A high-speed wire head and tail ear accurate control method and system

Abstract

The application discloses a high-speed wire head and tail ear accurate control method and system, belongs to the metal material rolling technical field, and the method comprises the following steps: collecting the rolling piece head and tail position, real-time temperature and dynamic carbon equivalent, and constructing a process fingerprint vector; taking the vector as input, using the rolling piece size prediction model, adopting the model predictive control algorithm to roll to solve the optimal tension adjustment amount and the roll gap compensation amount; the optimization result is executed, and the dynamic air pressure of the air cooling line fan is calculated based on the dynamic carbon equivalent and the temperature; the model parameters are adjusted online according to the measured size deviation, and the successful process data is stored in the process block chain. The application realizes the high-precision collaborative control of the head and tail ear, improves the yield, energy efficiency and equipment life, and has the self-adaptive learning and data traceability ability.

Description

Technical Field

[0001] This invention belongs to the field of metal material rolling technology, and particularly relates to a method and system for precise control of the head and tail loops of high-speed wire rods. Background Technology

[0002] High-speed wire rod, as a basic raw material, is widely used in industries such as construction, machinery manufacturing, and metal products. Dimensional accuracy is one of the core indicators for measuring product quality. In modern high-speed rolling production, the rolled piece deforms continuously at speeds of tens or even hundreds of meters per second, placing extremely high demands on the stability and precision of process control. Among these, the localized dimensional deviations at the head and tail of the rolled piece due to variations in process conditions, known as "head and tail ears," have long been a key technological challenge restricting product yield and quality consistency.

[0003] The formation mechanism of head and tail ears is complex, resulting from the coupled effects of multiple dynamic factors. First, during continuous rolling, the tension between stands undergoes a drastic change from zero to positive or vice versa as the head of the workpiece sequentially bites into each stand and the tail is ejected. This intensifies the lateral flow of the metal, which is no longer constrained by tension. Second, the head and tail sections of the workpiece have a longer contact time with air, guides, and rolls before entering the finishing mill, resulting in a significantly greater temperature drop than the middle section, which is in a stable rolling state. This leads to increased resistance to deformation and a stronger tendency for widening of the metal at the head and tail. Furthermore, the instantaneous fluctuations in the roll gap caused by the bite impact directly disrupt the stability of the deformation zone. These factors combine to cause the metal to overfill the die, forming ears at the head and tail.

[0004] To control the beginning and end of the rolling mill, the industry has developed various technologies, but these generally have limitations. Conventional methods mainly rely on the operator's experience, setting fixed tension compensation values ​​or roll gap adjustments in the rolling schedule, triggering execution when the detection system identifies the beginning and end positions. This method is essentially feedforward compensation based on static rules, and cannot respond in real time to the specific conditions of each steel piece, such as subtle fluctuations in material composition, real-time changes in the temperature field, and differences in equipment conditions. Therefore, the control effect is unstable and the adaptive capability is poor.

[0005] Furthermore, some advanced control systems have introduced online dimensional detection and closed-loop feedback mechanisms, such as using PID controllers to adjust deviations. However, these methods typically only aim to "eliminate the current dimensional deviation," resulting in lag in control actions. They also fail to incorporate production energy consumption, wear and tear on equipment such as rolls into the collaborative optimization framework, making it difficult to achieve overall optimal production efficiency. Although advanced algorithms such as Model Predictive Control (MPC) have been applied in process industries, existing solutions often lack a dedicated predictive model that accurately reflects the dynamic relationship between material properties (such as carbon equivalent), process conditions, and final dimensions when dealing with the specific problem of high-speed wire rod end fittings. This leads to insufficient prediction accuracy and limited optimization effectiveness.

[0006] A deeper challenge lies in knowledge management and system evolution. Existing control models largely rely on offline calibration and debugging; once set, they are fixed and cannot be used online for self-learning and self-optimization based on the continuous accumulation of production data. Successful process adjustment experience exists mostly as tacit knowledge, lacking a systematic, structured accumulation and reliable sharing mechanism. This leads to fluctuations in technical levels between different shifts and production lines, making it difficult to effectively inherit and iteratively upgrade process knowledge.

[0007] Therefore, existing technologies often exhibit problems such as insufficient control precision, low overall energy efficiency, weak adaptability, and difficulties in knowledge transfer when facing the demands of high-speed, multi-variety, and variable-condition wire production. There is an urgent need for a control method that can connect the entire chain of "material sensing - dynamic prediction - multi-objective collaboration - online evolution" to fundamentally achieve precise, intelligent, and sustainable control of head and tail defects. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a method and system for precise control of the head and tail lugs of high-speed wires, thus solving the aforementioned problems.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a method for precise control of the head and tail lugs of high-speed wires, comprising:

[0010] S1. Collect and calculate the head and tail positions, real-time temperature and dynamic carbon equivalent of the rolled piece, and construct a process fingerprint vector that includes material properties and process status.

[0011] S2. Using the process fingerprint vector as input, and employing a pre-established rolling mill size prediction model, with the primary objective of meeting the target size and the secondary objective of reducing energy consumption and equipment wear, a model predictive control algorithm is used to solve for the optimal tension adjustment sequence and roll gap compensation sequence within a finite time domain in the future.

[0012] S3. Send the optimal tension adjustment sequence and roll gap compensation sequence to the rolling mill control system for execution; at the same time, calculate and send the dynamic air pressure setpoint of the first fan of the Stellmore air-cooling line based on the dynamic carbon equivalent and the temperature when the head of the rolled piece leaves the finishing mill.

[0013] S4. Based on the deviation between the measured dimensions of the rolled piece and the output dimensions of the rolled piece size prediction model, adjust the internal parameters of the rolled piece size prediction model online; and after the rolled piece production is completed, encrypt and store the complete data packet of successful process fingerprint-control parameters-production results to the process blockchain.

[0014] Based on the above technical solutions, the present invention also provides the following optional technical solutions:

[0015] Further technical solution: In step S1, the formula for calculating the dynamic carbon equivalent is:

[0016]

[0017] in, Indicates dynamic carbon equivalent. Indicates the first The mass percentage of each element Indicates the first The elemental influence coefficient of a given element.

[0018] Further technical solution: In step S2, the rolled piece size prediction model is a machine learning model trained based on historical production data; the multi-objective rolling optimization control is achieved by solving the following optimization function:

[0019]

[0020] in, This indicates optimizing the objective function value. Indicates the prediction time step. This indicates that the future number of the rolled piece size prediction model outputs the following... Predicted dimensions of rolled parts Indicates the target size. This represents the tension adjustment sequence to be optimized. Indicates the tension reference value. This represents the energy consumption weighting coefficient. This represents the sequence of roll gap compensation amounts to be optimized. Indicates the reference value for roll gap. This represents the wear weighting coefficient.

[0021] Further technical solution: In step S3, the method for obtaining the dynamic air pressure setpoint of the first-stage fan of the Steyrmo air-cooled line is as follows:

[0022] The dynamic carbon equivalent and the measured temperature when the head leaves the finishing mill last stand are subjected to maximum-minimum normalization to obtain the dynamic carbon equivalent index and temperature index.

[0023] Based on the carbon equivalent index and temperature index, the dynamic air pressure setpoint of the first-stage fan of the Stellmore air-cooled line is obtained by the following formula:

[0024]

[0025] in, This indicates the dynamic air pressure of the first-stage fan in the Stellmore air-cooled line. This indicates the base wind pressure of this steel grade. Indicates the maximum allowable wind pressure adjustment. This represents the dynamic carbon equivalent index. Indicates the temperature index. and These represent the weighting coefficients for carbon equivalent and temperature, respectively, and their sum is 1.

[0026] Further technical solution: In step S4, the trigger condition for storing data to the process blockchain is:

[0027]

[0028] in, Indicates the storage trigger condition. This indicates optimizing the objective function value. Indicates the overall performance threshold. Indicates the actual size of the head. Indicates the target size. Indicates the allowable tolerance threshold for dimensions. Blockchain storage is triggered only if it is true.

[0029] A further technical solution: the tension reference value is taken as the maximum allowable tension adjustment amount of the corresponding frame, and the roll gap reference value is taken as the maximum allowable roll gap compensation amount of the corresponding frame.

[0030] A further technical solution: The rolling mill size prediction model in step S2 is a neural network model, whose input features include at least: dynamic carbon equivalent, rolling mill temperature, tension setting reference value, and roll gap reference value, and its output is the predicted rolling mill size value. The neural network model is obtained by using process parameter data from historical production as input features, the measured dimensions of the rolled piece at the corresponding time as the training target, and supervised training using the backpropagation algorithm.

[0031] Further technical solution: In step S4, the internal parameters of the rolling stock size prediction model are adjusted online by using the recursive least squares method or Kalman filter algorithm, with the real-time measured rolling stock size as the observation value, to identify and update the output layer weight parameters of the model online.

[0032] Further technical solution: The data packet stored in the process blockchain contains the following information: process fingerprint vector, optimal control parameter sequence, measured head and tail dimensions, optimized objective function value, and hash value of preceding data block.

[0033] A high-speed wire head and tail lug precision control system adopts the above-mentioned high-speed wire head and tail lug precision control method.

[0034] This invention provides a method and system for precise control of the head and tail lugs of high-speed wires, which has the following advantages compared with the prior art:

[0035] 1. This invention achieves accurate characterization of the dynamic behavior of rolled parts by constructing a process fingerprint vector that integrates material properties and process state, thereby improving the adaptability and foresight of control.

[0036] 2. This invention employs model predictive control and multi-objective rolling optimization, which effectively reduces energy consumption and equipment wear while ensuring the accuracy of the head and tail dimensions, thus achieving collaborative optimization;

[0037] 3. By introducing an online parameter adjustment and process blockchain storage mechanism, this invention enables the system to have self-learning capabilities and data traceability, supporting the continuous accumulation and secure sharing of process knowledge;

[0038] 4. This invention enhances the control capability of the phase transformation process by calculating the air pressure of the air-cooled fan through dynamic carbon equivalent coupled with temperature, and is especially suitable for the production of sensitive steel grades such as bainitic steel. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0041] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0042] Please see Figure 1 A method for precise control of the head and tail lugs of high-speed wires, provided in one embodiment of the present invention, includes:

[0043] S1. Collect and calculate the head and tail positions, real-time temperature and dynamic carbon equivalent of the rolled piece, and construct a process fingerprint vector that includes material properties and process status.

[0044] S2. Using the process fingerprint vector as input, and employing a pre-established workpiece size prediction model, with the primary objective of meeting the target size and the secondary objective of reducing energy consumption and equipment wear, a model predictive control (MPC) algorithm is used to solve the optimal tension adjustment sequence (representing the dynamic adjustment amount required relative to a certain "reference tension" (such as the set value when the middle of the rolling is stable)) and roll gap compensation sequence (representing the dynamic micro-compensation amount required relative to the "reference roll gap" or "theoretical set roll gap") within a future finite time domain (the estimated time required for the head / tail of the workpiece to move from the current position through N consecutive downstream finishing mill stands as the prediction time domain).

[0045] S3. Send the optimal tension adjustment sequence and roll gap compensation sequence to the rolling mill control system for execution; at the same time, calculate and send the dynamic air pressure setpoint of the first fan of the Stellmore air-cooling line based on the dynamic carbon equivalent and the temperature when the head of the rolled piece leaves the finishing mill.

[0046] S4. Based on the deviation between the measured dimensions of the rolled piece and the output dimensions of the rolled piece size prediction model, adjust the internal parameters of the rolled piece size prediction model online; and after the rolled piece production is completed, encrypt and store the complete data packet of successful process fingerprint-control parameters-production results to the process blockchain.

[0047] In summary, this invention fundamentally improves the control accuracy, overall energy efficiency, and adaptive capability of high-speed wire rod end-and-end defects by organically combining multi-source information perception, multi-objective collaborative optimization, online adaptive adjustment, and knowledge security accumulation. It also promotes the effective inheritance and iterative upgrading of process knowledge, bringing significant technological progress to high-speed wire rod production.

[0048] Preferably, in step S1, the formula for calculating the dynamic carbon equivalent is:

[0049]

[0050] in, Indicates dynamic carbon equivalent. Indicates the basic carbon equivalent. Indicates the first The mass percentage of each element Indicates the first The elemental influence coefficients of each element (constants that can be assigned by expert experience or obtained through material strengthening theory and regression analysis of historical data, specifically as described above) The following method was used to determine the contribution of each element to the high-temperature deformation resistance of the rolled piece: based on the historical data of the chemical composition of the target steel and the corresponding historical data of the rolling force, the data was fitted by a multiple linear regression method.

[0051] Dynamic carbon equivalent is a comprehensive index used to characterize the influence of steel's chemical composition on its high-temperature deformation behavior. It converts the various alloying elements in steel (excluding iron) into equivalent carbon elements to influence material properties through specific conversion relationships. During rolling, the material's deformation resistance is closely related to temperature, strain rate, and chemical composition. Dynamic carbon equivalent effectively simplifies complex chemical composition information into a single, quantifiable parameter, facilitating processing and application in real-time control systems. Its role is to provide a unified index reflecting the material's plastic deformation characteristics at high temperatures, supporting subsequent process parameter optimization and control. This calculation formula provides a structured method to quantify dynamic carbon equivalent by superimposing the base carbon content with the contributions of various alloying elements to form a comprehensive index. This linear superposition model intuitively reflects the independent influence of each element on material properties and uses coefficients for weighting. (Constant term) In the dynamic carbon equivalent calculation formula, the basic carbon equivalent of the matrix material is represented, or in other words, the basic deformation resistance characteristics exhibited by the material when the content of all alloying elements is zero. It provides a benchmark value for the entire calculation. This constant term can be determined experimentally based on the basic composition of a specific steel grade, or through regression analysis of a large amount of historical data. Element Influence Coefficient Quantified the first The relative contribution of each alloying element to the high-temperature deformation resistance of rolled steel. Different alloying elements, even with the same content, can have vastly different effects on the high-temperature deformation resistance of steel. The accuracy of these coefficients directly determines whether the dynamic carbon equivalent can truly reflect the material properties. Besides obtaining values ​​through expert experience or regression analysis of material strengthening theories and historical data, The influence of each element on lattice distortion and dislocation motion can also be predicted by combining physical metallurgical models with finite element simulations, or by first-principles calculations, thereby indirectly deriving its contribution to deformation resistance. Indicates the first in steel The actual content of each alloying element is usually expressed as a mass percentage. This is the direct input parameter for calculating dynamic carbon equivalent, reflecting the actual chemical composition of the steel. These mass percentage data are typically obtained in real-time or near real-time using chemical analysis methods such as furnace-front spectral analysis, direct-reading spectrometers, X-ray fluorescence spectrometry (XRF), or combustion-infrared absorption methods. By quantitatively characterizing the contribution of each element, dynamic carbon equivalent can become a physically meaningful indicator that is highly correlated with actual rolling behavior, thus providing a reliable basis for subsequent rolling force prediction, workpiece size prediction, and the formulation of control strategies.

[0052] The following is a specific example to illustrate this. In the high-speed wire rod rolling process, to accurately calculate the dynamic carbon equivalent, the chemical composition of the rolled piece can be obtained in real time using an online spectrometer. This includes the mass percentages of elements such as carbon (C), silicon (Si), manganese (Mn), phosphorus (P), sulfur (S), chromium (Cr), nickel (Ni), copper (Cu), molybdenum (Mo), vanadium (V), niobium (Nb), and titanium (Ti). For the influence coefficient of elements The determination of the target steel grade can be achieved through the following steps: First, collect historical production data for the target steel grade (e.g., a specific grade of low-carbon alloy steel) over a past period. This data should include detailed chemical composition (i.e., the mass percentage of each element) for each rolling batch and measured rolling force data for the corresponding stand during finishing rolling. Then, input this data into statistical analysis software, using rolling force as the dependent variable and the mass percentage of each element as the independent variable, and perform multiple linear regression analysis to obtain the contribution weight of each element to the rolling force. For example, for a specific steel grade, historical data regression analysis may yield the influence coefficient of the carbon element. The influence coefficient of silicon is 1.0. The influence coefficient of manganese is 0.3. The dynamic carbon equivalent is 0.25, while the influence coefficients of other trace elements may be even smaller. When the real-time detected carbon content of the rolled piece is 0.15%, silicon content is 0.20%, and manganese content is 0.60%, the dynamic carbon equivalent is... Based on these coefficients and real-time content, combined with the constant term... Perform calculations. For example, if If it is 0.05, then =0.41. This calculated dynamic carbon equivalent value will be used as part of the process fingerprint vector and input into the roll size prediction model to guide subsequent tension adjustment and roll gap compensation.

[0053] Through the above technical solution, this application can significantly improve the accuracy and robustness of precise control of the head and tail lugs of high-speed wire rods. Since the dynamic carbon equivalent is no longer a rough estimate based on experience or fixed parameters, but rather a data-driven multiple linear regression method combined with historical chemical composition and rolling force data, it accurately fits the contribution of each element to the high-temperature deformation resistance of the rolled piece. This allows the calculated dynamic carbon equivalent to more realistically and precisely reflect the material properties of the rolled piece during the dynamic rolling process. This high-precision dynamic carbon equivalent, as a key component of the process fingerprint vector, provides a more reliable input for the rolled piece size prediction model, thereby significantly improving the model's prediction accuracy for rolled piece size changes. Based on more accurate predictions, the model predictive control algorithm can generate more precise tension adjustment and roll gap compensation sequences, effectively suppressing dimensional deviations in the head and tail lugs of the rolled piece and ensuring product quality. Furthermore, the accurate dynamic carbon equivalent also provides a more solid foundation for the dynamic air pressure setting of the first-stage fan in the Steyrmo air-cooled line, further optimizing the cooling process and contributing to more uniform microstructure properties. Overall, this solution improves the intelligence level and control effect of the entire control system from the source of material property characterization, effectively solving the problem of insufficient control accuracy caused by inaccurate material property characterization.

[0054] Preferably, in step S2, the rolled piece size prediction model is a machine learning model trained based on historical production data; the multi-objective rolling optimization control is achieved by solving the following optimization function:

[0055]

[0056] in, This indicates optimizing the objective function value. Indicates the prediction time step. This indicates that the future number of the rolled piece size prediction model outputs the following... Predicted dimensions of rolled parts Indicates the target size. This represents the sequence of tension adjustments to be optimized (representing the amount of dynamic adjustment required relative to a certain "reference tension" (such as the set value when the rolling middle is stable). Indicates the tension reference value. This represents the energy consumption weighting coefficient. This represents the sequence of roll gap compensation amounts to be optimized (representing the dynamic micro-compensation amount required relative to the "reference roll gap" or "theoretical set roll gap"). Indicates the reference value for roll gap. This represents the wear and tear weighting coefficient;

[0057] Wherein, the tension reference value is taken as the maximum allowable tension adjustment amount of the corresponding frame, and the roll gap reference value is taken as the maximum allowable roll gap compensation amount of the corresponding frame;

[0058] In step S2, the rolling mill size prediction model is a neural network model. Its input features include at least: dynamic carbon equivalent, rolling mill temperature, tension setting reference value, and roll gap reference value. Its output is the predicted rolling mill size value. The neural network model is obtained by using process parameter data from historical production as input features, the measured dimensions of the rolled piece at the corresponding time as the training target, and supervising training using the backpropagation algorithm.

[0059] The neural network model is a feedforward neural network with three hidden layers. The number of neurons in each layer is [input layer dimension, 64, 32, 16, 1], respectively. The hidden layers use the ReLU activation function, and the output layer uses the linear activation function. The input features are constructed as follows: the four features of dynamic carbon equivalent, rolling temperature, tension setting reference value, and roll gap reference value are Z-score standardized and then concatenated into a one-dimensional input vector. The specific steps of the supervised training include: using the data of the stable section in the middle of the rolling mill in historical data as the training set, using the mean square error between the predicted size and the actual size as the loss function, using the Adam optimizer for training, and using the early stopping method to prevent overfitting.

[0060] The rolling mill size prediction model is specifically designed as a machine learning model trained on historical production data. This model learns from a large amount of historical data and discovers complex nonlinear relationships, thereby improving prediction accuracy. For example, besides a neural network model, this machine learning model can also be a support vector machine (SVM) model, which maps the input data to a high-dimensional space using a kernel function and searches for the optimal hyperplane in that space for regression prediction; or it can be a random forest model, which improves the model's generalization ability and robustness by integrating the prediction results of multiple decision trees. The multi-objective rolling optimization control is achieved by solving a specific optimization function. This function comprehensively considers the deviation between the rolling mill size and the target size, energy consumption, and equipment wear. The size deviation term aims to ensure the accuracy of the rolling mill size, the energy consumption term aims to reduce energy consumption during the rolling process, and the wear term aims to reduce equipment (such as rolls) wear. By introducing weighting coefficients, the importance of different objectives can be adjusted according to actual production needs, achieving a flexible multi-objective balance. In the optimization function, the tension reference value... The roll gap reference value is taken as the maximum permissible tension adjustment amount for the corresponding frame. The maximum permissible roll gap compensation amount for the corresponding stand is taken as the reference value. These reference values ​​serve as normalization factors, ensuring that adjustment amounts of different dimensions are comparable in the optimization function and guaranteeing that the optimization process is carried out within physically permissible limits. For example, the maximum permissible tension adjustment amount can be determined based on mill design parameters or historical operating experience to avoid damage to the workpiece or equipment; the maximum permissible roll gap compensation amount can be set based on the adjustment range of the rolls and the deformation capacity of the workpiece.

[0061] Furthermore, the workpiece size prediction model in step S2 is specifically set as a neural network model. Neural network models, due to their powerful nonlinear mapping and self-learning capabilities, are particularly suitable for handling complex, multi-factor coupled size prediction problems during rolling. The input features of this neural network model include at least dynamic carbon equivalent, workpiece temperature, tension setting reference value, and roll gap reference value; these parameters are key factors affecting the workpiece size. Its output is the predicted workpiece size. This neural network model is obtained through supervised training using historical production process parameter data as input features and the measured dimensions of the rolled piece at the corresponding time point as the training target, employing the error backpropagation algorithm. The error backpropagation algorithm is an efficient training method that adjusts the weights and biases within the network based on the prediction error, thereby gradually approximating the true mapping relationship. Specifically, the neural network model is configured as a feedforward neural network with three hidden layers, with the number of neurons in each layer being [input layer dimension, 64, 32, 16, 1]. This multi-layer structure allows the network to learn deeper and more abstract feature representations. The hidden layers use the ReLU (Rectified Linear Unit) activation function, which effectively alleviates the gradient vanishing problem and accelerates network training. The output layer uses a linear activation function to adapt to continuous dimension prediction tasks. The input features are constructed by performing Z-score normalization on four features: dynamic carbon equivalent, rolled piece temperature, tension setting reference value, and roll gap reference value, and then concatenating them into a one-dimensional input vector. Z-score standardization eliminates dimensional differences between different features, making the data distribution more uniform and thus improving the training efficiency and predictive performance of the model. The specific steps of supervised training include: using data from the stable section of the rolled piece in historical data as the training set; using the mean squared error (MSE) between the predicted and actual dimensions as the loss function; training with the Adam optimizer; and employing early stopping to prevent overfitting. Using stable section data as the training set ensures the quality and representativeness of the training data. The mean squared error as the loss function quantifies the deviation between the predicted and true values. The Adam optimizer is an adaptive learning rate optimization algorithm that efficiently adjusts model parameters. Early stopping effectively avoids overfitting on the training set by stopping training when the performance on the validation set no longer improves, thereby improving the model's generalization ability.

[0062] The proposed solution significantly improves the accuracy of rolled piece size prediction by concretizing the rolled piece size prediction model into a machine learning model, particularly a neural network model, trained based on historical production data. During the rolling process, the size variation of the rolled piece is influenced by a complex array of factors, including material properties (such as dynamic carbon equivalent) and process conditions (such as rolled piece temperature, tension, and roll gap). These factors exhibit highly nonlinear and coupled relationships. Traditional physical models often struggle to accurately describe these complex relationships. By employing a neural network model, using a large amount of historical production process parameter data as input features, and corresponding measured rolled piece dimensions as training targets, and utilizing a backpropagation algorithm for supervised training, this model can learn from massive amounts of data and capture these complex nonlinear mapping relationships, thereby providing more accurate rolled piece size predictions. Based on the high-precision prediction model, this application further achieves multi-objective rolling optimization control by solving a multi-objective optimization function. This optimization function not only takes the deviation between the rolled piece size and the target size as the main optimization objective, but also considers energy consumption and equipment wear. This is achieved by introducing an energy consumption weighting coefficient into the optimization function. and wear weighting coefficient And set the tension reference value. The roll gap reference value represents the maximum allowable tension adjustment for the corresponding frame. To accommodate the maximum permissible roll gap compensation for the corresponding stand, the optimizer can effectively balance energy consumption and equipment wear while maintaining dimensional accuracy. For example, when a deviation from the target workpiece size is predicted, the optimizer calculates the optimal tension adjustment sequence. and roll gap compensation sequence These sequences not only enable the rolled piece dimensions to return to the target as quickly as possible, but also prioritize adjustment methods that minimize energy consumption and equipment wear. This control strategy, based on accurate prediction and multi-objective collaborative optimization, allows the mill control system to make more intelligent and precise dynamic adjustments, effectively solving the problems of insufficient control accuracy of the rolled piece head and tail dimensions, as well as the difficulty in balancing energy consumption and equipment wear.

[0063] As a specific implementation method, the rolled piece size prediction model can be built using a deep learning framework based on Python, such as TensorFlow or PyTorch. This neural network model can be deployed on an industrial computer within the rolling mill control system. During the training phase, rolling data for a specific steel grade can be collected over the past year, including the dynamic carbon equivalent, real-time temperature, tension settings for each stand, roll gap settings, and the corresponding measured dimensions of the rolled piece at the finishing mill stand entrance. This data, after Z-score standardization, forms the training dataset. For example, the dynamic carbon equivalent can range from 0.3% to 0.8%, and the rolled piece temperature can range from 900℃ to 1050℃. During training, data from the stable rolling section in the middle of the rolled piece can be selected to avoid the influence of unstable data from the beginning and end on the model's generalization ability. The loss function can be the mean squared error (MSE), the optimizer can be Adam, and the learning rate can be set to 0.001. To prevent overfitting, a validation set can be set. Early stopping is triggered when the loss on the validation set no longer decreases for 10 consecutive epochs. After training, the trained neural network model parameters are solidified and loaded into the rolling mill control system. During actual rolling, when a new process fingerprint vector (including dynamic carbon equivalent, workpiece temperature, etc.) is input, the neural network model can output real-time predicted workpiece dimensions for the next N finishing mill stands. These predicted values ​​are then fed into the model predictive control algorithm, combined with preset energy consumption weighting coefficients. and wear weighting coefficient Using the maximum allowable tension adjustment amount and roll gap compensation amount as reference values, the optimal tension adjustment sequence is obtained through rolling calculation. and roll gap compensation sequence The instructions were then issued to the rolling mill for execution.

[0064] Through the above technical solution, this application can significantly improve the control accuracy of the head and tail lug dimensions of high-speed wire rods. Employing a machine learning model trained on historical production data, particularly a neural network model, can more accurately predict the dimensional changes of the rolled piece during complex rolling processes, overcoming the limitations of traditional models in handling nonlinear and multi-factor coupled problems. Furthermore, by introducing a multi-objective optimization function, energy consumption and equipment wear are effectively balanced while ensuring the rolled piece dimensions accurately meet the objectives, avoiding the negative impacts that may arise from single-objective optimization. For example, while pursuing dimensional accuracy, it can avoid energy consumption spikes or excessive equipment wear caused by over-adjustment. This comprehensive control strategy results in smaller fluctuations in the head and tail lug dimensions of the rolled piece, higher product qualification rates, reduced production and equipment maintenance costs, and improved production efficiency and economic benefits.

[0065] Preferably, in step S3, the dynamic air pressure setpoint of the first-stage fan of the Stellmore air-cooled line is obtained as follows:

[0066] The dynamic carbon equivalent and the measured temperature when the head leaves the finishing mill last stand are subjected to maximum-minimum normalization to obtain the dynamic carbon equivalent index and temperature index.

[0067] Based on the carbon equivalent index and temperature index, the dynamic air pressure setpoint of the first-stage fan of the Stellmore air-cooled line is obtained by the following formula:

[0068]

[0069] in, This indicates the dynamic air pressure of the first-stage fan in the Stellmore air-cooled line. This indicates the base wind pressure of this steel grade. Indicates the maximum allowable wind pressure adjustment. This represents the dynamic carbon equivalent index. Indicates the temperature index. and These represent the weighting coefficients for carbon equivalent and temperature, respectively, and their sum is 1;

[0070] Among them, the and Values ​​are assigned through expert experience or obtained through offline thermo-mechanical coupling simulation calibration based on the phase transformation characteristics of steel grades. For steel grades sensitive to bainitic phase transformation, settings are provided. .

[0071] To achieve precise control over the cooling process at the head of the rolled piece, it is first necessary to standardize the key parameters affecting the cooling effect. This involves performing maximum-minimum normalization on the dynamic carbon equivalent and the measured temperature at the head of the rolled piece leaving the finishing mill stand. This aims to eliminate the influence of different physical dimensions and numerical ranges on subsequent calculations, thereby obtaining the dynamic carbon equivalent index and temperature index. This normalization maps the original data to a unified range (e.g., 0 to 1), making the weighting of different parameters in the formula more reasonable and effective. This processing can employ linear normalization, i.e., subtracting the minimum value from the original value and then dividing by the difference between the maximum and minimum values ​​to obtain the normalized result; alternatively, nonlinear normalization methods, such as using the Sigmoid or Tanh functions, can be used to accommodate certain nonlinear relationships or provide better robustness to outliers.

[0072] After obtaining the normalized dynamic carbon equivalent index and temperature index, the dynamic wind pressure setpoint of the first-stage fan of the Steilmo air-cooled line is calculated using a preset mathematical formula. This formula... Dynamic wind pressure was defined. It is the wind pressure of the steel foundation Add a carbon equivalent index and temperature index The composition of the weighted portfolio adjustment, where, The maximum permissible wind pressure adjustment range is defined to ensure that the adjustment is within the equipment's capacity. This formula can be directly integrated into the programmable logic controller (PLC) or distributed control system (DCS) of the rolling mill control system. It can perform rapid calculations using real-time collected index values ​​and send the calculation results as instructions to the fan controller. Alternatively, the formula can also run as part of an advanced control algorithm on a host computer or industrial control computer, sending the calculated wind pressure setpoint to the fan control unit via an industrial communication network.

[0073] Weighting coefficients in the formula and Accurately reflecting the relative importance of carbon equivalent and temperature to cooling effects is crucial. The determination of these weighting coefficients can be varied. One method is through expert experience assignment, where engineers with extensive knowledge of rolling and cooling processes directly set these coefficients based on historical production data and process rules. A more precise method involves calibration using offline thermo-mechanical coupling simulation based on the phase transformation characteristics of the steel grade. This typically involves using specialized numerical simulation software to establish thermodynamic and phase transformation kinetic models of the rolled piece during cooling. By simulating cooling curves and microstructure evolution under different parameter combinations, the weighting coefficients that best optimize product performance can be derived. In particular, for bainitic phase transformation-sensitive steels, whose phase transformation behavior is more sensitive to temperature changes, the weight of temperature is deliberately increased when setting the weights. Set a weight greater than the carbon equivalent. This is to ensure that cooling control can more effectively guide the desired phase change path.

[0074] The proposed solution normalizes the dynamic carbon equivalent and measured head temperature of the rolled piece, and then dynamically calculates the setpoint of the air pressure for the first-stage fan in the Stellmore air-cooling line, achieving refined and adaptive control of the head cooling process. This dynamic air pressure adjustment method allows the cooling intensity to respond in real time to the actual composition and thermal state of the rolled piece, avoiding the problems of insufficient or excessive cooling caused by traditional fixed air pressure or simple segmented air pressure.

[0075] Specifically, in high-speed wire rod production, as the head of the rolled piece leaves the finishing stand, the control system acquires its dynamic carbon equivalent (e.g., calculated via an online composition analyzer or based on a heat composition and decarburization model) and measured temperature (e.g., via an infrared thermometer) in real time. Assuming that for a specific steel grade, the historical range of its carbon equivalent is 0.3% to 0.6%, and the historical temperature range is 850... Up to 1000 If the current dynamic carbon equivalent is 0.45%, and the measured temperature is 950°C... Then, through max-min normalization, the dynamic carbon equivalent index and temperature index can be calculated. For example, the dynamic carbon equivalent index might be 0.5, and the temperature index might be 0.67. For this steel grade, if it belongs to the bainitic transformation-sensitive steel grade, then based on the offline thermo-mechanical coupling simulation results, the carbon equivalent weight... Temperature weighting These can be set to 0.4 and 0.6 respectively. Simultaneously, the base wind pressure for this steel grade can be set. The maximum allowable wind pressure adjustment is 10 kPa. The setpoint is 5 kPa. Substituting these parameters into the formula, the dynamic wind pressure setpoint can be calculated. 13.01 kPa. The calculated dynamic wind pressure setpoint was then sent to the control system of the first section of the Stellmore air-cooled line fan to precisely control the cooling intensity at the head of the rolled piece.

[0076] Through the above technical solution, this application further improves the internal quality and uniformity of the rolled piece head while ensuring the dimensional accuracy of the rolled piece. This dynamic air pressure adjustment method can effectively suppress unwanted phase transformations and optimize the microstructure of the rolled piece head, thereby significantly improving its mechanical properties and product quality. Compared with solutions that only focus on dimensional control, this application reduces defects caused by improper cooling, such as uneven hardness, excessive residual stress, or coarse microstructure, ultimately improving the overall yield and added value of high-speed wire rod products.

[0077] Preferably, in step S4, the trigger condition for storing data to the process blockchain is:

[0078]

[0079] in, Indicates the storage trigger condition. This indicates optimizing the objective function value. Indicates the overall performance threshold. Indicates the actual size of the head. Indicates the target size. Indicates the allowable tolerance threshold for dimensions. Blockchain storage is triggered only if the condition is true.

[0080] The data packets stored in the process blockchain include the following information: process fingerprint vector, optimal control parameter sequence, measured head and tail dimensions, and optimized objective function value. and the hash value of the preceding data block;

[0081] In step S4, the internal parameters of the rolling mill size prediction model are adjusted online using recursive least squares or Kalman filtering algorithm. The real-time measured rolling mill size is used as the observation value to identify and update the weight parameters of the output layer of the model online. Specifically, the recursive least squares method is used to adjust the parameters online: all weights of the neural network model except the output layer are fixed, the output layer is regarded as a linear layer, the output of the last hidden layer of the network is used as the input feature, the deviation between the measured size and the model predicted size is used as the observation error, and the recursive least squares method with forgetting factor is used to update the weight parameters of the output layer online.

[0082] This data storage trigger condition is used to determine whether the current rolling batch meets preset performance and quality standards, thereby deciding whether to store relevant data in the process blockchain. Its function is to filter high-quality production data, preventing abnormal or unqualified production data from being included in the blockchain, ensuring the reliability and validity of the data in the blockchain. This condition can be based on preset thresholds for logical judgment. For example, data storage is triggered when the optimization objective function value is lower than a certain comprehensive performance threshold, and the deviation between the actual size of the rolled piece head and the target size is less than a certain allowable deviation threshold. (Optimization objective function value) This is a comprehensive performance index obtained by the model predictive control algorithm when solving for the optimal control sequence. It reflects the achievement of multiple optimization objectives such as workpiece dimensional accuracy, energy consumption, and equipment wear. Its function is to quantify the overall performance of the current rolling process. (Comprehensive performance threshold) It is a preset upper limit value used to evaluate the overall performance of the rolling process, when optimizing the objective function value. When the performance falls below this threshold, it indicates that the overall performance of the current rolling process has reached or exceeded expectations. This threshold can be set based on statistical analysis of historical production data, expert experience, or process requirements. For example, it can be determined by analyzing a large number of successful rolling batches. The values ​​are statistically analyzed, and their average or a certain percentile is taken as the average. Actual head size This refers to the actual measured dimension of the workpiece head after passing through the finishing mill. Its function is to directly reflect the control accuracy of the workpiece head dimension. This dimension can be obtained in real time using equipment such as online diameter gauges and laser measurement systems. Target dimension. This refers to the ideal dimension that the rolled piece needs to achieve during the finishing rolling process. Its function is to serve as a reference for dimensional control, and this dimension is usually preset by product specifications or process requirements. Dimensional tolerance threshold. This is the maximum permissible deviation between the actual and target dimensions of the rolled piece head. When the actual deviation is less than this threshold, the rolled piece head size is considered acceptable. This threshold can be set according to product quality standards, customer requirements, or industry specifications; for example, it could be set to ±0.5% or ±0.1mm of the target size. The data package stored in the process blockchain contains key information for successful rolling batches, including the process fingerprint vector, optimal control parameter sequence, measured head and tail dimensions, and optimized objective function value. The data package, along with the hash values ​​of preceding data blocks, serves to provide a reliable and verifiable data foundation for subsequent process optimization, quality traceability, and model training. The data package can be constructed using structured data formats such as JSON or XML, ensuring the integrity and accuracy of all fields. The online adjustment of the internal parameters of the rolled piece size prediction model aims to dynamically correct the model's parameters based on real-time production data, thereby improving the model's prediction accuracy and adaptability. Its purpose is to enable the prediction model to adapt to changes in the production environment and maintain its effectiveness. This adjustment can employ various online learning algorithms; for example, in addition to recursive least squares or Kalman filtering, adaptive neural networks and online support vector machines can also be used. Recursive least squares and Kalman filtering are commonly used online parameter identification and state estimation methods. Recursive least squares is suitable for linear or linearizable systems, updating parameters by minimizing the sum of squared prediction errors; Kalman filtering is an optimal estimation method suitable for noisy dynamic systems, estimating system state and parameters by fusing prediction and observation information. Their role is to provide a mathematically rigorous and computationally efficient online parameter update mechanism. Fixing all weights in the neural network model except for the output layer, treating the output layer as a linear layer, simplifies the computational complexity of online tuning. It transforms the complex online learning problem of nonlinear neural networks into a relatively simple linear regression problem, reducing the computational burden and improving the real-time performance of parameter updates. This method can be applied to various neural network architectures, as long as the output layer can be effectively approximated as a linear layer. Using the output of the last hidden layer as the input feature is a crucial step in combining the nonlinear feature extraction capability of the neural network with a linear model. The output of the last hidden layer already contains a high-dimensional feature representation of the original input after multiple nonlinear transformations. Its purpose is to leverage the powerful feature extraction capability of the neural network while maintaining the linearity of the output layer parameter updates. Using the deviation between the measured size and the model's predicted size as the observation error, which is the difference between the model's prediction and the actual measurement, directly reflects the model's prediction accuracy. Its function is to serve as a feedback signal for online parameter tuning, driving the model parameters to update in a direction that reduces error. The weight parameters of the output layer are updated online using recursive least squares with a forgetting factor. Recursive least squares with a forgetting factor is an improved online learning algorithm that assigns smaller weights to older data during parameter updates, enabling the model to adapt more quickly to new data and environmental changes. Its purpose is to improve the model's adaptability to time-varying characteristics, prevent the model from being "locked in" by old data, and ensure that the model always reflects the latest process conditions. The forgetting factor can be adjusted according to the rate of process change and the degree of dependence on historical data; for example, its value is typically between 0.9 and 1.

[0083] The proposed solution ensures the quality of process data and the accuracy of the prediction model during the rolling mill production process by introducing strict data storage trigger conditions and a refined online model parameter adjustment mechanism. Specifically, after the rolling mill production is completed, the system does not unconditionally store all production data in the process blockchain; instead, it first evaluates the overall performance and product quality of the rolling process. This is achieved by optimizing the objective function value... Compared with the preset comprehensive performance threshold Compare and simultaneously determine the actual dimensions of the rolled piece head. With target size Deviation and dimensional tolerance threshold The system compares and only triggers data storage when both conditions are met simultaneously, indicating that the rolling process has achieved high standards in both control effectiveness and product quality. This dual-criteria screening mechanism effectively avoids recording low-quality or abnormal production data into the blockchain, thus ensuring the purity and reliability of data in the process blockchain and providing a solid foundation for subsequent process optimization, quality traceability, and model retraining. Simultaneously, to ensure the rolled piece size prediction model maintains high accuracy in constantly changing production environments, this solution employs an online adjustment strategy for the model's internal parameters. This adjustment does not involve comprehensive online training of the entire complex neural network; instead, it cleverly focuses the adjustment on the output layer weight parameters. By fixing all weights of the neural network model except for the output layer and treating the output layer as a linear layer, the system can utilize the output of the last hidden layer as input features, transforming a complex nonlinear problem into a relatively simple linear problem. Based on this, a recursive least squares method with a forgetting factor is used, with the deviation between the real-time measured rolled piece size and the model's predicted size as the observation error, to identify and update the output layer weight parameters online. This method not only significantly reduces the complexity of online computation and ensures real-time parameter updates, but also, by introducing a forgetting factor, enables the model to adapt more flexibly to new process conditions and material properties, avoiding excessive reliance on historical data and thus continuously improving the accuracy and robustness of the rolled product size prediction model. In this way, high-quality production data is selectively stored and used to continuously optimize the prediction model, forming a data-driven, self-improving closed-loop control system that significantly enhances the precision control level of the head and tail lug dimensions of high-speed wire rods.

[0084] The following is a concrete example to illustrate this. In the high-speed wire rod rolling process, after a batch of rolled products is completed, the control system first obtains the optimized objective function value $J$ and the actual head size of the rolled product for that batch. Assuming a preset overall performance threshold The dimensional tolerance threshold is 0.01. It is 0.05 mm. If the calculated value is... The value is 0.008, and With target size The deviation is 0.03 mm, therefore, due to and If both conditions are met, store the trigger condition. If true, then the system will use the process fingerprint vector of this rolling process, the optimal tension adjustment sequence and roll gap compensation sequence obtained during the rolling process, the measured head and tail dimensions of the rolled piece, and the optimization objective function value of this rolling process. The data, along with the hash value of the previous stored data block, is packaged, encrypted, and stored in the process blockchain. Regarding the online adjustment of the rolled piece size prediction model, the hidden layer weights of the neural network model are fixed after offline training. When a new rolled piece passes through the finishing mill, the online diameter gauge measures the rolled piece size in real time. The control system inputs the current dynamic carbon equivalent, rolled piece temperature, tension setting reference value, and roll gap reference value of the rolled piece into the neural network model, obtaining the output of the last hidden layer. This output is considered a feature vector, which, along with the deviation between the measured size and the model's predicted size, is input into a recursive least squares algorithm with a forgetting factor. This algorithm calculates and updates the weight parameters of the neural network output layer in real time based on this information. For example, the forgetting factor can be set to 0.98, meaning that recent measurement data has a greater impact on parameter updates, while the impact of older data gradually weakens. In this way, even if there are slight changes in the composition of the rolled material or the ambient temperature, the prediction model can quickly adapt and maintain its prediction accuracy, thus providing more accurate size predictions for subsequent rolling processes.

[0085] Through the above technical solutions, this application can significantly improve the reliability and adaptability of precise control of the head and tail lugs of high-speed wire rods. By setting strict data storage trigger conditions, the system only stores production data that meets high-performance and high-precision standards into the process blockchain, effectively avoiding interference from low-quality or abnormal data on subsequent process optimization and model training, thereby ensuring the purity and value of the data in the blockchain. This provides a solid foundation for building a high-quality process knowledge base and realizing traceable production management. Simultaneously, by adopting an online parameter adjustment strategy focused on the output layer, combined with recursive least squares with a forgetting factor, the rolled piece size prediction model can quickly adapt to changes in the production environment and fluctuations in material properties while maintaining computational efficiency, continuously improving prediction accuracy. This online adaptive capability ensures the robustness and effectiveness of the model in long-term operation, thereby improving the control accuracy of the head and tail dimensions of rolled pieces and the product qualification rate, while reducing scrap rate and production costs.

[0086] A high-speed wire head and tail lug precision control system adopts the above-mentioned high-speed wire head and tail lug precision control method.

[0087] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0088] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A high-speed wire head-tail ear sub-accurate control method, characterized in that, include: S1. Collect and calculate the head and tail positions, real-time temperature and dynamic carbon equivalent of the rolled piece, and construct a process fingerprint vector that includes material properties and process status. S2. Using the process fingerprint vector as input, and employing a pre-established rolling mill size prediction model, with the primary objective of meeting the target size and the secondary objective of reducing energy consumption and equipment wear, a model predictive control algorithm is used to solve for the optimal tension adjustment sequence and roll gap compensation sequence within a finite time domain in the future. S3. Send the optimal tension adjustment sequence and roll gap compensation sequence to the rolling mill control system for execution; at the same time, calculate and send the dynamic air pressure setpoint of the first fan of the Stellmore air-cooling line based on the dynamic carbon equivalent and the temperature when the head of the rolled piece leaves the finishing mill. S4. Based on the deviation between the measured dimensions of the rolled piece and the output dimensions of the rolled piece size prediction model, adjust the internal parameters of the rolled piece size prediction model online; and after the rolled piece production is completed, encrypt and store the complete data packet of successful process fingerprint-control parameters-production results to the process blockchain.

2. The method for precise control of the head and tail lugs of high-speed wires according to claim 1, characterized in that, In step S1, the formula for calculating the dynamic carbon equivalent is: in, Indicates dynamic carbon equivalent. Indicates the first The mass percentage of each element Indicates the first The elemental influence coefficient of a given element.

3. The method for precise control of the head and tail lugs of high-speed wires according to claim 1, characterized in that, In step S2, the rolled piece size prediction model is a machine learning model trained based on historical production data; the multi-objective rolling optimization control is achieved by solving the following optimization function: in, This indicates optimizing the objective function value. Indicates the prediction time step. This indicates that the future number of the rolled piece size prediction model outputs the following... Predicted dimensions of rolled parts Indicates the target size. This represents the tension adjustment sequence to be optimized. Indicates the tension reference value. This represents the energy consumption weighting coefficient. This represents the sequence of roll gap compensation amounts to be optimized. Indicates the reference value for roll gap. This represents the wear weighting coefficient.

4. The method for precise control of the head and tail lugs of high-speed wires according to claim 3, characterized in that, In step S3, the dynamic air pressure setpoint of the first-stage fan of the Stellmore air-cooled line is obtained as follows: The dynamic carbon equivalent and the measured temperature when the head leaves the finishing mill last stand are subjected to maximum-minimum normalization to obtain the dynamic carbon equivalent index and temperature index. Based on the carbon equivalent index and temperature index, the dynamic air pressure setpoint of the first-stage fan of the Stellmore air-cooled line is obtained by the following formula: in, This indicates the dynamic air pressure of the first-stage fan in the Stellmore air-cooled line. This indicates the base wind pressure of this steel grade. Indicates the maximum allowable wind pressure adjustment. This represents the dynamic carbon equivalent index. Indicates the temperature index. and These represent the weighting coefficients for carbon equivalent and temperature, respectively, and their sum is 1.

5. The method for precise control of the head and tail lugs of high-speed wires according to claim 3, characterized in that, In step S4, the trigger condition for storing data to the process blockchain is: in, Indicates the storage trigger condition. This indicates optimizing the objective function value. Indicates the overall performance threshold. Indicates the actual size of the head. Indicates the target size. Indicates the allowable tolerance threshold for dimensions. Blockchain storage is triggered only if it is true.

6. The method for precise control of the head and tail lugs of high-speed wires according to claim 3, characterized in that, The tension reference value is taken as the maximum allowable tension adjustment amount of the corresponding frame, and the roll gap reference value is taken as the maximum allowable roll gap compensation amount of the corresponding frame.

7. The method for precise control of the head and tail lugs of high-speed wires according to claim 3, characterized in that, The rolling mill size prediction model in step S2 is a neural network model. Its input features include at least: dynamic carbon equivalent, rolling mill temperature, tension setting reference value, and roll gap reference value. Its output is the predicted rolling mill size value. The neural network model is obtained by using process parameter data from historical production as input features, the measured dimensions of the rolled piece at the corresponding time as the training target, and supervised training using the backpropagation algorithm.

8. The method for precise control of the head and tail lugs of high-speed wires according to claim 5, characterized in that, In step S4, the internal parameters of the rolling stock size prediction model are adjusted online by using the recursive least squares method or the Kalman filter algorithm, with the real-time measured rolling stock size as the observation value, to identify and update the output layer weight parameters of the model online.

9. The method for precise control of the head and tail lugs of high-speed wires according to claim 5, characterized in that, The data packets stored in the process blockchain contain the following information: process fingerprint vector, optimal control parameter sequence, measured head and tail dimensions, optimization objective function value, and hash value of preceding data block.

10. A high-speed wire end lug precision control system, characterized in that, The method for precise control of the head and tail lugs of high-speed wires as described in any one of claims 1-5 is adopted.

Images