A method and device for classifying slab bending patterns based on machine learning

Abstract

The present invention discloses a method and device for classifying slab bending modes based on machine learning. The method includes: obtaining the original data of the slab bending measured by the detection instrument corresponding to the slab to be classified, and preprocessing the obtained original data to obtain the slab full-length centerline curve data corresponding to the slab to be classified; respectively calculating the distances between the slab full-length centerline curve data and each slab bending template curve data in the preset standard template library; using the distances between the slab full-length centerline curve data and all slab bending template curve data in the standard template library as the input of the preset machine learning model to achieve the classification of slab bending modes. The present invention uses the distances between the bending curves and all the curves in the template library as the input of the machine learning model, and through the training of the model, realizes the classification of slab bending modes, improving the classification accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of hot continuous rolling roughing slab camber recognition and classification, and particularly relates to a slab bending mode classification method and device based on machine learning. Background Art

[0002] In the process of hot continuous rolling of strip steel, due to different mill rigidities on the operating side and the driving side, uneven cooling resulting in a temperature difference between the two sides of the slab, unreasonable tilting values on both sides of the rolling mill, etc., the slab will have a planar shape bend. After multiple passes of rolling, the bending shape of the slab will become very complex. Since the bending of the slab will cause accidents such as roll jamming, rolling seizure, strip breakage, tearing, and steel piling in subsequent rolling, seriously affecting the stability and continuity of production. Therefore, the accurate classification of the slab bending shape is an important part of the evaluation of the planar shape quality and an important prerequisite for realizing the automatic control of the slab planar shape. However, at present in the production site, it mainly relies on the subjective observation and experience judgment of technicians. This process has a large amount of analysis data, low efficiency, and most of the work is repetitive each time, resulting in a large waste of human resources. Summary of the Invention

[0003] The present invention provides a slab bending mode classification method and device based on machine learning to solve the technical problems of low efficiency and high human cost in the prior art.

[0004] To solve the above technical problems, the present invention provides the following technical solutions:

[0005] On the one hand, the present invention provides a slab bending mode classification method based on machine learning, including:

[0006] Obtaining the original data of the slab bending measured by the detection instrument corresponding to the slab to be classified, and preprocessing the obtained original data to obtain the slab full-length center line curve data corresponding to the slab to be classified;

[0007] Calculating the distances between the slab full-length center line curve data corresponding to the slab to be classified and each slab bending template curve data in a preset standard template library; wherein, the standard template library includes slab bending template curve data corresponding to various different types of slab bending forms;

[0008] Using the distances between the slab full-length center line curve data corresponding to the slab to be classified and all the slab bending template curve data in the standard template library as the input of a preset machine learning model, and realizing the classification of the slab bending mode of the slab to be classified through the machine learning model.

[0009] Further, the preprocessing of the obtained original data includes:

[0010] Normalize both the horizontal and vertical coordinates of the acquired raw data to [0, 1], and perform linear interpolation; where the number of data points after interpolation is n, and the value range of n is 150 - 200.

[0011] Furthermore, the standard template library includes:

[0012] The slab bending template curve data of the "C" bend on the operating side, and its function expression is:

[0013] y = -13.108x 6 + 39.324x 5 - 65.84x 4 + 66.14x 3 - 36.333x 2 + 9.8171x

[0014] The slab bending template curve data of the "C" bend on the drive side, and its function expression is:

[0015] y = 13.108x 6 - 39.324x 5 + 65.84x 4 - 66.14x 3 + 36.333x 2 - 9.8171x + 1

[0016] The slab bending template curve data of the "L" bend on the operating side at the head, and its function expression is:

[0017] y = -6.554x 6 + 35.751x 5 - 73.144x 4 + 73.493x 3 - 38.578x 2 + 10.049x

[0018] The slab bending template curve data of the "L" bend on the drive side at the head, and its function expression is:

[0019] y = 6.554x 6 - 35.751x 5 + 73.144x 4 - 73.493x 3 + 38.578x 2 - 10.049x + 1

[0020] The slab bending template curve data of the "L" bend on the operating side at the tail, and its function expression is:

[0021] y = -6.554x6 +3.5726x 5 +7.3036x 4 -7.3533x 3 +2.2448x 2 -0.2322x + 1

[0022] Slab bending template curve data for the "L"-shaped bend on the tail drive side, and its function expression is:

[0023] y = 6.554x 6 -3.5726x 5 -7.3036x 4 +7.3533x 3 -2.2448x 2 +0.2322x

[0024] Slab bending template curve data for the "S"-shaped bend on the drive side, and its function expression is:

[0025] y = -16.089x 5 +40.224x 4 -40.423x 3 +20.411x 2 -5.1407x + 1

[0026] Slab bending template curve data for the "S"-shaped bend on the operator side, and its function expression is:

[0027] y = 16.089x 5 -40.224x 4 +40.423x 3 -20.411x 2 +5.1407x

[0028] Among them, x represents the abscissa of the corresponding slab bending template curve, y represents the ordinate of the corresponding slab bending template curve, and the number of data points of each slab bending template curve is m, and the value range of m is 150 - 200.

[0029] Furthermore, calculating the distances between the slab full-length centerline curve data corresponding to the to-be-classified slab and each slab bending template curve data in the preset standard template library includes:

[0030] Using the dynamic time warping DTW algorithm to calculate the distances between the slab full-length centerline curve data corresponding to the to-be-classified slab and each slab bending template curve data in the preset standard template library respectively.

[0031] Furthermore, the machine learning model is a support vector machine SVM model.

[0032] Furthermore, the construction process of the machine learning model includes:

[0033] Obtain the full-length centerline curve of the hot-rolled slab as sample data;

[0034] Construct a training sample set: {(x i , y i ) | x i = {d i1 , d i2 , d i3 , d i4 , d i5 , d i6 , d i7 , d i8} ∈ R 8 , y i ∈ R, i = 1, 2,..., M+, where x i represents the sequence composed of the distances between the i-th sample curve S i and all slab bending template curve data in the standard template library, d ij represents the distance between the i-th sample curve S i and the j-th slab bending template curve data in the standard template library, j = 1, 2, 3, 4, 5, 6, 7, 8; y i represents the slab bending mode type corresponding to the i-th sample curve S i , and M represents the total number of samples;

[0035] Use the Gaussian function to map the data samples to a high-dimensional feature space and construct a classification function based on the SVM model in the high-dimensional space: where ω and b are the regression parameters of the SVM model;

[0036] Initialize the parameters C and g of the SVM model, and represent the parameter solution of the SVM model as the following constrained optimization problem:

[0037]

[0038] Satisfy the constraint ε i ≥ 0; where ε i is the slack variable, and the calculation formula is represents the Gaussian function corresponding to x i ;

[0039] Train the SVM model based on the training sample set, optimize the SVM model according to the training results, save the model parameters with the highest score after tuned training, and obtain the machine learning model.

[0040] Further, tuning the SVM model according to the training result includes:

[0041] Optimizing the parameters C and g of the SVM model by using the K-fold cross-validation method, including: evenly dividing the training sample set into K groups, using 1 group as the verification data sample, and the remaining K - 1 groups as the training data samples, and each group of data is taken as the verification data sample in turn; in each verification, taking the parameter C and g to take value combinations at a certain step size within the given value range, the step size of the parameter C is (C max -C min ) / K, and the step size of the parameter g is (g max -g min ) / K; where C max , C min , g max , g min are the maximum and minimum values of the parameters C and g respectively; under each group of parameter (C, g) combinations, perform K calculations respectively, and take the mean value of the model test accuracy of the K calculations as the score score of the model under this group of (C, g):

[0042]

[0043] where f p is the number of correctly classified for each category, and S is the number of the test sample set each time.

[0044] On the other hand, the present invention also provides a slab bending mode classification device based on machine learning, and the slab bending mode classification device based on machine learning includes:

[0045] A slab full-length centerline curve data acquisition module, configured to acquire the original data of the slab bending measured by the detection instrument corresponding to the slab to be classified, and preprocess the acquired original data to obtain the slab full-length centerline curve data corresponding to the slab to be classified;

[0046] A curve distance calculation module, configured to calculate the distances between the slab full-length centerline curve data corresponding to the slab to be classified obtained by the slab full-length centerline curve data acquisition module and each slab bending template curve data in a preset standard template library; where the standard template library includes slab bending template curve data corresponding to various different types of slab bending forms;

[0047] A slab bending mode classification module is configured to calculate the distance between the slab full-length centerline curve data corresponding to the slab to be classified calculated by the curve distance calculation module and all slab bending template curve data in the standard template library, and use it as the input of a preset machine learning model. Through the machine learning model, the classification of the slab bending mode of the slab to be classified is realized.

[0048] In another aspect, the present invention also provides an electronic device, which includes a processor and a memory; wherein, at least one instruction is stored in the memory, and the instruction is loaded and executed by the processor to implement the above method.

[0049] In yet another aspect, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored, and the instruction is loaded and executed by the processor to implement the above method.

[0050] The beneficial effects brought by the technical solution provided by the present invention at least include:

[0051] The present invention can accurately classify the slab bending form by calculating the distance between the slab bending curve and the template curve in a machine learning manner, and embed the slab bending form classification model into the slab bending quality evaluation system, so as to automatically identify the bending form of each slab and realize the classification of the slab bending mode. Compared with the traditional judgment method based on the minimum distance, the present invention greatly improves the accuracy of classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0053] Figure 1 is a schematic execution flow diagram of a method for classifying slab bending modes based on machine learning provided by an embodiment of the present invention;

[0054] Figure 2 is a schematic diagram of the full-length centerline curve of a hot-rolled slab provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will further describe the embodiments of the present invention in detail with reference to the drawings.

[0056] First Embodiment

[0057] This embodiment provides a method for classifying slab bending patterns based on machine learning. This method is based on the curve data of the full-length center line of the slab and realizes the classification of slab bending patterns through machine learning. This method can be implemented by an electronic device. The execution process of this method is as Figure 1 shown, and includes the following steps:

[0058] S1. Obtain the original data of the slab bending measured by the detection instrument corresponding to the slab to be classified, and preprocess the obtained original data to obtain the curve data of the full-length center line of the slab corresponding to the slab to be classified;

[0059] Specifically, in this embodiment, the preprocessing of the obtained original data is carried out in the following manner:

[0060] Normalize both the abscissa and ordinate of the obtained original data to [0, 1], and perform linear interpolation; among them, the number of data points after interpolation is n, and the value range of n is 150 - 200.

[0061] S2. Calculate the distances between the curve data of the full-length center line of the slab corresponding to the slab to be classified and the curve data of each slab bending template in the preset standard template library; among them, the standard template library includes the curve data of slab bending templates corresponding to various different types of slab bending forms;

[0062] Specifically, in this embodiment, the standard template library includes the following 8 bending curve functions:

[0063] The curve data of the slab bending template for the "C" - shaped bend on the operator side, and its function expression is:

[0064] y = - 13.108x 6 +39.324x 5 -65.84x 4 +66.14x 3 -36.333x 2 +9.8171x

[0065] The curve data of the slab bending template for the "C" - shaped bend on the drive side, and its function expression is:

[0066] y = 13.108x 6 -39.324x 5 +65.84x 4 -66.14x 3 +36.333x 2 -9.8171x + 1

[0067] The curve data of the slab bending template for the "L" - shaped bend on the operator side at the head, and its function expression is:

[0068] y = -6.554x 6 + 35.751x 5 - 73.144x 4 + 73.493x 3 - 38.578x 2 + 10.049x

[0069] The curve data of the slab bending template for the "L" bend on the head drive side, and its function expression is:

[0070] y = 6.554x 6 - 35.751x 5 + 73.144x 4 - 73.493x 3 + 38.578x 2 - 10.049x + 1

[0071] The curve data of the slab bending template for the "L" bend on the tail operation side, and its function expression is:

[0072] y = -6.554x 6 + 3.5726x 5 + 7.3036x 4 - 7.3533x 3 + 2.2448x 2 - 0.2322x + 1

[0073] The curve data of the slab bending template for the "L" bend on the tail drive side, and its function expression is:

[0074] y = 6.554x 6 - 3.5726x 5 - 7.3036x 4 + 7.3533x 3 - 2.2448x 2 + 0.2322x

[0075] The curve data of the slab bending template for the "S" bend on the drive side, and its function expression is:

[0076] y = -16.089x 5 + 40.224x 4 - 40.423x 3 + 20.411x 2 - 5.1407x + 1

[0077] The curve data of the slab bending template for the "S" bend on the operation side, and its function expression is:

[0078] y = 16.089x 5 - 40.224x 4+40.423x 3 -20.411x 2 +5.1407x

[0079] Among them, x ∈ [0, 1], which is the abscissa of the corresponding slab bending template curve, y is the ordinate of the corresponding slab bending template curve, the number of data points of each slab bending template curve is m, and the value range of m is 150 - 200.

[0080] Furthermore, the distances between the slab full-length centerline curve data corresponding to the slab to be classified and the curve data of each slab bending template curve in the preset standard template library are calculated as follows: The distances between the slab full-length centerline curve data corresponding to the slab to be classified and the curve data of each slab bending template curve in the preset standard template library are calculated respectively by using the dynamic time warping (DTW) algorithm. The process is as follows:

[0081] For two sequences Q and P with lengths n and m respectively, where Q = q1, q2,..., q i ,...q n , P = p1, p2,..., p j ,...p m ; where q i represents the i-th element in sequence Q, i = 1, 2,..., n, and p j represents the j-th element in sequence P, j = 1, 2,..., m;

[0082] The DTW distance is calculated using the following dynamic programming formula:

[0083] γ(i, j) = d(q i , p j ) + min{γ(i - 1, j - 1), γ(i - 1, j), γ(i, j - 1)}

[0084] where d(q i , p j ) = (q i - p j ) 2 ; γ(i, j) represents the minimum cumulative distance corresponding to p j and q i .

[0085] For the slab full-length centerline curve data obtained in S1, calculate the distances between it and the curves corresponding to 8 standard template library functions respectively, and obtain {d i1 , d i2 , d i3 , d i4 , d i5 , d i6 , di7 , d i8}。

[0086] S3, to calculate the distance between the slab full-length centerline curve data corresponding to the slab to be classified and all the slab bending template curve data in the standard template library, and use it as the input of a preset machine learning model. Through the machine learning model, the classification of the slab bending mode of the slab to be classified is realized.

[0087] Specifically, in this embodiment, the machine learning model is a support vector machine SVM model.

[0088] The construction process of the machine learning model includes:

[0089] Obtain the original bending data of hot-rolled slabs, and preprocess the original bending data to obtain the hot-rolled slab full-length centerline curve as sample data; in this embodiment, the 10 curves obtained are as Figure 2 shown.

[0090] Construct a training sample set: {(x i , y i ) | x i = {d i1 , d i2 , d i3 , d i4 , d i5 , d i6 , d i7 , d i8} ∈ R 8 , y i ∈ R, i = 1, 2,..., M}, where x i represents the sequence composed of the distances between the i-th sample curve S i and all the slab bending template curve data in the standard template library, d ij represents the distance between the i-th sample curve S i and the j-th slab bending template curve data in the standard template library, j = 1, 2, 3, 4, 5, 6, 7, 8; y i represents the slab bending mode type corresponding to the i-th sample curve S i , and M represents the total number of samples;

[0091] In this embodiment, Figure 2 the distance calculation results between the 10 curves shown in and the curves corresponding to 8 standard template library functions are as shown in Table 1 below:

[0092] Table 1 Distance calculation results

[0093]

[0094]

[0095] In this embodiment, 2000 slab bending curves are used as training samples.

[0096] Using the Gaussian function map the data samples to a high-dimensional feature space, and construct a classification function based on the SVM model in the high-dimensional space: where ω and b are the regression parameters of the SVM model;

[0097] Initialize the parameters C and g of the SVM model, and express the parameter solution of the SVM model as the following constrained optimization problem:

[0098]

[0099] Satisfy the constraint ε i ≥0; where ε i is a slack variable, and the calculation formula is represents the Gaussian function corresponding to x i ;

[0100] Train the SVM model based on the training sample set, optimize the SVM model according to the training results, save the model parameters with the highest score after the optimized training, and obtain the machine learning model.

[0101] Furthermore, the optimizing the SVM model according to the training results includes:

[0102] Adopt the K-fold cross-validation method to optimize the parameters C and g of the SVM model, including: evenly divide the training sample set into K groups, 1 group as the verification data sample, and the remaining K-1 groups as the training data samples, and each group of data takes turns as the verification data sample; in each verification, take the parameter C and g in the given value range according to a certain step size for value combination, the step size of parameter C is (C max -C min ) / K, the step size of parameter g is (g max -g min ) / K; where C max , C min , g max , g min are the maximum and minimum values of parameters C and g respectively; under each group of parameter (C, g) combinations, perform K calculations respectively, and obtain the average value of the model test accuracies of the K calculations as the score score of this group (C, g):

[0103]

[0104] Among them, f p is the number of correct classifications for each category, and S is the number of samples in each test sample set.

[0105] Specifically, in this embodiment, K = 5, C max = 10, C min = 0.1, g max = 0.1, g min = 0.001; After tuning calculation, the optimal (C, g) result is (1, 0.01), and the corresponding score score = 0.975.

[0106] Embedding the above-trained model into the slab bending quality evaluation system can obtain the bending form of each slab online and record it in the database. Among them, in this embodiment, Figure 2 For the 10 curves in, the classification results by the slab bending mode classification method of this embodiment are shown in Table 2 below:

[0107] Table 2 Curve Classification Results

[0108] Model Output Corresponding Classification Curve 1 6 Tail Drive Side "L" Bend Curve 2 5 Tail Operation Side "L" Bend Curve 3 6 Tail Drive Side "L" Bend Curve 4 1 Operation Side "C" Bend Curve 5 2 Drive Side "C" Bend Curve 6 2 Drive Side "C" Bend Curve 7 2 Drive Side "C" Bend Curve 8 4 Head Drive Side "L" Bend Curve 9 5 Tail Operation Side "L" Bend Curve 10 5 Tail Operation Side "L" Bend

[0109] In summary, this embodiment provides a slab bending mode classification method based on machine learning. After calculating the distance between the bending curve and the standard template library function, instead of judging by the minimum distance to achieve the classification of the bending form, the distances between the bending curve and all functions in the sample library are used as the input of the machine learning model, and through the training of the model, the classification of the slab bending mode is achieved. Compared with the traditional judgment method based on the minimum distance, the method of this embodiment greatly improves the classification accuracy.

[0110] Second Embodiment

[0111] This embodiment provides a slab bending mode classification device based on machine learning, including:

[0112] A slab full-length centerline curve data acquisition module, configured to acquire the original data of the slab bending measured by the detection instrument corresponding to the slab to be classified, and preprocess the acquired original data to obtain the slab full-length centerline curve data corresponding to the slab to be classified;

[0113] A curve distance calculation module, configured to calculate the distances between the slab full-length centerline curve data corresponding to the slab to be classified obtained by the slab full-length centerline curve data acquisition module and each slab bending template curve data in a preset standard template library; wherein, the standard template library includes slab bending template curve data corresponding to various different types of slab bending forms;

[0114] A slab bending mode classification module is configured to calculate the distance between the slab full-length centerline curve data corresponding to the slab to be classified calculated by the curve distance calculation module and all slab bending template curve data in the standard template library, and use it as the input of a preset machine learning model. Through the machine learning model, the classification of the slab bending mode of the slab to be classified is realized.

[0115] The slab bending mode classification device based on machine learning in this embodiment corresponds to the slab bending mode classification method based on machine learning in the above first embodiment; among them, the functions realized by each functional module in the slab bending mode classification device based on machine learning in this embodiment correspond one by one to each process step in the slab bending mode classification method based on machine learning in the first embodiment; therefore, it will not be elaborated here.

[0116] The third embodiment

[0117] This embodiment provides an electronic device, which includes a processor and a memory; among them, at least one instruction is stored in the memory, and the instruction is loaded and executed by the processor to implement the method of the first embodiment.

[0118] This electronic device may vary greatly due to configuration or performance differences, and may include one or more processors (central processing units, CPUs) and one or more memories. Among them, at least one instruction is stored in the memory, and the instruction is loaded and executed by the processor to perform the following steps:

[0119] The fourth embodiment

[0120] This embodiment provides a computer-readable storage medium, in which at least one instruction is stored, and the instruction is loaded and executed by the processor to implement the above method. Among them, the computer-readable storage medium may be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc. The instructions stored therein can be loaded and executed by the processor in the terminal to perform the following steps:

[0121] In addition, it should be noted that the present invention can be provided as a method, a device or a computer program product. Therefore, the embodiments of the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0122] Embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing terminal device generate a device for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0123] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks. These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, such that a series of operation steps are executed on the computer or other programmable terminal device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable terminal device provide steps for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0124] It should also be noted that, in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. The term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or terminal device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or terminal device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the existence of additional identical elements in the process, method, article or terminal device comprising the element.

[0125] Finally, it should be noted that the above description is the preferred embodiment of the present invention. It should be pointed out that although the preferred embodiments of the present invention have been described, for those skilled in the art of this technology, once the basic creative concept of the present invention is known, several improvements and refinements can be made without departing from the principle described in the present invention. These improvements and refinements should also be regarded as the protection scope of the present invention. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A method for classifying slab bending patterns based on machine learning, characterized in that, Including: Obtain the original data of the slab bending measured by the inspection instrument corresponding to the slab to be classified, and preprocess the obtained original data to obtain the slab full-length centerline curve data corresponding to the slab to be classified; Calculate the distances between the slab full-length centerline curve data corresponding to the slab to be classified and each slab bending template curve data in the preset standard template library respectively; wherein, the standard template library includes slab bending template curve data corresponding to various different types of slab bending forms; Use the calculated distances between the slab full-length centerline curve data corresponding to the slab to be classified and all slab bending template curve data in the standard template library as the input of the preset machine learning model, and through the machine learning model, realize the classification of the slab bending mode of the slab to be classified; The machine learning model is a support vector machine SVM model; The construction process of the machine learning model includes: Obtain the hot-rolled slab full-length centerline curve as sample data; Construct a training sample set: {(x i , y i ) | x i = {d i1 , d i2 , d i3 , d i4 , d i5 , d i6 , d i7 , d i8} ∈ R 8 , y i ∈ R, i = 1, 2,..., M}, where x i represents the sequence composed of the distances between the i-th sample curve S i and all the slab bending template curve data in the standard template library, d ij represents the distance between the i-th sample curve S i and the j-th slab bending template curve data in the standard template library, j = 1, 2, 3, 4, 5, 6, 7, 8; y i represents the slab bending mode type corresponding to the i-th sample curve S i , and M represents the total number of samples; Using the Gaussian function Map the data samples to a high-dimensional feature space and construct a classification function based on the SVM model in the high-dimensional space: where ω and b are the regression parameters of the SVM model; Initialize the parameters C and g of the SVM model, and represent the parameter solution of the SVM model as the following constrained optimization problem: Meet the constraints where ε i is a slack variable, and its calculation formula is represents the Gaussian function corresponding to x i ; Train the SVM model based on the training sample set, optimize the SVM model according to the training results, save the model parameters with the highest score after optimized training, and obtain the machine learning model; The optimizing the SVM model according to the training results includes: Optimize the parameters C and g of the SVM model using the K-fold cross-validation method, including: evenly dividing the training sample set into K groups, using 1 group as the validation data sample, and the remaining K - 1 groups as the training data samples, and each group of data is used as the validation data sample in turn; in each validation, take the parameter C and g to take value combinations at a certain step size within the given value range, the step size of the parameter C is (C max -C min ) / K, and the step size of the parameter g is (g max -g min ) / K; where C max , C min , g max , g min are the maximum and minimum values of the parameters C and g respectively; under each group of parameter (C, g) combinations, perform K calculations respectively, and take the mean of the model test accuracies of the K calculations as the score of the model under this group (C, g): where f p is the number of correct classifications for each category, and S is the number of samples in each test set.

2. The method for classifying slab bending patterns based on machine learning according to claim 1, characterized in that, The preprocessing of the obtained original data includes: Normalize the horizontal and vertical coordinates of the obtained original data to [0, 1], and perform linear interpolation; wherein, the number of data points after interpolation is n, and the value range of n is 150 - 200.

3. The method for classifying slab bending patterns based on machine learning according to claim 1, characterized in that, The standard template library includes: The slab bending template curve data of the "C" bend on the operator side, and its function expression is: y = -13.108x 6 +39.324x 5 -65.84x 4 +66.14x 3 -36.333x 2 +9.8171x The slab bending template curve data of the "C" bend on the drive side, and its function expression is: y = 13.108x 6 - 39.324x 5 + 65.84x 4 - 66.14x 3 + 36.333x 2 - 9.8171x + 1 The slab bending template curve data of the "L" bend on the operator side at the head, and its function expression is: y = -6.554x 6 +35.751x 5 -73.144x 4 +73.493x 3 -38.578x 2 +10.049x The slab bending template curve data of the "L" bend on the drive side at the head, and its function expression is: y = 6.554x 6 - 35.751x 5 + 73.144x 4 - 73.493x 3 + 38.578x 2 - 10.049x + 1 The slab bending template curve data of the "L" bend on the operator side at the tail, and its function expression is: y = -6.554x 6 +3.5726x 5 +7.3036x 4 -7.3533x 3 +2.2448x 2 -0.2322x + 1 The slab bending template curve data of the "L" bend on the drive side at the tail, and its function expression is: y = 6.554x 6 - 3.5726x 5 - 7.3036x 4 + 7.3533x 3 - 2.2448x 2 + 0.2322x The slab bending template curve data of the "S" bend on the drive side, and its function expression is: y = -16.089x 5 +40.224x 4 -40.423x 3 +20.411x 2 -5.1407x + 1 The slab bending template curve data of the "S" bend on the operator side, and its function expression is: y = 16.089x 5 - 40.224x 4 + 40.423x 3 - 20.411x 2 + 5.1407x Wherein, x represents the abscissa of the corresponding slab bending template curve, y represents the ordinate of the corresponding slab bending template curve, and the number of data points of each slab bending template curve is m, and the value range of m is 150 - 200.

4. The method for classifying slab bending patterns based on machine learning according to claim 1, characterized in that, The calculating the distances between the slab full-length centerline curve data corresponding to the slab to be classified and each slab bending template curve data in the preset standard template library respectively includes: Use the dynamic time warping DTW algorithm to calculate the distances between the slab full-length centerline curve data corresponding to the slab to be classified and each slab bending template curve data in the preset standard template library respectively.

5. A device for classifying slab bending patterns based on machine learning, characterized in that, Including: The slab full-length centerline curve data acquisition module is used to acquire the original data of the slab bending measured by the detection instrument corresponding to the slab to be classified, and preprocess the acquired original data to obtain the slab full-length centerline curve data corresponding to the slab to be classified; The curve distance calculation module is used to calculate the distances between the slab full-length centerline curve data corresponding to the slab to be classified obtained by the slab full-length centerline curve data acquisition module and each slab bending template curve data in the preset standard template library respectively; wherein, the standard template library includes slab bending template curve data corresponding to various different types of slab bending forms; The slab bending mode classification module is used to use the distances between the slab full-length centerline curve data corresponding to the slab to be classified calculated by the curve distance calculation module and all slab bending template curve data in the standard template library as the input of a preset machine learning model, and realize the classification of the slab bending mode of the slab to be classified through the machine learning model; The machine learning model is a support vector machine SVM model; The construction process of the machine learning model includes: Obtain the hot-rolled slab full-length centerline curve as sample data; Construct a training sample set: {(x i , y i ) | x i = {d i1 , d i2 , d i3 , d i4 , d i5 , d i6 , d i7 , d i8} ∈ R 8 , y i ∈ R, i = 1, 2,..., M}, where x i represents the sequence composed of the distances between the i-th sample curve S i and all slab bending template curve data in the standard template library, d ij represents the distance between the i-th sample curve S i and the j-th slab bending template curve data in the standard template library, j = 1, 2, 3, 4, 5, 6, 7, 8; y i represents the slab bending mode type corresponding to the i-th sample curve S i , and M represents the total number of samples; Using the Gaussian function Map the data samples to a high-dimensional feature space and construct a classification function based on the SVM model in the high-dimensional space: where ω and b are the regression parameters of the SVM model; Initialize the parameters C and g of the SVM model, and represent the parameter solution of the SVM model as the following constrained optimization problem: Meet the constraints where ε i is a slack variable, and the calculation formula is represents the Gaussian function corresponding to x i ; Train the SVM model based on the training sample set, optimize the SVM model according to the training results, save the model parameters with the highest score after the optimized training, and obtain the machine learning model; The optimizing the SVM model according to the training results includes: Optimize the parameters C and g of the SVM model using the K-fold cross-validation method, including: evenly dividing the training sample set into K groups, using 1 group as the validation data sample, and the remaining K - 1 groups as the training data samples, and each group of data is used as the validation data sample in turn; in each validation, take the parameter C and g to take value combinations at a certain step size within the given value range, the step size of the parameter C is (C max -C min ) / K, and the step size of the parameter g is (g max -g min ) / K; where C max , C min , g max , g min are the maximum and minimum values of the parameters C and g respectively; under each group of parameter (C, g) combinations, perform K calculations respectively, and take the mean of the model test accuracies of the K calculations as the score of the model under this group of (C, g): Among them, f p is the number of correct classifications for each category, and S is the number of samples in each test set.

Images