Roll gap control method, device and storage medium

Abstract

This invention discloses a roll gap control method, device, and storage medium, belonging to the field of intelligent control technology for steel rolling processes. The method includes: acquiring the actual roll gap value, attenuation factor data, historical deviation data, and billet data to be rolled after the end of the current rolling cycle; calculating the actual roll gap attenuation and inputting it, along with the historical deviation data and billet data, into a time-series prediction model to obtain a predicted roll gap value; generating control parameters to be verified based on the difference between the predicted and actual roll gap values, inputting them into a digital twin model for simulation verification, and obtaining a tolerance simulation value; when the tolerance simulation value meets a preset threshold, adjusting the actual roll gap to the predicted roll gap value before the next cycle's bar enters the rolling mill, thus completing roll gap pre-control. This invention solves the technical problem in related technologies where roll gap adjustment relies on post-event feedback, leading to control lag and potential batch quality risks.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for steel rolling processes, and in particular to a roll gap control method, device, and storage medium. Background Technology

[0002] During bar rolling, the coupled effects of multiple factors, such as roll wear, mill thermal deformation, rolling force fluctuations, and differences in the plastic deformation of steel grades, can trigger a phenomenon called "dynamic attenuation" of the roll gap, directly affecting the dimensional accuracy of the finished bar. Current methods for adjusting the roll gap in bar rolling rely on operator experience or various model systems providing roll gap adjustment parameters to control mill adjustments. These methods, based on measurements obtained after rolling, suffer from lag issues. Furthermore, over time, the mill experiences wear, excessive temperature, and deformation. Therefore, roll gap adjustments predicted by models trained using historical data will be inaccurate even when dealing with the same type of bar.

[0003] In related technologies, the principles of existing solutions are as follows: Figure 1 As shown, typically after rolling, the current bar tolerance value is obtained, and operators adjust the roll gap based on experience or large-scale model predictions to achieve the roll gap adjustment objective. However, existing roll gap adjustment technologies have significant drawbacks: First, they lack awareness of roll gap attenuation modeling, relying solely on a post-hoc feedback mechanism of "set roll gap value - measured deviation" to adjust PID parameters, without considering the gradual characteristics of roll gap attenuation. This results in PID parameter adjustments lagging behind actual roll gap changes, often only discovering tolerance exceedances after billet rolling is complete. Second, the attenuation factor is considered in a single way, modeling only based on different bar specifications. Even when some technologies incorporate roll wear factors, they only use a fixed wear rate for calculation, failing to integrate key influencing factors such as thermal deformation and rolling force fluctuations, leading to large errors in roll gap attenuation calculations. Third, there is a lack of effective parameter verification mechanisms. Newly generated PID parameters are directly sent to the rolling mill for execution without verifying their adaptability to roll gap attenuation trends, easily causing production risks of batch bar tolerance non-compliance.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] One of the technical problems that this invention aims to solve is that the roller gap adjustment in related technologies relies on post-event feedback, which leads to control lag and easily causes batch quality risks.

[0006] To address the aforementioned technical problems, in a first aspect, embodiments of the present invention provide a roll gap control method, the method comprising: after the end of the current rolling cycle, acquiring the current actual roll gap value collected from the rolling mill, measured data of multiple attenuation factors, historical deviation time-series data, and characteristic data of the billet to be rolled; calculating the actual roll gap attenuation amount of the current rolling cycle based on the measured data of the multiple attenuation factors; inputting the actual roll gap attenuation amount, the historical deviation time-series data, and the characteristic data of the billet to be rolled into a pre-trained target time-series prediction model to obtain a predicted roll gap value for at least one future rolling cycle, wherein the predicted roll gap value is a target roll gap value that satisfies the finished product tolerance requirements; based on The difference between the predicted roll gap value for the next rolling cycle and the current actual roll gap value generates a control parameter to be verified. The control parameter to be verified, the predicted roll gap value, and the measured data of the multiple attenuation factors are input into the roll gap digital twin model for simulation verification to obtain a tolerance simulation value. The roll gap digital twin model is used to simulate the mapping relationship between the dynamic changes of the roll gap during rolling and the tolerance of the finished bar. When the tolerance simulation value meets the preset tolerance threshold, the control parameter to be verified is determined as the target control parameter. Before the bar enters the mill in the next rolling cycle, the actual roll gap is adjusted to the predicted roll gap value according to the target control parameter to complete the roll gap control.

[0007] Optionally, calculating the actual roll gap attenuation in the current rolling cycle based on the measured data of the plurality of attenuation factors includes: determining the attenuation component corresponding to each attenuation factor in the measured data of the plurality of attenuation factors, wherein the measured data of the plurality of attenuation factors includes at least roll wear factor data, mill thermal deformation factor data, rolling force fluctuation factor data, and steel grade plastic deformation factor data; and superimposing the attenuation components corresponding to each attenuation factor to obtain the actual roll gap attenuation in the current rolling cycle.

[0008] Optionally, the target time-series prediction model is trained in the following way: Historical rolling data stored in the steel rolling process database is acquired. This historical rolling data includes multiple sets of sample data collected from bars of the same steel grade and specification during historical rolling cycles. Each set of sample data includes input features and a corresponding output label. The input features include the actual roll gap attenuation, sample deviation time-series data, and the characteristic data of the steel billet to be rolled. The output label is the actual roll gap value for the corresponding rolling cycle of each set of sample data. Based on predefined data partitioning rules, the historical rolling data is divided into a training set, a validation set, and a test set. A transfer learning strategy is used to initialize the initial time-series prediction model, resulting in an initialized model. The initial time-series prediction model comprises an input layer, a hidden neural network layer, a fully connected layer, and an output layer. The input layer is used to input the input features, the hidden neural network layer is used to extract the time-series features of roll gap changes, the fully connected layer is used to map the time-series features to the predicted roll gap value, and the output layer is used to output the predicted roll gap value for at least one future rolling cycle. The initial model is trained using the training set to obtain a model to be validated. The model to be validated is validated based on the validation set, and the model parameters are adjusted according to the validation results to obtain an optimized model. The optimized model is tested using the test set, and the target time-series prediction model is output when the test error is less than a preset error threshold.

[0009] Optionally, the method further includes: after each rolling batch is completed, obtaining the actual roll gap data of the batch and the predicted data of the target time-series prediction model, and calculating the deviation; based on the deviation, using an optimization algorithm to fine-tune the weight parameters of the target time-series prediction model, and using the updated model for roll gap prediction of the next rolling batch.

[0010] Optionally, generating the control parameters to be verified based on the difference between the predicted roll gap value for the next rolling cycle and the actual roll gap value for the current rolling cycle includes: obtaining preset initial control parameters, wherein the initial control parameters include an initial proportional parameter, an initial integral parameter, and an initial derivative parameter; determining the difference between the predicted roll gap value for the next rolling cycle and the actual roll gap value for the current rolling cycle as a first control component; accumulating multiple historical deviation values ​​obtained from the historical deviation time series data to obtain a second control component; obtaining the historical deviation value of the previous rolling cycle from the historical deviation time series data, and calculating the difference between the difference and the historical deviation value of the previous rolling cycle to obtain a third control component; correcting the initial proportional parameter according to the first control component, correcting the initial integral parameter according to the second control component, and correcting the initial derivative parameter according to the third control component to obtain the control parameters to be verified.

[0011] Optionally, the method further includes: when the tolerance simulation value does not meet the preset tolerance threshold, repeating the following operations until the newly obtained tolerance simulation value meets the preset tolerance threshold: updating the control parameter to be verified based on the deviation between the current tolerance simulation value and the preset tolerance threshold, and re-inputting the updated parameter into the roll gap digital twin model for simulation verification; determining the control parameter to be verified corresponding to meeting the preset tolerance threshold as the target control parameter.

[0012] Optionally, before the bar enters the mill in the next rolling cycle, the actual roll gap is adjusted to the predicted roll gap value according to the target control parameters. After completing the roll gap control, the method further includes: obtaining the actual tolerance of the finished bar after rolling; comparing the actual tolerance of the finished bar with the target tolerance to obtain the tolerance deviation; adjusting the weight coefficient of the corresponding attenuation factor in the target time series prediction model according to the tolerance deviation, and / or adjusting the parameters of the corresponding simulation module in the roll gap digital twin model.

[0013] On the other hand, the present invention also provides a roll gap control device, the device comprising: an acquisition module, configured to acquire, after the end of the current rolling cycle, the current actual roll gap value collected from the rolling mill, measured data of multiple attenuation factors, historical deviation time series data, and characteristic data of the billet to be rolled; a calculation module, configured to calculate the actual roll gap attenuation amount of the current rolling cycle based on the measured data of the multiple attenuation factors; an input module, configured to input the actual roll gap attenuation amount, the historical deviation time series data, and the characteristic data of the billet to be rolled into a pre-trained target time series prediction model to obtain a predicted roll gap value for at least one future rolling cycle, wherein the predicted roll gap value is a target roll gap value that satisfies the finished product tolerance requirements; and a parameter generation module, configured to... The difference between the predicted roll gap value for the next rolling cycle and the current actual roll gap value generates a control parameter to be verified. A parameter verification module is used to input the control parameter to be verified, the predicted roll gap value, and the measured data of multiple attenuation factors into a roll gap digital twin model for simulation verification to obtain a tolerance simulation value. The roll gap digital twin model is used to simulate the mapping relationship between the dynamic changes in roll gap during rolling and the tolerance of the finished bar. A parameter control module is used to determine the control parameter to be verified as the target control parameter when the tolerance simulation value meets a preset tolerance threshold. Before the bar enters the mill in the next rolling cycle, the actual roll gap is adjusted to the predicted roll gap value according to the target control parameter to complete roll gap control.

[0014] On the other hand, the present invention also provides a machine-readable storage medium storing instructions for causing a machine to perform: a roll gap control method according to any one of the foregoing.

[0015] On the other hand, the present invention also provides an electronic device including one or more processors and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement any one of the roll gap control methods.

[0016] In this embodiment of the invention, the actual attenuation of the roll gap is calculated based on the real-time collected measured data of the attenuation factor. The actual attenuation of the roll gap, historical deviation time series data, and characteristic data of the billet to be rolled are input into a pre-trained target time series prediction model to obtain the predicted value of the roll gap. The difference between the predicted value of the roll gap and the current actual value of the roll gap is used to generate control parameters to be verified. The control parameters to be verified are then simulated and verified using a roll gap digital twin model. This achieves forward-looking prediction of the dynamic changes of the roll gap and adaptive adjustment of the control parameters. It can complete the precise pre-adjustment of the roll gap under different steel grades, specifications, and rolling conditions, thereby improving the roll gap control accuracy, reducing the risk of tolerance exceeding the standard of finished bars, and reducing the scrap rate. This solves the technical problem in related technologies where roll gap adjustment relies on ex-post feedback, leading to control lag and easy batch quality risks. Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a block diagram illustrating the principle of roller gap adjustment in the prior art provided by this invention; Figure 2 This is a flowchart of a roll gap control method provided in an embodiment of the present invention; Figure 3 This is a flowchart of an optional roll gap closed-loop control method provided in an embodiment of the present invention; Figure 4 This is a block diagram of an optional roll gap closed-loop control system provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of a roll gap control device provided in an embodiment of the present invention. Detailed Implementation

[0018] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0019] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with relevant laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0020] This invention provides a method for controlling roller gap. Figure 2 This is a flowchart of a roll gap control method provided in an embodiment of the present invention, such as... Figure 2 As shown, the method includes the following steps: Step S102: After the current rolling cycle ends, acquire the current actual value of the roll gap, measured data of multiple attenuation factors, historical deviation time series data, and characteristic data of the billet to be rolled from the rolling mill.

[0021] Specifically, after the current rolling cycle ends, the data acquisition layer acquires the current actual roll gap value, measured data of multiple attenuation factors, historical deviation time series data, and characteristic data of the billet to be rolled. The data acquisition layer includes a laser thickness gauge, temperature sensor, pressure sensor, encoder, and process database interface. The laser thickness gauge is used to capture the roll wear status in real time; the temperature sensor accurately collects temperature data of the mill and rolls; the pressure sensor acquires force fluctuation information during the rolling process; the encoder monitors changes in roll speed; and the process database interface is used to retrieve basic process parameters such as the plastic deformation coefficient of the steel grade and the thermal expansion coefficient of the mill. Multiple devices collaborate to achieve synchronous acquisition and standardized preprocessing of multi-dimensional data.

[0022] In this embodiment, the measured data of the multiple attenuation factors include roll wear attenuation and mill thermal deformation attenuation. Roll wear attenuation is quantified by combining roll wear rate, rolling time, and steel hardness characteristics. Mill thermal deformation attenuation is calculated based on the thermal expansion characteristics of the mill stand, temperature change range, and key stress dimensions of the stand. The historical deviation time series data is a data sequence composed of the deviation between the actual value and the target value of the roll gap for a preset number of rolling cycles in the past, arranged in chronological order. The characteristic data of the billet to be rolled is obtained through the process database interface, including steel grade, cross-sectional dimensions, entry temperature, and hardness grade. The above data are collected synchronously through a standardized communication protocol, and after preprocessing such as filtering, noise reduction, and unit unification, they are stored in the system cache.

[0023] Step S104: Calculate the actual attenuation of the roll gap in the current rolling cycle based on measured data of multiple attenuation factors.

[0024] In one optional embodiment, the actual roll gap attenuation in the current rolling cycle is calculated based on measured data of multiple attenuation factors, including: determining the attenuation component corresponding to each attenuation factor in the measured data of multiple attenuation factors, wherein the measured data of multiple attenuation factors includes at least roll wear factor data, mill thermal deformation factor data, rolling force fluctuation factor data, and steel grade plastic deformation factor data; and superimposing the attenuation components corresponding to each attenuation factor to obtain the actual roll gap attenuation in the current rolling cycle.

[0025] Optionally, based on the measured data of the multiple attenuation factors obtained in step S102, the actual roll gap attenuation amount for the current rolling cycle is calculated. The specific formula is as follows:

[0026] in, This refers to the amount of roll gap reduction caused by roll wear. This refers to the attenuation caused by thermal deformation of the rolling mill. This is the attenuation caused by fluctuations in rolling force. This is the amount of attenuation compensation caused by the plastic deformation of the steel grade.

[0027] The attenuation components are determined as follows: the attenuation component corresponding to the roll wear factor data is calculated by combining the roll wear amount measured by the laser thickness gauge with the roll wear rate, the current rolling time, and the hardness characteristics of the steel grade; the attenuation component corresponding to the mill thermal deformation factor data is calculated based on the thermal expansion characteristics of the mill stand, the temperature change amplitude collected by the temperature sensor, and the key stress dimensions of the stand; the attenuation component corresponding to the rolling force fluctuation factor data is calculated by combining the rolling force fluctuation data obtained by the pressure sensor with the mill stiffness characteristics; and the attenuation component corresponding to the steel grade plastic deformation factor data is calculated by retrieving the steel grade plastic deformation coefficient from the process database interface.

[0028] Step S106: Input the actual roll gap attenuation, historical deviation time series data, and steel billet characteristic data into the pre-trained target time series prediction model to obtain the roll gap prediction value for at least one future rolling cycle. The roll gap prediction value is the roll gap target value that makes the finished product tolerance meet the requirements.

[0029] In one optional embodiment, the target time-series prediction model is trained as follows: Historical rolling data stored in the steel rolling process database is acquired. This historical rolling data includes multiple sets of sample data collected from bars of the same steel grade and specification during historical rolling cycles. Each set of sample data includes input features and a corresponding output label. The input features include the actual roll gap attenuation, sample deviation time-series data, and the characteristic data of the steel billet to be rolled. The output label is the actual roll gap value for each set of sample data corresponding to the rolling cycle. Based on predefined data partitioning rules, the historical rolling data is divided into a training set, a validation set, and a test set. A transfer learning strategy is then used to initially train the time-series prediction model. The process involves several steps: First, an initial model is obtained, comprising an input layer, a hidden neural network layer, a fully connected layer, and an output layer. The input layer receives input features, the hidden neural network layer extracts temporal features of roll gap changes, the fully connected layer maps these temporal features to predicted roll gap values, and the output layer outputs predicted roll gap values ​​for at least one future rolling cycle. The initial model is then trained using a training set to obtain a model to be validated. This model is then validated using a validation set, and the model parameters are adjusted based on the validation results to obtain an optimized model. Finally, the optimized model is tested using a test set, and the target temporal prediction model is output when the test error is less than a preset error threshold.

[0030] Optionally, in practical applications, the actual roll gap attenuation calculated in step S104, together with the historical deviation time-series data and the characteristic data of the billet to be rolled obtained in step S102, constitute the input of the target time-series prediction model. Specifically, the input of the target time-series prediction model is an 8-dimensional input feature vector, including: the time-series characteristics of roll gap attenuation in the first three rolling cycles, the deviation of the incoming material cross-sectional dimensions, the real-time value of the rolling temperature, the average value of the rolling force, the roll wear, the mill stand temperature, the hardness coefficient of the steel grade, and the rolling speed. The above features are standardized to the [0,1] interval, and the standardization formula is: xnorm=(x-xmin) / (xmax-xmin), where xmin and xmax are the historical minimum and maximum values ​​of each feature, extracted and determined from the process database to eliminate the influence of dimensions.

[0031] Furthermore, the aforementioned target time-series prediction model adopts a lightweight gated recurrent unit (GRU) architecture that integrates rolling mechanism. Its core computational logic extracts the core features of roll gap time-series data through dynamic updates of the update gate, reset gate, and hidden state. Let... The 8-dimensional feature vector input at time t, The hidden state from the previous rolling cycle is represented by the following model calculation process: Update Gate: z t = σ(Wz·[h t-1 ,x t]+bz), used to control the degree of historical information retention; reset gate: r t = σ(Wr·[h t-1 ,x t ]+br), used to control the degree of historical information ignoring; Candidate hidden state: t = tanh(Wh·[r t ⊙h t-1 ,x t ]+bh), integrate current input with historical features to extract the temporal pattern of the roll gap; current hidden state: h t = (1-z t )⊙h t-1 +z t ⊙ t The system integrates and updates the gate and candidate states; finally, it outputs the result through roll gap prediction. t = Wy·h t + by maps to the actual roll gap value. Here, Wz and Wr have dimensions of 64×(64+8), Wh has a dimension of 64×(64+8), Wy has a dimension of 5×64, bz, br, and bh have dimensions of 64×1, and by has a dimension of 5×1; σ is the Sigmoid activation function, used to control the on / off degree of the gating unit; tanh is the hyperbolic tangent activation function, used to handle the nonlinear mapping of the hidden state; ⊙ is the Hadamard product, realizing element-wise multiplication.

[0032] It should be noted that the above model network structure adopts a series relationship of input layer → GRU hidden layer → fully connected layer → output layer. The input layer has an 8-dimensional dimension, corresponding to the above 8 types of core input features; the GRU hidden layer is a single-layer unidirectional temporal structure with 64 neurons, and uses batch normalization to accelerate the model training convergence speed. There is no dropout layer to avoid overfitting to industrial small sample data; the fully connected layer has 32 neurons and uses the ReLU (Linear Rectified Activation Function) activation function (ReLU(x)=max(0,x)) to realize the nonlinear mapping from the temporal features of the GRU hidden layer to the roll gap prediction value; the output layer has a 5-dimensional dimension, no activation function, and the output value is denormalized to restore the actual roll gap value (the denormalization formula is x=xnorm×(xmax-xmin)+xmin), directly outputting the roll gap prediction value for the next 1-5 rolling cycles.

[0033] The training process for the above-mentioned target time series prediction model is as follows: First, historical rolling data of bars of the same steel grade and specification were extracted from the process database. Each sample group included the aforementioned 8-dimensional input features and the actual roll gap values ​​corresponding to 1-5 rolling cycles. The dataset was divided into training, validation, and test sets in a 7:2:1 ratio. Second, a transfer learning strategy was used to initialize the model, based on the pre-trained model weights of the same series of steel grades, to shorten the training cycle and improve the model's generalization ability. Then, the training parameters were set: the optimizer was Adam, the initial learning rate was set to 0.001, and it decayed to 0.8 times its original value every 10 training epochs; the batch size was set to 32, the number of training epochs was set to 50, and an early stopping mechanism (patience=5) was set, i.e., training was terminated early when the validation set loss did not decrease for 5 consecutive epochs to avoid overfitting. The loss function used was mean squared error (MSE).

[0034] Where N is the sample size. i y is the model's predicted value. i This is the actual roll gap value.

[0035] During the training process described above, the training set data is input into the initialized model, and the weight matrices and bias terms of each layer are updated through backpropagation. After each training round, the model performance is verified using the validation set data and the learning rate is adjusted. After training is completed, the model accuracy is tested using the test set data to ensure that the prediction error is within the allowable range.

[0036] After the model is deployed, the actual roll gap attenuation ΔS_total calculated in step S104, together with the historical deviation time series data and the characteristic data of the billet to be rolled, constitute the input feature vector. By inputting the vector into the model, the predicted roll gap value for at least one future rolling cycle can be obtained.

[0037] In an optional embodiment, after outputting the target time-series prediction model, the method further includes: after each rolling batch is completed, obtaining the actual roll gap data of the batch and the prediction data of the target time-series prediction model, and calculating the deviation; based on the deviation, using an optimization algorithm to fine-tune the weight parameters of the target time-series prediction model, and using the updated model for roll gap prediction of the next rolling batch.

[0038] Optionally, the optimization algorithm can be either gradient descent or the Adam optimization algorithm. For example, the Adam optimization algorithm can be used in scenarios requiring fast convergence, while stochastic gradient descent can be used in scenarios with limited computational resources.

[0039] Optionally, to improve the stability and efficiency of model updates, the fine-tuning process can be constrained by combining a random forest incremental learning algorithm. Specifically, the absolute value of the deviation between the actual roll gap data and the model prediction data of the current batch is compared with a preset deviation threshold; the fine-tuning operation of the weight parameters is triggered only when the deviation exceeds the threshold; if the deviation is within the threshold range, the current model parameters remain unchanged.

[0040] Furthermore, a sliding window mechanism can be used for the above fine-tuning process. Each time a fine-tuning is performed, not only is the rolling data of the current batch used, but historical data from the most recent M rolling batches are also acquired to form a fine-tuning training set, where M is a positive integer. This enhances the model's adaptability to recent changes in operating conditions while suppressing the negative impact of outlier data from a single batch on model accuracy. After fine-tuning, the updated model weights are saved to the process database and simultaneously deployed to the online inference environment for roll gap prediction in the next rolling batch.

[0041] It should be noted that by incrementally updating the model after each rolling batch, the target time-series prediction model can track the changing patterns of gradually changing working conditions such as roll wear and thermal deformation in a timely manner, and always maintain the prediction accuracy of the current rolling process. At the same time, through the deviation threshold triggering mechanism and sliding window strategy, the interference of abnormal data in a single batch is effectively suppressed, and the model overfitting is avoided. Under the premise of ensuring prediction accuracy, the frequency of online updates and computational overhead are reduced.

[0042] Step S108: Based on the difference between the predicted roll gap value for the next rolling cycle and the actual roll gap value for the current rolling cycle, generate control parameters to be verified.

[0043] In one optional embodiment, generating control parameters to be verified based on the difference between the predicted roll gap value of the next rolling cycle and the actual roll gap value of the current rolling cycle includes: obtaining preset initial control parameters, wherein the initial control parameters include an initial proportional parameter, an initial integral parameter, and an initial derivative parameter; determining the difference between the predicted roll gap value of the next rolling cycle and the actual roll gap value of the current rolling cycle as a first control component; accumulating multiple historical deviation values ​​obtained from historical deviation time series data to obtain a second control component; obtaining the historical deviation value of the previous rolling cycle from the historical deviation time series data, and calculating the difference between the difference and the historical deviation value of the previous rolling cycle to obtain a third control component; correcting the initial proportional parameter according to the first control component, correcting the initial integral parameter according to the second control component, and correcting the initial derivative parameter according to the third control component to obtain the control parameters to be verified.

[0044] For the above embodiments and optional embodiments, the determination of initial control parameters adopts a three-level process: process benchmark matching, offline simulation tuning, and on-site trial rolling calibration. In the process benchmark matching stage, the historical best rolling process package for bars of the same steel grade and specification is retrieved from the process database, and the benchmark PID parameters under the corresponding mill type, rolling speed, and target roll gap are extracted as initial estimated values. For new steel grades and specifications, the initial benchmark parameters are calculated using the critical proportionality method based on the mill's rated parameters, roll gap adjustment range, and rolling force upper limit. In the offline simulation tuning stage, the estimated values ​​are input into the digital twin model to simulate no-load and low-load rolling conditions, observe the roll gap adjustment response speed, overshoot, and steady-state error, and optimize the initial parameters through simulation iteration. In the on-site trial rolling calibration stage, a small batch of steel billets is selected for trial rolling, and the actual roll gap value, rolling force, and finished product tolerance data are collected during the trial rolling process. The PID parameters are fine-tuned to ensure that the roll gap adjustment steady-state error is within the allowable range, and finally, the initial control parameters under this condition are determined and stored in the process database for later use.

[0045] Secondly, three control components are calculated: the difference between the predicted roll gap value of the next rolling cycle and the actual roll gap value of the current rolling cycle is determined as the first control component; multiple historical deviation values ​​obtained from historical deviation time series data are accumulated to obtain the second control component; the historical deviation value of the previous rolling cycle is obtained from the historical deviation time series data, and the difference between the difference and the historical deviation value of the previous rolling cycle is calculated to obtain the third control component.

[0046] Then, the initial control parameters are corrected based on the three control components to obtain the control parameters to be verified. The specific correction rules are as follows: the initial proportional parameter is corrected based on the first control component; when the first control component increases, the initial proportional parameter is increased, and when the first control component decreases, the initial proportional parameter is decreased. The initial integral parameter is corrected based on the second control component; when the second control component is continuously present, the initial integral parameter is increased, and when the second control component changes rapidly, the initial integral parameter is decreased. The initial derivative parameter is corrected based on the third control component; when the third control component increases, the initial derivative parameter is increased, and when the third control component decreases, the initial derivative parameter is decreased.

[0047] It should be noted that the next rolling cycle is the rolling cycle adjacent to the current rolling cycle among at least one future rolling cycle mentioned above. That is, by using the difference between the predicted roll gap value of the next rolling cycle and the actual roll gap value of the current rolling cycle as the first control component, a forward-looking response to future roll gap deviations is achieved, allowing the control parameters to adapt to upcoming roll gap changes in advance. The second control component performs integral correction on historical accumulated deviations, effectively eliminating steady-state errors caused by slowly changing factors such as roll wear and thermal deformation. The third control component performs differential correction on the deviation change trend, suppressing dynamic deviations caused by factors such as fluctuations in incoming material temperature and sudden changes in rolling force. The synergistic effect of the three levels of control components achieves adaptive dynamic adjustment of proportional, integral, and derivative parameters. While ensuring the speed of adjustment response, it avoids problems such as overshoot, integral saturation, and system oscillation, fundamentally avoiding the lag defects of traditional ex-post adjustments, and significantly improving the accuracy and stability of roll gap control.

[0048] Step S110: Input the control parameters to be verified, the predicted roll gap value, and the measured data of multiple attenuation factors into the roll gap digital twin model for simulation verification to obtain the tolerance simulation value. The roll gap digital twin model is used to simulate the mapping relationship between the dynamic changes of the roll gap and the tolerance of the finished bar during the rolling process.

[0049] Specifically, the roll gap digital twin model adopts a three-layer deep fusion architecture: a geometric twin layer, a physical twin layer, and a data twin layer. The geometric twin layer, based on rolling mill equipment drawings and on-site measured data, recreates the geometric shape, dimensional parameters, and assembly relationships of core components such as rolls, mill stands, hydraulic pressing actuators, and billet conveying mechanisms at a 1:1 scale. Core parameters include roll diameter, roll length, mill stand stress length, hydraulic pressing actuator stroke, and billet conveying speed. This layer is linked with the laser thickness gauge and displacement sensors in the data acquisition layer to acquire data such as roll wear and equipment deformation in real time. It dynamically updates the roll surface profile, equipment dimensions, and other geometric parameters every rolling cycle to ensure synchronization with the actual rolling mill equipment status.

[0050] The physical twin layer serves as the core simulation layer, integrating the four core factors of roll gap dynamic decay (roll wear, mill thermal deformation, rolling force fluctuation, and steel grade plastic deformation) to construct a four-field coupled simulation logic encompassing the wear field, temperature field, mechanical field, and strip plastic deformation field. The specific simulation model is as follows: The roll wear simulation adopts the Archard wear model, with the expression: W=K×F×S / H Where W is the roll wear amount (mm), K is the wear coefficient, F is the rolling force (kN), S is the contact area between the roll and the billet (mm²), and H is the roll hardness (HV). The thermal deformation simulation uses the heat conduction equation, which is expressed as follows: T / t=a( ²T / x²+ ²T / y²+ ²T / z²) Where T is temperature (°C), t is time (s), a is the thermal diffusivity, and x, y, z are spatial coordinates. The input is real-time temperature data of the mill stand and rolls collected by temperature sensors to simulate roll gap changes caused by thermal deformation of the mill. The mechanical field simulation uses an elastic deformation model based on Hooke's law, expressed as: σ = Eε, where σ is stress (MPa), E is the elastic modulus, and ε is strain. Input is rolling force monitoring data to simulate roll gap offset caused by elastic deformation of the mill stand and rolls. The plastic deformation simulation uses the Mises yield criterion, expressed as:

[0051] in, For equivalent stress, σ1, σ2, and σ3 are the principal stresses in the three directions, respectively. To simulate the effect of plastic deformation of steel on roll gap, the simulation step size was set to 0.5s, synchronized with the actual rolling cycle, to ensure that the simulation process is completely consistent with the actual rolling process.

[0052] The data twin layer, serving as the core linkage layer, connects to the edge computing gateway via an industrial communication protocol. It receives in real-time attenuation factor data from the data acquisition layer, control parameters to be verified from the model calculation layer, and roll gap prediction values ​​output by the target time-series prediction model. It also outputs core indicators such as the simulated roll gap change trend, predicted tolerance values ​​for finished bars, and roll gap adjustment overshoot in real-time, transmitting them to the model calculation layer, execution control layer, and process database. Every rolling cycle, the simulation data is compared with actual rolling data, and the core parameters of the physical twin layer are calibrated using the least squares method to ensure a high degree of consistency between simulation accuracy and actual operating conditions. After verification, the control parameters, simulated roll gap change data, tolerance prediction data, and deviation analysis results from this verification are stored in the process database as the parameter initialization basis for subsequent rolling of new steel grades and new bar specifications.

[0053] Through the collaborative work of the above three-layer architecture, after receiving the control parameters to be verified, the digital twin model starts multi-physics coupling simulation to simulate the dynamic decay process of the roll gap in multiple future rolling cycles, and simultaneously records the real-time roll gap value, roll gap decay rate, virtual finished bar tolerance value and roll gap adjustment overshoot, and finally outputs the tolerance simulation value.

[0054] It should be noted that by using a digital twin model to simulate and verify the control parameters to be verified, the effectiveness of the control parameters can be predicted before actual execution, achieving parameter verification with zero trial and error cost, and avoiding the risk of batch bar material tolerance non-compliance that may be caused by directly issuing new parameters for execution.

[0055] Step S112: If the tolerance simulation value meets the preset tolerance threshold, the control parameter to be verified is determined as the target control parameter. Before the bar enters the mill in the next rolling cycle, the actual roll gap is adjusted to the predicted roll gap value according to the target control parameter to complete the roll gap control.

[0056] In one optional embodiment, when the tolerance simulation value does not meet the preset tolerance threshold, the following operations are repeated until the newly obtained tolerance simulation value meets the preset tolerance threshold: the control parameter to be verified is updated based on the deviation between the current tolerance simulation value and the preset tolerance threshold, and the updated parameter is re-input into the roll gap digital twin model for simulation verification; the control parameter to be verified corresponding to the preset tolerance threshold is determined as the target control parameter.

[0057] Specifically, when the simulated tolerance value is within the allowable range, the target control parameters are sent to the roll gap position closed-loop system. This system uses a proportional-integral control valve to drive the rolls to press down or lift up, thereby adjusting the roll gap. Before the bar enters the mill in the next rolling cycle, the roll gap is pre-adjusted to bring the actual roll gap to the predicted value.

[0058] Optionally, the roll gap position closed-loop system uses roll gap displacement sensors and pressure sensors as sensing foundations to continuously collect real-time operating data such as the actual roll gap position and rolling force. This data is transmitted to the signal processing module through a standardized data interface. After noise reduction, filtering, and deviation calculation from the predicted roll gap value, a deviation signal is generated. Combined with the target control parameters verified by the digital twin model, the closed-loop control core module generates control commands through proportional-integral operations. These commands are transmitted to the hydraulic pressing actuator, which drives the servo valve to adjust the oil inlet and outlet of the hydraulic cylinder, thereby achieving precise movement of the roll position and completing the roll gap adjustment.

[0059] It should be noted that by determining the control parameter to be verified as the target control parameter and issuing it for execution only when the tolerance simulation value meets the preset tolerance threshold, it is ensured that only verified and effective parameters can be used for actual roll gap control. This avoids the risk of batch bar tolerance non-compliance that may be caused by the direct execution of new parameters in traditional methods. At the same time, by completing the roll gap pre-adjustment before the bars enter the mill in the next rolling cycle, the roll gap control is changed from passive correction to active prediction, which effectively avoids the control lag defects caused by ex-post feedback and reduces the scrap rate.

[0060] In one optional embodiment, before the bar enters the mill in the next rolling cycle, the actual roll gap is adjusted to the predicted roll gap value according to the target control parameters. After the roll gap control is completed, the method further includes: obtaining the actual tolerance of the finished bar after rolling; comparing the actual tolerance of the finished bar with the target tolerance to obtain the tolerance deviation; adjusting the weight coefficient of the corresponding attenuation factor in the target time series prediction model according to the tolerance deviation, and / or adjusting the parameters of the corresponding simulation module in the roll gap digital twin model.

[0061] Specifically, after the billet is rolled, the actual tolerance data is obtained through the finished product tolerance inspection process. This actual tolerance is compared with the preset target tolerance to obtain the tolerance deviation, which is then simultaneously transmitted to the target time-series prediction model and the roll gap digital twin model. For the target time-series prediction model, the weight allocation of relevant attenuation factors such as roll wear, thermal deformation, and rolling force influence in the model is adjusted by comparing the actual roll gap attenuation corresponding to the actual tolerance with the attenuation results predicted by the model. For the roll gap digital twin model, the simulation logic and parameters of the geometric twin layer, physical twin layer, and data twin layer are corrected by comparing the actual rolling process with the simulation process.

[0062] The model correction process employs targeted optimization logic, focusing on adjusting the weight coefficients of the corresponding attenuation factors in the target time-series prediction model and the parameters of the corresponding simulation modules in the roll gap digital twin model, specifically targeting the influencing factors indicated by tolerance deviations. Simultaneously, a random forest incremental learning algorithm is used to update only model parameters where errors exceed acceptable limits, avoiding excessive iteration that could impact system stability. This feedback information is also stored as historical data, providing a reference for subsequent similar operating conditions. Through continuous feedback iteration, the target time-series prediction model and the roll gap digital twin model develop adaptive evolutionary capabilities.

[0063] It should be noted that by using a closed-loop iterative mechanism driven by finished product tolerance feedback, the core parameters of the target time-series prediction model and the simulation logic of the roll gap digital twin model are continuously calibrated, thereby realizing the adaptive optimization and iterative upgrade of the control system and continuously improving its adaptability and adjustment accuracy to complex rolling conditions.

[0064] Through steps S102 to S112, the actual attenuation of the roll gap is calculated based on the real-time collected measured data of the attenuation factor. The actual attenuation of the roll gap, historical deviation time series data, and characteristic data of the billet to be rolled are input into the pre-trained target time series prediction model to obtain the predicted value of the roll gap. The control parameters to be verified are generated based on the difference between the predicted value of the roll gap and the current actual value of the roll gap. The control parameters to be verified are then simulated and verified in conjunction with the roll gap digital twin model. This achieves forward-looking prediction of the dynamic changes of the roll gap and adaptive adjustment of the control parameters. It can complete the precise pre-adjustment of the roll gap under different steel grades, different specifications, and different rolling conditions, thereby improving the roll gap control accuracy, reducing the risk of tolerance exceeding the standard of finished bar, and reducing the scrap rate. This solves the technical problem in related technologies where roll gap adjustment relies on ex-post feedback, resulting in control lag and easy to cause batch quality risks.

[0065] Based on the above embodiments and optional embodiments, the present invention proposes an implementation method for an optional roll gap control method. Figure 3 A flowchart of an optional roll gap closed-loop control method provided in an embodiment of the present invention is shown below. Figure 3 As shown, this embodiment uses a bar production line as the application scenario, the mill roll gap as the control object, and the finished bar tolerance as the final control objective, combined with... Figure 4 The layered architecture shown (data acquisition layer, model calculation layer, parameter verification layer, execution control layer, and feedback iteration layer) has the following overall control flow steps: Step S202, Data Acquisition and Standardization Processing: After the current rolling cycle ends, the actual value of the current roll gap, measured data of multiple attenuation factors, historical deviation time series data, characteristic data of the billet to be rolled, and the target roll gap value for the next cycle are collected by the data acquisition layer through the collaborative collection of various devices. After filtering and noise reduction, outlier removal and unit unification preprocessing, the data are stored in the system cache area for subsequent steps to call.

[0066] Step S204, based on the above-collected measured data of the attenuation factor, in Figure 4 In the model calculation layer shown, the actual roll gap attenuation for the current rolling cycle is calculated according to the formula provided in the previous embodiments. The measured attenuation factor data includes at least roll wear factor data, mill thermal deformation factor data, rolling force fluctuation factor data, and steel grade plastic deformation factor data. The actual roll gap attenuation is the sum of the attenuation components corresponding to each attenuation factor.

[0067] Step S206 further executes the relevant steps of model prediction in the above-mentioned model calculation layer. The actual roll gap attenuation amount from step S204, the historical deviation time-series data from step S202, and the characteristic data of the billet to be rolled are input into the pre-trained GRU time-series prediction model, which outputs the predicted roll gap values ​​for the next 1-5 rolling cycles. The predicted roll gap values ​​are the target roll gap values ​​that ensure the finished product tolerances meet the requirements. These predicted roll gap values ​​are synchronously transmitted to subsequent steps and the digital twin model.

[0068] Step S208: The model calculation layer generates the control parameters. Preset initial control parameters are obtained, including initial proportional parameters, initial integral parameters, and initial derivative parameters. The difference between the predicted roll gap value for the next rolling cycle and the actual roll gap value for the current rolling cycle is determined as the first control component. Multiple historical deviation values ​​from the historical deviation time series data are accumulated to obtain the second control component. The historical deviation value from the previous rolling cycle is obtained from the historical deviation time series data, and the difference between the difference and the historical deviation value from the previous rolling cycle is calculated to obtain the third control component. The initial proportional parameters are corrected based on the first control component, the initial integral parameters are corrected based on the second control component, and the initial derivative parameters are corrected based on the third control component to obtain the control parameters to be verified.

[0069] The physical quantities on which the correction of control parameters (PID parameters) is based are acquired in real time by the data acquisition layer or predicted by the model calculation layer. These include: physical quantities related to roll gap (real-time actual roll gap value, roll gap deviation value, roll gap deviation change rate), physical quantities of mill operating conditions (real-time rolling force, rolling force fluctuation amplitude, roll temperature, mill stand deformation, cumulative roll wear), and physical quantities of incoming material and finished product (bar incoming cross-sectional dimension deviation, rolling temperature, actual tolerance of finished product, tolerance deviation value).

[0070] It should be noted that the initial parameters are corrected according to the following PID adjustment rules: The proportional coefficient P is adjusted as follows: when the roll gap deviation is larger and the rolling force fluctuation is more severe, the P value is appropriately increased to improve the adjustment response speed; when the roll gap is close to the set value and the operating condition is stable, the P value is decreased to avoid overshoot. The integral coefficient I is adjusted as follows: when the roll gap steady-state deviation persists and the roll wear or thermal deformation decays slowly, the I value is increased to eliminate static error; when the deviation changes rapidly and the decay rate is fast, the I value is decreased to prevent integral saturation. The derivative coefficient D is adjusted as follows: when the roll gap deviation changes rapidly and the incoming material temperature fluctuates greatly, the D value is increased to suppress dynamic deviation; when the operating condition is stable and the deviation changes slowly, the D value is decreased to avoid system oscillation. Through the above corrections, the final PID control parameters to be verified are generated.

[0071] Step S210, in Figure 4The parameter verification layer performs relevant parameter verification steps, specifically: inputting the control parameter to be verified, the predicted roll gap value, and the measured data of multiple attenuation factors into the roll gap digital twin model, and conducting simulation verification according to a three-layer architecture of geometric twin layer, physical twin layer, and data twin layer, and outputting the tolerance simulation value; in this step, if the tolerance simulation value meets the preset tolerance threshold, the control parameter to be verified is determined as the target control parameter, and the process proceeds to step S212; if the threshold is not met, the control parameter to be verified is updated based on the deviation between the current tolerance simulation value and the preset tolerance threshold, and is re-input into the roll gap digital twin model for simulation verification until the preset tolerance threshold is met.

[0072] Step S212, this step is by Figure 4 The execution control layer completes the process. Specifically, when the tolerance simulation value meets the preset tolerance threshold, the target control parameters are sent to the roll gap position closed-loop system. This system collects real-time operating data such as the actual roll gap position and rolling force, and combines it with the target control parameters to drive the hydraulic pressing actuator through a proportional-integral control valve. Before the bar enters the mill in the next rolling cycle, the roll gap is pre-adjusted to adjust the actual roll gap to the predicted roll gap value. During the rolling process, dynamic micro-corrections are made based on real-time monitoring data to ensure roll gap accuracy.

[0073] Step S214, this step is by Figure 4 The feedback iteration layer is completed. Specifically, after rolling, the actual tolerance of the finished bar is obtained, compared with the target tolerance to obtain the tolerance deviation, and simultaneously fed back to the target time-series prediction model and the roll gap digital twin model. For the target time-series prediction model, the weight coefficients of the corresponding attenuation factors are adjusted; for the roll gap digital twin model, the parameters of the corresponding simulation modules are adjusted. The model correction process adopts directional optimization logic, focusing on adjusting the influencing factors pointed to by the tolerance deviation, and at the same time combining the random forest incremental learning algorithm to update only the model parameters whose errors exceed the acceptable range, avoiding excessive iteration that affects system stability. The above feedback information is also stored as historical data to achieve adaptive optimization of the control system.

[0074] This invention also provides a roll gap control device. Figure 5 This is a schematic diagram of a roll gap control device provided in an embodiment of the present invention, as shown below. Figure 5 As shown, the device includes the following steps: an acquisition module 502, a calculation module 504, an input module 506, a parameter generation module 508, a parameter verification module 510, and a parameter control module 512, wherein: The acquisition module 502 is used to acquire the current roll gap actual value, multiple attenuation factor measured data, historical deviation time series data and steel billet characteristic data collected from the rolling mill after the end of the current rolling cycle. The calculation module 504, connected to the acquisition module 502, is used to calculate the actual attenuation of the roll gap in the current rolling cycle based on the measured data of multiple attenuation factors. The input module 506, connected to the calculation module 504, is used to input the actual roll gap attenuation, historical deviation time series data and the characteristic data of the billet to be rolled into the pre-trained target time series prediction model to obtain the roll gap prediction value for at least one future rolling cycle. The roll gap prediction value is the roll gap target value that makes the finished product tolerance meet the requirements. The parameter generation module 508, connected to the input module 506, is used to generate control parameters to be verified based on the difference between the predicted roll gap value of the next rolling cycle and the actual roll gap value of the current rolling cycle. The parameter verification module 510 is connected to the parameter generation module 508. It is used to input the control parameters to be verified, the predicted roll gap value, and the measured data of multiple attenuation factors into the roll gap digital twin model for simulation verification to obtain the tolerance simulation value. The roll gap digital twin model is used to simulate the mapping relationship between the dynamic change of roll gap and the tolerance of finished bar during the rolling process. The parameter control module 512 is connected to the parameter verification module 510. It is used to determine the control parameter to be verified as the target control parameter when the tolerance simulation value meets the preset tolerance threshold. Before the bar enters the mill in the next rolling cycle, the actual roll gap is adjusted to the predicted roll gap value according to the target control parameter to complete the roll gap control.

[0075] This invention also provides a machine-readable storage medium storing instructions for causing a machine to execute: according to any of the roll gap control methods described above.

[0076] This invention also provides a processor for running a program, wherein the program is executed to perform any of the roll gap control methods described above.

[0077] This invention also provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements any of the roll gap control methods described above.

[0078] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0079] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0080] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0081] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0082] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0083] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0084] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0085] It should also be noted that 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 process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0086] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for controlling roll gap, characterized in that, include: After the current rolling cycle ends, acquire the current actual value of the roll gap, measured data of multiple attenuation factors, historical deviation time series data, and characteristic data of the billet to be rolled from the rolling mill; Based on the measured data of the multiple attenuation factors, the actual attenuation of the roll gap in the current rolling cycle is calculated. The actual roll gap attenuation, the historical deviation time series data, and the characteristic data of the billet to be rolled are input into the pre-trained target time series prediction model to obtain the roll gap prediction value for at least one future rolling cycle, wherein the roll gap prediction value is the roll gap target value that makes the finished product tolerance meet the requirements. Based on the difference between the predicted roll gap value for the next rolling cycle and the actual roll gap value for the current rolling cycle, control parameters to be verified are generated. The control parameters to be verified, the predicted roll gap value, and the measured data of the multiple attenuation factors are input into the roll gap digital twin model for simulation verification to obtain the tolerance simulation value. The roll gap digital twin model is used to simulate the mapping relationship between the dynamic changes in roll gap during rolling and the tolerance of the finished bar. If the simulated tolerance value meets the preset tolerance threshold, the control parameter to be verified is determined as the target control parameter. Before the bar enters the mill in the next rolling cycle, the actual roll gap is adjusted to the predicted roll gap value according to the target control parameter to complete the roll gap control.

2. The method according to claim 1, characterized in that, The calculation of the actual roll gap attenuation in the current rolling cycle based on the measured data of the multiple attenuation factors includes: The attenuation components corresponding to each attenuation factor in the measured data of the plurality of attenuation factors are determined respectively. The measured data of the plurality of attenuation factors include at least roll wear factor data, mill hot deformation factor data, rolling force fluctuation factor data and steel grade plastic deformation factor data. The actual roll gap attenuation is obtained by superimposing the attenuation components corresponding to the attenuation factors for the current rolling cycle.

3. The method according to claim 1, characterized in that, The target time series prediction model is trained in the following way: The historical rolling data stored in the steel rolling process database is obtained. The historical rolling data includes multiple sets of sample data collected in the historical rolling cycle of bars of the same steel grade and specification. Each set of sample data includes input features and corresponding output labels. The input features include the actual attenuation of the sample roll gap, the time series data of the sample deviation, and the characteristic data of the sample billet to be rolled. The output label is the actual roll gap value of the rolling cycle corresponding to each set of sample data. Based on predefined data partitioning rules, the historical rolling data is divided into a training set, a validation set, and a test set; A transfer learning strategy is used to initialize the initial time series prediction model, resulting in an initialized model. The initial time series prediction model includes an input layer, a hidden neural network layer, a fully connected layer, and an output layer. The input layer is used to input the input features, the hidden neural network layer is used to extract the time series features of roll gap changes, the fully connected layer is used to map the time series features to the roll gap prediction value, and the output layer is used to output the roll gap prediction value for at least one future rolling cycle. The initial model is trained using the training set to obtain the model to be validated; The model to be verified is verified based on the verification set, and the model parameters are adjusted according to the verification results to obtain the optimized model. The optimized model is tested using the test set, and the target time series prediction model is output when the test error is less than the preset error threshold.

4. The method according to claim 3, characterized in that, The method further includes: After each rolling batch is completed, the actual roll gap data of that batch is obtained and the predicted data of the target time series prediction model is compared, and the deviation is calculated. Based on the deviation, the weight parameters of the target time-series prediction model are fine-tuned using an optimization algorithm, and the updated model is used for roll gap prediction in the next rolling batch.

5. The method according to claim 1, characterized in that, The difference between the predicted roll gap value based on the next rolling cycle and the actual roll gap value at present generates control parameters to be verified, including: Obtain preset initial control parameters, wherein the initial control parameters include initial proportional parameters, initial integral parameters, and initial derivative parameters; The difference between the predicted roll gap value for the next rolling cycle and the actual roll gap value for the current rolling cycle is determined as the first control component. The second control component is obtained by summing multiple historical deviation values ​​obtained from the historical deviation time series data. The historical deviation value of the previous rolling cycle is obtained from the historical deviation time series data, and the difference between the difference value and the historical deviation value of the previous rolling cycle is calculated to obtain the third control component. The initial proportional parameter is corrected according to the first control component, the initial integral parameter is corrected according to the second control component, and the initial derivative parameter is corrected according to the third control component to obtain the control parameter to be verified.

6. The method according to claim 1, characterized in that, The method further includes: When the tolerance simulation value does not meet the preset tolerance threshold, repeat the following operation until the newly obtained tolerance simulation value meets the preset tolerance threshold: The control parameters to be verified are updated based on the deviation between the current tolerance simulation value and the preset tolerance threshold, and the updated parameters are re-input into the roll gap digital twin model for simulation verification. The control parameter to be verified that meets the preset tolerance threshold is determined as the target control parameter.

7. The method according to any one of claims 1-6, characterized in that, Before the bar enters the mill in the next rolling cycle, the actual roll gap is adjusted to the predicted roll gap value according to the target control parameters. After completing the roll gap control, the method further includes: Obtain the actual tolerance of the finished bar stock after rolling; The actual tolerance of the finished bar stock is compared with the target tolerance to obtain the tolerance deviation; Based on the tolerance deviation, adjust the weight coefficient of the corresponding attenuation factor in the target time series prediction model, and / or adjust the parameters of the corresponding simulation module in the roll gap digital twin model.

8. A roll gap control device, characterized in that, include: The acquisition module is used to acquire the current roll gap actual value, multiple attenuation factor measured data, historical deviation time series data and steel billet characteristic data collected from the rolling mill after the end of the current rolling cycle. The calculation module is used to calculate the actual roll gap attenuation amount in the current rolling cycle based on the measured data of the multiple attenuation factors. The input module is used to input the actual roll gap attenuation, the historical deviation time series data, and the characteristic data of the billet to be rolled into the pre-trained target time series prediction model to obtain the roll gap prediction value for at least one future rolling cycle, wherein the roll gap prediction value is the roll gap target value that makes the finished product tolerance meet the requirements. The parameter generation module is used to generate control parameters to be verified based on the difference between the predicted roll gap value and the actual roll gap value in the next rolling cycle. The parameter verification module is used to input the control parameters to be verified, the predicted roll gap value, and the measured data of the multiple attenuation factors into the roll gap digital twin model for simulation verification, thereby obtaining the tolerance simulation value. The roll gap digital twin model is used to simulate the mapping relationship between the dynamic changes in the roll gap during rolling and the tolerance of the finished bar. The parameter control module is used to determine the control parameter to be verified as the target control parameter when the tolerance simulation value meets the preset tolerance threshold, and adjust the actual roll gap to the predicted roll gap value according to the target control parameter before the bar enters the mill in the next rolling cycle, thereby completing the roll gap control.

9. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores instructions for causing the machine to perform: the roll gap control method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, It includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the roll gap control method according to any one of claims 1 to 7.

Images