High-speed continuous rolling cold pilger mill control system and method based on Internet of Things

Abstract

The invention belongs to the technical field of control, and discloses a high-speed continuous rolling cold pilger mill control system and method based on the Internet of Things. Comprising the following steps: collecting pipe data, high-speed continuous rolling cold pilger mill load data, working environment data and high-speed continuous rolling cold pilger mill traction data; aiming at the pipe data, optimizing a support vector regression model by using an LZ sea image optimization algorithm to predict pipe tension parameter data; comparing the predicted pipe tension parameter data with the standard pipe tension parameter data to obtain tension error data; calculating a compensation tension factor by using a multivariate compensation model, and compensating the tension error data to obtain compensation tension error data; based on the compensation tension error data, the traction force of the high-speed continuous rolling cold pilger mill is adjusted through the control quantity output by the PID control model, and the pipe tension parameters are controlled through the adjusted traction force so that the pipe tension parameters can be balanced; the prediction and control precision of the pipe tension is effectively improved, and the quality problem caused by unbalanced tension is reduced.

Description

Technical Field

[0001] The present invention relates to the field of control technology, and specifically to a control system and method for a high-speed continuous rolling cold rolling tube mill based on the Internet of Things. Background Art

[0002] The existing control systems for high-speed continuous rolling cold rolling tube mills have many problems, which affect the intelligence and accuracy of the systems. First of all, the existing control systems for high-speed continuous rolling cold rolling tube mills do not have the function of real-time data collection and analysis, and cannot comprehensively and meticulously monitor the entire production process.

[0003] Secondly, the existing control systems rely on real-time detection of the tube tension parameter data, and adjust the traction force through the control system to maintain the stability of the tension. However, usually this can only make reactive adjustments, cannot predict the tension parameter data, but only make adjustments according to the current tension parameter data.

[0004] In addition, for the tension error data, the influence of other surrounding factors on it is not considered, and the calculation of compensating the tension error data is not carried out, resulting in unbalanced tube tension parameters during rolling, thus affecting the tube quality.

[0005] In view of this, the present invention proposes a control system and method for a high-speed continuous rolling cold rolling tube mill based on the Internet of Things to solve the above problems. Summary of the Invention

[0006] In order to overcome the above defects of the prior art and to achieve the above object, the present invention provides the following technical solutions. A control system for a high-speed continuous rolling cold rolling tube mill based on the Internet of Things includes:

[0007] A data acquisition module: acquiring tube data, the load data of the high-speed continuous rolling cold rolling tube mill, the working environment data, and the traction force data of the high-speed continuous rolling cold rolling tube mill;

[0008] A tube tension prediction module: for the tube data, using the LZ walrus optimization algorithm to optimize the support vector regression model to predict the tube tension parameter data; wherein, for the LZ walrus optimization algorithm, a multi-stage cascaded Logistic mapping is used to generate the initial positions of the walrus individuals; a multi-switching point Huber loss function is used to evaluate the fitness values of each walrus individual, and a dynamic adjustment method is used to set the switching points in the multi-switching point Huber loss function;

[0009] A tension error compensation module: comparing the predicted tube tension parameter data with the standard tube tension parameter data to obtain the tension error data; using the load data of the high-speed continuous rolling cold rolling tube mill and the working environment data as the input of the multi-variable compensation model, and outputting the compensation tension factor; using the compensation tension factor to compensate the tension error data to obtain the compensated tension error data;

[0010] Automatic adjustment control module: Based on the compensated tension error data, the control quantity output by the PID control model is used to adjust the traction force of the high-speed continuous rolling cold rolling tube mill, and the adjusted traction force is used to control the tube tension parameters to balance them.

[0011] Furthermore, the tube data includes tube diameter data, tube inner diameter data, tube thickness data, rolling temperature data, cooling rate data, tube density data, and lubricant dosage data;

[0012] The load data of the high-speed continuous rolling cold rolling tube mill includes rolling force data, roll spacing data, and working duration data;

[0013] The working environment data includes air humidity data, air pressure data, and ambient temperature data.

[0014] Furthermore, the specific method of using the LZ walrus optimization algorithm to optimize the support vector regression model for predicting tube tension parameter data for the tube data includes:

[0015] Perform missing value, outlier, and normalization processing on the tube data to obtain a preprocessed tube data set;

[0016] Input the preprocessed tube data set, including M_n groups of samples, where each group of samples includes a group of preprocessed tube data and the corresponding tube tension parameter data;

[0017] Use the LZ walrus optimization algorithm to find the best hyperparameter combination of the SVR model, including the penalty coefficient C, epsilon parameter, and gamma parameter; use the random method to initialize the weight coefficient and bias term;

[0018] Calculate the initial loss function value of the support vector regression model Among them, represents regularization, w represents the initialized weight vector, b represents the initialized bias term, represents the error term, C is the penalty coefficient, M_n is the total number of samples in the sample set, q is the sample index, ξ represents the error of the qth group of samples on the upper deviation, ξ * represents the error of the qth group of samples on the lower deviation;

[0019] During the iteration process, use the gradient descent method to update the weight vector and bias term, and calculate the new loss function;

[0020] When the set number of iterations is reached, stop the iteration and output the predicted tension parameter data.

[0021] Furthermore, the specific method of using the LZ walrus optimization algorithm to find the best hyperparameter combination of the SVR model includes:

[0022] Step SS1: Initialize the positions of walrus individuals using a multi - cascaded Logistic map, i.e., the hyperparameter combination (C, epsilon, gamma); set the number of walrus individuals; set the maximum number of iterations; set the initial parameters CS 0 and P 0 , where CS 0 is the initialized exploration factor and P 0 is the initialized migration factor;

[0023] Step SS2: Evaluate the fitness value of each walrus individual position using a multi - switching - point Huber loss function;

[0024] Step SS3: According to the fitness values corresponding to each walrus individual position obtained, select the walrus individual position with the maximum fitness as the best individual position X leader , and guide other walrus individuals to search; where X leader =(C leader , epsilon leader , gamma leader ), C leader , epsilon leader and gamma leader represent the best penalty coefficient, epsilon parameter, and gamma parameter respectively;

[0025] Step SS4: Introduce an adaptive parameter adjustment strategy to adjust the exploration factor and migration factor,

[0026] The formula for the adaptive exploration factor is: where t is the number of iterations, T max is the maximum number of iterations, and CS(t) is the adaptive exploration factor at the t - th iteration;

[0027] The formula for the adaptive migration factor is: where P(t) is the adaptive migration factor at the t - th iteration;

[0028] Use the adaptive exploration factor and adaptive migration factor to update the positions of walrus individuals. When t < 0.4T max , enter the exploration stage for global search. The position update formula in the exploration stage is: X i，EP (t + 1)=X i (t)+CS(t)*(X leader -X i )(t)); X i，EP (t + 1) represents the position of the i - th walrus at the (t + 1) - th iteration in the exploration stage, X leader is the position of the walrus individual with the maximum fitness, X i(t) is the position of the i-th walrus at the t-th iteration;

[0029] When 0.4T max ≤ t < 0.7T max it enters the migration stage and conducts local search. The position update formula in the migration stage is: X i，MP (t + 1) = X i (t) + P(t) * (X leader - X i (t)) + Y * (X i,best - X i (t)); where X i，MP (t + 1) is the position of the i-th walrus at the (t + 1)-th iteration in the migration stage, X i，best is the historical optimal position of the i-th walrus, and γ is the control local search factor, with a range of (0, 1);

[0030] When t ≥ 0.7T max it enters the exploitation stage and conducts fine search. The position update formula in the exploitation stage is: X i，XP (t + 1) = X i (t) + P(t) * (X leader - X i (t)) * ∈; where X i，XP (t + 1) is the position of the i-th walrus at the (t + 1)-th iteration in the exploitation stage, and ∈ is the fine-tuning factor;

[0031] Step SS5: When the set number of iterations is reached, stop the iteration and output the optimal hyperparameter combination.

[0032] Furthermore, the specific method for initializing the walrus individual positions using multi-cascaded Logistic mapping includes:

[0033] Set MY multi-cascaded Logistic mappings. Among them, the formula for the first cascaded Logistic mapping is: k 1 = kz 1 * k 0 * (1 - k 0 ); where k 0 represents the mapping initial value, which is selected using the random method, k 1 represents the mapping value output by the first cascaded Logistic mapping, and kz 1 represents the control parameter of the first cascaded Logistic mapping;

[0034] The formula for the second cascaded Logistic mapping is: Based on the output of the first cascaded Logistic mapping, a non-linear transformation is performed; where k 2Represents the mapping value output by the first cascaded Logistic mapping; kz 2 Represents the control parameter of the first cascaded Logistic mapping; θ 2 Represents the non-linearity parameter of the second cascaded Logistic mapping;

[0035] The formula for the MY-th cascaded Logistic mapping is: Where, k MY And k MY-1 Represent the mapping values output by the MY-th and MY - 1-th cascaded Logistic mappings, kz MY Represents the control parameter of the MY-th cascaded Logistic mapping, θ MY Represents the non-linearity parameter of the MY-th cascaded Logistic mapping;

[0036] And initialize the walrus individual position with the k MY mapping value to obtain the initial hyperparameter combination (C, epsilon, gamma).

[0037] Furthermore, the specific method for evaluating the fitness value of each walrus individual position using the multi-switching-point Huber loss function includes:

[0038] The formula for evaluating the fitness value of each walrus individual position using the multi-switching-point Huber loss function is:

[0039] Where, Is the Huber loss function, δ 1 , δ 2 And δ 3 Are the dynamic switching point 1, dynamic switching point 2, and dynamic switching point 3 respectively; y m Is the pipe tension parameter data, Is the predicted pipe tension parameter data;

[0040] Use the dynamic adjustment method to set the switching points in the multi-switching-point Huber loss function. The specific method includes:

[0041] Calculate the error standard deviation σ initial between the tension parameter data of the M_n group of samples and the predicted tension parameter data of the M_n group of samples, and set the switching points as multiples of the error standard deviation to obtain the initial switching points. The formula is: δ 1，cs = σ initial ; δ 2，cs = 2 * σ initial ; δ 3，cs = 3 * σ initial Where, δ1，cs is the initial switching point 1, δ 2，cs is the initial switching point 2, δ 3，cs is the initial switching point 3;

[0042] Calculate the rate of change of the mean square error, and the formula is: where, RE t is the rate of change of the mean square error at the t-th round, MSE t is the mean square error at the t-th round, MSE t-1 is the mean square error in the (t - 1)-th round;

[0043] Use the exponentially weighted moving average method to smooth the rate of change of the mean square error, and the formula is: where, as is the smoothing factor, ranging from 0 to 1, controlling the degree of smoothing, RE t-1 is the rate of change of the mean square error in the (t - 1)-th round, is the smoothed rate of change of the mean square error at the t-th round;

[0044] Use the initial switching point and the smoothed rate of change of the mean square error to set the dynamic switching point, and the formula is: where, τ is the attenuation factor, ranging from [0.01, 0.1].

[0045] Furthermore, the specific method of comparing the predicted pipe tension parameter data with the standard pipe tension parameter data to obtain the tension error data includes:

[0046] Calculate the error between the predicted pipe tension parameter data and the standard pipe tension parameter data, and the formula is: WC raw (sj) = ZL pred (sj) - ZL std , where, WC raw (sj) represents the tension error data at the time point sj, ZL pred (sj) represents the predicted pipe tension parameter data at the time point sj, ZL std is the standard pipe tension parameter data, which is obtained by referring to the pipe design specification.

[0047] Furthermore, the specific method of using the high-speed continuous rolling cold rolling mill load data and the working environment data as the input of the multi-variable compensation model, outputting the compensation tension factor; and compensating the tension error data with the compensation tension factor to obtain the compensated tension error data includes:

[0048] Use the high-speed continuous rolling cold rolling mill load data and the working environment data as the input feature data;

[0049] Use the Pearson analysis method to determine whether the input feature data is linear data or non-linear data;

[0050] Input linear data or non - linear data into the multivariate compensation model. Calculate the compensation tension factor through the multivariate compensation model, and obtain the compensated tension error data using the compensation tension factor.

[0051] The formula for calculating the compensation tension factor through the multivariate compensation model is: Where, ΔWE com is the compensation tension factor, cc 0 is the intercept, u is the total number of linear data, r is the index of linear data, QZ r (sj) represents the weight of the r - th linear data at the sj time point, cc r represents the regression coefficient of the r - th linear data, TZ r (sj) is the data value of the r - th linear data at the sj time point, J is the total number of non - linear data, j is the index of non - linear data, TZ j (sj) is the data value of the j - th non - linear data at the sj time point;

[0052] Use the compensation tension factor to compensate the tension error data to obtain the compensated tension error data. The formula is: WC ten (sj) = WC raw (sj)*(1 + ΔWE com (sj)), where, WC ten is the compensated tension error data at the time point Sj.

[0053] Furthermore, based on the compensated tension error data, the specific method of using the control quantity output by the PID control model to adjust the traction force of the high - speed continuous rolling cold - rolling tube mill and using the adjusted traction force to control the tube tension parameters to make them balanced includes:

[0054] Input the compensated tension error data into the PID control model. The formula for outputting the control quantity is:

[0055]

[0056] Where, SC(sj) represents the output control quantity at the time point sj, Kp represents the proportional gain, Ki represents the integral gain, represents the integral of the compensated tension error from time 0 to the time point sj, Kd represents the derivative gain, represents the change rate of the compensated tension error at the time point sj;

[0057] Use the output control quantity to adjust the traction force of the high - speed continuous rolling cold - rolling tube mill. The formula is: QY tj (sj) = QY gj(sj) + SC(sj); where QY tj (sj) is the traction force of the high-speed continuous rolling cold rolling tube mill adjusted at time point sj, QY gj (sj) is the traction force of the high-speed continuous rolling cold rolling tube mill at time point sj, and the traction force of the adjusted high-speed continuous rolling cold rolling tube mill is used to control the tension parameter data of the tube;

[0058] The traction force of the high-speed continuous rolling cold rolling tube mill directly affects the tension of the tube. By adjusting the traction force of the high-speed continuous rolling cold rolling tube mill, precise control of the tension parameter data of the tube is achieved.

[0059] The control method of the high-speed continuous rolling cold rolling tube mill based on the Internet of Things is implemented based on the above-mentioned control system of the high-speed continuous rolling cold rolling tube mill based on the Internet of Things, and includes:

[0060] Step S1, collect tube data, high-speed continuous rolling cold rolling tube mill load data, working environment data, and high-speed continuous rolling cold rolling tube mill traction force data;

[0061] Step S2, for the tube data, use the LZ walrus optimization algorithm to optimize the support vector regression model to predict the tube tension parameter data;

[0062] Step S3, compare the predicted tube tension parameter data with the standard tube tension parameter data to obtain the tension error data; use the high-speed continuous rolling cold rolling tube mill load data and working environment data as the input of the multi-variable compensation model, and output the compensation tension factor; use the compensation tension factor to compensate the tension error data to obtain the compensated tension error data;

[0063] Step S4, based on the compensated tension error data, use the control quantity output by the PID control model to adjust the traction force of the high-speed continuous rolling cold rolling tube mill, and use the adjusted traction force to control the tube tension parameter to make it balanced.

[0064] The technical effects and advantages of the control system and method of the high-speed continuous rolling cold rolling tube mill based on the Internet of Things of the present invention:

[0065] By using the LZ walrus optimization algorithm to optimize the support vector regression model, the present invention improves the accuracy of tension prediction, enhances the optimization ability by optimizing the initial value of the multi-stage cascade Logistic mapping optimization algorithm, and makes the prediction model more robust to outliers and reduces the error impact by introducing the multi-switching point Huber loss function;

[0066] The error compensation module combines the load data and environmental data of the high-speed continuous rolling cold rolling tube mill, distinguishes linear and non-linear data through the Pearson analysis method, calculates the compensation tension factor using the multi-variable compensation model, and accurately obtains the compensated tension error data using the compensation tension factor to improve the system stability;

[0067] At the control level, the system adopts a PID control model to dynamically adjust the traction force according to the compensated tension error data, making the pipe tension tend to be balanced;

[0068] The whole system is highly intelligent, capable of real-time collecting and analyzing data, while improving the system's adaptability to complex working conditions. Precise tension control not only reduces the quality defects of pipes, but also reduces material waste and improves production efficiency;

[0069] Generally speaking, the system has significant advantages in terms of intelligence, stability, calculation efficiency and production quality, providing an efficient and reliable solution for the intelligent control of high-speed continuous rolling cold tube mills. Brief Description of the Drawings

[0070] Figure 1 It is a schematic diagram of the control system of the high-speed continuous rolling cold tube mill based on the Internet of Things according to the present invention;

[0071] Figure 2 It is a schematic diagram of the control method of the high-speed continuous rolling cold tube mill based on the Internet of Things according to the present invention. Detailed Embodiments

[0072] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0073] Embodiment 1

[0074] Please refer to Figure 1 As shown, the control system of the high-speed continuous rolling cold tube mill based on the Internet of Things in this embodiment includes:

[0075] Data acquisition module: acquire pipe data, the load data of the high-speed continuous rolling cold tube mill, the working environment data, and the traction force data of the high-speed continuous rolling cold tube mill;

[0076] Pipe tension prediction module: for the pipe data, use the LZ walrus optimization algorithm to optimize the support vector regression model to predict the pipe tension parameter data; among them, for the LZ walrus optimization algorithm, use a multi-cascade Logistic mapping to generate the initial position of the walrus individuals; use a multi-switching point Huber loss function to evaluate the fitness value of each walrus individual, and use a dynamic adjustment method to set the switching point in the multi-switching point Huber loss function;

[0077] Tension error compensation module: Compare the predicted pipe tension parameter data with the standard pipe tension parameter data to obtain tension error data; Use the high-speed continuous rolling cold rolling mill load data and working environment data as the input of the multi-variable compensation model, and output the compensation tension factor; Use the compensation tension factor to compensate the tension error data to obtain the compensated tension error data;

[0078] Automatic adjustment control module: Based on the compensated tension error data, use the control quantity output by the PID control model to adjust the traction force of the high-speed continuous rolling cold rolling mill, and use the adjusted traction force to control the pipe tension parameters to balance them;

[0079] Among them, the traction force data of the high-speed continuous rolling cold rolling mill is collected in real time through a traction force sensor.

[0080] Pipe data includes pipe diameter data, pipe inner diameter data, pipe thickness data, rolling temperature data, cooling rate data, pipe density data, and lubricant dosage data;

[0081] The high-speed continuous rolling cold rolling mill load data includes rolling force data, roll spacing data, and working duration data;

[0082] The working environment data includes air humidity data, air pressure data, and ambient temperature data;

[0083] Among them, the pipe diameter data and pipe inner diameter data are collected in real time using a laser rangefinder; the pipe thickness data is collected in real time through a displacement sensor; the rolling temperature data is collected in real time through a thermocouple; the cooling rate data is collected in real time through a flow sensor; the pipe density data is collected in real time through an ultrasonic density sensor; the lubricant dosage data is collected in real time through a flow meter;

[0084] The rolling force data is collected in real time through a strain gauge; the roll spacing data is collected in real time through a displacement sensor; the working duration data is collected in real time through a timer;

[0085] The air humidity data is collected in real time through a humidity sensor; the air pressure data is collected in real time through an air pressure sensor; the ambient temperature data is collected in real time through a temperature sensor.

[0086] Compared with the genetic algorithm and the particle swarm optimization algorithm, the walrus optimization algorithm has higher computational efficiency when optimizing the SVR model. The GA requires cumbersome crossover and mutation operations, while the PSO needs to calculate the individual and global optimal solutions of each particle, resulting in a heavy computational burden; The walrus optimization algorithm adopts an iterative update mechanism. By dynamically adjusting the individual positions, it does not require complex gene operations or optimal value calculations, reducing the computational overhead;

[0087] For pipe data, the specific method of using the LZ walrus optimization algorithm to optimize the support vector regression model to predict the pipe tension parameter data includes:

[0088] Perform missing value, outlier, and normalization processing on the pipe data to obtain a preprocessed pipe data set;

[0089] Input the preprocessed pipe data set, including M_n groups of samples. Each group of samples includes a group of preprocessed pipe data and the corresponding pipe tension parameter data. Among them, the pipe tension parameter data is obtained through a tension sensor;

[0090] Use the LZ walrus optimization algorithm to find the optimal hyperparameter combination of the SVR model, including the penalty coefficient C, epsilon parameter, and gamma parameter; use the random method to initialize the weight coefficient and bias term;

[0091] Calculate the initial loss function value of the support vector regression model Among them, represents regularization, w represents the initialized weight vector, b represents the initialized bias term, represents the error term, C is the penalty coefficient, M_n is the total number of samples in the sample set, q is the sample index, ξ represents the error of the qth group of samples on the upper deviation, and ξ * represents the error of the qth group of samples on the lower deviation;

[0092] During the iteration process, use the gradient descent method to update the weight vector and bias term, and calculate the new loss function;

[0093] When the set number of iterations is reached, stop the iteration and output the predicted tension parameter data;

[0094] Among them, for pipe data, the interpolation method is used for missing value and outlier processing, and the min-max normalization method is used for normalization processing;

[0095] The interpolation method predicts the value of unknown data points through known data points;

[0096] The min-max normalization method is a method of compressing data into a specified range, usually between 0 and 1.

[0097] The walrus optimization algorithm is a swarm intelligence optimization algorithm that simulates the group behavior of walrus populations. It is inspired by the group cooperation behavior of walruses in the natural environment. The walrus optimization algorithm is usually divided into three stages: exploration stage, migration stage, and exploitation stage;

[0098] In the walrus optimization algorithm, random initialization is used to set the positions of walrus individuals. Although the random initialization method can cover all regions of the search space, it consumes a large amount of time and has low search efficiency.

[0099] The specific method of using the LZ walrus optimization algorithm to find the optimal hyperparameter combination of the SVR model includes:

[0100] Step SS1: Initialize the positions of walrus individuals, i.e., the hyperparameter combination (C, epsilon, gamma), using a multi-cascade Logistic map; set the number of walrus individuals; set the maximum number of iterations; use the random method to set the initialization parameters CS 0 and P 0 , where CS 0 is the initialized exploration factor, and P 0 is the initialized migration factor;

[0101] Step SS2: Evaluate the fitness value of each walrus individual position using the multi-switching point Huber loss function;

[0102] Step SS3: According to the obtained fitness values corresponding to each walrus individual position, select the walrus individual position with the maximum fitness as the best individual position X leader , and guide other walrus individuals to search; where, X leader =(C leader , epsilon leader , gamma leader ), C leader , epsilon leader and gamma leader represent the best penalty coefficient, epsilon parameter, and gamma parameter, respectively;

[0103] Step SS4: Introduce an adaptive parameter adjustment strategy to adjust the exploration factor and migration factor to better balance the global search and local search capabilities;

[0104] The formula for the adaptive exploration factor is: where t is the number of iterations, T max is the maximum number of iterations, and CS(t) is the adaptive exploration factor at the t-th iteration;

[0105] The formula for the adaptive migration factor is: where P(t) is the adaptive migration factor at the t-th iteration;

[0106] Update the positions of walrus individuals using the adaptive exploration factor and adaptive migration factor. When t < 0.4T maxWhen it enters the exploration stage for global search, the position update formula in the exploration stage is: X i，EP (t + 1) = X i (t) + CS(t) * (X leader - X i (t)); X i,EP (t + 1) represents the position of the i-th walrus in the exploration stage at the (t + 1)-th iteration, X leader is the position of the walrus individual with the maximum fitness, X i (t) is the position of the i-th walrus at the t-th iteration;

[0107] When 0.4T max ≤ t < 0.7T max it enters the migration stage for local search. The position update formula in the migration stage is: X i,MP (t + 1) = X i (t) + P(t) * (X leader - X i (t)) + Y * (X i,best - X i (t)); where X i，MP (t + 1) is the position of the i-th walrus at the (t + 1)-th iteration in the migration stage, X i,best is the historical optimal position of the i-th walrus, and γ is the control local search factor with a range of (0, 1);

[0108] When t ≥ 0.7T max it enters the exploitation stage for fine search. The position update formula in the exploitation stage is: X i，XP (t + 1) = X i (t) + P(t) * (X leader - X i (t)) * ∈; where X i，XP (t + 1) is the position of the i-th walrus at the (t + 1)-th iteration in the exploitation stage, and ∈ is the fine-tuning factor with a range of (0, 0.1);

[0109] Step SS5: When the set number of iterations is reached, stop the iteration and output the optimal hyperparameter combination.

[0110] The specific way to initialize the walrus individual positions using multi-cascade Logistic mapping includes:

[0111] Set MY multi-cascade Logistic mappings. Among them, the formula for the first cascade Logistic mapping is: k 1 = kz 1 * k 0 * (1 - k 0 ); where k 0Represents the initial value of the mapping, selected using the random method, k 1 Represents the mapping value output by the first cascaded Logistic mapping, kz 1 Represents the control parameter of the first cascaded Logistic mapping, selected using the random method;

[0112] The formula for the second cascaded Logistic mapping is: Based on the output of the first cascaded Logistic mapping, a non-linear transformation is performed; where, k 2 Represents the mapping value output by the first cascaded Logistic mapping; kz 2 Represents the control parameter of the first cascaded Logistic mapping, selected using the random method; θ 2 Represents the non-linearity parameter of the second cascaded Logistic mapping, set by the empirical method;

[0113] The formula for the MY-th cascaded Logistic mapping is: Where, k MY and k MY-1 Represent the mapping values output by the MY-th and MY-1-th cascaded Logistic mappings, kz MY Represents the control parameter of the MY-th cascaded Logistic mapping, selected using the random method, θ MY Represents the non-linearity parameter of the MY-th cascaded Logistic mapping, set by the empirical method;

[0114] And initialize the walrus individual positions with the k MY mapping value to obtain the initial hyperparameter combination (C, epsilon, gamma);

[0115] For example, set 3 multi-cascaded Logistic mappings, set the number of walrus individuals to 5. First, use the random method to set the control parameter kz of the first cascade 1 = 4, the control parameter kz of the second cascade 2 = 3, the control parameter kz of the third cascade 3 = 2, and the non-linearity parameters θ 2 and θ 3 are each set to 2; the initial values of the mapping for the 5 walrus individuals are each set to 0.1, 0.3, 0.5, 0.7, and 0.9 using the random method; calculate using the formula of the cascaded Logistic mapping. The calculation steps for the position of the first walrus individual are as follows:

[0116] The mapping value output by the first cascaded Logistic mapping is: k 1 = 4 * 0.1 * (1 - 0.1) = 0.36; the mapping value output by the second cascaded Logistic mapping is: k2 = 3 * (0.36 * (1 - 0.36)) 2 = 0.1593; The mapping value output by the third - stage cascaded Logistic mapping is: k 3 = 2 * (0.1593 * (1 - 0.1593)) 2 = 0.0354; So the mapping value is 0.0354;

[0117] Use the mapping value to initialize the position of the first walrus individual. Referring to relevant literature, the mapping range of the penalty coefficient C is [0.1, 10], the mapping range of the epsilon parameter is [0.01, 0.1], and the mapping range of the gamma parameter is [0.1, 1]. C = 0.1+(0.0354 * (10 - 0.1)) = 0.45; epsilon = 0.01+(0.0354 * (0.1 - 0.01)) = 0.0132; gamma = 0.1+(0.0354 * (1 - 0.1)) = 0.1319. So the position of the first walrus individual is [0.45, 0.0132, 0.1319]. The calculation steps for the positions of other walrus individuals are the same as above;

[0118] The multi - stage cascaded Logistic mapping cascades multiple Logistic mappings, enabling the positions of walrus individuals to follow certain regularities during initialization, which is more structured than simple random initialization; and the multi - stage cascaded mapping introduces more complexity and diversity through non - linear transformations, making the initialized individual position distribution more extensive and uniform, and able to cover a wider search area.

[0119] The specific method of using the multi - switching - point Huber loss function to evaluate the fitness value of each walrus individual position includes:

[0120] The formula for using the multi - switching - point Huber loss function to evaluate the fitness value of each walrus individual position is:

[0121] where, is the Huber loss function, δ 1 、δ 2 and δ 3 are the dynamic switching point 1, dynamic switching point 2, and dynamic switching point 3 respectively; y m is the pipe tension parameter data, is the predicted pipe tension parameter data;

[0122] Use the dynamic adjustment method to set the switching points in the multi - switching - point Huber loss function. The specific method includes:

[0123] Calculate the standard deviation of the error σ between the tension parameter data of the \(M_n\) groups of samples and the predicted tension parameter data of the \(M_n\) groups of samples initial , and set the switching point as a multiple of the standard deviation of the error to obtain the initial switching point. The formula is: \(\delta\) 1，cs =\(\sigma\) initial ; \(\delta\) 2，cs = 2*\(\sigma\) initial ; \(\delta\) 3，cs = 3*\(\sigma\) initial , where \(\delta\) 1，cs is the initial switching point 1, \(\delta\) 2，cs is the initial switching point 2, \(\delta\) 3，cs is the initial switching point 3;

[0124] Calculate the rate of change of the mean squared error. The formula is: where \(RE\) t is the rate of change of the mean squared error at the \(t\)th round, \(MSE\) t is the mean squared error at the \(t\)th round, \(MSE\) t-1 is the mean squared error in the \((t - 1)\)th round;

[0125] Use the exponentially weighted moving average method to smooth the rate of change of the mean squared error. The formula is: where \(a_s\) is the smoothing factor, ranging from 0 to 1, controlling the degree of smoothing, \(RE\) t-1 is the rate of change of the mean squared error in the \((t - 1)\)th round, is the smoothed rate of change of the mean squared error at the \(t\)th round;

[0126] Use the initial switching point and the smoothed rate of change of the mean squared error to set the dynamic switching point. The formula is: where \(\tau\) is the attenuation factor, ranging from [0.01, 0.1];

[0127] The walrus optimization algorithm uses the mean squared error as the loss function to evaluate the fitness value of each walrus individual. This method is sensitive to noise in the data, easily leading to inaccurate fitness evaluation and thus affecting the optimization result;

[0128] The multi - cut - point Huber loss function combines the advantages of the mean squared error and the absolute error. For small errors, it performs squared punishment like the mean squared error; for medium errors, it performs non - linear punishment; and for large errors, it performs linear punishment. This characteristic makes it less sensitive to noisy data, thus reducing the impact of noisy data on fitness evaluation.

[0129] The specific ways to compare the predicted pipe tension parameter data with the standard pipe tension parameter data include:

[0130] Calculate the error between the predicted pipe tension parameter data and the standard pipe tension parameter data. The formula is: WC raw (sj) = ZL pred (sj) - ZL std , where WC raw (sj) represents the tension error data at time point sj, ZL pred (sj) represents the predicted pipe tension parameter data at time point sj, and ZL std is the standard pipe tension parameter data, which is obtained by referring to the pipe design specification.

[0131] Use the high-speed continuous rolling cold rolling mill load data and working environment data as the input of the multi-variable compensation model, and output the compensation tension factor; the specific method of using the compensation tension factor to compensate the tension error data includes:

[0132] Use the high-speed continuous rolling cold rolling mill load data and working environment data as the input feature data;

[0133] Use the Pearson analysis method to determine whether the input feature data is linear data or non-linear data;

[0134] The calculation formula of the Pearson analysis method is: where xx F represents the correlation coefficient of the Fth input feature data, A represents the total number of data points of the Fth input feature data, a data point refers to a specific data value, TZ F，a represents the ath data point of the Fth input feature data, represents the average value of the Fth input feature data, ZL pred，a represents the ath predicted tension parameter data, represents the average value of the predicted tension parameter data; obtained by referring to relevant literature, when |xx F | > 0.3, it indicates that there is a linear relationship between the data, and when 0 ≤ |Xx F | ≤ 0.3, it indicates that there is a non-linear relationship between the data;

[0135] Input the linear data or non-linear data into the multi-variable compensation model, calculate the compensation tension factor through the multi-variable compensation model, and obtain the compensated tension error data using the compensation tension factor;

[0136] The formula for calculating the compensation tension factor through the multi-variable compensation model is: where ΔWE com is the compensation tension factor, CC 0 is the intercept, which is determined by the random method, u is the total number of linear data, r is the index of the linear data, QZr (sj) represents the weight of the r-th linear data at the sj time point, cc r represents the regression coefficient of the r-th linear data, which is determined by the random method, TZ r (sj) is the data value of the r-th linear data at the sj time point, J is the total number of non-linear data, j is the index of the non-linear data, TZ j (sj) is the data value of the j-th non-linear data at the sj time point;

[0137] Compensate the tension error data using the compensation tension factor to obtain the compensated tension error data. The formula is: WC ten (sj) = WC raw (sj) * (1 + ΔWE com (sj)), where WC ten is the compensated tension error data at the sj time point;

[0138] Adopt the Pearson analysis method to dynamically judge the linear or non-linear characteristics of the input feature data, so as to select a suitable compensation method for different types of data. Use the multi-variable compensation model to comprehensively consider various factors, accurately calculate the compensation tension factor, and improve the accuracy and stability of tension control through the compensated tension error data.

[0139] Based on the compensated tension error data, use the control quantity output by the PID control model to adjust the traction force of the high-speed continuous rolling cold rolling tube mill, and use the adjusted traction force to control the tube tension parameters to make it balanced. The specific methods include:

[0140] Input the compensated tension error data into the PID control model. The formula for outputting the control quantity is:

[0141]

[0142] Among them, SC(sj) represents the output control quantity at the sj time point, Kp represents the proportional gain, Ki represents the integral gain, represents the integral of the compensated tension error from time 0 to the sj time point, Kd represents the differential gain, represents the change rate of the compensated tension error at the sj time point;

[0143] Use the output control quantity to adjust the traction force of the high-speed continuous rolling cold rolling tube mill. The formula is: QY tj (sj) = QY gj (sj) + SC(sj); where QY tj (sj) is the adjusted traction force of the high-speed continuous rolling cold rolling tube mill at the sj time point, QY gj(sj) is the traction force of the high-speed continuous rolling cold rolling tube mill at time point sj. The traction force of the adjusted high-speed continuous rolling cold rolling tube mill is used to control the tension parameter data of the tube.

[0144] The traction force of the high-speed continuous rolling cold rolling tube mill directly affects the tension of the tube. By adjusting the traction force of the high-speed continuous rolling cold rolling tube mill, precise control of the tension parameter data of the tube is achieved.

[0145] In this embodiment, the LZ walrus optimization algorithm is used to optimize the support vector regression model to improve the accuracy of tension prediction, and the initial value of the multi-stage cascaded Logistic mapping optimization algorithm is optimized to enhance the optimization ability. The introduction of the multi-switching point Huber loss function makes the prediction model more robust to outliers and reduces the error impact.

[0146] The error compensation module combines the load data and environmental data of the high-speed continuous rolling cold rolling tube mill, distinguishes linear and non-linear data through Pearson analysis, calculates the compensation tension factor using a multi-variable compensation model, and accurately obtains the compensation tension error data using the compensation tension factor to improve the system stability.

[0147] At the control level, the system adopts a PID control model to dynamically adjust the traction force according to the compensation tension error data, making the tube tension tend to balance.

[0148] The entire system is highly intelligent, capable of real-time data collection and analysis, while improving the system's adaptability to complex working conditions. Precise tension control not only reduces the quality defects of the tube, but also reduces material waste and improves production efficiency.

[0149] Generally speaking, the system has significant advantages in terms of intelligence, stability, calculation efficiency, and production quality, providing an efficient and reliable solution for the intelligent control of high-speed continuous rolling cold rolling tube mills.

[0150] Embodiment 2

[0151] Please refer to Figure 2 As shown, for the parts not described in detail in this embodiment, refer to the description content of Embodiment 1. A control method for a high-speed continuous rolling cold rolling tube mill based on the Internet of Things is provided, including:

[0152] Step S1: Collect tube data, high-speed continuous rolling cold rolling tube mill load data, working environment data, and high-speed continuous rolling cold rolling tube mill traction force data.

[0153] Step S2: For the tube data, use the LZ walrus optimization algorithm to optimize the support vector regression model to predict the tube tension parameter data.

[0154] Step S3: Compare the predicted pipe tension parameter data with the standard pipe tension parameter data to obtain the tension error data; use the high-speed continuous rolling cold rolling mill load data and the working environment data as the input of the multi-variable compensation model, and output the compensation tension factor; use the compensation tension factor to compensate the tension error data to obtain the compensated tension error data;

[0155] Step S4: Based on the compensated tension error data, use the control quantity output by the PID control model to adjust the traction force of the high-speed continuous rolling cold rolling mill, and use the adjusted traction force to control the pipe tension parameters to make them balanced.

[0156] Embodiment 3

[0157] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the operation mode of the above-provided high-speed continuous rolling cold rolling mill control system and method based on the Internet of Things.

[0158] Since the electronic device introduced in this embodiment is the electronic device adopted for implementing the high-speed continuous rolling cold rolling mill control system and method based on the Internet of Things in the embodiments of the present application, based on the high-speed continuous rolling cold rolling mill control system and method introduced in the embodiments of the present application, those skilled in the art can understand the specific implementation manners and various variations of the electronic device in this embodiment. Therefore, the specific implementation of how this electronic device implements the method in the embodiments of the present application will not be described in detail here. As long as those skilled in the art implement the electronic device adopted for the high-speed continuous rolling cold rolling mill control system and method based on the Internet of Things in the embodiments of the present application, it falls within the scope of protection of the present application.

[0159] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to obtain a formula closest to the actual situation. The preset parameters and threshold selections in the formulas are set by those skilled in the art according to the actual situation.

[0160] The above is only the preferred embodiment of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for those ordinary technical users in the technical field, without departing from the principle of the present invention, several improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. The high-speed continuous cold rolling mill control system based on the Internet of Things is characterized by: include: Data collection module: collects pipe data, high-speed continuous cold rolling mill load data, working environment data and high-speed continuous cold rolling mill traction data; Pipe tension prediction module: For pipe data, the support vector regression model is optimized using the LZ walrus optimization algorithm to predict pipe tension parameter data; for the LZ walrus optimization algorithm, the initial position of the walrus individual is generated using multi-cascade logistic mapping; the fitness value of each walrus individual is evaluated using the multi-switch point Huber loss function, and the dynamic adjustment method is used to set the switching point in the multi-switch point Huber loss function; in, is the Huber loss function, δ1, δ2 and δ3 are dynamic switching points 1, 2 and 3 respectively; y m is the pipe tension parameter data, is the predicted pipe tension parameter data; Tension error compensation module: Compare the predicted pipe tension parameter data with the standard pipe tension parameter data to obtain tension error data; use the load data and working environment data of the high-speed continuous cold rolling mill as the input of the multivariate compensation model, and output the compensation tension factor; use the compensation tension factor to compensate the tension error data to obtain the compensation tension error data; Automatic adjustment control module: Based on the compensation tension error data, the control quantity output by the PID control model is used to adjust the traction force of the high-speed continuous cold rolling tube mill, and the adjusted traction force is used to control the tube tension parameters to balance them.

2. The high-speed continuous cold rolling mill control system based on the Internet of Things according to claim 1 is characterized in that: The pipe data includes pipe diameter data, pipe inner diameter data, pipe thickness data, rolling temperature data, cooling speed data, pipe density data and lubricant dosage data; The load data of high-speed continuous cold rolling mill includes rolling force data, roll spacing data and working time data; Working environment data includes air humidity data, air pressure data, and ambient temperature data.

3. The high-speed continuous cold rolling mill control system based on the Internet of Things according to claim 2 is characterized in that: The specific method of using the LZ walrus optimization algorithm to optimize the support vector regression model to predict the pipe tension parameter data for the pipe data includes: The pipe data is processed for missing values, outliers and normalization to obtain the preprocessed pipe data set; Input the pre-processed pipe data set, including Mn groups of samples, each group of samples includes a group of pre-processed pipe data and corresponding pipe tension parameter data; Use the LZ walrus optimization algorithm to find the optimal hyperparameter combination of the SVR model, including the penalty coefficient C, epsilon parameter, and gamma parameter; use the random method to initialize the weight coefficient and bias term; Calculate the initial loss function value of the support vector regression model in, represents regularization, w represents the initialization weight vector, b represents the initialization bias term, represents the error term, C is the penalty coefficient, M_n is the total number of samples in the sample set, q is the sample index, ξ represents the error of the qth group of samples on the upper deviation, ξ * It represents the error of the qth group of samples on the lower deviation; During the iteration, the gradient descent method is used to update the weight vector and bias term, and calculate the new loss function; When the set number of iterations is reached, the iteration is stopped and the predicted tension parameter data is output.

4. The high-speed continuous cold rolling mill control system based on the Internet of Things according to claim 3 is characterized in that: The specific method of using the LZ walrus optimization algorithm to find the optimal hyperparameter combination of the SVR model includes: Step SS1, use multi-cascade Logistic mapping to initialize the individual positions of walruses; set the number of walruses; set the maximum number of iterations; set the initialization parameters CS0 and P0 of the walrus optimization algorithm, where CS0 is the initialization exploration factor and P0 is the initialization migration factor; Step SS2, using the multi-switch point Huber loss function to evaluate the fitness value of each walrus individual position; Step SS3: According to the obtained fitness value corresponding to each walrus individual position, the walrus individual position with the largest fitness is selected as the best individual position, and other walrus individuals are guided to search; Step SS4: Introduce an adaptive parameter adjustment strategy to adjust the exploration factor and migration factor. The formula of the adaptive exploration factor is: Where t is the number of iterations, T max is the maximum number of iterations, CS(t) is the adaptive exploration factor at the tth iteration; The formula for the adaptive migration factor is: Where P(t) is the adaptive migration factor at the tth iteration; Use adaptive exploration factor and adaptive migration factor to update the position of individual walruses when t<0.4T max When 0.4T max ≤t<0.7T max When t≥0.7T max Enter the development phase; Step SS5: When the set number of iterations is reached, stop the iteration and output the optimal hyperparameter combination.

5. The high-speed continuous cold rolling mill control system based on the Internet of Things according to claim 4 is characterized in that: The specific method of initializing the individual positions of walruses using multi-cascade logistic mapping includes: Set MY multi-cascade Logistic mappings, where the formula of the first cascade Logistic mapping is: k1=kz1*k0*(1-k0); where k0 represents the initial value of the mapping, which is selected by random method, k1 represents the mapping value of the first cascade Logistic mapping, and kz1 represents the control parameter of the first cascade Logistic mapping; The formula for the second cascaded logistic map is: A nonlinear transformation is performed based on the output of the first cascade Logistic mapping; wherein k2 represents the mapping value of the first cascade Logistic mapping; kz2 represents the control parameter of the first cascade Logistic mapping; θ2 represents the nonlinearity parameter of the second cascade Logistic mapping; The formula for the MYth cascade logistic map is: Among them, k MY and k MY-1 Indicates the mapping value of the MY-th cascade and the MY-1-th cascade Logistic mapping, kz MY represents the control parameter of the MY-th cascade logistic map, θ MY Represents the nonlinearity parameter of the MYth cascaded Logistic map; And k MY The mapping values ​​are used to initialize the individual positions of walruses and obtain the initial hyperparameter combination (C, epsilon, gamma).

6. The high-speed continuous cold rolling mill control system based on the Internet of Things according to claim 5 is characterized in that: The dynamic adjustment method is used to set the switching point in the multi-switching point Huber loss function, and the specific method includes: Calculate the error standard deviation σ between the tension parameter data of the Mn group samples and the tension parameter data predicted by the Mn group samples initial , and set the switching point to a multiple of the error standard deviation to obtain the initial switching point, the formula is: δ 1，cs =σ initial ; δ 2，cs =2*σ initial ; δ 3，cs =3*σ initial , where δ 1，cs is the initial switching point 1, δ 2，cs is the initial switching point 2, δ 3，cs is the initial switching point 3; Calculate the mean square error change rate, the formula is: Among them, RE t is the rate of change of mean square error in round t, MSE t is the mean square error of round t, MSE t-1 is the mean square error in round t-1; The exponentially weighted moving average method is used to smooth the mean square error change rate. The formula is: Among them, as is the smoothing factor, ranging from 0 to 1, RE t-1 is the rate of change of mean square error in round t-1, is the rate of change of the smoothed mean square error in round t; The dynamic switching point is set using the initial switching point and the smoothed mean square error change rate. The formula is: Where τ is the attenuation factor, ranging from [0.01, 0.1].

7. The high-speed continuous cold rolling mill control system based on the Internet of Things according to claim 6 is characterized in that: The specific method of comparing the predicted pipe tension parameter data with the standard pipe tension parameter data to obtain the tension error data includes: The error calculation between the predicted pipe tension parameter data and the standard pipe tension parameter data is performed using the formula: WC raw (sj)=ZL pred (sj)-ZL std , among which, WC raw (sj) represents the tension error data at time point sj, ZL pred (sj) represents the predicted pipe tension parameter data at time point sj, ZL std It is the standard pipe tension parameter data.

8. The high-speed continuous cold rolling mill control system based on the Internet of Things according to claim 7 is characterized in that: The high-speed continuous cold rolling mill load data and working environment data are used as inputs of the multivariate compensation model to output a compensation tension factor; The specific method of compensating the tension error data by using the compensation tension factor to obtain the compensation tension error data includes: Using the load data and working environment data of the high-speed continuous cold rolling mill as input feature data; Use Pearson analysis to determine whether the input feature data is linear data or nonlinear data; Inputting linear data or nonlinear data into the multivariate compensation model, calculating the compensation tension factor through the multivariate compensation model, and obtaining the compensation tension error data using the compensation tension factor; The formula for calculating the compensation tension factor through the multivariate compensation model is: Among them, ΔWE com is the compensation tension factor, cc0 is the intercept, u ​​is the total number of linear data, r is the index of linear data, QZ r (sj) represents the weight of the rth linear data at the sj time point, cc r Represents the regression coefficient of the rth linear data, TZ r (sj) is the data value of the rth linear data at time point sj, J is the total number of nonlinear data, j is the index of nonlinear data, TZ j (sj) is the data value of the jth nonlinear data at time point sj; The tension error data is compensated by using the compensation tension factor to obtain the compensation tension error data. The formula is: WC ten (sj)=WC raw (sj)*(1+ΔWE com (sj)), where WC ten is the compensation tension error data at time point sj.

9. The high-speed continuous cold rolling mill control system based on the Internet of Things according to claim 8 is characterized in that: The specific method of using the control quantity output by the PID control model to adjust the traction force of the high-speed continuous cold rolling mill based on the compensation tension error data, and using the adjusted traction force to control the tube tension parameter to balance it includes: Input the compensation tension error data into the PID control model, and the formula for the output control quantity is: Among them, SC(sj) represents the output control amount at time point sj, Kp represents the proportional gain, Ki represents the integral gain, represents the compensation tension error integral from time 0 to time point sj, Kd represents the differential gain, Indicates the rate of change of the compensation tension error at time point sj; The output control quantity is used to adjust the traction force of the high-speed continuous cold rolling mill. The formula is: QY tj (sj)=QY gj (sj)+SC(sj); where QY tj (sj) is the traction force of the high-speed continuous cold rolling mill adjusted at time point sj, QY gj (sj) is the traction force of the high-speed continuous cold rolling mill at the time point sj. The tension parameters of the pipe are controlled by adjusting the traction force of the high-speed continuous cold rolling mill.

10. A high-speed continuous rolling cold rolling mill control method based on the Internet of Things, which is based on the high-speed continuous rolling cold rolling mill control system based on the Internet of Things according to any one of claims 1 to 9, characterized in that: include: Step S1, collecting pipe data, high-speed continuous cold rolling mill load data, working environment data and high-speed continuous cold rolling mill traction data; Step S2: for the pipe data, using the LZ walrus optimization algorithm to optimize the support vector regression model to predict the pipe tension parameter data; Step S3, comparing the predicted pipe tension parameter data with the standard pipe tension parameter data to obtain tension error data; The load data and working environment data of the high-speed continuous cold rolling mill are used as the input of the multivariate compensation model, and the compensation tension factor is output; The tension error data is compensated by using the compensation tension factor to obtain the compensated tension error data; Step S4: Based on the compensation tension error data, the control amount output by the PID control model is used to adjust the traction force of the high-speed continuous cold rolling mill, and the adjusted traction force is used to control the tube tension parameter to balance it.

Images