CN112037209A - Steel plate roller wear loss prediction method and system - Google Patents

Steel plate roller wear loss prediction method and system Download PDF

Info

Publication number
CN112037209A
CN112037209A CN202010908672.8A CN202010908672A CN112037209A CN 112037209 A CN112037209 A CN 112037209A CN 202010908672 A CN202010908672 A CN 202010908672A CN 112037209 A CN112037209 A CN 112037209A
Authority
CN
China
Prior art keywords
sample
steel plate
neural network
training
hidden layer
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010908672.8A
Other languages
Chinese (zh)
Inventor
王龙
冀秀梅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Shanghai for Science and Technology
Original Assignee
University of Shanghai for Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Shanghai for Science and Technology filed Critical University of Shanghai for Science and Technology
Priority to CN202010908672.8A priority Critical patent/CN112037209A/en
Publication of CN112037209A publication Critical patent/CN112037209A/en
Pending legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/0002Inspection of images, e.g. flaw detection
    • G06T7/0004Industrial image inspection
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Computing arrangements based on biological models using neural network models
    • G06N3/04Architectures, e.g. interconnection topology
    • G06N3/0454Architectures, e.g. interconnection topology using a combination of multiple neural nets
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Computing arrangements based on biological models using neural network models
    • G06N3/08Learning methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20081Training; Learning
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20084Artificial neural networks [ANN]
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30108Industrial image inspection
    • G06T2207/30136Metal

Abstract

The invention relates to a method and a system for predicting the abrasion loss of a steel plate roller, and relates to the field of steel rolling. According to the method, training sample data corresponding to influence factors influencing rolling kilometers in the sample data is determined through a grey correlation degree analysis method, so that the input variable of a neural network is optimized; training the neural network of the extreme learning machine by using the training sample data, establishing a neural network model, improving the prediction precision of the rolling kilometers, predicting the rolling kilometers of each set of working roll by using the trained neural network model, and predicting the wear loss of the working roll by using the predicted rolling kilometers, so that the prediction deviation of the wear loss of the working roll is reduced, and the defect of plate shape caused by excessive wear of the working roll is avoided.

Description

Steel plate roller wear loss prediction method and system
Technical Field
The invention relates to the field of steel rolling, in particular to a method and a system for predicting wear loss of a steel plate roller.
Background
At present, the production scheduling method of the medium plate mainly comprises the steps that based on experience, a worker estimates the abrasion loss of a roller according to the total rolling tonnage in combination with the specific steel plate variety and specification floating, and then plans and schedules the roller and the roller changing period. In the rolling process, a roller is worn, fig. 1 is a working roller wear curve diagram, the wear amount of a working roller is a key parameter for plate shape control, the wear of the working roller can cause roll gap deviation between two sides and the middle of the working roller, and the roll gap balance needs to be adjusted through a bending roller to keep a certain roller convexity. When the abrasion loss is too large, a larger roll bending force is needed to maintain a certain convexity, and if the limit of the roll bending force is reached, the shape of the plate is out of control. The existing medium plate scheduling method has large influence of human factors, so that the deviation between the estimated roller wear loss and the actual wear loss is large, and the shape of a medium plate rolled at the last stage of a roller is poor.
Disclosure of Invention
The invention aims to provide a method and a system for predicting the abrasion loss of a steel plate roller.
In order to achieve the purpose, the invention provides the following scheme:
a method for predicting the abrasion loss of a steel plate roller comprises the following steps:
acquiring sample data; the sample data comprises parameter data of a sample steel plate and the number of rolled kilometers of the sample; the parameters of the sample steel plate comprise a blank size and a target size;
determining training sample data by a grey correlation degree analysis method according to the sample data; the training sample data comprises: influence factor data influencing the rolling kilometers in the parameter data and the sample rolling kilometers of the sample steel plate;
taking the influence factor data of the sample steel plate as input, taking the sample rolling kilometers of the sample steel plate as output, and training a neural network of a limit learning machine to obtain a trained neural network model;
acquiring influence factor data of a steel plate to be predicted;
inputting the influence factor data of the steel plate to be predicted into the neural network model to obtain the rolling kilometer number of the steel plate to be predicted;
acquiring the abrasion loss of the roller sample corresponding to the sample data;
determining the relationship between the abrasion loss of the roller and the rolling kilometers by utilizing a regression analysis method according to the abrasion loss of the roller sample and the rolling kilometers of the sample;
and obtaining the predicted roller wear amount by utilizing the relationship between the roller wear amount and the rolling kilometers according to the rolling kilometers of the steel plate to be predicted.
Optionally, the determining training sample data by a gray correlation analysis method according to the sample data specifically includes:
taking the rolled kilometer number of the sample as a reference number series, and taking the parameter data as a comparison number series;
carrying out non-dimensionalization processing on the reference number sequence and the comparison number sequence to obtain a reference non-quantitative number sequence and a comparison non-quantitative number sequence;
calculating a difference series of the reference scalar series and the comparison scalar series;
calculating a maximum difference and a minimum difference of the reference scalar number series and the comparison scalar number series using the difference number series;
calculating a gray correlation coefficient of each element in the comparison scalar series with the reference scalar series using the maximum difference and the minimum difference;
respectively calculating the grey correlation degree of each element in the comparison scalar number series and the reference scalar number series by using the grey correlation coefficient;
determining elements in the comparative scalar number series corresponding to the gray relevance degrees larger than the preset gray relevance degrees as influence factors influencing rolling kilometers, wherein data corresponding to the influence factors in the sample data are the influence factor data.
Optionally, the method includes the steps of taking influence factor data of the sample steel plate as input, taking a sample rolling kilometer number of the sample steel plate as output, training a neural network of a limit learning machine, and obtaining a trained neural network model, and specifically includes:
determining the number of hidden layer nodes of the neural network of the extreme learning machine by using ten-fold cross validation by taking the influence factor data of the sample steel plate as input and the number of rolled kilometers of a sample of the sample steel plate as output;
and determining the output weight of the neural network of the extreme learning machine by using the number of nodes of the hidden layer to obtain a neural network model.
Optionally, the method for determining the number of hidden layer nodes of the neural network of the extreme learning machine by using the influence factor data of the sample steel plate as input and the number of rolled kilometers of the sample steel plate as output and using cross validation includes:
acquiring a preset range of the node number of the hidden layer;
sequentially taking all the nodes in the preset range of the number of the nodes as hidden layer training nodes of the neural network of the extreme learning machine;
taking influence factor data of the sample steel plate as input, taking the number of rolled kilometers of a sample of the sample steel plate as output, training the neural network of the extreme learning machine corresponding to the number of training nodes of each hidden layer by using cross-folding cross validation, and training preset training times to obtain a well-trained neural network training model of cross-folding cross validation of each time corresponding to the number of training nodes of each hidden layer;
calculating the root mean square error of each neural network training model corresponding to the number of the hidden layer training nodes;
calculating the mean value of the root mean square errors of the extreme learning machine neural network corresponding to each hidden layer training node number by using all the root mean square errors corresponding to each hidden layer training node number and the preset training times;
and comparing the root mean square error average values of the extreme learning machine neural network corresponding to all the nodes in the preset range of the node number, and determining the node number corresponding to the minimum root mean square error average value as the hidden layer node number of the extreme learning machine neural network.
Optionally, the method includes the steps of taking influence factor data of the sample steel plate as input, taking a sample rolling kilometer number of the sample steel plate as output, determining output weight of the extreme learning machine neural network by using the number of hidden layer nodes, and obtaining a neural network model, and specifically includes:
the number of nodes of a hidden layer of the extreme learning machine neural network is the number of nodes of the hidden layer, and the number of nodes of an input layer of the extreme learning machine neural network is the number of influence factors in the influence factor data;
taking the influence factor data of the sample steel plate as the input of the extreme learning machine neural network, taking the sample rolling kilometer number of the sample steel plate as the output of the extreme learning machine neural network, and solving the following formula to determine the output weight of the extreme learning machine neural network to obtain a trained neural network model;
wherein L represents the total number of hidden layer nodes, i is 1, 2. Beta is aiAn output weight representing the number of ith hidden layer nodes; g () represents an activation function; wiInput weights, W, representing the number of ith hidden layer nodesi=[wi1,wi2,…,wim']T,wim'Representing the input weight of the connection of the mth input layer node number and the ith hidden layer node number, wherein m 'represents the node number of the input layer, and m' is 5; xkDenotes an input variable, X, of the k-th steel platek=[xk1,xk2,…,xkm']T,xkm'An input variable representing the number of nodes of the m' th input layer corresponding to the kth sample steel plate; wi·XkRepresents WiAnd XkInner product of (d); biA bias representing the number of ith hidden layer nodes; t is tkRepresenting the sample rolling kilometers of a sample steel plate; n represents the total number of sample steel plates.
A steel plate roll wear amount prediction system comprising:
the sample data acquisition module is used for acquiring sample data; the sample data comprises parameter data of a sample steel plate and the number of rolled kilometers of the sample; the parameters of the sample steel plate comprise a blank size and a target size;
the training sample data determining module is used for determining training sample data through a grey correlation degree analysis method according to the sample data; the training sample data comprises: influence factor data influencing the rolling kilometers in the parameter data and the sample rolling kilometers of the sample steel plate;
the neural network training module is used for training the neural network of the extreme learning machine by taking the influence factor data of the sample steel plate as input and the sample rolling kilometers of the sample steel plate as output to obtain a trained neural network model;
the influence factor data acquisition module is used for acquiring influence factor data of the steel plate to be predicted;
the rolling kilometer number prediction module is used for inputting the influence factor data of the steel plate to be predicted into the neural network model to obtain the rolling kilometer number of the steel plate to be predicted;
the roller sample abrasion loss acquisition module is used for acquiring the roller sample abrasion loss corresponding to the sample data;
the roller abrasion loss and rolling kilometer number relation determining module is used for determining the relation between the roller abrasion loss and the rolling kilometer number by utilizing a regression analysis method according to the roller sample abrasion loss and the sample rolling kilometer number;
and the roller wear amount prediction module is used for obtaining the predicted roller wear amount by utilizing the relationship between the roller wear amount and the rolling kilometers according to the rolling kilometers of the steel plate to be predicted.
Optionally, the training sample data determining module specifically includes:
a reference comparison sequence determining unit, configured to use the sample rolling kilometer number as a reference sequence, and use the parameter data as a comparison sequence;
the non-dimensionalization processing unit is used for carrying out non-dimensionalization processing on the reference number sequence and the comparison number sequence to obtain a reference non-dimensionalization number sequence and a comparison non-dimensionalization number sequence;
a difference series calculation unit for calculating a difference series of the reference scalar series and the comparison scalar series;
a maximum-minimum difference calculation unit for calculating a maximum difference and a minimum difference between the reference scalar number sequence and the comparison scalar number sequence using the difference number sequence;
a gray correlation coefficient calculation unit for calculating a gray correlation coefficient of each element in the comparison scalar number series with the reference scalar number series using the maximum difference and the minimum difference;
the grey correlation degree calculating unit is used for calculating the grey correlation degree of each element in the comparison scalar number series and the reference scalar number series respectively by utilizing the grey correlation coefficient;
and the training sample data determining unit is used for determining elements in the comparative unamount number series corresponding to the gray correlation degree which is greater than the preset gray correlation degree as influence factors influencing the rolling kilometers, and the data corresponding to the influence factors in the sample data are the influence factor data.
Optionally, the neural network training module specifically includes:
the hidden layer node number determining unit is used for determining the hidden layer node number of the neural network of the extreme learning machine by using the influence factor data of the sample steel plate as input and the sample rolling kilometer number of the sample steel plate as output and using cross validation;
and the output weight determining unit is used for determining the output weight of the neural network of the extreme learning machine by taking the influence factor data of the sample steel plate as input and the sample rolling kilometer number of the sample steel plate as output and utilizing the number of nodes of the hidden layer to obtain a neural network model.
Optionally, the hidden layer node number determining unit specifically includes:
a node number preset range obtaining subunit, configured to obtain a node number preset range of the hidden layer;
a hidden layer training node number determining subunit, configured to sequentially use all the node numbers within the preset node number range as hidden layer training node numbers of the extreme learning machine neural network;
the training subunit is used for taking the influence factor data of the sample steel plate as input, taking the number of rolled kilometers of a sample of the sample steel plate as output, training the extreme learning machine neural network corresponding to each hidden layer training node number by using cross-folding cross validation, training preset training times, and obtaining a trained neural network training model of each cross-folding cross validation corresponding to each hidden layer training node number;
the root mean square error calculation subunit is used for calculating the root mean square error of each neural network training model corresponding to the number of the hidden layer training nodes;
the root mean square error average value calculating operator unit is used for calculating the root mean square error average value of the extreme learning machine neural network corresponding to each hidden layer training node number by using all the root mean square errors corresponding to each hidden layer training node number and the preset training times;
and the comparison subunit is used for comparing the root mean square error average values of the extreme learning machine neural network corresponding to all the nodes within the preset node number range, and determining the node number corresponding to the minimum root mean square error average value as the hidden layer node number of the extreme learning machine neural network.
Optionally, the output weight determining unit specifically includes:
the node number determining subunit is used for determining the number of nodes of a hidden layer of the extreme learning machine neural network as the number of nodes of the hidden layer, and determining the number of nodes of an input layer of the extreme learning machine neural network as the number of influence factors in the influence factor data;
the output weight determining subunit is used for taking the influence factor data of the sample steel plate as the input of the extreme learning machine neural network, taking the sample rolling kilometer number of the sample steel plate as the output of the extreme learning machine neural network, and solving the following formula to determine the output weight of the extreme learning machine neural network to obtain a trained neural network model;
wherein L represents the total number of hidden layer nodes, i is 1, 2. Beta is aiAn output weight representing the number of ith hidden layer nodes; g () represents an activation function; wiInput weights, W, representing the number of ith hidden layer nodesi=[wi1,wi2,…,wim']T,wim'Representing the input weight of the connection of the mth input layer node number and the ith hidden layer node number, wherein m 'represents the node number of the input layer, and m' is 5; xkDenotes an input variable, X, of the k-th steel platek=[xk1,xk2,…,xkm']T,xkm'An input variable representing the number of nodes of the m' th input layer corresponding to the kth sample steel plate; wi·XkRepresents WiAnd XkInner product of (d); biA bias representing the number of ith hidden layer nodes; t is tkRepresenting the sample rolling kilometers of a sample steel plate; n represents the total number of sample steel plates.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention provides a method and a system for predicting wear loss of a steel plate roller. The method comprises the following steps: acquiring sample data; the sample data comprises parameter data of a sample steel plate and the number of rolled kilometers of the sample; parameters of the sample steel plate include a blank size and a target size; determining training sample data by a grey correlation degree analysis method according to the sample data; the training sample data includes: influence factor data influencing the rolling kilometers in the parameter data and the sample rolling kilometers of the sample steel plate; taking the influence factor data of the sample steel plate as input, taking the sample rolling kilometers of the sample steel plate as output, and training the neural network of the extreme learning machine to obtain a trained neural network model; acquiring influence factor data of a steel plate to be predicted; inputting the influence factor data of the steel plate to be predicted into a neural network model to obtain the rolling kilometer number of the steel plate to be predicted; acquiring a roller sample abrasion loss corresponding to the sample data; determining the relationship between the abrasion loss of the roller and the rolling kilometers by using a regression analysis method according to the abrasion loss of the roller sample and the rolling kilometers of the sample; and obtaining the predicted roller wear amount according to the rolling kilometers of the steel plate to be predicted by utilizing the relationship between the roller wear amount and the rolling kilometers. According to the method, training sample data corresponding to influence factors influencing rolling kilometers in the sample data is determined through a grey correlation degree analysis method, so that the input variable of a neural network is optimized; training the neural network of the extreme learning machine by using the training sample data, establishing a neural network model, and improving the prediction precision of rolling kilometers; the trained neural network model is used for predicting the rolling kilometers of each set of working roll, and then the predicted rolling kilometers are used for predicting the abrasion loss of the working roll, so that the prediction deviation of the abrasion loss of the roll is reduced, and the defect of plate shape caused by excessive abrasion of the working roll is avoided.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a graph of work roll wear;
FIG. 2 is a flowchart of a method for predicting wear of a roll of a steel plate according to an embodiment of the present invention;
FIG. 3 is a linear fit relationship diagram of rolling tonnage and work roll wear loss provided by the embodiment of the present invention;
FIG. 4 is a graph showing a linear fit relationship between rolling kilometers and work roll wear according to an embodiment of the present invention;
FIG. 5 is a block diagram of an ELM neural network provided by an embodiment of the present invention;
fig. 6 is a structural diagram of a steel plate roll wear amount prediction system according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a method and a system for predicting the abrasion loss of a steel plate roller.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
The present embodiment provides a method for predicting wear loss of a steel plate roll, fig. 2 is a flowchart of the method for predicting wear loss of a steel plate roll according to the embodiment of the present invention, and referring to fig. 2, the method for predicting wear loss of a steel plate roll includes:
step 101, obtaining sample data; the sample data comprises parameter data of a sample steel plate and the number of rolled kilometers of the sample; parameters of the sample steel plate include the billet size, target size, and number of rolling passes. The blank dimensions include: slab thickness, slab width and slab length, the target size includes: a target thickness of steel sheet, a target width of steel sheet, and a target length of steel sheet.
Generally, the data which can be obtained in the on-site rolling schedule table of the steel plate factory are the number of rolling passes of each steel plate and the rolling length of the pass, and the total rolling length of each steel plate cannot be directly obtained, so the cumulative total rolling length of each steel plate, namely the rolling kilometer number of each steel plate, is obtained by summing the number of the steel plates and the rolling length of each pass by using an excel data perspective method. In this embodiment, 109585 actual production data are continuously collected for 3 months in a single-stand non-rotating steel full-longitudinal rolling medium plate mill, and subjected to data perspective summation processing, so as to obtain sample data of 9842 steel plates. Since the production line selected in this embodiment is the non-rotating steel rolling, the slab width is substantially the same as the target width of the steel plate, and therefore only the slab width is considered.
The present embodiment utilizes a regression analysis method to calculate the relationship between the roll wear loss and the rolling kilometer number and the relationship between the roll wear loss and the rolling tonnage through actual data (sample data), and obtains the relationship between the roll wear loss and the rolling kilometer number as follows: the roll wear (um) is-0.105-5.886 rolling kilometers (km), and the relationship between the roll wear and the rolling tonnage is as follows: the abrasion loss (um) of the roller is-32.06-0.06867 rolling tonnage (t). Referring to fig. 3 and 4, the fitting degree of the linear relation between the abrasion loss of the working roll and the rolling kilometers is higher than that of the working roll, so that the abrasion loss of the rolling roll is more reasonable to estimate by replacing the rolling tonnage with the rolling kilometers. The rolling kilometers refer to the sum of all rolling passes of all steel plates in the rolling process in one roll changing period, and the rolling kilometers of one steel plate are predicted in the embodiment. The standard error S of a fitting straight line of the relation between the abrasion loss of the roller and the rolling tonnage, namely the standard deviation of a residual error is 30.9043, the goodness of fit R-Sq is 68.1%, and the R-Sq (adjustment) is 68.1%. The standard error S of a fitting straight line of the relation between the abrasion loss of the roller and the rolling kilometers is 16.4878, the goodness of fit R-Sq is 90.9%, and the R-Sq (adjustment) is 90.9%. The standard error S represents the standard distance between a data value (point) and a fitted straight line, and for a given research, the smaller the S value is, the better the prediction effect of the regression model on the prediction variable is; the goodness of fit R-Sq refers to the degree of fit of a fitted straight line to a data value, the maximum value of the R-Sq is 1, and the closer the value of the R-Sq is to 1, the better the degree of fit of the fitted straight line to the data value is; conversely, a smaller R-Sq value indicates a poorer degree of fit of the fitted straight line to the data values. R-Sq (adjusted) refers to adjusted R-Sq, and the closer R-Sq (adjusted) is to R-Sq, the more reliable the regression model is. Therefore, the prediction effect of predicting the roller wear amount by using the rolling kilometers is better than that of predicting the roller wear amount by using the rolling tonnage, and the fitting degree of the fitting straight line of the relationship between the roller wear amount and the rolling kilometers to the roller wear amount data is better than that of the fitting straight line of the relationship between the roller wear amount and the rolling tonnage to the roller wear amount data.
Step 102, determining training sample data through a grey correlation degree analysis method according to the sample data; the training sample data includes: and influencing factor data influencing the rolling kilometers in the parameter data and the sample rolling kilometers of the sample steel plate. And analyzing and determining the influence degree of each factor in the parameters by combining a gray correlation analysis method based on Python programming, wherein the factors comprise: slab thickness, slab width, slab length, steel plate target thickness, steel plate target length, and rolling pass number.
Step 102 specifically includes:
and taking the rolling kilometers of the sample as a reference series and the parameter data as a comparison series.
The reference series is noted as: y ' (Y ' (1), Y ' (2), …, Y ' (k), …, Y ' (n))T,n=9842;
In the formula, Y' (k) represents the number of rolled sample kilometers of the kth sample steel plate, and k is 1, 2.
The comparison series is recorded as:
X′j=(X′j(1),X′j(2),…,X′j(k),…,X′j(n))T,j=1,2,…m,m=6,n=9842;
in the formula, Xj' (k) denotes parameter data of a kth sample steel plate; j represents a factor number in the parameters, and j is 1,2,3,4,5 and 6 respectively represents the slab thickness, the slab width, the slab length, the steel plate target thickness, the steel plate target length and the rolling pass number; m represents the number of factors in the parameter.
And carrying out non-dimensionalization processing on the reference number sequence and the comparison number sequence to obtain a reference non-quantitative number sequence and a comparison non-quantitative number sequence. And carrying out non-dimensionalization on the reference number sequence and the comparison number sequence, and adopting initialization processing, namely dividing each number in the number sequence by the first number to obtain a new number sequence.
Reference to the infinite series:
comparison of the scalar series:
the non-dimensionalized data sequence forms a new matrix as follows:
wherein m is 6, n is 9842.
The difference series Δ D of the reference and comparison scalar series is calculated. Calculating the absolute difference value of corresponding elements of the comparison sequence and the reference sequence, namely:
ΔD=|Y(k)-Xj(k)|,k=1,2,…n,j=1,2,…m,m=6,n=9842
the difference series is used to calculate the maximum difference and the minimum difference of the reference scalar series and the comparison scalar series. Calculating a maximum value, and determining a maximum difference delta max and a minimum difference delta min:
and calculating and comparing the gray correlation coefficient of each element in the scalar number series with the reference scalar number series by using the maximum difference and the minimum difference. Calculating and comparing the jth factor of the kth sample steel plate in the scalar sequence with the gray correlation coefficient eta of the reference scalar sequence by using the maximum difference and the minimum differencej(k)。
Wherein k is 1,. n, n is 9842; j-1, 2,. m, m-6; ρ is a resolution coefficient, where 0< ρ <1, the smaller ρ is, the larger the difference between the gray correlation coefficients is, the stronger the discrimination capability is, and ρ is usually 0.5.
Using ashAnd the color correlation coefficient respectively calculates and compares the gray correlation degree of each element in the scalar number series with the reference scalar number series. Calculating the association sequence, namely calculating the mean value of the association coefficients of the elements corresponding to the reference scalar sequence respectively for the comparison scalar sequence to reflect the association relationship between the comparison scalar sequence and the reference scalar sequence to obtain the grey association degree r of each factor0jIt is written as:
wherein k is 1,. n, n is 9842; j-1, 2,. m, m-6.
In this embodiment, the gray correlation of each factor in the parameters is shown in table 1, regardless of the index weight.
TABLE 1 Grey correlation of factors
And determining elements in the comparative infinite number series corresponding to the gray correlation degree which is greater than the preset gray correlation degree as influence factors influencing the rolling kilometers, wherein data corresponding to the influence factors in the sample data are influence factor data. In this embodiment, the preset gray relevance is 0.8, and the element with the gray relevance greater than 0.8 is selected as an influencing factor, that is: slab thickness, slab width, slab length, steel plate target thickness, steel plate target length.
And 103, training the neural network of the extreme learning machine by taking the influence factor data of the sample steel plate as input and the sample rolling kilometers of the sample steel plate as output to obtain a trained neural network model. The input variable of an Extreme Learning Machine (ELM) neural network is influence factor data of each steel plate, and the output variable is the rolling kilometer number corresponding to each steel plate.
Step 103 specifically comprises:
and determining the number of hidden layer nodes of the neural network of the extreme learning machine by using the influence factor data of the sample steel plate as input and the number of rolled kilometers of the sample steel plate as output and using cross validation of ten folds. In order to ensure good generalization performance, it is important to determine the appropriate number of hidden layer nodes. The number of hidden layer nodes is too small, the network performance is poor, the number of hidden layer nodes is too large, the training time is long, and overfitting is easy to occur. Regarding the number of hidden layer nodes, because there is no scientific and unified determination method at present, in this embodiment, on the premise that the number of hidden layer nodes is smaller than the number of training samples-1 to ensure the generalization capability of the neural network of the extreme learning machine, a ten-fold cross validation method is adopted between 30 and 300 number of hidden layer nodes, that is, a data set of training samples is divided into 10 parts, 9 parts of the data set are taken as training data in turn, 1 part of the data set is taken as test data, and the test is performed, and the cycle is sequentially performed until all training sample data are selected once. The method specifically comprises the following steps:
and acquiring a preset range of the node number of the hidden layer. The number of nodes is within a preset range of 30-300.
And sequentially taking all the nodes in the preset range of the node number as hidden layer training node numbers of the neural network of the extreme learning machine.
And taking the influence factor data of the sample steel plate as input, taking the sample rolling kilometer number of the sample steel plate as output, training the extreme learning machine neural network corresponding to each hidden layer training node number by using ten-fold cross validation, and training preset training times to obtain a neural network training model which is trained by each ten-fold cross validation and corresponds to each hidden layer training node number. And (3) taking the influence factor data of the sample steel plate as a data set of the training sample, dividing the data set into 10 parts, taking 9 parts as training data and 1 part as test data in turn, inputting the test data into the extreme learning machine neural network corresponding to the number of hidden layer training nodes for testing, and circulating in sequence until all the training sample data are selected once. The preset training times are 20 times, and each training is performed by adopting ten-fold cross validation. In this embodiment, ten-fold cross validation training is performed on different hidden layer training node numbers, so that each hidden layer training node number needs to be trained 20 times, and each time ten-fold cross validation training corresponding to training can obtain 1 neural network training model, that is, 20 neural network training models can be obtained correspondingly by training each hidden layer training node number.
And calculating the root mean square error of each neural network training model corresponding to each hidden layer training node number. In the embodiment, the root mean square error RMSE is selected as an evaluation index of the neural network training model. The root mean square error, RMSE, is expressed as:
wherein n is1=1,2,...,N1,N1Represents the total amount of training data or test data entered;is n th1The predicted value of the neural network training model corresponding to the input data,is n th1The real value of the rolling kilometer number corresponding to the input data.
The calculating of the root mean square error of the neural network training model obtained by the ten-fold cross validation training specifically comprises the following steps: inputting training data of ten-fold cross validation training into an RMSE expression to obtain a root mean square error RMSE1, inputting test data corresponding to the training data into the RMSE expression to obtain a root mean square error RMSE2, and according to a weight formula: and (3) obtaining the final RMSE by using the RMSE (0.48 RMSE1+0.6 RMSE 2), and using the final RMSE as an evaluation index of a neural network training model corresponding to the cross-validation training.
And calculating the mean value of the root mean square errors of the neural network of the extreme learning machine corresponding to each hidden layer training node number by using all the root mean square errors corresponding to each hidden layer training node number and preset training times.
And comparing the root mean square error average values of the extreme learning machine neural network corresponding to all the nodes in the preset range of the node number, and determining the node number corresponding to the minimum root mean square error average value as the hidden layer node number of the extreme learning machine neural network. In this embodiment, each hidden layer training node number is continuously trained for 20 times, and each training corresponds to one root mean square error, each hidden layer training node number corresponds to 20 root mean square errors, an average value of the 20 root mean square errors is taken, the node number corresponding to the minimum root mean square error average value in the root mean square error average values of all the hidden layer training node numbers is determined as the hidden layer node number of the extreme learning machine neural network, and finally the hidden layer node number selected in this embodiment is 60.
When determining the number of hidden layer nodes, the output weight of the neural network of the extreme learning machine is obtained by substituting the number of hidden layer training nodes into a formulaAnd obtaining the output weight of the neural network training model corresponding to the number of the training nodes of each hidden layer. Where L ' represents the total number of hidden layer training nodes, i ' is 1, 2.., L '; beta'i′An output weight representing the number of ith' hidden layer training nodes; g () represents an activation function; w'i′An input weight representing the number of ith' hidden layer training nodes; xkRepresenting input variables of the kth sample steel plate; w'i′·XkRepresents W'i′And XkInner product of (d); b'i′A bias representing the number of ith' hidden layer training nodes; t is tkThe rolling number of the sample steel sheet is expressed in kilometers. W'i′And b'i′And (4) randomly determining.
And determining the output weight of the neural network of the extreme learning machine by using the number of nodes of the hidden layer to obtain the neural network model. The method specifically comprises the following steps:
the number of nodes of the hidden layer of the neural network of the extreme learning machine is the number of nodes of the hidden layer, and the number of nodes of the input layer of the neural network of the extreme learning machine is the number of influence factors in the influence factor data. In this embodiment, the Sigmoid function is selected as the activation function of the hidden layer of the extreme learning machine neural network, so that the finally determined basic parameters of the extreme learning machine neural network are shown in table 2.
TABLE 2 basic parameters of ELM neural networks
And (3) taking the influence factor data of the sample steel plate as the input of the neural network of the extreme learning machine, taking the rolled kilometers of the sample steel plate as the output of the neural network of the extreme learning machine, and solving the following formula to determine the output weight of the neural network of the extreme learning machine so as to obtain a trained neural network model. The structure of the neural network model is shown in fig. 5, and the neural network model includes: the system comprises an input layer, a hidden layer and an output layer, wherein the slab thickness, the slab width, the slab length, the steel plate target thickness and the steel plate target length are input into the input layer, and the output layer outputs the predicted rolling kilometers.
Wherein L represents the total number of hidden layer nodes, i is 1, 2. Beta is aiAn output weight representing the number of ith hidden layer nodes; g () represents an activation function; wiInput weights, W, representing the number of ith hidden layer nodesi=[wi1,wi2,…,wim']T,wim'Representing the input weight of the connection of the mth input layer node number and the ith hidden layer node number, wherein m 'represents the node number of the input layer, and m' is 5; xkDenotes an input variable, X, of the k-th steel platek=[xk1,xk2,…,xkm']T,xkm'An input variable representing the number of nodes of the m' th input layer corresponding to the kth sample steel plate; wi·XkRepresents WiAnd XkInner product of (d); biA bias representing the number of ith hidden layer nodes; t is tkRepresenting the sample rolling kilometers of a sample steel plate; n represents the total number of sample steel plates. WiAnd biAnd (4) randomly determining.
The solving process of the output weight specifically includes:
the sample data is (X)k,Y′(k)),Xk=[xk1,xk2,…,xkm']T,XkThe k-th sample steel plate has m 'input variables, k is 1,2, 9842, and m' is 5; y' (k) represents the sample rolling kilometers of the kth sample steel sheet. For a single hidden layer neural network with 60 hidden layer nodes, it can be expressed as:
wherein g () is an activation function; wi=[wi1,wi2,…,wim']TI ═ 1,2, …,60, m ═ 5 as input weights; beta is aiAn output weight which is the number of the ith hidden layer node; biIs the bias of the ith hidden layer node number; o iskAnd the rolling kilometers obtained by calculation of the single hidden layer neural network are shown, namely the predicted rolling kilometers. Wi·XkRepresents WiAnd XkThe inner product of (d).
The goal of single-hidden-layer neural network learning is to minimize the output error, which can be expressed as:
wherein, OkTo predict the number of rolled kilometers, tkY' (k) is the sample rolling kilometer of the kth sample steel sheet. I.e. the presence of betai,WiAnd biSo thatThe matrix can be expressed as H β ═ T, where H is the output of the hidden layer node, β is the output weight, T is the desired output (sample rolled kilometers), and can be expressed as:
wherein β ═ β11,…,βL]T,T=[t1,t1,…,tn]T
When the number L of hidden layer nodes is less than the number n of samples, in the ELM algorithm, when W is randomly determinediAnd biThen, the output matrix H of the hidden layer is uniquely determined, and T is a known quantity, so that training the single hidden layer neural network can be converted into solving a linear system H β ═ T. By applying the generalized inverse method, the output weight β, β ═ H can be determined+T, where H + is the Moore-Penrose generalized inverse of the output matrix H. And the linear system H beta is T, the norm of the obtained solution is minimum and unique, namely the rolling kilometer error predicted by the ELM neural network is minimum.
And 104, acquiring influence factor data of the steel plate to be predicted.
And 105, inputting the influence factor data of the steel plate to be predicted into the neural network model to obtain the rolling kilometers of the steel plate to be predicted.
And 106, acquiring the abrasion loss of the roller sample corresponding to the sample data.
And step 107, determining the relationship between the abrasion loss of the roller and the rolling kilometers by using a regression analysis method according to the abrasion loss of the roller sample and the rolling kilometers of the sample. The relationship between the abrasion loss of the roller and the rolling kilometers in the embodiment is as follows: the abrasion loss of the roller is-0.105-5.886 rolling kilometers.
And 108, obtaining the predicted roller wear amount by utilizing the relationship between the roller wear amount and the rolling kilometers according to the rolling kilometers of the steel plate to be predicted.
The embodiment also provides a steel plate roll wear amount prediction system, and fig. 6 is a structural diagram of the steel plate roll wear amount prediction system provided by the embodiment of the invention. Referring to fig. 6, the steel plate roll wear amount prediction system includes:
a sample data obtaining module 201, configured to obtain sample data; the sample data comprises parameter data of a sample steel plate and the number of rolled kilometers of the sample; parameters of the sample steel plate include the billet size, target size, and number of rolling passes. The blank dimensions include: slab thickness, slab width and slab length, the target size includes: a target thickness of steel sheet, a target width of steel sheet, and a target length of steel sheet.
A training sample data determining module 202, configured to determine training sample data according to the sample data by using a gray correlation analysis method; the training sample data includes: and influencing factor data influencing the rolling kilometers in the parameter data and the sample rolling kilometers of the sample steel plate. And analyzing and determining the influence degree of each factor in the parameters by combining a gray correlation analysis method based on Python programming, wherein the factors comprise: slab thickness, slab width, slab length, steel plate target thickness, steel plate target length, and rolling pass number.
The training sample data determining module 202 specifically includes:
and the reference comparison sequence determining unit is used for taking the rolling kilometers of the sample as a reference sequence and taking the parameter data as a comparison sequence.
The reference series is noted as: y ' (Y ' (1), Y ' (2), …, Y ' (k), …, Y ' (n))T
In the formula, Y' (k) represents the number of rolled sample kilometers of the kth sample steel plate, k is 1, 2.
The comparison series is recorded as: xj′=(Xj′(1),Xj′(2),…,Xj′(k),…,Xj′(n))T
In the formula, Xj' (k) denotes parameter data of a kth sample steel plate; j represents a factor number in the parameters, and j is 1,2,3,4,5 and 6 respectively represents the slab thickness, the slab width, the slab length, the steel plate target thickness, the steel plate target length and the rolling pass number; m represents the number of factors in the parameter.
And the non-dimensionalization processing unit is used for carrying out non-dimensionalization processing on the reference number sequence and the comparison number sequence to obtain a reference non-dimensionalized number sequence and a comparison non-dimensionalized number sequence. Reference to the infinite series:comparison of the scalar series:
a difference sequence calculation unit for calculating a difference sequence Δ D of the reference and comparison scalar sequences; Δ D ═ y (k) -Xj(k)|。
And a maximum and minimum difference calculation unit for calculating a maximum difference and a minimum difference of the reference scalar sequence and the comparison scalar sequence using the difference sequence. Maximum differenceSum minimum difference
And the gray correlation coefficient calculating unit is used for calculating the gray correlation coefficient for comparing each element in the scalar number series with the reference scalar number series by using the maximum difference and the minimum difference. Calculating and comparing the jth factor of the kth sample steel plate in the scalar sequence with the gray correlation coefficient eta of the reference scalar sequence by using the maximum difference and the minimum differencej(k)。
Wherein k is 1,. n, n is 9842; j-1, 2,. m, m-6; ρ is a resolution coefficient, where 0< ρ <1, the smaller ρ is, the larger the difference between the gray correlation coefficients is, the stronger the discrimination capability is, and ρ is usually 0.5.
And the grey correlation degree calculating unit is used for respectively calculating and comparing the grey correlation degree of each element in the scalar number series with the grey correlation degree of the reference scalar number series by using the grey correlation coefficient. Respectively calculating the average value of the correlation coefficients of the elements corresponding to the reference scalar number series according to the comparison scalar number series to reflect the correlation between the comparison scalar number series and the reference scalar number series, and obtaining the grey correlation degree of each factor
And the training sample data determining unit is used for determining elements in the comparative invariant number series corresponding to the gray correlation degree which is greater than the preset gray correlation degree as influence factors influencing the rolling kilometers, and the data corresponding to the influence factors in the sample data are influence factor data. In this embodiment, the preset gray relevance is 0.8, and the element with the gray relevance greater than 0.8 is selected as an influencing factor, that is: slab thickness, slab width, slab length, steel plate target thickness, steel plate target length.
And the neural network training module 203 is used for training the neural network of the extreme learning machine by taking the influence factor data of the sample steel plate as input and the sample rolling kilometers of the sample steel plate as output to obtain a trained neural network model. The input variable of an Extreme Learning Machine (ELM) neural network is influence factor data of each steel plate, and the output variable is the rolling kilometer number corresponding to each steel plate.
The neural network training module 203 specifically includes:
and the hidden layer node number determining unit is used for determining the hidden layer node number of the neural network of the extreme learning machine by using the influence factor data of the sample steel plate as input and the sample rolling kilometer number of the sample steel plate as output and using cross validation. In this embodiment, on the premise that the number of hidden layer nodes is smaller than-1 of the number of training samples to ensure the generalization capability of the neural network of the extreme learning machine, a ten-fold cross validation method is adopted between 30 and 300 of the number of hidden layer nodes, that is, a data set of training samples is divided into 10 parts, 9 parts of the data set are taken as training data and 1 part of the data set is taken as test data in turn, and the test is performed and the cycle is performed in sequence until all the training sample data are selected once.
The hidden layer node number determining unit specifically includes:
and the node number preset range acquiring subunit is used for acquiring the node number preset range of the hidden layer. The number of nodes is within a preset range of 30-300.
And the hidden layer training node number determining subunit is used for sequentially using all the node numbers in the node number preset range as hidden layer training node numbers of the neural network of the extreme learning machine.
And the training subunit is used for taking the influence factor data of the sample steel plate as input, taking the sample rolling kilometer number of the sample steel plate as output, training the extreme learning machine neural network corresponding to each hidden layer training node number by utilizing cross-folding cross validation, training preset training times, and obtaining a well-trained neural network training model of each time of cross-folding cross validation corresponding to each hidden layer training node number.
And the root mean square error calculating subunit is used for calculating the root mean square error of each neural network training model corresponding to each hidden layer training node number. In the embodiment, the root mean square error RMSE is selected as an evaluation index of the neural network training model. The root mean square error, RMSE, is expressed as:
wherein n is1=1,2,...,N1,N1Represents the total amount of training data or test data entered;is n th1The predicted value of the neural network training model corresponding to the input data,is n th1The real value of the rolling kilometer number corresponding to the input data.
The root mean square error calculating subunit is used for inputting training data of ten-fold cross validation training into an RMSE expression to obtain a root mean square error RMSE1, inputting test data corresponding to the training data into the RMSE expression to obtain a root mean square error RMSE2, and according to a weight formula: RMSE 0.48 RMSE1+0.6 RMSE2 gave the final RMSE.
And the root mean square error average value calculating operator unit is used for calculating the root mean square error average value of the extreme learning machine neural network corresponding to each hidden layer training node number by utilizing all root mean square errors corresponding to each hidden layer training node number and preset training times.
And the comparison subunit is used for comparing the root mean square error average values of the extreme learning machine neural network corresponding to all the nodes within the preset node number range, and determining the node number corresponding to the minimum root mean square error average value as the hidden layer node number of the extreme learning machine neural network. In this embodiment, each hidden layer training node number is continuously trained for 20 times, and each training corresponds to one root mean square error, each hidden layer training node number corresponds to 20 root mean square errors, an average value of the 20 root mean square errors is taken, the node number corresponding to the minimum root mean square error average value in the root mean square error average values of all the hidden layer training node numbers is determined as the hidden layer node number of the extreme learning machine neural network, and finally the hidden layer node number selected in this embodiment is 60.
And the output weight determining unit is used for determining the output weight of the neural network of the extreme learning machine by taking the influence factor data of the sample steel plate as input and the sample rolling kilometer number of the sample steel plate as output and utilizing the number of nodes of the hidden layer to obtain the neural network model.
The output weight determining unit specifically includes:
and the node number determining subunit is used for determining the number of the nodes of the hidden layer of the extreme learning machine neural network as the number of the nodes of the hidden layer, and determining the number of the nodes of the input layer of the extreme learning machine neural network as the number of the influence factors in the influence factor data. In this embodiment, a Sigmoid function is selected for the activation function of the hidden layer of the extreme learning machine neural network, so that the finally determined basic parameters of the extreme learning machine neural network include: the number of nodes of the input layer is 5, the number of nodes of the hidden layer is 60, and the activation function is a Sigmoid function.
And the output weight determining subunit is used for taking the influence factor data of the sample steel plate as the input of the extreme learning machine neural network, taking the sample rolling kilometer number of the sample steel plate as the output of the extreme learning machine neural network, and solving the following formula to determine the output weight of the extreme learning machine neural network so as to obtain the trained neural network model.
Wherein L represents the total number of hidden layer nodes, and i is 1,2,...,L;βiAn output weight representing the number of ith hidden layer nodes; g () represents an activation function; wiInput weights, W, representing the number of ith hidden layer nodesi=[wi1,wi2,…,wim']T,wim'Representing the input weight of the connection of the mth input layer node number and the ith hidden layer node number, wherein m 'represents the node number of the input layer, and m' is 5; xkDenotes an input variable, X, of the k-th steel platek=[xk1,xk2,…,xkm']T,xkm'An input variable representing the number of nodes of the m' th input layer corresponding to the kth sample steel plate; wi·XkRepresents WiAnd XkInner product of (d); biA bias representing the number of ith hidden layer nodes; t is tkRepresenting the sample rolling kilometers of a sample steel plate; n represents the total number of sample steel plates. WiAnd biAnd (4) randomly determining.
And the influencing factor data acquisition module 204 is used for acquiring the influencing factor data of the steel plate to be predicted.
And the rolling kilometer number prediction module 205 is configured to input the influence factor data of the steel plate to be predicted into the neural network model, so as to obtain the rolling kilometer number of the steel plate to be predicted.
And a roller sample wear amount obtaining module 206, configured to obtain a roller sample wear amount corresponding to the sample data.
And the roller wear amount and rolling kilometer number relation determining module 207 is used for determining the relation between the roller wear amount and the rolling kilometer number by utilizing a regression analysis method according to the roller sample wear amount and the sample rolling kilometer number. The relationship between the abrasion loss of the roller and the rolling kilometers in the embodiment is as follows: the abrasion loss of the roller is-0.105-5.886 rolling kilometers.
And the roller wear amount prediction module 208 is configured to obtain the predicted roller wear amount according to the rolling kilometers of the steel plate to be predicted and by using the relationship between the roller wear amount and the rolling kilometers.
With the increasing demand of the market for product quality, the method of arranging production plans purely according to experience has been gradually lagged behind and is prone to deviation. Meanwhile, with the development of big data and intelligent manufacturing technology, scientific production scheduling becomes inevitable. The invention provides a steel plate roller abrasion loss prediction method and a steel plate roller abrasion loss prediction system based on a single hidden layer feedforward neural network ELM extreme learning machine algorithm, wherein firstly, according to the blank size, the target size and the rolling track times in actual historical data of a large number of rolling processes, influence factors influencing the rolling kilometers are determined by utilizing a gray level correlation analysis method, and input variables of the ELM neural network are optimized; optimizing the number of hidden layer nodes of the ELM neural network by utilizing ten-fold cross validation; finally, a high-precision rolling kilometer number prediction model is established through an ELM artificial neural network, the rolling kilometer number can be accurately predicted, the relation between the roller abrasion loss and the rolling kilometer number is determined by utilizing a regression analysis method, the roller change period is arranged by predicting the abrasion loss of the working roller by using the rolling kilometer number, the roller change period is closer to the real abrasion loss of the working roller, the defects of the traditional experience mode of estimating the roller change period according to the rolling tonnage and the variety specification are overcome, and the situation that the shape of the roller at the last stage is difficult to control due to excessive abrasion of the working roller caused by the arrangement of a production plan according to the traditional experience is avoided; the rolling kilometers are predicted through the neural network model, theoretical basis is laid for scientifically and reasonably arranging a production plan, and the production plan can be scientifically and reasonably arranged. Relevant researches show that compared with traditional neural networks such as BP (back propagation) neural network and RBF (radial basis function) neural network, the ELM has the advantage of higher learning speed, overcomes the problems of local optimization, low training speed and the like easily occurring in the traditional algorithm, and has better generalization capability.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.

Claims (10)

1. A method for predicting the wear amount of a roll of a steel plate, comprising:
acquiring sample data; the sample data comprises parameter data of a sample steel plate and the number of rolled kilometers of the sample; the parameters of the sample steel plate comprise a blank size and a target size;
determining training sample data by a grey correlation degree analysis method according to the sample data; the training sample data comprises: influence factor data influencing the rolling kilometers in the parameter data and the sample rolling kilometers of the sample steel plate;
taking the influence factor data of the sample steel plate as input, taking the sample rolling kilometers of the sample steel plate as output, and training a neural network of a limit learning machine to obtain a trained neural network model;
acquiring influence factor data of a steel plate to be predicted;
inputting the influence factor data of the steel plate to be predicted into the neural network model to obtain the rolling kilometer number of the steel plate to be predicted;
acquiring the abrasion loss of the roller sample corresponding to the sample data;
determining the relationship between the abrasion loss of the roller and the rolling kilometers by utilizing a regression analysis method according to the abrasion loss of the roller sample and the rolling kilometers of the sample;
and obtaining the predicted roller wear amount by utilizing the relationship between the roller wear amount and the rolling kilometers according to the rolling kilometers of the steel plate to be predicted.
2. The method for predicting the wear loss of the steel plate roll according to claim 1, wherein the determining training sample data by a grey correlation analysis method according to the sample data specifically comprises:
taking the rolled kilometer number of the sample as a reference number series, and taking the parameter data as a comparison number series;
carrying out non-dimensionalization processing on the reference number sequence and the comparison number sequence to obtain a reference non-quantitative number sequence and a comparison non-quantitative number sequence;
calculating a difference series of the reference scalar series and the comparison scalar series;
calculating a maximum difference and a minimum difference of the reference scalar number series and the comparison scalar number series using the difference number series;
calculating a gray correlation coefficient of each element in the comparison scalar series with the reference scalar series using the maximum difference and the minimum difference;
respectively calculating the grey correlation degree of each element in the comparison scalar number series and the reference scalar number series by using the grey correlation coefficient;
determining elements in the comparative scalar number series corresponding to the gray relevance degrees larger than the preset gray relevance degrees as influence factors influencing rolling kilometers, wherein data corresponding to the influence factors in the sample data are the influence factor data.
3. The method for predicting the abrasion loss of the steel plate roll according to claim 1, wherein the step of training the neural network of the limit learning machine by taking the data of the influencing factors of the sample steel plate as input and the number of rolled kilometers of the sample steel plate as output to obtain a trained neural network model specifically comprises the following steps:
determining the number of hidden layer nodes of the neural network of the extreme learning machine by using ten-fold cross validation by taking the influence factor data of the sample steel plate as input and the number of rolled kilometers of a sample of the sample steel plate as output;
and determining the output weight of the neural network of the extreme learning machine by using the number of nodes of the hidden layer to obtain a neural network model.
4. The method for predicting the abrasion loss of the steel plate roll according to claim 3, wherein the step of determining the number of hidden layer nodes of the extreme learning machine neural network by using the influence factor data of the sample steel plate as input and the number of sample rolled kilometers of the sample steel plate as output and using cross validation comprises the following specific steps:
acquiring a preset range of the node number of the hidden layer;
sequentially taking all the nodes in the preset range of the number of the nodes as hidden layer training nodes of the neural network of the extreme learning machine;
taking influence factor data of the sample steel plate as input, taking the number of rolled kilometers of a sample of the sample steel plate as output, training the neural network of the extreme learning machine corresponding to the number of training nodes of each hidden layer by using cross-folding cross validation, and training preset training times to obtain a well-trained neural network training model of cross-folding cross validation of each time corresponding to the number of training nodes of each hidden layer;
calculating the root mean square error of each neural network training model corresponding to the number of the hidden layer training nodes;
calculating the mean value of the root mean square errors of the extreme learning machine neural network corresponding to each hidden layer training node number by using all the root mean square errors corresponding to each hidden layer training node number and the preset training times;
and comparing the root mean square error average values of the extreme learning machine neural network corresponding to all the nodes in the preset range of the node number, and determining the node number corresponding to the minimum root mean square error average value as the hidden layer node number of the extreme learning machine neural network.
5. The method for predicting the abrasion loss of the steel plate roll according to claim 4, wherein the method for predicting the abrasion loss of the steel plate roll by using the influence factor data of the sample steel plate as input and the sample rolling kilometers of the sample steel plate as output determines the output weight of the extreme learning machine neural network by using the number of the hidden layer nodes to obtain the neural network model specifically comprises the following steps:
the number of nodes of a hidden layer of the extreme learning machine neural network is the number of nodes of the hidden layer, and the number of nodes of an input layer of the extreme learning machine neural network is the number of influence factors in the influence factor data;
taking the influence factor data of the sample steel plate as the input of the extreme learning machine neural network, taking the sample rolling kilometer number of the sample steel plate as the output of the extreme learning machine neural network, and solving the following formula to determine the output weight of the extreme learning machine neural network to obtain a trained neural network model;
wherein L represents the total number of hidden layer nodes, i is 1, 2. Beta is aiAn output weight representing the number of ith hidden layer nodes; g () represents an activation function; wiInput weights, W, representing the number of ith hidden layer nodesi=[wi1,wi2,…,wim']T,wim'Representing the input weight of the connection of the mth input layer node number and the ith hidden layer node number, wherein m 'represents the node number of the input layer, and m' is 5; xkDenotes an input variable, X, of the k-th steel platek=[xk1,xk2,…,xkm']T,xkm'An input variable representing the number of nodes of the m' th input layer corresponding to the kth sample steel plate; wi·XkRepresents WiAnd XkInner product of (d); biA bias representing the number of ith hidden layer nodes; t is tkRepresenting the sample rolling kilometers of a sample steel plate; n represents the total number of sample steel plates.
6. A steel plate roll wear amount prediction system, comprising:
the sample data acquisition module is used for acquiring sample data; the sample data comprises parameter data of a sample steel plate and the number of rolled kilometers of the sample; the parameters of the sample steel plate comprise a blank size and a target size;
the training sample data determining module is used for determining training sample data through a grey correlation degree analysis method according to the sample data; the training sample data comprises: influence factor data influencing the rolling kilometers in the parameter data and the sample rolling kilometers of the sample steel plate;
the neural network training module is used for training the neural network of the extreme learning machine by taking the influence factor data of the sample steel plate as input and the sample rolling kilometers of the sample steel plate as output to obtain a trained neural network model;
the influence factor data acquisition module is used for acquiring influence factor data of the steel plate to be predicted;
the rolling kilometer number prediction module is used for inputting the influence factor data of the steel plate to be predicted into the neural network model to obtain the rolling kilometer number of the steel plate to be predicted;
the roller sample abrasion loss acquisition module is used for acquiring the roller sample abrasion loss corresponding to the sample data;
the roller abrasion loss and rolling kilometer number relation determining module is used for determining the relation between the roller abrasion loss and the rolling kilometer number by utilizing a regression analysis method according to the roller sample abrasion loss and the sample rolling kilometer number;
and the roller wear amount prediction module is used for obtaining the predicted roller wear amount by utilizing the relationship between the roller wear amount and the rolling kilometers according to the rolling kilometers of the steel plate to be predicted.
7. The steel plate roll wear prediction system according to claim 6, wherein the training sample data determination module specifically includes:
a reference comparison sequence determining unit, configured to use the sample rolling kilometer number as a reference sequence, and use the parameter data as a comparison sequence;
the non-dimensionalization processing unit is used for carrying out non-dimensionalization processing on the reference number sequence and the comparison number sequence to obtain a reference non-dimensionalization number sequence and a comparison non-dimensionalization number sequence;
a difference series calculation unit for calculating a difference series of the reference scalar series and the comparison scalar series;
a maximum-minimum difference calculation unit for calculating a maximum difference and a minimum difference between the reference scalar number sequence and the comparison scalar number sequence using the difference number sequence;
a gray correlation coefficient calculation unit for calculating a gray correlation coefficient of each element in the comparison scalar number series with the reference scalar number series using the maximum difference and the minimum difference;
the grey correlation degree calculating unit is used for calculating the grey correlation degree of each element in the comparison scalar number series and the reference scalar number series respectively by utilizing the grey correlation coefficient;
and the training sample data determining unit is used for determining elements in the comparative unamount number series corresponding to the gray correlation degree which is greater than the preset gray correlation degree as influence factors influencing the rolling kilometers, and the data corresponding to the influence factors in the sample data are the influence factor data.
8. The steel plate roll wear prediction system of claim 6, wherein the neural network training module specifically comprises:
the hidden layer node number determining unit is used for determining the hidden layer node number of the neural network of the extreme learning machine by using the influence factor data of the sample steel plate as input and the sample rolling kilometer number of the sample steel plate as output and using cross validation;
and the output weight determining unit is used for determining the output weight of the neural network of the extreme learning machine by taking the influence factor data of the sample steel plate as input and the sample rolling kilometer number of the sample steel plate as output and utilizing the number of nodes of the hidden layer to obtain a neural network model.
9. The steel plate roll wear amount prediction system according to claim 8, wherein the hidden layer node number determination unit specifically includes:
a node number preset range obtaining subunit, configured to obtain a node number preset range of the hidden layer;
a hidden layer training node number determining subunit, configured to sequentially use all the node numbers within the preset node number range as hidden layer training node numbers of the extreme learning machine neural network;
the training subunit is used for taking the influence factor data of the sample steel plate as input, taking the number of rolled kilometers of a sample of the sample steel plate as output, training the extreme learning machine neural network corresponding to each hidden layer training node number by using cross-folding cross validation, training preset training times, and obtaining a trained neural network training model of each cross-folding cross validation corresponding to each hidden layer training node number;
the root mean square error calculation subunit is used for calculating the root mean square error of each neural network training model corresponding to the number of the hidden layer training nodes;
the root mean square error average value calculating operator unit is used for calculating the root mean square error average value of the extreme learning machine neural network corresponding to each hidden layer training node number by using all the root mean square errors corresponding to each hidden layer training node number and the preset training times;
and the comparison subunit is used for comparing the root mean square error average values of the extreme learning machine neural network corresponding to all the nodes within the preset node number range, and determining the node number corresponding to the minimum root mean square error average value as the hidden layer node number of the extreme learning machine neural network.
10. The steel plate roll wear amount prediction system according to claim 9, wherein the output weight determination unit specifically includes:
the node number determining subunit is used for determining the number of nodes of a hidden layer of the extreme learning machine neural network as the number of nodes of the hidden layer, and determining the number of nodes of an input layer of the extreme learning machine neural network as the number of influence factors in the influence factor data;
the output weight determining subunit is used for taking the influence factor data of the sample steel plate as the input of the extreme learning machine neural network, taking the sample rolling kilometer number of the sample steel plate as the output of the extreme learning machine neural network, and solving the following formula to determine the output weight of the extreme learning machine neural network to obtain a trained neural network model;
wherein L represents the total number of hidden layer nodes, i is 1, 2. Beta is aiAn output weight representing the number of ith hidden layer nodes; g () represents an activation function; wiInput weights, W, representing the number of ith hidden layer nodesi=[wi1,wi2,…,wim']T,wim'Representing the input weight of the connection of the mth input layer node number and the ith hidden layer node number, wherein m 'represents the node number of the input layer, and m' is 5; xkDenotes an input variable, X, of the k-th steel platek=[xk1,xk2,…,xkm']T,xkm'An input variable representing the number of nodes of the m' th input layer corresponding to the kth sample steel plate; wi·XkRepresents WiAnd XkInner product of (d); biA bias representing the number of ith hidden layer nodes; t is tkRepresenting the sample rolling kilometers of a sample steel plate; n represents the total number of sample steel plates.
CN202010908672.8A 2020-09-02 2020-09-02 Steel plate roller wear loss prediction method and system Pending CN112037209A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010908672.8A CN112037209A (en) 2020-09-02 2020-09-02 Steel plate roller wear loss prediction method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010908672.8A CN112037209A (en) 2020-09-02 2020-09-02 Steel plate roller wear loss prediction method and system

Publications (1)

Publication Number Publication Date
CN112037209A true CN112037209A (en) 2020-12-04

Family

ID=73591101

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010908672.8A Pending CN112037209A (en) 2020-09-02 2020-09-02 Steel plate roller wear loss prediction method and system

Country Status (1)

Country Link
CN (1) CN112037209A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112893476A (en) * 2021-01-15 2021-06-04 山信软件股份有限公司 Method for calculating rolling mileage of cold rolling mill
CN113111092A (en) * 2021-03-15 2021-07-13 中冶南方工程技术有限公司 Silicon steel iron loss prediction method based on cold rolling full-process data

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112893476A (en) * 2021-01-15 2021-06-04 山信软件股份有限公司 Method for calculating rolling mileage of cold rolling mill
CN112893476B (en) * 2021-01-15 2022-03-22 山信软件股份有限公司 Method for calculating rolling mileage of cold rolling mill
CN113111092A (en) * 2021-03-15 2021-07-13 中冶南方工程技术有限公司 Silicon steel iron loss prediction method based on cold rolling full-process data
CN113111092B (en) * 2021-03-15 2022-06-24 中冶南方工程技术有限公司 Silicon steel iron loss prediction method based on cold rolling full-process data

Similar Documents

Publication Publication Date Title
CN112037209A (en) Steel plate roller wear loss prediction method and system
CN106448151B (en) Short-term traffic flow prediction method
CN101346676B (en) Method and device for tuning and control
CN101863088A (en) Method for forecasting Mooney viscosity in rubber mixing process
CN106991242A (en) A kind of control method of plate property optimization
CN111241657A (en) Cold-rolled strip steel plate shape prediction method based on deep learning
CN109657878A (en) A kind of Air-conditioning Load Prediction method and device
KR100186903B1 (en) Method of controlling form of strip in rolling mill
Park et al. Prediction for manufacturing factors in a steel plate rolling smart factory using data clustering-based machine learning
WO2022116571A1 (en) Lstm-based roll-bending force prediction method for hot rolling
CN103316928B (en) On-line cold-rolled flatness signal pattern recognition system
CN109637128B (en) Markov-based gray Verhulst short-time traffic flow prediction method and system
KR100660231B1 (en) Method for decision optimal target shape of plate in continous rolling process
JPH08117827A (en) Rolling device
Chandrasekara et al. Forecasting exchange rates using artificial neural networks
KR100384121B1 (en) Method for controlling shape of cold rolled strip using neural network
CN108470227A (en) Based on the Regional Macro economic System prediction modeling method of assessment GDP growth figures and application
Xu et al. Application of Takagi–Sugeno fuzzy model for slab camber control in a hot strip rougher mill
CN113102516B (en) Hot continuous rolling strip steel head width prediction method integrating rolling mechanism and deep learning
CN114192587B (en) Cold-rolled plate shape control parameter setting method based on error back propagation algorithm
CN114117919B (en) Instant learning soft measurement modeling method based on sample collaborative representation
Feng et al. Roller Parameters Prediction of Steel Tube Based on Principal Component Analysis and BP Neural Network
CN112330012B (en) Building energy consumption prediction method and equipment based on transfer learning
JPH07204718A (en) Rolling device
Kim et al. Sensitivity analysis on a neural network for analyzing the camber in hot rolling process

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination