A method for cross-process shape feedforward control of strip steel applicable to continuous annealing and tempering units

Abstract

The present invention relates to a method for cross-process shape feedforward control of strip steel applicable to a continuous annealing and tempering unit, comprising: performing pattern recognition on the shape of the strip steel at the outlet of the cold tandem rolling mill; based on the recognized pattern, using a defect prediction model to predict the waviness defect data of the strip steel shape at the outlet of the cold tandem rolling mill; based on the waviness defect data, using a quality prediction model for the inheritance and evolution of the strip steel shape quality before and after continuous annealing to predict the actual shape quality of the strip steel after continuous annealing; converting the actual shape quality into the IU value distribution of the strip steel at the inlet of the tempering mill, and based on the IU value distribution of the strip steel at the inlet of the tempering mill, performing bending roll force feedforward control of the strip steel shape. The present invention can solve the problem of poor control accuracy of the shape control in the existing continuous annealing and tempering unit, thereby significantly improving the shape quality of the strip steel product.

Description

Technical Field

[0001] The present invention relates to the technical field of strip steel rolling production, and particularly to a cross-process method for shape feedforward control of strip steel in a continuous annealing and tempering unit. Background Art

[0002] The strip ratio is a key indicator to measure the technical level of a country's steel industry, and the strip shape is the core technical indicator of strip steel products. With the continuous emergence of high-end cold-rolled strip products, the quality index requirements of users for products are becoming increasingly stringent. At the same time, the continuous annealing and tempering unit is the last process of most cold-rolled strip products, which is crucial and important. Therefore, it is becoming more and more important to ensure the strip shape quality of the strip steel in the continuous annealing and tempering unit. At present, most of the tempering mills configured in continuous annealing and tempering units are 6-high tempering mills or 4-high tempering mills, and most of them are not equipped with detection devices such as shape meters, resulting in the inability of the tempering mills in the continuous annealing and tempering unit to achieve closed-loop feedback control of strip shape quality. Only presetting or feedforward control can be relied on to achieve strip steel shape control. However, neither the presetting nor the feedforward in the tempering mill of the continuous annealing and tempering unit has considered the inheritance of the strip shape of the incoming material strip steel. Therefore, it is impossible to perform real-time online feedforward adjustment according to the strip shape quality of the cold-rolled and hardened strip steel after annealing, which restricts the further improvement of strip shape quality.

[0003] At present, the production process flow of most domestic cold-rolled strip products includes pickling, cold rolling, annealing, and tempering. Tempering, as the last process before the strip steel becomes a finished product, is particularly important for the strip shape quality of the finished strip steel. However, most of the tempering mills in domestic continuous annealing and tempering units are not equipped with detection devices such as shape meters, resulting in the inability of the tempering mills in the continuous annealing and tempering unit to achieve closed-loop feedback control of strip shape quality. Only presetting or feedforward control can be relied on to achieve strip steel shape control. In order to better ensure the strip shape quality of the strip steel in the tempering unit, a cross-process method for shape feedforward control of strip steel in a continuous annealing and tempering unit is urgently needed. Summary of the Invention

[0004] The purpose of the present invention is to propose a cross-process method for shape feedforward control of strip steel in a continuous annealing and tempering unit to solve the problems existing in the above-mentioned prior art. Through numerical simulation modeling, accumulation of production big data, and by means of advanced machine learning and other methods, a prediction model of strip steel shape inheritance and evolution from the acid rolling unit to the inlet of the tempering mill is established. On this basis, shape feedforward control of the tempering mill is carried out to solve the problem of poor control accuracy of strip shape control in the existing continuous annealing and tempering unit, thereby significantly improving the strip shape quality of strip steel products.

[0005] To achieve the above purpose, the present invention provides the following solution:

[0006] A cross-process method for shape feedforward control of strip steel in a continuous annealing and tempering unit includes:

[0007] Perform pattern recognition on the strip shape of the cold tandem rolling mill exit strip;

[0008] Based on the recognized strip shape pattern of the cold tandem rolling mill exit strip, use a preset defect prediction model to predict the waviness defect data of the strip shape of the cold tandem rolling mill exit strip; wherein, the defect prediction model is constructed based on a long short-term memory neural network;

[0009] Based on the waviness defect data, use a preset quality prediction model for the inheritance and evolution of the strip shape quality before and after continuous annealing to predict the physical strip shape quality of the strip after continuous annealing; wherein, the quality prediction model is constructed based on a Stacking ensemble learning model;

[0010] Convert the physical strip shape quality into the IU value distribution of the strip at the entry of the temper mill, and based on the IU value distribution of the strip at the entry of the temper mill, perform feedforward control of the bending roll force of the strip shape.

[0011] Optionally, performing pattern recognition on the strip shape of the cold tandem rolling mill exit strip includes:

[0012] Obtain the IU value distribution of the strip shape at the exit of the cold tandem rolling mill: Divide the width and thickness of the strip according to a preset range, and for each steel grade, statistically calculate the average IU value of the strip shape according to the division of width and thickness;

[0013] Based on the IU value distribution of the strip shape, use Legendre orthogonal polynomials to identify the strip shape pattern of the cold tandem rolling mill exit strip.

[0014] Optionally, using Legendre orthogonal polynomials to identify the strip shape pattern of the cold tandem rolling mill exit strip includes:

[0015] Perform mean removal processing on the IU value distribution of the strip shape;

[0016] Based on the IU value distribution of the strip shape after mean removal processing, calculate the strip shape deviation and calculate the maximum value of the strip shape deviation;

[0017] Normalize the strip shape deviation to obtain a normalized sample;

[0018] Extract the basic strip shape pattern coefficients;

[0019] Compare the magnitudes and positive / negative signs of the basic strip shape pattern coefficients;

[0020] Determine the largest basic strip shape pattern coefficient, and the strip shape defect corresponding to the largest pattern coefficient is the strip shape defect of this strip.

[0021] Optionally, using a preset defect prediction model to predict the waviness defect data of the strip shape of the cold tandem rolling mill exit strip includes:

[0022] Based on the strip shape mode at the outlet of the tandem cold rolling mill recognized, carry out finite element simulation modeling of the pre-buckling and post-buckling of cold-rolled strip;

[0023] Based on the finite element simulation of the pre-buckling and post-buckling of cold-rolled strip, convert the IU value distribution of various typical strip shape defects into the form of physical strip shape defects, and obtain the wave height, wave spacing and wave width of the physical strip shape defects corresponding to the IU value distribution of each strip shape;

[0024] Based on various typical wave shape defects obtained from finite element simulation calculations, the initial IU value of the strip, the strip thickness and the strip width corresponding to the wave height, wave spacing and wave width, construct a buckling database;

[0025] Use the buckling database to train the long short-term memory neural network to obtain the defect prediction model;

[0026] Input the width, thickness and IU value of the strip into the defect prediction model to predict the wave shape defect data of the strip shape at the outlet of the tandem cold rolling mill; wherein, the wave shape defect data includes: wave height, wave spacing and wave width.

[0027] Optionally, carrying out finite element simulation modeling of the pre-buckling and post-buckling of cold-rolled strip includes:

[0028] Construct a buckling model;

[0029] Based on the buckling model, carry out pre-buckling behavior analysis of the wave shape to obtain the first-order buckling modes of center wave, edge wave, quarter wave, and edge-center composite wave;

[0030] Based on the buckling model, carry out post-buckling behavior analysis of the wave shape to obtain the post-buckling path of the strip.

[0031] Optionally, constructing the buckling model includes:

[0032] Set boundary constraint conditions;

[0033] Set the temperature field function and distribution form;

[0034] The boundary constraint conditions are:

[0035]

[0036] Among them, u (L,y) is the x-direction displacement of the strip at x = L, u (-L,y) is the y-direction displacement of the strip at x = -L, v (L,y) is the y-direction displacement of the strip at x = L, v (-L,y) is the y-direction displacement of the strip at x = L, w (L,y) is the z-direction displacement of the strip at x = L, w (-L,y) is the z-direction displacement of the strip at x = L, u(0,0) is the displacement in the x - direction at the center point of the strip steel, v (0,0) is the displacement in the y - direction at the center point of the strip steel; x and y refer to the coordinates of the strip steel;

[0037] The temperature field function and its distribution form are as follows:

[0038] Center wave

[0039] Edge wave

[0040] Quarter wave

[0041] Edge - center composite wave

[0042] where Y is the different widths of the strip steel and T is the temperature field function.

[0043] Optionally, predicting the physical shape quality of the strip steel after continuous annealing includes:

[0044] Collecting raw data; where the raw data includes: cold - rolling exit waviness data, online platform data, and manual observation data;

[0045] Cleaning and normalizing the raw data;

[0046] Extracting features from the processed raw data;

[0047] Using the extracted features to train a Stacking ensemble learning model to obtain a quality prediction model for the inheritance and evolution of the strip steel shape quality before and after continuous annealing.

[0048] Optionally, using the extracted features to train a Stacking ensemble learning model includes:

[0049] In the first stage, using the extracted features to train several individual machine learning models to obtain several prediction results; where the several individual machine learning models include: multi - layer perceptron, support vector machine, and random forest algorithm;

[0050] In the second stage, using the several prediction results to train a multiple linear regression model;

[0051] Based on the trained several individual machine learning models and the trained multiple linear regression model, a quality prediction model for the inheritance and evolution of the strip steel shape quality before and after continuous annealing is constructed.

[0052] Optionally, the cold - rolling exit waviness data includes: the IU value of the strip steel shape;

[0053] The online platform data includes: various strip steel specifications, average temperatures at various parts of the annealing furnace, and tension differences between the inlet and outlet;

[0054] The manually observed data includes: the strip steel shape grades at the inlet of each annealing furnace and the strip steel shape grades at the outlet of the annealing furnace.

[0055] The beneficial effects of the present invention are as follows:

[0056] The present invention performs pattern recognition on the strip steel shape at the outlet of the cold tandem rolling mill; it can visualize the distribution of the measured strip steel shape IU value by the shape meter as various physical strip steel shape defects and classify and summarize them; based on the recognized strip steel shape pattern at the outlet of the cold tandem rolling mill, the present invention uses a preset defect prediction model to predict the waviness defect data of the strip steel shape at the outlet of the cold tandem rolling mill, obtains the wave height, wave distance, and wave width of the physical strip steel shape defects corresponding to the distribution of each strip steel shape IU value, and forms a simulation-based large database based on this; based on the large database, the present invention uses a preset quality prediction model for the inheritance and evolution of the strip steel shape quality before and after continuous annealing to predict the physical strip steel shape quality of the strip steel after continuous annealing; based on the predicted physical strip steel shape defects of the strip steel after annealing, the present invention converts them into the distribution of the strip steel shape IU value, and based on the shape control effect of the temper mill, realizes the feedforward control of the bending roll force of the strip steel shape, that is, adjusts the bending roll force of the temper mill in real time according to the strip steel shape defects of the incoming material. Finally, the control and improvement of the strip steel shape quality of the temper mill are realized. This comprehensive control idea effectively combines data analysis, prediction models, and real-time control technologies, providing a scientific and intelligent solution for the strip steel shape control of the continuous annealing and tempering mill. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0058] Figure 1 It is a statistical schematic diagram of the strip steel at the outlet of the cold tandem rolling mill of the M3 mild steel series in the embodiment of the present invention; among them, (a) is M3A45 IU_50, and (b) is M3A45 IU_ALL;

[0059] Figure 2 It is a statistical schematic diagram of the strip steel at the outlet of the cold tandem rolling mill of the household appliance plate SDX51D in the embodiment of the present invention; among them, (a) is SDX51D IU_50, and (b) is SDX51D IU_ALL;

[0060] Figure 3Statistical schematic diagram of the strip steel at the outlet of the CSA01 cold tandem rolling mill in the embodiment of the present invention; among them, (a) is CSA01 IU_50, and (b) is CSA01 IU_ALL;

[0061] Figure 4 Schematic diagram of eight standard mode curves in the embodiment of the present invention; among them, (a) is the primary component, (b) is the secondary component, (c) is the tertiary component, and (d) is the quaternary component;

[0062] Figure 5 Legendre decomposition pattern recognition of a certain cross-section in the embodiment of the present invention;

[0063] Figure 6 Schematic diagram of the buckling model and grid model in the embodiment of the present invention;

[0064] Figure 7 Schematic diagram of four waviness temperature field function diagrams in the embodiment of the present invention; among them, (a) is the center wave, (b) is the edge wave, (c) is the quarter wave, and (d) is the edge-center composite wave;

[0065] Figure 8 Schematic diagram of four waviness first-order buckling modes in the embodiment of the present invention;

[0066] Figure 9 Schematic diagram of the influence of different thicknesses on the buckling critical value in the embodiment of the present invention; among them, (a) is the center wave, (b) is the edge wave, (c) is the quarter wave, and (d) is the edge-center composite wave;

[0067] Figure 10 Schematic diagram of the influence of different widths on the buckling critical value in the embodiment of the present invention; among them, (a) is the center wave, (b) is the edge wave, (c) is the quarter wave, and (d) is the edge-center composite wave;

[0068] Figure 11 Schematic diagram of the post-buckling paths of strip steels with different thicknesses for four wavinesses in the embodiment of the present invention; among them, (a) is the center wave, (b) is the edge wave, (c) is the quarter wave, and (d) is the edge-center composite wave;

[0069] Figure 12 Schematic diagram of the LSTM memory cell structure in the embodiment of the present invention;

[0070] Figure 13 Schematic diagram of the waviness defect prediction model based on data in the embodiment of the present invention;

[0071] Figure 14 Schematic diagram of the training of the Stacking model in the embodiment of the present invention;

[0072] Figure 15Schematic diagram of the establishment process of the strip shape inheritance and evolution model dataset according to the embodiments of the present invention;

[0073] Figure 16 Schematic diagram of the process for establishing a Stacking integrated learning waviness prediction model according to the embodiments of the present invention;

[0074] Figure 17 Schematic diagram of the strip shape feedforward control process of the continuous annealing and tempering mill according to the embodiments of the present invention;

[0075] Figure 18 Schematic diagram of the method process of a cross-process strip shape feedforward control applicable to the strip steel of the continuous annealing and tempering mill unit according to the embodiments of the present invention;

[0076] Figure 19 Schematic diagram of the principle of cross-process feedforward control of the tempering mill according to the embodiments of the present invention. Detailed implementation manners

[0077] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying 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 of 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.

[0078] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0079] First, briefly summarize that the technical steps of this embodiment mainly include:

[0080] 1) Pattern recognition and classification of the strip shape at the outlet of the cold tandem rolling mill unit

[0081] In the entire strip steel production line, only the cold tandem rolling mill unit is equipped with detection devices such as strip shape meters at the end. Therefore, the strip shape of the cold tandem rolling mill unit is statistically analyzed, and at the same time, pattern recognition is performed on the strip shape at the cold rolling outlet, and the quality of the strip shape at the cold rolling outlet is classified.

[0082] 2) Finite element simulation modeling of pre-buckling and post-buckling of cold-rolled strip steel and establishment of prediction models for the wave height, wave width, and wave distance of various typical wave shape defects driven by data

[0083] Based on the classification of the strip shape quality of cold-rolled export strips, a pre-buckling and post-buckling calculation model of typical waviness defects is established through finite element simulation. The buckling modes, critical conditions and post-buckling paths corresponding to the strip width, strip thickness, initial IU value size and distribution under different typical waviness defects are solved. The influence laws of various influencing factors on the post-buckling wave height, wave distance and wave width of the strip are studied, a large database of various typical waviness defects based on simulation is formed, and a prediction model of wave height, wave distance and wave width based on simulation big data is established on this basis.

[0084] 3) Research on the prediction model of strip shape inheritance and evolution in the continuous annealing process based on data-driven

[0085] Accumulate a large amount of strip shape quality data of strip steel before and after continuous annealing at the production site, and propose a strip shape inheritance and evolution model before and after the annealing furnace based on the Stacking integrated learning model. First, through steps such as original data collection, data cleaning and processing, and normalization processing, use a one-dimensional convolutional neural network to extract eigenvalue, input the extracted eigenvalue and waviness type into the integrated learning model for training to obtain a 1DCNN-Stacking waviness classification model to accurately predict the strip shape inheritance and evolution law of strip steel.

[0086] 4) Research on the strip shape feedforward control system of the continuous annealing and tempering mill

[0087] On the basis of the original strip shape preset model of the tempering mill unit, superimpose the strip shape feedforward control idea and method considering cross-process proposed in this patent. The specific implementation method is to classify and pattern recognize the strip shape defects of the strip steel measured by the strip shape gauge at the exit of the acid rolling mill unit. On this basis, predict the wave height, wave distance and wave width of the cold-hardened coiled strip steel, obtain the actual strip shape quality of the strip steel after continuous annealing with the help of the prediction model of strip shape inheritance and evolution before and after continuous annealing, convert the actual strip shape quality into the IU value distribution of the strip steel at the entrance of the tempering mill, and carry out feedforward control for the strip shape incoming material defects on this basis.

[0088] A method for cross-process strip shape feedforward control applicable to strip steel of the continuous annealing and tempering mill unit proposed in this embodiment, as Figure 19 shown, specifically includes the following steps:

[0089] Step 1, pattern recognize the strip shape of the strip steel at the exit of the cold tandem rolling mill unit;

[0090] Furthermore, the pattern recognition of the strip shape of the strip steel at the exit of the cold tandem rolling mill unit includes:

[0091] Obtain the IU value distribution of the strip shape of the strip steel at the exit of the cold tandem rolling mill unit: divide the width and thickness of the strip steel according to a preset range, and statistically calculate the average strip shape IU value for each steel type according to the division of width and thickness;

[0092] Based on the IU value of the strip shape distribution, the strip shape pattern of the cold tandem rolling mill outlet strip is identified using Legendre orthogonal polynomials.

[0093] Specifically, in this embodiment, the pattern recognition of the strip shape of the cold tandem rolling mill outlet strip:

[0094] Throughout the strip production line, only the end of the cold tandem rolling mill is equipped with detection devices such as a strip shape meter. Therefore, the strip shape of the cold tandem rolling mill strip is statistically analyzed, and the pattern recognition of the cold rolling outlet strip shape is carried out. A strip shape pattern recognition model is designed based on Legendre orthogonal polynomials. The normalized samples of the measured strip shape are subjected to mathematical operations to extract the basic strip shape pattern coefficients. The basic strip shape pattern coefficients are compared in terms of magnitude and positive / negative, and the largest basic strip shape pattern coefficient is determined. The strip shape defect corresponding to the largest basic strip shape pattern coefficient is the strip shape defect of the strip.

[0095] 1) Statistical analysis of the IU distribution of the cold tandem rolling mill outlet strip shape

[0096] Statistical analysis is carried out on the strip shapes of key steel grades produced by the cold tandem rolling mill. First, the width and thickness of the strip are divided into certain ranges. For each steel grade, the average IU value of the strip shape is statistically calculated according to the width and thickness divisions and described in the form of a heat map. The average IU value of the strip shape and the number of rolled coils in each division interval are marked. The size of the IU value is represented by colors of different depths, and the darker the color, the larger the IU value. The average IU value of the first 50 meters of the strip head and the average IU value of the entire strip length are mainly statistically analyzed.

[0097] The following is the statistics of the cold tandem rolling mill outlet strip of a certain cold tandem rolling mill:

[0098] For the M3 soft steel series M3A45, a total of 1922 coils were rolled in the first quarter, and the overall strip shape was good. The average strip shape value of the first 50 meters of the strip head was 2.624 IU. After being subdivided by width and thickness, the average IU values of narrow and thin and wide and thick materials were relatively large; the average strip shape value of the entire strip length was 1.529, and the average IU value of narrow and thin, that is, with a thickness of 0.16 - 0.65 mm and a width of 700 - 1100 mm, was relatively large, reaching 2.411. As Figure 1 shown; Figure 1 (a) of is M3A45IU_50, Figure 1 (b) of is M3A45 IU_ALL.

[0099] For the household appliance plate SDX51D, which is mainly of thin specifications of 0.6 and below, a total of 725 coils were rolled in the first quarter of SDX51D. The average strip shape value of the first 50 meters of the strip head was 2.756 IU. After being subdivided by width and thickness, the average IU value of the thickness of 0.65 - 1.25 mm and the width of 700 - 1100 mm was relatively large, reaching 3.205 IU; the average strip shape value of the entire strip length was 1.634, and the overall IU value was very small. As Figure 2 shown; Figure 2(a) is SDX51D IU_50, Figure 2 (b) is SDX51D IU_ALL.

[0100] In the first quarter, CSA01 rolled a total of 1,448 coils. The average shape value of the leading 50-meter strip is 2.841 IU. After being segmented by width and thickness, the average IU value of the strip with a thickness of 1.25 - 2.6 and a width of 700 - 1,100 is relatively large, reaching 4.346 IU; the average shape value of the entire strip length is 1.468, and the overall shape is good. As Figure 3 shown; Figure 3 (a) is CSA01 IU_50, Figure 3 (b) is CSA01 IU_ALL.

[0101] 2) Pattern recognition of the strip shape at the exit of the cold tandem rolling mill

[0102] There are eight standard patterns for the strip shape of cold-rolled strip steel: left edge wave, right edge wave, center wave, double edge wave, right triple wave, left triple wave, quarter wave, and edge-center wave. The basic patterns of the strip shape are represented by the first, second, third, and fourth Legendre orthogonal polynomials p1(x), p2(x), p3(x), and p4(x) respectively. To represent the strip shape curves of different strip widths, first normalize the strip shape along the strip width direction, and the normalized strip width range is [-1, 1]. These eight standard patterns can be represented by the following normalized equations:

[0103] Standard normalized equation for left edge wave:

[0104] Y1 = p1(x) = x

[0105] Standard normalized equation for right edge wave:

[0106] Y2 = -p1(x) = -x

[0107] Standard normalized equation for center wave:

[0108]

[0109] Standard normalized equation for double edge wave:

[0110]

[0111] Standard normalized equation for right triple wave:

[0112]

[0113] Standard normalized equation for left triple wave:

[0114]

[0115] Standard normalized equation for quarter wave:

[0116]

[0117] Edge - center wave standard normalization equation:

[0118]

[0119] The eight basic pattern curves of strip shape are as Figure 4 shown, among which, Figure 4 (a) of it is the first - order component, Figure 4 (b) of it is the second - order component, Figure 4 (c) of it is the third - order component, Figure 4 (d) of it is the fourth - order component.

[0120] The strip - shape deviation curve can be expressed as a linear combination of eight standard patterns and the strip - shape error term:

[0121] Δε(y) = a1p1(x)+a2p2(x)+a3p3(x)+a4p4(x)

[0122] In the formula, a i (i = 1, 2, 3, 4) respectively represent the characteristic coefficients of the first - order, second - order, third - order, and fourth - order strip - shape patterns. Their magnitudes represent the contents of each order of strip - shape patterns, and the positive and negative respectively represent the two types corresponding to the left and right, middle and both sides of the strip - shape standard pattern.

[0123] Among the eight patterns of cold - rolled strip steel shape, the most typical wavy patterns are four: center wave, double - edge wave, quarter - wave, and edge - center wave, which respectively correspond to the second - order and fourth - order polynomials in Legendre polynomials. At this time, the strip shape can be expressed as:

[0124]

[0125] Interpolate the data of the cold - rolled exit strip - shape meter to obtain multiple segments of discrete values. The identification of strip shape is through mathematical operations to extract the characteristic coefficients of the strip - shape pattern from the discrete values of strip shape. Since the coefficients can be either positive or negative, they can be respectively expressed as the magnitudes of the four basic strip - shape pattern components. In this way, the discrete strip shape can be transformed into Legendre polynomial coefficients. Compare the magnitudes and positive and negative of the obtained Legendre polynomial coefficients to determine the largest strip - shape basic pattern coefficient. The strip - shape defect corresponding to the largest strip - shape basic pattern coefficient is the strip - shape defect of this strip steel, as Figure 5 is the result after the strip - shape pattern recognition by Legendre orthogonal polynomials.

[0126] The process of strip - shape pattern recognition based on Legendre orthogonal polynomials is as follows:

[0127] ① Obtain the strip - shape distribution value;

[0128] ② Perform mean - value removal processing on the strip - shape distribution value;

[0129] ③ Calculate the shape deviation;

[0130] ④ Normalize the shape deviation;

[0131] ⑤ Extract the basic shape mode coefficients;

[0132] ⑥ Compare the magnitudes and signs of the basic shape mode coefficients;

[0133] ⑦ Determine the maximum mode coefficient, and the shape defect corresponding to the maximum mode coefficient is the shape defect of the strip.

[0134] Step 2: Based on the identified shape mode of the strip at the exit of the cold tandem rolling mill, use a preset defect prediction model to predict the waviness defect data of the strip at the exit of the cold tandem rolling mill; wherein, the defect prediction model is constructed based on a long short-term memory neural network;

[0135] Further, predicting the waviness defect data of the strip at the exit of the cold tandem rolling mill using a preset defect prediction model includes:

[0136] Based on the identified shape mode of the strip at the exit of the cold tandem rolling mill, conduct finite element simulation modeling for the pre-buckling and post-buckling of the cold-rolled strip;

[0137] Based on the finite element simulation of the pre-buckling and post-buckling of the cold-rolled strip, convert the IU value distribution of various typical shape defects into the form of physical shape defects, and obtain the wave height, wave distance, and wave width of the physical shape defects corresponding to the IU value distribution of each shape;

[0138] Based on various typical waviness defects obtained from finite element simulation calculations, the initial IU value of the strip, and the wave height, wave distance, and wave width corresponding to the strip thickness and strip width, construct a buckling database;

[0139] Use the buckling database to train the long short-term memory neural network to obtain the defect prediction model;

[0140] Input the width, thickness, and IU value of the strip into the defect prediction model to predict the waviness defect data of the strip at the exit of the cold tandem rolling mill; wherein, the waviness defect data includes: wave height, wave distance, and wave width.

[0141] Conducting finite element simulation modeling for the pre-buckling and post-buckling of the cold-rolled strip includes:

[0142] Construct a buckling model;

[0143] Based on the buckling model, conduct pre-buckling behavior analysis of the waviness to obtain the first-order buckling modes of center waves, edge waves, quarter waves, and edge-center composite waves;

[0144] Based on the buckling model, conduct post-buckling behavior analysis of the waviness.

[0145] The pre-buckling result is a mode, and the post-buckling result is a displacement. The wave distance and wave width of the wave shape can be obtained in the pre-buckling result buckling mode; the post-buckling behavior analysis can obtain the post-buckling path of the strip, that is, the relationship between the wave height and stress of the strip wave shape. The stress can be converted to the IU value, so different IU values can be obtained corresponding to different wave heights of the strip.

[0146] Specifically, in this embodiment, the finite element simulation modeling of the pre-buckling and post-buckling of the cold-rolled strip and the establishment of the wave height, wave width and wave distance prediction model of each typical wave shape defect driven by data include:

[0147] 1) Establishment of buckling model

[0148] The research object is a steel strip with a thickness of less than 2.4 mm. The ratio of thickness to width and length is very small, so the S4R unit in ABAQUS is used. The width of the steel strip is 2B = 1400 mm, and the length is 2L = 6000 mm. The overall mesh size is set to 25 mm × 25 mm. The material parameters are set to: elastic modulus E = 2.1 × 10 5 MPa, Poisson's ratio μ=0.3, thermal expansion coefficient α x =1×10 -5 ℃ -1 , α y =α z = 0, Gauss integration is used as the thickness integration rule, and the thickness integration point is set to 5. The buckling model and mesh model are as follows Figure 6 shown.

[0149] The boundary constraint conditions are as follows: set a node set at the center point of the model and constrain its displacement in the X and Y directions to prevent the rigid body displacement of the model. At the same time, in order to make the model bear force and improve the constraint, all the translational degrees of freedom of the corresponding points on both sides of the lateral side (X = ± L) are coupled. Boundary conditions:

[0150]

[0151] Among them, (x, y) refers to the coordinates of the strip, the x range is [-L, L] and the y range is [-B, B], and u (x,y) It means that at the coordinate (x,

[0152] y), the displacement in the x direction, v (x,y) refers to the displacement in the y direction at the coordinate (x,y), w (x,y) It refers to the displacement in the z direction at the coordinate (x,y). (L,y) and u (-L,y) It refers to the x-direction displacement of all points with abscissa L, -L (x=L, -L), v (L,y) and v (-L,y)refers to the displacement in the y direction of all the points on the ordinate with abscissa L and -L (x = L, -L), w (L,y) and w (-L,y) refers to the displacement in the z direction of all the points on the ordinate with abscissa L and -L (x = L, -L), u (0,0) and v (0,0) refers to the displacements in the x and y directions of the point with coordinates (0, 0).

[0153] Different wavy shapes have different boundary conditions, so the parts that need to be constrained are also different. From the wavy shape pattern of the center wave, it can be seen that there is no displacement in the Z direction on both longitudinal sides (Y = ±B) of the model, so the displacements in the Z direction of these two sides can be constrained; from the wavy shape pattern of the edge wave, it can be seen that there is no displacement in the Z direction on the longitudinal center line (Y = 0) of the model, so the displacements in the Z direction of this center line can be constrained; the buckling of the quarter wave occurs between the edge line and the center line, and the peak is at a distance of one - quarter of the plate width from the edge. Therefore, the displacements in the Z direction of both longitudinal sides and the center line are constrained; the wavy shape of the edge - center composite wave is opposite to that of the quarter wave, and the buckling occurs at the center line and both sides, and there is no buckling at a distance of one - quarter of the plate width from the edge. Therefore, the displacement in the Z direction of the plate width at a distance of one - quarter from the edge is constrained.

[0154] Such as Figure 7 As shown in the temperature field function diagrams of the four wavy shapes, the buckled wavy shape undergoes buckling deformation through a specific initial strain (temperature field). Loading an uneven initial strain distribution in the strip width direction will cause the strip to buckle and generate periodic wavy shapes; among them, Figure 7 (a) of Figure 7 is the center wave, Figure 7 (b) of Figure 7 is the edge wave,

[0155] Center wave

[0156] Edge wave

[0157] Quarter wave

[0158] Edge - center composite wave

[0159] Among them, Y is the different widths of the strip, the range of Y is [-B, B], and T is the temperature field function.

[0160] 2) Pre - buckling analysis of the wavy shape

[0161] According to the principle of minimum potential energy, in the buckling analysis of strip steel, the first-order mode should be extracted as the corresponding mode for the pre-buckling critical condition, and the pre-buckling critical stress value of the strip steel can be solved through the eigenvalue corresponding to the first-order mode. The four waviness modes are as follows Figure 8 shown, which are the first-order buckling modes of center wave, edge wave, quarter wave, and combined edge and center wave respectively.

[0162] ① The influence law of thickness on the buckling critical value, as Figure 9 shown;

[0163] ② The influence of width on the buckling critical value, as Figure 10 shown.

[0164] 3) Post-buckling behavior analysis of waviness

[0165] When performing post-buckling analysis, the arc-length method (Riks) is generally used for solution. The arc-length method can solve post-buckling problems including stable buckling and unstable buckling. However, the discontinuous response at the buckling point makes it difficult to accurately solve the post-buckling problem. Therefore, how to convert the discontinuous response problem into a continuous response problem becomes a key to accurately analyzing the post-buckling problem. Introducing initial defects in the arc-length method helps to solve the emergence of equilibrium path bifurcation and at the same time overcomes the convergence problem of using the arc-length method for nonlinear buckling analysis.

[0166] When using the arc-length method for solution, certain initial defects need to be introduced. Generally speaking, there are three methods:

[0167] ① Static analysis displacement. The geometric distribution after deformation obtained by solving the static analysis in ABAQUS / Standard is used as the input of the initial defect for post-buckling calculation.

[0168] ② Introduction of buckling modes. Each mode obtained by solving the Linear Perturbation - Buckle module is introduced as the initial defect for post-buckling, either superimposed or individually depending on the situation.

[0169] ③ Direct definition. In specific cases, such as when the defect values of each node of the element are clear, the defects can be directly introduced through the node coordinate and coordinate perturbation value table.

[0170] In this embodiment, the initial defects are applied by introducing buckling modes. It should be noted here that the defect scaling factor is an empirical value. For shell elements, generally taking a few percent of the thickness is sufficient. According to the experience and calculation correction of the research group, the scaling factor is taken as 0.001. There are three conditions for the termination of the arc-length method solution:

[0171] ① Exceeding the set maximum number of increment steps;

[0172] ② Reaching the set maximum load scaling factor;

[0173] ③The set maximum displacement and degrees of freedom are reached.

[0174] The post-buckling paths of strip steel with a width of 1400 mm and different thicknesses are solved, and the calculation results are as Figure 11 shown; among them, Figure 11 (a) of Figure 11 (b) of Figure 11 (c) of Figure 11 (d) of

[0175] 4) Establishment of the buckling database

[0176] Through the finite element simulation of the pre-buckling and post-buckling of cold-rolled strip steel, the IU value distribution of various typical shape defects can be converted into the form of physical shape defects, and the wave height, wave distance, and wave width of the physical shape defects corresponding to the IU value distribution of each shape can be obtained, and a large simulation-based database can be formed based on this.

[0177] 5) Fast prediction models for the wave height, wave distance, and wave width of physical shape defects of strip steel with various typical wave shape defects based on data

[0178] Based on the various typical wave shape defects, the initial IU values of strip steel, the thickness of strip steel, and the large simulation database of the wave height, wave distance, and wave width corresponding to the strip width obtained from the aforementioned finite element simulation calculations, fast prediction models for the wave height, wave distance, and wave width of physical shape defects of strip steel with various typical wave shape defects are established.

[0179] ①Long short-term memory neural network (LSTM)

[0180] The long short-term memory network (LSTM) is a special recurrent neural network. Its basic unit is the LSTM cell, and each LSTM cell contains a cell state and three gates: an input gate, a forget gate, and an output gate. As Figure 12 shown.

[0181] a. Cell state

[0182] The cell state is the main line running through the entire LSTM cell and is used to transmit long-term information. Its update mainly relies on the forget gate and the input gate.

[0183] b. Forget gate

[0184] The forget gate determines how much information to forget from the cell state. It generates a value between 0 and 1 through a sigmoid function, and then multiplies it with the cell state to determine which information to retain. The formula for the forget gate is:

[0185] f t= σ(W f · [h t-1 , x t + b f )

[0186] Wherein, f t is the output of the forget gate; x t is the input at the current moment, h t-1 is the output of the hidden layer at the previous moment, σ is the activation coefficient of the four gates, W f , b f are the weight matrix and the bias matrix respectively.

[0187] c. Input gate

[0188] The input gate determines how much new information to add to the cell state. It consists of a sigmoid function and a tanh function. The sigmoid function determines which values to update, and the tanh function generates new candidate values. The input gate formula is:

[0189] i t = σ(W i · [h t-1 , x t + b i )

[0190] C t = tanh(W C · [h t-1 , x t + b C )

[0191] Wherein, i t is the proportional coefficient for saving the input to the state storage unit, C t is the corrected state storage unit, W i , b i are the corresponding weight matrix and bias matrix respectively.

[0192] d. Update cell state

[0193] The cell state is updated through the forget gate and the input gate. The forget gate determines which information needs to be forgotten, and the input gate determines which new information to add.

[0194] C t = f t * C t-1 + i t * C t

[0195] Wherein, C t , C t-1 are the state storage units at the current moment and the previous moment respectively.

[0196] e. Output gate

[0197] The output gate determines how much information is output from the cell state. It determines the final output through a sigmoid function and a tanh function.

[0198] o t = σ(W o ·[h t-1 , x t +b o )

[0199] h t = o t *tanh(C t )

[0200] In the formula, o t is the output of the corrected output gate, h t is the final output of the hidden layer, W o , b o are the corresponding weight matrix and bias matrix respectively.

[0201] ② The waviness defect prediction model based on data is as Figure 13 shown;

[0202] The input of the waviness defect prediction model is the width, thickness and IU value of the strip steel. Through the long short-term neural network model and combined with the simulation database of wave height, wave distance and wave width obtained by finite element calculation, a rapid prediction model of the actual strip shape can be obtained.

[0203] Step 3: Based on the waviness defect data, use the preset quality prediction model of the strip shape quality inheritance and evolution before and after continuous annealing to predict the actual strip shape quality of the strip after continuous annealing; wherein, the quality prediction model is constructed based on the Stacking ensemble learning model;

[0204] Furthermore, predicting the actual strip shape quality of the strip after continuous annealing includes:

[0205] Collecting the original data; wherein, the original data includes: cold rolling exit waviness data, online platform data and manual observation data;

[0206] Cleaning and normalizing the original data;

[0207] Performing feature extraction on the processed original data;

[0208] Training the Stacking ensemble learning model with the extracted features to obtain the quality prediction model of the strip shape quality inheritance and evolution before and after continuous annealing.

[0209] Further, training the Stacking ensemble learning model using the extracted features includes:

[0210] In the first stage, several individual machine learning models are trained using the extracted features to obtain several prediction results; among them, the several individual machine learning models include: multi-layer perceptron, support vector machine, and random forest algorithm;

[0211] In the second stage, a multiple linear regression model is trained using the several prediction results;

[0212] Based on the trained several individual machine learning models and the trained multiple linear regression model, a quality prediction model for the shape quality inheritance and evolution of strip steel before and after continuous annealing is constructed.

[0213] Specifically, in this embodiment, the research on the inheritance and evolution law of strip steel shape driven by continuous annealing process data is as follows:

[0214] The strip steel at the cold rolling exit is annealed in the annealing furnace, and the strip steel shape will be greatly improved. There is a shape meter detection device at the cold rolling exit. Only knowing the waviness of the strip steel at the cold rolling exit cannot provide an accurate initial shape value for the temper mill. The waviness data of the actual strip steel after annealing is required. However, a shape measurement device cannot be installed during the continuous annealing process, so manual observation of the shape data is needed. But manual observation takes a lot of time and effort, with low efficiency, and it is impossible to observe all specifications of strip steel. Traditional machine learning algorithms have problems such as difficulty in parameter optimization. Therefore, a shape inheritance and evolution model before and after the annealing furnace based on an ensemble learning model is proposed.

[0215] 1) Stacking ensemble learning model based on small sample data

[0216] As Figure 14 shown, the Stacking model includes two parts: a base learner and a meta-learner. The base learner is trained based on the original dataset, and the output results are integrated and passed into the meta-learner. Finally, the results of the meta-learner are obtained for evaluation and analysis.

[0217] The first-layer base learner has two requirements: excellent model prediction performance and model diversity. Excellent means that the prediction performance of each base learner should be good, and the classification prediction performance should be at the same level; diversity means that there should be large differences in the predictions of each base learner, learning and training from different angles, and making full use of the advantages of each model to achieve better performance. Therefore, three base learners are selected, namely multi-layer perceptron, support vector machine, and random forest algorithm.

[0218] The selection of the second-layer meta-learner has a great impact on the generalization performance of the Stacking model. Due to the differences in the predictions of the first-layer base learners, a suitable meta-learner is required to optimize the prediction performance of the Stacking model. Therefore, the Stacking model must meet four requirements for the meta-learner: good algorithm performance, robustness, stability, and computational efficiency. The second layer uses the multiple linear regression method to integrate the classification results of the first-layer base learners and train the final prediction results.

[0219] 2) Establishment of the flatness genetic and evolution model dataset;

[0220] Such as Figure 15 Before constructing the integrated learning waviness prediction model as shown, it is necessary to construct the flatness dataset. Only by constructing a reasonable dataset can the model learn the important information and exert its more accurate prediction performance. The flatness prediction dataset establishment process is as Figure 15 shown.

[0221] ① Raw data collection:

[0222] The raw data includes cold rolling exit waviness data, online platform data, and manual observation data. Among them, the cold rolling exit data mainly includes the flatness IU value of the steel plate, and the data platform data includes 56 features such as the specifications of each strip steel (width, thickness, strength grade), the average temperature at each part of the annealing furnace, and the tension difference between the inlet and outlet. The manual observation data includes the flatness grade of the strip steel at the inlet of each annealing furnace and the flatness grade of the strip steel at the outlet of the annealing furnace.

[0223] ② Data cleaning and processing:

[0224] Because the original dataset is very large, especially in the strip steel rolling process with complex technological processes, data may be lost or abnormal. Therefore, corresponding cleaning work needs to be carried out on the original data, and the 3σ principle is used to filter the abnormal data, and the rows where the filtered abnormal data is located are also deleted. After cleaning the original data, the flatness measurement values are processed. The flatness IU mean value is statistically calculated for each steel type according to the width and thickness, and the long-term observation experience and the statistical value of the cold rolling strip steel exit IU value are converted into the strip steel waviness grade, and finally the flatness mode input features and the waviness grade classification prediction target values are obtained.

[0225] Normalization processing:

[0226] Some of the input features in the dataset are dimensional data, such as the average temperature at each part of the annealing furnace, the outlet tension, and the thickness, etc., and some data are dimensionless, such as the average elongation of the temper mill. The difference between dimensional data and dimensionless data may reach 10 5If the heights are of different units, it is not conducive to the training of the model, which will cause the features with larger dimensional numerical values to cover the dimensionless features. Therefore, it is necessary to normalize the data and scale it to a specified range.

[0227] ③ Feature extraction of the ensemble learning model:

[0228] The convolutional neural network is a feedforward neural network generated by imitating the receptive fields of cells in the mammalian visual cortex. It can automatically extract the effective features in the input signal, make a judgment on the current signal through network selection and classification output, screen out the features, and provide the screened features to the ensemble learning prediction model to improve the prediction accuracy.

[0229] ④ Establish a Stacking ensemble learning waviness prediction model:

[0230] The ensemble learning method trains the model in two stages. In the first stage, three individual machine learning models are trained to obtain three prediction results respectively. In the second stage, the multiple linear regression model is used as the meta-learner of the ensemble model, and the prediction results of the three models are used as input elements to train again to obtain the final ensemble prediction model.

[0231] Use the one-dimensional convolutional neural network model to extract the eigenvalue, input the extracted eigenvalue and the waviness type into the ensemble learning model for training to obtain the 1DCNN-Stacking waviness classification model. As Figure 16 shown

[0232] Step 5: Convert the physical strip shape quality into the IU value distribution of the strip steel at the entry of the temper mill, and based on the IU value distribution of the strip steel at the entry of the temper mill, perform the feedforward control of the bending roll force of the strip steel shape.

[0233] Specifically, in this embodiment, the shape feedforward control of the tandem temper mill is as Figure 17 shown. Based on the predicted physical strip shape defects after annealing, convert them into the distribution of the strip steel shape IU value, and based on the shape regulation effect of the temper mill, realize the feedforward control of the bending roll force of the strip steel shape, that is, adjust the bending roll force of the temper mill in real time according to the strip shape defects of the incoming material. Finally, realize the control and improvement of the strip shape quality of the temper mill unit.

[0234] Based on the pattern recognition and classification of the strip steel shape at the exit of the cold tandem rolling mill unit, the distribution of the strip steel shape IU value measured by the shape meter can be visualized as various physical strip shape defects and classified and summarized.

[0235] Based on the finite element simulation modeling of the pre-buckling and post-buckling of cold-rolled strip steel, the IU value distribution of various typical strip shape defects can be converted into the form of physical strip shape defects, and the wave height, wave distance and wave width of the physical strip shape defects corresponding to the IU value distribution of each strip shape can be obtained, and a simulation-based large database is formed based on this.

[0236] Establishment of prediction models for the wave height, wave width, and wave distance of various typical wavy defects driven by data. Based on the various typical wavy defects obtained from the aforementioned finite element simulation calculations, the initial IU value of the strip steel, and the simulation database of the wave height, wave distance, and wave width corresponding to the strip steel thickness and width, a rapid prediction model for the strip steel wave height, wave distance, and wave width of physical shape defects of various typical wavy defects based on data is established.

[0237] Research on the prediction model of strip steel shape inheritance and evolution in the continuous annealing process driven by data. Based on the actual measurement of a large amount of production data, a data model for strip steel shape inheritance and evolution before and after continuous annealing is established, which can realize the rapid and accurate prediction of the wave height, wave distance, and wave width of the physical shape of strip steel for various steel grades and specifications during the continuous annealing process.

[0238] Shape feedforward control of the continuous annealing and tempering mill. Based on the predicted physical shape defects of the strip steel after annealing, convert them into the distribution of the strip steel shape IU value, and based on the shape control effect of the tempering mill, realize the feedforward control of the bending roll force of the strip steel shape, that is, adjust the bending roll force of the tempering mill in real time according to the shape defects of the incoming strip steel. Finally, realize the control and improvement of the shape quality of the tempering mill unit.

[0239] Based on the statistical analysis of the strip steel shape at the outlet of the cold rolling mill unit and historical data, pattern recognition of the strip steel shape is carried out. With the help of the research on the law of strip steel shape inheritance and evolution driven by data in the continuous annealing process, predict the shape quality of the strip steel after annealing and provide it as the initial shape value for the tempering mill unit. Finally, through the established database of the shape control effect of strip steel of different specifications in the tempering rolling process, feedforward control is carried out on the incoming strip steel of the tempering mill unit to obtain high-quality finished products. This comprehensive control idea effectively combines data analysis, prediction models, and real-time control technologies, providing a scientific and intelligent solution for the shape control of the continuous annealing and tempering mill.

[0240] The embodiments described above are only descriptions of the preferred embodiments of the present invention and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for cross-process shape feedforward control of strip steel in a continuous annealing and tempering unit, characterized in that, Including: Performing pattern recognition on the strip shape of the cold tandem rolling mill outlet strip; Based on the recognized strip shape pattern of the cold tandem rolling mill outlet strip, using a preset defect prediction model to predict the waviness defect data of the strip shape of the cold tandem rolling mill outlet strip; wherein, the defect prediction model is constructed based on a long short-term memory neural network; Based on the waviness defect data, using a preset quality prediction model for the genetic and evolution of strip shape quality before and after continuous annealing to predict the physical strip shape quality of the strip after continuous annealing; wherein, the quality prediction model is constructed based on a Stacking ensemble learning model; Converting the physical strip shape quality into the IU value distribution of the strip at the entry of the skin pass mill, and based on the IU value distribution of the strip at the entry of the skin pass mill, performing feedforward control of the work roll bending force for the strip shape.

2. The method for strip shape feedforward control applicable to continuous annealing and tempering unit across processes according to claim 1, wherein Performing pattern recognition on the strip shape of the cold tandem rolling mill outlet strip includes: Obtaining the IU value distribution of the strip shape at the outlet of the cold tandem rolling mill: dividing the width and thickness of the strip according to a preset range, and statistically calculating the average IU value of the strip shape for each steel grade according to the division of width and thickness; Based on the IU value distribution of the strip shape, using Legendre orthogonal polynomials to identify the strip shape pattern of the cold tandem rolling mill outlet strip.

3. The method for cross-process shape feedforward control of strip steel applicable to continuous annealing and tempering units according to claim 2, characterized in that, Using Legendre orthogonal polynomials to identify the strip shape pattern of the cold tandem rolling mill outlet strip includes: Performing mean removal processing on the IU value distribution of the strip shape; Based on the IU value distribution of the strip shape after mean removal processing, calculating the strip shape deviation and the maximum value of the strip shape deviation; Normalizing the strip shape deviation to obtain a normalized sample; Extracting the basic pattern coefficients of the strip shape; Comparing the magnitudes, positive and negative values of the basic pattern coefficients of the strip shape; Determining the largest basic pattern coefficient of the strip shape, and the strip shape defect corresponding to the largest pattern coefficient is the strip shape defect of the strip.

4. The method for cross-process shape feedforward control of strip steel applicable to continuous annealing and tempering unit according to claim 1, characterized in that, Using a preset defect prediction model to predict the waviness defect data of the strip shape of the cold tandem rolling mill outlet strip includes: Based on the recognized strip shape pattern of the cold tandem rolling mill outlet strip, performing finite element simulation modeling on the pre-buckling and post-buckling of cold-rolled strip; Based on the finite element simulation of the pre-buckling and post-buckling of cold-rolled strip, converting the IU value distribution of various typical strip shape defects into the form of physical strip shape defects, and obtaining the wave height, wave distance, and wave width of the physical strip shape defects corresponding to the IU value distribution of each strip shape; Based on various typical waviness defects obtained from finite element simulation calculations, the initial IU value of the strip, the wave height, wave distance, and wave width corresponding to the strip thickness and strip width, constructing a buckling database; Using the buckling database to train a long short-term memory neural network to obtain the defect prediction model; Inputting the width, thickness, and IU value of the strip into the defect prediction model to predict the waviness defect data of the strip shape of the cold tandem rolling mill outlet strip; wherein, the waviness defect data includes: wave height, wave distance, and wave width.

5. The method for cross-process shape feedforward control of strip steel applicable to continuous annealing and tempering units according to claim 4, characterized in that, Performing finite element simulation modeling on the pre-buckling and post-buckling of cold-rolled strip includes: Constructing a buckling model; Based on the buckling model, performing pre-buckling behavior analysis of the waviness to obtain the first-order buckling modes of center waves, edge waves, quarter waves, and edge-center composite waves; Based on the buckling model, performing post-buckling behavior analysis of the waviness to obtain the post-buckling path of the strip.

6. The method for cross-process shape feedforward control of strip steel applicable to continuous annealing and tempering unit according to claim 5, characterized in that Constructing the buckling model includes: Setting boundary constraint conditions; Setting the temperature field function and distribution form; The boundary constraint conditions are: where u (L,y) is the displacement in the x - direction at x = L of the strip, u (-L,y) is the displacement in the y - direction at x = - L of the strip, v (L,y) is the displacement in the y - direction at x = L of the strip, v (-L,y) is the displacement in the y - direction at x = L of the strip, w (L,y) is the displacement in the z - direction at x = L of the strip, w (-L,y) is the displacement in the z - direction at x = L of the strip, u (0,0) is the displacement in the x - direction at the center point of the strip, v (0,0) is the displacement in the y - direction at the center point of the strip; x, y refer to the coordinates of the strip; The temperature field function and its distribution form are as follows: Medium wave Edge wave One-quarter wave Edge-center composite wave Among them, Y is the different widths of the strip steel, and T is the temperature field function.

7. The method for cross-process shape feedforward control of strip steel applicable to continuous annealing and tempering unit according to claim 1, characterized in that Predicting the physical shape quality of the strip steel after continuous annealing includes: Collecting the original data; among them, the original data includes: cold rolling exit waviness data, online platform data, and manual observation data; Cleaning and normalizing the original data; Extracting features from the processed original data; Using the extracted features to train a Stacking ensemble learning model to obtain a quality prediction model for the inheritance and evolution of the strip steel shape quality before and after continuous annealing.

8. The method for strip steel shape feedforward control applicable to the continuous annealing and tempering unit across processes according to claim 7, characterized in that Using the extracted features to train a Stacking ensemble learning model includes: In the first stage, using the extracted features to train several individual machine learning models to obtain several prediction results; among them, the several individual machine learning models include: multi-layer perceptron, support vector machine, and random forest algorithm; In the second stage, using the several prediction results to train a multiple linear regression model; Based on the trained several individual machine learning models and the trained multiple linear regression model, a quality prediction model for the inheritance and evolution of the strip steel shape quality before and after continuous annealing is constituted.

9. The method for strip steel shape feedforward control applicable to the continuous annealing and tempering unit across processes according to claim 7, characterized in that The cold rolling exit waviness data includes: the IU value of the strip steel shape; The online platform data includes: the specifications of each strip steel, the average temperature at each part of the annealing furnace, and the tension difference between the inlet and outlet; The manual observation data includes: the strip steel shape grades at the inlet of each annealing furnace and the strip steel shape grades at the outlet of the annealing furnace.

Images