An electric furnace temperature segmentation intelligent control method based on large model smelting time series data

Abstract

This invention discloses a segmented intelligent control method for electric arc furnace (EAF) temperature based on large-scale smelting time-series data, relating to the field of intelligent control technology for EAF smelting. The method acquires the current furnace smelting time-series data and existing endpoint temperature prediction results, generates fixed-stage data results according to fixed time windows, and extracts stage power consumption increment, stage oxygen supply increment, stage furnace door opening duration, stage temperature change, and scrap steel addition interval. Based on historical normal stage reference results, it calculates stage consistency deviation results. In the second fixed stage, if power consumption, oxygen supply, and temperature rise are insufficient, and the furnace door opening is too long and the impact of material addition has not ended, a melting hysteresis trigger result is generated. The relevant results are input into a smelting time-series stage identification model to obtain the actual stage identification result. When identified as a melting hysteresis stage, stage reassignment is performed, generating electrode power correction, melting hold duration correction, and oxygen supply limitation results. Based on feedback from the next sampling time, the method performs retention or limit updates.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for electric furnace smelting, and more specifically, to a segmented intelligent control method for electric furnace temperature based on large-scale smelting time-series data. Background Technology

[0002] Current electric arc furnace (EAF) steelmaking processes typically utilize smelting time-series data for endpoint temperature prediction. This data can include scrap addition, molten iron temperature, power consumption, oxygen supply, furnace door status, and temperature measurement results. To improve the computability of endpoint temperature prediction, existing technologies often divide a heat cycle into multiple stages according to a fixed time window. Features such as power consumption increments, oxygen supply increments, and temperature changes within each stage are then extracted and combined with machine learning or neural network models to output the endpoint temperature prediction. This type of method provides a data foundation for determining tapping temperature and controlling the smelting process.

[0003] The existing technology has the following shortcomings: The electric arc furnace smelting process is greatly affected by factors such as scrap size, scrap amount, initial iron temperature, furnace door opening time, and charging rhythm. The stages obtained by dividing the time window into fixed stages may not be consistent with the actual smelting state in the furnace. Especially in furnaces with delayed scrap melting, although the current sampling time has entered the second fixed stage according to time, the metal charge in the furnace may not have been fully melted. If the existing algorithm directly uses the data of this stage as the data of the oxidation reaction stage, it is easy to lead to inaccurate basis for oxygen supply control, power control, or temperature control. Existing endpoint temperature prediction methods usually focus on whether the predicted value is close to the target endpoint temperature, lack judgment on the accuracy of fixed stage division, and also lack a processing mechanism for real stage re-identification, stage re-assignment, and segmented control correction when stage mismatch occurs. There is a need for a segmented intelligent control method for electric furnace temperature that can identify mismatches in fixed stages by combining smelting time sequence data and generate control correction results based on the actual stages.

[0004] To address the above problems, this invention proposes a solution. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an intelligent segmented temperature control method for electric furnaces based on large-scale smelting time-series data, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A segmented intelligent control method for electric furnace temperature based on large-scale smelting time-series data includes the following steps; Step S1: Obtain the smelting start time, current sampling time, scrap steel addition time, cumulative power consumption, cumulative oxygen supply, furnace door opening status, most recent temperature measurement, target endpoint temperature, and current endpoint temperature prediction for the current furnace batch of the target electric furnace. Based on the time difference between the current sampling time and the smelting start time, divide the smelting time series data into multiple fixed stage data results, and generate stage power consumption increment results, stage oxygen supply increment results, stage furnace door opening duration results, stage temperature measurement change results, and scrap steel addition interval results. Step S2: Calculate the stage consistency deviation result of the current fixed stage based on the historical normal stage reference results. When the current fixed stage is the second fixed stage, and the stage consistency deviation result exceeds the stage consistency threshold result, the stage power consumption increment result, the stage oxygen supply increment result, and the stage temperature measurement change result are all lower than the corresponding lower limit boundary result, the stage furnace door opening time result is higher than the corresponding upper limit boundary result, and the scrap steel addition interval result is less than the feeding influence time threshold result, a melting hysteresis trigger result is generated. Step S3: Input the melting hysteresis triggering result, the current fixed stage data result and the existing endpoint temperature prediction result into the smelting time sequence stage identification model to generate the true stage identification result. When the true stage identification result is the melting hysteresis stage, reassign the current sampled data from the second fixed stage data result to the melting hysteresis stage data result, and generate the electrode power correction result, melting hold time correction result and oxygen supply limitation result. Step S4: Obtain the temperature measurement result and the existing endpoint temperature prediction result at the next sampling time. Recalculate the stage consistency deviation result and the endpoint temperature deviation result. When both the stage consistency deviation result and the endpoint temperature deviation result are improved, retain the melting hysteresis reassignment mark result and the segmented control correction result. Otherwise, perform a single review and limit the update of the segmented control correction result.

[0007] In a preferred embodiment, step S1 includes the following: The smelting elapsed time result is generated based on the time difference between the current sampling time and the smelting start time. The smelting time sequence data of the current furnace is divided into fixed time windows according to the smelting elapsed time result to obtain multiple fixed stage data results. The fixed-stage data results include the first fixed-stage data results, the second fixed-stage data results, the third fixed-stage data results, and the fourth fixed-stage data results; Extract the stage increment from the data results of each fixed stage to form the stage verification data results; The phase verification data results include phase power consumption increment results, phase oxygen supply increment results, phase furnace door opening time results, phase temperature measurement change results, and scrap steel addition interval results; The incremental power consumption result of a stage is obtained by calculating the difference between the cumulative power consumption at the end of the fixed stage and the cumulative power consumption at the beginning of the fixed stage. The incremental oxygen supply in a phase is obtained by calculating the difference between the cumulative oxygen supply at the end of the fixed phase and the cumulative oxygen supply at the beginning of the fixed phase. The stage furnace door opening time result is obtained by accumulating the duration during which the furnace door is in the open state within the fixed stage; The phased temperature change results are obtained by calculating the difference between the available temperature measurement results at the end of the fixed phase and the available temperature measurement results at the beginning of the fixed phase. The scrap steel addition interval is obtained by calculating the time difference between the current sampling time and the most recent scrap steel addition time; A reference result for the historical normal furnace phase is constructed based on the set of historical normal furnace batches; The historical normal furnace set is processed according to the same fixed time window division method as the current furnace set, and the data results of the first fixed stage, the second fixed stage, the third fixed stage, and the fourth fixed stage are obtained respectively. For each fixed stage, the increment of power consumption, increment of oxygen supply, furnace door opening time and temperature change are statistically analyzed to generate reference power consumption increment results, reference oxygen supply increment results, reference furnace door opening time results and reference temperature change results.

[0008] In a preferred embodiment, step S2 includes the following: The phase consistency deviation result is calculated based on the current fixed phase data result, phase verification data result, historical normal phase reference result, and reference boundary result. When the phase consistency deviation result does not exceed the phase consistency threshold result, the fixed phase division result is retained. When the stage consistency deviation exceeds the stage consistency threshold, it is further determined whether the deviation direction meets the melting hysteresis condition. The melting lag conditions are as follows: the current fixed stage is the second fixed stage, and the stage power consumption increment is lower than the lower limit boundary of the reference power consumption increment of the second fixed stage, the stage oxygen supply increment is lower than the lower limit boundary of the reference oxygen supply increment of the second fixed stage, the stage temperature change is lower than the lower limit boundary of the reference temperature change of the second fixed stage, the stage furnace door opening time is higher than the upper limit boundary of the reference furnace door opening time of the second fixed stage, and the scrap steel addition interval is less than the feeding influence time threshold. When all of the above conditions are met, a melting hysteresis trigger result is generated.

[0009] In a preferred embodiment, step S3 includes the following: When generating melting hysteresis triggering results, the current fixed stage data results, stage verification data results, historical normal stage reference results, stage consistency deviation results, the real stage identification results at the previous sampling time, the current endpoint temperature prediction results, the target endpoint temperature results, and the melting hysteresis triggering results are input into the smelting time sequence stage identification model trained with historical furnace stage samples to generate candidate real stage matching scores. The real stage identification result is determined based on the candidate real stage matching score and stage continuity constraints; When the candidate real stage with the highest matching score does not meet the stage continuity constraint, the candidate real stage that meets the stage continuity constraint and has the highest matching score is selected as the real stage identification result. When the actual stage identification result is the melting hysteresis stage, the current sampled data is reassigned from the second fixed stage data result to the melting hysteresis stage data result, and a melting hysteresis reassignment label result is generated. Based on the melting hysteresis reassignment labeling results, endpoint temperature deviation results, and control boundary results, segmented control correction results are generated. The segmented control correction results include electrode power correction results, melting hold time correction results, and oxygen supply limitation results.

[0010] In a preferred embodiment, step S4 includes the following: After executing the segmented control correction results, obtain the temperature measurement results, cumulative power consumption results, cumulative oxygen supply results, furnace door opening status results, existing endpoint temperature prediction results, and stage verification data results at the next sampling time. Recalculate the phase consistency deviation and endpoint temperature deviation results based on the data from the next sampling time. Compare the phase consistency deviation result at the next sampling time with the phase consistency deviation result before execution control, and compare the endpoint temperature deviation result at the next sampling time with the endpoint temperature deviation result before execution control. When the phase consistency deviation result decreases and the absolute value of the endpoint temperature deviation result decreases, a reassignment retention result is generated. When the stage consistency deviation result does not decrease or the absolute value of the endpoint temperature deviation result does not decrease, a reassignment verification result is generated, and the true stage identification result and segmented control correction result are corrected in a single iteration. The single-iteration correction involves merging the stage verification data results at the current sampling time with the stage verification data results at the next sampling time into a short-term state change result, re-inputting it into the smelting time sequence stage identification model, obtaining the updated candidate true stage matching score result, and reconfirming the true stage identification result based on the updated candidate true stage matching score result and stage continuity constraints.

[0011] The technical effects and advantages of the present invention, which describes a segmented intelligent control method for electric furnace temperature based on large-scale smelting time-series data, are as follows: By adding a stage consistency deviation judgment after dividing the fixed time window into stages, the electric arc furnace (EAF) smelting temperature control no longer directly relies on the fixed stage results obtained by time division. Instead, it first judges whether the current fixed stage is consistent with the actual furnace conditions. In the case of insufficient power consumption, oxygen supply, and temperature measurement changes, excessive furnace door opening, and incomplete scrap steel addition effects in the second fixed stage, the system can identify melting hysteresis trigger results and reassign the corresponding data from the second fixed stage to the melting hysteresis stage, thereby reducing the situation where data of incomplete melting is mistakenly used as data of the oxidation reaction stage. The system generates the actual stage identification results through the smelting sequence stage identification model, and prioritizes the generation of electrode power correction results and melting hold time correction results in the melting hysteresis stage, while forming oxygen supply limitation results, so that the control actions match the current furnace condition stage. Feedback verification is performed through the stage consistency deviation results and endpoint temperature deviation results at the next sampling time. The system can retain, review, or limit the update of melting hysteresis reassignment marking results and segmented control correction results, thereby improving the stability and pertinence of the segmented control process of the EAF temperature. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the structure of an electric furnace temperature segmented intelligent control method based on large model smelting time series data according to the present invention.

[0013] Figure 2 This is a schematic diagram illustrating the fixed-stage mismatch identification and real-stage reassignment of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] Example Please see Figures 1-2 As shown, this invention discloses a segmented intelligent control method for electric furnace temperature based on large-scale smelting time-series data, comprising the following steps: Step S1: Obtain the smelting start time, current sampling time, scrap steel addition time, cumulative power consumption, cumulative oxygen supply, furnace door opening status, most recent temperature measurement, target endpoint temperature, and current endpoint temperature prediction for the current furnace batch of the target electric furnace. Based on the time difference between the current sampling time and the smelting start time, divide the smelting time series data into multiple fixed stage data results, and generate stage power consumption increment results, stage oxygen supply increment results, stage furnace door opening duration results, stage temperature measurement change results, and scrap steel addition interval results. Step S2: Calculate the stage consistency deviation result of the current fixed stage based on the historical normal stage reference results. When the current fixed stage is the second fixed stage, and the stage consistency deviation result exceeds the stage consistency threshold result, the stage power consumption increment result, the stage oxygen supply increment result, and the stage temperature measurement change result are all lower than the corresponding lower limit boundary result, the stage furnace door opening time result is higher than the corresponding upper limit boundary result, and the scrap steel addition interval result is less than the feeding influence time threshold result, a melting hysteresis trigger result is generated. Step S3: Input the melting hysteresis triggering result, the current fixed stage data result and the existing endpoint temperature prediction result into the smelting time sequence stage identification model to generate the true stage identification result. When the true stage identification result is the melting hysteresis stage, reassign the current sampled data from the second fixed stage data result to the melting hysteresis stage data result, and generate the electrode power correction result, melting hold time correction result and oxygen supply limitation result. Step S4: Obtain the temperature measurement result and the existing endpoint temperature prediction result at the next sampling time. Recalculate the stage consistency deviation result and the endpoint temperature deviation result. When both the stage consistency deviation result and the endpoint temperature deviation result are improved, retain the melting hysteresis reassignment mark result and the segmented control correction result. Otherwise, perform a single review and limit the update of the segmented control correction result.

[0016] In step S1, the following data are acquired for the current furnace run: smelting start time, current sampling time, scrap steel addition time, cumulative power consumption, cumulative oxygen supply, furnace door opening status, most recent temperature measurement, target endpoint temperature, and current endpoint temperature prediction. Based on the time difference between the current sampling time and the smelting start time, the smelting time series data is divided into multiple fixed-stage data results, and stage power consumption increment results, stage oxygen supply increment results, stage furnace door opening duration results, stage temperature change results, and scrap steel addition interval results are generated. Specific content includes: Acquire the following information for the target electric furnace: smelting start time, current sampling time, initial iron temperature, scrap steel addition amount, scrap steel addition time, cumulative power consumption, cumulative oxygen supply, furnace door opening status, most recent temperature measurement, target endpoint temperature, and current endpoint temperature prediction.

[0017] Based on the time difference between the current sampling time and the smelting start time, the smelting elapsed time result is generated; Based on the smelting time results, the smelting time sequence data of the current furnace is divided into fixed time windows to obtain the data results of the first fixed stage, the second fixed stage, the third fixed stage, and the fourth fixed stage.

[0018] It should be noted that the fixed time window division in step S1 is used to retain the basic stage construction method in the existing endpoint temperature prediction algorithm, rather than directly assuming that the current furnace has actually entered the corresponding smelting stage.

[0019] This step first generates basic stage data results according to a fixed time window, so that subsequent steps can determine whether the fixed stage division results are consistent with the current actual state inside the furnace. It can correct potential stage mismatches caused by fixed time window stage division without disrupting the existing endpoint temperature prediction process.

[0020] For example, data with a smelting time of 0 to 8 minutes can be classified into the first fixed stage data result, data with a smelting time of 9 to 16 minutes can be classified into the second fixed stage data result, data with a smelting time of 17 to 24 minutes can be classified into the third fixed stage data result, and data with a smelting time of 25 minutes to the end of smelting can be classified into the fourth fixed stage data result. The first fixing stage is usually used to form basic data related to melting; the second fixing stage is usually used to form basic data related to oxidation reaction; the third fixing stage is usually used to form basic data related to slag formation adjustment; and the fourth fixing stage is usually used to form basic data related to temperature rise adjustment.

[0021] For example, when the current sampling time is eleven minutes away from the start of smelting, according to the fixed time window division rule, the sampling data is classified into the second fixed stage data result, which usually corresponds to the stage where the oxidation reaction is more active; In furnaces with excessive scrap steel addition, large scrap steel block size, low initial iron temperature, or prolonged furnace door opening time, the furnace charge at that sampling time may not have been fully melted. This step only generates the second fixed stage data result and does not directly generate the true result of the oxidation reaction stage. It is necessary to determine whether the second fixed stage data result can be used as oxidation reaction stage data through stage accuracy judgment.

[0022] Furthermore, the data results of each fixed stage are subjected to stage increment extraction to form stage verification data results; The phase verification data results include phase power consumption increment results, phase oxygen supply increment results, phase furnace door opening time results, phase temperature measurement change results, and scrap steel addition interval results; The incremental power consumption result of a stage is obtained by calculating the difference between the cumulative power consumption at the end of the fixed stage and the cumulative power consumption at the beginning of the fixed stage. The incremental oxygen supply in a phase is obtained by calculating the difference between the cumulative oxygen supply at the end of the fixed phase and the cumulative oxygen supply at the beginning of the fixed phase. The stage furnace door opening time result is obtained by accumulating the duration during which the furnace door is in the open state within the fixed stage; The phased temperature change results are obtained by calculating the difference between the available temperature measurement results at the end of the fixed phase and the available temperature measurement results at the beginning of the fixed phase. The scrap steel addition interval is obtained by calculating the time difference between the current sampling time and the most recent scrap steel addition time; It should be noted that the phase verification data results only retain the data required for subsequent phase accuracy judgment and real phase re-identification; Data that does not directly participate in subsequent stage judgments or segmented control corrections can be saved as original smelting records, rather than as a necessary field in the stage verification data results. Furthermore, the results of phased temperature changes can be generated either directly from temperature measurement data or by interpolating temperature measurement data. When there are temperature measurement results at the beginning and end of a certain fixed period, the direct temperature measurement data should be used first to generate the temperature change results for that period. When there is no complete direct temperature measurement data within a fixed period, the most recent temperature measurement result before and after the fixed period is used for interpolation to obtain the interpolated temperature measurement result at the beginning of the period and the interpolated temperature measurement result at the end of the period. Then, the temperature change result of the period is generated, and the temperature interpolation mark result is generated simultaneously. The temperature interpolation marking results are used to reduce the weight of the judgment items corresponding to temperature changes in subsequent stages of accuracy judgment, so as to avoid misjudgment in stages due to the uncertainty of interpolation data.

[0023] Furthermore, data reliability processing was performed on the phased power consumption increment results, phased oxygen supply increment results, phased furnace door opening duration results, and phased temperature measurement change results; If a variable experiences data collection interruptions, communication interruptions, or missing device records within a fixed period, a corresponding data missing marker result will be generated. If the value of a certain variable exceeds the theoretical recording range of the device, and adjacent variables do not change in tandem, an abnormal acquisition marker result will be generated. If a variable deviates from the historical normal range, but changes in power consumption, oxygen supply, furnace door opening, or temperature measurement are coordinated, it will not be deleted as a data collection error, but will be retained as furnace condition change data that can be used for stage judgment.

[0024] Furthermore, construct reference results for normal historical periods; The historical normal stage reference results are derived from a set of historical normal furnace batches; The historical normal furnace set refers to the historical furnaces where the final temperature is within the allowable range of the target final temperature, no key equipment failures are recorded, no temperature measurement abnormalities are recorded, no long-term communication interruptions are recorded, and the total smelting time is within the allowable range of the corresponding process for the target steel grade. The historical normal furnace set is processed according to the same fixed time window division method as the current furnace set, and the data results of the first fixed stage, the second fixed stage, the third fixed stage, and the fourth fixed stage are obtained respectively. It should be noted that the reference results for the historical normal stage are not arbitrary empirical values, but are obtained from the statistics of the historical normal furnace set; For each fixed stage, the power consumption increment, oxygen supply increment, furnace door opening time and temperature change are statistically analyzed to generate reference power consumption increment results, reference oxygen supply increment results, reference furnace door opening time results and reference temperature change results. The above reference results are used to determine whether the current fixed-stage data results are close to the historical normal state.

[0025] For example, the reference power consumption increment result can be the median value of the power consumption increment in the same fixed period of a historical normal furnace batch; The reference oxygen supply increment result can be the median value of the oxygen supply increment during the same fixed period in the historical normal furnace batches; The reference temperature change results can be the median value of the temperature change during the same fixed period in historical normal furnace batches; The median value is used because historical data from electric arc furnace sites may include individual furnaces with minor disturbances, and the median value can reduce the impact of extreme values ​​on the reference results.

[0026] Furthermore, in order to ensure the feasibility of subsequent threshold judgments, corresponding reference boundary results are generated for the historical normal stage reference results; For variables such as stage power consumption increment, stage oxygen supply increment, and stage temperature measurement change used to determine whether the normal progress level of the stage has been reached, generate lower limit boundary results; For variables such as the duration of furnace door opening in each stage, which are used to assess the risk of heat loss or melting delay, upper limit boundary results are generated.

[0027] For example, the lower boundary result can be taken as the 25th quartile result of the corresponding variable in the same fixed stage of a historical normal furnace; the upper boundary result can be taken as the 75th quartile result of the corresponding variable in the same fixed stage of a historical normal furnace. When the number of historical normal furnace cycles is insufficient to stably calculate the quantile results, the mean of the corresponding variable in the same fixed period of historical normal furnace cycles minus one standard deviation can be taken as the lower boundary result, and the mean plus one standard deviation can be taken as the upper boundary result.

[0028] Furthermore, obtain the existing endpoint temperature prediction results; The existing endpoint temperature prediction results can be output by a trained endpoint temperature prediction model. The inputs of the endpoint temperature prediction model include the initial temperature of molten iron, the amount of scrap steel added, the results of the stage power consumption increment, the results of the stage oxygen supply increment, the smelting time, and the most recent temperature measurement results. It should be noted that the existing endpoint temperature prediction results are not directly used as control commands in this embodiment, but rather as the basis for subsequent determination of the direction of temperature deviation.

[0029] This step utilizes existing prediction models to obtain a predictive basis for whether the current furnace's endpoint temperature may be too high or too low. However, it does not directly determine oxygen supply, electrode power, or holding time based on this predictive basis. It obtains fixed-stage data results, stage verification data results, historical normal stage reference results, reference boundary results, temperature measurement interpolation marking results, data missing marking results, and current endpoint temperature prediction results. The fixed time window division results are used as subsequent verification objects, thereby solving the problem that existing algorithms directly rely on time windows in the stage data construction stage and lack a basis for subsequent accuracy judgment. This step retains the data basis of existing endpoint temperature prediction algorithms and provides a unified data source and comparison object for calculating the stage consistency deviation results in step S2, enabling subsequent judgment on whether mismatch has occurred in the fixed stage division.

[0030] In step S2, the stage consistency deviation result of the current fixed stage is calculated based on the historical normal stage reference results. When the current fixed stage is the second fixed stage, and the stage consistency deviation result exceeds the stage consistency threshold result, the stage power consumption increment result, the stage oxygen supply increment result, and the stage temperature measurement change result are all lower than the corresponding lower limit boundary result, the stage furnace door opening time result is higher than the corresponding upper limit boundary result, and the scrap steel addition interval result is less than the feeding influence time threshold result, a melting hysteresis trigger result is generated, the specific content of which includes: Based on the current fixed-stage data results, stage verification data results, historical normal-stage reference results, and reference boundary results, calculate the stage consistency deviation results for the current fixed stage; The phase consistency deviation result is used to determine whether the current second fixed phase data result has the characteristics of a normal oxidation reaction phase. It is not a single variable anomaly judgment result, but a comprehensive deviation result determined by the power consumption deviation sub-result, oxygen supply deviation sub-result, temperature measurement deviation sub-result, and furnace door deviation sub-result. The sum of the power consumption verification coefficient, oxygen supply verification coefficient, temperature measurement verification coefficient and furnace door verification coefficient is one. The historical normal second fixed stage reference result is obtained from the statistics of historical furnaces with qualified endpoint temperature, no key equipment failure, no abnormal temperature measurement and no long-term communication interruption. The corresponding lower boundary results and the corresponding upper boundary results are determined by the quantile results or mean standard deviation results of the historical normal second fixed stage reference results; The stage consistency deviation result is used to determine whether the second fixed stage obtained by dividing the fixed time window is consistent with the actual furnace condition stage, rather than simply determining whether the collected data is abnormal.

[0031] The accuracy of the current fixed phase division is determined by comparing the phase consistency deviation result with the phase consistency threshold result. When the phase consistency deviation result does not exceed the phase consistency threshold result, the fixed phase division result is retained. When the stage consistency deviation exceeds the stage consistency threshold, it is further determined whether the deviation direction meets the melting hysteresis condition. When the current fixed stage is the second fixed stage, and the results of the stage power consumption increment, the stage oxygen supply increment, and the stage temperature change are all lower than the lower limit boundary results corresponding to the second fixed stage, and the results of the stage furnace door opening time are higher than the upper limit boundary results corresponding to the second fixed stage, and the results of the scrap steel addition interval are less than the feeding influence time threshold results, a melting hysteresis trigger result is generated. It should be noted that the core of this step is not simply to determine whether the data is abnormal, but to determine whether the current fixed stage obtained by dividing the data into fixed time windows is consistent with the actual furnace condition stage.

[0032] In the existing endpoint temperature prediction algorithm, as long as the current sampled data is within the second fixed time window, it will enter the second fixed stage data result. However, this embodiment further determines whether the second fixed stage data result really has the characteristics of power consumption, oxygen supply, temperature rise and furnace door status that the oxidation reaction stage should have. If none of these characteristics reach the reference level of the second fixed stage in historical normality, it indicates that there is a risk of inaccuracy in the fixed time window stage division.

[0033] The phase consistency deviation result is expressed by the following calculation relationship: the phase consistency deviation result is equal to the sum of the following: the power consumption deviation sub-result multiplied by the power consumption verification coefficient, the oxygen supply deviation sub-result multiplied by the oxygen supply verification coefficient, the temperature measurement deviation sub-result multiplied by the temperature measurement verification coefficient, and the furnace door deviation sub-result multiplied by the furnace door verification coefficient. Among them, the power consumption deviation sub-result is obtained by dividing the absolute value of the difference between the stage power consumption increment result of the current fixed stage and the reference power consumption increment result of the corresponding fixed stage by the reference power consumption increment result; The oxygen supply deviation sub-result is obtained by dividing the absolute value of the difference between the current fixed stage oxygen supply increment result and the corresponding fixed stage reference oxygen supply increment result by the reference oxygen supply increment result. The temperature measurement deviation result is obtained by dividing the absolute value of the difference between the temperature change result of the current fixed stage and the reference temperature change result of the corresponding fixed stage by the absolute value of the reference temperature change result plus a stable constant. The furnace door deviation result is obtained by dividing the absolute value of the difference between the furnace door opening time result of the current fixed stage and the reference furnace door opening time result of the corresponding fixed stage by the reference furnace door opening time result plus a stable constant. The stability constant is a positive number used to avoid the denominator being zero. It can be set to a value smaller than the smallest effective order of magnitude of each reference variable, depending on the sampling precision. The sum of the power consumption verification coefficient, oxygen supply verification coefficient, temperature measurement verification coefficient, and furnace door verification coefficient is one, which is used to represent the relative role of different variables in the stage consistency judgment; It should be noted that each verification coefficient can be dynamically determined based on the current data reliability of the variable.

[0034] When the temperature interpolation mark exists, reduce the temperature verification coefficient and allocate the reduced weight to the power consumption verification coefficient and oxygen supply verification coefficient. When missing oxygen supply data is present, reduce the oxygen supply verification coefficient and increase the power consumption verification coefficient and furnace door verification coefficient. When multiple data channels are complete, the power consumption verification coefficient, oxygen supply verification coefficient, temperature measurement verification coefficient, and furnace door verification coefficient can be set to the same weight or determined based on historical verification results.

[0035] Furthermore, the phase consistency threshold result is generated based on the historical set of normal furnace cycles; For each furnace in the same fixed stage within the set of historical normal furnaces, the corresponding historical stage consistency deviation results are generated according to the above calculation relationship, forming a set of historical stage consistency deviations. When the number of historical normal furnace cycles reaches the preset sample size, the high quantile result of the historical stage consistency deviation set is used as the stage consistency threshold result for that fixed stage. When the number of historical normal furnace cycles does not reach the preset sample size, the mean of the historical stage consistency deviation set plus a standard deviation is used as the stage consistency threshold result for that fixed stage. For example, the preset sample size can be set to no less than thirty furnaces, and the higher quantile result can be the ninetieth place result; If there is insufficient historical data for the new production line, an initial threshold result can be formed by using process experience samples of the same steel grade or furnace type, and then updated with actual production data.

[0036] Furthermore, for the second fixed stage, all data from the second fixed stage in the historical normal furnace batches are selected first, and the consistency deviation results of the historical second fixed stage are calculated respectively to obtain the set of consistency deviations of the historical second fixed stage. Then, based on this set, a second-stage consistency threshold result is generated; When the current furnace is in the second fixed stage, if the stage consistency deviation result is greater than the second stage consistency threshold result, it indicates that there is a significant deviation between the current furnace's second fixed stage and the historical normal second fixed stage. However, it cannot be immediately identified as melting lag. It is still necessary to determine whether the deviation direction conforms to the melting lag mechanism. For example, when the consistency deviation result of the current furnace at the eleventh minute is greater than the consistency threshold result of the second stage, it can only indicate that the second fixed stage data corresponding to the sampling time deviates from the historical normal second fixed stage data. It is still necessary to determine whether it simultaneously meets the conditions of insufficient power consumption, insufficient oxygen supply, insufficient temperature rise, excessive furnace door opening time, and the effect of feeding not ending.

[0037] Furthermore, the melting hysteresis condition is determined jointly based on the directional deviations of multiple variables, specifically: First, the current fixed phase must be the second fixed phase; Second, if the current stage power consumption increment is lower than the lower limit of the reference power consumption increment in the second fixed stage, it indicates that the effective heat input level in the current stage is insufficient. Third, the current oxygen supply increment result is lower than the lower limit boundary result of the reference oxygen supply increment result in the second fixed stage, indicating that the current stage has not yet entered the normal oxidation reaction oxygen supply rhythm; Fourth, if the temperature change result in the current stage is lower than the lower limit boundary of the reference temperature change result in the second fixed stage, it means that the temperature rise in the current stage has not reached the normal level of the second fixed stage. Fifth, if the current furnace door opening time is higher than the upper limit of the reference furnace door opening time in the second fixed stage, it indicates that there may be heat loss or feeding disturbance caused by furnace door opening in the current stage. Sixth, if the scrap steel addition interval is less than the feeding influence time threshold, it means that the current sampling time is still within the influence time range after the most recent scrap steel addition. When all of the above conditions are met, a melting hysteresis trigger result is generated; It should be noted that the threshold for the duration of the impact of the feeding is also determined by the set of historical normal heats. Specifically, the time required for the temperature change to recover to the lower limit of the corresponding fixed stage reference temperature change after each addition of scrap steel in the historical normal heats is counted to form the set of historical durations of the impact of the feeding. The higher quantile of this set is used as the threshold result for the duration of the feeding effect; If historical data is insufficient, production experience values ​​of the same furnace type, the same scrap steel structure, or adjacent steel grades can be used as the initial threshold for the duration of material feeding influence, and updated after subsequent effective furnace cycles are added. For example, the higher quantile result of the historical duration of the added fuel effect can be the 75th quantile result; This result represents the upper reference value of the time required for most normal heat cycles to recover to the normal temperature rise range after the addition of scrap steel, and is used to determine whether the current sampling time is still affected by the most recent feeding.

[0038] Furthermore, when the stage consistency deviation result exceeds the stage consistency threshold result, but the melting hysteresis condition is not fully met, no melting hysteresis trigger result is generated. Instead, other re-identification trigger results or data verification results are generated based on the deviation direction. If the oxygen supply increment result is higher than the upper limit of the reference oxygen supply increment result, and the temperature change result is higher than the upper limit of the reference temperature change result, a strong oxidation reaction trigger result can be generated. If only a single temperature measurement result deviates, while power consumption, oxygen supply, and furnace door opening status remain unchanged, a temperature measurement data verification result is generated, and the actual stage re-identification is not directly performed.

[0039] Through step S2, we obtain accurate results for the fixed stage, melting hysteresis triggering results, other re-identification triggering results, or data verification results; If accurate results are obtained for a fixed stage, the current furnace batch will continue to use the fixed stage data results from step S1 and the existing endpoint temperature prediction results. If a melting hysteresis trigger result is obtained, the current fixed stage data result, stage verification data result, historical normal stage reference result, existing endpoint temperature prediction result, and melting hysteresis trigger result are input into step S3 to perform real stage re-identification and segmented control correction.

[0040] This step compares the current fixed-stage data results with historical normal-stage reference results. When the current fixed stage is the second fixed stage and the power consumption, oxygen supply, temperature changes, furnace door opening duration, and scrap steel addition interval all meet the melting hysteresis condition, a melting hysteresis trigger result is generated. This solves the problem that existing fixed-time-window stage division algorithms only divide stages based on the smelting time and cannot determine whether the second fixed stage has truly entered the oxidation reaction stage. This step uses stage consistency deviation results and multiple directional conditions to jointly determine whether there is a mismatch risk in the fixed stage, improving the pertinence and reliability of melting hysteresis identification. The melting hysteresis trigger result output in step S2 is further used as the trigger condition for step S3 to re-identify the true stage and correct the control.

[0041] In step S3, the melting hysteresis triggering result, the current fixed stage data result, and the existing endpoint temperature prediction result are input into the smelting time-series stage identification model to generate the true stage identification result. When the true stage identification result is the melting hysteresis stage, the current sampled data is reassigned from the second fixed stage data result to the melting hysteresis stage data result, and electrode power correction result, melting hold time correction result, and oxygen supply limitation result are generated. The specific contents include: When generating melting hysteresis triggering results, the current fixed stage data results, stage verification data results, historical normal stage reference results, the real stage identification results at the previous sampling time, the current endpoint temperature prediction results, the target endpoint temperature results, and the melting hysteresis triggering results are input into the smelting time series stage identification model to generate candidate real stage matching score results. Based on the candidate real stage matching score results and stage continuity constraints, the real stage identification result is determined; When the actual stage identification result is the melting hysteresis stage, the current sampled data is reassigned from the second fixed stage data result to the melting hysteresis stage data result, and a melting hysteresis reassignment label result is generated. Based on the melting hysteresis heavy assignment results, the endpoint temperature deviation results, and the control boundary results, the segmented control correction results are generated. It should be noted that the smelting time sequence stage identification model can be a smelting time sequence stage identification model trained or fine-tuned with historical furnace stage samples, or it can be a stage identification model with time sequence coding capability. The smelting time series stage identification model adopts structured input and structured output. The smelting time series stage identification model does not output natural language control suggestions, but outputs the matching score results corresponding to each candidate real stage. The real stage identification result is determined by the candidate real stage matching score and the stage continuity constraint. When the candidate real stage with the highest matching score does not meet the stage continuity constraint, the candidate real stage that meets the stage continuity constraint and has the highest matching score is selected as the real stage identification result. Stage continuity constraints are used to limit situations where the process jumps directly from the melting lag stage to the slag-forming adjustment stage or the temperature rise adjustment stage. The input and output of this model are limited to structured data. The model input includes structured numerical sequence results and stage prompt label results. The structured numerical sequence results include stage power consumption increment results, stage oxygen supply increment results, stage furnace door opening time results, stage temperature measurement change results, scrap steel addition interval results, stage consistency deviation results, and endpoint temperature deviation results. The stage prompt label results include the current fixed stage name, re-identification trigger type, the true stage identification result at the previous sampling time, the temperature interpolation label result, and the data missing label result; The model output is the candidate true stage matching score for each candidate true stage, rather than an unverifiable natural language description.

[0042] Furthermore, the candidate real stages include the melting lag stage, the oxidation reaction stage, the slag formation adjustment stage, and the temperature rise adjustment stage; The melting lag stage indicates that although the current furnace has entered the second fixed stage according to the smelting time, the melting state of the furnace charge has not yet reached the conditions for entering the oxidation reaction stage normally. The oxidation reaction stage is used to indicate that the oxygen supply increment, temperature rise change and reaction rhythm of the current furnace batch are close to the normal oxidation reaction state; The slag-forming adjustment stage is used to indicate the main impacts of auxiliary material addition, slag state, and heat loss changes on temperature changes. The temperature adjustment stage indicates that the current furnace is close to the target endpoint temperature and mainly involves correcting the endpoint temperature.

[0043] The matching score of a candidate true stage is expressed by the following calculation relationship: the matching score of a candidate true stage is equal to the evidence score of that candidate true stage divided by the sum of the evidence scores of all candidate true stages; Among them, the evidence score of the candidate true stage is generated by the model based on the structured numerical sequence results, stage prompt label results and historical furnace stage samples; All candidate real stages include the melting delay stage, oxidation reaction stage, slag formation adjustment stage, and temperature rise adjustment stage; It should be noted that the candidate true stage matching score is used to indicate the degree of closeness between the current sampled data and each candidate true stage. The larger the candidate true stage matching score, the closer the current sampled data is to the corresponding candidate true stage.

[0044] Furthermore, for feasible model training methods, the stages in historical furnaces that are jointly confirmed by manual verification, temperature measurement records, furnace charge records, and final temperature qualification records can be used as training labels, and the corresponding stage verification data results can be used as training inputs to obtain model parameters through supervised training. For newly built production lines with insufficient historical labels, initial training samples can be formed using historical samples of the same furnace type or operating procedures. Then, during subsequent feedback verification, the confirmed reassignment retention results can be written into the sample library for continuous supplementation of training data.

[0045] Furthermore, stage continuity constraints are applied to the candidate true stage matching scores. These stage continuity constraints include: If the actual stage identification result at the previous sampling time is the melting lag stage, then the current sampling time is allowed to remain in the melting lag stage or transfer to the oxidation reaction stage, but is not allowed to directly transfer to the slag adjustment stage or the temperature adjustment stage. If the actual stage identification result at the previous sampling time is the oxidation reaction stage, then the current sampling time is allowed to remain in the oxidation reaction stage or move to the slag-forming adjustment stage; If the actual stage identification result at the previous sampling time is the slag-forming adjustment stage, then the current sampling time is allowed to remain in the slag-forming adjustment stage or move to the heating adjustment stage; If the actual stage identification result at the previous sampling time is the temperature rise and adjustment stage, then the current sampling time is allowed to remain in the temperature rise and adjustment stage; If the candidate real stage with the highest stage matching score does not meet the stage continuity constraint, then the candidate real stage that meets the stage continuity constraint and has the highest stage matching score is selected as the real stage identification result. It should be noted that the stage continuity constraint is used to prevent the model from outputting stage jumps that do not conform to the smelting process due to fluctuations at a single sampling point. For example, the current furnace should not jump directly from the melting lag stage to the temperature adjustment stage. If the matching score of the candidate true stage in the temperature adjustment phase increases temporarily due to temperature interpolation error or instantaneous power consumption fluctuation, it will be excluded by stage continuity constraint, thereby improving the stability of the true stage identification result.

[0046] Furthermore, when the actual stage identification result is the melting hysteresis stage, the current sampled data is reassigned to a different stage. The stage reassignment process includes: retaining the data records that originally belonged to the second fixed stage for the current sampled data, writing the current sampled data into the melting hysteresis stage data results, and generating melting hysteresis reassignment marker results; The melting lag reassignment labeling results are used to indicate that the data belongs to the second fixed stage in the sense of a fixed time window, but belongs to the melting lag stage in the sense of actual furnace conditions; By retaining both the original fixed-stage records and the actual-stage records, it is possible to trace the source of the fixed-stage division in the future and avoid the loss of original data due to reassignment processing.

[0047] Furthermore, based on the difference between the target endpoint temperature result and the existing endpoint temperature prediction result, an endpoint temperature deviation result is generated; When the endpoint temperature deviation result is positive, it means that the current endpoint temperature prediction result is lower than the target endpoint temperature result; When the endpoint temperature deviation result is negative, it indicates that the current endpoint temperature prediction result is higher than the target endpoint temperature result. When the endpoint temperature deviation is close to zero, it indicates that the current endpoint temperature prediction result is close to the target endpoint temperature result. The endpoint temperature deviation result is obtained by subtracting the existing endpoint temperature prediction result from the target endpoint temperature result.

[0048] Furthermore, the generated segmented control correction results include electrode power correction results, melting hold time correction results, and oxygen supply limitation results; Electrode power correction results are used to indicate whether to make a positive or maintain correction to the current electrode power within the device's permissible power range; The melt hold time correction result is used to indicate whether the duration of the melt phase is increased or maintained before entering the next stage of control; The oxygen supply limitation result is used to indicate that direct oxygen supply enhancement is prohibited or the increase in oxygen supply is limited during the melting lag phase; It should be noted that the control objective of the melting lag stage is not to directly compensate for the endpoint temperature by increasing oxygen supply, but to first ensure that the melting state of the furnace charge is close to the transition conditions of the normal stage. Therefore, the control priority is that the electrode power correction result takes precedence over the melting hold time correction result, and the melting hold time correction result takes precedence over the oxygen supply enhancement result.

[0049] The electrode power correction result is expressed by the following calculation relationship: the electrode power correction result is equal to the product of the endpoint temperature deviation adjustment result, the melting hysteresis reliable adjustment result, and the power correction reference result, and is limited by the electrode power allowable boundary result; The result of the endpoint temperature deviation adjustment is determined by the proportional relationship between the endpoint temperature deviation result and the allowable deviation range of the target endpoint temperature. The reliability adjustment result of melting hysteresis is determined by the matching score of the candidate true stage corresponding to the melting hysteresis stage. The higher the matching score of the candidate true stage, the greater the reliability adjustment result of melting hysteresis. The power correction baseline result is determined by the target electric furnace's rated power, current power, and equipment operating procedures; The permissible limit of electrode power is determined by the capacity of the electric furnace equipment and the process safety range; It should be noted that the limiting process is used to ensure that the electrode power correction result does not exceed the device's allowable power range.

[0050] The result of the melting and holding time correction is expressed by the following calculation relationship: the result of the melting and holding time correction is equal to the sum of the results of the insufficient temperature measurement adjustment, the furnace door heat loss adjustment, and the feeding influence adjustment, and is limited by the result of the allowable boundary of melting and holding time. Among them, the temperature measurement insufficiency adjustment result is determined by the difference between the reference temperature measurement change lower limit boundary result of the second fixed stage and the temperature measurement change result of the current stage; The result of furnace door heat loss adjustment is determined by the degree to which the current stage furnace door opening time exceeds the upper limit boundary of the second fixed stage furnace door opening time. The adjustment result of the feeding effect is determined by the difference between the feeding effect duration threshold result and the scrap steel addition interval result; The allowable boundary result for melting and holding time is determined by the smelting rhythm of the target steel grade, equipment load, and the allowable smelting cycle of the process. This calculation method provides a clear data source and calculation approach for the melt hold time correction results. For example, when the endpoint temperature deviation result is positive and the candidate true stage matching score result of the melting hysteresis stage is higher than the preset confidence score result, a positive electrode power correction result and a melting hold time correction result are generated. When the endpoint temperature deviation result is positive, but the matching score of the candidate true stage in the melting hysteresis stage is only slightly higher than that of other candidate true stages, the melting hold time correction result is generated first, and the electrode power correction result is slightly limited. When the endpoint temperature deviation result is not positive, but the actual stage identification result is still the melting hysteresis stage, an oxygen supply limitation result is generated to prevent the oxygen supply enhancement control from being directly executed due to the fixed stage entering the second fixed stage.

[0051] In step S3, the melting hysteresis triggering result obtained in step S2 is further converted into the real stage identification result, and the segmented control correction result is generated based on the real stage identification result. This step allows this embodiment to move beyond simply predicting the endpoint temperature and instead transform the judgment of whether the current second fixed stage is truly suitable for entering oxidation reaction control into specific control adjustments for electrode power, melting holding time, and oxygen supply limitations.

[0052] This step, based on the risk of melting lag identified in step S2, uses a smelting timing stage identification model to identify the actual stage of the current sampled data. When identified as a melting lag stage, the current sampled data is reassigned from the second fixed stage data results to the melting lag stage data results. Simultaneously, electrode power correction results, melting hold time correction results, and oxygen supply limitation results are generated. This step solves the problem that while existing endpoint temperature prediction algorithms can output endpoint temperature prediction results, they cannot determine how the control actions should be adjusted in the current stage. It further transforms the judgment of whether the fixed stage is accurate into the execution result of how the current stage should be controlled, so that the data that has not yet melted in the second fixed stage is no longer directly treated as oxidation reaction stage data, thereby improving the matching degree between segmented control actions and actual furnace conditions. The segmented control correction results generated in step S3 provide execution objects for the feedback verification in step S4.

[0053] In step S4, the temperature measurement result and the existing endpoint temperature prediction result are obtained at the next sampling time. The stage consistency deviation result and the endpoint temperature deviation result are recalculated. When both the stage consistency deviation result and the endpoint temperature deviation result are improved, the melting hysteresis reassignment mark result and the segmented control correction result are retained. Otherwise, a single review is performed and the segmented control correction result is updated with limited amplitude. The specific contents include: After executing the segmented control correction results, obtain the temperature measurement results, cumulative power consumption results, cumulative oxygen supply results, furnace door opening status results, existing endpoint temperature prediction results, and stage verification data results at the next sampling time. Recalculate the phase consistency deviation and endpoint temperature deviation results based on the data from the next sampling time. Compare the phase consistency deviation result at the next sampling time with the phase consistency deviation result before execution control, and compare the endpoint temperature deviation result at the next sampling time with the endpoint temperature deviation result before execution control. When the phase consistency deviation result decreases and the absolute value of the endpoint temperature deviation result decreases, a reassignment retention result is generated. When the stage consistency deviation result does not decrease or the absolute value of the endpoint temperature deviation result does not decrease, a reassignment verification result is generated, and the true stage identification result and segmented control correction result are corrected in a single iteration. It should be noted that the feedback verification in step S4 is not to train the model an infinite number of times, nor is it to continuously and repeatedly change the control actions. Instead, it is to confirm the real stage identification results and segmented control correction results generated in step S3 within the current sampling period.

[0054] Furthermore, the feedback verification results include the improvement results of stage deviation and the improvement results of temperature deviation; The improvement results of the stage deviation are obtained by comparing the stage consistency deviation results before the execution control with the stage consistency deviation results at the next sampling time. If the phase consistency deviation result at the next sampling time is less than the phase consistency deviation result before the execution control, then a phase deviation improvement result is generated. If the phase consistency deviation result at the next sampling time is not less than the phase consistency deviation result before the control is executed, then a phase deviation unimproved result is generated. The temperature deviation improvement result is obtained by comparing the absolute value of the endpoint temperature deviation result before the control is executed with the absolute value of the endpoint temperature deviation result at the next sampling time. If the absolute value of the endpoint temperature deviation at the next sampling time is less than the absolute value of the endpoint temperature deviation before control is executed, then a temperature deviation improvement result is generated. If the absolute value of the endpoint temperature deviation at the next sampling time is not less than the absolute value of the endpoint temperature deviation before the control is executed, then a result indicating that the temperature deviation has not been improved is generated. It should be noted that the stage deviation improvement results are used to determine whether the real stage reassignment makes the current sampled data closer to the reasonable furnace condition evolution, and the temperature deviation improvement results are used to determine whether the segmented control correction results improve the direction of the final temperature prediction.

[0055] Reassignment retention results are only generated when both stage deviation improvement results and temperature deviation improvement results exist simultaneously. If only the temperature deviation improves but the stage deviation does not, it means that the temperature prediction may improve in the short term due to changes in other variables, which cannot fully prove that the melting hysteresis reassignment is correct. If the stage deviation improves but the temperature deviation does not, it indicates that the current stage identification may be correct, but the magnitude or direction of the segmented control correction needs to be verified. For example, if the data of the current furnace in the second fixed stage is reassigned to the melting hysteresis stage in step S3, and positive correction of electrode power and correction of melting holding time are performed, the temperature change result at the next sampling time increases, the stage consistency deviation result decreases, and the existing endpoint temperature prediction result is closer to the target endpoint temperature result, then the reassignment retention result is generated.

[0056] After the reassignment retention results are generated, the fixed stage data results, real stage identification results, melting hysteresis triggering results, segmented control correction results, and feedback verification results of the current furnace are written into the similar furnace stage sample library as a reference for subsequent judgment of similar furnaces.

[0057] Furthermore, when generating the reassignment verification results, a single-iteration correction is performed, which includes: The phase verification data results at the current sampling time and the phase verification data results at the next sampling time are combined into a short-term state change result; The short-term state change results are re-inputted into the smelting time sequence stage identification model to obtain the updated candidate real stage matching score results. The results of identifying the true stage are reconfirmed based on the updated candidate true stage matching scores and stage continuity constraints. If the updated true stage identification result is still the melting hysteresis stage, the melting hysteresis reassignment label result is retained, but the magnitude of the segment control correction result is reduced. If the updated true stage identification result changes to the oxidation reaction stage, then the melting hysteresis reassignment label result is revoked, and the oxygen supply correction or conventional endpoint temperature control corresponding to the oxidation reaction stage is switched. It should be noted that the reduction in the amplitude of the segmented control correction result is not arbitrary, but rather a limit update is performed based on the control response direction result.

[0058] The control response direction result is used to indicate the direction of the effect of the current segmented control correction result on the temperature deviation; If the absolute value of the endpoint temperature deviation decreases after performing positive correction of electrode power, then the control response direction result corresponding to the electrode power is a valid positive result. If the absolute value of the endpoint temperature deviation does not decrease after performing positive correction of electrode power, then the control response direction result corresponding to the electrode power is invalid or too weak.

[0059] Based on the control response direction, the corresponding control correction is retained, reduced, or revoked. The constrained update of the segmented control correction results is expressed by the following calculation relationship: The segmented control correction result at the next sampling time is equal to the current segmented control correction result plus the product of the stage correction step size result, the current endpoint temperature deviation result, and the control response direction result, and is then subjected to amplitude limiting processing based on the control boundary result corresponding to the actual stage. Among them, the segmented control correction result at the next sampling time is the segmented control correction result used for the next control cycle after feedback iteration; The current segmented control correction result is the segmented control correction result that has already been executed in step S3; The step size for stage correction is determined based on the actual stage identification results; The current endpoint temperature deviation is the difference between the target endpoint temperature and the current predicted endpoint temperature. The direction of the control response is determined based on whether the endpoint temperature deviation decreases after the control is executed. The control boundary results include the allowable boundary results for electrode power, the allowable boundary results for melting and holding time, and the allowable boundary results for oxygen supply; When the actual stage identification result is the melting hysteresis stage, the segmented control correction result is generated according to the control priority.

[0060] The first priority is the electrode power correction result, which is used to correct the current electrode power within the allowable boundary of the electrode power; The second priority is the result of the melt holding time correction, which is used to correct the duration of the melting phase within the allowable boundary of the melt holding time. The third priority is the result of oxygen supply limitation, which is used to limit the enhancement of oxygen supply or the magnitude of the increase in oxygen supply.

[0061] Unless the feedback verification at the next sampling moment shows that the actual stage identification result has turned into the oxidation reaction stage, oxygen supply enhancement control will not be executed directly. When updating the limit, the control boundary results should include at least the electrode power allowable boundary results, the melting hold time allowable boundary results, and the oxygen supply allowable boundary results; The permissible limit of electrode power is determined by the capacity of the electric furnace equipment and the process safety range; The allowable boundary result for melting and holding time is determined by the smelting rhythm of the target steel grade, equipment load, and the allowable smelting cycle of the process. The oxygen supply limit is determined by the oxygen supply capacity of the oxygen lance and the process safety range of the corresponding smelting stage. Limiting is used to ensure that the updated segmented control correction results do not exceed the equipment capacity and process safety limits.

[0062] Furthermore, when the actual stage identification result is the melting hysteresis stage, the actual control boundary activated is mainly based on the electrode power allowable boundary result and the melting hold time allowable boundary result, while the oxygen supply allowable boundary result is used for amplitude verification of the oxygen supply limitation result. When the actual stage identification result is converted to the oxidation reaction stage, the oxygen supply allowable boundary result is used for oxygen supply correction; When the actual stage identification result is the temperature adjustment stage, the control boundary results include the allowable boundary results of the temperature holding time. It should be noted that the oxygen supply correction during the melting lag stage is constrained by the oxygen supply limitation results. Unless subsequent feedback verification shows that the actual stage has turned into the oxidation reaction stage, oxygen supply enhancement control will not be directly implemented.

[0063] Furthermore, step S4 also includes restricted updates to historical reference data; When the reassignment retention result is generated and the final endpoint temperature of the current furnace is within the allowable range of the target endpoint temperature, the melting hysteresis stage data result, stage verification data result, segmented control correction result and feedback verification result of the furnace are written into the similar furnace stage sample library. When the number of samples of the same type of melting lag in the sample library of similar furnace stages reaches the reference update threshold, the reference power consumption increment, reference oxygen supply increment, reference temperature change, reference furnace door opening time, and charging influence time threshold are recalculated for the melting lag stage. It should be noted that this restricted update is only used for accuracy judgment and real stage identification in subsequent stages, and does not change the fixed time window division rule in step S1, so as to ensure that the boundary between the existing technology base and the improved part of this embodiment is clear. For example, the reference update quantity threshold result can be set to no less than ten similar samples that have been verified to be valid through feedback. This quantity is only used to ensure that there is a certain sample basis when updating the reference result. In actual implementation, it can be set according to the production rhythm of the target electric furnace, the quantity of steel grades and the scale of historical data.

[0064] This step verifies the actual response to the real stage re-identification and segmented control correction in step S3. When the feedback results prove that the reassignment and control correction are effective, the results are retained and accumulated as experience for similar furnace runs; When the feedback results cannot prove the validity, only one review and limit update are performed to avoid excessive fluctuations in control. It forms a complete closed loop from fixed-stage data construction, stage accuracy judgment, real stage re-identification, segmented control correction to feedback verification and update, and can provide an implementable software control solution for the problem of stage mismatch in melting lag furnaces with fixed time window stage division. For example, if a certain furnace is in the second fixed stage at the current sampling time, the historical normal second fixed stage reference results show that the stage power consumption increment, stage oxygen supply increment, and stage temperature change results have corresponding lower limit boundary results, and the stage furnace door opening time result has corresponding upper limit boundary results. The current furnace cycle's stage power consumption increment, stage oxygen supply increment, and stage temperature change results are all lower than the corresponding lower limit boundary results, the stage furnace door opening duration result is higher than the corresponding upper limit boundary result, and the scrap steel addition interval result is less than the charging influence duration threshold result, generating a melting hysteresis trigger result.

[0065] The stage verification data results, stage consistency deviation results, true stage identification results from the previous sampling time, target endpoint temperature results, existing endpoint temperature prediction results, and melting hysteresis triggering results are input into the smelting time series stage identification model. If the candidate true stage matching score of the melting hysteresis stage is the highest and meets the stage continuity constraint, then the true stage identification result is determined to be the melting hysteresis stage. At this time, the data records that originally belonged to the second fixed stage are retained, and the current sampled data is written into the melting hysteresis stage data results to generate melting hysteresis reassignment label results.

[0066] When the endpoint temperature deviation results indicate that the existing endpoint temperature prediction results are lower than the target endpoint temperature results, the electrode power correction results are generated first within the electrode power allowable boundary results, and the melting and holding time correction results are generated within the melting and holding time allowable boundary results. At the same time, the oxygen supply limitation results are generated to limit the enhancement of direct oxygen supply. At the next sampling time, the stage consistency deviation result and the endpoint temperature deviation result are recalculated. If the stage consistency deviation result decreases and the absolute value of the endpoint temperature deviation result decreases, the melting hysteresis reassignment mark result and the segmented control correction result are retained. Otherwise, the phase verification data results of the current sampling time and the next sampling time are merged into short-term state change results, re-input into the smelting time sequence phase identification model for single verification, and the segmented control correction results are updated within the control boundary results.

[0067] After executing the segmented control correction results, this step uses the temperature measurement results at the next sampling time, the existing endpoint temperature prediction results, and the recalculated stage consistency deviation results and endpoint temperature deviation results to verify the actual stage identification results and segmented control correction results of step S3. This solves the problem that stage reassignment and control correction may be misjudged based on a single judgment. When both the stage consistency deviation results and endpoint temperature deviation results improve, the melting hysteresis reassignment marker results and segmented control correction results are retained. When both do not improve simultaneously, a single review is performed and the segmented control correction results are updated with limited amplitude, thereby avoiding repeated fluctuations or over-adjustments in control actions. This step writes the effective melting hysteresis samples into the similar furnace stage sample library, providing an updated basis for stage identification and control correction in subsequent furnaces. This makes the method form a complete closed loop from fixed stage data construction, stage mismatch identification, actual stage reassignment, segmented control correction to feedback verification and update.

[0068] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0069] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0070] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0071] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0072] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An electric furnace temperature segmentation intelligent control method based on large model smelting time series data, characterized by, Includes steps; Step S1: Obtain the current furnace smelting time sequence data, target endpoint temperature results and existing endpoint temperature prediction results, generate fixed stage data results according to fixed time windows, and extract stage power consumption increment results, stage oxygen supply increment results, stage furnace door opening duration results, stage temperature measurement change results and scrap steel addition interval results to form stage verification data results. Step S2: Calculate the stage consistency deviation result of the second fixed stage based on the historical normal stage reference results and reference boundary results. When the deviation exceeds the stage consistency threshold result, and the stage power consumption increment result, stage oxygen supply increment result, and stage temperature measurement change result are all lower than the corresponding lower limit boundary result, the furnace door opening time is higher than the corresponding upper limit boundary result, and the scrap steel addition interval is less than the feeding influence time threshold result, a melting hysteresis trigger result is generated. Step S3: Input the stage verification data results, stage consistency deviation results, real stage identification results at the previous sampling time, and melting hysteresis triggering results into the smelting time sequence stage identification model to obtain the candidate real stage matching score results, and determine the real stage identification results in combination with the stage continuity constraints. Step S4: When the actual stage identification result is the melting hysteresis stage, generate the melting hysteresis reassignment label result, and generate the electrode power correction result, melting hold time correction result, and oxygen supply limitation result based on the endpoint temperature deviation result and control boundary result. Then, based on the stage consistency deviation result and endpoint temperature deviation result at the next sampling time, retain or limit the update of the melting hysteresis reassignment label result and the segmented control correction result.

2. The electric furnace temperature segmentation intelligent control method based on large model smelting time series data according to claim 1, characterized in that, The smelting elapsed time result is generated based on the time difference between the current sampling time and the smelting start time. The smelting time sequence data of the current furnace is divided into fixed time windows according to the smelting elapsed time result to obtain multiple fixed stage data results. The fixed-stage data results include the first fixed-stage data results, the second fixed-stage data results, the third fixed-stage data results, and the fourth fixed-stage data results.

3. The electric furnace temperature segmentation intelligent control method based on large model smelting time series data according to claim 2, characterized in that, Extract the stage increment from the data results of each fixed stage to form the stage verification data results; The phase verification data results include phase power consumption increment results, phase oxygen supply increment results, phase furnace door opening time results, phase temperature measurement change results, and scrap steel addition interval results; The incremental power consumption result for each stage is obtained by calculating the difference between the cumulative power consumption at the end of the fixed stage and the cumulative power consumption at the beginning of the fixed stage. The incremental oxygen supply in a phase is obtained by calculating the difference between the cumulative oxygen supply at the end of the fixed phase and the cumulative oxygen supply at the beginning of the fixed phase. The stage furnace door opening time result is obtained by accumulating the duration during which the furnace door is in the open state within the fixed stage; The phased temperature change results are obtained by calculating the difference between the available temperature measurement results at the end of the fixed phase and the available temperature measurement results at the beginning of the fixed phase. The scrap steel addition interval is obtained by calculating the time difference between the current sampling time and the most recent scrap steel addition time.

4. The electric furnace temperature segmentation intelligent control method based on large model smelting time series data according to claim 3, characterized in that, A reference result for the historical normal furnace phase is constructed based on the set of historical normal furnace batches; The historical normal furnace set is processed according to the same fixed time window division method as the current furnace set, and the data results of the first fixed stage, the second fixed stage, the third fixed stage, and the fourth fixed stage are obtained respectively. For each fixed stage, the increment of power consumption, increment of oxygen supply, furnace door opening time and temperature change are statistically analyzed to generate reference power consumption increment results, reference oxygen supply increment results, reference furnace door opening time results and reference temperature change results.

5. The electric furnace temperature segmentation intelligent control method based on large model smelting time series data according to claim 1, characterized in that, The phase consistency deviation result is calculated based on the current fixed phase data result, phase verification data result, historical normal phase reference result, and reference boundary result. When the phase consistency deviation result does not exceed the phase consistency threshold result, the fixed phase division result is retained. When the stage consistency deviation exceeds the stage consistency threshold, it is further determined whether the deviation direction meets the melting hysteresis condition.

6. The electric furnace temperature segmentation intelligent control method based on large model smelting time series data according to claim 5, characterized in that, The specific conditions for melting hysteresis are: The current fixed stage is the second fixed stage, and the stage power consumption increment result is lower than the lower limit boundary result of the reference power consumption increment result of the second fixed stage, the stage oxygen supply increment result is lower than the lower limit boundary result of the reference oxygen supply increment result of the second fixed stage, the stage temperature change result is lower than the lower limit boundary result of the reference temperature change result of the second fixed stage, the stage furnace door opening time result is higher than the upper limit boundary result of the reference furnace door opening time result of the second fixed stage, and the scrap steel addition interval result is less than the feeding influence time threshold result. When all of the above conditions are met, a melting hysteresis trigger result is generated.

7. The electric furnace temperature segmentation intelligent control method based on large model smelting time series data according to claim 1, characterized in that, When generating melting hysteresis triggering results, the current fixed stage data results, stage verification data results, historical normal stage reference results, stage consistency deviation results, the real stage identification results at the previous sampling time, the current endpoint temperature prediction results, the target endpoint temperature results, and the melting hysteresis triggering results are input into the smelting time sequence stage identification model trained with historical furnace stage samples to generate candidate real stage matching scores. The candidate real stages include the melting delay stage, the oxidation reaction stage, the slag formation adjustment stage, and the temperature rise adjustment stage; The real stage identification result is determined based on the candidate real stage matching score and stage continuity constraints; When the candidate real stage with the highest matching score does not meet the stage continuity constraint, the candidate real stage that meets the stage continuity constraint and has the highest matching score is selected as the real stage identification result.

8. The electric furnace temperature segmentation intelligent control method based on large model smelting time series data according to claim 7, characterized in that, When the actual stage identification result is the melting hysteresis stage, retain the data record of the current sampled data that originally belonged to the second fixed stage, and write the current sampled data into the melting hysteresis stage data result to generate the melting hysteresis reassignment label result; The endpoint temperature deviation result is generated based on the difference between the target endpoint temperature result and the existing endpoint temperature prediction result; The segmented control correction results are generated based on the melting hysteresis reassignment labeling results, the endpoint temperature deviation results, the candidate true stage matching score results corresponding to the melting hysteresis stage, and the control boundary results. The segmented control correction results include electrode power correction results, melting hold time correction results, and oxygen supply limitation results; When the identification result in the real stage remains in the melting hysteresis stage, the oxygen supply limitation result is used to limit the oxygen supply enhancement, and the electrode power correction result and the melting hold time correction result are generated with priority over the oxygen supply enhancement result.

9. The method for segmented intelligent control of electric furnace temperature based on large-scale smelting time-series data according to claim 1, characterized in that, After executing the segmented control correction results, obtain the temperature measurement results, cumulative power consumption results, cumulative oxygen supply results, furnace door opening status results, existing endpoint temperature prediction results, and stage verification data results at the next sampling time. Recalculate the phase consistency deviation and endpoint temperature deviation results based on the data from the next sampling time. The phase consistency deviation result at the next sampling time is compared with the phase consistency deviation result before the execution control, and the endpoint temperature deviation result at the next sampling time is compared with the endpoint temperature deviation result before the execution control.

10. The method for segmented intelligent control of electric furnace temperature based on large-scale smelting time-series data according to claim 9, characterized in that, When the phase consistency deviation result decreases and the absolute value of the endpoint temperature deviation result decreases, a reassignment retention result is generated. When the stage consistency deviation result does not decrease or the absolute value of the endpoint temperature deviation result does not decrease, a reassignment verification result is generated, and the true stage identification result and segmented control correction result are corrected in a single iteration. The single-iteration correction involves merging the stage verification data results at the current sampling time with the stage verification data results at the next sampling time into a short-term state change result, re-inputting it into the smelting time sequence stage identification model, obtaining the updated candidate true stage matching score result, and reconfirming the true stage identification result based on the updated candidate true stage matching score result and stage continuity constraints.

Images