater than the first deviation threshold and less than or equal to the second deviation threshold (e.g., between 1% and 3%), it indicates that the model has a certain degree of drift, and a medium calibration method is required to adjust the intermediate and output layer parameters of the model. If the relative deviation is extremely large, i.e., greater than the second deviation threshold (e.g., greater than 3%), it means that the model has a large deviation or is not suitable for the current operating conditions, and a heavy calibration method is required, such as retraining the model with new training samples or adjusting the model's structural parameters. The calibration objective of this embodiment is to minimize the deviation between the predicted carbon content and the measured carbon content of the secondary lance. In light calibration, optimization algorithms such as gradient descent can be used to fine-tune the output layer parameters with the goal of reducing the deviation. In medium calibration, in addition to adjusting the output layer parameters, the weights and biases of the intermediate layers also need to be optimized. In heavy calibration, the multi-source process data and corresponding measured values ​​at the secondary lance measurement time are added to the model's training set as new training samples, and the model is retrained using methods such as mini-batch gradient descent to adapt it to the current steelmaking conditions, thereby reducing the prediction deviation. After calibration, the multi-source process data before and after the secondary lance measurement time are used for verification. If the relative deviation between the calibrated model's prediction deviation and the measured value of the secondary lance on the verification data is less than a first deviation threshold, the calibration is considered valid.

[0046] In conclusion, this embodiment concludes by using the calibrated carbon content prediction model as the subsequent prediction model. By employing the measured carbon content of the auxiliary lance for online calibration, this embodiment allows for timely correction of prediction deviations caused by changes in operating conditions and equipment aging during long-term use, based on actual measurement results. This ensures the model maintains high prediction accuracy, thereby improving the accuracy and reliability of online steel carbon content measurement. This enables the model to better adapt to various changes in the converter steelmaking process, providing strong support for precise control in converter steelmaking.

[0047] In one embodiment of this application, the calibration method includes: a parameter update mode and a state reset mode; the calibration method for online calibration of the carbon content prediction model based on relative deviation includes: If the blowing process is in the early or middle stage, and the average relative deviation at the time of measurement of the secondary gun is greater than the first deviation threshold for N consecutive times, and the deviation direction consistency rate is greater than the preset direction threshold, then the parameter update mode is selected. If the blowing process is in the later stage and the single relative deviation is greater than the second deviation threshold, the state reset mode is selected first, or the parameter update mode and the state reset mode are executed simultaneously. If the blowing process is at any stage and the single relative deviation is greater than the third deviation threshold, then the state reset mode is selected. The relative deviation is the difference between the relative deviation and the measured value of the carbon content of the secondary gun. The third deviation threshold is greater than the second deviation threshold, and the second deviation threshold is greater than the first deviation threshold.

[0048] In this embodiment, the parameter update mode mainly adjusts and optimizes the existing parameters of the carbon content prediction model without changing the model's structure and core framework, reducing prediction bias by fine-tuning the parameters. The state reset mode, on the other hand, restores the model to a preset initial state or a validated baseline state, essentially restarting the model, and is used when the model exhibits significant deviations or anomalies. In this embodiment, if the blowing process is in the early or middle stage, and the average relative deviation at the secondary gun measurement time is greater than the first deviation threshold for N consecutive times, while the deviation direction consistency rate is greater than the preset direction threshold, then the parameter update mode is selected. This situation indicates that the model has experienced continuous and regular deviations in the early or middle stage. By adjusting the model parameters in a targeted manner through the parameter update mode, the deviations can be effectively corrected without requiring significant modifications to the model, which is beneficial for maintaining the model's stability.

[0049] Where N is the preset number of times, such as 3 times; the first deviation threshold is set according to the allowable deviation range of the model in the early and middle stages, such as 2%; the deviation direction consistency rate refers to the proportion of deviation directions that remain consistent in N consecutive relative deviations, and the preset direction threshold can be set to 80%.

[0050] In this embodiment, the later stage is a critical stage in the steelmaking process, requiring higher accuracy in carbon content prediction. Therefore, the second deviation threshold is set more strictly than the first. When a single relative deviation exceeds this threshold, it indicates a significant anomaly in the model at this critical stage. The state reset mode can be prioritized to quickly restore the model to a reliable state, ensuring the accuracy of subsequent predictions. If the situation is more complex, a parameter update mode can be combined simultaneously to further optimize the parameters based on the reset. In this embodiment, if the blowing process is at any stage and a single relative deviation exceeds the third deviation threshold, the state reset mode is selected. The third deviation threshold is the most stringent of the three, and its value is greater than the second deviation threshold, which in turn is greater than the first deviation threshold. When a single relative deviation exceeds the third deviation threshold, it indicates that the model has a serious deviation, caused by sudden changes in operating conditions, data anomalies, or significant problems with the model itself. In this case, the state reset mode must be used to restore the model to its initial or baseline state to avoid erroneous predictions from severely impacting the steelmaking process.

[0051] In this embodiment, different calibration methods are selected based on the different stages of the blowing process and the varying relative deviations, demonstrating strong targeting and flexibility. In the early and middle stages, when large, consecutive deviations occur in the same direction, a parameter update mode is used to gradually adjust the model parameters. In the later stages, when a single deviation is large, a state reset mode is prioritized to quickly bring the model back to the correct prediction trajectory. At any stage, when the deviation is extremely large, the state reset mode is directly selected to ensure the model can correct errors promptly. This embodiment can intervene promptly when the model exhibits serious anomalies, ensuring the reliability of online measurements.

[0052] In one embodiment of this application, the parameter update mode includes: The correction amount of the weight matrix is ​​determined based on the product of the Kalman gain matrix and the prediction residual; The weight matrix of the carbon content prediction model is updated based on recursive least squares method and correction. The state reset mode includes resetting the hidden state of the recurrent neural network in the carbon content prediction model to the initial state calculated based on the current input feature vector and the measured value of the secondary gun carbon content.

[0053] In this embodiment, for the parameter update mode, the correction amount of the weight matrix is ​​first determined based on the product of the Kalman gain matrix and the prediction residual. The Kalman gain matrix dynamically adjusts the gain according to the statistical characteristics of the prediction error, making the correction more consistent with the actual situation; the prediction residual is the difference between the predicted carbon content and the measured carbon content of the secondary lance, representing the degree of deviation in the model prediction. The correction amount obtained by multiplying these two values ​​accurately indicates the magnitude and direction of the weight matrix adjustment. Secondly, the weight matrix of the carbon content prediction model is updated based on the recursive least squares method and the aforementioned correction amount. The recursive least squares method can update the model parameters in real time as new data is continuously acquired, allowing the weight matrix to better adapt to changes in the actual steelmaking process, thereby improving the model's prediction accuracy. For the state reset mode, the current input feature vector is extracted. This previous input feature vector includes information extracted from multi-source process data such as the current exhaust gas composition, furnace flame multispectral data, and process parameters. Based on the measured carbon content of the secondary lance, the initial hidden state of the recurrent neural network is calculated using a preset initialization algorithm. The initialization algorithm can use a mapping relationship trained on historical data to map the input feature vector and the measured carbon content of the secondary lance to appropriate hidden state values, ensuring that the initial hidden state can represent the current smelting conditions. The hidden state of the recurrent neural network is reset to the initial state calculated above, thereby eliminating the influence of previously accumulated errors and allowing the model to make subsequent predictions based on the accurate initial state, quickly restoring the model's predictive performance.

[0054] Specifically, the hidden state of the recurrent neural network in the carbon content prediction model is reset to an initial state calculated based on the current input feature vector and the measured carbon content of the secondary gun. The hidden state of the recurrent neural network reflects past input information. When the model exhibits significant deviations, resetting the hidden state eliminates the influence of past errors. By recalculating the initial state based on the current accurate input feature vector and the measured carbon content of the secondary gun, the model can make predictions from a new starting point, improving the accuracy of subsequent predictions.

[0055] In the parameter update mode of this embodiment, the correction amount of the weight matrix is ​​determined based on the product of the Kalman gain matrix and the prediction residual, and then updated using the recursive least squares method. This method can effectively utilize new measurement data to optimize the model, improving the model's adaptability and accuracy. In the state reset mode of this embodiment, the hidden state of the recurrent neural network is reset to the initial state calculated based on the current input feature vector and the measured value of the secondary gun carbon content. This can quickly eliminate the accumulated erroneous information in the model, allowing the model to return to an accurate prediction state.

[0056] In one embodiment of this application, the method for online determination of carbon content in molten steel during converter steelmaking further includes: Acquire laser-induced breakdown spectral data and extract the intensity values ​​of carbon characteristic peaks; The laser carbon content value is calculated based on the intensity value and the molten pool temperature using a calibration curve. The quality of the laser-induced breakdown spectral data was assessed, and the quality assessment results were obtained. Based on the stage of the blowing process and the quality assessment results, the first weight corresponding to the laser carbon content value is determined. The target carbon content is obtained by weighting the first weight, the laser carbon content value, and the fused carbon content.

[0057] In this embodiment, laser-induced breakdown spectroscopy (LAS) uses a laser to excite molten steel to generate plasma. The elemental composition of the molten steel can be analyzed based on the spectrum emitted by the plasma. Carbon will form a characteristic peak at a specific wavelength. Extracting the intensity value of this characteristic peak can indicate the carbon content level in the molten steel.

[0058] This embodiment calculates the laser carbon content value based on the extracted intensity value and molten pool temperature using a calibration curve. The calibration curve is plotted using standard samples with known carbon content and represents the relationship between the carbon characteristic peak intensity value and the carbon content. A function is fitted with carbon content on the x-axis and the carbon characteristic peak intensity value on the y-axis. Since the molten pool temperature affects the measurement results of the laser-induced breakdown spectrum, the intensity value needs to be corrected for the molten pool temperature during the calculation process to improve the accuracy of the laser carbon content value. The quality assessment in this embodiment includes indicators such as the signal-to-noise ratio of the spectrum, the clarity of characteristic peaks, and the stability of the spectrum. If the signal-to-noise ratio of the spectrum is high, the characteristic peaks are clear, and the stability is good, it indicates that the data quality is high and the corresponding laser carbon content value is highly reliable; conversely, the data quality is low and the reliability of the laser carbon content value is poor. In this embodiment, due to the large changes in the composition of molten steel during the early and middle stages of blowing, the laser-induced breakdown spectrum measurement will be subject to more interference. If the data quality is high, the first weight can be set to a lower value, such as 0.1-0.2; if the data quality is low, the first weight should be even lower, such as 0-0.05.

[0059] In the later stages of blowing, the composition of molten steel is relatively stable, which improves the accuracy of laser-induced breakdown spectroscopy measurement. If the data quality is high, the first weight can be appropriately increased, for example, by 0.2-0.3; if the data quality is low, the first weight should remain at a low level. In this embodiment, the target carbon content is obtained by weighted calculation based on the first weight, the laser carbon content value, and the fused carbon content. The calculation formula is as follows: Target carbon content = laser carbon content value × first weight + fusion carbon content × (1 - first weight).

[0060] In this embodiment, laser-induced breakdown spectral data is newly incorporated. By extracting the intensity values ​​of the carbon characteristic peaks and combining them with the molten pool temperature to calculate the laser carbon content value, a new method for determining the carbon content of molten steel is provided. Secondly, the quality of the laser-induced breakdown spectral data is assessed, and its weight is determined based on the blowing stage and the quality assessment results. This weighted calculation, combined with the fused carbon content, fully utilizes the advantages of laser measurement and further improves the accuracy of carbon content determination.

[0061] In one embodiment of this application, the method for online determination of carbon content in molten steel during converter steelmaking, after performing a quality assessment on laser-induced breakdown spectral data and obtaining the quality assessment result, further includes: determining whether to discard the laser-induced breakdown spectral data based on the assessment result; wherein, Calculate the fluctuation coefficient of the carbon characteristic peak intensity. If the fluctuation coefficient is greater than the preset fluctuation threshold, it is determined that it is affected by splashing interference. Calculate the signal-to-noise ratio (SNR) of the characteristic peak to the background noise. If the SNR is less than the preset SNR threshold, the signal is deemed invalid. If there is no interference from splashes and the signal is valid, then A quality score is calculated based on the quality assessment results. The quality score ranges from 0 to 1, with higher quality scores having larger values. The stage factor is determined based on the refining process. The stage factor for the early and middle stages is a first fixed value, and the stage factor for the late stage is a second fixed value, which is greater than the first fixed value. The first weight corresponding to the laser carbon content value is obtained by weighting the stage factor and the mass fraction.

[0062] In this embodiment, the evaluation results are first used to determine whether to discard the laser-induced breakdown spectrum data. The specific determination rules are as follows: The fluctuation coefficient of the carbon characteristic peak intensity is calculated. The fluctuation coefficient is the ratio of the standard deviation to the mean of the carbon characteristic peak intensity within a certain time window; this fluctuation coefficient indicates the degree of change in the carbon characteristic peak intensity over a certain period. If the fluctuation coefficient is greater than a preset fluctuation threshold, it indicates that the laser-induced breakdown spectrum data has been interfered with by molten steel splashing. Splashing causes instability in the spectral signal, and in this case, the data is determined to be affected by splashing interference and should be discarded. The preset fluctuation threshold is based on experimental calibration.

[0063] The signal-to-noise ratio (SNR) of the characteristic peak to the background noise is calculated. The SNR is the ratio of the peak intensity of the characteristic peak to the average intensity of the background noise near the characteristic peak, and it is a crucial indicator of signal quality. If the SNR is less than a preset threshold, it indicates that the effective signal in the spectrum is overwhelmed by noise, making it impossible to accurately extract carbon characteristic peak information. In this case, the signal is deemed invalid, and the data must be discarded. The SNR threshold is experimentally calibrated.

[0064] Secondly, if the laser-induced breakdown spectral data is not affected by sputtering and the signal is valid, then the first weight is calculated, and the specific steps are as follows: A quality score is calculated based on the quality assessment results. The quality score ranges from 0 to 1, with higher quality scores resulting in larger values. Factors influencing the quality score include spectral signal-to-noise ratio, characteristic peak sharpness, and spectral stability. For example, a high spectral signal-to-noise ratio, sharp characteristic peaks, and good stability will result in a quality score close to 1; conversely, a lower quality score will occur.

[0065] For example, quality assessment is performed on laser-induced breakdown spectral data. Assessment indicators include the signal-to-noise ratio of the spectral signal, the clarity of carbon characteristic peaks, and spectral stability. Based on these indicators, the quality assessment results are categorized into four levels: excellent, good, average, and poor. The specific calculation method for the quality score is as follows: excellent, 0.9; good, 0.7; average, 0.5; and poor, 0.3.

[0066] The stage factor is determined based on the stage of the blowing process. The stage factor for the early and middle stages is a first fixed value, while the stage factor for the later stage is a second fixed value, with the second fixed value being greater than the first. This is because the steel composition is relatively stable in the later stage, leading to higher accuracy in laser-induced breakdown spectroscopy measurements; therefore, a higher stage factor is assigned to reflect its importance at this stage. For example, the first fixed value could be set to 0.3, and the second fixed value could be set to 0.7. Finally, the first weight corresponding to the laser carbon content value is calculated based on the stage factor and the mass fraction. The calculation formula can be: First weight = stage factor × mass fraction.

[0067] For example, if a certain moment is in the later stage, the stage factor is 0.7, and the quality score calculated by the quality assessment is 0.7, then the first weight = 0.7 × 0.7 = 0.49; if it is in the early stage, the stage factor is 0.3, and the quality score is 0.7, then the first weight = 0.3 × 0.7 = 0.21.

[0068] The method for quality assessment of laser-induced breakdown spectral data in this embodiment includes determining whether there is sputtering interference and whether the signal is valid, as well as calculating the quality score and stage factor based on the quality assessment results, and then determining the first weight. This quality assessment and weight determination method ensures that only high-quality laser carbon content values ​​are included in the final calculation, improving the reliability of the entire measurement method.

[0069] In one embodiment of this application, the method for online determination of carbon content in molten steel during converter steelmaking further includes: when the blowing process is in the later stage and the predicted carbon content is lower than a preset threshold, adding a collaborative calibration mechanism between the acoustic emission signal of the molten pool and the temperature field, specifically including: Acoustic emission sensor arrays arranged at the bottom of the furnace are used to collect acoustic signals of molten pool bubble bursting, and the energy ratio and signal entropy value of preset frequency bands are extracted. The radial temperature gradient and axial temperature fluctuation variance of the molten pool were constructed based on the furnace wall thermocouple assembly. The energy percentage, signal entropy, radial temperature gradient, and axial temperature fluctuation variance are input into a pre-trained carbon activity calibration model to obtain the calibration factor. The fused carbon content is corrected based on the calibration factor to obtain the corrected carbon content; When the calibration factor is between the preset lower limit and the preset upper limit, the corrected carbon content is used as the target carbon content. When the calibration factor is less than the preset lower limit or greater than the preset upper limit, laser carbon determination of the furnace mouth splash is initiated. If a valid laser carbon determination result is obtained, that result will be used; otherwise, secondary gun carbon determination will be triggered and the measured value of the secondary gun will be used.

[0070] In this embodiment, firstly, acoustic emission sensor arrays arranged at the furnace bottom are used to collect acoustic signals from the bursting of bubbles in the molten pool, and the energy proportion and signal entropy value of a preset frequency band are extracted. The bursting of bubbles in the molten pool generates acoustic signals at specific frequencies. The size, number, and bursting intensity of bubbles vary with different carbon contents, resulting in different frequency band energy distributions and complexities of the acoustic signals. The preset frequency band is typically selected from a range of frequencies related to the characteristics of bubble bursting. The energy proportion indicates the contribution of that frequency band to the total signal, while the signal entropy value reflects the disorder and complexity of the signal; a smaller entropy value indicates a more regular signal, and vice versa.

[0071] Secondly, the radial temperature gradient and axial temperature fluctuation variance of the molten pool are constructed based on the thermocouple array on the furnace wall. Thermocouples arranged at different positions on the furnace wall can measure the temperature at different radial and axial locations in real time. The radial temperature gradient refers to the rate of temperature change along the radius of the molten pool, showing the uniformity of the radial temperature distribution. The axial temperature fluctuation variance refers to the dispersion of temperature change over time along the height of the molten pool, reflecting the stability of the axial temperature. When the carbon content is low, the temperature distribution and fluctuation characteristics of the molten pool will change accordingly. Then, the extracted energy percentage, signal entropy, radial temperature gradient, and axial temperature fluctuation variance are input into a pre-trained carbon activity calibration model to obtain a calibration factor. The carbon activity calibration model is trained using a large amount of historical data and establishes a mapping relationship between the input parameters and the carbon content calibration amount. The magnitude of the calibration factor indicates the magnitude and direction of the correction needed to the fused carbon content. Next, the fused carbon content is corrected based on the calibration factor to obtain the corrected carbon content. The correction formula can be expressed as: Corrected carbon content = fused carbon content × calibration factor. When the calibration factor is between the preset lower limit and the preset upper limit (for example, the preset lower limit is 0.95 and the preset upper limit is 1.05), it indicates that the calibration factor is within a reasonable range and the corrected carbon content has high reliability. In this case, the corrected carbon content is used as the target carbon content. When the calibration factor is less than the preset lower limit or greater than the preset upper limit, it indicates that the current calibration result will have a large error, and laser carbon determination of the molten steel splashed from the furnace mouth needs to be initiated. Laser carbon determination obtains carbon content information by performing laser-induced breakdown spectral analysis on the molten steel particles splashed from the furnace mouth. If a valid laser carbon determination result is obtained, i.e., the spectral quality meets the requirements and the carbon characteristic peaks are clearly distinguishable, then this result is used as the target carbon content; if the laser carbon determination result is invalid, then secondary lance carbon determination is triggered, and the carbon content of the molten steel is directly measured through the secondary lance, and the measured value of the secondary lance is used as the target carbon content. This collaborative calibration mechanism can further improve the accuracy and reliability of carbon content determination during the critical stage of low carbon content in the later stages of blowing, ensuring precise control of the steelmaking endpoint.

[0072] In this embodiment, when the predicted carbon content is lower than a preset threshold during the later stages of blowing, a collaborative calibration mechanism between the acoustic emission signal of the molten pool and the temperature field is added. By collecting the acoustic signal of bubble bursting in the molten pool and constructing information such as the temperature gradient of the molten pool, a calibration factor is obtained using a carbon activity calibration model to correct the fused carbon content. When the calibration factor is not within a reasonable range, laser carbon determination using furnace mouth splashes or secondary lance carbon determination is further initiated. This multi-level calibration and supplementary measurement mechanism ensures the accuracy of carbon content measurement during the critical later stages of blowing, providing a strong guarantee for accurate control of molten steel quality.

[0073] In one embodiment of this application, a method for online determination of carbon content in molten steel during converter steelmaking further includes: when the blowing process is in the later stage and the predicted carbon content is lower than a preset threshold... Based on the correlation between the intensity attenuation rate of the multispectral flame at the furnace mouth and historical smelting data, a spectral reliability coefficient is generated. The thermodynamic model deviation coefficient is calculated based on the degree to which the ratio of carbon monoxide to carbon dioxide in the exhaust gas deviates from the thermodynamic equilibrium value. The confidence weighting factor is corrected to be the product of the spectral reliability coefficient and the original confidence weighting factor; The applicability weighting factor is corrected to be the product of the thermodynamic model deviation coefficient and the original applicability weighting factor.

[0074] In this embodiment, First, a spectral reliability coefficient is generated based on the correlation between the intensity attenuation rate of the furnace mouth flame multispectral data and historical smelting data. The intensity of the furnace mouth flame multispectral data decreases as the blowing process progresses, and the intensity attenuation rate varies under different carbon contents and smelting conditions. By comparing the intensity attenuation rate of the current flame multispectral data with the attenuation rate under the same conditions in historical smelting data, the correlation coefficient between the two is calculated, such as the Pearson correlation coefficient. This coefficient can be used as the spectral reliability coefficient. The higher the correlation, the stronger the consistency between the current spectral data and historical reliable data, and the closer the spectral reliability coefficient is to 1; conversely, the lower the correlation, the smaller the spectral reliability coefficient, indicating that the current spectral data may be abnormal or severely interfered with. Secondly, the thermodynamic model deviation coefficient is calculated based on the degree to which the ratio of carbon monoxide to carbon dioxide in the exhaust gas deviates from the thermodynamic equilibrium value. Under ideal thermodynamic equilibrium, the ratio of carbon monoxide to carbon dioxide in the exhaust gas has a definite theoretical value, which can be calculated using thermodynamic formulas based on parameters such as the molten pool temperature. If the actual measured ratio of carbon monoxide to carbon dioxide deviates from this theoretical equilibrium value, it indicates a discrepancy between the assumptions of the thermodynamic model and the actual situation; the greater the deviation, the more significant the deviation. The thermodynamic model deviation coefficient can be calculated as follows: when the deviation rate between the actual ratio and the equilibrium value is less than or equal to a preset deviation threshold, the deviation coefficient is 1; when the deviation rate is greater than the preset deviation threshold, the deviation coefficient decreases linearly with the increase of the deviation rate, and the minimum deviation coefficient is not lower than 0. This method quantifies the degree of deviation between the thermodynamic model and the actual situation. Then, the confidence weighting factor is corrected to be the product of the spectral reliability coefficient and the original confidence weighting factor. The original confidence weighting factor is mainly determined based on the uncertainty estimate of the carbon content prediction model, while the corrected confidence weighting factor further incorporates the reliability of the furnace flame multispectral data. When the spectral reliability coefficient is high, the corrected confidence weighting factor is closer to the original weight, indicating that the spectral data supports the prediction model's results; when the spectral reliability coefficient is low, the corrected confidence weighting factor will decrease accordingly, reducing the weight of the prediction model's results in the fusion, thus representing the impact of the unreliability of the spectral data on the carbon content prediction model.

[0075] Finally, the applicability weighting factor is corrected to be the product of the thermodynamic model deviation coefficient and the original applicability weighting factor. The original applicability weighting factor represents the basic applicability of the thermodynamic model at the current stage, while the corrected applicability weighting factor incorporates the influence of the deviation of the exhaust gas component ratio from the thermodynamic equilibrium value. When the thermodynamic model deviation coefficient is 1, the corrected applicability weighting factor is consistent with the original weight, indicating that the exhaust gas component data conforms to the thermodynamic equilibrium assumption and the thermodynamic model has high reliability. When the thermodynamic model deviation coefficient is less than 1, the corrected applicability weighting factor decreases, reflecting the decreased reliability of the thermodynamic model due to actual deviations from the equilibrium state, thereby reducing the weight of the theoretical carbon content value in the fusion process. This embodiment employs a correction method. It generates a spectral reliability coefficient based on the correlation between the intensity attenuation rate of the multispectral flame at the furnace mouth and historical smelting data. It calculates the thermodynamic model deviation coefficient based on the degree to which the ratio of carbon monoxide to carbon dioxide in the exhaust gas deviates from the thermodynamic equilibrium value. Then, it corrects the confidence weight factor and applicability weight factor. This correction method dynamically adjusts the weights of the two model results based on the actual spectral and exhaust gas composition conditions. This makes the confidence weight factor and applicability weight factor more closely reflect the reliability of actual smelting data, further improving the accuracy and reliability of carbon content determination. This, in turn, improves the accuracy of fused carbon content, making it more adaptable to the complex converter steelmaking process, especially in the critical stage of low carbon content during the later stages of blowing, providing a more reliable basis for accurately controlling the steelmaking endpoint.

[0076] Corresponding to the method for online determination of carbon content in molten steel during converter steelmaking in the above embodiment, Figure 2 This is a structural block diagram of a system for online determination of carbon content in molten steel during converter steelmaking, provided as an embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 2 The system 20 for online determination of carbon content in molten steel during converter steelmaking includes: a data acquisition module 21, a feature extraction module 22, a data prediction module 23, a theoretical calculation module 24, and a data fusion module 25.

[0077] Among them, the data acquisition module 21 is used to acquire multi-source process data, which includes: exhaust gas composition, furnace flame multispectral data and process parameters; Feature extraction module 22 is used to perform time synchronization and feature extraction on multi-source process data to generate feature vectors; Data prediction module 23 is used to input feature vectors into the carbon content prediction model to obtain the predicted carbon content value and its uncertainty estimate. The theoretical calculation module 24 is used to obtain the molten pool temperature and, based on the molten pool temperature, waste gas composition and process parameters, calculate the theoretical carbon content value through a thermodynamic carbon content calculation model. The data fusion module 25 is used to perform weighted fusion of the predicted carbon content value and the theoretical carbon content value to generate the fused carbon content as the target carbon content.

[0078] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of the modules in the aforementioned device embodiments, for example... Figure 2 The functions of the data acquisition module 21, feature extraction module 22, data prediction module 23, theoretical calculation module 24, and data fusion module 25 are shown.

[0079] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0080] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0081] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store device type information.

[0082] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation methods described in any embodiment of the method for online determination of carbon content in molten steel in converter steelmaking provided in the embodiments of this application, or they can execute the implementation methods of the electronic devices described in the embodiments of this application, which will not be repeated here.

[0083] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0084] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0085] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0086] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0087] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or units, or it may be an electrical, mechanical, or other form of connection.

[0088] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0089] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0090] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for online determination of carbon content in molten steel during converter steelmaking, characterized in that, include: Acquire multi-source process data, which includes: waste gas composition, furnace flame multispectral data, and process parameters; The multi-source process data is time-synchronized and feature-extracted to generate feature vectors; The feature vector is input into the carbon content prediction model to obtain the predicted carbon content and its uncertainty estimate. The molten pool temperature is obtained, and based on the molten pool temperature, the waste gas composition, and the process parameters, the theoretical carbon content value is calculated using a thermodynamic carbon content calculation model. The predicted carbon content value and the theoretical carbon content value are weighted and fused to generate a fused carbon content as the target carbon content.

2. The method for online determination of carbon content in molten steel during converter steelmaking according to claim 1, characterized in that, The step of weightedly fusing the predicted carbon content value and the theoretical carbon content value to generate a fused carbon content includes: The confidence weighting factor is calculated based on the aforementioned uncertainty estimate. Based on the current blowing stage and the estimated error values ​​of the molten pool temperature and the estimated error values ​​of the waste gas component measurement, the applicability weighting factor is calculated; The weighted fusion of the confidence weight factor, the applicability weight factor, the predicted carbon content value and the theoretical carbon content value is used to generate the fused carbon content. The confidence weight factor is the weight corresponding to the predicted carbon content value, and the applicability weight factor is the weight corresponding to the theoretical carbon content value.

3. The method for online determination of carbon content in molten steel during converter steelmaking according to claim 2, characterized in that, The confidence weighting factor calculated based on the uncertainty estimate includes: The uncertainty estimate is input into an exponential decay function to obtain the initial confidence weight, wherein the exponential decay function satisfies that the initial confidence weight decreases monotonically as the uncertainty estimate increases. The initial confidence weights are normalized to generate the confidence weight factors.

4. The method for online determination of carbon content in molten steel during converter steelmaking according to claim 3, characterized in that, The applicability weighting factor, calculated based on the current blowing stage, the estimated error value of the molten pool temperature, and the estimated error value of the waste gas component measurement, includes: The blowing process is divided into early stage, middle stage and late stage based on the proportion of cumulative oxygen supply to planned total oxygen supply. If the blowing process is in the early or middle stage, and the estimated value of the waste gas component measurement error is greater than the preset component error threshold or the estimated value of the molten pool temperature error is greater than the preset temperature error threshold, then the applicability weight factor is 0; otherwise, the applicability weight factor is the preset base value. If the blowing process is in the later stage, obtain the error estimate of the molten pool temperature sensor and the error estimate of the waste gas component measurement; The applicability attenuation coefficient is calculated by inputting the estimated error values ​​of the molten pool temperature and the measurement error values ​​of the exhaust gas components into a preset function; wherein, the preset function satisfies the condition that the attenuation coefficient monotonically decreases as the estimated error value increases; The applicability weight factor is calculated based on the preset basic weights and error attenuation coefficient.

5. The method for online determination of carbon content in molten steel during converter steelmaking according to claim 1, characterized in that, Also includes: At the preset measurement time of the secondary gun, the measured value of the carbon content of the secondary gun is obtained; Calculate the relative deviation between the predicted carbon content value and the measured carbon content value of the secondary gun at the measurement time; The calibration method for online calibration of the carbon content prediction model is selected based on the relative deviation. The carbon content prediction model is calibrated online based on the selected calibration method and calibration objective to obtain the calibrated carbon content prediction model; wherein, the calibration objective is to minimize the deviation between the predicted carbon content value and the measured carbon content value of the secondary gun. The calibrated carbon content prediction model will be used as the carbon content prediction model for subsequent predictions.

6. The method for online determination of carbon content in molten steel during converter steelmaking according to claim 5, characterized in that, The calibration methods include: parameter update mode and state reset mode; the calibration method for selecting online calibration of the carbon content prediction model based on the relative deviation includes: If the blowing process is in the early or middle stage, and the average relative deviation at the time of measurement of the secondary gun is greater than the first deviation threshold for N consecutive times, and the deviation direction consistency rate is greater than the preset direction threshold, then the parameter update mode is selected. If the blowing process is in the later stage and the single relative deviation is greater than the second deviation threshold, the state reset mode is selected first, or the parameter update mode and the state reset mode are executed simultaneously. If the blowing process is at any stage and the single relative deviation is greater than the third deviation threshold, then the state reset mode is selected. Wherein, the relative deviation is the difference between the relative deviation and the measured value of the carbon content of the sub-gun; The third deviation threshold is greater than the second deviation threshold, and the second deviation threshold is greater than the first deviation threshold.

7. The method for online determination of carbon content in molten steel during converter steelmaking according to claim 6, characterized in that, The parameter update modes include: The correction amount of the weight matrix is ​​determined based on the product of the Kalman gain matrix and the prediction residual; The weight matrix of the carbon content prediction model is updated based on the recursive least squares method and the correction amount. The state reset mode includes: resetting the hidden state of the recurrent neural network in the carbon content prediction model to the initial state calculated based on the current input feature vector and the measured value of the secondary gun carbon content.

8. A system for online determination of carbon content in molten steel during converter steelmaking, characterized in that, include: The data acquisition module is used to acquire multi-source process data, which includes: exhaust gas composition, furnace flame multispectral data, and process parameters; The feature extraction module is used to perform time synchronization and feature extraction on the multi-source process data to generate feature vectors; The data prediction module is used to input the feature vector into the carbon content prediction model to obtain the predicted carbon content value and its uncertainty estimate. The theoretical calculation module is used to obtain the molten pool temperature and, based on the molten pool temperature, the waste gas composition, and the process parameters, calculate the theoretical carbon content value through a thermodynamic carbon content calculation model. The data fusion module is used to perform weighted fusion of the predicted carbon content value and the theoretical carbon content value to generate a fused carbon content as the target carbon content.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images