Thick plate rolling force prediction method, system and equipment fusing mechanism and data

Abstract

The invention discloses a mechanism and data fused thick plate rolling force prediction method, system and equipment. The method comprises the steps that rolling force and deformation resistance mechanism models are selected according to characteristics in the hot rolling process; a deformation resistance mechanism model coefficient is corrected by using an improved particle swarm algorithm, a Sims (rolling force) mechanism model is substituted, an optimized Sims mechanism model calculation theoretical value is obtained, and a radial basis function neural network model is constructed based on industrial big data and deviation between the theoretical value and a measured value. A radial basis function neural network model for controlling rolling force deviation and an optimized Sims mechanism model are organically combined through an addition compensation method to obtain a rolling force integration model, the rolling force prediction value is calculated in real time through the integration model, the function expression relation of a traditional mathematical model can be embodied, the prediction precision of a neural network can be considered, and the prediction accuracy of the rolling force is improved. Online prediction of the rolling force is achieved, the high-precision rolling production requirement is met, and the finished product quality of a hot-rolled plate is improved.

Description

Technical Field

[0001] This application relates to the field of automatic control technology for rolling processes, and in particular to a method, system, and equipment for predicting rolling force of thick plates that integrates mechanisms and data. Background Technology

[0002] As one of the most important force parameters in the rolling process, the accuracy of rolling force prediction has a significant impact on the dimensional accuracy and product quality of strip and sheet. Therefore, constructing an accurate rolling force model has important practical engineering significance. Summary of the Invention

[0003] In view of this, this application provides a method, system and equipment for predicting rolling force of thick plates that integrates mechanism and data. The integrated model can reflect the functional expression relationship of traditional mathematical models and take into account the prediction accuracy of neural networks. It can realize online prediction of rolling force, meet the requirements of high-precision rolling production and improve the finished product quality of hot-rolled plates.

[0004] According to one aspect of this application, a method for predicting rolling force in thick plates by integrating mechanisms and data is provided, applied to a hot-rolled strip steel production system for performing flat-roll rolling of strip steel under hot rolling conditions. The method includes: Collect strip steel production data and normalize the strip steel production data to obtain a training dataset. The strip steel production data includes the inlet thickness, outlet thickness, reduction rate, rolling speed, rolling temperature, strip width, and measured rolling force. A rolling force mechanism model is established for the rolling force experienced by strip steel during flat roll rolling, and a deformation resistance mechanism model is established for the deformation resistance generated by strip steel to resist plastic deformation during flat roll rolling. After the least squares method is used to perform the initial regression correction on the deformation resistance mechanism model, the improved particle swarm optimization algorithm is used to perform the regression correction on the deformation resistance mechanism model after the initial regression correction. Based on the regression correction results of the improved particle swarm optimization algorithm under different inertia weight improvement strategies, the optimal deformation resistance mechanism model is determined. The improved particle swarm optimization algorithm is obtained by introducing multiple inertia weight improvement strategies into the original particle swarm optimization algorithm. Using inlet thickness, outlet thickness, reduction rate, rolling speed, rolling temperature, and strip width as input parameters, the deformation resistance is calculated based on the input parameters through the optimal deformation resistance mechanism model and substituted into the rolling force mechanism model. The rolling force mechanism model, combined with the input parameters and the calculated deformation resistance, calculates the rolling force on the strip under the ideal conditions corresponding to the input parameters, which is then used as the theoretical value of the rolling force. Using the input parameters as input and the rolling force deviation between the theoretical and measured rolling force values ​​as output, a radial basis function neural network model is constructed to learn the rolling force deviation law of strip steel under non-ideal conditions. A thick plate rolling force prediction model is obtained by fusing a radial basis function neural network model and a rolling force mechanism model using an additive strategy. This model integrates the mechanism and data. The model is then trained using a training dataset to predict the rolling force based on newly acquired input parameters. Specifically, when the trained model predicts the rolling force based on the newly acquired input parameters, the deformation resistance of the strip under the new input parameters is calculated using an optimal deformation resistance model and substituted into the rolling force mechanism model. The rolling force mechanism model then calculates the theoretical rolling force value based on the deformation resistance. The rolling force deviation is calculated using a radial basis function neural network model as compensation and fused into the theoretical rolling force value to obtain the final predicted rolling force value.

[0005] According to another aspect of this application, a thick plate rolling force prediction system integrating mechanism and data is provided, applied to a hot-rolled strip steel production system, the hot-rolled strip steel production system being used for flat-roll rolling of strip steel under hot rolling conditions, the system comprising: The data acquisition module is used to collect strip steel production data and normalize the strip steel production data to obtain a training dataset. The strip steel production data includes inlet thickness, outlet thickness, reduction rate, rolling speed, rolling temperature, strip width, and measured rolling force. The mechanism model building module is used to establish a rolling force mechanism model for the rolling force experienced by strip steel during the flat roll rolling process, and to establish a deformation resistance mechanism model for the deformation resistance generated by strip steel to resist plastic deformation during the flat roll rolling process. The mechanism model optimization module is used to perform an initial regression correction on the deformation resistance mechanism model using the least squares method in the particle swarm algorithm, and then to perform a second regression correction on the deformation resistance mechanism model after the initial regression correction using the improved particle swarm algorithm. Based on the regression correction results of the improved particle swarm algorithm under different inertia weight improvement strategies, the optimal deformation resistance mechanism model is determined. The improved particle swarm algorithm is obtained by introducing multiple inertia weight improvement strategies into the original particle swarm algorithm. The deviation compensation module is used to take the inlet thickness, outlet thickness, reduction rate, rolling speed, rolling temperature and strip width as input parameters, calculate the deformation resistance based on the input parameters through the optimal deformation resistance mechanism model, and substitute it into the rolling force mechanism model. The rolling force mechanism model combines the input parameters and the calculated deformation resistance to calculate the rolling force of the strip under the ideal conditions corresponding to the input parameters as the theoretical value of the rolling force. The deviation compensation module is also used to construct a radial basis function neural network model for learning the rolling force deviation law of strip steel under non-ideal conditions, taking input parameters as input and the rolling force deviation between the theoretical value and the measured value of rolling force as output. The rolling force prediction module is used to fuse the radial basis function neural network model and the rolling force mechanism model through an additive strategy to obtain a thick plate rolling force prediction model that integrates mechanism and data. The thick plate rolling force prediction model is trained with a training dataset so that the trained thick plate rolling force prediction model can predict the rolling force based on the newly acquired input parameters. Specifically, when the trained thick plate rolling force prediction model predicts the rolling force based on the newly acquired input parameters, the deformation resistance of the strip under the newly acquired input parameters is calculated by the optimal deformation resistance model and substituted into the rolling force mechanism model. The rolling force mechanism model calculates the theoretical value of the rolling force based on the deformation resistance. The rolling force deviation is calculated by the radial basis function neural network model as compensation and fused into the theoretical value of the rolling force to obtain the final rolling force prediction value.

[0006] According to another aspect of this application, an apparatus is provided, including a medium, a processor, and a computer program stored on the medium and executable on the processor, wherein the processor executes the program to implement the above-described method for predicting the rolling force of thick plates by fusing the mechanism and data.

[0007] By employing the above technical solutions, this application provides a method, system, and equipment for predicting rolling force in thick plates that integrates mechanisms and data. Based on the characteristics of the hot rolling process, a rolling force and deformation resistance mechanism model is selected. An improved particle swarm optimization algorithm is used to correct the coefficients of the deformation resistance mechanism model, which are then substituted into the Sims (rolling force) mechanism model to obtain the optimized Sims mechanism model for calculating theoretical values. Based on industrial big data and the deviation between theoretical and measured values, a radial basis function neural network model is constructed. Through an additive compensation method, the radial basis function neural network model controlling the rolling force deviation and the optimized Sims mechanism model are organically combined to obtain an integrated rolling force model. This integrated model calculates the predicted rolling force value in real time, reflecting both the functional expression of traditional mathematical models and the prediction accuracy of neural networks. This enables online prediction of rolling force, meeting the requirements of high-precision rolling production and improving the finished product quality of hot-rolled plates.

[0008] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a method for predicting the rolling force of thick plates that integrates mechanisms and data, as provided in an embodiment of this application, is shown. Figure 2 This paper illustrates a flowchart of an improved particle swarm optimization algorithm for regressing deformation resistance model according to an embodiment of this application. Figure 3 This illustration shows a BP neural network structure provided in an embodiment of this application; Figure 4 This paper shows a scatter plot of predicted and measured values ​​of a least squares regression deformation resistance model provided in an embodiment of this application. Figure 5 This paper illustrates the regression effect diagram of predicted and measured values ​​of a thick plate rolling force model based on fusion mechanism and data, provided in an embodiment of this application. Figure 6 This paper shows a partial relative error diagram between the predicted and measured values ​​of a thick plate rolling force model based on fusion mechanism and data, provided in an embodiment of this application. Figure 7 A schematic diagram of a thick plate rolling force prediction system that integrates mechanism and data is shown in an embodiment of this application. Detailed Implementation

[0010] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0011] This embodiment provides a method for predicting the rolling force of thick plates by integrating mechanisms and data, applied to a hot-rolled strip steel production system. The hot-rolled strip steel production system is used for flat-roll rolling of strip steel under hot rolling conditions. The method includes: Step 101: Collect strip steel production data and normalize the strip steel production data to obtain a training dataset. The strip steel production data includes inlet thickness, outlet thickness, reduction rate, rolling speed, rolling temperature, strip width, and measured rolling force. Step 102: Establish a rolling force mechanism model for the rolling force experienced by the strip during the flat roll rolling process, and establish a deformation resistance mechanism model for the deformation resistance generated by the strip during the flat roll rolling process to resist plastic deformation. Step 103: After the initial regression correction of the deformation resistance mechanism model using the least squares method, the improved particle swarm optimization algorithm is used to perform a second regression correction on the deformation resistance mechanism model after the initial regression correction. Based on the regression correction results of the improved particle swarm optimization algorithm under different inertia weight improvement strategies, the optimal deformation resistance mechanism model is determined. The improved particle swarm optimization algorithm is obtained by introducing multiple inertia weight improvement strategies into the original particle swarm optimization algorithm. Step 104: Using the inlet thickness, outlet thickness, reduction rate, rolling speed, rolling temperature, and strip width as input parameters, the deformation resistance is calculated based on the input parameters using the optimal deformation resistance mechanism model and substituted into the rolling force mechanism model. The rolling force mechanism model, combined with the input parameters and the calculated deformation resistance, calculates the rolling force of the strip under the ideal conditions corresponding to the input parameters as the theoretical value of the rolling force. Step 105: Using the input parameters as input and the rolling force deviation between the theoretical rolling force value and the measured rolling force value as output, construct a radial basis function neural network model for learning the rolling force deviation law of strip steel under non-ideal conditions; Step 106: The radial basis function neural network model and the rolling force mechanism model are fused using an additive strategy to obtain a thick plate rolling force prediction model that integrates mechanism and data. The thick plate rolling force prediction model is then trained using a training dataset so that the trained thick plate rolling force prediction model can predict the rolling force based on the newly acquired input parameters. Specifically, when the trained thick plate rolling force prediction model predicts the rolling force based on the newly acquired input parameters, the deformation resistance of the strip under the newly acquired input parameters is calculated using the optimal deformation resistance model and substituted into the rolling force mechanism model. The rolling force mechanism model calculates the theoretical value of the rolling force based on the deformation resistance. The rolling force deviation is calculated using the radial basis function neural network model as compensation and fused into the theoretical value of the rolling force to obtain the final predicted value of the rolling force.

[0012] Currently, the deformation resistance of materials is one of the most important factors affecting the accuracy of rolling force. Early scholars mainly focused on experimental research for deformation resistance, without studying deformation resistance models based on measured rolling force data of steel rolling equipment, and summarizing a set of deformation resistance models suitable for the application of this equipment.

[0013] Research on rolling force models mainly falls into two categories: one is theoretical models based on traditional rolling mechanisms. These theoretical models introduce many assumptions and are not suitable for the control process of hot strip rolling, which is characterized by multiple variables, strong coupling, nonlinearity, and time-varying properties. The other category is modeling methods based on artificial intelligence and big data. Although these methods have higher prediction accuracy, they cannot specifically elucidate the relationship between strip input and output during the rolling process.

[0014] Therefore, in the above embodiments of this application, a high-precision deformation resistance model is established using intelligent algorithms. Based on the theoretical model of traditional rolling mechanisms and artificial intelligence models, a method for predicting the rolling force of thick plates that integrates mechanisms and data is proposed. Specifically, as... Figure 1 As shown, the rolling force mechanism model is used to calculate the force required for rolling under ideal conditions based on physical laws (such as the principle of material deformation and roll stress analysis). The inputs to the rolling force mechanism model are the inlet / outlet thickness, reduction rate, rolling speed, temperature, and workpiece width. The output is the theoretical calculated value of the rolling force, that is, the theoretical value of the rolling force (assuming perfect conditions such as uniform strip material and no equipment wear).

[0015] The deformation resistance mechanism model is used to quantify the strip's ability to resist deformation (i.e., the "hardness" of the material). Specifically, key factors affecting deformation resistance include temperature (higher temperatures result in softer materials), rolling speed (faster speeds result in "harder" materials), and reduction rate (greater deformation leads to stronger resistance). The optimization process of the deformation resistance mechanism model involves first adjusting the model parameters using the least squares method based on measured data; then, a modified particle swarm optimization algorithm (dynamically adjusting the optimization strategy) is used for secondary correction to ensure the model closely matches the actual material properties.

[0016] Next, data-driven deviation compensation is performed, namely, constructing a radial basis function neural network model (RBF model) to capture errors in actual production that cannot be explained by theory. The input to the RBF model is the same parameters (thickness, temperature, etc.) as the rolling force model. The output is the deviation between the theoretical rolling force and the measured value (e.g., additional resistance caused by equipment aging or material fluctuations). By learning from historical data "when theoretical calculations are too high / too low," an error pattern database is constructed.

[0017] Finally, the fusion model predicts the rolling force, that is, the thick plate rolling force prediction model = optimized rolling force model + RBF model.

[0018] Specifically, the prediction steps for the rolling force in the model are as follows: 1. Calculate the theoretical value: The newly acquired real-time parameters are input into the deformation resistance model, and the current "softness and hardness" of the strip is output; the "softness and hardness" is substituted into the optimized rolling force model, and the theoretical value of the basic rolling force is output.

[0019] The same real-time parameters are input into the RBF model, and the output error compensation value is (e.g., "+50 tons" means that an additional 50 tons of force is actually required).

[0020] Final output: Predicted rolling force = Basic theoretical value + Error compensation value The rolling force prediction model for thick plates ensures that the predicted values ​​conform to the basic laws of the rolling process (such as the decrease in rolling force as temperature increases), avoiding results that violate physical principles. The RBF model is used to compensate for actual disturbances. Through error compensation values ​​learned from data, it covers implicit factors not considered by the theoretical model (such as sensor drift and slight wear of rolls), making the predicted values ​​closer to the real scenario. The theoretical value provides a reliable foundation, and the compensation value dynamically corrects deviations. The combination of the two retains the interpretability of the mechanism and has the high accuracy of data-driven approaches. The mechanistic model calculates the basic rolling force value that conforms to physical laws; the data model learns the actual deviations; and the results are simply added together to directly output a scientific and practical rolling force prediction value, providing core parameters for the precise control of hot-rolled strip steel production.

[0021] Specifically, the collected strip steel production data is shown in Table 1.

[0022] Table 1

[0023] Optionally, in step 103, various inertia weight improvement strategies are introduced into the original particle swarm optimization algorithm, including: Step 1031: Based on the original particle swarm optimization algorithm, a non-linearly decreasing convex function is introduced as the first inertia weight improvement strategy, and a sigmoid-like function is introduced as the second inertia weight improvement strategy to replace the inertia weight term in the velocity update rule of the original particle swarm optimization algorithm. The first inertia weight improvement strategy is as follows: , The second inertia weight improvement strategy is as follows: , t is the current iteration number, t max Let ω be the total number of iterations, and ω(t) be the inertia weight at the t-th iteration. start ω represents the initial inertia weight at the start of the iteration. end The final inertia weight at the end of the iteration. This is a coefficient used to correlate the current iteration number with the total number of iterations.

[0024] In the above embodiments of this application, the inertia weight of the original particle swarm optimization algorithm is usually set to 1 by default. The influence of the inertia weight is mainly reflected in the update of particle velocity, thus affecting the convergence ability of the algorithm. Since the original particle swarm optimization algorithm has poor convergence ability, the inertia weight has been improved to obtain an improved particle swarm optimization algorithm. Its original velocity update rule formula is as follows: , c1 and c2 are learning factors, which can be set to 2; r1 and r2 are random numbers between [0,1]. pi For the optimal position of particle i, p g The global optimal position is given by ω, where ω is the inertia weight and v is the position. i+1 v represents the update rate of the (i+1)th particle. i Let x represent the velocity of the i-th particle. i Let be the position of the i-th particle.

[0025] Furthermore, two methods were used to improve the inertia weight. The first is a non-linearly decreasing convex function inertia weight, and the second is a novel sigmoid-like inertia weight, which has the advantage of combining linear and non-linear inertia weights. Specifically: Inertia weight is a parameter in particle swarm optimization (PSO) that controls whether particles maintain their previous state of motion. A large weight allows particles to "run fast" (search for the optimal solution globally), while a small weight allows particles to "slow down" (perform a fine search in a local area). The inertia weight in the original PSO algorithm is often fixed or changes linearly, which can easily lead to "not finding a good region in the early stages" or "missing the optimal solution in the later stages."

[0026] For the first inertia weight improvement strategy, the non-linearly decreasing "convex function" weights are: , The formula's logic is that the weight decreases slowly in the early stages and quickly in the later stages; specifically, ω... start : Initial weight (e.g., 0.9, allowing the particles to "explore widely" first); ω end The final weight (e.g., 0.4, to let the particles "focus on details").

[0027] For example, in a total of 100 iterations, the initial inertia weight at the start of the iteration is 0.9 and the ending weight is 0.4: In the first 50 iterations (early stage): the weight remains above 0.7 (slow decrease), and the particles quickly "sweep" across a large area to find the possible optimal solution direction; In the next iteration (later stage): the weight drops rapidly from 0.7 to 0.4 (fast decrease), and the particles slow down to carefully search for the optimal point in the "candidate region".

[0028] For the second inertia weight improvement strategy, segmentation plus "Sigmoid curve" weighting is used: , The formula logic is to control the weights in stages, with rapid exploration in the early stage and slow convergence in the later stage. α=0.2, which represents the first 20% of iterations (for example, out of a total of 100 iterations, the first 20 iterations belong to the "early stage").

[0029] Specifically, in the first 20% of iterations (early stage): the weight is fixed at 0.9 (super-large weight), and the particles "run at full speed" to quickly cover the global area and find the potential optimal range; in the next 80% of iterations (late stage): a function similar to a "Sigmoid curve" is used to gradually and smoothly reduce the weight from 0.9 to around 0.4. For example: just entering the late stage (21st iteration): the weight is still close to 0.9, and the search range continues to expand; near the end (100th iteration): the weight drops to around 0.4, the particles "slow down", and are precisely adjusted in the found area.

[0030] Next, in the original particle swarm optimization algorithm, the particle velocity update rule is: The particle's next velocity = inertial weight × previous velocity + velocity toward its "optimal position" + velocity toward the "group optimal position".

[0031] The core improvement is replacing the "original fixed / linear weights" with formulas, making speed updates more intelligent: 1. For the first strategy: In each iteration, use a "quadratic function" to calculate the current weight, maintain a large weight in the early stage (global search), and decrease it rapidly in the later stage (local search); 2. For the second strategy: use fixed large weights for the first 20% of iterations (aggressive search), and then use the Sigmoid function to make the weights "gradually decrease" (refined search); The ultimate goal is to achieve a balance between "finding the right direction" (global search) and "finding the right position" (local search) in the particle swarm, and to more accurately optimize the parameters of the deformation resistance model (such as making the calculation of deformation resistance more consistent with actual production data).

[0032] These two strategies enable the particle swarm optimization algorithm to find the optimal parameters of the deformation resistance model more efficiently (such as the influence coefficients of temperature and speed on the "material hardness"). With more accurate deformation resistance, the "theoretical rolling force" calculated by substituting it into the rolling force mechanism model becomes even more accurate; coupled with the compensation for "actual deviations" by the RBF neural network, the final predicted rolling force value can both conform to physical laws and closely match actual production conditions.

[0033] Therefore, such as Figure 2 As shown, after "Calculate the fitness of each particle": based on the current iteration number t and the total iteration number t... maxThe algorithm calculates ω(t) using one of the two strategies mentioned above. During "updating particle velocity and position": the calculated ω(t) is substituted into the velocity update formula to dynamically adjust the particle velocity. Before "meeting the termination condition": the weights are continuously adjusted according to the iteration progress to ensure a dynamic balance between global and local searches. By dynamically adjusting the inertia weights, the algorithm can automatically switch search modes according to the iteration progress, avoiding getting trapped in local optima. Smaller weights in the later stages allow particles to finely adjust their positions, increasing the probability of finding the global optimum. This is suitable for complex optimization problems, such as optimizing deformation resistance model parameters in thick plate rolling force prediction. By introducing a nonlinear decreasing convex function or a piecewise sigmoid function to adjust the inertia weights, the particle swarm optimization algorithm can more intelligently balance global and local searches, improving convergence speed and accuracy, thereby more effectively optimizing model parameters and improving the accuracy of thick plate rolling force prediction.

[0034] Specifically, the constructed BP neural network structure diagram is as follows: Figure 3 As shown, Figure 3 In the figure, P is the predicted value of the output rolling force, V is the rolling speed, T is the rolling temperature, h0 is the inlet thickness, h1 is the outlet thickness, b is the width of the rolled piece, and ε is the reduction rate.

[0035] Optionally, in step 102, the rolling force mechanism model is the Sims model, and the rolling force mechanism model is as follows: , P is the theoretical value of the rolling force corresponding to the rolling force, B is the width of the rolled piece, and l c The horizontal projection length of the deformed area after flattening is calculated using the radius of the flattened roll and the reduction amount. Q is the friction influence coefficient, and K... m The parameters for determining deformation conditions are obtained through deformation resistance calculations.

[0036] In the above embodiments of this application, a rolling force mechanism model can be determined based on the characteristics of the flat roll rolling process. The rolling force model can be a Sims model. P is the theoretical value of the rolling force corresponding to the rolling force, kN; B is the width of the rolled piece, mm; l c The horizontal projection length of the deformed area after flattening is calculated using the radius of the flattened roll and the reduction amount, in mm; Q is the friction force influence coefficient; K m The K value for hot-rolled plates depends on the chemical composition of the metallic material and the physical conditions of deformation. m =1.15σ (σ is the deformation resistance, MPa).

[0037] Optionally, in step 102, the deformation resistance mechanism model is the Zhou Jihua-Guan Kezhi model, and the deformation resistance mechanism model is as follows: , σ0 represents the deformation resistance, σ0 represents the baseline deformation resistance, and u represents the strain rate, which is calculated using the rolling speed, inlet thickness, and outlet thickness. The degree of deformation is calculated by the reduction rate or based on the inlet and outlet thicknesses. T is the rolling temperature, and a1 to a6 are the first to sixth regression parameters, respectively.

[0038] In the above embodiments of this application, σ0 is the reference deformation resistance, MPa; u is the strain rate, which is calculated using the rolling speed, inlet thickness, and outlet thickness. The degree of deformation is calculated by the reduction rate or based on the inlet and outlet thicknesses; T is the rolling temperature, T = (273 + t) / 100, ℃; a1 to a6 are regression parameters that depend on the chemical materials.

[0039] Furthermore, the degree of deformation The true strain is the degree of logarithmic deformation. In rolling theory, the true strain reflects the essence of cumulative deformation, and its calculation formula is: , It is the input inlet thickness. It is the input output thickness.

[0040] If the input pressure rate is known : , It can be derived from the reduction rate : , therefore, Essentially, it is the quantification of "thickness change" in the input parameters, which directly corresponds to "reduction rate" or "inlet + outlet thickness".

[0041] Strain rate u is the rate of deformation of the workpiece per unit time during the rolling process. In flat roll rolling, the average thickness method is commonly used for calculation. , v is the input rolling speed. It is the input inlet thickness. It is the input output thickness.

[0042] The model regression parameters of particle swarm optimization algorithms with different inertia weights obtained by MATLAB software are shown in Table 2.

[0043] Table 2

[0044] By comparing the accuracy of different weighting improvement strategies, the deformation resistance model was finally determined as follows: .

[0045] Optionally, in step 104, the calculation of the rolling force on the strip under the ideal conditions corresponding to the input parameters, using the rolling force mechanism model combined with the input parameters and the calculated deformation resistance, as the theoretical value of the rolling force, includes: Step 1041: Use the back calculation formula to back calculate the deformation resistance and obtain the back-calculated deformation resistance; Step 1042: Using the rolling force mechanism model combined with the input parameters and the back-calculated deformation resistance, calculate the rolling force on the strip under the ideal conditions corresponding to the input parameters as the theoretical value of the rolling force. The back-calculation formula is as follows: , , , , , The calculated deformation resistance is given by P, where P is the theoretical value of the rolling force corresponding to the rolling force, b is the width of the rolled piece, and l is the resistance to deformation. c The horizontal projection length of the deformed area after flattening is given by the radius R of the flattened roll. d and reduction amount The calculations show that R is the initial rolling radius, Q is the friction influence coefficient, ε is the reduction rate, H is the pre-rolling height of the workpiece, and h is the post-rolling height of the workpiece. The average height of the workpiece before and after rolling.

[0046] In the above embodiments of this application, before performing regression correction on the deformation resistance model, it is first necessary to back-calculate the deformation resistance. The back-calculated deformation resistance is shown in Table 3: Table 3

[0047] Optionally, in step 103, the least squares method is used to perform an initial regression correction on the deformation resistance mechanism model, including: Step 1032: By adjusting the regression parameters of the deformation resistance mechanism model in the nonlinear least squares curve fitting function, the sum of squared residuals between the regression fitted value and the measured value is minimized. The nonlinear least squares curve fitting function is: , C represents the fitted deformation resistance model parameters, lsqcurvefit represents the nonlinear least squares curve fitting function, fun represents the deformation resistance mechanism model to be fitted, Pos represents the initial values ​​of the deformation resistance model parameters, x represents the set of values ​​for strain rate, deformation degree, and deformation temperature, and y represents the deformation resistance obtained by substituting and back-calculating.

[0048] In the above embodiments of this application, the least squares method is based on the lsqcurvefit function in MATLAB. The principle of the lsqcurvefit function is to solve the function for fitting nonlinear curves using the least squares method. The basic principle of least squares regression fitting parameters is that, after regressing the parameters, the sum of squared residuals between the function regression fitting value and the measured value is minimized.

[0049] The deformation resistance model obtained through regression using the lsqcurvefit function has coefficients a1 = -0.1203, a2 = 5.9426, a3 = 0.1914, a4 ​​= 0.1252, a5 = -0.3493, and a6 = 0.0032. Substituting these regression coefficients into the deformation resistance formula yields the deformation resistance model regressed using the traditional least squares method. The model is as follows: , The deformation resistance formula is regressed using the lsqcurvefit function based on the least squares method; a scatter plot of the predicted and measured deformation resistance values ​​is obtained, as shown below. Figure 4 As shown.

[0050] Optionally, in step 101, the strip steel production data is normalized to obtain a training dataset, including: Step 1011: For any type of strip steel production data, normalize the data using a normalization function. Based on the normalized strip steel production data, obtain a training dataset. The normalization function is: , Y represents the normalized strip steel production data, and x represents the original strip steel production data. min x is the minimum value in the original strip steel production data. max This is the maximum value in the original strip steel production data.

[0051] In the above embodiments of this application, the normalization process can be performed using the mapminmax function in Matlab software.

[0052] Optionally, in step 101, after normalizing the strip steel production data, the method further includes: Step 107: In the strip steel production data after normalization, a training dataset and a test dataset are divided according to a preset ratio. The training dataset is used to train the thick plate rolling force prediction model, and the test dataset is used to test whether the training of the thick plate rolling force prediction model has been completed.

[0053] In the above embodiments of this application, the division of the training set and the test set can be based on a 7:3 ratio.

[0054] By applying the technical solution of this embodiment, it is possible to achieve rolling force prediction that inherits both the structural form of the theoretical model and the accuracy of the big data model.

[0055] In one specific embodiment, a BP neural network and a radial basis function neural network are established respectively. The divided data are then substituted into the two models respectively, and the prediction accuracies of the two models are obtained as shown in the table below: Table 4

[0056] Furthermore, a radial basis function neural network model can be used to predict rolling force deviation. The deviation between the predicted and measured values ​​of the rolling force mechanism model is calculated using the optimized Sims mechanism model. Based on industrial big data, a radial basis function neural network model for rolling force deviation is constructed. An integrated rolling force model is obtained by organically combining the radial basis function neural network model for rolling force deviation and the optimized Sims mechanism model using an additive compensation method. This integrated model is used to calculate the predicted rolling force value in real time. Some predicted rolling force values ​​are shown in Table 5. Table 5

[0057] Plot the regression effects of the predicted rolling force values ​​for the integrated model test set, as follows: Figure 5 As shown; To more intuitively demonstrate the accuracy of each set of data, an error analysis was performed on the prediction data of the integrated model's test set, such as... Figure 6 As shown in Table 6, in order to further analyze the error and distribution of the test set data, an explanation is provided.

[0058] Table 6

[0059] Fit99 is the percentage of data with an error within 1% of the total sample, referred to as the 99% hit rate metric. Fit95 is the percentage of data with an error within 5% of the total sample, referred to as the 97% hit rate metric. Fit90 is the percentage of data with an error within 10% of the total sample, referred to as the 90% hit rate metric. In summary, the research results demonstrate the accuracy and superiority of the proposed deviation-compensated integrated rolling force prediction model, and this research provides a new approach to rolling force prediction.

[0060] By applying the technical solution of this embodiment, the functional expression relationship of the traditional mathematical model can be reflected, while the prediction accuracy of the neural network can be taken into account. It has the characteristics of complementary advantages, can realize online prediction of rolling force, meet the requirements of high-precision rolling production, and improve the finished product quality of hot-rolled plates.

[0061] Furthermore, as Figure 1 In terms of specific implementation, this application provides a thick plate rolling force prediction system that integrates mechanism and data, such as... Figure 7 As shown, this is applied to a hot-rolled strip steel production system, which is used to perform flat-roll rolling of strip steel under hot rolling conditions. The system includes: The data acquisition module 201 is used to collect strip steel production data and normalize the strip steel production data to obtain a training dataset. The strip steel production data includes inlet thickness, outlet thickness, reduction rate, rolling speed, rolling temperature, strip width, and measured rolling force. The mechanism model establishment module 202 is used to establish a rolling force mechanism model for the rolling force experienced by the strip during the flat roll rolling process, and to establish a deformation resistance mechanism model for the deformation resistance generated by the strip to resist plastic deformation during the flat roll rolling process. The mechanism model optimization module 203 is used to perform an initial regression correction on the deformation resistance mechanism model using the least squares method in the particle swarm algorithm, and then to perform a second regression correction on the deformation resistance mechanism model after the initial regression correction using the improved particle swarm algorithm. Based on the regression correction results of the improved particle swarm algorithm under different inertia weight improvement strategies, the optimal deformation resistance mechanism model is determined. The improved particle swarm algorithm is obtained by introducing multiple inertia weight improvement strategies into the original particle swarm algorithm. The deviation compensation construction module 204 is used to take the inlet thickness, outlet thickness, reduction rate, rolling speed, rolling temperature and strip width as input parameters, calculate the deformation resistance based on the input parameters through the optimal deformation resistance mechanism model, and substitute it into the rolling force mechanism model. The rolling force mechanism model combines the input parameters and the calculated deformation resistance to calculate the rolling force of the strip under the ideal conditions corresponding to the input parameters as the theoretical value of the rolling force. The deviation compensation construction module 204 is also used to construct a radial basis function neural network model for learning the rolling force deviation law of strip steel under non-ideal conditions, with input parameters as input and rolling force deviation between theoretical rolling force value and measured rolling force value as output. The rolling force prediction module 205 is used to fuse the radial basis function neural network model and the rolling force mechanism model through an additive strategy to obtain a thick plate rolling force prediction model that integrates mechanism and data. The thick plate rolling force prediction model is trained with a training dataset so that the trained thick plate rolling force prediction model can predict the rolling force based on the newly acquired input parameters. Specifically, when the trained thick plate rolling force prediction model predicts the rolling force based on the newly acquired input parameters, the deformation resistance of the strip under the newly acquired input parameters is calculated by the optimal deformation resistance model and substituted into the rolling force mechanism model. The rolling force mechanism model calculates the theoretical value of the rolling force based on the deformation resistance. The rolling force deviation is calculated by the radial basis function neural network model as compensation and fused into the theoretical value of the rolling force to obtain the final rolling force prediction value.

[0062] It should be noted that other corresponding descriptions of the functional units involved in the thick plate rolling force prediction system that integrates mechanism and data provided in this application embodiment can be found in the following references. Figure 1 The corresponding descriptions in the method will not be repeated here.

[0063] Based on the above, Figure 1 The method shown, and Figure 7 To achieve the above objectives, the virtual system embodiment shown in this application also provides a device, which may be a personal computer, server, network device, etc. This device includes a medium and a processor; the medium is used to store a computer program; the processor is used to execute the computer program to achieve the above-described objectives. Figure 1 The fusion mechanism and data are presented in the thick plate rolling force prediction method.

[0064] Optionally, the device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Bluetooth interfaces, Wi-Fi interfaces), etc.

[0065] Those skilled in the art will understand that the device structure provided in this embodiment does not constitute a limitation on the device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0066] The medium may also include an operating system and a network communication module. The operating system is a program that manages and stores the device's hardware and software resources, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the medium, as well as communication with other hardware and software within the physical device.

[0067] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms, or by hardware implementation. The rolling force and deformation resistance mechanism models are selected according to the characteristics of the hot rolling process. An improved particle swarm optimization algorithm is used to correct the coefficients of the deformation resistance mechanism model, which are then substituted into the Sims (rolling force) mechanism model to obtain the optimized Sims mechanism model to calculate theoretical values. Based on industrial big data and the deviation between theoretical and measured values, a radial basis function neural network model is constructed. By using an additive compensation method, the radial basis function neural network model of rolling force deviation and the optimized Sims mechanism model are organically combined to obtain an integrated rolling force model. This integrated model calculates the predicted rolling force value in real time, which can reflect the functional expression relationship of the traditional mathematical model while also taking into account the prediction accuracy of the neural network. This enables online prediction of rolling force, meets the requirements of high-precision rolling production, and improves the finished product quality of hot-rolled plates.

[0068] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the system of the embodiment scenario can be distributed throughout the system of the embodiment scenario as described, or they can be modified to reside in one or more systems different from this embodiment scenario. The modules of the above-described embodiment scenario can be combined into one module, or further divided into multiple sub-modules.

[0069] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any modifications that can be made by those skilled in the art should fall within the protection scope of this application.

Claims

1. A method for predicting the rolling force of thick plates by integrating mechanisms and data, characterized in that, An application to a hot-rolled strip steel production system, wherein the hot-rolled strip steel production system is used to perform flat-roll rolling of strip steel under hot rolling conditions, the method comprising: Collect strip steel production data and normalize the strip steel production data to obtain a training dataset. The strip steel production data includes the inlet thickness, outlet thickness, reduction rate, rolling speed, rolling temperature, strip width, and measured rolling force. A rolling force mechanism model is established for the rolling force experienced by strip steel during flat roll rolling, and a deformation resistance mechanism model is established for the deformation resistance generated by strip steel to resist plastic deformation during flat roll rolling. After the least squares method is used to perform the initial regression correction on the deformation resistance mechanism model, the improved particle swarm optimization algorithm is used to perform the regression correction on the deformation resistance mechanism model after the initial regression correction. Based on the regression correction results of the improved particle swarm optimization algorithm under different inertia weight improvement strategies, the optimal deformation resistance mechanism model is determined. The improved particle swarm optimization algorithm is obtained by introducing multiple inertia weight improvement strategies into the original particle swarm optimization algorithm. Using inlet thickness, outlet thickness, reduction rate, rolling speed, rolling temperature, and strip width as input parameters, the deformation resistance is calculated based on the input parameters through the optimal deformation resistance mechanism model and substituted into the rolling force mechanism model. The rolling force mechanism model, combined with the input parameters and the calculated deformation resistance, calculates the rolling force on the strip under the ideal conditions corresponding to the input parameters, which is then used as the theoretical value of the rolling force. Using the input parameters as input and the rolling force deviation between the theoretical and measured rolling force values ​​as output, a radial basis function neural network model is constructed to learn the rolling force deviation law of strip steel under non-ideal conditions. A thick plate rolling force prediction model is obtained by fusing a radial basis function neural network model and a rolling force mechanism model using an additive strategy. This model integrates the mechanism and data. The model is then trained using a training dataset to predict the rolling force based on newly acquired input parameters. Specifically, when the trained model predicts the rolling force based on the newly acquired input parameters, the deformation resistance of the strip under the new input parameters is calculated using an optimal deformation resistance model and substituted into the rolling force mechanism model. The rolling force mechanism model then calculates the theoretical rolling force value based on the deformation resistance. The rolling force deviation is calculated using a radial basis function neural network model as compensation and fused into the theoretical rolling force value to obtain the final predicted rolling force value.

2. The method according to claim 1, characterized in that, Several inertia weighting improvement strategies are introduced into the original particle swarm optimization algorithm, including: Based on the original particle swarm optimization (PSO) algorithm, a non-linearly decreasing convex function is introduced as the first inertia weight improvement strategy, and a sigmoid-like function is introduced as the second inertia weight improvement strategy. These strategies replace the inertia weight term in the velocity update rule of the original PSO algorithm. The first inertia weight improvement strategy is as follows: , The second inertia weight improvement strategy is as follows: ， t is the current iteration number, t max Let ω be the total number of iterations, and ω(t) be the inertia weight at the t-th iteration. start ω represents the initial inertia weight at the start of the iteration. end The final inertia weight at the end of the iteration. This is a coefficient used to correlate the current iteration number with the total number of iterations.

3. The method according to claim 1, characterized in that, The rolling force mechanism model is the Sims model, and the rolling force mechanism model is as follows: ， P is the theoretical value of the rolling force corresponding to the rolling force, B is the width of the rolled piece, and l c The horizontal projection length of the deformed area after flattening is calculated using the radius of the flattened roll and the reduction amount. Q is the friction influence coefficient, and K... m The parameters for determining deformation conditions are obtained through deformation resistance calculations.

4. The method according to claim 1, characterized in that, The deformation resistance mechanism model is the Zhou Jihua-Guan Kezhi model, and the deformation resistance mechanism model is as follows: ， σ0 represents the deformation resistance, σ0 represents the baseline deformation resistance, and u represents the strain rate, which is calculated using the rolling speed, inlet thickness, and outlet thickness. The degree of deformation is calculated by the reduction rate or based on the inlet and outlet thicknesses. T is the rolling temperature, and a1 to a6 are the first to sixth regression parameters, respectively.

5. The method according to claim 1, characterized in that, The rolling force is calculated as a theoretical value by combining the rolling force mechanism model with input parameters and the calculated deformation resistance, and then using this model to calculate the rolling force on the strip under the ideal conditions corresponding to the input parameters. The deformation resistance is calculated by using the inverse calculation formula to obtain the inverse deformation resistance; The rolling force is calculated as the theoretical value of the rolling force under ideal conditions corresponding to the input parameters by combining the rolling force mechanism model with the input parameters and the back-calculated deformation resistance. The back-calculation formula is as follows: ， ， ， ， ， The calculated deformation resistance is given by P, where P is the theoretical value of the rolling force corresponding to the rolling force, b is the width of the rolled piece, and l is the resistance to deformation. c The horizontal projection length of the deformed area after flattening is given by the radius R of the flattened roll. d and reduction amount The calculations show that R is the initial rolling radius, Q is the friction influence coefficient, ε is the reduction rate, H is the pre-rolling height of the workpiece, and h is the post-rolling height of the workpiece. The average height of the workpiece before and after rolling.

6. The method according to claim 5, characterized in that, The least squares method was used to perform an initial regression correction on the deformation resistance mechanism model, including: By adjusting the regression parameters of the deformation resistance mechanism model in the nonlinear least squares curve fitting function, the sum of squared residuals between the regression fitted value and the measured value is minimized. The nonlinear least squares curve fitting function is as follows: ， C represents the fitted deformation resistance model parameters, lsqcurvefit represents the nonlinear least squares curve fitting function, fun represents the deformation resistance mechanism model to be fitted, Pos represents the initial values ​​of the deformation resistance model parameters, x represents the set of values ​​for strain rate, deformation degree, and deformation temperature, and y represents the deformation resistance obtained by substituting and back-calculating.

7. The method according to claim 1, characterized in that, The normalization process of the strip steel production data yields a training dataset, including: For any type of strip steel production data, a normalization function is used to normalize the data. Based on the normalized strip steel production data, a training dataset is obtained, where the normalization function is: ， Y represents the normalized strip steel production data, and x represents the original strip steel production data. min x is the minimum value in the original strip steel production data. max This is the maximum value in the original strip steel production data.

8. The method according to any one of claims 1 to 7, characterized in that, After normalizing the strip steel production data, the method further includes: In the normalized strip steel production data, a training dataset and a test dataset are divided according to a preset ratio. The training dataset is used to train the thick plate rolling force prediction model, and the test dataset is used to test whether the training of the thick plate rolling force prediction model has been completed.

9. A thick plate rolling force prediction system integrating mechanism and data, characterized in that, An application in a hot-rolled strip steel production system, the hot-rolled strip steel production system being used for flat-roll rolling of strip steel under hot rolling conditions, the system comprising: The data acquisition module is used to collect strip steel production data and normalize the strip steel production data to obtain a training dataset. The strip steel production data includes inlet thickness, outlet thickness, reduction rate, rolling speed, rolling temperature, strip width, and measured rolling force. The mechanism model building module is used to establish a rolling force mechanism model for the rolling force experienced by strip steel during the flat roll rolling process, and to establish a deformation resistance mechanism model for the deformation resistance generated by strip steel to resist plastic deformation during the flat roll rolling process. The mechanism model optimization module is used to perform an initial regression correction on the deformation resistance mechanism model using the least squares method in the particle swarm algorithm, and then to perform a second regression correction on the deformation resistance mechanism model after the initial regression correction using the improved particle swarm algorithm. Based on the regression correction results of the improved particle swarm algorithm under different inertia weight improvement strategies, the optimal deformation resistance mechanism model is determined. The improved particle swarm algorithm is obtained by introducing multiple inertia weight improvement strategies into the original particle swarm algorithm. The deviation compensation module is used to take the inlet thickness, outlet thickness, reduction rate, rolling speed, rolling temperature and strip width as input parameters, calculate the deformation resistance based on the input parameters through the optimal deformation resistance mechanism model, and substitute it into the rolling force mechanism model. The rolling force mechanism model combines the input parameters and the calculated deformation resistance to calculate the rolling force of the strip under the ideal conditions corresponding to the input parameters as the theoretical value of the rolling force. The deviation compensation module is also used to construct a radial basis function neural network model for learning the rolling force deviation law of strip steel under non-ideal conditions, taking input parameters as input and the rolling force deviation between the theoretical value and the measured value of rolling force as output. The rolling force prediction module is used to fuse the radial basis function neural network model and the rolling force mechanism model through an additive strategy to obtain a thick plate rolling force prediction model that integrates mechanism and data. The thick plate rolling force prediction model is trained with a training dataset so that the trained thick plate rolling force prediction model can predict the rolling force based on the newly acquired input parameters. Specifically, when the trained thick plate rolling force prediction model predicts the rolling force based on the newly acquired input parameters, the deformation resistance of the strip under the newly acquired input parameters is calculated by the optimal deformation resistance model and substituted into the rolling force mechanism model. The rolling force mechanism model calculates the theoretical value of the rolling force based on the deformation resistance. The rolling force deviation is calculated by the radial basis function neural network model as compensation and fused into the theoretical value of the rolling force to obtain the final rolling force prediction value.

10. An apparatus comprising a medium, a processor, and a computer program stored on the medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method for predicting the rolling force of thick plates by fusing the mechanism and data as described in any one of claims 1 to 7.

Images