A temperature prediction method and apparatus

Abstract

The application provides a temperature prediction method and device, the method obtains the production performance data of the furnace in the molten steel smelting process, extracts the features related to the preset parameters from the production performance data of the furnace, obtains the sample features, and determines the actual outgoing temperature of the pre-continuous casting process corresponding to the sample features, trains the preset temperature prediction model according to the nonlinear mapping relationship between the sample features and the actual outgoing temperature of the pre-continuous casting process, obtains the target temperature prediction model, calculates the target outgoing time of the molten steel based on the continuous casting steel pouring live data, production planning scheduling information and process production performance data, updates at least part of the features related to the preset parameters based on the target outgoing time, obtains the target features, inputs the target features into the target temperature prediction model, and predicts the target outgoing temperature of the molten steel in the pre-continuous casting process; the above method provides a more practical and more accurate temperature prediction method for predicting the target outgoing temperature of the pre-continuous casting process.

Description

Technical Field

[0001] This application relates to the field of iron and steel metallurgy technology, and in particular to a temperature prediction method and apparatus. Background Technology

[0002] In the steelmaking-continuous casting production process, the molten steel exit temperature in the pre-casting stage is a key process parameter connecting "refining and upgrading" with "continuous casting billet formation". If the temperature is too high, it will cause a sharp increase in the heat flux density of the crystallizer during continuous casting, which can easily induce quality accidents such as billet cracks and steel leakage, while also increasing fuel consumption and production costs. If the temperature is too low, it will lead to a decrease in the fluidity of the molten steel, causing problems such as nozzle blockage and billet composition segregation, and may even lead to the interruption of continuous casting production.

[0003] However, in early production, the determination of the molten steel exit temperature in the pre-casting process mainly relied on operator experience, setting static temperature values ​​based on historical production data for a specific steel grade. This method was ill-suited to changes in operating conditions such as fluctuations in refining time and adjustments in continuous casting rhythm, resulting in insufficient precision in temperature control. With the maturation of metallurgical thermodynamics and heat transfer theories, the industry gradually introduced mechanistic models based on the "steel composition-refining process-continuous casting requirements" framework. These models achieved preliminary quantification of temperature recommendations by calculating parameters such as cooling curves and alloying endothermic / exothermic effects. However, these models had weak real-time response capabilities to dynamic operating conditions, making it difficult to meet the complex production needs of high-value-added steel grades such as pipeline steel and high-strength steel for automobiles. Clearly, a new temperature prediction method is urgently needed to address at least one of the aforementioned problems.

[0004] It should be noted that the above content only provides background information related to this application and does not necessarily constitute prior art. Summary of the Invention

[0005] This application provides a temperature prediction method and apparatus to solve the technical problem that related solutions cannot fully consider the multidimensional factors affecting the temperature change of molten steel under fluctuating operating conditions, thus making it difficult to accurately predict the target outlet temperature of the continuous casting process.

[0006] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0007] This application provides a temperature prediction method, comprising: acquiring furnace production performance data during the steel smelting process; extracting features related to preset parameters from the furnace production performance data to obtain sample features, and determining the actual outlet temperature of the continuous casting pre-process corresponding to the sample features, wherein the preset parameters include at least steel properties, continuous casting performance data, and ladle thermal state; training a preset temperature prediction model based on the nonlinear mapping relationship between the sample features and the actual outlet temperature of the continuous casting pre-process to obtain a target temperature prediction model; calculating the target outlet time of molten steel based on continuous casting performance data, production planning scheduling information, and process production performance data; updating at least some features related to the preset parameters based on the target outlet time to obtain target features, and inputting the target features into the target temperature prediction model to predict the target outlet temperature of molten steel in the continuous casting pre-process.

[0008] This application also provides a temperature prediction device, comprising: a data acquisition module for acquiring furnace production performance data during the steel smelting process; a feature extraction module for extracting features related to preset parameters from the furnace production performance data to obtain sample features, and determining the actual outlet temperature of the continuous casting pre-process corresponding to the sample features, wherein the preset parameters include at least steel properties, continuous casting performance data, and ladle thermal state; a model training module for training a preset temperature prediction model based on the nonlinear mapping relationship between the sample features and the actual outlet temperature of the continuous casting pre-process to obtain a target temperature prediction model; a calculation module for calculating the target outlet time of molten steel based on continuous casting performance data, production planning scheduling information, and process production performance data; a feature update module for updating at least some features related to the preset parameters based on the target outlet time to obtain target features; and a temperature prediction module for inputting the target features into the target temperature prediction model to predict the target outlet temperature of molten steel in the continuous casting pre-process.

[0009] The beneficial effects of this application are as follows: By acquiring furnace production performance data during the steel smelting process, features related to preset parameters are extracted from the furnace production performance data to obtain multi-dimensional sample features. The actual outlet temperature of the continuous casting process corresponding to the sample features is determined. The preset parameters include at least steel properties, continuous casting performance, and ladle thermal state. The preset temperature prediction model is trained based on the nonlinear mapping relationship between the sample features and the actual outlet temperature of the continuous casting process to obtain a target temperature prediction model. Based on the continuous casting performance data, production planning and scheduling information, and process production performance data, the target outlet time of the molten steel is calculated. Based on the target outlet time, at least some features related to the preset parameters are updated to obtain multi-dimensional target features. The target features are input into the target temperature prediction model to predict the target outlet temperature of the molten steel. By modeling and mining the multi-dimensional factors affecting the change of molten steel temperature, the aim is to achieve a leap from post-event compensation to pre-event accurate prediction, thereby improving the thermal efficiency and operational stability of the continuous casting process from the source, so as to accurately control the molten steel temperature, ensure the smooth operation of the continuous casting process, improve the quality of the cast billet, and reduce the overall energy consumption.

[0010] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0011] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0012] In the attached diagram:

[0013] Figure 1 This is a schematic diagram illustrating an exemplary system architecture as shown in an exemplary embodiment of this application; Figure 2 This is a schematic flowchart illustrating a temperature prediction method in an exemplary embodiment of this application; Figure 3 This is a schematic diagram illustrating the distribution relationship between the continuous casting furnace sequence level and the actual exit temperature of the preceding continuous casting process, as shown in an exemplary embodiment of this application. Figure 4 This is a schematic diagram illustrating the ladle thermal state decision-making mechanism, as shown in an exemplary embodiment of this application. Figure 5 This is a schematic diagram illustrating the relationship between the thermal state of the ladle and the temperature difference between the actual outlet temperature and the tundish temperature in the process before continuous casting of molten steel, as shown in an exemplary embodiment of this application. Figure 6This is an exemplary embodiment of the present application illustrating the cumulative absolute deviation histogram of the target temperature prediction model; Figure 7 This is a block diagram illustrating a temperature prediction device in an exemplary embodiment of this application; Figure 8 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown. Detailed Implementation

[0014] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0015] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. The drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the shape, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0016] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the present application. However, it will be apparent to those skilled in the art that embodiments of the present application may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the present application.

[0017] First, it should be noted that in the field of iron and steel metallurgy, the pre-casting process is a specific process definition term. In a broad sense, it refers to all the process steps from when molten steel is tapped from the primary refining furnace (such as a converter or electric furnace) until it is injected into the crystallizer of the continuous casting machine.

[0018] Furnace production performance data refers to the actual production results recorded in continuous production processes such as steel smelting, with furnace as the basic unit, covering multiple dimensions such as output, quality, efficiency, and cost.

[0019] Binning is a technique for smoothing and discretizing data by examining neighboring data. It divides continuous data into different intervals (bins) and performs statistical processing on the data within each bin to achieve noise reduction and feature simplification. Its core process includes four steps: determining the number or boundaries of bins, data allocation, calculation of intra-bin statistics (such as mean / median), and data replacement. Binning methods are mainly divided into three categories: equal-depth binning (equal data volume within each bin), equal-width binning (fixed numerical range), and custom-defined interval binning. Equal-depth binning allocates data by sorting, while equal-width binning divides data evenly according to the numerical range. Data smoothing can be achieved by replacing bin means, medians, or boundary values. This method is applied in fields such as credit scoring and customer segmentation for discretizing continuous variables, improving data quality by reducing noise, and thus optimizing the performance of data mining models.

[0020] LF (Ladle Furnace) is a key piece of equipment used for ladle refining in the steel industry.

[0021] RH (Ruhrstahl-Heraeus, vacuum circulation degassing unit) is a secondary refining process unit in ladle refining. The RH process (Ruhstahl-Heraeus Process or Ruhstahl hausen Process, molten steel vacuum circulation degassing method or RH vacuum degassing method) is a vacuum degassing technology in the steelmaking process.

[0022] Ladle residue refers to the molten steel and / or slag remaining in the ladle after the main ladle has poured molten steel into the tundish during continuous casting. This residue is considered residue because it cannot be completely emptied. Ladle residue can also indicate the amount of molten steel remaining in the ladle during the pouring process.

[0023] CatBoost (Categorical Boosting) is a gradient boosting tree algorithm that can be easily integrated with deep learning frameworks, can handle multiple data types, and can help enterprises solve various problems.

[0024] Additive ensemble strategy is a core model combination method in ensemble learning, mainly used in boosting algorithms (such as AdaBoost, GBDT, XGBoost, etc.). It gradually adds weak learners, assigning a weight to each newly added model, and finally linearly weights and sums all the weak learners to form a strong learner.

[0025] The weight per meter of a cast billet refers to the weight of the billet per meter of length, usually expressed in kg / m or t / m. It is a key process parameter in steel continuous casting production, used to control length cutting and quality control.

[0026] The furnace batch casting machine number is the continuous casting machine number designated for that furnace batch. It is a key parameter for achieving coordinated production scheduling of steelmaking and continuous casting, improving the hot charging ratio and logistics efficiency.

[0027] The ladle turret is a key piece of equipment in modern continuous casting technology used to transport and support ladles for pouring. It is located between the columns of the molten steel receiving span and the pouring span.

[0028] Quick change refers to the process in which, during multi-furnace continuous casting, when the currently used tundish reaches the end of its service life (e.g., severe erosion of refractory materials, excessive slag accumulation) or needs to be changed to a different steel grade, the old tundish is removed directly without interrupting the casting process, and a new, preheated tundish is quickly installed and casting begins immediately.

[0029] "Small-fast change" refers to the process where, when a continuous casting machine needs to switch from producing steel grade A to steel grade B, if the alloy element content of the two grades differs to a certain extent but the compatibility is controllable (or the difference is large but acceptable through calculation), the casting process is not interrupted, but the molten steel of the new steel grade is directly injected.

[0030] CART (classification and regression tree) is a decision tree learning method that uses a binary tree structure to recursively divide the feature space to predict conditional probability distributions. It is suitable for classification and regression tasks.

[0031] It should also be noted that the relevant technical solutions are usually set by operators based on limited information such as steel grade and planned continuous casting sequence, relying on their personal experience to set a fixed target outlet temperature or a relatively wide temperature range for the pre-casting process. This method is highly subjective and cannot accurately quantify the many dynamically changing factors in the production process, such as process rhythm, equipment conditions, and ambient temperature. This leads to inconsistent decision-making standards between different shifts and different heats, making it difficult to guarantee production stability and consistency.

[0032] Furthermore, most mainstream temperature assessment systems rely on standard smelting times to evaluate the average temperature drop of molten steel, providing only rough compensation for insufficient heat storage in the ladle or tundish. They lack dynamic consideration of each process node for each heat of molten steel under different production conditions, resulting in a static temperature assessment system with poor adaptability to dynamic operating conditions. For example, key variables such as ladle thermal state, refining time fluctuations, and differences in tundish heat storage capacity are not included in the calculations. This leads to significant differences in actual continuous casting superheat under the same target temperature control for the same steel grade in different heats, easily causing quality fluctuations or production accidents.

[0033] Furthermore, the assessment of the impact of related technologies on key variables such as the thermal state of the ladle and the insulation characteristics of the tundish is superficial. These variables are often simply substituted into the model as ordinary input parameters without establishing a refined quantitative correlation mechanism. This results in insufficient consideration of key influencing factors and coarse control granularity. For example, it fails to fully account for the uncertainty of heat absorption in the ladle and the temperature drop differences caused by the limitations of the refractory materials in the tundish.

[0034] Finally, many production sites, limited by operational habits, costs, or equipment layout, do not actually measure the temperature of molten steel arriving at the continuous casting ladle turret. This lack of crucial measured data fundamentally hinders the implementation of a technology route centered on ladle temperature measurement at the turret, leading to a break in the control chain and model inaccuracies. Without continuous feeding and feedback from on-site data, the machine learning algorithms they rely on will gradually deviate from actual production parameters, such as the transfer temperature drop rate and the ladle pouring temperature drop rate. Ultimately, this results in the predicted target outlet temperature of the preceding continuous casting processes deviating from the actual requirements, becoming nothing more than theoretical calculations lacking factual basis.

[0035] Therefore, this application utilizes artificial intelligence algorithms to deeply mine and model the multiple factors affecting the temperature drop of molten steel, aiming to achieve a leap from post-event compensation to pre-event accurate prediction, thereby fundamentally improving the thermal efficiency and stability of the entire continuous casting process.

[0036] Figure 1 This is a schematic diagram illustrating an exemplary system architecture as shown in an exemplary embodiment of this application.

[0037] Reference Figure 1As shown, the system architecture may include a data acquisition device 101 and a computer device 102. The computer device 102 may be at least one of a desktop graphics processing unit (GPU) computer, a GPU computing cluster, or a neural network computer. The data acquisition device 101 is used to collect furnace production performance data during the steel smelting process. In this embodiment, after acquiring the data, the data acquisition device 101 provides it to the computer device 102 for processing. Technical personnel can use the computer device 102 to extract features related to preset parameters from the furnace production performance data, obtain sample features, and determine the actual outlet temperature of the pre-casting process corresponding to the sample features. The preset parameters include at least steel properties, continuous casting performance, and ladle thermal state. The preset temperature prediction model is trained based on the nonlinear mapping relationship between the sample features and the actual outlet temperature of the pre-casting process to obtain a target temperature prediction model. Based on the continuous casting performance data, production scheduling information, and process production performance data, the target outlet time of the molten steel is calculated. Features at least partially related to the preset parameters are updated based on the target outlet time to obtain target features. These target features are then input into the target temperature prediction model to predict the target outlet temperature of the molten steel in the pre-casting process. It should be noted that the data acquisition device 101 and computer device 102 provided in this embodiment are merely examples and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0038] It should be noted that the temperature prediction method provided in this application embodiment is generally executed by computer device 102, and correspondingly, the temperature prediction device is generally installed in computer device 102.

[0039] Figure 2 This is a schematic flowchart illustrating a temperature prediction method in an exemplary embodiment of this application. The temperature prediction method can be executed by a computing processing device, which may be... Figure 1 The computer device 102 shown is illustrated. (Refer to...) Figure 2 As shown, the temperature prediction method includes at least steps S210 to S240, which are described in detail below: In step S210, the actual production data of each furnace during the steel smelting process are obtained.

[0040] In one embodiment of this application, before obtaining the actual production performance data of each heat in the steel smelting process, historical production data of the steel smelting process within a preset time range is obtained. The historical production data of the sample features includes ladle turnover history, tundish usage records, process specifications parameters, and production performance data of each process. The ladle turnover history, tundish usage records, process specifications parameters, and production performance data of each process are associated with the sample features using timestamps to determine the heat number corresponding to each of the sample features. The ladle turnover history, tundish usage records, process specifications parameters, and production performance data of each process are merged according to the sample feature heat number to obtain merged production data. The merged production data of the sample features is then cleaned to obtain the actual production performance data of each heat.

[0041] In this embodiment, historical production data within a preset time range (e.g., the historical production data of steel plant A for the past year) is obtained. The historical production data includes, but is not limited to, ladle turnover history, intermediate ladle usage records, process specification parameters, and production performance data of each process. The historical production data is merged based on the furnace number to obtain merged production data, which can be represented as dataset D. It is assumed that dataset D has a total of 34,390 furnace samples.

[0042] In this embodiment, the ladle turnover history and production performance data of each process include the heat number. The tundish usage record and process specification parameters are matched with the corresponding heat number through timestamps. The steps to determine the heat number corresponding to the tundish usage record include: obtaining the actual start time and end time of each heat from the continuous casting process production performance, as the heat time interval of each heat in the continuous casting process; extracting the tundish start time and end time of each tundish from the tundish usage record, constituting the tundish usage time interval of that tundish; comparing the heat time interval of each heat with the tundish usage time interval one by one to determine whether there is a time overlap, usually requiring the start time of the heat to fall between the start time and end time of the tundish; if the heat time interval and the tundish usage time interval overlap (or the inclusion relationship is established), then the tundish usage record is associated with that heat, that is, the heat number corresponding to the tundish usage record is determined. One tundish can correspond to multiple consecutive heats, and one heat corresponds to only one tundish. The steps for determining the heat number corresponding to the process specification parameters include: first, extracting the steel grade, cross-sectional dimensions, and production date for each heat from the production performance data; then, obtaining the process specification parameter table, which records the valid versions of the process parameters for each steel grade and cross-section, along with their effective start and end dates; next, for each heat, searching the parameter table for records where the steel grade and cross-section are completely matched and the heat's production date falls within the effective range; if multiple versions exist for the same steel grade and cross-section, selecting the version with the latest effective start date as the currently valid parameter; finally, binding this parameter record with the corresponding heat number to complete the association. It should be noted that the steps for determining the heat number corresponding to the intermediate package usage record and process specification parameters in this embodiment are merely illustrative. In actual applications, modifications can be made according to requirements, and this application does not impose any limitations on this.

[0043] The ladle turnover history records the complete time sequence of the ladle lining in each stage of "steeling-refining-pouring-idle-baking", including one or more of the following: the end time of the previous pour, idle time, number of turnovers, and baking regime. The tundish usage record covers one or more of the following: tundish service life, lining erosion status, baking temperature and time curve, and usage time of sizing nozzles or submerged nozzles. The process specification parameters are the benchmark documents that define the operational boundaries and target values ​​of each process. Specifically, they include one or more of the following: the liquidus temperature corresponding to the steel grade (i.e., the liquidus temperature of the steel grade), the superheat control range of the tundish (e.g., 7~15℃), the cooling intensity of the crystallizer, the casting speed curve, and the argon flow rate for protective pouring. These parameters constitute... The system establishes the target framework and constraints for molten steel temperature control. Any operational fluctuations deviating from the procedures will be reflected in real-time changes in the thermal state of the molten steel. The production performance data for each process consists of massive time-series records generated during actual production, objectively reflecting the dynamic evolution of operating conditions. Specifically, this includes one or more data points from the steelmaking process, such as converter tapping temperature, tapping time, and alloy additive addition; one or more data points from the refining process, such as LF / RH treatment cycle, energizing time, composition adjustment records, soft argon blowing time, and outlet temperature; and one or more data points from the continuous casting process, such as ladle start / end time, tundish molten steel level fluctuations, real-time casting speed changes, crystallizer heat flux density, secondary cooling water intensity, and the presence or absence of abnormal events such as sticking shutdowns or oxygen-induced flow interruptions. In the continuous casting process of steelmaking, sticking shutdown refers to the forced shutdown caused by the copper plate in the crystallizer sticking to the molten steel, leading to interruption of billet casting. By integrating the above-mentioned four-dimensional data, including ladle turnover history, intermediate ladle usage records, process specification parameters, and production performance data of each process, a full-process digital mapping from static process objectives to dynamic thermal state evolution can be constructed, providing complete feature support for target temperature prediction models based on mechanism and data fusion.

[0044] In this embodiment, data cleaning is performed on dataset D, including correcting the furnace number based on the smelting performance of molten steel in each process, further correcting the sequence number within the furnace casting by combining the tundish usage record, correcting the idling time of the ladle, deleting furnace samples containing missing / outlier values, etc., to obtain furnace production performance data, which can be denoted as dataset D1. It is assumed that dataset D1 has a total of 26,229 furnace samples.

[0045] In step S220, features related to preset parameters are extracted from the actual production data of each furnace to obtain sample features, and the actual exit temperature of the continuous casting pre-process corresponding to the sample features is determined.

[0046] Among them, the preset parameters include at least the properties of molten steel, the actual results of continuous casting, and the hot state of the ladle.

[0047] In one embodiment of this application, the sequence number within a casting cycle is numerically binned according to the actual outlet temperature of the preceding continuous casting process to obtain binning results. The continuous casting furnace sequence level is determined based on the binning results. The average temperature of the molten steel in the tundish is calculated and used as the tundish temperature. A sample of furnaces with sequence numbers within a preset sequence number range is extracted from the furnace production performance data, and target production data is constructed from the preset process parameters corresponding to the furnace samples. The preset process parameters include at least the ladle tare weight, steel loading time, cumulative energization time of the ladle refining furnace, cumulative active power consumption of the ladle refining furnace, ladle idling time, actual outlet temperature of the preceding continuous casting process, and tundish temperature. An optimal regression tree model is constructed, and the ladle thermal state is obtained based on the target production data and the optimal regression tree model.

[0048] In this embodiment, the tundish temperature calculated or selected for assessing whether the tundish steel temperature is qualified is used as the tundish temperature, such as the average value of continuous or manual temperature measurements of the tundish steel corresponding to certain specific ladle remaining tonnage or pouring time. It can be understood that certain specific ladle remaining tonnage refers to the weight of the remaining molten steel in the ladle reaching a preset value (e.g., 90t, 60t, 30t).

[0049] In one embodiment of this application, the optimal regression tree model is constructed as follows: initial independent variables are determined from preset process parameters, and an initial feature vector is constructed. The temperature difference of each furnace sample in the target production data is taken as the dependent variable, where the temperature difference is the difference between the actual outlet temperature and the tundish temperature of the preceding continuous casting process. The regression tree model is constructed with the initial feature vector as input and the dependent variable as output. Based on the node partitioning mechanism of the regression tree model, the feature importance score of each initial independent variable at the node partitioning is calculated, and the feature importance score is determined based on the node partitioning effect. The initial independent variables with feature importance scores greater than a preset score threshold are determined as target independent variables, and a target feature vector is constructed. The hyperparameter optimization method is adopted, with the minimization of the mean square error of cross-validation as the optimization objective, to optimize the hyperparameters of the regression tree model, determine the optimal hyperparameter combination, and train the optimal regression tree model based on the target feature vector.

[0050] Preferably, the regression tree model includes one or more of decision tree regression models, random forest regression models, and gradient boosting tree regression models, and more preferably, a decision tree regression model; preferably, the feature importance score is determined based on one or more of the following: reduction in mean squared error, reduction in Gini coefficient, and information gain; preferably, the hyperparameter optimization method is one of random search and grid search, and the hyperparameter optimization process uses K-fold cross-validation. Using a hyperparameter optimization method to optimize the hyperparameters of the regression tree model is to improve the model's prediction accuracy.

[0051] In this embodiment, determining the initial independent variables from the preset process parameters includes defining the ladle tare weight, steel loading time, cumulative energized time of the ladle refining furnace, cumulative active power consumption of the ladle refining furnace, and ladle idling time as initial independent variables. It is understood that in practical applications, the preset process parameters defined as initial independent variables can be added, reduced, or modified according to requirements. That is, one or more of the following can be defined as initial independent variables: ladle tare weight, steel loading time, cumulative energized time of the ladle refining furnace, cumulative active power consumption of the ladle refining furnace, and ladle idling time. When the regression tree model is a decision tree regression model, the optimal regression tree model is the optimal decision tree model.

[0052] In one embodiment of this application, the ladle thermal state is obtained based on sample feature target production data and sample feature optimal regression tree model, including: parsing the optimal regression tree model, extracting paths from the root node to each leaf node, each path corresponding to a set of classification rules consisting of target independent variables and corresponding preset thresholds; obtaining the value range of the dependent variable corresponding to each leaf node of the optimal regression tree model; mapping the value range of the dependent variable corresponding to each leaf node to the corresponding ladle thermal state classification label according to a preset temperature difference thermal state mapping relationship; obtaining the target independent variable value of the furnace to be tested from the sample feature target production data; inputting the target independent variable value into the optimal regression tree model, matching according to the path of the optimal regression tree model, determining the leaf node to which the furnace to be tested belongs, and thus obtaining the ladle thermal state.

[0053] In some embodiments, to simplify data complexity and highlight core trends in the casting process, the numerical variable, the sequence number within the casting cycle, is numerically binned based on the distribution of actual outlet temperatures from the preceding continuous casting processes. The resulting variable is named the continuous casting furnace sequence level. (Reference) Figure 3 As shown, Figure 3 This is a schematic diagram illustrating the distribution relationship between the actual exit temperatures of the continuous casting furnace sequence level and the pre-casting processes, as shown in an exemplary embodiment of this application. It is understood that the actual exit temperatures of the pre-casting processes corresponding to any two consecutive casting furnace sequence levels do not necessarily need to be different; rather, the distribution of actual exit temperatures of the pre-casting processes corresponding to different consecutive casting furnace sequence levels only needs to appear different from an overall trend perspective. Furthermore, Figure 3 The distribution of the continuous casting furnace sequence level and the actual exit temperature of the preceding process in the continuous casting is only for illustrative purposes and is not intended to limit the scope of this application.

[0054] To integrate scattered information and form a global perspective on the thermal state of the ladle, based on all key operational parameters that can be collected during the ladle turnover process, and under the premise of continuous casting and pouring with stable tundish thermal state, the temperature difference between the actual outlet temperature of the pre-casting process and the tundish temperature is used as the target variable. A tree model is used to construct a ladle thermal state index that can comprehensively and quantitatively characterize the real-time thermal performance of the ladle.

[0055] Seventeen sample features related to steel properties, continuous casting performance, and ladle thermal state were extracted from dataset D1 to construct the feature set required for the target outlet temperature prediction model of the continuous casting pre-process. These 17 sample features are: net weight of molten steel entering the continuous casting pre-process, liquidus temperature of the steel grade, waiting time for steel transfer (time interval between outlet temperature measurement and ladle pouring), pouring cycle, tundish temperature (the calculated / selected tundish steel temperature used to assess whether the tundish steel temperature is qualified, such as the average or maximum value of continuous / manual temperature measurement of tundish steel corresponding to certain ladle pouring tonnage or pouring time), continuous casting furnace sequence level, tundish steel temperature at ladle pouring start (for the first pouring, this feature corresponds to the tundish online temperature), net weight of molten steel at ladle pouring start, and special pouring modes (except for the first pouring, such as quick change (online tundish change, i.e., changing the tundish but not stopping the billet pulling) / the previous quick change / small quick change). The 17 features include: steel grades with varying alloy element content (e.g., casting modes requiring higher pre-casting temperatures compared to normal continuous casting furnaces), ladle tare weight, steel loading time (time interval between the end of primary furnace tapping and the pre-casting temperature measurement), cumulative LF refining energization time (i.e., cumulative energization time of the ladle refining furnace), cumulative LF refining active power consumption (i.e., cumulative active power consumption of the ladle refining furnace), ladle idling time (time interval between the end of continuous casting corresponding to the previous ladle turnaround and the start of primary furnace tapping corresponding to the current turnaround), ladle thermal state, season, and tundish type (classification of the tundish based on the type of refractory material used in the tundish working layer and the effective service life of the corresponding refractory material). By constructing the above 17 features, a comprehensive input feature set can be formed, covering steel properties, production plans, equipment status, process performance, environmental factors, etc., to fully cover key influencing dimensions.

[0056] In some embodiments, the numerical variable sequential number within the casting cycle is numerically binned according to the value distribution of the outlet temperature of the preceding continuous casting process. The resulting variable, continuous casting furnace sequence level, is still considered a numerical variable, and its characteristic correlation with the actual outlet temperature of the preceding continuous casting process is as follows: Figure 3 As shown, the pre-casting process includes LF refining and RH vacuum treatment, with LF refining accounting for more than 80% of the furnaces.

[0057] According to the production management regulations of steel plant A, the average of the continuous temperature measurements of molten steel in the tundish corresponding to ladle refill amounts of 90t, 60t, and 30t is taken as the tundish temperature. Balancing the timeliness and accuracy of data acquisition, five key process parameters are selected as initial independent variables: ladle tare weight (X1), steel loading time (X2), cumulative LF refining energization time (X3), cumulative LF refining active power consumption (X4), and ladle idling time (X5), denoted as the initial feature vector. Extract the sequence number values ​​within each pouring cycle from dataset D1. The five key process parameters corresponding to the heat sample [11,96], along with the actual outlet temperature and tundish temperature of the preceding continuous casting process, are used to construct target production data, which can be denoted as dataset D2. For dataset D2, using the temperature difference between the actual outlet temperature and the tundish temperature of the preceding continuous casting process as the dependent variable Y, a decision tree regression model is constructed and classification rules are extracted, ultimately forming the ladle thermal state variable. The specific process is as follows: First, a decision tree regression model is constructed using the initial feature vector as input and the dependent variable as output.

[0058] The decision tree regression model recursively partitions the feature space to minimize the mean squared error (MSE) of the sample set corresponding to each partition node on the dependent variable Y. For the m-th partition node, the partitioning optimization objective is: Equation (1) Among them, X j (j=1,2,3,4,5) represents the initial independent variable to be divided, and t is the corresponding division threshold; n m Let n be the total number of samples at the m-th partition node. L n R These represent the number of samples in the left and right child nodes after the partitioning; MSE L MSE R Let be the mean squared errors of the left and right child node samples on the dependent variable Y, respectively, calculated using the following formula: Equation (2) in, Let Y be the mean of the samples within a node. Let Y be the value of the i-th dependent variable Y. The interpretations of other parameters are given in Equation (1) and will not be repeated here.

[0059] Secondly, the dataset D2 is divided into a training set and a test set by random sampling at a preset ratio (e.g., 8:2) to ensure that the distribution characteristics of the two sets of data on the dependent variable Y are consistent after the division, thus completing the sample set division.

[0060] Then, based on the feature importance of the decision tree, i.e., quantified by the reduction in MSE (mean squared error reduction) during node partitioning, independent variables with weak explanatory power for the dependent variable Y (i.e., initial independent variables with feature importance scores less than or equal to a preset score threshold) such as steel ladle weight X1 and cumulative LF refining energizing time X3 are removed, retaining the core set of independent variables (i.e., target independent variables), denoted as... Further, based on this core set of independent variables, a combination of random search and K-fold cross-validation (e.g., 5-fold cross-validation) is used to optimize the key hyperparameters of the decision tree: maximum depth, minimum number of split samples, minimum number of leaf nodes, maximum number of leaf nodes, cost complexity pruning parameters, maximum number of features for node splits, splitting strategy, and minimum impurity reduction. The optimal combination of hyperparameters is determined with the goal of minimizing the MSE of the training set cross-validation, ultimately yielding the optimal decision tree model. By completing feature selection and parameter tuning through the above steps, a more suitable decision tree model can be obtained, leading to a more accurate understanding of the ladle's thermal state.

[0061] Finally, the mapping between the optimal decision tree partitioning rule and the hot state of the ladle is completed through the following steps: The optimal decision tree model is partitioned into three layers of nodes, forming a set of core independent variables. The clear partitioning rules (i.e., classification rules) are used to classify each leaf node. The interval mapping is used to classify the thermal state labels of steel ladles, where the thermal state of steel ladles includes, but is not limited to, high thermal stability state, medium-high thermal state, medium thermal equilibrium state, medium-low thermal state, and low thermal activity state. Figure 4 This is a schematic diagram illustrating the ladle thermal state decision-making mechanism as shown in an exemplary embodiment of this application, such as... Figure 4As shown, the decision-making mechanism, i.e., the preset temperature difference thermal state mapping relationship, is as follows: If the ladle idling time is greater than the first preset time, the value range of the dependent variable corresponding to each leaf node is mapped to the low thermal activity state; if the ladle idling time is less than or equal to the first preset time (e.g., 170.20 minutes), the cumulative LF refining active power consumption is less than or equal to the preset active power consumption (e.g., 4986 kWh), and the steel loading time is less than or equal to the second preset time (e.g., 38.44 minutes), the value range of the dependent variable corresponding to each leaf node is mapped to the medium-high thermal state; if the ladle idling time is less than or equal to the first preset time (e.g., 170.20 minutes), the cumulative LF refining active power consumption is less than or equal to the preset active power consumption (e.g., 4986 kWh), and the steel loading time is greater than the second preset time (e.g., 38.44 minutes). If the ladle idling time is less than or equal to the first preset time (e.g., 170.20 minutes), the cumulative active power consumption of LF refining is greater than the preset active power consumption (e.g., 4986 kWh), and the ladle idling time is less than or equal to the third preset time (e.g., 59.81 minutes), then the value range of the dependent variable corresponding to each leaf node is mapped to the medium thermal equilibrium state, where the third preset time is less than the first preset time; if the ladle idling time is less than or equal to the first preset time (e.g., 170.20 minutes), the cumulative active power consumption of LF refining is greater than the preset active power consumption (e.g., 4986 kWh), and the ladle idling time is greater than the third preset time (e.g., 59.81 minutes), then the value range of the dependent variable corresponding to each leaf node is mapped to the medium-low thermal state. For the core independent variable values ​​(i.e., target independent variable values) of each heat sample in dataset D1, the classification results of the ladle thermal state for each heat are obtained through path matching using the optimal decision tree model. The temperature difference distribution between the actual outlet temperature and the tundish temperature of the preceding continuous casting process corresponding to each ladle thermal state is as follows: Figure 5 As shown, Figure 5 This is a schematic diagram illustrating the distribution of the temperature difference between the ladle's thermal state and the actual outlet temperature of the molten steel in the pre-casting process, as well as the tundish temperature, in an exemplary embodiment of this application. It is understood that... Figure 5 The temperature drop from the outlet temperature measurement to the tundish temperature measurement in the pre-casting process of molten steel refers to the temperature difference between the actual outlet temperature and the tundish temperature in the pre-casting process. The average temperature drop refers to the median of the temperature difference. Figure 5 The values ​​of the parameters are for illustrative purposes only, and this application does not impose any restrictions on them.

[0062] Based on all key operational parameters during the ladle turnover process, such as ladle loading time, cumulative LF refining energization time, cumulative LF refining active power consumption, and ladle idling time, a comprehensive quantitative ladle thermal state index is constructed through a tree model to accurately characterize the real-time thermal performance of the ladle.

[0063] Based on the aforementioned steps, 17 features related to molten steel properties, continuous casting performance, and ladle thermal state were extracted from dataset D1 to obtain sample features. Based on these sample features, the feature set required for the preset temperature prediction model of the target outlet temperature in the pre-casting process was constructed. These 17 features, along with their respective data types, units of measurement, and numerical ranges, are shown in Table 1. Table 1 is the feature details table corresponding to the target sample model features.

[0064] Table 1

[0065] It should be noted that Table 1 is for illustrative purposes only. The parameters corresponding to the features can be increased, decreased, or changed according to actual needs, and their values ​​can also be modified according to actual needs. This application does not impose any restrictions on this.

[0066] In step S230, the preset temperature prediction model is trained based on the nonlinear mapping relationship between the sample characteristics and the actual outlet temperature of the continuous casting process to obtain the target temperature prediction model.

[0067] In one embodiment of this application, the actual outlet temperature of the continuous casting pre-process is used as the prediction target. A nonlinear mapping model is constructed using a machine learning model to obtain a preset temperature prediction model. The sample features and the actual outlet temperature of the continuous casting pre-process are divided into a training set and a validation set. The optimization objective is to minimize the mean square error between the actual outlet temperature of the continuous casting pre-process in the validation set and the predicted outlet temperature of the continuous casting pre-process in the preset temperature prediction model. A hyperparameter optimization algorithm is used to optimize the hyperparameter combination of the preset temperature prediction model. Based on the optimal hyperparameter combination obtained by optimization and the training set, the preset temperature prediction model is trained to obtain the target temperature prediction model.

[0068] In this embodiment, preferably, the machine learning model includes one or more of regression tree models, support vector regression models, and neural network regression models; more preferably, it is a gradient boosting tree model; and even more preferably, it is a CatBoost regression model. Preferably, the hyperparameter optimization algorithm is one or more of Bayesian optimization algorithms, random search algorithms, and grid search algorithms. The prediction accuracy of the validation set is used as the core evaluation criterion. The optimal values ​​of the hyperparameters are accurately determined by combining the objective function with the feedback from the validation set. Determining the optimal hyperparameter values ​​based on the prediction performance feedback from the validation set ensures that the optimized hyperparameter combination enables the model to have good generalization ability.

[0069] In one embodiment of this application, the actual outlet temperature of the continuous casting pre-process is used as the prediction target. A nonlinear mapping model is constructed using a machine learning model to obtain a preset temperature prediction model. This includes: performing multiple rounds of iterative training on the preset temperature prediction model until a preset iteration threshold is reached or a preset convergence condition is met; during each round of iterative training, the model parameters are adjusted based on the prediction deviation of the current model to update the current model, wherein the prediction deviation is the difference between the actual outlet temperature of the continuous casting pre-process and the predicted outlet temperature of the continuous casting pre-process in the current model; after the iterative training is completed, the model output results of each round of iteration are integrated to obtain the final model output, thereby determining the preset temperature prediction model.

[0070] In this embodiment, the preset temperature prediction model is first initialized, and the basic parameters of the model are configured to ensure that the model can carry out subsequent training normally. When the machine learning model is a CatBoost regression model, an initial predicted value can be obtained after model initialization. The initial predicted value is the mean of the prediction target. During iterative training, regression base learners are built, and the model parameters are adjusted based on the prediction residuals. After iteration, the preset temperature prediction model is obtained by combining the initial predicted value with the weighted prediction sum of each base learner. If other types of machine learning models such as neural networks or support vector regression are selected, their corresponding initialization methods, iterative training logic, and result integration methods can be adopted to ensure that the model has good generalization ability. It can be understood that the regression base learner refers to the underlying basic model used to perform the regression task in the ensemble learning framework. Its prediction results are integrated by higher-level meta-learners to generate the final regression prediction.

[0071] In one embodiment of this application, the actual outlet temperature of the continuous casting pre-process is used as the prediction target. A gradient boosting tree algorithm is used to construct a nonlinear mapping model to obtain a preset temperature prediction model. This includes: using the mean of the sample feature prediction target as the initial predicted value of the sample feature preset temperature prediction model; performing multiple iterations, constructing a decision tree in each iteration until a preset iteration threshold is reached or a preset convergence condition is met. Each iteration includes: constructing a decision tree based on the prediction residual of the current model to obtain a newly constructed decision tree, where the prediction residual of the sample feature current model is the difference between the actual outlet temperature of the continuous casting pre-process and the temperature prediction value of the current model; adding the product of the newly constructed decision tree and a preset learning rate parameter to the temperature prediction value of the sample feature current model to update the current model; calculating the weighted sum of the prediction values ​​of the decision trees constructed in all iterations to obtain a weighted prediction sum; and adding the initial prediction value of the sample feature to the weighted prediction sum of the sample feature to obtain the sample feature preset temperature prediction model. The current model is the model at the current iteration.

[0072] In this embodiment, the target temperature prediction model for predicting the target outlet temperature of the continuous casting process is built through the following steps: taking the actual outlet temperature of the continuous casting process as the target variable, a suitable machine learning algorithm is selected by comprehensively considering the data scale, data dimension, data type, deployment scenario, etc., to learn the nonlinear mapping relationship between the feature set and the actual outlet temperature of the continuous casting process, complete the model training, and obtain the target temperature prediction model to ensure the calculation accuracy and generalization ability.

[0073] In this embodiment, the actual exit temperature of the continuous casting process is taken as the prediction target Y. A nonlinear mapping model is constructed using the CatBoost gradient boosting tree algorithm to obtain a preset temperature prediction model, thereby achieving high-precision temperature prediction. The specific process is as follows: CatBoost is based on the gradient boosting tree framework. It gradually builds multiple decision trees through an additive ensemble strategy and combines their predictions in a weighted manner to form a strong learner. The working mechanism is as follows: The mean of the dependent variable Y (i.e., the mean of the prediction target) is used as the initial predicted value for the preset temperature prediction model. Where n is the sample size, Y i The actual exit temperature of the i-th sample in the pre-casting process is denoted as .

[0074] Iteratively construct the m-th tree. The construction process for each decision tree includes: defining the loss function of the decision tree as mean squared error; calculating the residuals using negative gradients to determine the optimization direction of the current model; fitting the sample feature residuals using a classification / regression tree to generate the leaf node splitting rules for the sample feature decision tree; and controlling the contribution weight of the sample feature decision tree prediction results using a preset learning rate parameter. Specifically, it includes the following steps: The loss function is defined as mean squared error. ,in, It is the predicted outlet temperature of the pre-casting process in the continuous casting stage, based on a preset temperature prediction model. It is the actual outlet temperature of the process before continuous casting.

[0075] The residuals are calculated using the negative gradient to find the optimization direction of the current model: Equation (3) Equation (3) is the calculation result after normalizing and simplifying the negative gradient, where r mi Let be the residual of the i-th sample in the m-th iteration (i.e., the prediction residual of the current model). Y represents the predicted value of the current model for the i-th sample after the (m-1)-th iteration (i.e., the temperature predicted by the current model); i The actual exit temperature of the i-th sample in the pre-casting process is denoted as .

[0076] By fitting the residuals using a CART tree, the leaf node partitioning rule T of the m-th tree is obtained. m (X): Equation (4) in, Let J be the predicted value of the m-th tree for the i-th sample; m R represents the total number of leaf nodes in the m-th tree; mj This represents the sample region corresponding to the j-th leaf node of the m-th tree; For the indicator function, if the sample Belongs to region R mj If the value is 1, then the value is 1; otherwise, the value is 0. w is a temporary variable representing the weight of the candidate leaf node. mj The weight of the j-th leaf node in the m-th tree: Equation (5) To avoid model overfitting, a preset learning rate parameter is used. Control the contribution weight of each tree's prediction result: Equation (6) After M iterations, the final prediction of the model is the sum of the weighted predictions of the decision trees built in all iterations (i.e., the weighted sum of predictions) and the initial prediction: Equation (7) It is understandable that M is the preset iteration number threshold or the total number of iterations when the preset convergence condition is met. The explanations of the other parameters have been provided in the corresponding positions above, and will not be repeated here.

[0077] Based on the values ​​of the four sub-variables in the above feature set—special casting mode, ladle thermal state, season, and tundish type—stratified random sampling is used with a preset division ratio (e.g., 7:1.5:1.5) to divide the feature set, along with the dependent variable, namely the actual outlet temperature of the continuous casting process, into a training set, a validation set, and a test set, ensuring that the distribution characteristics of the three sets of data on the dependent variable are consistent after the division.

[0078] To minimize the mean squared error (MSE) of the verification set, a maximum number of iterations (e.g., 200 iterations) or a preset convergence condition is set. The hyperparameters are optimized using the Tree-structured Parzen Estimator (TPE) Bayesian optimization algorithm. These hyperparameters include the number of iterations, tree depth, learning rate, randomness intensity, sampling randomness control parameter, number of discretization boundaries for numerical features control parameter, L2 regularization coefficient for leaf nodes, and minimum number of samples per leaf node. This yields the optimal CatBoost regression prediction model, i.e., the target temperature prediction model. The adjusted coefficient of determination R0 of this model on the training and test sets is then used. 2 The values ​​are 0.82 and 0.76 respectively, indicating that the model exhibits good fitting and generalization abilities and is suitable for practical application. Furthermore, the prediction bias of this optimal CatBoost regression prediction model on the test set is as follows: Figure 6 As shown, Figure 6 This is an exemplary embodiment of the present application illustrating the cumulative histogram of absolute deviation of the target temperature prediction model.

[0079] In step S240, the target time for molten steel to leave the station is calculated based on the actual data of continuous casting, production planning and scheduling information and process production performance data.

[0080] In one embodiment of this application, if it is determined that the current process of the molten steel is a process before continuous casting, then the planned casting machine number for the heat is obtained, wherein the actual production data of the process includes the current process and the planned casting machine number for the heat; based on the planned casting machine number for the heat, the actual continuous casting data is obtained, wherein the actual continuous casting data includes the remaining amount of molten steel in the ladle of the current casting heat, the billet weight per meter of each runner at the current moment, and the casting speed of each runner at the current moment; based on the remaining amount of molten steel in the ladle of the current casting heat, the billet weight per meter of each runner at the current moment, and the casting speed of each runner at the current moment; The remaining pouring time for the current pouring heat is calculated by taking the current billet weight per meter and the current pouring speed of each runner. Based on the current time, the remaining pouring time for the current pouring heat, the total planned pouring cycle of heats scheduled to start before the target heat, the subcontracting time corresponding to the planned casting machine number of the heat, the waiting time for steel, and the steel transfer time, the target exit time is calculated. The production planning and scheduling information includes the total planned pouring cycle, subcontracting time, waiting time for steel, and steel transfer time.

[0081] In this embodiment, during the pre-casting process of molten steel, the target time for molten steel to leave the station is calculated in real time based on the actual data of continuous casting, production planning and scheduling information and actual production data of the process.

[0082] In this embodiment, when the molten steel reaches the LF refining or RH vacuum treatment position, the planned process path for the heat is obtained, and it is determined whether the current process of the molten steel is a process before continuous casting. If so, the target exit time of the molten steel is calculated at a preset frequency (e.g., an update frequency of once every 5 seconds) until the molten steel leaves the processing position of that process.

[0083] If we denote the heat number corresponding to the molten steel as A, and the current process as LF1, then we obtain the planned casting machine number for the heat, denoted as CCM1, and calculate the remaining casting duration (RCD, min) of the heat currently being cast on the ladle turret corresponding to CCM1. It can be understood that min refers to the unit of time, minutes.

[0084] Equation (8) Where RCD is the remaining pouring time, W is the remaining ladle steel volume (tons) of the current heat being poured on the CCM1 ladle tumbler; weight_per_meter_j is the weight per meter of the billet in channel j of CCM1 at the current moment (tons / meter). If the weight per meter of the billet is not maintained on site according to parameters such as steel grade and billet specifications, it can be calculated based on the specifications of the bottom of the crystallizer in channel j at the current moment and the billet density. The specifications of the bottom of the crystallizer can be approximated by the billet width and thickness / diameter; casting_speed_j is the casting speed (meters / minute) of channel j in CCM1 at the current moment; n represents the total number of channels currently running on the continuous casting machine, and j is the j-th channel.

[0085] Similarly, based on the production plan for each heat, the total planned casting cycle (TPCC, min) of heats that have not yet been cast but are planned to be cast before the target heat A can be calculated. Assuming that a total of m heats are found, the target departure time of the target heat A is the current time + RCD + TPCC + (m+1) × r - t1 - t2, where r is the subcontracting time of CCM1; t1 is the waiting time associated with the planned steel grade of heat A, CCM1 and other parameters; and t2 is the steel transfer time from the LF1 processing position to the CCM1 ladle position.

[0086] Based on the actual data of continuous casting steel pouring, production planning and scheduling information and process production performance data, the target exit time is calculated, taking into full account the current production rhythm. Using this as the top-level constraint, the target exit time is transformed into core derived features, including steel loading time, steel transfer waiting time, etc., to ensure that relevant features are strongly bound to time nodes.

[0087] In step S250, at least some features related to preset parameters are updated based on the target exit time to obtain target features. The target features are then input into the target temperature prediction model to predict the target exit temperature of molten steel in the pre-casting process.

[0088] In one embodiment of this application, the target features include the net weight of molten steel entering the continuous casting station, the liquidus temperature of the steel grade, the waiting time for steel transfer, the casting cycle, the tundish temperature, the continuous casting furnace sequence level, the molten steel temperature in the tundish before the ladle starts casting, the net weight of molten steel in the tundish before the ladle starts casting, special casting mode, ladle tare weight, steel holding time, the cumulative energized time of the ladle refining furnace, the cumulative active power consumption of the ladle refining furnace, the idling time of the ladle, the thermal state of the ladle, the season, and the type of tundish.

[0089] In this embodiment, it can be understood that updating features related to at least some preset parameters based on the target departure time includes updating some features related to preset parameters based on the target departure time, and also updating all features related to preset parameters based on the target departure time. The sample features and the target features are only different in value. Based on the calculated target exit time, the actual or planned information corresponding to the target characteristics is obtained or calculated. Using the target temperature prediction model, the target exit temperature at the target exit time is calculated. The characteristics that differ from the sample characteristics used in model building are as follows: liquidus temperature of the steel grade (planned liquidus temperature of the steel grade), steel transfer waiting time (the sum of the steel transfer time from the current process position to the planned casting machine ladle position and the planned casting machine waiting time), pouring cycle (planned pouring cycle, which can be calculated by referring to the RCD relationship expression of the remaining pouring time mentioned above), tundish temperature (target tundish temperature), continuous pouring furnace sequence level (planned continuous pouring furnace sequence level), tundish steel temperature at the start of ladle pouring (current tundish steel temperature), net weight of tundish steel at the start of ladle pouring (current net weight of tundish steel), special pouring mode (planned special pouring mode), and steel slack time (the time interval from the end of primary furnace tapping to the target exit). By combining planned scheduling with real-time production performance, the characteristic values ​​are dynamically obtained, ensuring their timeliness and accuracy. By detecting abnormal operating conditions through special casting modes, the temperature prediction results are automatically adjusted to ensure the stability of the continuous casting process.

[0090] After the target temperature prediction model was put into operation, a total of 2300 heats with continuous production and normal data acquisition were tracked. Among them, the proportion of heats controlled according to the model prediction was 92.70% (2109 / 2300). The hit rate of the continuous casting tundish temperature is shown in Table 2. As can be seen from Table 2, when the target outlet temperature was controlled according to the model prediction, the hit rate of the continuous casting target tundish temperature reached 95.02% (2004 / 2109); while when the target outlet temperature was not controlled according to the model prediction, the hit rate of the continuous casting target tundish temperature was only 77.49% (148 / 191). This shows that this application can significantly improve the hit rate of the continuous casting tundish temperature and accurately control the tundish temperature.

[0091] Table 2

[0092] In some embodiments, historical production data within a preset time range is first acquired, including ladle turnover history, tundish usage records, process specifications, and production performance data for each process. Then, data cleaning and feature engineering are performed. A machine learning algorithm is selected to build a target temperature prediction model for the constructed feature set (sample features), and the target outlet time for the molten steel being smelted in the pre-continuous casting process is calculated. Finally, the target outlet temperature corresponding to the target outlet time is predicted for the molten steel being smelted in the pre-continuous casting process using the target temperature prediction model. This application is applicable to the prediction of target outlet time and corresponding target outlet temperature when the primary refining furnace argon station, LF refining, and RH vacuum treatment are used as pre-continuous casting processes. Controlling the production rhythm and outlet temperature of the pre-continuous casting process based on the target outlet time and target outlet temperature predicted by the target temperature prediction model helps ensure smooth production and guarantees the hit rate of the target tundish temperature for continuous casting.

[0093] This application replaces manual experience-based decision-making with a data-driven approach. Based on multiple multi-dimensional features, it calculates the target outlet temperature in real time, avoiding inconsistencies in standards caused by subjective experience differences between different shifts and heats. This ensures production stability and consistency, eliminates reliance on operator experience, and improves the objectivity and consistency of decision-making. This application can respond in real time to dynamic changes in the production process, such as process rhythm and equipment operating conditions. By quantifying key variables, such as ladle thermal state and tundish insulation characteristics, it automatically adjusts the calculated and predicted target outlet temperature to adapt to the specific conditions of different heats, reducing differences in superheating in continuous casting and avoiding... This approach avoids quality fluctuations or production accidents, enabling dynamic temperature regimes and enhancing adaptability to changes in operating conditions. Furthermore, it allows for in-depth quantification of key factors such as the uncertainty of ladle heat absorption and the limitations of refractory material insulation in the tundish through feature engineering. This refines the consideration of key factors, improving the granularity of control from coarse compensation to precise prediction, ensuring that temperature prediction matches the actual temperature drop pattern, thereby optimizing energy utilization and process efficiency. Moreover, this application does not rely on the measured temperature of molten steel on the continuous casting ladle turret, but rather merges multiple temperature drop sub-nodes, solving the problem of control chain breakage, possessing high universality, compensating for missing key data, and improving the integrity of the control chain.

[0094] Figure 7 This is a block diagram illustrating a temperature prediction device according to an exemplary embodiment of this application. The device can be applied to... Figure 1 The implementation environment shown is specifically configured in computer device 102. This device can also be applied to other exemplary implementation environments and specifically configured in other devices. This embodiment does not limit the implementation environment to which this device is applicable.

[0095] like Figure 7 As shown, the exemplary temperature prediction device includes: a data acquisition module 710, a feature extraction module 720, a model training module 730, a calculation module 740, a feature update module 750, and a temperature prediction module 760.

[0096] The system includes the following modules: a data acquisition module 710, used to acquire furnace production performance data during the steel smelting process; a feature extraction module 720, used to extract features related to preset parameters from the furnace production performance data to obtain sample features, and determine the actual outlet temperature of the continuous casting process corresponding to the sample features, wherein the preset parameters include at least steel properties, continuous casting performance, and ladle thermal state; a model training module 730, used to train a preset temperature prediction model based on the nonlinear mapping relationship between the sample features and the actual outlet temperature of the continuous casting process to obtain a target temperature prediction model; a calculation module 740, used to calculate the target outlet time of molten steel based on continuous casting performance data, production plan scheduling information, and process production performance data; a feature update module 750, used to update at least some features related to preset parameters based on the target outlet time to obtain target features; and a temperature prediction module 760, used to input the target features into the target temperature prediction model to predict the target outlet temperature of molten steel in the continuous casting process.

[0097] It should be noted that the temperature prediction device and the temperature prediction method provided in the above embodiments belong to the same concept. The specific ways in which each module and unit performs its operation have been described in detail in the method embodiments and will not be repeated here. In practical applications, the temperature prediction device provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. This is not a limitation here.

[0098] Embodiments of this application also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a temperature prediction method as described in any of the above embodiments.

[0099] Figure 8 A schematic diagram of a computer system suitable for implementing the embodiments of this application is shown. It should be noted that... Figure 8 The computer system 800 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of the embodiments of this application.

[0100] like Figure 8As shown, the computer system 800 includes a Central Processing Unit (CPU) 801, which can perform various appropriate actions and processes, such as executing the methods provided in the various embodiments above, based on a program stored in Read-Only Memory (ROM) 802 or a program loaded from storage portion 808 into Random Access Memory (RAM) 803. The RAM 803 also stores various programs and data required for system operation. The CPU 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0101] The following components are connected to I / O interface 805: an input section 806 including a keyboard, mouse, etc.; an output section 807 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to I / O interface 805 as needed. A removable medium 811, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 810 as needed so that computer programs read from it can be installed into storage section 808 as needed.

[0102] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 809, and / or installed from removable medium 811. When the computer program is executed by central processing unit (CPU) 801, it performs various functions defined in the system of this application.

[0103] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0104] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0105] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A temperature prediction method, characterized in that, include: Obtain actual production data for each furnace during the steel smelting process; Features related to preset parameters are extracted from the actual production data of the furnace batch to obtain sample features, and the actual exit temperature of the continuous casting pre-process corresponding to the sample features is determined. The preset parameters include at least the properties of molten steel, the actual continuous casting performance, and the hot state of the ladle. The preset temperature prediction model is trained based on the nonlinear mapping relationship between the sample features and the actual outlet temperature of the continuous casting pre-process to obtain the target temperature prediction model. Based on the actual data of continuous casting, production planning and scheduling information and process production performance data, the target time for molten steel to leave the station is calculated. Based on the target exit time update, which is at least partially related to the preset parameters, the target features are obtained. The target features are then input into the target temperature prediction model to predict the target exit temperature of molten steel in the pre-casting process.

2. The temperature prediction method according to claim 1, characterized in that, Before obtaining the actual production data of each heat during the steel smelting process, the method further includes: Acquire historical production data of the steel smelting process within a preset time range. The historical production data includes ladle turnover history, intermediate ladle usage records, process specification parameters, and production performance data of each process. The ladle turnover history, the intermediate ladle usage record, the process specification parameters, and the production performance data of each process are associated with timestamps to determine the furnace number corresponding to the ladle turnover history, the intermediate ladle usage record, the process specification parameters, and the production performance data of each process, respectively. Based on the furnace number, the ladle turnover history, the intermediate ladle usage record, the process specification parameters, and the production performance data of each process are merged to obtain merged production data. The merged production data is cleaned to obtain the actual production performance data for the furnace batch.

3. The temperature prediction method according to claim 1, characterized in that, After obtaining the actual production performance data of each heat during the steel smelting process, the method further includes: Based on the actual exit temperature of the preceding continuous casting process, the sequence number within the casting cycle is numerically divided into boxes to obtain the boxing results. The continuous casting furnace sequence level is then determined based on the boxing results. Calculate the average temperature of the molten steel in the tundish, and use the average temperature as the tundish temperature; Extract furnace samples with a sequence number within a preset sequence number range from the furnace production performance data, and construct target production data from the preset process parameters corresponding to the furnace samples. The preset process parameters include at least the ladle tare weight, steel loading time, cumulative energizing time of the ladle refining furnace, cumulative active power consumption of the ladle refining furnace, ladle idling time, actual outlet temperature of the continuous casting process, and tundish temperature. An optimal regression tree model is constructed, and the thermal state of the ladle is obtained based on the target production data and the optimal regression tree model.

4. The temperature prediction method according to claim 3, characterized in that, The optimal regression tree model is constructed using the following method: Initial independent variables are determined from the preset process parameters, and an initial feature vector is constructed. The temperature difference of each furnace sample in the target production data is taken as the dependent variable. The temperature difference is the difference between the actual outlet temperature and the tundish temperature of the process before continuous casting. Using the initial feature vector as input and the dependent variable as output, a regression tree model is constructed; Based on the node partitioning mechanism of the regression tree model, the feature importance score of each initial independent variable during node partitioning is calculated; The initial independent variables whose feature importance scores are greater than a preset score threshold are determined as target independent variables, and a target feature vector is constructed. A hyperparameter optimization method is adopted, with the minimization of the mean square error of cross-validation as the optimization objective, to optimize the hyperparameters of the regression tree model, determine the optimal hyperparameter combination, and train the optimal regression tree model based on the target feature vector.

5. The temperature prediction method according to claim 3, characterized in that, Based on the target production data and the optimal regression tree model, the hot state of the ladle is obtained, including: The optimal regression tree model is analyzed, and the path from the root node to each leaf node is extracted. Each path corresponds to a set of classification rules consisting of the target independent variable and the corresponding preset threshold. Obtain the value range of the dependent variable corresponding to each leaf node; Based on the preset temperature difference thermal state mapping relationship, the value range of the dependent variable corresponding to each leaf node is mapped to the corresponding ladle thermal state classification label. Obtain the target independent variable values ​​for the furnace to be tested from the target production data; The target independent variable value is input into the optimal regression tree model, and the leaf node to which the furnace to be tested belongs is determined by matching according to the path of the optimal regression tree model, thereby obtaining the hot state of the ladle.

6. The temperature prediction method according to claim 1, characterized in that, The preset temperature prediction model is trained based on the nonlinear mapping relationship between the sample features and the actual outlet temperature of the continuous casting pre-process to obtain the target temperature prediction model, including: Using the actual outlet temperature of the continuous casting process as the prediction target, a nonlinear mapping model is constructed using a machine learning model to obtain the preset temperature prediction model. The sample features and the actual outlet temperature of the continuous casting pre-process are divided into a training set and a validation set; With the goal of minimizing the mean square error between the actual outlet temperature of the continuous casting pre-process in the validation set and the predicted outlet temperature of the continuous casting pre-process in the preset temperature prediction model, a hyperparameter optimization algorithm is used to optimize the hyperparameter combination of the preset temperature prediction model. The preset temperature prediction model is trained based on the optimal hyperparameter combination obtained through optimization and the training set to obtain the target temperature prediction model.

7. The temperature prediction method according to claim 6, characterized in that, Using the actual outlet temperature of the preceding continuous casting process as the prediction target, a nonlinear mapping model is constructed using a machine learning model to obtain the preset temperature prediction model, including: The preset temperature prediction model is trained iteratively for multiple rounds until the preset number of iterations threshold is reached or the preset convergence condition is met. In each round of iterative training, the model parameters are adjusted based on the prediction deviation of the current model to update the current model, wherein the prediction deviation is the difference between the actual outlet temperature of the continuous casting pre-process and the predicted outlet temperature of the continuous casting pre-process in the current model. After the iterative training is completed, the model output results of each iteration are integrated to obtain the final model output, thereby determining the preset temperature prediction model.

8. The temperature prediction method according to any one of claims 1 to 7, characterized in that, Based on continuous casting data, production scheduling information, and process production performance data, the target time for molten steel to exit the station is calculated, including: If it is determined that the current process of the molten steel is a process before continuous casting, then the planned casting machine number for the heat is obtained, wherein the actual production data of the process includes the current process and the planned casting machine number for the heat; According to the planned casting machine number of the heat, the continuous casting steel pouring status data is obtained, wherein the continuous casting steel pouring status data includes the remaining molten steel in the ladle of the current pouring heat, the billet weight per meter of each runner at the current time, and the casting speed of each runner at the current time. The remaining pouring time for the current pouring furnace is calculated based on the remaining molten steel in the ladle for the current pouring furnace, the billet weight per meter in each runner at the current moment, and the casting speed in each runner at the current moment. The target departure time is calculated based on the current time, the remaining pouring time of the current pouring furnace, the total planned pouring cycle of furnaces scheduled to start pouring before the target furnace, the subcontracting time corresponding to the planned casting machine number of the furnace, the waiting time for steel, and the steel transfer time. The production plan scheduling information includes the total planned pouring cycle, the subcontracting time, the waiting time for steel, and the steel transfer time.

9. The temperature prediction method according to any one of claims 1 to 7, characterized in that, The target characteristics include the net weight of molten steel entering the station before continuous casting, the liquidus temperature of the steel grade, the waiting time for steel transfer, the pouring cycle, the tundish temperature, the continuous casting furnace sequence level, the temperature of molten steel in the tundish before the ladle starts pouring, the net weight of molten steel in the tundish before the ladle starts pouring, special pouring mode, ladle tare weight, steel holding time, the cumulative energized time of the ladle refining furnace, the cumulative active power consumption of the ladle refining furnace, the idling time of the ladle, the thermal state of the ladle, the season, and the type of tundish.

10. A temperature prediction device, characterized in that, include: The data acquisition module is used to acquire furnace production performance data during the steel smelting process; The feature extraction module is used to extract features related to preset parameters from the production performance data of the furnace, obtain sample features, and determine the actual exit temperature of the continuous casting pre-process corresponding to the sample features. The preset parameters include at least the properties of molten steel, continuous casting performance, and ladle hot state. The model training module is used to train a preset temperature prediction model based on the nonlinear mapping relationship between the sample features and the actual outlet temperature of the continuous casting pre-process, so as to obtain a target temperature prediction model. The calculation module is used to calculate the target time for molten steel to leave the station based on the actual data of continuous casting, production planning and scheduling information and process production performance data. The feature update module is used to update at least some features related to the preset parameters based on the target departure time to obtain the target features; The temperature prediction module is used to input the target features into the target temperature prediction model to predict the target outlet temperature of molten steel in the process before continuous casting.

Images