Converter endpoint slag composition prediction and converter protection guidance method and system

Abstract

The application discloses a converter endpoint slag composition prediction and converter protection guiding method and system, and belongs to the technical field of steel metallurgy automation. The application acquires static characteristics and dynamic time sequence characteristics of the last 30 seconds at the end of blowing by means of an automatic data synchronization gateway, and constructs derived characteristic variables with metallurgical physical and chemical significance. A composite algorithm architecture composed of a multi-machine learning regression array, an adaptive KNN and a physical information neural network PINN is used for prediction, and an intelligent evaluator dynamically schedules an optimal fusion strategy to output endpoint slag multi-components and total slag quantity. Further based on saturated MgO solubility and carbon thermal reduction molecular equilibrium constant, the feeding instructions of raw dolomite, magnesium balls and coke butter are quantitatively settled, the flux synergistic effect is automatically deducted and integral quantization limiting is executed, and closed-loop automatic converter protection is realized. The application greatly improves the prediction accuracy and robustness, realizes quantitative converter protection control without human intervention, significantly reduces auxiliary material consumption and prolongs the service life of the furnace lining.

Description

Technical Field

[0001] This invention belongs to the field of automation control and intelligent steelmaking technology in iron and steel metallurgy, specifically involving a method for predicting the final slag composition of a converter based on mechanism-data fusion, a method for guiding furnace protection, and a system. Background Technology

[0002] Converter slag splashing is a core process for extending furnace lining life and reducing refractory material consumption, and its success or failure is directly dependent on the physicochemical properties of the final slag. In traditional industrial settings, obtaining the final slag composition relies on on-site feedback from laboratory tests, which has a 10-15 minute time lag, forcing workers to rely solely on manual experience for slag adjustment and splashing. If the final slag composition becomes uncontrolled, it can easily lead to safety accidents such as large-scale slag layer detachment, severe abnormal erosion of the furnace lining, or even slag splashing.

[0003] Currently, the prediction and control of converter end-point slag mainly relies on two types of existing technologies: The first type is the traditional static empirical scheme based on static material balance and multiple linear regression. It relies entirely on the static initial conditions in the early stage of blowing and uses multiple linear empirical formulas or tables to estimate the final slag composition. This scheme ignores the changes in the micro-reaction field at the end of blowing and cannot detect slag over-oxidation or furnace lining erosion caused by severe post-blowing or excessively high lance position.

[0004] The second category is independent machine learning regression schemes based on pure data-driven approaches, such as backpropagation neural networks and support vector machines. These models only pursue the minimization of mathematical loss functions and lack the constraints of metallurgical physicochemical mechanisms. They are prone to outputting predictions that violate common sense in metallurgy when raw material fluctuations or steel grade changes occur, resulting in poor generalization ability.

[0005] In addition, existing systems generally cannot directly and quantitatively convert the predicted slag over-oxidation degree or MgO saturation into charging instructions for the slag modification facility. Operators still rely on subjective experience to estimate the charging, resulting in excessive or insufficient flux, which seriously affects the furnace lining life and production safety.

[0006] Therefore, there is an urgent need for an intelligent method that can accurately predict the final slag composition and automatically and quantitatively guide the furnace protection feeding. Summary of the Invention

[0007] This invention aims to provide a method, system, and equipment for predicting the composition of converter slag at the end point and guiding furnace protection, in order to solve the technical problems of prediction lag, easy distortion of pure data models, and inability to automatically close the loop control in the prior art. It achieves high-precision prediction of the multi-component and total slag quantity of the slag at the end point, and automatically and quantitatively issues slag modification and furnace protection control commands based on saturated MgO theory and heat balance rules.

[0008] To achieve the above objectives, the technical solution provided by the present invention is as follows: This invention provides a method for predicting the final slag composition of a converter and providing guidance for furnace protection, comprising the following steps: S1. Feature Data Extraction Steps: Through an automated data synchronization gateway, static and dynamic time-series feature values ​​from the converter secondary control system are acquired in real time. The static feature values ​​include at least the iron content, scrap steel addition amount, auxiliary material addition amount, historical slag analysis data, historical steel analysis data, and dynamic measurement data of the auxiliary lance. The dynamic time-series feature values ​​include the average lance position, average oxygen flow rate, average CO content in flue gas, and average CO2 content in flue gas during the last 30 seconds of the blowing process. S2, Multi-dimensional input feature construction steps: Perform feature engineering processing based on metallurgical boundaries on the feature data extracted in step S1 to construct derived feature variables with metallurgical physicochemical significance. The derived feature variables include at least lime melting efficiency CaO, theoretical slag basicity R, and dephosphorization index Lp. S3. Multi-objective ensemble learning prediction step: Input the feature vector constructed in step S2 into the composite algorithm architecture, which includes: The first prediction branch is a machine learning regression array consisting of a deep neural network MLP, a histogram gradient boosting tree HistGBDT, a random forest RF, and a limit tree ET. The second prediction branch, based on the adaptive nearest neighbor regression network KNN, outputs an inference value space with local spatiotemporal similarity. The third prediction branch uses a deep neural network to specifically fit the residual between the calculated values ​​of metallurgical thermodynamics theory and the actual dynamic measurement values ​​of physicochemical processes, and outputs the residual correction and deduction value space. The intelligent evaluator scans the dynamic feature vector space layout of the current furnace in real time, predicts the posterior probability distribution of multiple fusion strategies, and schedules the optimal fusion strategy according to the principle of maximizing probability, outputting the predicted values ​​of each component of the final slag and the theoretically calculated total slag volume. S4. Closed-loop control steps for furnace protection feeding: Based on the predicted value of the final slag composition and the theoretically calculated total slag amount output in step S3, the theoretical amount of slag adjustment material is dynamically calculated according to the preset slag modification and furnace protection feeding rules. The calculated amount of slag adjustment material is written into the L2 central database through the local area network and automatic feeding is triggered.

[0009] Preferably, in step S2, the lime melting efficiency CaO is constructed according to the following formula: Where T hm Where is the temperature of the molten iron, [Si] is the silicon content of the molten iron, and C is the target carbon content at the final stage. The dephosphorization index Lp is constructed according to the following formula: Where Tsteel The final temperature of the molten steel. To estimate the CaO percentage, TFe is the median total iron in the slag from the previous period.

[0010] Preferably, step S1 further includes a self-healing data processing mechanism, which includes: Establish a threshold value based on the basics of metallurgical thermodynamics. When the recorded value of a furnace exceeds the threshold value, it is judged as abnormal data and blocked. For missing data, recent production data of the same steel grade as the currently missing heat are selected from the historical database, and the average value of that steel grade is calculated to fill in the missing data. The true values ​​of sudden fluctuations in slag composition during abnormal splashing or peroxidation processes are retained.

[0011] Preferably, in step S1, the dynamic time-series characteristic values ​​are obtained by continuously sampling the actual position trajectory of the oxygen lance, the height of the oxygen lance position movement trajectory, and the time-series variation rate of CO and CO2 in the flue gas from the last 30 sample points at the end of the blowing process. These values ​​are used to dynamically characterize the over-oxidation tendency of the nonlinear reaction in the furnace and the physical-fluid properties of the slag.

[0012] Preferably, in step S3, the training of the machine learning regression array adopts a time exponential decay function, so that historical samples closer to the current time node receive higher training weights; the intelligent evaluator extracts the fourth power of the inverse mean absolute percentage error of the test set as a dynamic weighting benchmark.

[0013] Preferably, in step S4, the slag modification and furnace protection feeding rules include: Basicity deficit control rule: When the predicted basicity R < 2 and the oxide mass fraction difference equation 2 × SiO2 - CaO > 0 is satisfied, raw dolomite is dynamically replenished according to the following formula: Saturated MgO furnace protection rules: Based on the safe threshold of saturated MgO solubility, magnesium balls are dynamically replenished according to the following formula: Where ΔMgO is the thermodynamic deficit between the current slag and the safe saturation MgO solubility red line, ΔMgO = max(0, 8%-MgO); Carbothermic reduction oxygen reduction rule: When the predicted total iron content (TFe) > 25%, dynamically supplement coking coal according to the following formula: in, This represents the total iron content.

[0014] Preferably, in step S4, the calculated amount of raw dolomite, magnesium balls, and coke is uniformly calculated in whole-volume increments of 50 kg, and a single-furnace safety limit is set: raw dolomite ≤ 800 kg, magnesium balls ≤ 400 kg, and coke ≤ 300 kg. In the saturated MgO furnace protection rule, the system first deducts the MgO mass fraction brought in by the addition of raw dolomite to supplement alkalinity. If there is still a MgO saturation gap, the addition of magnesium balls is triggered.

[0015] Preferably, in step S2, the multi-dimensional input feature values ​​further include: the theoretical SiO2 content and the theoretical CaO mass ratio of auxiliary material dissolution based on dynamic calculation of the material balance equation for low-silicon molten iron; the combined characteristics of scrap steel cooling equivalent and carbon oxidation heating equivalent introduced for high scrap steel ratio; a 3×3×3 working condition grid constructed by three-dimensional orthogonal segmentation of the P content of molten iron entering the furnace, the target C content at the endpoint, and the oxygen supply intensity; and the real TFe, MgO, and basicity data of the previous furnace.

[0016] This application also provides an electronic device, including: a memory and a processor, wherein the memory stores computer program instructions for execution on the processor, and when the processor executes the computer program instructions, it implements the risk stratification generation method for autonomous driving test scenarios based on the conditional diffusion model as described above.

[0017] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method for generating risk stratification of autonomous driving test scenarios based on the conditional diffusion model as described above.

[0018] Beneficial effects Compared with the prior art, the present invention has the following advantages: 1) Significantly improve prediction accuracy: By introducing the dynamic time-series characteristics (oxygen lance trajectory, flue gas CO / CO2 variation rate) of the last 30 seconds of the blowing process, instead of relying solely on the static charge amount, the average absolute error of the prediction of total slag amount, TFe and MgO is reduced to within ±1.5%, and the multi-component prediction hit rate is increased by more than 12%.

[0019] 2) Strong robustness and safety: A dedicated Physical Information Neural Network (PINN) branch is set up in the model architecture to fit the nonlinear residuals between the thermodynamic theoretical boundary and the measured value. Ten machine learning algorithms are integrated to build a multi-model integrated prediction network to ensure that the prediction output is always constrained within the metallurgical thermodynamic / kinetic boundary under extreme working conditions such as steel grade switching and drastic fluctuations in raw materials, thus avoiding the output of prediction values ​​that violate physical common sense from pure data models.

[0020] 3) Achieve closed-loop quantitative control: Directly connect the secondary control chain, based on the red line of saturated MgO solubility and the molecular equilibrium constant of carbothermic reduction, quantitatively calculate the precise feeding instructions of raw dolomite, magnesium balls and coke, and automatically deduct the synergistic effect between fluxes. Implement 50kg step-size quantification and safety upper limit limit to achieve completely unattended closed-loop control of slag adjustment and splashing furnace protection. On average, the cost of auxiliary materials per furnace is reduced by about 15%, and the furnace lining life is significantly extended. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic flowchart of a converter endpoint slag composition prediction and furnace protection guidance method according to an embodiment of the present invention.

[0023] Figure 2 This is a diagram of the actual industrial operation interface and the charging instruction board for furnace number 26224687 in this embodiment of the invention. Detailed Implementation

[0024] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] Please see Figure 1 This invention provides a method for predicting the final slag composition of a converter and providing guidance for furnace protection, comprising the following steps: S1. Feature Data Extraction Steps: Through an automated data synchronization gateway, static and dynamic time-series feature values ​​from the converter secondary control system are acquired in real time. The static feature values ​​include at least the iron content, scrap steel addition amount, auxiliary material addition amount, historical slag analysis data, historical steel analysis data, and dynamic measurement data of the auxiliary lance. The dynamic time-series feature values ​​include the average lance position, average oxygen flow rate, average CO content in flue gas, and average CO2 content in flue gas during the last 30 seconds of the blowing process. S2, Multi-dimensional input feature construction steps: Perform feature engineering processing based on metallurgical boundaries on the feature data extracted in step S1 to construct derived feature variables with metallurgical physicochemical significance. The derived feature variables include at least lime melting efficiency CaO, theoretical slag basicity R, and dephosphorization index Lp. S3. Multi-objective ensemble learning prediction step: Input the feature vector constructed in step S2 into the composite algorithm architecture, which includes: The first prediction branch is a machine learning regression array consisting of a deep neural network MLP, a histogram gradient boosting tree HistGBDT, a random forest RF, and a limit tree ET. The second prediction branch, based on the adaptive nearest neighbor regression network KNN, outputs an inference value space with local spatiotemporal similarity. The third prediction branch uses a deep neural network to specifically fit the residual between the calculated values ​​of metallurgical thermodynamics theory and the actual dynamic measurement values ​​of physicochemical processes, and outputs the residual correction and deduction value space. The intelligent evaluator scans the dynamic feature vector space layout of the current furnace in real time, predicts the posterior probability distribution of multiple fusion strategies, and schedules the optimal fusion strategy according to the principle of maximizing probability, outputting the predicted values ​​of each component of the final slag and the theoretically calculated total slag volume. S4. Closed-loop control steps for furnace protection feeding: Based on the predicted value of the final slag composition and the theoretically calculated total slag amount output in step S3, the theoretical amount of slag adjustment material is dynamically calculated according to the preset slag modification and furnace protection feeding rules. The calculated amount of slag adjustment material is written into the L2 central database through the local area network and automatic feeding is triggered.

[0026] In one embodiment, in step S2, the lime melting efficiency CaO is constructed according to the following formula: Where T hm Where is the temperature of the molten iron, [Si] is the silicon content of the molten iron, and C is the target carbon content at the final stage. The dephosphorization index Lp is constructed according to the following formula: Where T steel The final temperature of the molten steel. To estimate the CaO percentage, TFe is the median total iron in the slag from the previous period.

[0027] In one embodiment, step S1 further includes a self-healing data processing mechanism, which includes: Establish a threshold value based on the basics of metallurgical thermodynamics. When the recorded value of a furnace exceeds the threshold value, it is judged as abnormal data and blocked. For missing data, recent production data of the same steel grade as the currently missing heat are selected from the historical database, and the average value of that steel grade is calculated to fill in the missing data. The true values ​​of sudden fluctuations in slag composition during abnormal splashing or peroxidation processes are retained.

[0028] In one embodiment, in step S1, the dynamic time-series characteristic values ​​are obtained by continuously sampling the actual position trajectory of the oxygen lance, the height of the oxygen lance position movement trajectory, and the time-series variation rate of CO and CO2 in the flue gas from the last 30 sample points at the end of the blowing process. These values ​​are used to dynamically characterize the over-oxidation tendency of the nonlinear reaction in the furnace and the physical-fluid properties of the slag.

[0029] In one embodiment, in step S3, the training of the machine learning regression array adopts a time exponential decay function, so that historical samples closer to the current time node receive higher training weights; the intelligent evaluator extracts the fourth power of the inverse mean absolute percentage error of the test set as a dynamic weighting benchmark.

[0030] In one embodiment, in step S4, the modified slag protection furnace feeding rules include: Basicity deficit control rule: When the predicted basicity R < 2 and the oxide mass fraction difference equation 2 × SiO2 - CaO > 0 is satisfied, raw dolomite is dynamically replenished according to the following formula: Saturated MgO furnace protection rules: Based on the safe threshold of saturated MgO solubility, magnesium balls are dynamically replenished according to the following formula: Where ΔMgO is the thermodynamic deficit between the current slag and the safe saturation MgO solubility red line, ΔMgO = max(0, 8%-MgO); Carbothermic reduction oxygen reduction rule: When the predicted total iron content (TFe) > 25%, dynamically supplement coking coal according to the following formula: in, This represents the total iron content.

[0031] In one embodiment, in step S4, the calculated amount of raw dolomite, magnesia balls, and coke are uniformly calculated in whole-volume increments of 50 kg, and a single-furnace safety limit is set: raw dolomite ≤ 800 kg, magnesia balls ≤ 400 kg, and coke ≤ 300 kg. In the saturated MgO furnace protection rule, the system first deducts the MgO mass fraction brought in by the addition of raw dolomite to supplement alkalinity. If there is still a MgO saturation gap, the addition of magnesia balls is triggered.

[0032] In one embodiment, step S2 further includes the following multi-dimensional input feature values: the theoretical SiO2 content and the theoretical CaO mass ratio of additive dissolution, dynamically calculated based on the material balance equation for low-silicon molten iron; the combined characteristics of scrap steel cooling equivalent and carbon oxidation heating equivalent introduced for high scrap steel ratio; a 3×3×3 working condition grid constructed by three-dimensional orthogonal segmentation of the P content of the molten iron entering the furnace, the target C content at the endpoint, and the oxygen supply intensity; and the actual TFe, MgO, and basicity data of the previous furnace.

[0033] Example 1 In this embodiment, the actual smelting furnace number of the 100t converter is 26224687, and the steel grade is HRB400-G1.

[0034] I. Automatic Extraction of Feature Data As heat 26224687 neared its end, the heterogeneous characteristic data gateway automatically intercepted and synchronized the basic production characteristic values ​​and end-stage kinetic characteristics of the current heat: Static data: This included the steel grade, hot metal charge (67.19t), scrap steel charge, initial hot metal temperature (1317℃), initial elemental content of the hot metal (mass percentage of [C], [Si], [Mn], [P], and [S]), and the total amount of conventional slag flux previously added. All data were verified to be within normal ranges and no abnormal warnings were triggered.

[0035] Dynamic characteristics: The system extracts the oxygen lance position movement trajectory height and flue gas composition in real time during the last 30 seconds of blowing, calculates the average lance position, average oxygen flow rate, and average change rate of CO and CO2 in the flue gas at the end of the blowing period, and forms a dynamic characteristic field vector.

[0036] II. Joint Prediction of Heterogeneous Algorithm Networks and Mechanistic Boundaries The extracted feature matrix is ​​input into the intelligent control engine. First, metallurgical physicochemical mechanism formulas such as the dynamic melting efficiency of active lime (CaO) and the dephosphorization index (Lp) are injected into the feature engineering as constraint terms. Then, parallel simulation is performed: compensation calculations are carried out using model groups such as MLP and HistGBDT, combined with the blowing conditions of the most recent 5 heats.

[0037] Finally, the intelligent evaluator constructed by the secondary machine learning classifier dynamically scans the current feature field layout, adaptively schedules the optimal composite strategy space, and outputs the predicted values ​​of the final slag chemical composition and the calculated slag quantity for 26,224,687 heats. Predicted alkalinity R: 2.5148 (within the safe range of 2.0 to 2.8) Predicted MgO: 6.9057% (below the 8.0% red line for thermodynamic saturation solubility of slag) Predicted TFe: 27.757% (exceeding the 20%~25% safety threshold) The final theoretical slag volume for the current furnace run is calculated to be 9.5229t. III. System slag adjustment and furnace protection quantitative control settlement Input the predicted value into the feeding control module: R=2.5148>2 was detected, and low alkalinity whitening compensation was not triggered.

[0038] MgO = 6.9057% < 8%, activating the magnesium replenishment and furnace protection logic. ΔMgO = 8% - 6.9057% = 1.0943%. Raw white metal (effective MgO mass percentage 19.485%) is used for loss compensation. The addition amount of raw whitening agent was determined to be 650 kg (not exceeding the upper limit of 800 kg) by implementing a step-by-step adjustment of 50 kg.

[0039] Adding 650kg of raw white MgO actually replenishes 0.904% of the MgO mass, which is (650×19.485%) / 14000. This covers the deficit, and the system automatically resets the amount of magnesium balls added to zero.

[0040] The predicted TFe value is 18.5% < 25%, so the kerosene control channel is not triggered.

[0041] IV. Industrial Operation Interface Display Please see Figure 2 It intuitively displays the core components of the final slag calculated in real time and the final precise quantitative feeding execution command issued.

[0042] Comparative Example To verify the effectiveness of this invention, 50 heats using the traditional BP neural network prediction model (existing technical solution two) were selected as comparative examples on the same converter. The results show: When switching steel grades, the traditional BP model shows that the predicted MgO value deviates from the thermodynamic saturation boundary (below 2% or above 15%) in 18% of cases, while the model of this invention predicts values ​​within a reasonable range in all heats.

[0043] Traditional models cannot achieve automatic closed-loop control, resulting in an average feed deviation of ±120kg per furnace for operators. In contrast, the present invention automatically calculates the deviation to be less than ±25kg.

[0044] After adopting the present invention, the density of the slag layer is significantly improved, and the scouring life of the furnace lining is extended by about 22%.

[0045] The above results demonstrate that the present invention is significantly superior to the prior art in terms of prediction accuracy, robustness, and automation level.

[0046] This application also provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the risk stratification generation method for autonomous driving test scenarios based on the conditional diffusion model described above.

[0047] It is understood that the above scenarios are merely examples and do not constitute a limitation on the application scenarios of the technical solutions provided in the embodiments of this application. The technical solutions of this application can also be applied to other scenarios. For example, as those skilled in the art will know, with the evolution of system architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0048] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0049] The steps in the method of this application embodiment can be adjusted, combined, or deleted according to actual needs.

[0050] The units in the device of this application embodiment can be merged, divided, and deleted according to actual needs.

[0051] In this application, the same or similar terms, concepts, technical solutions and / or application scenario descriptions are generally described in detail only when they appear for the first time. When they appear again, they are generally not repeated for the sake of brevity. When understanding the technical solutions and other contents of this application, the same or similar terms, concepts, technical solutions and / or application scenario descriptions that are not described in detail later can be referred to their previous relevant detailed descriptions.

[0052] In this application, the descriptions of the various embodiments have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0053] The technical features of the present application can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present application.

[0054] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, controlled terminal, or network device, etc.) to execute the methods of each embodiment of this application.

[0055] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a storage medium or transmitted from one storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, storage disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)).

[0056] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting the final slag composition of a converter and providing guidance for furnace protection, characterized in that, Includes the following steps: S1. Through an automated data synchronization gateway, the static characteristic values ​​and dynamic time-series characteristic values ​​of the converter secondary control system are acquired in real time. The static characteristic values ​​include at least the iron content, scrap steel addition amount, auxiliary material addition amount, slag analysis historical data, molten steel analysis historical data, and auxiliary lance dynamic measurement data. The dynamic time-series characteristic values ​​include the average lance position, average oxygen flow rate, average CO content in flue gas, and average CO2 content in flue gas in the last 30 seconds of the blowing process. S2. Perform feature engineering processing on the feature data extracted in step S1 based on the metallurgical boundary to construct derived feature variables with metallurgical physicochemical significance. The derived feature variables include at least lime melting efficiency CaO, theoretical slag basicity R, and dephosphorization index Lp. S3. Input the feature vector constructed in step S2 into the composite algorithm architecture, which includes: The first prediction branch is a machine learning regression array consisting of a deep neural network MLP, a histogram gradient boosting tree HistGBDT, a random forest RF, and a limit tree ET. The second prediction branch, based on the adaptive nearest neighbor regression network KNN, outputs an inference value space with local spatiotemporal similarity. The third prediction branch uses a deep neural network to specifically fit the residual between the calculated values ​​of metallurgical thermodynamics theory and the actual dynamic measurement values ​​of physicochemical processes, and outputs the residual correction and deduction value space. The intelligent evaluator scans the dynamic feature vector space layout of the current furnace in real time, predicts the posterior probability distribution of multiple fusion strategies, and schedules the optimal fusion strategy according to the principle of maximizing probability, outputting the predicted values ​​of each component of the final slag and the theoretically calculated total slag volume. S4. Based on the predicted value of the final slag composition and the theoretically calculated total slag quantity output in step S3, dynamically calculate the theoretical amount of slag adjustment material according to the preset slag modification and furnace protection feeding rules, and write the calculated amount of slag adjustment material into the L2 central database through the local area network and trigger automatic feeding.

2. The method for predicting converter endpoint slag composition and providing furnace protection guidance according to claim 1, characterized in that, In step S2, the lime melting efficiency CaO is constructed according to the following formula: Where T hm Where is the temperature of the molten iron, [Si] is the silicon content of the molten iron, and C is the target carbon content at the final stage. The dephosphorization index Lp is constructed according to the following formula: Where T steel The final temperature of the molten steel. To estimate the CaO percentage, TFe is the median total iron in the slag from the previous period.

3. The method for predicting converter end-point slag composition and providing furnace protection guidance according to claim 1, characterized in that, Step S1 also includes a self-healing data processing mechanism, which includes: Establish a threshold value based on the basics of metallurgical thermodynamics. When the recorded value of a furnace exceeds the threshold value, it is judged as abnormal data and blocked. For missing data, recent production data of the same steel grade as the currently missing heat are selected from the historical database, and the average value of that steel grade is calculated to fill in the missing data. The true values ​​of sudden fluctuations in slag composition during abnormal splashing or peroxidation processes are retained.

4. The method for predicting converter endpoint slag composition and providing furnace protection guidance according to claim 1, characterized in that, In step S1, the dynamic time-series characteristic values ​​are obtained by continuously sampling the actual position trajectory of the oxygen lance, the height of the oxygen lance position movement trajectory, and the time-series variation rate of CO and CO2 in the flue gas from the last 30 sample points at the end of the blowing process. These values ​​are used to dynamically characterize the over-oxidation tendency of the nonlinear reaction in the furnace and the physical-fluid properties of the slag.

5. The method for predicting converter end-point slag composition and providing furnace protection guidance according to claim 1, characterized in that, In step S3, the training of the machine learning regression array adopts a time exponential decay function, so that historical samples closer to the current time node receive higher training weights; the intelligent evaluator extracts the fourth power of the inverse mean absolute percentage error of the test set as a dynamic weighting benchmark.

6. The method for predicting converter endpoint slag composition and providing furnace protection guidance according to claim 1, characterized in that, In step S4, the slag modification and furnace protection feeding rules include: Basicity deficit control rule: When the predicted basicity R < 2 and the oxide mass fraction difference equation 2 × SiO2 - CaO > 0 is satisfied, raw dolomite is dynamically replenished according to the following formula: Saturated MgO furnace protection rules: Based on the safe threshold of saturated MgO solubility, magnesium balls are dynamically replenished according to the following formula: Where ΔMgO is the thermodynamic deficit between the current slag and the safe saturation MgO solubility red line, ΔMgO = max(0, 8%-MgO); Carbothermic reduction oxygen reduction rule: When the predicted total iron content (TFe) > 25%, dynamically supplement coking coal according to the following formula: in, This represents the total iron content.

7. The method for predicting converter endpoint slag composition and providing furnace protection guidance according to claim 6, characterized in that, In step S4, the calculated amounts of raw dolomite, magnesia balls, and coke are uniformly calculated using a physical step of 50 kg, and a single furnace safety limit is set: raw dolomite ≤ 800 kg, magnesia balls ≤ 400 kg, and coke ≤ 300 kg. In the saturated MgO furnace protection rule, the system first deducts the MgO mass fraction brought in by the addition of raw dolomite to supplement alkalinity. If there is still a MgO saturation gap, the addition of magnesia balls is triggered.

8. The method for predicting converter end-point slag composition and providing guidance for furnace protection according to claim 1, characterized in that, In step S2, the multi-dimensional input feature values ​​also include: the theoretical SiO2 content and the theoretical CaO mass ratio of auxiliary material dissolution based on dynamic calculation of the material balance equation for low-silicon molten iron; the combined characteristics of scrap steel cooling equivalent and carbon oxidation heating equivalent introduced for high scrap steel ratio; a 3×3×3 working condition grid constructed by three-dimensional orthogonal segmentation of the P content of molten iron entering the furnace, the target C content at the endpoint, and the oxygen supply intensity; and the real TFe, MgO, and basicity data of the previous furnace.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1 to 8.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1 to 8.

Images