Cold heading steel continuous casting billet hot straight sending process billet quality feedforward compensation method and system

Abstract

This application provides a method and system for feedforward compensation of billet quality during the hot direct delivery process of cold-heading steel continuous casting billets. The method includes: acquiring multi-dimensional quality status assessment data of the billet, performing matching analysis through a preset historical feedforward compensation parameter database to obtain an initial feedforward compensation parameter set, then combining the multi-dimensional quality status assessment data with a preset feedforward compensation optimization model and a preset hot direct delivery digital twin model to obtain an optimized feedforward compensation parameter set, and executing feedforward compensation for the hot direct delivery of cold-heading steel continuous casting billets. This application achieves accurate feedforward compensation of billet quality during the hot direct delivery process by integrating multi-dimensional quality online monitoring and fusion, performing initial similarity matching with feedforward parameters based on historical experience, conducting simulation analysis and self-correction through the preset feedforward compensation optimization model and the preset hot direct delivery digital twin model, and executing graded feedforward compensation. Simultaneously, it implements proactive feedforward prevention and control of hot delivery cracks.

Description

Technical Field

[0001] This application relates to the field of integrated production technology of continuous casting and rolling in iron and steel metallurgy, and more specifically, to a method and system for feedforward compensation of billet quality in the hot direct delivery process of cold heading steel continuous casting billets. Background Technology

[0002] Cold heading steel is the main material for making threaded fasteners. To achieve energy conservation and emission reduction, existing technologies often use hot direct delivery of continuously cast billets for production. Traditional hot direct delivery processing of cold heading steel billets has the following problems: First, it mainly focuses on single-point temperature measurement or limited-point geometric dimension detection of cold heading steel, failing to effectively form a multi-dimensional detection system for temperature, geometric dimensions, surface quality, and internal quality under harsh conditions of high temperature and high dust. Internal defects cannot be effectively detected, resulting in a single detection dimension. Second, existing feedforward compensation technologies are often limited to a single stage, such as feedforward for secondary cooling water distribution or rolling force feedforward, failing to effectively cover continuous casting, hot direct delivery, and subsequent rolling. The traditional feedforward compensation system, which mainly uses fixed compensation coefficients or relies solely on empirical judgment, cannot adapt to the dynamic changes in the quality fluctuations of continuously cast billets and the rolling conditions. Thirdly, it lacks effective feedforward control methods for hot-feeding crack defects. Cold heading steel is prone to surface cracks during hot feeding in the γ+α two-phase region. Traditional technologies mainly employ avoidance strategies such as slow offline cooling or rapid surface cooling, failing to form a feedforward-based proactive control system based on evolution prediction. Fourthly, traditional feedforward technologies are designed for general-purpose carbon steel or low-alloy steel, without targeted feedforward compensation based on the chemical composition characteristics, microstructure evolution, and cold heading performance requirements of cold heading steel, resulting in insufficient compensation effectiveness.

[0003] Effective technical solutions are urgently needed to address the above problems. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for feedforward compensation of billet quality in the hot direct delivery process of cold heading steel continuous casting billets. It can achieve accurate feedforward compensation of billet quality in the hot direct delivery process by means of multi-dimensional online quality monitoring and fusion, initial similarity matching by combining feedforward parameters based on historical experience, simulation analysis and self-correction by using a preset feedforward compensation optimization model and a preset hot direct delivery digital twin model, and perform hierarchical feedforward compensation. At the same time, it can actively prevent and control hot delivery cracks, thereby achieving accurate feedforward compensation of billet quality in the hot direct delivery process of cold heading steel continuous casting billets.

[0005] In the first aspect, this application provides a method for feedforward compensation of billet quality during the hot direct delivery process of cold heading steel continuous casting billets, including the following steps: Acquire real-time quality detection datasets of cold heading steel billets and perform analysis and processing to obtain multi-dimensional quality status assessment data of the billets; Based on the multidimensional quality status evaluation data of the billet, an initial set of feedforward compensation parameters is obtained by matching and analyzing the preset historical feedforward compensation parameter database. Based on the initial feedforward compensation parameter set and the multidimensional quality status evaluation data of the billet, the optimized feedforward compensation parameter set is obtained through analysis and processing using a preset feedforward compensation optimization model and a preset hot direct delivery digital twin model. Based on the optimized feedforward compensation parameter set and the multi-dimensional quality status evaluation data of the billet, hot direct feedforward compensation for cold heading steel continuous casting billets is performed.

[0006] Optionally, in the billet quality feedforward compensation method for the hot direct delivery process of cold heading steel continuous casting billets described in this application, the step of acquiring the real-time detection dataset of cold heading steel billet quality and performing analysis and processing to obtain multi-dimensional quality status assessment data of the billet includes: Obtain a real-time quality detection dataset for cold heading steel billets, including the billet's three-dimensional temperature field, cross-sectional geometric parameters, surface defect characteristics, and internal quality index prediction data. The three-dimensional temperature field of the billet, the geometric parameters of the billet cross section, the surface defect characteristic data of the billet, and the predicted data of the internal quality index of the billet are spatiotemporally registered and fused to obtain multi-dimensional quality status assessment data of the billet.

[0007] Optionally, in the billet quality feedforward compensation method for the hot direct delivery process of cold heading steel continuous casting billets described in this application, the step of obtaining an initial feedforward compensation parameter set by matching and analyzing the multi-dimensional quality status evaluation data of the billets through a preset historical feedforward compensation parameter database includes: The multidimensional quality status evaluation data of the billet is subjected to feature extraction to obtain the billet quality quantification feature vector. Based on the quantification feature vector of the billet quality, a similarity matching analysis is performed through a preset historical feedforward compensation parameter database to obtain a preset number of candidate feedforward compensation parameter sets and corresponding cosine similarities. The cosine similarity is squared and normalized to obtain the weight values ​​corresponding to the candidate feedforward compensation parameter set. The candidate feedforward compensation parameter set is weighted and fused according to the corresponding weight values ​​to obtain the initial feedforward compensation parameter set.

[0008] Optionally, in the billet quality feedforward compensation method for the hot direct delivery process of cold heading steel continuous casting billets described in this application, the step of analyzing and processing the initial feedforward compensation parameter set in conjunction with the multi-dimensional quality state evaluation data of the billet through a preset feedforward compensation optimization model and a preset hot direct delivery digital twin model to obtain an optimized feedforward compensation parameter set includes: The three-dimensional temperature field of the billet and the corresponding preset temperature compensation coefficient, the geometric parameters of the billet cross section and the corresponding preset size compensation coefficient, the surface defect characteristic data of the billet and the corresponding preset defect compensation strength coefficient, the predicted data of the internal quality index of the billet and the corresponding internal quality compensation coefficient, as well as the initial feedforward compensation parameter set and the preset cold heading steel material characteristic parameters are input into the preset feedforward compensation optimization model for analysis and processing to obtain the feedforward compensation decision parameter set. Based on the multidimensional quality status assessment data of the billet and the set of feedforward compensation decision parameters, the continuous casting side source compensation, the pre-compensation of the hot direct delivery process and the rolling compensation simulation analysis are performed through the preset hot direct delivery digital twin model to obtain the post-rolling quality prediction data. The post-rolling quality prediction data is compared with the preset quality target data to obtain the compensation deviation parameter; If the compensation deviation parameter is greater than the preset compensation deviation warning threshold, iterative adjustment is performed using the preset gradient descent algorithm, and cyclic monitoring is executed. If the compensation deviation parameter is less than or equal to the preset compensation deviation warning threshold, then the feedforward compensation decision parameter set is determined to be the optimized feedforward compensation parameter set.

[0009] Optionally, in the billet quality feedforward compensation method for the hot direct delivery process of cold heading steel continuous casting billets described in this application, the step of performing hot direct delivery feedforward compensation of cold heading steel continuous casting billets based on the optimized feedforward compensation parameter set and the multi-dimensional quality status evaluation data of the billets includes: The multidimensional quality status evaluation data of the billet is analyzed and processed to obtain feedforward compensation trigger commands, including continuous casting side source compensation commands, hot direct delivery process pre-compensation commands and rolling compensation commands. According to the continuous casting side source compensation instruction, the hot direct delivery process pre-compensation instruction, and the rolling compensation instruction, the hot direct delivery feedforward compensation of the cold heading steel continuous casting billet is executed according to the optimized feedforward compensation parameter set.

[0010] Optionally, the billet quality feedforward compensation method for the hot direct delivery process of cold heading steel continuous casting billets described in this application further includes: The roll reduction values ​​at the mill inlet and outlet are obtained and analyzed using a pre-defined recursive least squares method to obtain roll wear compensation parameters. The actual temperature of the roll is obtained and analyzed in combination with the preset thermal expansion coefficient to obtain the predicted reduction of roll gap. The optimized theoretical adjustment amount of the roll gap is extracted based on the optimized feedforward compensation parameter set, and then corrected based on the roll wear compensation parameters and the predicted reduction of the roll gap to obtain the optimized actual adjustment amount of the roll gap.

[0011] Optionally, the billet quality feedforward compensation method for the hot direct delivery process of cold heading steel continuous casting billets described in this application further includes: Data is extracted based on the three-dimensional temperature field of the billet to obtain the surface temperature of the preset billet node at a preset time point, and a real-time temperature-time curve is constructed. The cooling rate is obtained by analyzing the real-time temperature and time curves. The three-dimensional temperature field, cooling rate, and preset cold heading steel material characteristic parameters of the billet are input into the preset billet surface crack prediction model for analysis and processing to obtain the predicted value of the billet surface crack. The predicted value of surface cracks on the billet is compared with the preset threshold value for the occurrence of surface cracks on the billet, and a graded feedforward control strategy is obtained based on the threshold range that falls within it.

[0012] Secondly, this application provides a billet quality feedforward compensation system for the hot direct delivery process of cold-heading steel continuous casting billets. The system includes a memory and a processor. The memory contains a program for a billet quality feedforward compensation method for the hot direct delivery process of cold-heading steel continuous casting billets. When the program for the billet quality feedforward compensation method for the hot direct delivery process of cold-heading steel continuous casting billets is executed by the processor, it implements the following steps: Acquire real-time quality detection datasets of cold heading steel billets and perform analysis and processing to obtain multi-dimensional quality status assessment data of the billets; Based on the multidimensional quality status evaluation data of the billet, an initial set of feedforward compensation parameters is obtained by matching and analyzing the preset historical feedforward compensation parameter database. Based on the initial feedforward compensation parameter set and the multidimensional quality status evaluation data of the billet, the optimized feedforward compensation parameter set is obtained through analysis and processing using a preset feedforward compensation optimization model and a preset hot direct delivery digital twin model. Based on the optimized feedforward compensation parameter set and the multi-dimensional quality status evaluation data of the billet, hot direct feedforward compensation for cold heading steel continuous casting billets is performed.

[0013] Optionally, in the billet quality feedforward compensation system for the hot direct delivery process of cold heading steel continuous casting billets described in this application, the step of acquiring the real-time detection dataset of cold heading steel billet quality and performing analysis and processing to obtain multi-dimensional quality status assessment data of the billet includes: Obtain a real-time quality detection dataset for cold heading steel billets, including the billet's three-dimensional temperature field, cross-sectional geometric parameters, surface defect characteristics, and internal quality index prediction data. The three-dimensional temperature field of the billet, the geometric parameters of the billet cross section, the surface defect characteristic data of the billet, and the predicted data of the internal quality index of the billet are spatiotemporally registered and fused to obtain multi-dimensional quality status assessment data of the billet.

[0014] Optionally, in the billet quality feedforward compensation system for the hot direct delivery process of cold heading steel continuous casting billets described in this application, the step of obtaining an initial feedforward compensation parameter set by matching and analyzing the multi-dimensional quality status evaluation data of the billets through a preset historical feedforward compensation parameter database includes: The multidimensional quality status evaluation data of the billet is subjected to feature extraction to obtain the billet quality quantification feature vector. Based on the quantification feature vector of the billet quality, a similarity matching analysis is performed through a preset historical feedforward compensation parameter database to obtain a preset number of candidate feedforward compensation parameter sets and corresponding cosine similarities. The cosine similarity is squared and normalized to obtain the weight values ​​corresponding to the candidate feedforward compensation parameter set. The candidate feedforward compensation parameter set is weighted and fused according to the corresponding weight values ​​to obtain the initial feedforward compensation parameter set.

[0015] As can be seen from the above, the billet quality feedforward compensation method and system provided in this application for the hot direct delivery process of cold heading steel billets achieves accurate feedforward compensation of billet quality during the hot direct delivery process by integrating multi-dimensional online quality monitoring and fusion, performing initial similarity matching based on feedforward parameters from historical experience, conducting simulation analysis and self-correction through a preset feedforward compensation optimization model and a preset hot direct delivery digital twin model, and performing graded feedforward compensation. At the same time, it actively prevents and controls hot delivery cracks, thereby achieving accurate feedforward compensation of billet quality during the hot direct delivery process of cold heading steel billets.

[0016] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing embodiments of this application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart of a billet quality feedforward compensation method for the hot direct delivery process of cold heading steel continuous casting billets provided in this application embodiment; Figure 2 A flowchart illustrating the process of obtaining the optimized feedforward compensation parameter set for the hot direct delivery process of cold heading steel continuous casting billet provided in this application embodiment; Figure 3The flowchart illustrates the process for obtaining the optimized actual adjustment amount of the roll gap in the hot direct delivery process of cold heading steel continuous casting billet provided in this application embodiment. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0020] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] Please refer to Figure 1 , Figure 1 This is a flowchart of a billet quality feedforward compensation method for the hot direct delivery process of cold-heading steel continuous casting billets, as described in some embodiments of this application. This billet quality feedforward compensation method for the hot direct delivery process of cold-heading steel continuous casting billets is used in terminal devices, such as computers and mobile terminals. The billet quality feedforward compensation method for the hot direct delivery process of cold-heading steel continuous casting billets includes the following steps: S11. Obtain the real-time detection dataset of cold heading steel billet quality, and perform analysis and processing to obtain multi-dimensional quality status assessment data of the billet. S12. Based on the multidimensional quality status evaluation data of the billet, a matching analysis is performed through a preset historical feedforward compensation parameter database to obtain an initial feedforward compensation parameter set. S13. Based on the initial feedforward compensation parameter set and the multidimensional quality status evaluation data of the billet, the optimized feedforward compensation parameter set is obtained by analyzing and processing the data through the preset feedforward compensation optimization model and the preset hot direct delivery digital twin model. S14. Perform hot direct feedforward compensation for cold heading steel continuous casting billets based on the optimized feedforward compensation parameter set and the multi-dimensional quality status evaluation data of the billets.

[0022] It should be noted that, in order to achieve accurate feedforward compensation for cold heading steel continuous casting billets, firstly, multi-dimensional data including temperature, geometric dimensions, surface defects, and internal quality are simultaneously collected and dynamically fused to construct a complete digital profile of the billet quality. Then, an initial set of feedforward compensation parameters is selected based on historical experience, and optimized through a preset feedforward compensation optimization model and a preset direct heat transfer digital twin model to construct a three-level collaborative compensation architecture for the continuous casting side, direct heat transfer, and rolling process. Real-time corrections are also made by comprehensively considering roll wear and thermal expansion. The preset direct heat transfer digital twin model is based on the finite element-discrete element coupling method and includes a heat transfer module. (Solve the Fourier heat conduction equation, input is a three-dimensional temperature field, output is temperature evolution), phase transformation module (based on the JMAK equation, input is cooling rate and steel composition, output is phase transformation volume fraction), deformation module (based on rigid-plastic finite element method, input is rolling force and roll gap, output is post-rolling dimension prediction), boundary conditions (inlet temperature, cooling water volume, rolling speed, etc.) are collected in real time by sensors, and the boundary condition parameters are updated every 10ms. The model is calibrated offline using the gradient descent method. Using historical production data as samples, parameters such as heat transfer coefficient and friction coefficient are optimized to minimize the deviation between simulation output and measured results.

[0023] According to an embodiment of the present invention, the step of acquiring a real-time quality detection dataset of cold heading steel billets and performing analysis and processing to obtain multi-dimensional quality status assessment data of the billets includes: Obtain a real-time quality detection dataset for cold heading steel billets, including the billet's three-dimensional temperature field, cross-sectional geometric parameters, surface defect characteristics, and internal quality index prediction data. The three-dimensional temperature field of the billet, the geometric parameters of the billet cross section, the surface defect characteristic data of the billet, and the predicted data of the internal quality index of the billet are spatiotemporally registered and fused to obtain multi-dimensional quality status assessment data of the billet.

[0024] It should be noted that the cold heading steel billet is divided into a head section, a middle section, and a tail section according to a preset length, and a preset number of matrix infrared temperature sensors and infrared thermal imagers are arranged along the direct heat delivery direction. The thermal imagers collect surface temperature images of the billet at a preset frequency, evaluate the surface temperature distribution, and input them into a preset solidification heat transfer model (obtained by training with a large number of historical samples of surface temperature and corresponding billet core temperature). The billet core temperature is then calculated, and a coordinate system is established with the end of the billet head section as the origin. The determined temperature is mapped to the corresponding node, thereby constructing a three-dimensional temperature field of the billet. An example is shown in Table 1. Table 1

[0025] Dual-wavelength (660nm and 980nm, respectively for high-precision contour measurement and signal calibration) line laser contour scanners are arranged on both sides of the hot direct conveying roller. Based on the preset line laser triangulation principle, the complete cross-sectional contour is collected at every preset length, and the width, thickness, bulge and curvature are extracted. See Table 2 for examples. Table 2

[0026] A predetermined number of CCD arrays are arranged circumferentially at the hot direct delivery inlet to acquire images of the billet surface. After preprocessing, defects are identified and classified based on a predetermined deep learning algorithm, and the defect quantification value is reconstructed in three dimensions to obtain the defect feature data of the billet surface. See Table 3 for an example. Table 3

[0027] Extracting parameters including casting speed, superheat, water volume in each secondary cooling stage, and electromagnetic stirring parameters from the casting process data, a pre-trained PSO-LSTM neural network is used. The input parameters are casting speed, superheat, water volume in each secondary cooling stage, and electromagnetic stirring current / frequency. The output parameters are center segregation index, center porosity level, and shrinkage tendency index. The network structure is a three-layer LSTM with 128-64-32 hidden layer nodes. The particle swarm optimization parameters are set to a population size of 30 and 100 iterations. The coupling logic uses low-magnification test results as labels and adopts the mean square error loss function to output predicted data of the internal quality indicators of the billet. See Table 4 for an example. Table 4

[0028] Deep learning is used to uncover the deep nonlinear correlation between process parameters and internal defects, thereby improving detection accuracy and efficiency.

[0029] According to an embodiment of the present invention, the step of obtaining an initial feedforward compensation parameter set by matching and analyzing the multidimensional quality state evaluation data of the cast billet through a preset historical feedforward compensation parameter database includes: The multidimensional quality status evaluation data of the billet is subjected to feature extraction to obtain the billet quality quantification feature vector. Based on the quantification feature vector of the billet quality, a similarity matching analysis is performed through a preset historical feedforward compensation parameter database to obtain a preset number of candidate feedforward compensation parameter sets and corresponding cosine similarities. The cosine similarity is squared and normalized to obtain the weight values ​​corresponding to the candidate feedforward compensation parameter set. The candidate feedforward compensation parameter set is weighted and fused according to the corresponding weight values ​​to obtain the initial feedforward compensation parameter set.

[0030] It should be noted that, based on successful cases with a cold upsetting pass rate ≥98% in historical production data, the multidimensional quality status data of each billet is input into an autoencoder for unsupervised training, compressed into a 16-dimensional quality fingerprint vector, and stored in association with the successful compensation strategy corresponding to that billet to construct a preset historical feedforward compensation parameter database. Similarity matching is performed based on the fingerprint vector of the real-time acquired multidimensional quality status evaluation data of the billet (i.e., the low-dimensional, high-information-density quantized feature vector obtained after feature extraction: billet quality quantified feature vector). In this embodiment, cosine similarity (focusing on the direction of the vector rather than its modulus) is selected. It is more suitable for measuring the shape similarity of fingerprint vectors than the magnitude of their amplitude. The weight values ​​of the candidate feedforward compensation parameters are determined and then weighted and fused to form the initial set of feedforward compensation parameters. These parameters include the percentage of water adjustment corresponding to the preset section of the secondary cooling zone on the continuous casting side, the power setting value and conveying speed adjustment percentage of each section of the induction heating on the hot delivery side, and the adjustment amount of the roll gap, the percentage of the rolling speed adjustment, the percentage of the reduction increase, the adjustment amount of the rolling force, the adjustment amount of the tension, and the adjustment amount of the bending force on the rolling side. It makes full use of the historically accumulated production experience, avoids re-execution of decisions, and improves the calculation speed to meet real-time requirements.

[0031] Please refer to Figure 2 , Figure 2 This is a flowchart illustrating the process of obtaining an optimized feedforward compensation parameter set for the hot direct delivery process of cold heading steel continuous casting billets in some embodiments of this application. According to embodiments of the present invention, the step of obtaining an optimized feedforward compensation parameter set by analyzing and processing the initial feedforward compensation parameter set in conjunction with the multi-dimensional quality status evaluation data of the billet through a preset feedforward compensation optimization model and a preset hot direct delivery digital twin model includes: S21. Input the three-dimensional temperature field of the billet and the corresponding preset temperature compensation coefficient, the geometric parameters of the billet cross section and the corresponding preset size compensation coefficient, the surface defect characteristic data of the billet and the corresponding preset defect compensation strength coefficient, the predicted data of the internal quality index of the billet and the corresponding internal quality compensation coefficient, as well as the initial feedforward compensation parameter set and the preset cold heading steel material characteristic parameters into the preset feedforward compensation optimization model for analysis and processing to obtain the feedforward compensation decision parameter set; S22. Based on the multidimensional quality status assessment data of the billet and the feedforward compensation decision parameter set, the continuous casting side source compensation, the hot direct delivery process pre-compensation and the rolling compensation simulation analysis are performed through the preset hot direct delivery digital twin model to obtain the post-rolling quality prediction data. S23. Compare the post-rolling quality prediction data with the preset quality target data to obtain the compensation deviation parameter; S241. If the compensation deviation parameter is greater than the preset compensation deviation warning threshold, iterative adjustment is performed using the preset gradient descent algorithm, and cyclic monitoring is executed. S242. If the compensation deviation parameter is less than or equal to the preset compensation deviation warning threshold, then the feedforward compensation decision parameter set is determined to be the optimized feedforward compensation parameter set.

[0032] It should be noted that after determining the initial feedforward compensation parameter set according to historical experience, further refined optimization and dynamic adaptive adjustment are carried out; according to the initial feedforward compensation parameter set, the three-dimensional temperature field of the billet, the geometric parameters of the billet cross-section, the characteristic data of the billet surface defects, and the predicted data of the internal quality indexes of the billet, combined with the preset cold heading steel material characteristic parameters (including composition, phase transformation point, and thermoplastic curve parameters), multi-objective optimization and non-dominated sorting genetic algorithm solution analysis are carried out to minimize the exit size deviation, maximize the predicted value of the cold upsetting qualification rate, and minimize the risk of rolling force exceeding the limit, so as to obtain the feedforward compensation decision parameter set. For the preset temperature compensation coefficient, preset size compensation coefficient, preset defect compensation intensity coefficient, and internal quality compensation coefficient in the preset feedforward compensation optimization model (obtained by training with the three-dimensional temperature field of the billet and the corresponding preset temperature compensation coefficient, the geometric parameters of the billet cross-section and the corresponding preset size compensation coefficient, the characteristic data of the billet surface defects and the corresponding preset defect compensation intensity coefficient, the predicted data of the internal quality indexes of the billet and the corresponding internal quality compensation coefficient, the initial feedforward compensation parameter set, and the preset cold heading steel material characteristic parameters and the corresponding feedforward compensation decision parameter set), by introducing the SAC (Soft Actor-Critic) deep reinforcement learning algorithm, the dynamic adaptive adjustment of the compensation coefficient is realized. The structures of the policy network and the value network are both three-layer fully connected, 256-128-64. The state space is 8 types of features, a total of 38 dimensions. The action space includes 6 compensation coefficients, each with boundaries. The specific expression of the reward function includes four items: dimensional accuracy, cold upsetting qualification rate, rolling force safety, and energy consumption penalty. Among them, the energy consumption refers to the energy consumption increment calculated by comparing the real-time measured power consumption of the induction reheating device with the historical reference value through the watt-hour meter. The data is normalized to eliminate the dimension difference, and the weight coefficients are set to 0.5, 0.2, 0.2, and 0.1 respectively, and those skilled in the art can dynamically adjust according to the application situation. Using 10,000 billets of historical production data, the training set / validation set is divided according to the time series. The warm start mechanism is that the initial policy network parameters are pre-trained through supervised learning with the output of the feedforward compensation decision parameter set as the target, that is, the three-dimensional temperature field of the billet, the geometric parameters of the billet cross-section, the characteristic data of the billet surface defects, the predicted data of the internal quality indexes of the billet, and the real-time collected rolling conditions (roll wear, bearing temperature, real-time rolling force) are used as the state input, and a reward function is constructed with maximizing the cold upsetting qualification rate as the core optimization target to construct a SAC deep reinforcement learning agent based on the maximum entropy framework, and the dynamically adaptively adjusted compensation coefficient is output. The dynamic adaptive adjustment of the compensation coefficient in this embodiment is realized by inputting the initial compensation strategy as the warm start initial value into the constructed agent. Among them, the preset gradient descent algorithm is used for iterative adjustment, and the maximum number of iterations is 50 times, or the iteration is terminated when the change rate of the compensation deviation parameter is less than 1% for three consecutive times.

[0033] According to an embodiment of the present invention, the step of performing hot direct feedforward compensation for cold heading steel continuous casting billets based on the optimized feedforward compensation parameter set combined with the multi-dimensional quality state evaluation data of the casting billets includes: The multidimensional quality status evaluation data of the billet is analyzed and processed to obtain feedforward compensation trigger commands, including continuous casting side source compensation commands, hot direct delivery process pre-compensation commands and rolling compensation commands. According to the continuous casting side source compensation instruction, the hot direct delivery process pre-compensation instruction, and the rolling compensation instruction, the hot direct delivery feedforward compensation of the cold heading steel continuous casting billet is executed according to the optimized feedforward compensation parameter set.

[0034] It should be noted that, in order to parse and distribute the optimized feedforward compensation parameter set to the continuous casting side, the hot direct delivery process, and the rolling stage, corresponding triggering conditions are constructed. The source triggering conditions on the continuous casting side include: current billet surface temperature deviation > ±15℃ (instruction for adjusting water volume in the corresponding section of the secondary cooling zone); current billet core-to-surface temperature difference > 80℃ (instruction for increasing water volume in the later section of the secondary cooling zone to enhance surface cooling); and temperature gradient change rate > 30℃ / m (instruction for adjusting the water volume distribution ratio in each section of the secondary cooling zone). The triggering conditions for the hot direct delivery process include: predicted entry temperature below the lower limit of 90℃. 0℃ (start the induction heating device, set the power and frequency), head-to-tail temperature difference > 50℃ (instruction is segmented induction heating: only heat the tail section with a lower temperature), actual inlet temperature deviates from the predicted value beyond the threshold (instruction is fine-tuning the induction heating power); rolling trigger conditions include cross-sectional size deviation exceeding tolerance (instruction is adjusting the roll gap), presence of surface defects (instruction is mapping the defect location + adjusting the local rolling speed and rolling force), internal quality prediction exceeding the standard (instruction is increasing the rolling reduction by 5-15%), and uneven billet temperature (instruction is adjusting the rolling speed or rolling temperature).

[0035] Please refer to Figure 3 , Figure 3 This is a flowchart illustrating the process of obtaining the optimized roll gap actual adjustment amount in the hot direct delivery method for cold heading steel continuous casting billets in some embodiments of this application. According to embodiments of the present invention, it further includes: S311. Obtain the roll reduction values ​​at the mill inlet and the roll reduction values ​​at the mill outlet, and analyze and process them using a preset recursive least squares method to obtain roll wear compensation parameters. S312. Obtain the measured temperature of the roll and analyze it in conjunction with the preset thermal expansion coefficient to obtain the predicted reduction in roll gap. S32. Extract the theoretical adjustment amount of the optimized roll gap based on the optimized feedforward compensation parameter set, and perform correction processing based on the roll wear compensation parameters and the predicted reduction of roll gap to obtain the actual adjustment amount of the optimized roll gap.

[0036] It should be noted that, in order to improve the impact of roll gap wear and thermal expansion on roll gap accuracy during the rolling process, laser rangefinders are installed at the mill inlet and outlet to measure the actual reduction in real time. The recursive least squares method is used to estimate the roll wear compensation parameters online (which refers to the equivalent roll gap increment caused by roll wear). The predicted roll gap reduction caused by thermal expansion (which refers to the equivalent roll gap reduction caused by roll thermal expansion) is calculated by combining the bearing housing temperature sensor and the coefficient of thermal expansion. The theoretical compensation is comprehensively corrected by summing the optimized theoretical roll gap adjustment amount in the optimized feedforward compensation parameter set with the roll wear compensation parameters and then subtracting the predicted roll gap reduction.

[0037] According to an embodiment of the present invention, it further includes: Data is extracted based on the three-dimensional temperature field of the billet to obtain the surface temperature of the preset billet node at a preset time point, and a real-time temperature-time curve is constructed. The cooling rate is obtained by analyzing the real-time temperature and time curves. The three-dimensional temperature field, cooling rate, and preset cold heading steel material characteristic parameters of the billet are input into the preset billet surface crack prediction model for analysis and processing to obtain the predicted value of the billet surface crack. The predicted value of surface cracks on the billet is compared with the preset threshold value for the occurrence of surface cracks on the billet, and a graded feedforward control strategy is obtained based on the threshold range that falls within it.

[0038] It should be noted that the pre-defined billet surface crack prediction model consists of two coupled sub-models: a phase transformation prediction model and a second-phase precipitation model. The phase transformation prediction model is based on the JMAK equation. Phase transformation variables versus time curves at different cooling rates are obtained through thermal expansion experiments. The JMAK equation is used to regress and fit the isothermal phase transformation rate constant and Avramie index corresponding to the calibration temperature T, determining the phase transformation contribution probability. The second-phase precipitation model is based on the NbC / TiN solubility product equation and JMAK precipitation kinetics, outputting the precipitation volume fraction to determine the precipitation contribution probability. The coupled output of the two sub-models is the probability P of surface crack occurrence on the billet. c=min(1,w1*P1+w2*P2), with weights w1=0.4, w2=0.6, P1 being the phase transformation contribution probability, and P2 being the precipitation contribution probability. The inputs are the three-dimensional temperature field and cooling rate, respectively evaluating the contribution of microstructure transformation and second-phase precipitation to cracking. Based on a large number of historical samples, the three-dimensional temperature field of the continuously cast billet is obtained for analysis, and the cooling rate is obtained. The three-dimensional temperature field and cooling rate of the continuously cast billet are combined with the preset cold heading steel material characteristic parameters (chemical composition, phase transformation characteristic temperature, and second-phase precipitation kinetic parameters) to establish an austenite-ferrite phase transformation JMAK kinetic model (i.e., phase transformation prediction model) and a NbC, TiN second-phase precipitation thermodynamic-kinetic coupling model (i.e., second-phase precipitation). The model calculates the predicted value of surface cracks on the billet in real time, which is used to represent the probability of entering the third brittle zone (700 to 1000°C). In this embodiment, when the predicted value of surface cracks on the billet is <40%, it is judged as no risk and continuous monitoring is performed. When it falls into [40%, 60%], it is judged as low risk and the conveying speed is increased by 10% to 15% to shorten the residence time of the billet in the two-phase zone. If it falls into (60%, 80%), it is judged as medium risk and the induction heating device is used to quickly heat the surface of the billet at a preset heating rate so that it can quickly jump out of the two-phase zone. If the predicted value of surface cracks on the billet is >80%, it is judged as high risk and the surface is rapidly cooled and combined with induction heating for control.

[0039] It is worth mentioning that, according to embodiments of the present invention, it further includes: Obtain the measured quality data after rolling and compare it with the predicted quality data after rolling to obtain deviation data, including the predicted deviation of cold upsetting pass rate and the dimensional compensation deviation. The predicted deviation and dimensional compensation deviation of the cold upsetting pass rate are analyzed and processed by a preset gradient boosting tree algorithm to obtain a list of characteristic influence deviations and identify high-frequency deviation types. Based on the feature influence deviation list and high-frequency deviation type, the weight parameters of the preset feedforward compensation optimization model are adjusted to obtain the feedforward compensation correction model.

[0040] It should be noted that the acquired post-rolling quality prediction data and the cold upsetting test results of post-rolling products are correlated by batch to construct a deviation database containing prediction deviations. Then, the Gradient Boosting Tree (XGBoost) algorithm is used to mine the deep correlation rules between deviations and billet quality characteristics. Combined with SHAP interpretability analysis, the influence degree of each quality characteristic on the prediction deviation is ranked, and high-frequency deviation types are identified (such as the predicted value being too large when the segregation index is too high). High-frequency deviation types refer to deviation patterns that occur more frequently than a preset threshold (such as 10%) in the deviation database and are strongly correlated with specific quality characteristic combinations. Based on the above analysis results, the weight parameters (including compensation coefficients and digital twin model parameters) of the feedforward compensation decision model are updated offline weekly.

[0041] It is worth mentioning that, according to embodiments of the present invention, it further includes: The system acquires real-time rolling force, workpiece geometry parameters, roll parameters, and preset strip deformation resistance, and inputs them into a preset roll elastic flattening amount prediction model for analysis and processing to obtain roll elastic flattening amount change data. The rolling process parameters, roll material parameters, preset strip material parameters, preset friction coefficient and initial roll temperature are obtained, and the preset strip deformation resistance is input into the preset roll surface instantaneous thermal bulge change prediction model for analysis and processing to obtain the roll surface instantaneous thermal bulge change data. The system acquires the structural parameters of the supporting bearing, real-time roll offset, bearing housing and frame clearance, bearing temperature and cumulative bearing wear, and analyzes and processes them in conjunction with the real-time rolling force input to obtain equivalent roll gap compensation data by using a preset bearing clearance influence roll gap prediction model. The actual adjustment amount of the optimized roll gap is corrected based on the data of the change in the elastic flattening of the roll, the data of the instantaneous thermal convexity of the roll surface, and the data of the equivalent roll gap compensation, so as to obtain the actual adjustment amount of the roll gap.

[0042] It should be noted that during the rolling process, the multi-physics coupling effect at the roll-strip interface dynamically affects the roll gap accuracy. This includes factors such as roll elastic deformation caused by rolling force fluctuations, instantaneous thermal crown changes due to frictional heat between the roll and strip, and roll spatial position shifts caused by changes in roll bearing clearance. In this embodiment, rolling force sensors are arranged on the operating and driving sides of the mill to collect real-time rolling force. The geometric parameters of the workpiece include the inlet thickness, outlet thickness, and workpiece width. The roll parameters include roll geometric parameters (work roll radius, support roll radius, work roll crown, and support roll crown) and roll mechanical parameters (bending force and intermediate roll shifting). The preset strip deformation resistance is based on the steel grade. Based on the determined characteristics and measured temperature, the prediction model for the pre-set roll elastic flattening amount uses the influence function method to numerically solve for the elastic deformation of the roll system. This model is trained by acquiring a large amount of historical data on real-time rolling force, workpiece geometry parameters, roll parameters, pre-set strip deformation resistance, and corresponding roll elastic flattening amount variations. Rolling process parameters include rolling speed, real-time rolling force, forward slip ratio, emulsion temperature, and emulsion flow rate. Roll material parameters include roll thermal conductivity, specific heat capacity, density, and coefficient of thermal expansion. Pre-set strip material parameters include specific heat capacity and density. The pre-set friction coefficient refers to the friction coefficient under current lubrication conditions. The initial roll temperature refers to the ambient temperature or the temperature from the previous rolling operation. The temperature field at the end of the test is predicted using a two-dimensional finite difference method as the core of the instantaneous thermal convexity change prediction model on the roll surface. This model is trained by acquiring a large amount of historical data on rolling process parameters, roll material parameters, preset strip material parameters, preset friction coefficient, initial roll temperature, preset strip deformation resistance, and corresponding instantaneous thermal convexity change data on the roll surface. Support bearing structural parameters include bearing type, number of rolling elements, rolling element diameter, bearing pitch circle diameter, and bearing width. Real-time roll offset includes the roll offset in both horizontal and vertical directions. The clearance between the bearing housing and the frame is monitored in real-time by a clearance sensor, and the bearing temperature is determined by the bearing... Temperature sensors embedded in the bearing housing collect data, and the cumulative bearing wear value is estimated online using the recursive least squares method. A pre-set bearing clearance influence roll gap prediction model is established based on the bearing dynamics equation and Hertzian contact theory, creating a dynamic mapping relationship between bearing clearance, wear, and roll spatial position offset. This model is trained by acquiring a large number of historical samples of supporting bearing structural parameters, real-time roll offset, bearing housing and frame clearance values, bearing temperature, cumulative bearing wear value, real-time rolling force, and corresponding equivalent roll gap compensation data. Finally, the actual roll gap adjustment is optimized by adding the equivalent roll gap compensation data and subtracting the roll elastic flattening change data and instantaneous thermal convexity change data on the roll surface for dynamic optimization.

[0043] This invention also discloses a billet quality feedforward compensation system for the hot direct delivery process of cold-heading steel continuous casting billets, including a memory and a processor. The memory includes a billet quality feedforward compensation method program for the hot direct delivery process of cold-heading steel continuous casting billets. When the processor executes the billet quality feedforward compensation method program for the hot direct delivery process of cold-heading steel continuous casting billets, it implements the following steps: Acquire real-time quality detection datasets of cold heading steel billets and perform analysis and processing to obtain multi-dimensional quality status assessment data of the billets; Based on the multidimensional quality status evaluation data of the billet, an initial set of feedforward compensation parameters is obtained by matching and analyzing the preset historical feedforward compensation parameter database. Based on the initial feedforward compensation parameter set and the multidimensional quality status evaluation data of the billet, the optimized feedforward compensation parameter set is obtained through analysis and processing using a preset feedforward compensation optimization model and a preset hot direct delivery digital twin model. Based on the optimized feedforward compensation parameter set and the multi-dimensional quality status evaluation data of the billet, hot direct feedforward compensation for cold heading steel continuous casting billets is performed.

[0044] It should be noted that, in order to achieve accurate feedforward compensation for cold heading steel continuous casting billets, firstly, multi-dimensional data including temperature, geometric dimensions, surface defects, and internal quality are simultaneously collected and dynamically fused to construct a complete digital profile of the billet quality. Then, an initial set of feedforward compensation parameters is selected based on historical experience, and optimized through a preset feedforward compensation optimization model and a preset direct heat transfer digital twin model to construct a three-level collaborative compensation architecture for the continuous casting side, direct heat transfer, and rolling process. Real-time corrections are also made by comprehensively considering roll wear and thermal expansion. The preset direct heat transfer digital twin model is based on the finite element-discrete element coupling method and includes a heat transfer module. (Solve the Fourier heat conduction equation, input is a three-dimensional temperature field, output is temperature evolution), phase transformation module (based on the JMAK equation, input is cooling rate and steel composition, output is phase transformation volume fraction), deformation module (based on rigid-plastic finite element method, input is rolling force and roll gap, output is post-rolling dimension prediction), boundary conditions (inlet temperature, cooling water volume, rolling speed, etc.) are collected in real time by sensors, and the boundary condition parameters are updated every 10ms. The model is calibrated offline using the gradient descent method. Using historical production data as samples, parameters such as heat transfer coefficient and friction coefficient are optimized to minimize the deviation between simulation output and measured results.

[0045] According to an embodiment of the present invention, the step of acquiring a real-time quality detection dataset of cold heading steel billets and performing analysis and processing to obtain multi-dimensional quality status assessment data of the billets includes: Obtain a real-time quality detection dataset for cold heading steel billets, including the billet's three-dimensional temperature field, cross-sectional geometric parameters, surface defect characteristics, and internal quality index prediction data. The three-dimensional temperature field of the billet, the geometric parameters of the billet cross section, the surface defect characteristic data of the billet, and the predicted data of the internal quality index of the billet are spatiotemporally registered and fused to obtain multi-dimensional quality status assessment data of the billet.

[0046] It should be noted that the cold heading steel billet is divided into a head section, a middle section, and a tail section according to a preset length. A preset number of matrix infrared temperature sensors and infrared thermal imagers are arranged along the direct heat delivery direction. The thermal imagers collect surface temperature images of the billet at a preset frequency, evaluate the surface temperature distribution, and input them into a preset solidification heat transfer model (obtained by training with a large number of historical samples of surface temperature and corresponding billet core temperature). The billet core temperature is then calculated, and a coordinate system is established with the end of the billet head section as the origin. The determined temperature is mapped to the corresponding nodes, thereby constructing a three-dimensional temperature field of the billet. An example is shown in Table 1. Dual-wavelength (660nm and 980nm, used for high-precision profile measurement and signal calibration, respectively) line laser profile scanners are arranged on both sides of the direct heat delivery raceway. Based on the preset line laser triangulation principle, complete cross-sectional profiles are collected at every preset length, and the width, thickness, bulge, and curvature are extracted. An example is shown in Table 2. A preset number of C-type infrared sensors are arranged circumferentially at the direct heat delivery inlet. A CD array is used to acquire images of the billet surface. After preprocessing, defects are identified and classified based on a preset deep learning algorithm, and 3D digital reconstruction of defect quantification values ​​is performed to obtain surface defect feature data of the billet. Examples are shown in Table 3. From the casting process records, parameters including casting speed, superheat, water volume in each secondary cooling stage, and electromagnetic stirring parameters are extracted. Based on a pre-trained PSO-LSTM neural network, the input parameters are casting speed, superheat, water volume in each secondary cooling stage, and electromagnetic stirring current / frequency. The output parameters are center segregation index, center porosity level, and shrinkage tendency index. The network structure is a three-layer LSTM with 128-64-32 hidden layer nodes. The particle swarm optimization parameters are set to a population size of 30 and 100 iterations. The coupling logic uses low-magnification test results as labels and employs the mean square error loss function to output predicted data of the billet's internal quality indicators. Examples are shown in Table 4. Deep learning is used to mine the deep nonlinear correlation between process parameters and internal defects, improving detection accuracy and efficiency.

[0047] According to an embodiment of the present invention, the step of obtaining an initial feedforward compensation parameter set by matching and analyzing the multidimensional quality state evaluation data of the cast billet through a preset historical feedforward compensation parameter database includes: The multidimensional quality status evaluation data of the billet is subjected to feature extraction to obtain the billet quality quantification feature vector. Based on the quantification feature vector of the billet quality, a similarity matching analysis is performed through a preset historical feedforward compensation parameter database to obtain a preset number of candidate feedforward compensation parameter sets and corresponding cosine similarities. The cosine similarity is squared and normalized to obtain the weight values ​​corresponding to the candidate feedforward compensation parameter set. The candidate feedforward compensation parameter set is weighted and fused according to the corresponding weight values ​​to obtain the initial feedforward compensation parameter set.

[0048] It should be noted that, based on successful cases with a cold upsetting pass rate ≥98% in historical production data, the multidimensional quality status data of each billet is input into an autoencoder for unsupervised training, compressed into a 16-dimensional quality fingerprint vector, and stored in association with the successful compensation strategy corresponding to that billet to construct a preset historical feedforward compensation parameter database. Similarity matching is performed based on the fingerprint vector of the real-time acquired multidimensional quality status evaluation data of the billet (i.e., the low-dimensional, high-information-density quantized feature vector obtained after feature extraction: billet quality quantified feature vector). In this embodiment, cosine similarity (focusing on the direction of the vector rather than its modulus) is selected. It is more suitable for measuring the shape similarity of fingerprint vectors than the magnitude of their amplitude. The weight values ​​of the candidate feedforward compensation parameters are determined and then weighted and fused to form the initial set of feedforward compensation parameters. These parameters include the percentage of water adjustment corresponding to the preset section of the secondary cooling zone on the continuous casting side, the power setting value and conveying speed adjustment percentage of each section of the induction heating on the hot delivery side, and the adjustment amount of the roll gap, the percentage of the rolling speed adjustment, the percentage of the reduction increase, the adjustment amount of the rolling force, the adjustment amount of the tension, and the adjustment amount of the bending force on the rolling side. It makes full use of the historically accumulated production experience, avoids re-execution of decisions, and improves the calculation speed to meet real-time requirements.

[0049] According to an embodiment of the present invention, the step of analyzing and processing the initial feedforward compensation parameter set in conjunction with the multidimensional quality state evaluation data of the cast billet through a preset feedforward compensation optimization model and a preset hot direct delivery digital twin model to obtain an optimized feedforward compensation parameter set includes: The three-dimensional temperature field of the billet and the corresponding preset temperature compensation coefficient, the geometric parameters of the billet cross section and the corresponding preset size compensation coefficient, the surface defect characteristic data of the billet and the corresponding preset defect compensation strength coefficient, the predicted data of the internal quality index of the billet and the corresponding internal quality compensation coefficient, as well as the initial feedforward compensation parameter set and the preset cold heading steel material characteristic parameters are input into the preset feedforward compensation optimization model for analysis and processing to obtain the feedforward compensation decision parameter set. Based on the multidimensional quality status assessment data of the billet and the set of feedforward compensation decision parameters, the continuous casting side source compensation, the pre-compensation of the hot direct delivery process and the rolling compensation simulation analysis are performed through the preset hot direct delivery digital twin model to obtain the post-rolling quality prediction data. The post-rolling quality prediction data is compared with the preset quality target data to obtain the compensation deviation parameter; If the compensation deviation parameter is greater than the preset compensation deviation warning threshold, iterative adjustment is performed using the preset gradient descent algorithm, and cyclic monitoring is executed. If the compensation deviation parameter is less than or equal to the preset compensation deviation warning threshold, then the feedforward compensation decision parameter set is determined to be the optimized feedforward compensation parameter set.

[0050] It should be noted that after determining the initial feedforward compensation parameter set according to historical experience, further refined optimization and dynamic adaptive adjustment are carried out; based on the initial feedforward compensation parameter set, the three-dimensional temperature field of the billet, the geometric parameters of the billet cross-section, the characteristic data of the billet surface defects, and the predicted data of the internal quality indicators of the billet, combined with the preset cold heading steel material characteristic parameters (including composition, phase transformation points, and thermoplastic curve parameters), multi-objective optimization and non-dominated sorting genetic algorithm solution analysis are carried out to minimize the outlet size deviation, maximize the predicted value of the cold upsetting qualification rate, and minimize the risk of rolling force exceeding the limit, so as to obtain the feedforward compensation decision parameter set. For the preset temperature compensation coefficient, preset size compensation coefficient, preset defect compensation intensity coefficient, and internal quality compensation coefficient in the preset feedforward compensation optimization model (obtained by training with the three-dimensional temperature field of the billet and the corresponding preset temperature compensation coefficient, the geometric parameters of the billet cross-section and the corresponding preset size compensation coefficient, the characteristic data of the billet surface defects and the corresponding preset defect compensation intensity coefficient, the predicted data of the internal quality indicators of the billet and the corresponding internal quality compensation coefficient, the initial feedforward compensation parameter set, and the preset cold heading steel material characteristic parameters and the corresponding feedforward compensation decision parameter set), by introducing the SAC (Soft Actor-Critic) deep reinforcement learning algorithm, the dynamic adaptive adjustment of the compensation coefficient is realized. The structures of the policy network and the value network are both three-layer fully connected, 256-128-64. The state space has 8 types of features, a total of 38 dimensions, and the action space includes 6 compensation coefficients, each with boundaries. The specific expression of the reward function includes four items: dimensional accuracy, cold upsetting qualification rate, rolling force safety, and energy consumption penalty. Among them, the energy consumption refers to the energy consumption increment calculated by comparing the real-time measured power consumption of the induction reheating device with the historical reference value through the watt-hour meter, and the data is normalized to eliminate the dimensional difference, and the weight coefficients are set to 0.5, 0.2, 0.2, and 0.1 respectively, and those skilled in the art can dynamically adjust according to the application situation. Using 10,000 billets of historical production data, the training set / validation set is divided according to the time series. The hot start mechanism is that the initial policy network parameters are pre-trained by supervised learning with the output of the feedforward compensation decision parameter set as the target, that is, the three-dimensional temperature field of the billet, the geometric parameters of the billet cross-section, the characteristic data of the billet surface defects, the predicted data of the internal quality indicators of the billet, and the real-time collected rolling conditions (roll wear, bearing temperature, real-time rolling force) are used as the state input, and a reward function is constructed with maximizing the cold upsetting qualification rate as the core optimization target to construct a SAC deep reinforcement learning agent based on the maximum entropy framework, and the dynamically adaptively adjusted compensation coefficient is output. The dynamic adaptive adjustment of the compensation coefficient in this embodiment is realized by inputting the initial compensation strategy as the hot start initial value into the constructed agent. Among them, the preset gradient descent algorithm is used for iterative adjustment, and the maximum number of iterations is 50 times, or the iteration is terminated when the change rate of the compensation deviation parameter is less than 1% for three consecutive times.

[0051] According to an embodiment of the present invention, the step of performing hot direct feedforward compensation for cold heading steel continuous casting billets based on the optimized feedforward compensation parameter set combined with the multi-dimensional quality state evaluation data of the casting billets includes: The multidimensional quality status evaluation data of the billet is analyzed and processed to obtain feedforward compensation trigger commands, including continuous casting side source compensation commands, hot direct delivery process pre-compensation commands and rolling compensation commands. According to the continuous casting side source compensation instruction, the hot direct delivery process pre-compensation instruction, and the rolling compensation instruction, the hot direct delivery feedforward compensation of the cold heading steel continuous casting billet is executed according to the optimized feedforward compensation parameter set.

[0052] It should be noted that, in order to parse and distribute the optimized feedforward compensation parameter set to the continuous casting side, the hot direct delivery process, and the rolling stage, corresponding triggering conditions are constructed. The source triggering conditions on the continuous casting side include: current billet surface temperature deviation > ±15℃ (instruction for adjusting water volume in the corresponding section of the secondary cooling zone); current billet core-to-surface temperature difference > 80℃ (instruction for increasing water volume in the later section of the secondary cooling zone to enhance surface cooling); and temperature gradient change rate > 30℃ / m (instruction for adjusting the water volume distribution ratio in each section of the secondary cooling zone). The triggering conditions for the hot direct delivery process include: predicted entry temperature below the lower limit of 90℃. 0℃ (start the induction heating device, set the power and frequency), head-to-tail temperature difference > 50℃ (instruction is segmented induction heating: only heat the tail section with a lower temperature), actual inlet temperature deviates from the predicted value beyond the threshold (instruction is fine-tuning the induction heating power); rolling trigger conditions include cross-sectional size deviation exceeding tolerance (instruction is adjusting the roll gap), presence of surface defects (instruction is mapping the defect location + adjusting the local rolling speed and rolling force), internal quality prediction exceeding the standard (instruction is increasing the rolling reduction by 5-15%), and uneven billet temperature (instruction is adjusting the rolling speed or rolling temperature).

[0053] According to an embodiment of the present invention, it further includes: The roll reduction values ​​at the mill inlet and outlet are obtained and analyzed using a pre-defined recursive least squares method to obtain roll wear compensation parameters. The actual temperature of the roll is obtained and analyzed in combination with the preset thermal expansion coefficient to obtain the predicted reduction of roll gap. The optimized theoretical adjustment amount of the roll gap is extracted based on the optimized feedforward compensation parameter set, and then corrected based on the roll wear compensation parameters and the predicted reduction of the roll gap to obtain the optimized actual adjustment amount of the roll gap.

[0054] It should be noted that, in order to improve the impact of roll gap wear and thermal expansion on roll gap accuracy during the rolling process, laser rangefinders are installed at the mill inlet and outlet to measure the actual reduction in real time. The recursive least squares method is used to estimate the roll wear compensation parameters online (which refers to the equivalent roll gap increment caused by roll wear). The predicted roll gap reduction caused by thermal expansion (which refers to the equivalent roll gap reduction caused by roll thermal expansion) is calculated by combining the bearing housing temperature sensor and the coefficient of thermal expansion. The theoretical compensation is comprehensively corrected by summing the optimized theoretical roll gap adjustment amount in the optimized feedforward compensation parameter set with the roll wear compensation parameters and then subtracting the predicted roll gap reduction.

[0055] According to an embodiment of the present invention, it further includes: Data is extracted based on the three-dimensional temperature field of the billet to obtain the surface temperature of the preset billet node at a preset time point, and a real-time temperature-time curve is constructed. The cooling rate is obtained by analyzing the real-time temperature and time curves. The three-dimensional temperature field, cooling rate, and preset cold heading steel material characteristic parameters of the billet are input into the preset billet surface crack prediction model for analysis and processing to obtain the predicted value of the billet surface crack. The predicted value of surface cracks on the billet is compared with the preset threshold value for the occurrence of surface cracks on the billet, and a graded feedforward control strategy is obtained based on the threshold range that falls within it.

[0056] It should be noted that the pre-defined billet surface crack prediction model consists of two coupled sub-models: a phase transformation prediction model and a second-phase precipitation model. The phase transformation prediction model is based on the JMAK equation. Phase transformation variables versus time curves at different cooling rates are obtained through thermal expansion experiments. The JMAK equation is used to regress and fit the isothermal phase transformation rate constant and Avramie index corresponding to the calibration temperature T, determining the phase transformation contribution probability. The second-phase precipitation model is based on the NbC / TiN solubility product equation and JMAK precipitation kinetics, outputting the precipitation volume fraction to determine the precipitation contribution probability. The coupled output of the two sub-models is the probability P of surface crack occurrence on the billet. c=min(1,w1*P1+w2*P2), with weights w1=0.4, w2=0.6, P1 being the phase transformation contribution probability, and P2 being the precipitation contribution probability. The inputs are the three-dimensional temperature field and cooling rate, respectively evaluating the contribution of microstructure transformation and second-phase precipitation to cracking. Based on a large number of historical samples, the three-dimensional temperature field of the continuously cast billet is obtained for analysis, and the cooling rate is obtained. The three-dimensional temperature field and cooling rate of the continuously cast billet are combined with the preset cold heading steel material characteristic parameters (chemical composition, phase transformation characteristic temperature, and second-phase precipitation kinetic parameters) to establish an austenite-ferrite phase transformation JMAK kinetic model (i.e., phase transformation prediction model) and a NbC, TiN second-phase precipitation thermodynamic-kinetic coupling model (i.e., second-phase precipitation). The model calculates the predicted value of surface cracks on the billet in real time, which is used to represent the probability of entering the third brittle zone (700 to 1000°C). In this embodiment, when the predicted value of surface cracks on the billet is <40%, it is judged as no risk and continuous monitoring is performed. When it falls into [40%, 60%], it is judged as low risk and the conveying speed is increased by 10% to 15% to shorten the residence time of the billet in the two-phase zone. If it falls into (60%, 80%), it is judged as medium risk and the induction heating device is used to quickly heat the surface of the billet at a preset heating rate so that it can quickly jump out of the two-phase zone. If the predicted value of surface cracks on the billet is >80%, it is judged as high risk and the surface is rapidly cooled and combined with induction heating for control.

[0057] It is worth mentioning that, according to embodiments of the present invention, it further includes: Obtain the measured quality data after rolling and compare it with the predicted quality data after rolling to obtain deviation data, including the predicted deviation of cold upsetting pass rate and the dimensional compensation deviation. The predicted deviation and dimensional compensation deviation of the cold upsetting pass rate are analyzed and processed by a preset gradient boosting tree algorithm to obtain a list of characteristic influence deviations and identify high-frequency deviation types. Based on the feature influence deviation list and high-frequency deviation type, the weight parameters of the preset feedforward compensation optimization model are adjusted to obtain the feedforward compensation correction model.

[0058] It should be noted that the acquired post-rolling quality prediction data and the cold upsetting test results of post-rolling products are correlated by batch to construct a deviation database containing prediction deviations. Then, the Gradient Boosting Tree (XGBoost) algorithm is used to mine the deep correlation rules between deviations and billet quality characteristics. Combined with SHAP interpretability analysis, the influence degree of each quality characteristic on the prediction deviation is ranked, and high-frequency deviation types are identified (such as the predicted value being too large when the segregation index is too high). High-frequency deviation types refer to deviation patterns that occur more frequently than a preset threshold (such as 10%) in the deviation database and are strongly correlated with specific quality characteristic combinations. Based on the above analysis results, the weight parameters (including compensation coefficients and digital twin model parameters) of the feedforward compensation decision model are updated offline weekly.

[0059] It is worth mentioning that, according to embodiments of the present invention, it further includes: The system acquires real-time rolling force, workpiece geometry parameters, roll parameters, and preset strip deformation resistance, and inputs them into a preset roll elastic flattening amount prediction model for analysis and processing to obtain roll elastic flattening amount change data. The rolling process parameters, roll material parameters, preset strip material parameters, preset friction coefficient and initial roll temperature are obtained, and the preset strip deformation resistance is input into the preset roll surface instantaneous thermal bulge change prediction model for analysis and processing to obtain the roll surface instantaneous thermal bulge change data. The system acquires the structural parameters of the supporting bearing, real-time roll offset, bearing housing and frame clearance, bearing temperature and cumulative bearing wear, and analyzes and processes them in conjunction with the real-time rolling force input to obtain equivalent roll gap compensation data by using a preset bearing clearance influence roll gap prediction model. The actual adjustment amount of the optimized roll gap is corrected based on the data of the change in the elastic flattening of the roll, the data of the instantaneous thermal convexity of the roll surface, and the data of the equivalent roll gap compensation, so as to obtain the actual adjustment amount of the roll gap.

[0060] It should be noted that during the rolling process, the multi-physics coupling effect at the roll-strip interface dynamically affects the roll gap accuracy. This includes factors such as roll elastic deformation caused by rolling force fluctuations, instantaneous thermal crown changes due to frictional heat between the roll and strip, and roll spatial position shifts caused by changes in roll bearing clearance. In this embodiment, rolling force sensors are arranged on the operating and driving sides of the mill to collect real-time rolling force. The geometric parameters of the workpiece include the inlet thickness, outlet thickness, and workpiece width. The roll parameters include roll geometric parameters (work roll radius, support roll radius, work roll crown, and support roll crown) and roll mechanical parameters (bending force and intermediate roll shifting). The preset strip deformation resistance is based on the steel grade. Based on the determined characteristics and measured temperature, the prediction model for the pre-set roll elastic flattening amount uses the influence function method to numerically solve for the elastic deformation of the roll system. This model is trained by acquiring a large amount of historical data on real-time rolling force, workpiece geometry parameters, roll parameters, pre-set strip deformation resistance, and corresponding roll elastic flattening amount variations. Rolling process parameters include rolling speed, real-time rolling force, forward slip ratio, emulsion temperature, and emulsion flow rate. Roll material parameters include roll thermal conductivity, specific heat capacity, density, and coefficient of thermal expansion. Pre-set strip material parameters include specific heat capacity and density. The pre-set friction coefficient refers to the friction coefficient under current lubrication conditions. The initial roll temperature refers to the ambient temperature or the temperature from the previous rolling operation. The temperature field at the end of the test is predicted using a two-dimensional finite difference method as the core of the instantaneous thermal convexity change prediction model on the roll surface. This model is trained by acquiring a large amount of historical data on rolling process parameters, roll material parameters, preset strip material parameters, preset friction coefficient, initial roll temperature, preset strip deformation resistance, and corresponding instantaneous thermal convexity change data on the roll surface. Support bearing structural parameters include bearing type, number of rolling elements, rolling element diameter, bearing pitch circle diameter, and bearing width. Real-time roll offset includes the roll offset in both horizontal and vertical directions. The clearance between the bearing housing and the frame is monitored in real-time by a clearance sensor, and the bearing temperature is determined by the bearing... Temperature sensors embedded in the bearing housing collect data, and the cumulative bearing wear value is estimated online using the recursive least squares method. A pre-set bearing clearance influence roll gap prediction model is established based on the bearing dynamics equation and Hertzian contact theory, creating a dynamic mapping relationship between bearing clearance, wear, and roll spatial position offset. This model is trained by acquiring a large number of historical samples of supporting bearing structural parameters, real-time roll offset, bearing housing and frame clearance values, bearing temperature, cumulative bearing wear value, real-time rolling force, and corresponding equivalent roll gap compensation data. Finally, the actual roll gap adjustment is optimized by adding the equivalent roll gap compensation data and subtracting the roll elastic flattening change data and instantaneous thermal convexity change data on the roll surface for dynamic optimization.

[0061] The present invention discloses a method and system for feedforward compensation of billet quality in the hot direct delivery process of cold heading steel continuous casting billets. Through multi-dimensional online quality monitoring and fusion, and combining feedforward parameters based on historical experience for initial similarity matching, simulation analysis and self-correction are performed through a preset feedforward compensation optimization model and a preset hot direct delivery digital twin model, and graded feedforward compensation is executed. At the same time, active feedforward prevention and control of hot delivery cracks are carried out, thereby achieving accurate feedforward compensation of billet quality in the hot direct delivery process of cold heading steel continuous casting billets.

[0062] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0063] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0064] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0065] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0066] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

Claims

1. A method for feedforward compensation of billet quality during the hot direct delivery process of cold heading steel continuous casting billets, characterized in that, Includes the following steps: Acquire real-time quality detection datasets of cold heading steel billets and perform analysis and processing to obtain multi-dimensional quality status assessment data of the billets; Based on the multidimensional quality status evaluation data of the billet, an initial set of feedforward compensation parameters is obtained by matching and analyzing the preset historical feedforward compensation parameter database. Based on the initial feedforward compensation parameter set and the multidimensional quality status evaluation data of the billet, the optimized feedforward compensation parameter set is obtained through analysis and processing using a preset feedforward compensation optimization model and a preset hot direct delivery digital twin model. Based on the optimized feedforward compensation parameter set and the multi-dimensional quality status evaluation data of the billet, hot direct feedforward compensation for cold heading steel continuous casting billets is performed.

2. The billet quality feedforward compensation method for the hot direct delivery process of cold heading steel continuous casting billets according to claim 1, characterized in that, The process of acquiring and analyzing real-time quality monitoring datasets of cold heading steel billets to obtain multi-dimensional quality status assessment data includes: Obtain a real-time quality detection dataset for cold heading steel billets, including the billet's three-dimensional temperature field, cross-sectional geometric parameters, surface defect characteristics, and internal quality index prediction data. The three-dimensional temperature field of the billet, the geometric parameters of the billet cross section, the surface defect characteristic data of the billet, and the predicted data of the internal quality index of the billet are spatiotemporally registered and fused to obtain multi-dimensional quality status assessment data of the billet.

3. The billet quality feedforward compensation method for the hot direct delivery process of cold heading steel continuous casting billets according to claim 2, characterized in that, The initial feedforward compensation parameter set is obtained by matching and analyzing the multidimensional quality status evaluation data of the cast billet through a preset historical feedforward compensation parameter database, including: The multidimensional quality status evaluation data of the billet is subjected to feature extraction to obtain the billet quality quantification feature vector. Based on the quantification feature vector of the billet quality, a similarity matching analysis is performed through a preset historical feedforward compensation parameter database to obtain a preset number of candidate feedforward compensation parameter sets and corresponding cosine similarities. The cosine similarity is squared and normalized to obtain the weight values ​​corresponding to the candidate feedforward compensation parameter set. The candidate feedforward compensation parameter set is weighted and fused according to the corresponding weight values ​​to obtain the initial feedforward compensation parameter set.

4. The billet quality feedforward compensation method for the hot direct delivery process of cold heading steel continuous casting billets according to claim 3, characterized in that, The process involves analyzing and processing the initial feedforward compensation parameter set in conjunction with the multidimensional quality status evaluation data of the cast billet using a preset feedforward compensation optimization model and a preset hot direct delivery digital twin model to obtain an optimized feedforward compensation parameter set, including: The three-dimensional temperature field of the billet and the corresponding preset temperature compensation coefficient, the geometric parameters of the billet cross section and the corresponding preset size compensation coefficient, the surface defect characteristic data of the billet and the corresponding preset defect compensation strength coefficient, the predicted data of the internal quality index of the billet and the corresponding internal quality compensation coefficient, as well as the initial feedforward compensation parameter set and the preset cold heading steel material characteristic parameters are input into the preset feedforward compensation optimization model for analysis and processing to obtain the feedforward compensation decision parameter set. Based on the multidimensional quality status assessment data of the billet and the set of feedforward compensation decision parameters, the continuous casting side source compensation, the pre-compensation of the hot direct delivery process and the rolling compensation simulation analysis are performed through the preset hot direct delivery digital twin model to obtain the post-rolling quality prediction data. The post-rolling quality prediction data is compared with the preset quality target data to obtain the compensation deviation parameter; If the compensation deviation parameter is greater than the preset compensation deviation warning threshold, iterative adjustment is performed using the preset gradient descent algorithm, and cyclic monitoring is executed. If the compensation deviation parameter is less than or equal to the preset compensation deviation warning threshold, then the feedforward compensation decision parameter set is determined to be the optimized feedforward compensation parameter set.

5. The billet quality feedforward compensation method for the hot direct delivery process of cold heading steel continuous casting billets according to claim 4, characterized in that, The step of performing hot direct feedforward compensation for cold heading steel continuous casting billets based on the optimized feedforward compensation parameter set and the multi-dimensional quality status evaluation data of the casting billets includes: The multidimensional quality status evaluation data of the billet is analyzed and processed to obtain feedforward compensation trigger commands, including continuous casting side source compensation commands, hot direct delivery process pre-compensation commands and rolling compensation commands. According to the continuous casting side source compensation instruction, the hot direct delivery process pre-compensation instruction, and the rolling compensation instruction, the hot direct delivery feedforward compensation of the cold heading steel continuous casting billet is executed according to the optimized feedforward compensation parameter set.

6. The billet quality feedforward compensation method for the hot direct delivery process of cold heading steel continuous casting billets according to claim 5, characterized in that, Also includes: The roll reduction values ​​at the mill inlet and outlet are obtained and analyzed using a pre-defined recursive least squares method to obtain roll wear compensation parameters. The actual temperature of the roll is obtained and analyzed in combination with the preset thermal expansion coefficient to obtain the predicted reduction of roll gap. The optimized theoretical adjustment amount of the roll gap is extracted based on the optimized feedforward compensation parameter set, and then corrected based on the roll wear compensation parameters and the predicted reduction of the roll gap to obtain the optimized actual adjustment amount of the roll gap.

7. The method for feedforward compensation of billet quality in the hot direct delivery process of cold heading steel continuous casting billet according to claim 2, characterized in that, Also includes: Data is extracted based on the three-dimensional temperature field of the billet to obtain the surface temperature of the preset billet node at a preset time point, and a real-time temperature-time curve is constructed. The cooling rate is obtained by analyzing the real-time temperature and time curves. The three-dimensional temperature field, cooling rate, and preset cold heading steel material characteristic parameters of the billet are input into the preset billet surface crack prediction model for analysis and processing to obtain the predicted value of the billet surface crack. The predicted value of surface cracks on the billet is compared with the preset threshold value for the occurrence of surface cracks on the billet, and a graded feedforward control strategy is obtained based on the threshold range that falls within it.

8. A billet quality feedforward compensation system for the hot direct delivery process of cold heading steel continuous casting billets, characterized in that, The system includes a memory and a processor. The memory contains a program for a billet quality feedforward compensation method in the hot direct delivery process of cold-heading steel continuous casting billets. When the program for the hot direct delivery process of cold-heading steel continuous casting billets is executed by the processor, it performs the following steps: Acquire real-time quality detection datasets of cold heading steel billets and perform analysis and processing to obtain multi-dimensional quality status assessment data of the billets; Based on the multidimensional quality status evaluation data of the billet, an initial set of feedforward compensation parameters is obtained by matching and analyzing the preset historical feedforward compensation parameter database. Based on the initial feedforward compensation parameter set and the multidimensional quality status evaluation data of the billet, the optimized feedforward compensation parameter set is obtained through analysis and processing using a preset feedforward compensation optimization model and a preset hot direct delivery digital twin model. Based on the optimized feedforward compensation parameter set and the multi-dimensional quality status evaluation data of the billet, hot direct feedforward compensation for cold heading steel continuous casting billets is performed.

9. The billet quality feedforward compensation system for the hot direct delivery process of cold heading steel continuous casting billets according to claim 8, characterized in that, The process of acquiring and analyzing real-time quality monitoring datasets of cold heading steel billets to obtain multi-dimensional quality status assessment data includes: Obtain a real-time quality detection dataset for cold heading steel billets, including the billet's three-dimensional temperature field, cross-sectional geometric parameters, surface defect characteristics, and internal quality index prediction data. The three-dimensional temperature field of the billet, the geometric parameters of the billet cross section, the surface defect characteristic data of the billet, and the predicted data of the internal quality index of the billet are spatiotemporally registered and fused to obtain multi-dimensional quality status assessment data of the billet.

10. The billet quality feedforward compensation system for the hot direct delivery process of cold heading steel continuous casting billets according to claim 9, characterized in that, The initial feedforward compensation parameter set is obtained by matching and analyzing the multidimensional quality status evaluation data of the cast billet through a preset historical feedforward compensation parameter database, including: The multidimensional quality status evaluation data of the billet is subjected to feature extraction to obtain the billet quality quantification feature vector. Based on the quantification feature vector of the billet quality, a similarity matching analysis is performed through a preset historical feedforward compensation parameter database to obtain a preset number of candidate feedforward compensation parameter sets and corresponding cosine similarities. The cosine similarity is squared and normalized to obtain the weight values ​​corresponding to the candidate feedforward compensation parameter set. The candidate feedforward compensation parameter set is weighted and fused according to the corresponding weight values ​​to obtain the initial feedforward compensation parameter set.

Images