Tail packet transition intelligent decision-making method based on classification scoring model

Abstract

The invention relates to an intelligent decision-making method for tail ladle transition based on a classification scoring model, and belongs to the field of iron-making and steelmaking interface scheduling of iron and steel enterprises. The tail packet transition intelligent decision-making method comprises the following steps: collecting data of a field production process; generating a transition scheme set according to the collected data; extracting production parameter characteristics of each candidate iron notch-railway combination; a classification scoring model is constructed according to the production parameter features, quantitative evaluation is conducted on multiple schemes of same-furnace transition, different-furnace transition and in-situ waiting, and a candidate feasibility classification model and a candidate priority scoring model are included; and counting the judgment result of the classification model in the recent tail packet transition, and adaptively updating the threshold value. According to the intelligent decision-making method, accurate identification of transition opportunities and paths is achieved, the reasonability and accuracy of scheduling decision-making are remarkably improved, meanwhile, secondary transition can be reduced, temperature drop loss is reduced, decision-making time is shortened, and manual dependence is reduced.

Description

Technical Field

[0001] This invention relates to the field of scheduling at the ironmaking and steelmaking interface in iron and steel enterprises, and particularly to an intelligent decision-making method for tail package transfer based on a classification and scoring model. Background Technology

[0002] The tail ladle refers to the ladle of molten iron that is not filled after the blast furnace finishes tapping. Due to fluctuations in the amount of molten iron tapped each time, fluctuations in the amount of scrap steel, and changes in the state of the taphole, tail ladles are generated frequently. The way they are handled directly affects the turnover efficiency of molten iron ladles and the temperature drop of molten iron. These two indicators are important factors that restrict the continuous and stable operation of the ironmaking-steelmaking interface.

[0003] Currently, most steel companies use a rule-driven approach to handle tail flasks. Dispatchers rely on on-site experience and pre-defined rules to determine whether tail flasks need to be moved, which blast furnace taphole to move them to, and which railway to choose. Some companies even simply leave tail flasks in place without processing them.

[0004] Such rule-driven strategies typically employ a step-by-step selection process based on fixed rules: First, they determine whether a ladle is a tailing ladle based on indicators such as molten iron weight and scrap steel content. If the weight of a tailing ladle is below a preset threshold, it is directly considered an empty ladle and no longer transferred. Second, when one of the two tapholes of a blast furnace is tapping iron, the tailing ladle is preferentially transferred to the other taphole of the same blast furnace, provided that the taphole is within the planned tapping time and meets the adjustment conditions, such as the tapping time being more than 60 minutes from now, the remaining capacity of the ladle currently receiving iron being greater than 100 tons, or an empty ladle being positioned on another railway. If the same blast furnace cannot receive the tailing ladle, it will look for suitable tapholes in other blast furnaces. These tapholes are usually tapping iron or about to tap iron and can complete the adjustment before the tailing ladle arrives. Finally, the remaining amount of iron being received in the ladle is used to determine which railway under the blast furnace the tailing ladle should be placed on. If the remaining amount of iron being received in the ladle is large, the tailing ladle can be placed on the railway currently receiving iron; otherwise, another railway should be selected. If the ladle is not moved to another location to receive the molten iron, but instead waits in place, the next iron receiving will require waiting for the other ladle to finish one batch, which often takes 1-2 hours, resulting in a significant temperature drop loss in the molten iron. Improper handling of the ladle, leading to excessively long waiting times for receiving the molten iron, will also cause a significant temperature drop loss in the molten iron.

[0005] The existing tail pack relocation recommendation technology has the advantages of being simple to implement and easy to adjust based on on-site experience; however, it has the following shortcomings:

[0006] (1) The rules are too general and the decision-making is not precise enough. The thresholds are set by human experience, which cannot reflect the real-time production environment and make it difficult to comprehensively consider the trade-offs between various factors, such as transportation distance, waiting time, and predicted iron output.

[0007] (2) Poor scalability. Rules can only cover a limited number of scenarios. When the on-site production status is complex or undefined anomalies occur, manual intervention is often required.

[0008] (3) Difficulty in quantifying merits. The rules are mainly based on Boolean judgments, which cannot score and rank candidate transfer schemes. As a result, dispatchers sometimes still need to choose from multiple "qualified" schemes, which increases the workload and difficulty for dispatchers.

[0009] (4) Difficult to self-optimize. The existing solution does not support learning and correcting rules using historical data, and cannot dynamically adjust decisions based on equipment conditions, seasonal temperature differences, or molten iron quality indicators.

[0010] Therefore, there is an urgent need for an intelligent decision-making method that can use real-time data to comprehensively evaluate and quantitatively compare candidate transfer schemes, so that the tail pack can be quickly filled, the temperature drop can be reduced, and human intervention can be reduced. Summary of the Invention

[0011] Based on the above analysis, the present invention aims to provide an intelligent decision-making method for tail pack transfer based on a classification and scoring model, which enables tail packs to be quickly filled, reduces temperature drop, and reduces manual intervention.

[0012] On the one hand, this invention provides an intelligent decision-making method for tail packet transitions based on a classification and scoring model, comprising the following steps:

[0013] S1: Data Acquisition; collecting data from the on-site production process;

[0014] S2: Generate a set of transition schemes based on the collected data;

[0015] S3: Extract production parameter features for each candidate tap-railway combination;

[0016] S4: Construct a classification and scoring model based on production parameter characteristics to quantitatively evaluate various schemes such as transfer to the same furnace, transfer to different furnaces, and waiting in situ, including a candidate feasibility classification model and a candidate priority scoring model;

[0017] S5: Statistically analyze the classification model's judgment results in recent tail packet transitions and adaptively update the threshold.

[0018] Furthermore, in step S1, data acquisition includes blast furnace tapping plan, molten iron ladle status, track and swing status, and temperature drop model.

[0019] Furthermore, in step S2, the candidate transfer schemes include tail ladle determination and filtering: all molten iron ladles are compared with the steelmaking feeding requirements according to the "scrap steel quantity + molten iron quantity";

[0020] If the current molten iron content is less than the preset end-of-line ladle determination threshold, it is determined to be the end-of-line ladle and participates in the transfer decision; if the current molten iron content is less than the empty ladle threshold, it is regarded as an empty ladle and does not participate in the transfer decision.

[0021] It also includes candidate ladle enumeration: for each tail bale that needs to be transferred, a set of candidate ladles is constructed according to the principle of "priority for the same furnace"; for each candidate ladle, the two railways below it are further enumerated, and the specific railway is taken as a sub-candidate based on the remaining capacity of the ladle being received on the current railway and the empty bale alignment status of the other railway.

[0022] Furthermore, in step S2, the candidate transfer schemes include candidates for transfer within the same furnace, candidates for transfer to different furnaces, and in-situ waiting;

[0023] Candidates for transfer from the same furnace: If another tap is tapping or about to tapping iron from the blast furnace produced by the tail ladle, and the planned start time minus the estimated transportation time is within a certain range, then the tap is added to the candidate set;

[0024] Candidates for transfer to other furnaces: When the taphole of the same furnace does not meet the receiving conditions or the expected score is low, enumerate the tapholes of other blast furnaces that are tapping iron or are about to tap iron.

[0025] Waiting in place: Leaving the tail ladle in place to wait for the next batch of iron is a candidate solution, in case the transportation distance is long or the candidate taps are not enough to fill the tank.

[0026] Furthermore, in step S3, the extracted production parameter features include transportation time t. trans Waiting time t wait ; Full time t fill Temperature drop estimate ΔT: Probability of successful transfer in one operation p one Same furnace identification f same : Track oscillation condition f rail Other process constraints f const .

[0027] Furthermore, the temperature drop estimate ΔT = θ0 + θ1·t trans +θ2·t wait +θ3·(t trans ·t wait );

[0028] Where θ0, θ1, θ2, and θ3 are model parameters.

[0029] Furthermore, the same furnace identification f same If the candidate taphole is the same as the blast furnace produced by the tail ladle, set it to 1; otherwise, set it to 0. This is to reflect the principle of priority for the same furnace.

[0030] Track oscillation condition f railIf the remaining amount of iron being connected to the candidate railway is greater than the threshold or there is an empty package on another railway, then set it to 1; otherwise, set it to 0.

[0031] Other process constraints f const This reflects the special requirements of steelmaking, such as restrictions on tapping composition and equipment maintenance. If these requirements are met, set it to 1; otherwise, set it to 0.

[0032] Furthermore, the candidate feasibility classification model: First, it determines whether the candidate solution meets the basic feasibility requirements using a classification model.

[0033]

[0034] Where t max The maximum allowable waiting time for the tail ladle to be filled with molten iron is determined to ensure that the tail ladle can be filled before steelmaking feeding; this classification model makes judgments based on thresholds.

[0035] Furthermore, the candidate priority scoring model: for candidate solutions deemed feasible by the classification model, a comprehensive score is calculated using a linear or nonlinear scoring function.

[0036]

[0037] Where w i The weights for each feature are i = 1, 2, 3, 4, 5, or 6; these are determined through training with historical data or based on process experience.

[0038] Furthermore, the overall misjudgment cost function is calculated using the weights α and β set by the system:

[0039] C(t) = α·FP(t) + β·FN(t)

[0040] Where t is the classification threshold; FP(t) is the number of false acceptances, α is the weight coefficient of false acceptances; FN(t) is the number of false rejections, and β is the weight coefficient of false rejections.

[0041] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0042] 1. The intelligent decision-making method provided by this invention enables accurate identification of relocation timing and path, significantly improving the rationality and accuracy of scheduling decisions. It also reduces secondary relocation and temperature drop losses, shortens decision-making time, and reduces reliance on manual intervention. The system can automatically adjust model thresholds and weights using historical production data, enabling decision-making strategies to be continuously optimized according to production rhythm, equipment status, and seasonal changes, maintaining efficient and stable intelligent decision-making performance.

[0043] 2. Data acquisition and feature calculation are performed in real time. The model can quickly update its recommendations after changes in blast furnace tapping schedules or ladle status, ensuring timely decision-making. Model thresholds can be automatically adjusted based on historical data training, adapting to changes in equipment status, production rhythm, seasonal temperature, etc., avoiding situations where a single rule cannot adapt.

[0044] 3. This method provides a clear feature extraction and model framework. In the future, the decision-making effect can be further improved by adding new features (such as molten iron composition, fleet scheduling tension) or by adopting more advanced machine learning algorithms (such as reinforcement learning, deep learning).

[0045] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0046] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0047] Figure 1 This is a schematic diagram of the overall process of the intelligent decision-making method for tail package transfer;

[0048] Figure 2 This is a flowchart for the feasibility assessment process;

[0049] Figure 3 Flowchart for prioritizing candidates in a decision-making process. Detailed Implementation

[0050] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which constitute a part of the present invention and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0051] Currently, most steel companies use a rule-driven approach to handle tail flasks. Dispatchers rely on on-site experience and pre-defined rules to determine whether tail flasks need to be moved, which blast furnace taphole to move them to, and which railway to choose. Some companies even simply leave tail flasks in place without processing them.

[0052] Such rule-driven strategies typically employ a step-by-step selection process based on fixed rules: First, they determine whether a ladle is a tailing ladle based on indicators such as molten iron weight and scrap steel content. If the weight of a tailing ladle is below a preset threshold, it is directly considered an empty ladle and no longer transferred. Second, when one of the two tapholes of a blast furnace is tapping iron, the tailing ladle is preferentially transferred to the other taphole of the same blast furnace, provided that the taphole is within the planned tapping time and meets the adjustment conditions, such as the tapping time being more than 60 minutes from now, the remaining capacity of the ladle currently receiving iron being greater than 100 tons, or an empty ladle being positioned on another railway. If the same blast furnace cannot receive the tailing ladle, it will look for suitable tapholes in other blast furnaces. These tapholes are usually tapping iron or about to tap iron and can complete the adjustment before the tailing ladle arrives. Finally, the remaining amount of iron being received in the ladle is used to determine which railway under the blast furnace the tailing ladle should be placed on. If the remaining amount of iron being received in the ladle is large, the tailing ladle can be placed on the railway currently receiving iron; otherwise, another railway should be selected. If the ladle is not moved to another location to receive the next batch of iron, but instead waits in place for the next batch to finish, it will take 1-2 hours, resulting in a significant temperature drop loss in the molten iron. Improper handling of the ladle, leading to an excessively long waiting time for receiving the next batch, will also cause a significant temperature drop loss in the molten iron.

[0053] Therefore, this invention provides an intelligent decision-making method for tail packet transitions based on a classification scoring model, comprising the following steps:

[0054] S1: Data Acquisition; collecting data from the on-site production process;

[0055] S2: Generate a set of transition schemes based on the collected data;

[0056] S3: Extract production parameter features for each candidate tap-railway combination;

[0057] S4: Construct a classification and scoring model based on production parameter characteristics to quantitatively evaluate various schemes such as transfer to the same furnace, transfer to different furnaces, and waiting in situ, including a candidate feasibility classification model and a candidate priority scoring model;

[0058] S5: Statistically analyze the classification model's judgment results in recent tail packet transitions and adaptively update the threshold.

[0059] Compared with existing technologies, this invention collects data from the on-site production process, then constructs a set of transfer schemes based on the collected data; it then extracts the production parameter features of each candidate taphole-railway combination and constructs a classification and scoring model to quantitatively evaluate various schemes such as transfer from the same furnace, transfer from different furnaces, and in-situ waiting, including a candidate feasibility classification model and a candidate priority scoring model. Finally, it statistically analyzes the judgment results of the classification model in recent tail-bundle transfers, and adjusts and updates the thresholds as needed.

[0060] The intelligent decision-making method provided by this invention enables accurate identification of relocation timing and path, significantly improving the rationality and accuracy of scheduling decisions. It also reduces secondary relocation and temperature drop losses, shortens decision-making time, and reduces reliance on manual intervention. The system can automatically adjust model thresholds and weights using historical production data, enabling decision-making strategies to be continuously optimized according to production rhythm, equipment status, and seasonal changes, maintaining efficient and stable intelligent decision-making performance.

[0061] Specifically, in step S1, data acquisition includes blast furnace tapping plan, molten iron ladle status, track and swing status, and temperature drop model.

[0062] Specifically, the blast furnace tapping plan involves obtaining information such as the planned start time, planned tapping volume, and corresponding tapping tap number for each blast furnace in the next tapping cycle; and updating in real time the start time and volume of tapping at the tapping tapping tap.

[0063] Specifically, the status of molten iron ladles is collected in real time through the locomotive positioning system and the weighing system. The status information of each molten iron ladle includes its unique identifier, current location (blast furnace area or steelmaking area), current molten iron quantity, scrap steel quantity, whether it is receiving iron, and whether it is determined to be the last ladle.

[0064] Specifically, track and sway status: Obtain the current ladle position information of each railway under each tap, including the ladle number on each railway, whether the ladle is empty or not, the remaining capacity of the ladle being received, etc., and record the transportation distance and estimated travel time between each track.

[0065] Specifically, the temperature drop model: a temperature drop prediction model for molten iron ladles is established by combining historical temperature measurement data. The temperature drop during the transfer process is predicted based on the current temperature of the ladle, the proportion of scrap steel, and the expected waiting time, and is used by the scoring model.

[0066] It should be noted that in step S1, the information system first automatically collects the planned tapping start time, planned tapping quantity, and corresponding tapping tap number for each blast furnace in the next heat. It then updates the tapping start time and current tapping quantity at the tapping tap in real time. Combining the locomotive positioning system and weighing system, it obtains the unique identifier, current location, current molten iron and scrap steel quantity, whether it is currently receiving iron, and whether it is considered a tail ladle. It also acquires the current ladle position, empty ladle alignment, and remaining capacity of the ladle currently receiving iron on each railway track under each tapping tap, as well as track and swing status. To facilitate subsequent scoring, the system also establishes a temperature drop prediction model based on historical temperature data, estimating the temperature drop of the tail ladle during transportation based on the current temperature, scrap steel ratio, and expected waiting time.

[0067] Specifically, in step S2, the candidate transfer schemes include tail ladle determination and filtering: all molten iron ladles are compared with the steelmaking feeding requirements according to the "scrap steel quantity + molten iron quantity";

[0068] If the current molten iron content is less than the preset end-of-line ladle determination threshold, it is determined to be the end-of-line ladle and participates in the transfer decision; if the current molten iron content is less than the empty ladle threshold, it is regarded as an empty ladle and does not participate in the transfer decision.

[0069] It also includes candidate iron tap enumeration: for each tail bale that needs to be transferred, a set of candidate iron taps is constructed according to the principle of "priority for the same furnace"; for each candidate iron tap, the two railways below it are further enumerated (or multiple iron taps and multiple railways can be used), and the specific railway is taken as a sub-candidate based on the remaining capacity of the iron bale being received on the current railway and the empty bale alignment status of the other railway.

[0070] Specifically, in step S2, the candidate transfer schemes include candidates for transfer within the same furnace, candidates for transfer to different furnaces, and in-situ waiting;

[0071] Candidates for transfer from the same furnace: If another tap is tapping or about to tapping iron from the blast furnace produced by the tail ladle, and the planned start time minus the estimated transportation time is within a certain range, then the tap is added to the candidate set;

[0072] Candidates for transfer to other furnaces: When the taphole of the same furnace does not meet the receiving conditions or the expected score is low, enumerate the tapholes of other blast furnaces that are tapping iron or are about to tap iron.

[0073] Waiting in place: Leaving the tail ladle in place to wait for the next batch of iron is a candidate solution, in case the transportation distance is long or the candidate taps are not enough to fill the tank.

[0074] It should be noted that after obtaining the data of the on-site production process, the system compares all molten iron ladles with the steelmaking feeding requirements according to the "scrap steel quantity + molten iron quantity". Ladles with the current molten iron quantity lower than the preset end ladle judgment threshold are identified as end ladles, and molten iron quantities lower than the empty ladle threshold are regarded as empty ladles and excluded from the decision.

[0075] For each tail ladle requiring relocation, candidate taps are enumerated according to the principle of prioritizing taps from the same blast furnace. If another tap in the blast furnace where the tail ladle is located is tapping or about to tap, and the estimated transport time is within the planned tapping time, then that tap is included as a candidate. When the same blast furnace cannot meet the receiving requirements or the estimated score is low, taps from other blast furnaces that are tapping or about to tap can be selected as alternatives based on the tapping schedule. Remaining in place and waiting for the next tapping cycle is also a candidate option to address situations where the transport distance is too long or other options are insufficient to fill the tank in one relocation. For each candidate tap, the system further considers the current ladle positions on its two downstream railways, combining the remaining capacity of the currently receiving ladle with the empty ladle alignment status on the other railway to generate a specific railway plan.

[0076] Specifically, in step S3, the extracted production parameter features include transportation time t. trans Waiting time t wait; Full time t fill Temperature drop estimate ΔT: Probability of successful transfer in one operation p one Same furnace identification f same : Track oscillation condition f rail Other process constraints f const .

[0077] Specifically, the temperature drop estimate ΔT = θ0 + θ1·t trans +θ2·t wait +θ3·(t trans ·t wait );

[0078] Where θ0, θ1, θ2, and θ3 are model parameters.

[0079] It should be noted that, in order to quantitatively evaluate the candidate schemes, this invention extracts, but is not limited to, the following features for each candidate railway tap-railway combination:

[0080] Transportation time t trans Calculate the travel distance and estimated transportation time of the molten iron ladle based on the current location of the ladle and the location of the candidate iron tap track.

[0081] Waiting time t wait If the candidate tap is currently tapping iron, the waiting time for the last package to start receiving iron is predicted based on the expected amount of iron tapped and the remaining capacity of the currently receiving iron packages; if the tap is scheduled to tap iron but has not yet started, the calculation is the planned start time minus the current time and then the transportation time.

[0082] Full time t fill The time required to fill the tail ladle is calculated based on the remaining capacity of the tail ladle and the tapping rate of the candidate taphole (estimated by historical data or real-time measurement).

[0083] Temperature drop estimation ΔT: The predicted temperature drop of the ladle is based on transportation and waiting times. During transportation and waiting, the molten iron ladle dissipates heat to the environment through convection, radiation, and conduction, causing its temperature to decrease over time. Studies show that the temperature drop is rapid within the first 30 minutes after tapping, after which the rate of temperature drop stabilizes, averaging approximately 0.88℃ / min. Therefore, transportation and waiting times are key factors affecting the temperature drop of the ladle.

[0084] To quickly estimate temperature drop in scheduling decisions, a multiple linear model is used to establish the relationship between "transportation + waiting time" and the temperature drop value ΔT. The model assumes that the temperature drop is approximately linearly related to time over a short period of time and considers the transportation time t. trans Waiting time t wait And their interaction, the mathematical expression is as follows:

[0085] ΔT=θ0+θ1·t trans+θ2·t wait +θ3·(t trans ·t wait )

[0086] Where θ0, θ1, θ2, and θ3 are model parameters, which can be obtained from historical temperature measurement data through regression analysis. When the interaction effect is not obvious, it can also be simplified to a binary linear model ΔT = θ0 + θ1·t trans +θ2·t wait After the parameters are trained, the scheduling system can quickly estimate the temperature drop of the last package by measuring the transportation time and waiting time in real time, providing a reference for subsequent iron receiving scheme scoring and scheduling decisions.

[0087] The system uses a multiple linear regression method to train parameters on historical data samples, determining the optimal coefficients by minimizing the error between the predicted and measured temperature drops. After the model is deployed, it can automatically update parameters based on newly added temperature measurement data, achieving adaptive optimization.

[0088] The four parameters θ0, θ1, θ2, and θ3 are the regression coefficients of the temperature drop prediction model, representing the weights of the linear influence of different feature terms on the temperature drop value ΔT. Their meanings are as follows:

[0089] θ0 is a model constant term that reflects the reference temperature drop;

[0090] θ1 represents the transportation time t. trans The influence coefficient describes the contribution of transportation time to temperature drop;

[0091] θ2 is the waiting time t wait The influence coefficient describes the contribution of waiting time to temperature drop;

[0092] θ3 is the interaction term (t) trans ·t wait The influence coefficient is used to characterize the nonlinear effects of the combined effects of transportation and waiting time.

[0093] The probability p of a successful transition one Based on historical statistics, the probability that the candidate iron tap can be fully connected in one connection is estimated as the ratio of the remaining capacity of the tail package to the total expected iron capacity that the candidate iron tap can connect.

[0094] Same furnace marking f same If the candidate taphole is the same as the blast furnace produced by the tail ladle, set it to 1; otherwise, set it to 0. This is used to reflect the principle of priority for the same furnace.

[0095] Track oscillation condition f rail If the remaining amount of iron being connected to the candidate railway is greater than the threshold or there is an empty package on another railway, then set it to 1; otherwise, set it to 0.

[0096] Other process constraints fconst This reflects the special requirements of steelmaking, such as restrictions on tapping composition and equipment maintenance. If these requirements are met, set it to 1; otherwise, set it to 0.

[0097] It should be noted that, in order to quantitatively compare candidate taphole-railway combinations, this invention defines multiple evaluation features, including transportation time, waiting time, full-load time, temperature drop estimation, and the success probability of a single transfer. It also sets up indicative features such as co-furnace identification, track adjustment conditions, and other process constraints to reflect the co-furnace priority principle, track availability, and special limitations of the steelmaking process.

[0098] The transportation time is calculated based on the distance between the current position of the molten iron ladle and the candidate track position. The waiting time is calculated by combining the current iron receiving progress of the taphole and the planned tapping time to estimate the waiting time required to start receiving iron after arrival. The time to fill the ladle is estimated based on the remaining capacity of the tail ladle and the tapping rate of the taphole. The temperature drop estimate comes from the temperature drop model. The probability of a successful transfer is inferred based on the ratio of the remaining capacity of the tail ladle to the expected amount of iron that the taphole can receive. A series of indicator features are used to indicate whether the track is adjustable, whether it meets the process constraints, and whether the candidate taphole is from the same blast furnace produced by the tail ladle.

[0099] The aforementioned features can be further expanded as needed, for example, by considering blast furnace tapping time, historical temperature drop sensitivity, and locomotive scheduling tension. This invention is not limited to a specific feature set, but rather provides a structured feature extraction framework.

[0100] Specifically, the candidate feasibility classification model first determines whether the candidate solutions meet the basic feasibility requirements.

[0101]

[0102] Where t max The maximum allowable waiting time for the tail ladle to be filled with molten iron is determined to ensure that the tail ladle can be filled before steelmaking feeding; this classification model makes judgments based on thresholds.

[0103] Reference Figure 2 First, the feasibility classification model is used to determine whether candidate solutions meet the basic feasibility conditions. This model labels candidate solutions as "feasible" or "infeasible" based on thresholds such as track availability, process constraints, and whether the tail ladle can be fully loaded before steelmaking. For example, if the track can be adjusted and process constraints are met, and the tail ladle's waiting time for molten iron to be added does not exceed the allowable limit, it is marked as feasible; otherwise, it is considered infeasible. If feasible, it proceeds to the candidate priority scoring model for calculation; if infeasible, the system removes the solution from the candidate set and does not proceed to the subsequent priority scoring model calculation. When all candidate solutions are determined to be infeasible, the system automatically generates "wait in place" as the default recommended solution and prompts the dispatcher for manual confirmation or parameter adjustment.

[0104] Specifically, the candidate priority scoring model: For candidate solutions deemed feasible by the classification model, a comprehensive score is calculated using a linear or non-linear scoring function.

[0105]

[0106] Where w i The weights for each feature are set through training with historical data or based on process experience; the value range is generally [0,1], and all weights satisfy iΣwi=1 after normalization.

[0107] The initial weights are set based on on-site process experience. For example, transportation time and waiting time have higher weights, while temperature drop estimation has a lower weight. Subsequently, the system can be trained using the results data of historical transfer samples (such as the success rate of full connection in one go, temperature drop loss, etc.) and the weights can be automatically adjusted by minimizing the scoring error or optimizing the objective function. During system operation, the weights can also be periodically corrected according to changes in production characteristics (such as seasonal temperature or equipment status) to achieve continuous optimization.

[0108] t trans,max t wait,max and ΔT max p is the normalization factor for the corresponding feature. min The minimum threshold for the probability of full connection on the first attempt is given by combining on-site experience or data analysis. For all feasible candidate solutions, a comprehensive score is further calculated based on a weighted scoring function. The scoring function normalizes features such as transportation time, waiting time, probability of successful transfer on the first attempt, co-furnace identification, and temperature drop estimation, multiplies them by their corresponding weights, and sums them. The weights can be preset based on on-site experience or continuously adjusted and optimized based on historical production data. The higher the score, the better the candidate solution. A penalty is introduced; if p... one If the temperature is too low or the temperature drop is too large, negative points will be added.

[0109] The system calculates the score for all feasible candidate solutions for each tail packet and sorts them from highest to lowest score. When the highest score is below a preset threshold S... min At this time, it is recommended that the tail wrench remain in place and wait for the next tapping. min To enable offline analysis based on historical data before system deployment, a minimum confidence level is set as the initial threshold. Otherwise, the system selects the candidate with the highest score as the recommended transfer plan, outputting the ladle number, current location, recommended blast furnace number, taphole location, and railway line. It also provides the estimated transport and loading times, temperature drop estimates, and scoring criteria, displaying these information to the dispatcher via a list or floating window through the dispatch system interface. Dispatchers can choose to automatically execute the recommended plan or adjust it based on actual site conditions and experience. The system allows adjustment of feature weights and thresholds to accommodate special circumstances such as equipment maintenance or seasonal changes.

[0110] To continuously improve model performance, this invention emphasizes online learning and continuous iteration mechanisms. During system operation, the entire process data of each tail pack transfer is recorded, including candidate scheme characteristics, final selection, and execution results, such as whether a secondary transfer occurred, the time required to reach full capacity, and the actual temperature drop. Using this historical data, the thresholds or decision boundaries of the feasibility classification / candidate priority scoring model can be updated periodically, enabling the model to adaptively optimize according to changes in production rhythm and equipment status. In case of abnormal operating conditions, schedulers can also temporarily adjust parameters, allowing the system to quickly update recommendation results and achieve human-machine collaboration.

[0111] Specifically, the overall misjudgment cost function is calculated using the weights α and β set by the system:

[0112] C(t) = α·FP(t) + β·FN(t)

[0113] Where t is the classification threshold, used to distinguish between feasible and infeasible solutions. α is the weighting coefficient for false acceptances, used to measure the penalty for misclassifying an infeasible solution as a feasible solution (i.e., false acceptance). FP(t) is the number of false acceptances, representing the number or proportion of candidate solutions that the model judges as feasible but are actually infeasible when the classification threshold is t. β is the weighting coefficient for false rejections; FN(t) is the number of false rejections, representing the number or proportion of candidate solutions that the model judges as infeasible but are actually feasible when the classification threshold is t.

[0114] Based on the proportion of these two types of errors in the total sample and their actual costs, a comprehensive misjudgment cost function is calculated using weights α and β set by the system. The system iterates through a series of candidate thresholds in the background, selecting the threshold that minimizes C(t) as the new lower limit for judgment. When the system detects too many false acceptances, t is appropriately increased to make the judgment stricter; if there are too many false rejections, the threshold is correspondingly decreased to make the judgment more lenient. Such adjustments are completed automatically by the program without manual intervention. Users can also manually adjust the weights or thresholds under extreme operating conditions to quickly respond to production changes.

[0115] To more clearly describe the present invention, the following embodiments and comparative examples are provided for further illustration.

[0116] Example 1

[0117] Taking the example of a steel plant's iron-steel interface tail ladle relocation scenario, there are two blast furnaces (No. 1 and No. 2) on site, sharing three railway tracks. Blast furnace No. 1's taphole A is currently tapping iron, taphole B is waiting to tap iron, and blast furnace No. 2's taphole C is expected to tap iron in 20 minutes. Currently, the detected tail ladle volume at taphole A of blast furnace No. 1 is only 40 tons, which is determined to be a tail ladle. A decision needs to be made regarding its relocation and its target taphole.

[0118] (1) Data acquisition and feature calculation

[0119] The system collects the following data in real time:

[0120]

[0121] Based on the data, the system calculates:

[0122] Delivery time:

[0123] t trans B = 600 / 3 = 200s = 3.3min

[0124] t trans C = 1200 / 3 = 400s = 6.7min

[0125] Waiting time:

[0126] t wait B = 15 - 3.3 = 11.7 min

[0127] t wait C = 20 - 6.7 = 13.3 min

[0128] Expected full arrival time:

[0129] t fill B = 80 / 6 ≈ 13.3 min

[0130] t fill C = 80 / 8 = 10 min

[0131] (2) Feasibility classification judgment

[0132] The system has a preset total duration limit T. max =30min, used to determine whether the tail bale can be filled before steelmaking feeding.

[0133] Based on feasibility conditions:

[0134] t wait +t fill ≤T max =30min

[0135] The calculation results are as follows:

[0136] Iron tap B: 11.7min + 13.3min = 25.0min ≤ 30min → Feasible

[0137] Iron tap C: 13.3min + 10.0min = 23.3min ≤ 30min → Feasible

[0138] Both meet the basic feasibility requirements.

[0139] (3) Temperature drop estimation

[0140] A linear temperature drop prediction model is used:

[0141] ΔT=θ0+θ1·t trans +θ2·t wait +θ3·(t trans ·t wait )

[0142] Suppose θ0=-1.2; θ1=-0.03; θ2=-0.04; θ3=-0.001.

[0143] The calculation yields:

[0144] Iron tap B: ΔT = -1.2 - 0.03 × 3.3 - 0.04 × 11.7 - 0.001 × (3.3 × 11.7) ≈ -1.8℃

[0145] Iron tap C: ΔT = -1.2 - 0.03 × 6.7 - 0.04 × 13.3 - 0.001 × (6.7 × 13.3) ≈ -2.0℃

[0146] (4) Priority score calculation

[0147] The system calculates the comprehensive score for candidate solutions that are deemed "feasible" by the feasibility classification model using the following priority scoring function:

[0148]

[0149] in:

[0150] t trans : Transportation time; t trans,max Maximum transit time (30 minutes);

[0151] t wait Waiting time; t wait,max Waiting time limit (30 minutes);

[0152] p one : Probability of a successful transition;

[0153] f same Same furnace identifier, with a value of 1 (1 for the same furnace) or 0 (0 for different furnaces);

[0154] ΔT: Predicted temperature drop, ΔT max Upper limit of temperature drop (taken as 3℃);

[0155] p min Minimum acceptable probability of full coverage (e.g., 0.7);

[0156] w1-w6: The weights of each feature can be determined through experience or model learning.

[0157] According to the scenario of the embodiment, let:

[0158] w1:w2:w3:w4:w5:w6=0.2:0.2:0.2:0.2:0.1:0.1

[0159] Calculate the normalized value of each feature separately:

[0160]

[0161] Substituting, we get:

[0162] S B =0.2×0.89+0.2×0.61+0.2×0.9+0.2×1-0.1×0.6=0.78

[0163] S C =0.2×0.78+0.2×0.56+0.2×0.85+0.2×0-0.1×0.67=0.49

[0164] The system calculation results show that the comprehensive score of taphole B (in the same furnace) is higher than that of taphole C. Therefore, it is recommended to transfer the tail ladle to taphole B of blast furnace No. 1.

[0165] Compared to existing methods, the median waiting time for empty ladles was reduced by 12 minutes, the median interval between two iron connections at the tail ladle was reduced by 22 minutes, and the overall proportion of molten iron temperatures exceeding 1380℃ at the KR station tail ladle increased by 12.63 percentage points.

[0166] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A tail packet transition intelligent decision-making method based on a classification and scoring model, characterized in that, Includes the following steps: S1: Data Acquisition; Collect data on the on-site production process; S2: Generate a set of transition schemes based on the collected data; S3: Extract production parameter features for each candidate tap-railway combination; S4: Construct a classification and scoring model based on production parameter characteristics to quantitatively evaluate various schemes such as transfer to the same furnace, transfer to different furnaces, and waiting in situ, including a candidate feasibility classification model and a candidate priority scoring model; S5: Statistically analyze the classification model's judgment results in recent tail packet transitions and adaptively update the threshold.

2. The intelligent decision-making method for tail packet transition based on a classification scoring model according to claim 1, characterized in that, In step S1, data acquisition includes blast furnace tapping plan, molten iron ladle status, track and swing status, and temperature drop model.

3. The intelligent decision-making method for tail packet transition based on a classification scoring model according to claim 1, characterized in that, In step S2, the candidate transfer schemes include tail ladle determination and filtering: all molten iron ladles are compared with the steelmaking feeding requirements according to the "scrap steel quantity + molten iron quantity"; If the current molten iron content is less than the preset end-of-line ladle determination threshold, it is determined to be the end-of-line ladle and participates in the transfer decision; if the current molten iron content is less than the empty ladle threshold, it is regarded as an empty ladle and does not participate in the transfer decision. It also includes candidate ladle enumeration: for each tail ladle that needs to be transferred, a set of candidate ladles is constructed according to the principle of "priority for ladles in the same furnace"; for each candidate ladle, the two railways below it are further enumerated, and the specific railway is taken as a sub-candidate based on the remaining capacity of the ladle being received on the current railway and the empty ladle alignment status of the other railway.

4. The intelligent decision-making method for tail packet transition based on a classification scoring model according to claim 3, characterized in that, In step S2, the candidate transfer schemes include candidates for transfer within the same furnace, candidates for transfer to different furnaces, and in-situ waiting; Candidates for transfer from the same furnace: If another tap is tapping or about to tapping iron from the blast furnace produced by the tail ladle, and the planned start time minus the estimated transportation time is within a certain range, then the tap is added to the candidate set; Candidates for transfer to other furnaces: When the taphole of the same furnace does not meet the receiving conditions or the expected score is low, enumerate the tapholes of other blast furnaces that are tapping iron or are about to tap iron. Waiting in place: Leaving the tail ladle in place to wait for the next batch of iron is a candidate solution, in case the transportation distance is long or the candidate taps are not enough to fill the tank.

5. The intelligent decision-making method for tail packet transition based on a classification scoring model according to claim 1, characterized in that, In step S3, the extracted production parameter features include transportation time t. trans Waiting time t wait ; Full time t fill Temperature drop estimate ΔT: Probability of successful transfer in one operation p one Same furnace identification f same : Track oscillation condition f rail Other process constraints f const .

6. The intelligent decision-making method for tail packet transition based on a classification scoring model according to claim 5, characterized in that, The temperature drop estimate is ΔT = θ0 + θ1·t trans +θ2·t wait +θ3·(t trans ·t wait ); Where θ0, θ1, θ2, and θ3 are model parameters.

7. The intelligent decision-making method for tail packet transition based on a classification scoring model according to claim 5, characterized in that, Same furnace marking f same If the candidate taphole is the same as the blast furnace produced by the tail ladle, set it to 1; otherwise, set it to 0. This is to reflect the principle of priority for the same furnace. Track oscillation condition f rail If the remaining amount of iron being connected to the candidate railway is greater than the threshold or there is an empty package on another railway, then set it to 1; otherwise, set it to 0. Other process constraints f const This reflects the special requirements of steelmaking, such as restrictions on tapping composition and equipment maintenance. If these requirements are met, set it to 1; otherwise, set it to 0.

8. The intelligent decision-making method for tail packet transition based on a classification scoring model according to claim 1, characterized in that, Candidate Feasibility Classification Model: First, the classification model is used to determine whether the candidate solutions meet the basic feasibility requirements. Where t max The maximum allowable waiting time for the tail ladle to be filled with molten iron is determined to ensure that the tail ladle can be filled before steelmaking feeding; this classification model makes judgments based on thresholds.

9. The intelligent decision-making method for tail packet transition based on a classification scoring model according to claim 1, characterized in that, Candidate Priority Scoring Model: For candidate solutions deemed feasible by the classification model, a comprehensive score is calculated using a linear or non-linear scoring function. Where w i The weights for each feature are i = 1, 2, 3, 4, 5, or 6; these are determined through training with historical data or based on process experience.

10. The intelligent decision-making method for tail packet transition based on a classification scoring model according to claim 1, characterized in that, The overall misjudgment cost function is calculated using the weights α and β set by the system. C(t) = α·FP(t) + β·FN(t) Where t is the classification threshold; FP(t) is the number of false acceptances, α is the weight coefficient of false acceptances; FN(t) is the number of false rejections, and β is the weight coefficient of false rejections.

Images