An adaptive cold-rolled strip multi-channel shape prediction method

Abstract

The application discloses a kind of self-adapting cold-rolled strip multi-channel shape prediction method.The method mainly includes the following steps: collecting cold-rolled strip rolling production data and pretreatment;Based on extreme gradient boosting algorithm, a cold-rolled strip multi-channel shape prediction model is established;The training set is used to train the cold-rolled strip multi-channel shape prediction model to obtain the model CSFP_M;The noise shape channel optimization algorithm is fused with the model CSFP_M to obtain the self-adapting cold-rolled strip multi-channel shape prediction model Adaptive-CSFP_M;The model Adaptive-CSFP_M is embedded in the cold-rolled strip shape measurement and control system to realize the early prediction of shape.The method can accurately predict the shape under the condition of meeting the industrial production speed for different specifications of strip rolling production data, and can effectively eliminate the noise channel in the predicted shape to realize adaptive adjustment, which is of great significance to improve the quality of cold-rolled strip.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of plate shape measurement and control in the metallurgical rolling process, and particularly relates to a self-adaptive multi-channel plate shape prediction method for cold-rolled strips. BACKGROUND

[0002] Plate shape is a key quality index of strips, and its influencing factors are complex and changeable, with characteristics of multi-parameters, strong coupling and nonlinearity, etc., and it is very difficult to control. To improve the plate shape quality of strips, systematic research must be carried out in both plate shape detection and plate shape control. Among them, plate shape detection is the premise of realizing plate shape control, and belongs to the hardware of the plate shape measurement and control system, which is usually called "plate shape meter" in industry. Plate shape control is the final goal, and belongs to the software of the plate shape measurement and control system, which is usually called "plate shape control system" in industry, and the core of the plate shape control system is the plate shape prediction model. The essence of plate shape setting control is to take the plate shape prediction model as a target function calculator, and to optimize different plate shape control means through an optimization algorithm; plate shape predictive control also needs to be based on the plate shape prediction model to predict the plate shape state at the next moment in advance, so as to realize the pre-adjustment of the control means; when the plate shape feedback control is needed, the plate shape mode recognition is needed according to the deviation between the target plate shape and the measured plate shape, and then the control means is dynamically adjusted according to the result of the plate shape mode recognition, and the dynamic adjustment amount of various plate shape control means must be based on the calculation result of the plate shape prediction model.

[0003] Most of the researches on plate shape prediction are based on mechanism models, which have strong regularity and expansibility, but the model has many influencing factors, strong coupling and low robustness, and a large amount of simplification is needed on the basis of sacrificing accuracy. With the rapid development of computer technology and information collection technology, a large amount of data is collected in real time by collection equipment in industrial production. These production data contain rich significance and rules. So far, some scholars have combined plate shape production data with artificial intelligence algorithms to establish a plate shape prediction model, breaking the limitations of traditional mechanism models, but there are generally problems such as less training data, low diversification degree of rolling production parameters and plate shape, unreasonable selection of plate shape influencing factors, etc., although a relatively satisfactory prediction performance can be obtained, but model overfitting and other problems are prone to occur, which cannot be practically applied to the production site. Therefore, it is of great significance to invent a plate shape intelligent prediction model with stronger generalization ability and universality on the basis of meeting the prediction accuracy. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a self-adaptive multi-channel plate shape prediction method for cold-rolled strips, which can accurately predict the plate shape under the condition of meeting the requirements of industrial production.

[0005] In order to achieve the above object, the present application provides a kind of adaptive cold-rolled strip multi-channel shape prediction method, mainly including the following steps:

[0006] Collecting cold-rolled strip production data and preprocessing to obtain the training set and test set of the prediction model;

[0007] Based on extreme gradient boosting algorithm, a cold-rolled strip multi-channel shape prediction model is established;The cold-rolled strip multi-channel shape prediction model is a multi-channel parallel prediction model composed of multiple extreme gradient boosting algorithms, each extreme gradient boosting algorithm is trained as a sub-model to form a channel, and each sub-model is independent of each other;

[0008] The training set is used to train the cold-rolled strip multi-channel shape prediction model;

[0009] The noise shape channel optimization algorithm is integrated into the cold-rolled strip multi-channel shape prediction model to obtain an adaptive cold-rolled strip multi-channel shape prediction model;The noise shape channel is a channel that is more than the real shape channel when the prediction shape dimension is greater than the real shape dimension, and the noise shape channel optimization algorithm is used to eliminate the noise shape channel;

[0010] The adaptive cold-rolled strip multi-channel shape prediction model is embedded into the cold-rolled strip shape measurement and control system, and the current shape is predicted in real time according to the production data and compared with the set target shape to feedback to the control system to realize early control.

[0011] Further, the cold-rolled strip production data includes rolling parameters and shape data, and the cold-rolled strip production data is collected by sensors and shape meters respectively.

[0012] Further, the preprocessing of cold-rolled strip production data includes:

[0013] Extract the input and output related features of the model, take the outlet shape of the cold-rolled strip shape data as the output feature of the model, and select the following parameter features in the cold-rolled strip rolling parameters as the input features of the model, including: strip width, inlet thickness, outlet thickness, inlet thickness difference, outlet thickness difference, speed, rolling force, roll gap difference, driving side work bend, operation side work bend, driving side middle bend, operation side middle bend, inlet tension, outlet tension, segmented cooling duty ratio;

[0014] Extract stable rolling zone data, and extract data with rolling speed greater than or equal to rolling speed threshold value in the cold-rolled strip production data as stable rolling zone data;

[0015] Eliminate abnormal value data, and use isolated forest algorithm to eliminate abnormal value of a certain feature parameter at a certain time;

[0016] Further, the cold-rolled strip multi-channel plate shape prediction model is trained by using the training set, and the training includes:

[0017] The training parameters of the cold-rolled strip multi-channel plate shape prediction model are set, different parameter combinations are traversed by a grid search method, and the optimal parameters are determined: learning rate, iteration number, loss function, maximum tree depth, minimum loss reduction value required for node splitting, L1 regularization coefficient, and L2 regularization coefficient.

[0018] The model is trained by using the training set and adopting a five-fold cross-validation method, and the model with the minimum mean square error of the validation set in five training results is saved and used as the cold-rolled strip multi-channel plate shape prediction model.

[0019] Further, the noise plate shape channel optimization algorithm includes:

[0020] S1, a volume of strip data is taken from a test set, the actual width of the strip is denoted as W strip , the maximum number of channels of a plate shape detector roller is denoted as k, the single-channel width of the plate shape detector roller is denoted as W roll , and the predicted plate shape obtained after the cold-rolled strip multi-channel plate shape prediction model is input is denoted as a matrix M a×b .

[0021] S2: the theoretical effective channel number is calculated , and a temporary variable i is set as 0;

[0022] S3: it is judged whether k' % 2 = 0 is satisfied; if satisfied, go to step S4, otherwise go to step S5;

[0023] S4: k' = k'-1 is set;

[0024] S5: it is judged whether k' % 2 = 0 is satisfied ; if satisfied, go to step S6, otherwise go to step S8;

[0025] S6: M a×b [0:a][i] = 0, M a×b [0:a][k-i] = 0, and i = i+1 is set;

[0026] S7: it is judged whether k' % 2 = 0 is satisfied ; if satisfied, go to step S8, otherwise go to step S5;

[0027] S8: the optimized predicted plate shape M a×b is obtained.

[0028] Further, the adaptive cold-rolled strip multi-channel plate shape prediction model is verified by using the test set.

[0029] Compared with the prior art, the present application has the following beneficial effects:

[0030] In the present application, firstly, based on the cold-rolled strip industry production data, the production actual experience and data mining technology are combined to establish the cold-rolled strip production dataset CSR_Dataset, and the problems such as less training data, low diversification degree of production parameters and strip shape, and unreasonable selection of strip shape influencing factors in the existing strip shape prediction model are solved. Secondly, based on the extreme gradient boosting algorithm and using the dataset, a cold-rolled strip multi-channel strip shape prediction model is obtained, and finally, according to the model prediction result, a self-adaptive cold-rolled strip multi-channel strip shape prediction model is obtained by fusing the noise strip shape channel optimization algorithm. The model can accurately and quickly predict the strip shape for different specifications of cold-rolled strip production data, and can effectively eliminate the noise channel in the predicted strip shape to realize self-adaptive adjustment, which has important significance for improving the quality of cold-rolled strip. BRIEF DESCRIPTION OF DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0032] Figure 1 The flow chart of the self-adaptive cold-rolled strip multi-channel strip shape prediction method in the embodiment of the present application;

[0033] Figure 2 The cold-rolled strip multi-channel strip shape prediction model in the embodiment of the present application;

[0034] Figure 3 The five-fold cross-validation training schematic diagram in the embodiment of the present application;

[0035] Figure 4 The flow chart of the noise strip shape channel optimization algorithm in the embodiment of the present application;

[0036] Figure 5 The self-adaptive cold-rolled strip multi-channel strip shape prediction model fusing the noise strip shape channel optimization algorithm in the embodiment of the present application;

[0037] Figure 6 The average absolute error distribution histogram of the model Adaptive-CSFP_M for each sample prediction result of CSR_Testset in the embodiment of the present application;

[0038] Figure 7 The average absolute error scatter plot of the model Adaptive-CSFP_M for each sample prediction result of CSR_Testset in the embodiment of the present application;

[0039] Figure 8 This is a local comparison diagram of the actual and predicted flatness drawn by randomly extracting a row of data from CSR_Testset in an embodiment of the present invention;

[0040] Figure 9 This is a three-dimensional surface diagram of the actual and predicted export plate shapes of the 650mm-wide strip data in CSR_Testset in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0042] like Figure 1 As shown, an embodiment of the present invention provides an adaptive cold-rolled strip multi-channel flatness prediction method, comprising the following steps:

[0043] S1: Collect cold-rolled strip production data and preprocess it to obtain the training set and test set of the prediction model.

[0044] Cold rolling production lines generate a large amount of data in real time during the production process, mainly including rolling parameters and flatness data. The rolling parameters include: strip width, entry thickness, speed, rolling force, etc. Flatness data includes: entry flatness and exit flatness. These are collected in real time by sensors and flatness meters and other collection equipment with a time period of t = 20ms and transmitted back to the host computer to be saved as cold rolled strip production data. Assume that the maximum number of detection channels of the flatness meter detection roller of a certain cold rolling production line is recorded as k = 27, and the width of a single channel of the flatness meter detection roller is recorded as W roll =26mm, the collected production data of a roll of original cold-rolled strip is recorded as D, and the data D has n=34596 rows of data and m+2k=147 columns of features, including m=93 columns of rolling parameter features, k=27 columns of inlet flatness features and k=27 columns of outlet flatness features. The production data of a roll of original cold-rolled strip is recorded as

[0045] After collecting the cold strip production data, in order to facilitate the application of the shape prediction model, the collected data needs to be preprocessed, mainly including: extracting model input and output related features, extracting stable rolling zone data and removing abnormal value data. Among them: extracting model input and output related features specifically includes: about output features, taking the k=27 column exit shape of a roll of original cold strip production data D as the model output feature; selecting 15 rolling parameters in the cold strip production data, a total of 37 columns of data as the input features of the model, including: strip width, inlet thickness, outlet thickness, inlet thickness difference, outlet thickness difference, speed, rolling force, roll gap difference, driving side work bend, operation side work bend, driving side middle bend, operation side middle bend, inlet tension, outlet tension, sub-cooling duty cycle. That is, from the m=93 columns of data in a roll of original cold strip production data D, m'=37 columns of data are selected as the input features of the model, as shown in Table 1. The production data after extracting the model input and output features is denoted as

[0046] Table 1

[0047]

[0048] Extracting stable rolling zone data specifically includes: in the rolling start stage, first the strip thickness control is carried out, at this time the rolling speed is low, the feedback link of the shape control system is in the closed state and does not participate in the shape regulation and control, and when the speed is too low, the pressure signal detected by the shape detection roller is unstable, and the shape value has a large error. Therefore, a rolling speed threshold v=60 m / min is set, and the data with a speed greater than or equal to v=60 m / min in each roll is extracted as the stable rolling zone data. The extracted data has n'=32691 rows, and is denoted as

[0049]

[0050] Removing abnormal value data specifically includes: in the cold rolling production process, the sensor and other parameter acquisition devices may suddenly fluctuate due to various factors such as device vibration, strip deviation, transmission signal interference, etc., resulting in abnormal values in the collected data. Therefore, the isolated forest algorithm is used to remove the abnormal values in the data.

[0051] First, given a sample set X={x1, x2, …, x n |x i ∈R m′}, define an isolated tree, let T be a root node of the isolated tree, the node T can have two cases, one is an external node without child nodes, and the other is a node with two child nodes (T l , T r) and a test end, the test end contains an attribute q and a split point p, if q < p, the sample x i goes to child node T l , otherwise goes to child node T r . The path length of sample x i in the isolated tree is recorded as h(x), that is, x i goes through the number of nodes in the iTree. The isolated forest is constructed by r isolated trees, and the subset X' is randomly selected from the sample set X, and is put into the root node of the isolated tree. The average path length of the subset X' is recorded as , where H(i) is the harmonic number, which can be recorded as H(i) = ln(i) + ξ, ξ ≈ 0.5772 is the Euler constant. Finally, the anomaly score of sample x i is recorded as When the value of s of sample point x i is closer to 1, it means that the point is more likely to be an abnormal point; the value of s is closer to 0, which means it is more likely to be a normal point. The removed abnormal value data accounts for about 5.5% of the total data amount, and the cold-rolled strip shape data after removal has n" = 30896 rows, recorded as

[0052] Based on the above preprocessing steps, the prediction model data set is constructed and divided, mainly including the following steps: in order to improve the size of the model training set data and the diversification degree of the strip shape data, 5 volumes are selected from each of the 11, 13, 15, 17, 19, 21, 23, 25 total eight specifications of data in the collected cold-rolled strip production database, in order to avoid the occurrence of production accidents and other special situations, and to ensure that the data amount of each selected strip is greater than 5000 rows, and the s = 40 volumes are combined to form the cold-rolled strip original data set D i ={D1,D2,…,D 40}, the total data amount is 858414 rows, and the data distribution is shown in Table 2. After the batch data preprocessing step, the cold-rolled strip production data set D" i ={D" i ,D" 40} is obtained, recorded as CSR_Dataset.

[0053] Table 2

[0054]

[0055] The data set is divided, and 85% of the data is randomly selected from each volume of cold-rolled strip production data D" i (i = 1, …, 40) as a training set CSR_Trainset, and the rest as a test set CSR_Testset.

[0056] In the embodiment of the present application, based on the cold-rolled strip industrial production data, the production actual experience and data mining technology are combined to establish a cold-rolled pass production dataset CSR_Dataset, and the problem of less training data in the existing plate shape prediction model is solved.

[0057] S2: Establishing a cold-rolled strip multi-pass plate shape prediction model based on an extreme gradient boosting algorithm.

[0058] An extreme gradient boosting algorithm (XGBoost) based on a regression tree is used to construct a cold-rolled strip prediction model: given a sample set wherein x i represents a sample with an m-dimensional feature vector, y i represents a sample label. Let the expression of a decision tree model be wherein f(x i ) is the output of the decision tree; is the node value corresponding to each leaf node; and q(x i ) is the mapping relationship between the sample and the leaf node of the decision tree. XGBoost is an additive model containing multiple decision trees, and the expression of an XGBoost containing K decision trees is wherein represents the predicted value of the model for the sample x i ; is the cumulative result of K-1 trees; and f K (x i ) represents the result of the Kth regression tree. Since the outlet plate shape value of the cold-rolled strip dataset CSR_Dataset has 27 feature dimensions, 27 XGBoosts are used to construct a multi-pass parallel prediction model, each XGBoost is trained as a sub-model to form a pass, and the sub-models are independent of each other and do not have coupling effects between different passes. The model structure is shown in Figure 2 .

[0059] S3: Training the cold-rolled strip multi-pass plate shape prediction model using the training set to obtain the model CSFP_M.

[0060] The training of the cold-rolled strip multi-pass plate shape prediction model mainly includes the following steps:

[0061] The training parameters of the cold-rolled strip multi-channel plate shape prediction model are set, different combinations of parameters are traversed by the grid search method, and the optimal parameters are determined: the learning rate lr is set to 0.41, the iteration number num_estimators is set to 300, the loss function loss is set to mean square error, the maximum depth of the tree m_depth is set to 6, the minimum loss drop value required for node splitting gamma is set to 0.2, the L1 regularization coefficient reg_1 is set to 2.6, and the L2 regularization coefficient reg_2 is set to 0.5;

[0062] The model is trained by using the divided training set CSR_Trainset and adopting the five-fold cross-validation method, as shown in Figure 3 The model is trained by using the divided training set CSR_Trainset and adopting the five-fold cross-validation method, as shown in

[0063] S4: Fusion noise plate shape channel optimization algorithm, get adaptive cold-rolled strip multi-channel plate shape prediction model Adaptive-CSFP_M.

[0064] Noise plate shape channel optimization algorithm, select a roll of actual strip width W strip = 410 mm, and the effective channel number is 15. The real data of the outlet plate shape detected by the plate shape instrument is shown in Table 3 (taking the first 5 rows of data).

[0065] Table 3

[0066]

[0067]

[0068] The results of the CSFP_M model obtained by training for plate shape prediction appear noise plate shape, that is, the predicted plate shape channel dimension is always greater than or equal to the real plate shape channel dimension, and always equal to the maximum plate shape channel dimension in the training set. The extra plate shape channel dimension is always the edge channel, and the extra edge plate shape channel value is 100 times different from the average value of the middle normal channel plate shape value. Therefore, it is defined as noise plate shape channel, and an optimization algorithm needs to be designed to eliminate the noise plate shape channel. The algorithm flow chart is shown in Figure 4 The main steps include:

[0069] S41: From the test set CSR_Testset, take a roll of strip width W strip = 410 mm, the maximum channel number k of the plate shape instrument detection roller is 27, and the single channel width of the plate shape instrument detection roller is denoted as W roll= 26 mm, the data of actual channel number 15 are input into the model CSFP_M, and the obtained predicted plate shape value is recorded as matrix M a×b As shown in Table 4;

[0070] S42: calculating the theoretical effective channel number and setting temporary variable i = 0;

[0071] S43: judging whether k' % 2 = 0 is satisfied; if satisfied, going to step S44, otherwise going to step S45;

[0072] S44: k' = k' - 1;

[0073] S45: judging whether k' % 2 = 0 is satisfied; if satisfied, going to step S46, otherwise going to step S48;

[0074] S46: M a×b [0:a][i] = 0, M a×b [0:a][k-i] = 0, i = i + 1;

[0075] S47: judging whether k' % 2 = 0 is satisfied; if satisfied, going to step S48, otherwise going to step S45; S48: obtaining the optimized predicted plate shape M a×b , and the optimized part of data is shown in Table 5.

[0076] Table 4

[0077]

[0078]

[0079] Table 5

[0080]

[0081]

[0082] The noise plate shape channel optimization algorithm is fused into the constructed CSFP_M model to obtain an adaptive cold-rolled strip multi-channel plate shape prediction model Adaptive-CSFP_M, and the model structure is shown in Figure 5 The performance of the Adaptive-CSFP_M is evaluated by using the test set CSR_Testset, the average root mean square error of the overall samples of the test set CSR_Testset is 0.6816I, which meets the requirements of industrial production. In order to more intuitively observe the performance of the model Adaptive-CSFP_M on the test set CSR_Testset, the average absolute error MAE of each sample in the test set CSR_Testset is plotted into a scatter plot as Figure 6The shown and MAE distribution histogram is as Figure 7 shown. Figure 8 The real and predicted plate shape local contrast chart drawn from the randomly selected row data in the test set CSR_Testset, Figure 9 The real plate shape and the Adaptive-CSFP_M predicted plate shape three-dimensional surface chart is drawn from the data of 1017 rows with a bandwidth of 620mm and a channel number of 23 in the test set CSR_Testset. From the local and overall surface chart, it can be seen that the plate shape predicted by the model can basically fit the real plate shape, but the model performs detail blurring processing on the prediction in terms of detail texture. This situation shows that the model has certain generalization ability and can achieve good performance on different bandwidth test sets.

[0083] S5: Embed the final cold rolled strip multi-channel plate shape prediction model Adaptive-CSFP_M into the cold rolled strip plate shape measurement and control system, and compare the real-time predicted current plate shape with the set target plate shape to feedback to the control system to realize early control.

[0084] In the above embodiments, a cold rolled strip production data set of different production specifications and plate shapes is constructed based on a large amount of cold rolled strip production data, solving the problems of few training data, low degree of diversification of production parameters and plate shapes, and unreasonable selection of plate shape influencing factors in existing plate shape prediction models. Secondly, based on the extreme gradient boosting algorithm, a cold rolled strip multi-channel plate shape prediction model is established and trained using the data set. After the noise plate shape channel optimization algorithm and the cold rolled strip multi-channel prediction model are fused, an adaptive cold rolled strip multi-channel plate shape prediction model is obtained. This model can accurately and quickly predict the plate shape for different specifications of strip rolling production data, and can effectively eliminate the noise channels in the predicted plate shape to realize adaptive adjustment, which is of great significance to improve the quality of cold rolled strips.

[0085] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for adaptive cold rolled strip multi-pass flatness prediction, characterized in that, The application relates to a cold-rolled strip plate shape prediction method. Collecting and preprocessing cold-rolled strip production data to obtain a training set and a test set of a prediction model; A cold-rolled strip multi-channel plate shape prediction model is established based on an extreme gradient boosting algorithm; the cold-rolled strip multi-channel plate shape prediction model is a multi-channel parallel prediction model composed of multiple extreme gradient boosting algorithms; each extreme gradient boosting algorithm is used as a sub-model to train a channel; the sub-models are independent of each other; The training set is used to train the cold-rolled strip multi-channel plate shape prediction model; A noise plate shape channel optimization algorithm is fused into the cold-rolled strip multi-channel plate shape prediction model to obtain an adaptive cold-rolled strip multi-channel plate shape prediction model; the noise plate shape channel is a channel that is more than a real plate shape channel when the prediction plate shape dimension is greater than the real plate shape dimension; the noise plate shape channel optimization algorithm is used to eliminate the noise plate shape channel; The adaptive cold-rolled strip multi-channel plate shape prediction model is embedded into a cold-rolled strip plate shape measurement and control system; current plate shape is predicted in real time according to production data and compared with a set target plate shape to realize early control by feeding back to a control system.

2. The adaptive cold rolled strip multi-pass flatness prediction method of claim 1, wherein, The cold-rolled strip production data includes rolling parameters and plate shape data; the cold-rolled strip production data is collected in real time by sensors and a plate shape instrument.

3. The adaptive cold rolled strip multi-pass flatness prediction method of claim 2, wherein, The preprocessing of the cold-rolled strip production data includes: Extracting model input and output related features; taking outlet plate shape in the cold-rolled strip plate shape data as model output features; and selecting the following parameter features in the cold-rolled strip rolling parameters as model input features, including: strip width, inlet thickness, outlet thickness, inlet thickness difference, outlet thickness difference, speed, rolling force, roll gap difference, driving side work bend, operation side work bend, driving side middle bend, operation side middle bend, inlet tension, outlet tension, and segmented cooling duty cycle; Extracting stable rolling zone data; extracting data with a rolling speed greater than or equal to a rolling speed threshold value in the cold-rolled strip production data as stable rolling zone data; Eliminating abnormal value data; adopting an isolated forest algorithm to eliminate abnormal values of a certain feature parameter at a certain moment.

4. The adaptive cold rolled strip multi-pass flatness prediction method of claim 1, wherein, The training set is used to train the cold-rolled strip multi-channel plate shape prediction model, including: Setting training parameters of the cold-rolled strip multi-channel plate shape prediction model; determining optimal parameters by traversing different parameter combinations through a grid search method: learning rate, iteration number, loss function, maximum tree depth, minimum loss reduction value required for node splitting, L1 regularization coefficient, and L2 regularization coefficient; Using the training set and adopting a five-fold cross-validation method to train the constructed model; saving a model with the minimum mean square error of a validation set in five training results as the cold-rolled strip multi-channel plate shape prediction model.

5. The adaptive cold rolled strip multi-pass flatness prediction method of claim 1 wherein, The noise plate shape channel optimization algorithm includes: S1, take a strip data from the test set, the actual width of the strip is recorded as W strip , the maximum number of passages of the shape meter detection roller is recorded as k, and the single passage width of the shape meter detection roller is recorded as W roll , the predicted shape obtained after inputting the multi-pass strip shape prediction model of the cold-rolled strip is recorded as matrix M a×b ; S2: Calculate the theoretical effective channel number and set temporary variable i = 0; S3: judging whether k' % 2 = 0 is satisfied; if yes, going to step S4; otherwise, going to step S5; S4: k' = k'-1; S5: judge whether or not If yes, go to step S6, otherwise go to step S8; S6: Let M a×b [0:a][i] = 0, M a×b [0:a][k-i] = 0, i = i + 1; S7: determine whether or not If yes, go to step S8, otherwise go to step S5. S8: Obtain the optimized predicted plate shape M a×b .

6. The adaptive cold rolled strip multi-pass flatness prediction method of claim 1, wherein, The test set is used to verify the adaptive cold-rolled strip multi-channel plate shape prediction model.

Images