A method and system for predicting the quality of decarburized layer of a cord steel

Abstract

The application discloses a cord steel decarburization layer quality prediction method and system, and the method comprises the following steps: S1, acquiring sample data in a cord steel production process, wherein the sample data comprises scalar data and time series data, and the scalar data is lengthened to obtain multivariate time series data; S2, fitting the multivariate time series data into multivariate function type data by using a B-spline basis function; S3, projecting the multivariate function type data into a high-dimensional function type feature space by using a function type data kernel method; S4, calculating a function type data projection hyperplane that minimizes the intra-class distance and maximizes the inter-class distance based on the function type features in the high dimension; and S5, projecting each sample to obtain a final sample classification result. According to the application, the cord steel decarburization layer quality can be accurately and timely predicted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cord steel decarburization layer quality prediction, in particular to a cord steel decarburization layer quality prediction method and system. BACKGROUND

[0002] In the production process of cord steel, the detection of decarburization layer depth (decarburization layer quality) is usually carried out after the cord steel rod is offline, and usually needs to use a metallographic method and other detection methods, which has the characteristics of high cost and hysteresis, and cannot meet the real-time acquisition demand of modern steel enterprises for product quality. SUMMARY

[0003] The present application provides a cord steel decarburization layer quality prediction method and system to solve the problems existing in the prior art. The technical solution is as follows:

[0004] On the one hand, a cord steel decarburization layer quality prediction method is provided, comprising:

[0005] S1, obtaining sample data in the production process of cord steel, the sample data including scalar data and time series data, lengthening the scalar data to obtain multivariate time series data;

[0006] S2, fitting using B-spline basis function, fitting the multivariate time series data into multivariate functional data;

[0007] S3, projecting the multivariate functional data into a high-dimensional functional feature space using a functional data kernel method;

[0008] S4, based on the functional features in high dimensions, calculating a functional data projection hyperplane that minimizes the intra-class distance and maximizes the inter-class distance;

[0009] S5, projecting each sample to obtain the final sample classification result.

[0010] Optionally, in S1, the scalar data is lengthened to obtain multivariate time series data, specifically including:

[0011] The scalar data is converted into a constant function with a domain size of the heating furnace length, and is spliced with the time series data in the variable dimension to obtain the multivariate time series data.

[0012] Optionally, S2 specifically includes:

[0013] For functional data fitting, the function x obtained after fitting can be expressed as the inner product of the coefficient vector and the basis function: Where c is the coefficient vector, T represents matrix transposition, is the basis function.

[0014] The coefficient vector c is determined using the least squares method. In this case, the weighted fitting error term SMSSE(y) for the j-th variable at the i-th observation point is calculated. i,j |c i,j )=(y i,j -Ψ j c i,j ) T W i,j (y i,j -Ψ j c i,j ),in, Let D be the numerical matrix of the basis functions of the j-th variable. j The number of basis functions used when performing functional data fitting for the j-th variable. L is the l-th basis function used when fitting functional data of the j-th variable. j W represents the observation time length of the j-th variable. i,j Let be the weight matrix of the j-th curve for the i-th sample. It is defined as the inverse of the covariance matrix obtained by calculating the fitting residuals of all observation points of the curve of the j-th variable for the i-th sample.

[0015] Functional data fitting based on the least squares method loses some smoothness, so a roughness penalty term needs to be introduced to obtain the least squares fitting error PENSSE with the penalty term. m (y i,j |c i,j )=(y i,j -Ψ j c i,j ) T W i,j (y i,j -Ψ j c i,j )+λc i,j T R j c i,j , where R j It is a weight matrix that reflects the relationship between variables. The coefficient vector c is solved using the least squares method. i,j =(Ψ) j T W i,j Ψ j +λR j ) -1 Ψ j T W i,j y i,j , where λ is the smoothness coefficient.

[0016] Optionally, S3 specifically includes:

[0017] The selected functional Gaussian radial basis kernel function k er The multivariate functional data is projected to a high-dimensional space through a high-dimensional mapping, so that the linearly inseparable low-dimensional multivariate functional data is converted into linearly separable multivariate functional data in the high-dimensional space, and the inner product of the multivariate functional data in the high-dimensional space can be expressed as k er (x1(t),x2(t))=exp(||x1(t)-x2(t)|| 2 / S 2 ), wherein S is a parameter of the kernel function, and ||x1(t)-x2(t)|| 2 is expressed as:

[0018]

[0019] Therefore, the final functional kernel function k er can be expressed as the operation of the basis function .

[0020] Optionally, the S4 specifically includes:

[0021] The problem of minimizing the intra-class distance and maximizing the inter-class distance is converted into the problem of maximizing the Rayleigh quotient of multivariate functional data wherein is an inter-class scatter matrix, is an intra-class scatter matrix, p is the number of classes of the processed data, μ is a class center of one of the classes, N i is the number of samples in the ith class, and x i,j (t) is the jth multivariate function in the ith sample, and the Rayleigh quotient maximum value corresponds to ω, which is a high-dimensional functional projection hyperplane that minimizes the intra-class distance and maximizes the inter-class distance.

[0022] Since the multivariate functional data is linearly separable after being projected to the high-dimensional feature space, the high-dimensional functional projection hyperplane ω is converted into a linear representation of the high-dimensional multivariate functional data at this time wherein φ(x)=[φ1(x),φ2(x),...,φ I (x)] T represents the projection relationship from the low-dimensional space to the high-dimensional space, and α=[α1,α2,...,α I ] T is an I×1-dimensional projection coefficient vector, I is the total number of samples, and the projection of φ(x) in the high-dimensional space to the functional projection hyperplane ω is expressed as φ(x)ω=φ(x)φ(x) T α=k er α, so solving the Rayleigh quotient maximum value corresponding to ω is converted into solving the α problem.

[0023] By using the transformed form of high-dimensional functional projection hyperplane, and by using the functional data kernel function to transform the inner product of multivariate functional data under high dimension, the Rayleigh quotient of functional data is transformed into wherein,

[0024] is the jth sample of the kth class, k er (x i (t), x j (t)) is a kernel function, K i is an IxN i matrix, is a square matrix whose each element is , and for the element K i [a][b] in the ath row and bth column of K i , K er [a][b] = k a (x b (t), x i (t));

[0025] Solving the Rayleigh quotient is further transformed into solving the optimization problem At this time, using the Lagrange multiplier method, and taking the derivative of a can obtain M a = dN a, after solving the eigenvalue d and the eigenvector a, the functional projection hyperplane that can distinguish the pass and fail of the decarburized layer to the greatest extent is obtained.

[0026] On the other hand, a cord steel decarburized layer quality prediction system is provided, the system comprising:

[0027] An acquisition module is configured to acquire sample data in a cord steel production process, the sample data comprising scalar data and time series data, and to lengthen the scalar data to obtain multivariate time series data;

[0028] A fitting module is configured to use B-spline basis functions to fit the multivariate time series data as multivariate functional data;

[0029] A mapping module is configured to use a functional data kernel method to map the multivariate functional data to a high-dimensional functional feature space;

[0030] A calculation module is configured to calculate a functional data projection hyperplane that minimizes the intra-class distance and maximizes the inter-class distance based on the functional features under high dimension;

[0031] A projection module is configured to project each sample to obtain a final sample classification result.

[0032] Optionally, the acquisition module is specifically configured to:

[0033] The scalar data is converted into a constant function with a domain size of a furnace duration of the heating furnace, and is spliced with the time series data in a variable dimension to obtain the multivariate time series data.

[0034] Optionally, the fitting module is specifically configured to:

[0035] For functional data fitting, the function x obtained after fitting can be expressed as the inner product of the coefficient vector and the basis function: Wherein, c is the coefficient vector, T represents the matrix transpose, is the basis function.

[0036] For the determination of the coefficient vector c, the least square method is adopted, at this time, the weighted fitting error term SMSSE(y i,j |c i,j ) of the jth variable of the ith observation point is: i,j (y j -Ψ i,j c T ) i,j W i,j (y j -Ψ i,j c j ), wherein, is the basis function value matrix of the jth variable, D j is the number of basis functions used when the jth variable is fitted by functional data, is the lth basis function used when the jth variable is fitted by functional data, L i,j is the observation time length of the jth variable, W m is the jth curve weight matrix of the ith sample, which is defined as the inverse of the covariance matrix calculated from the fitting residuals of all observation points of the jth variable curve of the ith sample.

[0037] The functional data fitting based on the least square method will lose a certain degree of smoothness, so a roughness penalty term needs to be introduced to obtain the least square fitting error PENSSE i,j (y i,j |c i,j )=(y j -Ψ i,j c T ) i,j W i,j (y j -Ψ i,j c i,j )+λc T R j c i,j , wherein R jis a weight matrix embodying the relationship between variables, and the coefficient vector c is solved by using the least square method i,j = (Ψ j T W i,j Ψ j +λR j ) -1 Ψ j T W i,j y i,j , wherein λ is a smoothing coefficient.

[0038] Optionally, the mapping module is specifically configured to:

[0039] select a functional Gaussian radial basis kernel function k er , and project the multivariate functional data to a high-dimensional space through a high-dimensional mapping, so as to convert the linearly inseparable low-dimensional multivariate functional data into linearly separable multivariate functional data in the high-dimensional space, and the inner product of the multivariate functional data in the high-dimensional space can be expressed as k er (x1(t),x2(t))=exp(||x1(t)-x2(t)|| 2 / S 2 ), wherein S is a parameter of the kernel function, and ||x1(t)-x2(t)|| 2 is expressed as:

[0040]

[0041] Therefore, the final functional kernel function k er can be expressed as an operation of the basis function .

[0042] Optionally, the calculation module is specifically configured to:

[0043] convert the problem of minimizing the intra-class distance and maximizing the inter-class distance into a problem of maximizing the Rayleigh quotient of multivariate functional data , wherein is an inter-class scatter matrix, is an intra-class scatter matrix, p is the number of classes of the processed data, μ is a class center of one of the classes, N i is the number of the i-th class of samples, and x i,j (t) is the j-th multivariate function in the i-th class of samples, and the Rayleigh quotient maximum value corresponds to ω, which is a high-dimensional functional projection hyperplane that minimizes the intra-class distance and maximizes the inter-class distance.

[0044] Since the multivariate functional data is linearly separable after being projected to the high-dimensional feature space, the high-dimensional functional projection hyperplane ω is converted into a linear representation of the high-dimensional multivariate functional data at this time. where, φ(x) = [φ1(x), φ2(x),...,φ I (x)] T denotes the projection relationship from low dimension to high dimension, α = [α1, α2,...,α I ] T is the projection coefficient vector of I x 1 dimension, I is the total number of samples, and the projection of φ(x) in the high-dimensional space to the functional projection hyperplane ω is denoted as φ(x)ω = φ(x)φ(x) T α = k er α, so solving the maximum value of the Rayleigh quotient corresponds to solving the α problem;

[0045] By using the transformed form of the high-dimensional functional projection hyperplane and using the functional data kernel function to transform the inner product of the multivariate functional data in the high-dimensional form, the functional data Rayleigh quotient is transformed into where,

[0046] is the jth sample of the kth class, k er (x i (t), x j (t)) is a kernel function, K i is a matrix of I x N i , is a square matrix whose each element is , and for the element K i [a][b] in the a row and b column of K i , K er [a][b] = k a (x b (t), x i (t));

[0047] Solving the Rayleigh quotient is further transformed into solving the optimization problem At this time, using the Lagrange multiplier method and taking the derivative of α can obtain Mα = δNα, after solving the eigenvalue δ and the eigenvector α, the functional projection hyperplane that can distinguish the pass and fail of the decarburized layer to the greatest extent is obtained.

[0048] On the other hand, an electronic device is provided, the electronic device comprising a processor and a memory, the memory storing at least one instruction, the at least one instruction being loaded and executed by the processor to implement the above cord steel decarburized layer quality prediction method.

[0049] On the other hand, a computer readable storage medium is provided, the storage medium storing at least one instruction, the at least one instruction being loaded and executed by a processor to implement the above cord steel decarburized layer quality prediction method.

[0050] The technical scheme provided by the present application has at least the following beneficial effects:

[0051] By using the present application, the quality of the decarburized layer of the cord steel can be accurately and timely predicted. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical scheme in the embodiments of the present application, the drawings needed in the following embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0053] Figure 1 is a flow chart of a cord steel decarburized layer quality prediction method provided by an embodiment of the present application;

[0054] Figure 2 is a schematic diagram of a scalar data lengthening method provided by an embodiment of the present application;

[0055] Figure 3 is a block diagram of a cord steel decarburized layer quality prediction system provided by an embodiment of the present application;

[0056] Figure 4 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0057] In order to make the technical problems, technical schemes and advantages of the present application more clear, the following will be described in detail with reference to the drawings and specific embodiments.

[0058] An embodiment of the present application provides a cord steel decarburized layer quality prediction method, which can be realized by an electronic device, which can be a terminal or a server. Figure 1 As shown in a flow chart of a cord steel decarburized layer quality prediction method, the processing flow of the method can include the following steps:

[0059] S1, sample data in a cord steel production process is obtained, the sample data includes scalar data and time series data, the scalar data is lengthened to obtain multivariate time series data;

[0060] The cord steel production process generally includes converter-refining-continuous casting-heating furnace-rolling process, the scalar data includes composition test data of converter, refining and continuous casting process, and the time series data includes process control data of continuous casting, heating furnace and rolling.

[0061] Optionally, in the S1, the scalar data is lengthened to obtain multivariate time series data, and specifically includes:

[0062] The scalar data is converted into a constant function with a domain size of a constant of the furnace duration (such as Figure 2 The scalar data C content is taken as a function with a constant value, that is, a straight line, and the length of the straight line is the furnace duration, and the time series data is spliced in the variable dimension to obtain the multivariate time series data.

[0063] The multivariate time series data is expressed in a matrix form, each row of the matrix represents a sample, and each column of the matrix represents each single variable time series data.

[0064] S2, fitting by using a B-spline basis function, fitting the multivariate time series data into multivariate function type data;

[0065] Optionally, the S2 specifically includes:

[0066] For function type data fitting, the function x obtained after fitting can be expressed as the inner product of the coefficient vector and the basis function: Wherein, c is the coefficient vector, T represents the matrix transpose, is the basis function (for example, the basis function is selected as a 5-order B-spline basis function with strong fitting capability);

[0067] For determination of the coefficient vector c, the least square method is used, at this time, the weighted fitting error term SMSSE(y i,j |c i,j ) of the jth variable of the ith observation point is (y i,j -Ψ j c i,j ) T W i,j (y i,j -Ψ j c i,j ), wherein, is the basis function numerical matrix of the jth variable, D j is the number of basis functions used for function type data fitting of the jth variable, is the lth basis function used for function type data fitting of the jth variable, L j is the observation time length of the jth variable, W i,j is the jth curve weight matrix of the ith sample, which is defined as the inverse of the covariance matrix calculated from the fitting residuals of all observation points of the jth variable curve of the ith sample;

[0068] The function type data fitting based on the least square method will lose a certain smoothness, so a roughness penalty term is introduced to obtain the least square fitting error PENSSE m (y i,j |ci,j ) = (y i,j - Ψ j c i,j ) T W i,j (y i,j - Ψ j c i,j )+ λc i,j T R j c i,j , where R j is a weight matrix embodying the relationship between variables, and the coefficient vector c i,j is solved by using the least square method j T W i,j Ψ j + λR j ) -1 Ψ j T W i,j y i,j , where λ is a smoothing coefficient.

[0069] The function type data fitting is used to convert each numerical element of the matrix into each function element.

[0070] S3, the multivariate function type data is projected into a high-dimensional function type feature space by using a function type data kernel method.

[0071] Optionally, the S3 specifically includes:

[0072] A function type Gaussian radial basis kernel function k er is selected, and the multivariate function type data is projected into a high-dimensional space through a high-dimensional mapping, so that the linearly inseparable low-dimensional multivariate function type data is converted into linearly separable multivariate function type data in the high-dimensional space, and the inner product of the multivariate function type data in the high-dimensional space can be expressed as k er (x1(t), x2(t)) = exp(||x1(t)-x2(t)|| 2 / S 2 ), where S is a parameter of the kernel function, and ||x1(t)-x2(t)|| 2 is expressed as:

[0073]

[0074] Therefore, the final function type kernel function k er can be expressed as an operation of the basis function .

[0075] S4, based on the function type characteristics under high dimension, calculate the function type data projection hyperplane which makes the minimum intra-class distance and the maximum inter-class distance;

[0076] Optionally, the S4 specifically comprises:

[0077] Convert the minimum intra-class distance and the maximum inter-class distance problem into the maximum Rayleigh quotient problem of multivariate function type data Wherein, is the inter-class scatter matrix, is the intra-class scatter matrix, p is the number of classes of the processed data (the number of classes of the embodiment of the application is 2, two classes of qualified and unqualified), μ is the class center of one of the classes, N i is the number of the i-th class sample, x i,j (t) is the j-th multivariate function in the i-th class sample, and the ω corresponding to the maximum Rayleigh quotient is the high-dimensional function type projection hyperplane which makes the minimum intra-class distance and the maximum inter-class distance;

[0078] Since the multivariate function type data is linearly separable after being projected into the high-dimensional characteristic space, the high-dimensional function type projection hyperplane ω is converted into the linear representation of the high-dimensional multivariate function type data at this time Wherein, φ(x)=[φ1(x),φ2(x),...,φ I (x)] T Indicates the projection relationship from low dimension to high dimension, α=[α1,α2,...,α I ] T is the I×1-dimensional projection coefficient vector, I is the total number of samples, and the projection of φ(x) in the high-dimensional space to the function type projection hyperplane ω is expressed as φ(x)ω=φ(x)φ(x) T α=k er α, so solving the ω corresponding to the maximum Rayleigh quotient is converted into solving the α problem;

[0079] The transformed form of the high-dimensional function type projection hyperplane is used, and the inner product of the multivariate function type data under the high-dimensional form is transformed by using the kernel function of the function type data, at this time the function type data Rayleigh quotient is converted into Wherein,

[0080] is the j-th sample of the k-th class, k er (x i (t),x j (t)) is the kernel function, K i is the I×N i matrix, is the square matrix whose each element is , and K iThe element K in the a-th row and b-th column i [a][b]=k er (x a (t),x b i (t));

[0081] The solving of the Rayleigh quotient is further transformed into solving an optimization problem At this time, the Lagrange multiplier method is used, and derivation is performed on alpha to obtain M alpha = delta N alpha, after solving the eigenvalue delta and the eigenvector alpha, a functional projection hyperplane capable of distinguishing whether the decarburized layer is qualified or not to the greatest extent is obtained.

[0082] S5, projecting each sample to obtain a final sample classification result.

[0083] According to the form of the high-dimensional functional projection hyperplane, each sample is projected, the Euclidean distances of the samples from the centers of various categories are calculated, and the category with the minimum Euclidean distance is selected as the final belonging category of the sample.

[0084] As Figure 3 shown, the embodiment of the application further provides a cord steel decarburized layer quality prediction system, the system comprises:

[0085] An acquisition module 310 is configured to acquire sample data in a cord steel production process, the sample data comprising scalar data and time series data, and lengthen the scalar data to obtain multivariate time series data.

[0086] A fitting module 320 is configured to fit the multivariate time series data into multivariate functional data by using a B-spline basis function.

[0087] A mapping module 330 is configured to map the multivariate functional data into a high-dimensional functional feature space by using a functional data kernel method.

[0088] A calculation module 340 is configured to calculate a functional data projection hyperplane that minimizes the intra-class distance and maximizes the inter-class distance based on the functional features in the high dimension.

[0089] A projection module 350 is configured to project each sample to obtain a final sample classification result.

[0090] Optionally, the acquisition module is specifically configured to:

[0091] The scalar data is converted into a constant function with a domain size being the heating furnace length in the furnace, and is spliced with the time series data in the variable dimension to obtain the multivariate time series data.

[0092] Optionally, the fitting module is specifically configured to:

[0093] For functional data fitting, the fitted function x can be expressed as the inner product of the coefficient vector and the basis function: where c is the coefficient vector, T represents matrix transpose, is the basis function;

[0094] For the determination of the coefficient vector c, the least square method is adopted, at this time, the weighted fitting error term SMSSE(y i,j |c i,j ) of the jth variable of the ith observation point is: i,j j i,j T i,j i,j j i,j is the basis function value matrix of the jth variable, D j is the number of basis functions used for functional data fitting of the jth variable, is the lth basis function used for functional data fitting of the jth variable, L j is the length of observation of the jth variable, W i,j is the jth curve weight matrix of the ith sample, which is defined as the inverse of the covariance matrix calculated from the fitting residuals of all observation points of the jth variable curve of the ith sample;

[0095] The functional data fitting based on the least square method will lose a certain degree of smoothness, so a roughness penalty term needs to be introduced to obtain the least square fitting error PENSSE m (y i,j |c i,j ) of the ith observation point: i,j j i,j T i,j i,j j i,j i,j T j i,j j is the weight matrix reflecting the relationship between variables, and the coefficient vector c i,j is solved by the least square method: j T i,j j j -1 j ​​​​​​​​​​​​​​​​​​​​​​​​T W i,j y i,j where λ is a smoothing coefficient.

[0096] Optionally, the mapping module is specifically configured to:

[0097] a functional Gaussian radial basis kernel function k er The multivariate functional data is projected to a high-dimensional space through a high-dimensional mapping, so that the linearly inseparable low-dimensional multivariate functional data is converted into linearly separable multivariate functional data in the high-dimensional space, and the inner product of the multivariate functional data in the high-dimensional space can be expressed as k er (x1(t),x2(t))=exp(||x1(t)-x2(t)|| 2 / S 2 ), where S is a parameter of the kernel function, and ||x1(t)-x2(t)|| 2 is expressed as:

[0098]

[0099] Therefore, the final functional kernel function k er can be expressed as an operation of the basis function .

[0100] Optionally, the computing module is specifically configured to:

[0101] convert the problem of minimizing the intra-class distance and maximizing the inter-class distance into a problem of maximizing the Rayleigh quotient of multivariate functional data where is an inter-class scatter matrix, is an intra-class scatter matrix, p is the number of classes of the processed data, μ is a class center of one of the classes, N i is the number of the i-th class of samples, and x i,j (t) is the j-th multivariate function in the i-th class of samples, and the Rayleigh quotient maximum value corresponds to ω, which is a high-dimensional functional projection hyperplane that minimizes the intra-class distance and maximizes the inter-class distance.

[0102] Since the multivariate functional data is linearly separable after being projected to the high-dimensional feature space, the high-dimensional functional projection hyperplane ω is converted into a linear representation of the high-dimensional multivariate functional data at this time where φ(x)=[φ1(x),φ2(x),...,φ I (x)] T indicates the projection relationship from low-dimensional to high-dimensional, and α=[α1,α2,...,α I ] Tis the projection coefficient vector of Ix1 dimension, I is the total number of samples, and the functional projection hyperplane ω in the high-dimensional space is represented as T α=k er α, and solving the maximum value of the Rayleigh quotient corresponds to solving the problem of α.

[0103] By using the transformed form of the high-dimensional functional projection hyperplane and using the functional data kernel function to transform the inner product of the multivariate functional data in the high-dimensional form, the functional data Rayleigh quotient is transformed as wherein,

[0104] is the jth sample of the kth class, k er (x i (t), x j (t)) is a kernel function, K i is an IxN i matrix, is a square matrix whose each element is , and for the element K i [a][b] in the ath row and bth column of K i , K er [a][b]=k a (x b (t), x i (t));

[0105] Solving the Rayleigh quotient is further transformed into solving the optimization problem At this time, using the Lagrange multiplier method, and taking the derivative of α can obtain Mα=δNα, after solving the eigenvalue δ and the eigenvector α, the functional projection hyperplane capable of distinguishing the pass or fail of the decarburized layer to the greatest extent is obtained.

[0106] The function structure of the cord steel decarburized layer quality prediction system provided by the embodiment of the application corresponds to the cord steel decarburized layer quality prediction method provided by the embodiment of the application, and will not be described here.

[0107] Figure 4 is a structural schematic diagram of an electronic device 400 provided by the embodiment of the application, the electronic device 400 can have relatively large differences due to different configurations or performances, and can include one or more than one processor (central processing units, CPUs) 401 and one or more than one memory 402, wherein the memory 402 stores at least one instruction, the at least one instruction is loaded and executed by the processor 401 to realize the steps of the above-mentioned cord steel decarburized layer quality prediction method.

[0108] In the example embodiment, a computer readable storage medium, such as a memory including instructions executable by a processor in a terminal to perform the above-described method of predicting the quality of the decarburized layer of the cord steel, is also provided. For example, the computer readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.

[0109] Those of ordinary skill in the present art can understand that all or part of the steps of the above-described embodiments can be completed by hardware, or by a program instructing relevant hardware, and the program can be stored in a computer readable storage medium, such as a read-only memory, a magnetic disk, or an optical disk.

[0110] The above description is merely preferred embodiments of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method of predicting the mass of the decarburized layer of a cord steel, characterized in that, The method comprises: S1, acquiring sample data in a steel cord production process, the sample data comprising scalar data and time series data, lengthening the scalar data to obtain multivariate time series data, the steel cord production process comprising: converter-refining-continuous casting-heating furnace-rolling process, the scalar data comprising component assay data of converter, refining, and continuous casting processes, and the time series data comprising process control data of continuous casting, heating furnace, and rolling; S2, fitting the multivariate time series data into multivariate functional data by using a B-spline basis function; S3, projecting the multivariate functional data into a high-dimensional functional feature space by using a functional data kernel method; S4, calculating a functional data projection hyperplane that minimizes the intra-class distance and maximizes the inter-class distance based on the functional features in the high dimension; S5, projecting each sample to obtain a final sample classification result; In S1, the scalar data is lengthened to obtain the multivariate time series data, specifically comprising: The scalar data is converted into a constant function with a domain size of the heating furnace in-furnace time length, and is spliced with the time series data in the variable dimension to obtain the multivariate time series data; In S2, specifically comprising: For functional data fitting, the function x obtained after fitting is represented as The inner product of the coefficient vector and the basis function: where c is the coefficient vector, T denotes matrix transposition, is the basis function; The coefficient vector c is determined using the least squares method. In this case, the weighted fitting error term SMSSE(y) for the j-th variable at the i-th observation point is calculated. i,j |c i,j )=(y i,j -Ψ j c i,j ) T W i,j (y i,j -Ψ j c i,j ),in, Let D be the numerical matrix of the basis functions of the j-th variable. j The number of basis functions used when performing functional data fitting for the j-th variable. The l-th basis function used when performing functional data fitting for the j-th variable, L j W represents the observation time length of the j-th variable. i,j Let be the weight matrix of the j-th curve for the i-th sample. It is defined as the inverse of the covariance matrix obtained by calculating the fitting residuals of all observation points of the curve of the j-th variable for the i-th sample. Functional data fitting based on the least squares method loses some smoothness, so a roughness penalty term needs to be introduced to obtain the least squares fitting error PENSSE with the penalty term. m (y i,j |c i,j )=(y i,j -Ψ j c i,j ) T W i,j (y i,j -Ψ j c i,j )+λc i,j T R j c i,j , where R j It is a weight matrix that reflects the relationship between variables. The coefficient vector c is solved using the least squares method. i,j =(Ψ) j T W i,j Ψ j +λR j ) -1 Ψ j T W i,j y i,j , where λ is the smoothness coefficient; In S4, specifically comprising: The problem of minimum intra-class distance and maximum inter-class distance is converted into a problem of maximum Rayleigh quotient of multivariate function type data wherein, is an inter-class scatter matrix, is an intra-class scatter matrix, p is the number of classes of the data processed, μ is the class center of one of the classes, N k is the number of samples of the kth class, x k,i′ is the function of the ith sample in the kth class, and the Rayleigh quotient maximum value corresponds to ω, which is a high-dimensional function type projection hyperplane that minimizes the intra-class distance and maximizes the inter-class distance. Since the multivariate functional data is linearly separable after being projected into the high dimensional feature space, the high dimensional functional projection hyperplane ω is transformed into the linear representation of the high dimensional multivariate functional data where φ(x) = [φ1(x), φ2(x),...,φ I (x)] T represents the projection relationship from low dimension to high dimension, α = [α1, α2,...,α I ] T is the I × 1 dimensional projection coefficient vector, I is the total number of samples, and the projection of φ(x) in the high dimensional space to the functional projection hyperplane ω is represented as φ(x)ω = φ(x)φ(x) T α = k er α, so solving the maximum value of the Rayleigh quotient corresponding to ω is transformed into solving the α problem; By using the transformed form of high-dimensional functional projection hyperplane, and by using the functional data kernel function to transform the inner product of the multivariate functional data under high-dimensional form, the Rayleigh quotient of the functional data is transformed into where, is the ith sample of the kth class, is the kernel function, K k is the I x N k matrix, is the I x I matrix, for K k the element in the ath row and bth column, K k [a][b] = k er (x a (t), x b k (t)); The solution of the Rayleigh quotient is further transformed into the solution of an optimization problem At this time, using the Lagrange multiplier method, and the derivative of α can be obtained Mα=δNα, after solving the eigenvalue δ and eigenvector α, the functional projection hyperplane that can distinguish the pass and fail of the decarburized layer to the greatest extent is obtained.

2. The method of claim 1, wherein, In S3, specifically comprising: The selected functional Gaussian radial basis kernel function k er The multivariate functional data is projected to a high-dimensional space through a high-dimensional mapping, so that the linearly inseparable low-dimensional multivariate functional data is converted to linearly separable multivariate functional data in the high-dimensional space, and the inner product of the multivariate functional data in the high-dimensional space is represented as k er (x1(t),x2(t))=exp(||x1(t)-x2(t)|| 2 / S 2 ), wherein S is a parameter of the kernel function, and ||x1(t)-x2(t)|| 2 is represented as: Thus, the final functional kernel function k er is expressed as a basis function operation.

3. A cord steel decarburized layer mass prediction system characterized by, The system comprises: An acquisition module configured to acquire sample data in a steel cord production process, the sample data comprising scalar data and time series data, lengthen the scalar data to obtain multivariate time series data, and the steel cord production process comprising: converter-refining-continuous casting-heating furnace-rolling process, the scalar data comprising component assay data of converter, refining, and continuous casting processes, and the time series data comprising process control data of continuous casting, heating furnace, and rolling; A fitting module configured to fit the multivariate time series data into multivariate functional data by using a B-spline basis function; A mapping module configured to map the multivariate functional data into a high-dimensional functional feature space by using a functional data kernel method; A calculation module configured to calculate a functional data projection hyperplane that minimizes the intra-class distance and maximizes the inter-class distance based on the functional features in the high dimension; A projection module configured to project each sample to obtain a final sample classification result; The acquisition module is specifically configured to: convert the scalar data into a constant function with a domain size of the heating furnace in-furnace time length, and splice the constant function with the time series data in the variable dimension to obtain the multivariate time series data; The fitting module is specifically configured to: For functional data fitting, the function x obtained after fitting is represented as The inner product of the coefficient vector and the basis function: where c is the coefficient vector, T denotes matrix transposition, is the basis function; The coefficient vector c is determined using the least squares method. In this case, the weighted fitting error term SMSSE(y) for the j-th variable at the i-th observation point is calculated. i,j |c i,j )=(y i,j -Ψ j c i,j ) T W i,j (y i,j -Ψ j c i,j ),in, Let D be the numerical matrix of the basis functions of the j-th variable. j The number of basis functions used when performing functional data fitting for the j-th variable. The l-th basis function used when performing functional data fitting for the j-th variable, L j W represents the observation time length of the j-th variable. i,j Let be the weight matrix of the j-th curve for the i-th sample. It is defined as the inverse of the covariance matrix obtained by calculating the fitting residuals of all observation points of the curve of the j-th variable for the i-th sample. The functional data fitting based on least square method will lose some smoothness, so a roughness penalty term is introduced to get the penalized least square fitting error PENSSE m (y i,j -Ψ i,j c i,j ) j i,j T W i,j (y i,j -Ψ j c i,j )+λc i,j T R j c i,j , where R is the weight matrix embodying the relationship between variables, and the coefficient vector c is solved by least square method j i,j = (Ψ j T W i,j Ψ j +λR j ) -1 Ψ j T W i,j y i,j , where λ is the smoothness coefficient​​​ The calculation module is specifically configured to: The problem of minimum intra-class distance and maximum inter-class distance is converted into a problem of maximum Rayleigh quotient of multivariate function type data wherein, is an inter-class scatter matrix, is an intra-class scatter matrix, p is the number of classes of the data processed, μ is the class center of one of the classes, N k is the number of samples of the kth class, x k,i′ (t) is the function of the ith sample in the kth class, and the Rayleigh quotient maximum value corresponds to ω, which is a high-dimensional function type projection hyperplane that minimizes the intra-class distance and maximizes the inter-class distance. Since the multivariate functional data is linearly separable after being projected into the high dimensional feature space, the high dimensional functional projection hyperplane ω is transformed into the linear representation of the high dimensional multivariate functional data where φ(x) = [φ1(x), φ2(x),...,φ I (x)] T denotes the projection relationship from low dimensional to high dimensional, α = [α1, α2,...,α I ] T is the I × 1 dimensional projection coefficient vector, I is the total number of samples, and the projection of φ(x) in the high dimensional space to the functional projection hyperplane ω is denoted as φ(x)ω = φ(x)φ(x) T α = k er α, so solving the maximum value of the Rayleigh quotient corresponding to ω is transformed into solving the α problem; By using the transformed form of high-dimensional functional projection hyperplane and using the functional data kernel function to transform the inner product of multivariate functional data under high-dimensional form, the Rayleigh quotient of functional data is transformed into where, is the ith sample of the kth class, is the kernel function, K k is the I x N k matrix, is the I x I matrix, for the K k element in the ath row and bth column, K k [a][b] = k er (x a (t), x b k (t)); The solution of the Rayleigh quotient is further transformed into the solution of an optimization problem At this time, using the Lagrange multiplier method, and the derivative of α can be obtained Mα = δNα, after solving the eigenvalue δ and eigenvector α, the functional projection hyperplane that can distinguish the pass and fail of the decarburized layer to the greatest extent is obtained.

4. The system of claim 3, wherein, The mapping module is specifically configured to: The selected function type Gaussian radial basis kernel function k er The multivariate function type data is projected to a high-dimensional space through a high-dimensional mapping, so that the linearly inseparable low-dimensional multivariate function type data is converted to linearly separable multivariate function type data in the high-dimensional space, and the inner product of the multivariate function type data in the high-dimensional space is represented as k er (x1(t),x2(t))=exp(||x1(t)-x2(t)|| 2 / S 2 ), wherein S is a parameter of the kernel function, and ||x1(t)-x2(t)|| 2 is represented as: Thus, the final functional kernel function k er is expressed as a basis function operation.

Images