Converter steelmaking self-learning method and system based on data driving

Abstract

The invention discloses a converter steelmaking self-learning method and system based on data driving, and belongs to the technical field of metallurgical intelligent converter automatic control. The system comprises an information acquisition module, a sample labeling module, a sample management library, a data preprocessing module, an online decision module, a data perspective module and an optimization module, wherein the information acquisition module acquires raw and auxiliary material data, process control parameters and equipment state data; the sample labeling module is used for classifying and labeling the obtained data; the sample management library stores data; the data preprocessing module is used for preprocessing the collected data; the online decision-making module is used for judging and adjusting the smelting state; the data perspective module is used for grading and quantifying the smelting result; and the optimization module carries out optimal mode smelting mode, index correlation factor analysis and model parameter self-optimization through historical data. According to the invention, intelligent smelting of the converter is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of metallurgical intelligent converter automatic control, and in particular to a data-driven converter steelmaking self-learning method and system. Background Art

[0002] With the rapid development of the global steel industry, converter steelmaking, as one of the core processes of steel production, is facing many challenges in improving production efficiency, reducing energy consumption and ensuring stable product quality. At the same time, the steel industry is entering a critical period of digital transformation. With the increasing maturity of artificial intelligence and big data technologies, the intelligentization of the entire steel production process has become an irreversible trend.

[0003] The traditional converter steelmaking control method mainly relies on the operator's experience and preset process parameters. This method has the following major shortcomings: First, the process is complex and uncertain. The converter steelmaking process involves multiple complex physical and chemical reactions, which are affected by multiple factors such as molten iron composition, temperature, scrap steel ratio and furnace atmosphere. Due to the dynamic and uncertain nature of the process, traditional control methods are difficult to adjust and optimize in real time; second, it is highly dependent on experience. The current steelmaking process control relies heavily on the operator's experience. The skill level and experience of different operators lead to the instability of steelmaking quality, making it difficult to ensure the consistency and high quality of products; then, it lacks intelligence and automation. Although some automated control systems have been applied to converter steelmaking, most systems still rely on pre-set process parameters, lack adaptive and self-learning capabilities, and are difficult to cope with real-time changes and emergencies in the production process; finally, data utilization is insufficient: In the converter steelmaking process, a large amount of process data and production records are generated. However, the utilization rate of these data is not high, and they have not been fully applied to process optimization and quality control. The existing technology is still in its infancy in data-driven process optimization and has not fully utilized the value of data. Summary of the invention

[0004] The object of the present invention is to provide a data-driven converter steelmaking self-learning method and system to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a self-learning system for converter steelmaking driven by data, the system comprising: an information acquisition module, a sample annotation module, a sample management library, a data preprocessing module, an online decision-making module, a data perspective module and an optimization module, the information acquisition module is used to acquire raw and auxiliary material data, process control parameters and equipment status data; the sample annotation module is used to classify and annotate the acquired data according to the different requirements of the optimization calculation model, the online decision model and the data perspective model; the sample management library is used to store all acquired data; the data preprocessing module is used to preprocess the acquired data, including data cleaning, missing value filling, data standardization and feature engineering; the online decision-making module is used to judge and calculate and adjust the smelting state during the smelting process; the data perspective module is used to score and quantify the smelting results; the optimization module is used to perform optimal mode smelting mode, indicator correlation factor analysis and model parameter self-optimization through historical data.

[0006] Furthermore, the information acquisition module includes a raw material data acquisition unit, a process control parameter acquisition unit and an equipment status data acquisition unit. The raw material data acquisition unit is used to collect raw material data through the iron plant and the scrap steel workshop, and record it into the EMS system and then output it. The raw material data includes molten iron composition, molten iron temperature, slag content of molten iron, molten iron weight, proportion type of scrap steel, weight of scrap steel, temperature of scrap steel, type and composition of available materials; the process control parameter acquisition unit records the process control parameters collected by PLC into the L1 system through the staff and outputs it. The process control parameters include obtaining the composition requirements, temperature requirements, end slag requirements, weight requirements, flame information and sounds generated during the steelmaking process of the molten steel target; the equipment status data acquisition unit is recorded and input by on-site operators or the secondary system, and the equipment status data includes the furnace service, empty furnace time, lining information and bottom blowing status of the converter.

[0007] Furthermore, the online decision module judges and calculates and adjusts the smelting state during the smelting process by establishing an online decision model, and the operation process is described as follows: first, raw materials and process control parameters are obtained through the sample management library, and the obtained process control parameters and raw materials data are input into the preliminary processing stage of the online decision model to extract characteristic data of energy consumption, material consumption, equipment status, smelting control and time information related to steelmaking performance; secondly, the extracted characteristic data are used to conduct a quantitative analysis of the overall performance of steelmaking, and the quantitative results of the safety effect of converter steelmaking are obtained based on raw materials, process control parameters and current input; then, deep neural network technology is introduced to fit the relationship function between raw materials, process control parameters and steelmaking performance, and estimated indicators are obtained through performance deduction based on current real-time sensor information, and the estimated indicators are compared with acceptable indicators. By analyzing the difference between the estimated indicators and the acceptable indicators, an input judgment and correction model is obtained; finally, the final process control parameters and raw material inputs are obtained.

[0008] Furthermore, the data perspective module scores and quantifies the smelting results by establishing a data perspective model, analyzes and obtains converter smelting effect indicators such as endpoint hit rate, metal recovery rate, endpoint slag composition, dephosphorization rate, slag volume, smelting stability, endpoint oxygen level and material consumption, adjusts the weight of each indicator according to the production requirements of different enterprises, and uses regression analysis to explore the relationship between endpoint indicators, raw and auxiliary material data, process control parameters and equipment status, so as to quantitatively score the smelting results.

[0009] Furthermore, the optimization module divides the optimization process into two stages, local optimization and global optimization, by establishing an optimization calculation model, which is used to perform optimal mode smelting and indicator correlation factor analysis and model parameter self-optimization through historical data. The self-learning optimization process is as follows: First, the process control parameters are obtained through the sample management library, and after data preprocessing, the processed data is combined with the quantitative results of the safety effect of converter steelmaking in the sample management library, and the smelting process is further converted into measurable quantitative indicators, including local goals and global goals, through cost analysis. Secondly, local goal optimization and global goal optimization are performed through a self-learning algorithm based on converter steelmaking process data combined with a heuristic algorithm, an evolutionary algorithm and a simulated annealing technology method. Local target optimization: the self-learning algorithm based on the converter steelmaking process data determines the optimal state of the local target 1-M, and on the premise of ensuring that the local target 1-M reaches the required state, the optimization calculation model is used to optimize the process control parameters and find the optimal range of the process control parameter values; the global target optimization: by designing a global optimization mathematical model, the self-learning algorithm based on the converter steelmaking process data integrates the local targets 1-M into a global target, and based on the previous local optimization results, further optimizes the global control parameters and finds the optimal state of the global target. Finally, the self-learning algorithm based on the converter steelmaking process data outputs the optimal control parameters and smelting mode, and uses them to guide the next round of steelmaking operations, where M represents the number of local targets.

[0010] A data-driven converter steelmaking self-learning method comprises the following steps:

[0011] S1: Obtain raw material data, process control parameters and equipment status data;

[0012] S2: Set the automatic selection logic according to historical data and expert rules, load the machine learning model, load the target control curve, monitor the import results, and if the model is imported accurately and the status is normal, proceed to the next step, otherwise the alarm is transferred to manual processing;

[0013] S3: Start blowing, and the converter steelmaking self-learning system interacts with the L1 system based on data-driven;

[0014] S4: Online decision model raw materials, process control parameters, equipment status and current input information, calculates the smelting status and target curve deviation, and determines the converter smelting stage according to the deviation value;

[0015] S5: After smelting is completed, the converter information of the current furnace is managed and stored based on the principle of furnace number and time and space consistency;

[0016] S6: According to the on-site management requirements for converter smelting and the current furnace converter inspection and testing data, the endpoint hit rate, metal recovery rate, endpoint slag composition, dephosphorization rate, slag volume, smelting stability, endpoint oxygen level and material consumption converter smelting effect indicators are evaluated, and the weights of the adjustment indicators are adjusted. Then, the data perspective model is used to score and quantify the smelting results, and excellent furnaces are selected according to the scores to prepare for self-learning and further optimization of the smelting process;

[0017] S7: Based on the inspection and testing results of historical batches and the deviation of the calculation results of the optimization calculation model as the error function, the model parameters are automatically corrected through the self-learning optimization algorithm of gradient descent, least squares, statistical estimation, regression analysis and sliding average to minimize the error function, and the corrected online decision model parameters are saved to prepare for the next batch of production;

[0018] S8: According to the historical furnace smelting scores, adjust the smelting target curve, including the gas curve, audio curve, temperature curve and flame image curve of the flue gas analysis;

[0019] S9: Label samples as needed in the sample labeling module to provide sample support for subsequent training and self-learning.

[0020] Furthermore, in step S2: automatic selection logic is set according to historical data and expert rules, a machine learning model is loaded, and a target control curve is loaded, wherein the target control curve is determined according to actual production practice, and a curve with a higher furnace score is selected as the target curve.

[0021] Further, in step S4: the online decision model calculates the deviation between the smelting state and the target curve based on raw materials, process control parameters, equipment status and current input information, wherein the deviation types include: the comparison results of the gas concentration curve, audio curve, temperature curve, flame image of the flue gas analysis and the corresponding characteristic curve of the target, and the stage of converter smelting is determined by calculating the numerical deviation value and the time information deviation value on the characteristic point and the target curve, wherein the characteristic points include the following: the gas concentration characteristic points of the flue gas analysis include the curve inflection point, peak point and decline inflection point; the characteristic points of the audio curve include the frequency change point and the amplitude change point; the characteristic points of the flame image characteristic curve include the brightness change point and the color change point; the characteristic points of the temperature curve include the temperature rise point, the stabilization point and the decline point. The specific steps for determining the converter smelting stage are as follows:

[0022] S4-1: Determine the key smelting stages and the corresponding target curves. The converter steelmaking process can be divided into the following stages: Oxygen blowing start stage: At this time, the carbon-oxygen reaction has not yet started, and the CO content in the flue gas is low; Carbon-oxygen reaction enhancement stage: CO begins to be generated in large quantities, the CO content in the flue gas increases significantly, and the target curve has the first inflection point. The target curve is relatively flat at this stage; Intense reaction stage: The CO content reaches a peak, the reaction is most intense, and the target curve at this stage should show high and stable values; Carbon-oxygen reaction weakening stage: CO generation gradually decreases, and the target curve gradually falls back; End stage: The reaction is basically over, the CO content is close to zero, and the target curve is stable;

[0023] S4-2: By analyzing the curve of gas content changing with time, temperature curve, audio signal and flame image curve and comparing them with the target curve, the numerical deviation and time deviation are calculated;

[0024] S4-3: By analyzing the deviation inference stage, the initial stage: the actual CO curve has not shown a significant rise, and the temperature curve is in a stable stage, so the system can be judged to be still in the initial stage of oxygen blowing; the beginning stage of carbon-oxygen reaction: the actual CO curve begins to rise significantly and is close to the first inflection point of the target curve, it can be judged that it has entered the carbon-oxygen reaction enhancement stage. If the CO curve rises too fast, it means that the reaction is too intense and the temperature in the furnace is high; the intense reaction stage: when the CO curve reaches or exceeds the peak of the target curve and remains stable for a period of time, it indicates that the system is in an intense reaction stage. If the actual value is higher than the target value, it is necessary to adjust the oxygen blowing amount or the bottom blowing intensity; the weakened reaction stage: when the CO curve begins to decline and approaches the descending section of the target curve, it can be judged that it has entered the weakened reaction stage; the ending stage: when the CO curve basically drops to zero, and the temperature and other indicators tend to be stable, the reaction is basically over;

[0025] S4-4: At each stage, according to the size of the deviation, a multivariate regression or neural network regression model is jointly established to quantitatively calculate the correction amount for feeding, oxygen blowing, and bottom blowing, thereby linking the L1 system to perform oxygen gun control, feeding control, and bottom blowing control.

[0026] Furthermore, in step S7: establish an optimization calculation model. Specifically, the i-th local optimization objective function It can be expressed as:

[0027]

[0028] Among them, f i () represents the performance representation function of the i-th local objective, represents the jth decision variable associated with the i-th local objective;

[0029] Furthermore, taking steelmaking cost as the object, a global optimization mathematical model is established to find the global cost optimal parameters and steelmaking mode. The global optimization objective function G can be expressed as:

[0030]

[0031] Among them, ω i and are the importance weight and steelmaking cost corresponding to the i-th local objective respectively.

[0032] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0033] 1. Through the online decision-making model, the present invention can judge the smelting status in real time during the smelting process, and make calculations and adjustments according to the current process conditions, thereby realizing real-time optimization of the steelmaking process and overcoming the shortcomings of traditional methods in terms of dynamics and uncertainty.

[0034] 2. The data perspective model is used to score and quantify the smelting results, achieving an objective evaluation of the production effect of each furnace. In this way, the data-driven converter steelmaking self-learning system can systematically evaluate the performance of various indicators in the smelting process, provide data support for subsequent process optimization and parameter adjustment, and improve the efficiency and effect of the overall steelmaking process.

[0035] 3. Through the optimization calculation model, the present invention can select the optimal smelting mode and analyze the index correlation factors based on historical data, and self-optimize the model parameters through the optimization calculation model, thereby enhancing the adaptive ability of the data-driven self-learning system. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a structural diagram of a data-driven converter steelmaking self-learning system of the present invention;

[0037] Figure 2 It is a work flow chart of a data-driven converter steelmaking self-learning system of the present invention;

[0038] Figure 3 It is an operation process diagram of an online decision model based on a data-driven converter steelmaking self-learning method of the present invention;

[0039] Figure 4 It is a process diagram of a self-learning optimization algorithm based on a data-driven converter steelmaking self-learning method of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] like Figure 1-Figure 4 As shown, the present invention provides a technical solution: a self-learning system for converter steelmaking driven by data, the system comprising: an information acquisition module, a sample annotation module, a sample management library, a data preprocessing module, an online decision module, a data perspective module and an optimization module, the information acquisition module is used to acquire raw material data, process control parameters and equipment status data; the sample annotation module is used to classify and annotate the acquired data according to the different requirements of the optimization calculation model, the online decision model and the data perspective model; the sample management library is used to store all acquired data; the data preprocessing module is used to preprocess the acquired data, including data cleaning, missing value filling, data standardization and feature engineering; the online decision module is used to judge and calculate and adjust the smelting state during the smelting process; the data perspective module is used to score and quantify the smelting results; the optimization module is used to perform optimal mode smelting mode, indicator correlation factor analysis and model parameter self-optimization through historical data.

[0042] The information acquisition module includes a raw material data acquisition unit, a process control parameter acquisition unit and an equipment status data acquisition unit. The raw material data acquisition unit is used to collect raw material data through the iron plant and scrap steel workshop, and record it in the EMS system and then output it. The raw material data includes molten iron composition, molten iron temperature, slag content of molten iron, molten iron weight, proportion type of scrap steel, weight of scrap steel, temperature of scrap steel, type and composition of available materials; the process control parameter acquisition unit records the process control parameters collected by PLC into the L1 system through the staff and outputs it. The process control parameters include the composition requirements, temperature requirements, end slag requirements, weight requirements, flame information and sounds generated during the steelmaking process for obtaining the molten steel target; the equipment status data acquisition unit is recorded and input by on-site operators or the secondary system, and the equipment status data includes the furnace service, empty furnace time, lining information and bottom blowing status of the converter.

[0043] The online decision-making module judges and calculates and adjusts the smelting state during the smelting process by establishing an online decision-making model. The operation process is as follows: First, the raw materials and process control parameters are obtained through the sample management library, and the obtained process control parameters and raw materials data are input into the preliminary processing stage of the online decision-making model to extract the characteristic data of energy consumption, material consumption, equipment status, smelting control and time information related to steelmaking performance; secondly, the extracted characteristic data are used to conduct a quantitative analysis of the overall performance of steelmaking, and the quantitative results of the safety effect of converter steelmaking are obtained based on the raw materials, process control parameters and current input; then, the deep neural network technology is introduced to fit the relationship function between raw materials, process control parameters and steelmaking performance, and the estimated indicators are obtained through performance deduction based on the current real-time sensor information, and the estimated indicators are compared with the acceptable indicators. By analyzing the difference between the estimated indicators and the acceptable indicators, the input judgment and correction model is obtained; finally, the final process control parameters and raw material inputs are obtained.

[0044] The data perspective module scores and quantifies the smelting results by establishing a data perspective model, analyzes and obtains the converter smelting effect indicators of endpoint hit rate, metal recovery rate, endpoint slag composition, dephosphorization rate, slag volume, smelting stability, endpoint oxygen level and material consumption. According to the production requirements of different enterprises, the weight of each indicator is adjusted, and regression analysis is used to explore the relationship between endpoint indicators, raw and auxiliary material data, process control parameters and equipment status, so as to quantitatively score the smelting results.

[0045] The optimization module establishes an optimization calculation model and divides the optimization process into two stages: local optimization and global optimization. It is used to perform optimal mode smelting and indicator correlation factor analysis and model parameter self-optimization through historical data. The self-learning optimization process is as follows: First, the process control parameters are obtained through the sample management library, and after data preprocessing, the processed data is combined with the quantitative results of the safety effect of converter steelmaking in the sample management library. The smelting process is further converted into measurable quantitative indicators through cost analysis, including local goals and global goals. Secondly, local goal optimization and global goal optimization are performed through a self-learning algorithm based on converter steelmaking process data combined with heuristic algorithms, evolutionary algorithms and simulated annealing technology methods. The local goals are Target optimization: The self-learning algorithm based on the converter steelmaking process data determines the optimal state of the local target 1-M, and on the premise of ensuring that the local target 1-M reaches the required state, the optimization calculation model is used to optimize the process control parameters and find the optimal range of the process control parameter values; The global target optimization: By designing a global optimization mathematical model, the self-learning algorithm based on the converter steelmaking process data integrates the local targets 1-M into a global target, and based on the previous local optimization results, further optimizes the global control parameters and finds the optimal state of the global target. Finally, the self-learning algorithm based on the converter steelmaking process data outputs the optimal control parameters and smelting mode, and uses them to guide the next round of steelmaking operations, where M represents the number of local targets.

[0046] A data-driven converter steelmaking self-learning method comprises the following steps:

[0047] S1: Obtain raw material data, process control parameters and equipment status data;

[0048] S2: Set the automatic selection logic according to historical data and expert rules, load the machine learning model, load the target control curve, monitor the import results, and if the model is imported accurately and the status is normal, proceed to the next step, otherwise the alarm is transferred to manual processing;

[0049] S3: Start blowing, and the converter steelmaking self-learning system interacts with the L1 system based on data-driven;

[0050] S4: Online decision model raw materials, process control parameters, equipment status and current input information, calculates the smelting status and target curve deviation, and determines the converter smelting stage according to the deviation value;

[0051] S5: After smelting is completed, the converter information of the current furnace is managed and stored based on the principle of furnace number and time and space consistency;

[0052] S6: According to the on-site management requirements for converter smelting and the current furnace converter inspection and testing data, the endpoint hit rate, metal recovery rate, endpoint slag composition, dephosphorization rate, slag volume, smelting stability, endpoint oxygen level and material consumption converter smelting effect indicators are evaluated, and the weights of the adjustment indicators are adjusted. Then, the data perspective model is used to score and quantify the smelting results, and excellent furnaces are selected according to the scores to prepare for self-learning and further optimization of the smelting process;

[0053] S7: Based on the inspection and testing results of historical batches and the deviation of the calculation results of the optimization calculation model as the error function, the model parameters are automatically corrected through the self-learning optimization algorithm of gradient descent, least squares, statistical estimation, regression analysis and sliding average to minimize the error function, and the corrected online decision model parameters are saved to prepare for the next batch of production;

[0054] S8: According to the historical furnace smelting scores, adjust the smelting target curve, including the gas curve, audio curve, temperature curve and flame image curve of the flue gas analysis;

[0055] S9: Label samples as needed in the sample labeling module to provide sample support for subsequent training and self-learning.

[0056] In step S2: set automatic selection logic according to historical data and expert rules, load the machine learning model, load the target control curve, wherein the target control curve is determined according to actual production practice, and selects the curve with a higher furnace score as the target curve.

[0057] In step S4: the online decision model calculates the deviation between the smelting state and the target curve based on raw materials, process control parameters, equipment status and current input information, wherein the deviation types include: the comparison results of the gas concentration curve, audio curve, temperature curve, flame image and the corresponding characteristic curve of the target of the flue gas analysis, and the numerical deviation value and time information deviation value on the characteristic point and the target curve are calculated to determine the stage of converter smelting, wherein the characteristic points include the following: the gas concentration characteristic points of the flue gas analysis include curve inflection points, peak points and decline inflection points; the characteristic points of the audio curve include frequency change points and amplitude change points; the characteristic points of the flame image characteristic curve include brightness change points and color change points; the characteristic points of the temperature curve include temperature rise points, stable points and decline points. The specific steps for determining the converter smelting stage are as follows:

[0058] S4-1: Determine the key smelting stages and the corresponding target curves. The converter steelmaking process can be divided into the following stages: Oxygen blowing start stage: At this time, the carbon-oxygen reaction has not yet started, and the CO content in the flue gas is low; Carbon-oxygen reaction enhancement stage: CO begins to be generated in large quantities, the CO content in the flue gas increases significantly, and the target curve has the first inflection point. The target curve is relatively flat at this stage; Intense reaction stage: The CO content reaches a peak, the reaction is most intense, and the target curve at this stage should show high and stable values; Carbon-oxygen reaction weakening stage: CO generation gradually decreases, and the target curve gradually falls back; End stage: The reaction is basically over, the CO content is close to zero, and the target curve is stable;

[0059] S4-2: By analyzing the curve of gas content changing with time, temperature curve, audio signal and flame image curve and comparing them with the target curve, the numerical deviation and time deviation are calculated;

[0060] S4-3: By analyzing the deviation inference stage, the initial stage: the actual CO curve has not shown a significant rise, and the temperature curve is in a stable stage, so the system can be judged to be still in the initial stage of oxygen blowing; the beginning stage of carbon-oxygen reaction: the actual CO curve begins to rise significantly and is close to the first inflection point of the target curve, it can be judged that it has entered the carbon-oxygen reaction enhancement stage. If the CO curve rises too fast, it means that the reaction is too intense and the temperature in the furnace is high; the intense reaction stage: when the CO curve reaches or exceeds the peak of the target curve and remains stable for a period of time, it indicates that the system is in an intense reaction stage. If the actual value is higher than the target value, it is necessary to adjust the oxygen blowing amount or the bottom blowing intensity; the weakened reaction stage: when the CO curve begins to decline and approaches the descending section of the target curve, it can be judged that it has entered the weakened reaction stage; the ending stage: when the CO curve basically drops to zero, and the temperature and other indicators tend to be stable, the reaction is basically over;

[0061] S4-4: At each stage, according to the size of the deviation, a multivariate regression or neural network regression model is jointly established to quantitatively calculate the correction amount for feeding, oxygen blowing, and bottom blowing, thereby linking the L1 system to perform oxygen gun control, feeding control, and bottom blowing control.

[0062] In step S7: establish an optimization calculation model. Specifically, the i-th local optimization objective function It can be expressed as:

[0063]

[0064] Among them, f i () represents the performance representation function of the i-th local objective, represents the jth decision variable associated with the i-th local objective;

[0065] Furthermore, taking steelmaking cost as the object, a global optimization mathematical model is established to find the global cost optimal parameters and steelmaking mode. The global optimization objective function G can be expressed as:

[0066]

[0067] Among them, ω i and are the importance weight and steelmaking cost corresponding to the i-th local objective respectively.

[0068] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A data-driven converter steelmaking self-learning system, characterized in that: The system includes: an information acquisition module, a sample annotation module, a sample management library, a data preprocessing module, an online decision module, a data perspective module and an optimization module. The information acquisition module is used to acquire raw material and auxiliary material data, process control parameters and equipment status data; the sample annotation module is used to classify and annotate the acquired data according to the different requirements of the optimization calculation model, the online decision model and the data perspective model; the sample management library is used to store all acquired data; the data preprocessing module is used to preprocess the acquired data, including data cleaning, missing value filling, data standardization and feature engineering; the online decision module is used to judge and calculate and adjust the smelting state during the smelting process; the data perspective module is used to score and quantify the smelting results; the optimization module is used to perform optimal mode smelting mode, indicator correlation factor analysis and model parameter self-optimization through historical data.

2. The data-driven converter steelmaking self-learning system according to claim 1 is characterized in that: The information acquisition module includes a raw material data acquisition unit, a process control parameter acquisition unit and an equipment status data acquisition unit. The raw material data acquisition unit is used to collect raw material data through the iron plant and the scrap steel workshop, and record it in the EMS system and then output it. The raw material data includes molten iron composition, molten iron temperature, slag content of molten iron, molten iron weight, proportion type of scrap steel, weight of scrap steel, temperature of scrap steel, type and composition of available materials; the process control parameter acquisition unit records the process control parameters collected by PLC into the L1 system through the staff and outputs it. The process control parameters include the composition requirements, temperature requirements, end slag requirements, weight requirements, flame information and sounds generated during the steelmaking process for obtaining the molten steel target; the equipment status data acquisition unit is recorded and input by on-site operators or the secondary system, and the equipment status data includes the furnace service, empty furnace time, lining information and bottom blowing status of the converter.

3. The data-driven converter steelmaking self-learning system according to claim 1 is characterized in that: The online decision module judges and calculates and adjusts the smelting state during the smelting process by establishing an online decision model. The operation process is as follows: first, raw materials and process control parameters are obtained through the sample management library, and the obtained process control parameters and raw materials data are input into the preliminary processing stage of the online decision model to extract characteristic data of energy consumption, material consumption, equipment status, smelting control and time information related to steelmaking performance; Secondly, the extracted characteristic data is used to quantitatively analyze the overall performance of steelmaking, and the quantitative results of the safety effect of converter steelmaking are obtained based on raw materials, process control parameters and current inputs. Then, deep neural network technology is introduced to fit the relationship function between raw materials, process control parameters and steelmaking performance, and the estimated indicators are obtained through performance deduction based on the current real-time sensor information. The estimated indicators are compared with the acceptable indicators, and the input judgment and correction model is obtained by analyzing the difference between the estimated indicators and the acceptable indicators. Finally, the final process control parameters and raw material inputs are obtained.

4. The data-driven converter steelmaking self-learning system according to claim 1 is characterized in that: The data perspective module scores and quantifies the smelting results by establishing a data perspective model, analyzes and obtains converter smelting effect indicators such as endpoint hit rate, metal recovery rate, endpoint slag composition, dephosphorization rate, slag volume, smelting stability, endpoint oxygen level and material consumption, adjusts the weight of each indicator according to the production requirements of different enterprises, and uses regression analysis to explore the relationship between endpoint indicators, raw and auxiliary material data, process control parameters and equipment status, so as to quantitatively score the smelting results.

5. The data-driven converter steelmaking self-learning system according to claim 1 is characterized in that: The optimization module divides the optimization process into two stages: local optimization and global optimization by establishing an optimization calculation model, which is used to perform optimal mode smelting, index correlation factor analysis and model parameter self-optimization through historical data. The self-learning optimization process is as follows: First, the process control parameters are obtained through the sample management library, and after data preprocessing, the processed data are combined with the quantitative results of the safety effect of converter steelmaking in the sample management library, and the smelting process is further converted into measurable quantitative indicators through cost analysis, including local goals and global goals. Secondly, local goal optimization and global goal optimization are performed through a self-learning algorithm based on converter steelmaking process data combined with a heuristic algorithm, an evolutionary algorithm and a simulated annealing technology method. The local Target optimization: The self-learning algorithm based on the converter steelmaking process data determines the optimal state of the local target 1-M, and on the premise of ensuring that the local target 1-M reaches the required state, the optimization calculation model is used to optimize the process control parameters and find the optimal range of the process control parameter values; The global target optimization: By designing a global optimization mathematical model, the self-learning algorithm based on the converter steelmaking process data integrates the local targets 1-M into a global target, and based on the previous local optimization results, further optimizes the global control parameters and finds the optimal state of the global target. Finally, the self-learning algorithm based on the converter steelmaking process data outputs the optimal control parameters and smelting mode, and uses them to guide the next round of steelmaking operations, where M represents the number of local targets.

6. A data-driven converter steelmaking self-learning method, comprising the following steps: S1: Obtain raw material data, process control parameters and equipment status data; S2: Set the automatic selection logic according to historical data and expert rules, load the machine learning model, load the target control curve, monitor the import results, and if the model is imported accurately and the status is normal, proceed to the next step, otherwise the alarm is transferred to manual processing; S3: Start blowing, and the converter steelmaking self-learning system interacts with the L1 system based on data-driven; S4: Online decision model raw materials, process control parameters, equipment status and current input information, calculates the smelting status and target curve deviation, and determines the converter smelting stage according to the deviation value; S5: After smelting is completed, the converter information of the current furnace is managed and stored based on the principle of furnace number and time and space consistency; S6: According to the on-site management requirements for converter smelting and the current furnace converter inspection and testing data, the endpoint hit rate, metal recovery rate, endpoint slag composition, dephosphorization rate, slag volume, smelting stability, endpoint oxygen level and material consumption converter smelting effect indicators are evaluated, and the weights of the adjustment indicators are adjusted. Then, the data perspective model is used to score and quantify the smelting results, and excellent furnaces are selected according to the scores to prepare for self-learning and further optimization of the smelting process; S7: Based on the inspection and testing results of historical batches and the deviation of the calculation results of the optimization calculation model as the error function, the model parameters are automatically corrected through the self-learning optimization algorithm of gradient descent, least squares, statistical estimation, regression analysis and sliding average to minimize the error function, and the corrected online decision model parameters are saved to prepare for the next batch of production; S8: According to the historical furnace smelting scores, adjust the smelting target curve, including the gas curve, audio curve, temperature curve and flame image curve of the flue gas analysis; S9: Label samples as needed in the sample labeling module to provide sample support for subsequent training and self-learning.

7. The data-driven converter steelmaking self-learning method according to claim 6 is characterized in that: In step S2: set automatic selection logic according to historical data and expert rules, load the machine learning model, load the target control curve, wherein the target control curve is determined according to actual production practice, and selects the curve with a higher furnace score as the target curve.

8. The data-driven converter steelmaking self-learning method according to claim 6 is characterized in that: In step S4: the online decision model calculates the deviation between the smelting state and the target curve based on raw materials, process control parameters, equipment status and current input information, wherein the deviation types include: the comparison results of the gas concentration curve, audio curve, temperature curve, flame image and the corresponding characteristic curve of the target of the flue gas analysis, and the numerical deviation value and time information deviation value on the characteristic point and the target curve are calculated to determine the stage of converter smelting, wherein the characteristic points include the following: the gas concentration characteristic points of the flue gas analysis include the curve inflection point, peak point and decline inflection point; the characteristic points of the audio curve include the frequency change point and the amplitude change point; the characteristic points of the flame image characteristic curve include the brightness change point and the color change point; the characteristic points of the temperature curve include the temperature rise point, the stable point and the decline point. The specific steps for determining the converter smelting stage are as follows: S4-1: Determine the key smelting stages and the corresponding target curves. The converter steelmaking process can be divided into the following stages: Oxygen blowing start stage: At this time, the carbon-oxygen reaction has not yet started, and the CO content in the flue gas is low; Carbon-oxygen reaction enhancement stage: CO begins to be generated in large quantities, the CO content in the flue gas increases significantly, and the target curve has the first inflection point. The target curve is relatively flat at this stage; Intense reaction stage: The CO content reaches a peak, the reaction is most intense, and the target curve at this stage should show high and stable values; Carbon-oxygen reaction weakening stage: CO generation gradually decreases, and the target curve gradually falls back; End stage: The reaction is basically over, the CO content is close to zero, and the target curve is stable; S4-2: By analyzing the curve of gas content changing with time, temperature curve, audio signal and flame image curve and comparing them with the target curve, the numerical deviation and time deviation are calculated; S4-3: By analyzing the deviation inference stage, the initial stage: the actual CO curve has not shown a significant rise, and the temperature curve is in a stable stage, so the system can be judged to be still in the initial stage of oxygen blowing; the beginning stage of carbon-oxygen reaction: the actual CO curve begins to rise significantly and is close to the first inflection point of the target curve, it can be judged that it has entered the carbon-oxygen reaction enhancement stage. If the CO curve rises too fast, it means that the reaction is too intense and the temperature in the furnace is high; the intense reaction stage: when the CO curve reaches or exceeds the peak of the target curve and remains stable for a period of time, it indicates that the system is in an intense reaction stage. If the actual value is higher than the target value, it is necessary to adjust the oxygen blowing amount or the bottom blowing intensity; the weakened reaction stage: when the CO curve begins to decline and approaches the descending section of the target curve, it can be judged that it has entered the weakened reaction stage; the ending stage: when the CO curve basically drops to zero, and the temperature and other indicators tend to be stable, the reaction is basically over; S4-4: At each stage, according to the size of the deviation, a multivariate regression or neural network regression model is jointly established to quantitatively calculate the correction amount for feeding, oxygen blowing, and bottom blowing, thereby linking the L1 system to perform oxygen gun control, feeding control, and bottom blowing control.

9. The method for self-learning converter steelmaking based on data-driven according to claim 6, characterized in that: In step S7: establish an optimization calculation model. Specifically, the i-th local optimization objective function It can be expressed as: Among them, f i () represents the performance representation function of the i-th local objective, represents the jth decision variable associated with the i-th local objective; Furthermore, taking steelmaking cost as the object, a global optimization mathematical model is established to find the global cost optimal parameters and steelmaking mode. The global optimization objective function G can be expressed as: Among them, ω i and are the importance weight and steelmaking cost corresponding to the i-th local objective respectively.

Images