CN110264079B - Hot-rolled product quality prediction method based on CNN algorithm and Lasso regression model - Google Patents
Hot-rolled product quality prediction method based on CNN algorithm and Lasso regression model Download PDFInfo
- Publication number
- CN110264079B CN110264079B CN201910538744.1A CN201910538744A CN110264079B CN 110264079 B CN110264079 B CN 110264079B CN 201910538744 A CN201910538744 A CN 201910538744A CN 110264079 B CN110264079 B CN 110264079B
- Authority
- CN
- China
- Prior art keywords
- data
- model
- lasso regression
- product quality
- regression model
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000004519 manufacturing process Methods 0.000 claims abstract description 38
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000005098 hot rolling Methods 0.000 claims abstract description 5
- 238000013527 convolutional neural network Methods 0.000 claims description 35
- 238000012360 testing method Methods 0.000 claims description 20
- 238000011176 pooling Methods 0.000 claims description 18
- 230000006870 function Effects 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 13
- 230000009467 reduction Effects 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 9
- 239000000126 substance Substances 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 238000004140 cleaning Methods 0.000 claims description 6
- 238000010586 diagram Methods 0.000 claims description 5
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 230000009471 action Effects 0.000 claims description 3
- 230000005540 biological transmission Effects 0.000 claims description 3
- 238000010219 correlation analysis Methods 0.000 claims description 3
- 238000013461 design Methods 0.000 claims description 3
- 230000008030 elimination Effects 0.000 claims description 3
- 238000003379 elimination reaction Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000012937 correction Methods 0.000 claims description 2
- 230000009897 systematic effect Effects 0.000 claims description 2
- 229910000831 Steel Inorganic materials 0.000 abstract description 12
- 239000010959 steel Substances 0.000 abstract description 12
- 238000005096 rolling process Methods 0.000 abstract description 4
- 239000010410 layer Substances 0.000 description 24
- 230000008901 benefit Effects 0.000 description 4
- 239000002356 single layer Substances 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/21—Design, administration or maintenance of databases
- G06F16/215—Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06395—Quality analysis or management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Business, Economics & Management (AREA)
- Computational Linguistics (AREA)
- Marketing (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Development Economics (AREA)
- Educational Administration (AREA)
- Entrepreneurship & Innovation (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Manufacturing & Machinery (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Feedback Control In General (AREA)
Abstract
The invention belongs to the technical field of steel rolling product quality prediction, and particularly relates to a hot rolling product quality prediction method based on a CNN algorithm and a Lasso regression model. The method comprises the following steps of S1: obtaining sample data for modeling, wherein the sample data comprises training data, and determining a key input variable; s2: training the CNN by using key input variables of the sample data to obtain a feature vector model; s3: substituting key input variables of the training data into the feature vector model to obtain input variables substituted into the Lasso regression model; s4: determining an optimal regularization factor of the Lasso regression model, training the Lasso regression model by using the input variables in the S3 to obtain an uncorrected mixed prediction model, and correcting the uncorrected mixed prediction model to obtain a corrected mixed prediction model; s5: and inputting the production data of the future time period into the modified hybrid prediction model to obtain a prediction result of the production data. The method can improve the prediction precision of the model.
Description
Technical Field
The invention belongs to the technical field of steel rolling product quality prediction, and particularly relates to a hot rolling product quality prediction method based on a CNN algorithm and a Lasso regression model.
Background
In recent years, although steel enterprises have solved the situation of insufficient production of steel products, no proper control model exists for the product quality, so that the resource utilization rate in the production process is higher than the average level, the steel rolling production energy consumption ratio is too much, and compared with the foreign advanced technology, the method has a great progress space. At present, steel products are mainly used in various aspects of construction industry, aerospace industry, automobile manufacturing industry and the like, and the requirements of the industries on the quality of steel are very strict, so that a proper and accurate steel rolling product quality prediction model is established, and the method has great significance for product quality performance prediction.
The production process of steel is composed of a plurality of links, each link affects the product quality, in order to guarantee the product quality, a relation between production variables and product performance indexes needs to be established, and the aim of controlling the product quality is achieved by changing production parameters under the existing production conditions. After years of production experience, steel enterprises accumulate a great deal of production data which is easily polluted by factory noise, so how to process the data and use the data in modeling is preparation work before product quality modeling. At present, the method for establishing a prediction model for product quality mainly comprises two major categories, namely a regression algorithm and a neural network algorithm: the regression model has a simple structure, and although the generation of the over-fitting problem can be avoided, the regression model has the defect that the prediction result is greatly different from the expectation; all the nodes of the fully-connected neural network model are connected with each other, so that the quantity of extracted parameters is increased, and the computational complexity of the model is increased. Therefore, considering the above problems of the rolled steel product data and the modeling process, a proper modeling method needs to be selected to complete the product quality prediction.
Disclosure of Invention
Technical problem to be solved
Aiming at the existing technical problems, the invention provides the hot-rolled product quality prediction method based on the CNN algorithm and the Lasso regression model, which can relieve the overfitting problem of the model to a certain extent and improve the prediction precision of the model.
(II) technical scheme
In order to achieve the purpose, the invention adopts the main technical scheme that:
a hot-rolled product quality prediction method based on a CNN algorithm and a Lasso regression model comprises the following steps,
s1: obtaining sample data for modeling in historical data about the performance of the hot rolled product, wherein the sample data comprises training data, and determining a key input variable;
s2: training the CNN by using key input variables of the sample data to obtain a feature vector model;
s3: substituting key input variables of the training data into the feature vector model to obtain input variables substituted into the Lasso regression model;
s4: determining an optimal regularization factor of the Lasso regression model, training the Lasso regression model by using the input variables obtained in the step S3 to obtain an uncorrected hybrid prediction model, and correcting the uncorrected hybrid prediction model at least once to obtain a corrected hybrid prediction model for predicting production data of a future time period;
s5: and inputting the production data of the future time period into the modified hybrid prediction model to obtain a prediction result of the production data.
Specifically, in step S1, sample data for modeling in relation to the history data of the properties of the hot rolled product is acquired, including,
s11: acquiring historical data about the performance of hot rolling products in a preset time period, cleaning the historical data, and acquiring complete cleaned first-time data;
s12: sampling the cleaned first-time data to obtain sampled second-time data;
s13: performing dimensionality reduction on the sampled second data to obtain third data subjected to dimensionality reduction;
s14: and carrying out normalization processing on the third time data after dimension reduction to finally obtain sample data for modeling.
Specifically, after step S14, further comprising,
s15: determining the proportion of training data and test data in sample data;
s16: a key output variable is selected.
In particular, in step S4, the unmodified hybrid prediction model is modified at least once, including,
s41: substituting the key input variables of the test data into the feature vector model to obtain input variables for substituting into the unmodified hybrid prediction model;
s42: and substituting the input variables obtained in the step S41 into the unmodified hybrid prediction model to obtain a predicted value, comparing the predicted value with the key output variables of the test data to modify the hybrid prediction model for multiple times until the error between the predicted value and the key output variables in the test data is within a preset range, and obtaining the modified hybrid prediction model for predicting the production data of the current time period.
Specifically, the step S11 of obtaining the complete cleaned first data includes performing a cleaning operation on data having an abnormal value and a missing value by using a direct elimination processing mode;
the step S12 of obtaining the sampled second data includes sampling the data by using a system random sampling method.
Specifically, the data dimensionality reduction in S13 includes,
determining the degree of correlation between the input variable and the output variable in the sampled secondary data by using a correlation analysis method, calculating a correlation coefficient between the input variable and the output variable, determining a key input variable of the model according to the importance degree of the variables, and reducing the dimensionality of the input variable of the production data in the model, wherein the calculation mode of the correlation coefficient is as follows:
wherein n is the number of sample data sets, x i ，y i (i =1,2, … n) as input variables and output variables,
is the average of the input variable and the output variable.
Specifically, the specific method of the data normalization process in S14 is,
let x i ＝(x i1 ,x i2 ,…,x ip ) I ∈ 2313, p is the total number of features of the input variable, and the formula for normalizing the data is as follows:
wherein the content of the first and second substances,
is a value distributed between 0 and 1 after normalization treatment, k min Corresponding to the minimum value of the production variable, k max Corresponding to the maximum value of the production variable.
Specifically, in S15, the ratio of training data to test data in the sample data is determined, including,
modeling by using a regression method and a neural network method respectively, and comparing average relative errors of model predicted values and actual values in different proportions to obtain the proportion of training data and test data;
and in S16, three mechanical performance indexes of elongation at break, yield strength and tensile strength are selected as key output variables.
Specifically, the CNN is trained in S2 to obtain a feature vector model, including,
the process of establishing the feature vector model includes both forward propagation and backward propagation,
A. forward propagation
Integrating one-dimensional input variables into 6 × 6 two-dimensional data, using 32 convolution kernels with the size of 3 × 3 for the first convolution layer, using 64 convolution kernels with the same size in the design of the second convolution layer, and extracting layer by layer through the two convolution layers to finally obtain representative local data characteristics, wherein the convolution output calculation formula is as follows:
wherein the content of the first and second substances,
representing the output of the convolutional layer after adding an activation function, the activation function being a ReLU function, F j The number of the characteristic diagrams is shown,
is the input of the convolution characteristic of the previous layer,
in order to convolve the kernel matrix with the desired pattern,
is an offset;
after the convolution, the characteristic sampling after the convolution is carried out again, all data characteristics that draw the convolution layer promptly carry out statistical processing, choose for use maximum value pooling mode to sample, and two pooling layer windows all set to 2 x 2 size, and every process pooling is handled, and the data size all can reduce half, and the computational formula is:
wherein the content of the first and second substances,
is the output of the pond layer after the action,
pooling weight, pooling is pooling operation;
B. counter-propagating
The back propagation is a process of carrying out error calculation on the predicted value and the actual value, carrying out training through back transmission, and updating the weight, and the error objective function is defined as follows:
wherein E represents a back propagation error, n is the number of samples, y n Which represents the actual output of the device,
representing the prediction output.
Specifically, the optimal regularization factor of the Lasso regression model is determined in S4, which includes,
let X i (x i1 ，x i2 ，…，x ip ) T Is an input value of the model, wherein i =1,2, …, N is the number of samples, and p is a characteristic number of the production variable; y is i Is the output value of the model. Then the objective function optimized after the Lasso regression adds the L1 regularization term is:
wherein j =1,2, …, N, β = (β) 0 ，β 1 ，…，β j ) And λ is L1 regularization term coefficient for unknown parameters of dimension j × 1.
(III) advantageous effects
The invention has the beneficial effects that: according to the hot-rolled product quality prediction method based on the CNN algorithm and the Lasso regression model, the advantage of CNN extraction data characteristics is utilized, and Lasso regression is used for replacing the single-layer perceptron prediction output of the CNN, so that the problem of model overfitting can be relieved to a certain extent, and the prediction precision of the model is improved.
Drawings
FIG. 1 is a model diagram of a hot rolled product quality prediction method based on a CNN algorithm and a Lasso regression model;
FIG. 2 is a network architecture diagram for training a CNN;
FIG. 3 is a flow chart of a hot rolled product quality prediction method based on CNN algorithm and Lasso regression model.
Detailed Description
For the purpose of better explaining the present invention and to facilitate understanding, the present invention will be described in detail by way of specific embodiments with reference to the accompanying drawings.
The invention discloses a hot-rolled product quality prediction method based on a CNN algorithm and a Lasso regression model, which comprises the following steps,
s1: obtaining sample data for modeling in historical data of hot-rolled product performance, wherein the sample data comprises training data, and determining key input variables;
s2: training the CNN by using key input variables of the sample data to obtain a feature vector model;
s3: substituting key input variables of the training data into the feature vector model to obtain input variables substituted into the Lasso regression model;
s4: determining an optimal regularization factor of the Lasso regression model, training the Lasso regression model by using the input variables obtained in the step S3 to obtain an uncorrected hybrid prediction model, and correcting the uncorrected hybrid prediction model at least once to obtain a corrected hybrid prediction model for predicting production data of a future time period;
s5: and inputting the production data of the future time period into the modified hybrid prediction model to obtain a prediction result of the production data.
Specifically, in step S1, obtaining sample data for modeling in the history data on the properties of the hot rolled product includes:
s11: acquiring historical data about the performance of hot rolling products in a preset time period, cleaning the historical data, and acquiring complete cleaned first-time data;
s12: sampling the cleaned first data to obtain sampled second data;
s13: carrying out dimensionality reduction on the sampled second data to obtain third data subjected to dimensionality reduction;
s14: and carrying out normalization processing on the third time data after dimension reduction to finally obtain sample data for modeling.
Specifically, after step S14, further comprising,
s15: determining the proportion of training data and test data in sample data;
s16: a key output variable is selected.
Specifically, in step S4, the at least one correction of the unmodified hybrid prediction model includes:
s41: substituting the key input variables of the test data into the feature vector model to obtain input variables for substituting into the unmodified hybrid prediction model;
s42: and substituting the input variables obtained in the step S41 into the unmodified hybrid prediction model to obtain a predicted value, comparing the predicted value with the key output variables of the test data to modify the hybrid prediction model for multiple times until the error between the predicted value and the key output variables in the test data is within a preset range, and obtaining the modified hybrid prediction model for predicting the production data of the current time period.
Specifically, the specific method for acquiring the complete cleaned first data in step S11 includes: for data with abnormal values and missing values, a direct elimination processing mode is adopted to carry out cleaning work on the data;
the step S12 of obtaining the sampled second-time data includes: and sampling the data by using a systematic random sampling method.
Specifically, the data dimensionality reduction in S13 includes:
determining the degree of correlation between the input variable and the output variable in the sampled secondary data by using a correlation analysis method, calculating a correlation coefficient between the input variable and the output variable, determining a key input variable of the model according to the importance degree of the variables, and reducing the dimensionality of the input variable of the production data in the model, wherein the calculation mode of the correlation coefficient is as follows:
wherein n is the number of sample data sets, x i ，y i (i =1,2, … n) as input variables and output variables,
is the average of the input variable and the output variable.
In order to reduce the difference of the magnitude of the sample data, all variables need to be unified in the same magnitude, so that the data needs to be normalized, and the production variables can be converted from dimensional to non-dimensional. Specifically, the specific method of the data normalization process in S14 is,
let x i ＝(x i1 ,x i2 ,…,x ip ) I ∈ 2313, p is the special of the input variableAnd (4) characterizing the total number, wherein a formula for normalizing the data is as follows:
wherein the content of the first and second substances,
is a value distributed between 0 and 1 after normalization processing, k min Corresponding to the minimum value of the production variable, k max Corresponding to the maximum value of the production variable.
Specifically, the determining the ratio of the training data to the test data in the sample data in S15 includes:
modeling is carried out by a regression method and a neural network method respectively, and the average relative errors of the model predicted values and the model actual values under different proportions are compared to obtain the proportion of 8:2 between the training data and the test data, so 1850 groups of sample data are selected as the training data, and the rest 463 groups are selected as the test data.
Determining the factors influencing the mechanical properties of hot rolled products in production as 32 key input variables, and selecting three mechanical property indexes of elongation at break and yield strength) and tensile strength as key output variables.
Specifically, the CNN is trained in S2 to obtain a feature vector model, including,
the process of establishing the feature vector model comprises two aspects of forward propagation and backward propagation,
A. forward propagation
Integrating one-dimensional input variables into 6 × 6 two-dimensional data, using 32 convolution kernels with the size of 3 × 3 for the first convolution layer, using 64 convolution kernels with the same size in the design of the second convolution layer, and extracting layer by layer through the two convolution layers to finally obtain representative local data characteristics, wherein the convolution output calculation formula is as follows:
wherein the content of the first and second substances,
representing the output of the convolutional layer after adding an activation function, the activation function being a ReLU function, F j The number of the characteristic diagrams is shown,
is the input of the convolution characteristic of the previous layer,
in order to convolve the kernel matrix with the desired pattern,
is an offset;
after the convolution, the characteristic sampling after the convolution is carried out again, all data characteristics that draw the convolution layer promptly carry out statistical processing, choose for use maximum value pooling mode to sample, and two pooling layer windows all set to 2 x 2 size, and every process pooling is handled, and the data size all can reduce half, and the computational formula is:
wherein, the first and the second end of the pipe are connected with each other,
is the output of the pond layer after the action,
pooling weight, pooling is pooling operation;
B. counter-propagating
The back propagation is a process of carrying out error calculation on the predicted value and the actual value, carrying out training through back transmission, and updating the weight, and the error objective function is defined as follows:
wherein E represents a back propagation error, n is the number of samples, y n Which represents the actual output of the device,
representing the prediction output.
The feature vector model obtained by training the CNN is finally input by using a single-layer perceptron, a predicted value is output through a full-connection layer, the output mode can cause the over-fitting problem and further influences the prediction precision, and the Lasso regression method can relieve the over-fitting problem by changing the regularization coefficient, so that Lasso regression can be used for replacing the CNN single-layer perceptron output layer for prediction, and the improvement of the feature vector model obtained by training the CNN is realized. Lasso regression, i.e., minimum absolute value shrinkage and selection operators, alleviates the problem of overfitting of the model by adding L1 regularization.
Specifically, the optimal regularization factor of the Lasso regression model is determined in S4, which includes,
let X i (x i1 ，x i2 ，…，x ip ) T Is an input value of the model, wherein i =1,2, …, N is the number of samples, and p is a characteristic number of the production variable; y is i Is the output value of the model. Then the objective function optimized after the Lasso regression adds the L1 regularization term is:
wherein j =1,2, …, N, β = (β) 0 ，β 1 ，…，β j ) And λ is L1 regularization term coefficient for unknown parameters of dimension j × 1.
With the increase of the regularization term coefficient lambda, some model coefficients can be reduced to 0 even under the condition of keeping data characteristics, and the effect of reducing sum and selecting variables is achieved, so that the problem of model overfitting is effectively relieved.
The CNN is used as a special artificial neural network, has the ability of autonomous learning, and has the advantages of two functions of local connection and weight sharing. The function of replacing the whole characteristic by partial characteristic is realized by the local connection of the neuron and the neuron of the previous layer; the weight sharing means that the weights of the neurons in the same layer are the same, and the number of parameters processed by the model can be reduced. In addition, the Lasso regression can compress variables with smaller estimation parameters to zero by adding a regularization coefficient, so that the complexity and instability of a prediction model are reduced, and the over-fitting problem is avoided. According to the method, CNN is introduced to replace a full-connection neural network to model the quality prediction of the rolled steel product, and the CNN is easy to generate overfitting by using a mode of outputting a predicted value by a single-layer perceptron, so that the prediction precision of a test set is reduced, so that the hot rolled product quality prediction method based on the CNN algorithm and the Lasso regression model is provided, the advantage of CNN extraction data characteristics is kept, the advantage of the Lasso regression model for relieving the overfitting problem is also utilized, and the complexity and the instability of the model are reduced.
The technical principles of the present invention have been described above in connection with specific embodiments, which are intended to explain the principles of the present invention and should not be construed as limiting the scope of the present invention in any way. Based on the explanations herein, those skilled in the art will be able to conceive of other embodiments of the present invention without inventive efforts, which shall fall within the scope of the present invention.
Claims (10)
1. A hot-rolled product quality prediction method based on a CNN algorithm and a Lasso regression model is characterized by comprising the following steps,
s1: obtaining sample data for modeling in historical data of hot-rolled product performance, wherein the sample data comprises training data, and determining key input variables;
s2: training the CNN by using key input variables of the sample data to obtain a feature vector model;
s3: substituting key input variables of the training data into the feature vector model to obtain input variables substituted into the Lasso regression model;
s4: determining an optimal regularization factor of the Lasso regression model, training the Lasso regression model by using the input variables obtained in the step S3 to obtain an uncorrected hybrid prediction model, and correcting the uncorrected hybrid prediction model at least once to obtain a corrected hybrid prediction model for predicting production data of a future time period;
s5: and inputting the production data of the future time period into the modified hybrid prediction model to obtain a prediction result of the production data.
2. The CNN algorithm and Lasso regression model based hot rolled product quality prediction method according to claim 1, wherein said step S1 of obtaining sample data for modeling in historical data about hot rolled product properties comprises:
s11: acquiring historical data about the performance of hot rolling products in a preset time period, cleaning the historical data, and acquiring complete cleaned first-time data;
s12: sampling the cleaned first-time data to obtain sampled second-time data;
s13: carrying out dimensionality reduction on the sampled second data to obtain third data subjected to dimensionality reduction;
s14: and carrying out normalization processing on the third time data after dimensionality reduction to finally obtain sample data for modeling.
3. The CNN algorithm and Lasso regression model based hot rolled product quality prediction method according to claim 2, further comprising, after step S14,
s15: determining the proportion of training data and test data in sample data;
s16: a key output variable is selected.
4. The CNN algorithm and Lasso regression model based hot rolled product quality prediction method according to claim 3, wherein the step S4 of at least one correction of the uncorrected hybrid prediction model comprises:
s41: substituting the key input variables of the test data into the feature vector model to obtain input variables for substituting into the unmodified hybrid prediction model;
s42: and substituting the input variables obtained in the step S41 into the unmodified hybrid prediction model to obtain a predicted value, comparing the predicted value with the key output variables of the test data to modify the hybrid prediction model for multiple times until the error between the predicted value and the key output variables in the test data is within a preset range, and obtaining the modified hybrid prediction model for predicting the production data of the current time period.
5. The hot rolled product quality prediction method based on CNN algorithm and Lasso regression model according to claim 4,
step S11 of acquiring complete cleaned first data includes: for data with abnormal values and missing values, a direct elimination processing mode is adopted to carry out cleaning work on the data;
the step S12 of obtaining the sampled second-time data includes: and sampling the data by using a systematic random sampling method.
6. The CNN algorithm and Lasso regression model based hot rolled product quality prediction method according to claim 5, wherein the data dimensionality reduction in step S13 comprises,
determining the degree of correlation between the input variable and the output variable in the sampled secondary data by using a correlation analysis method, calculating a correlation coefficient between the input variable and the output variable, determining a key input variable of the model according to the importance degree of the variables, and reducing the dimensionality of the input variable of the production data in the model, wherein the calculation mode of the correlation coefficient is as follows:
wherein n is the number of sample data sets, x i ，y i (i =1,2, … n) as input variables and output variables,
is the average of the input variable and the output variable.
7. The hot rolled product quality prediction method based on CNN algorithm and Lasso regression model according to claim 6, wherein the data normalization in S14 is performed by,
let x i ＝(x i1 ,x i2 ,…,x ip ) I ∈ 2313, p is the total number of features of the input variable, and the formula for normalizing the data is as follows:
wherein the content of the first and second substances,
is a value distributed between 0 and 1 after normalization treatment, k min Corresponding to the minimum value of the production variable, k max Corresponding to the maximum value of the production variable.
8. The hot rolled product quality prediction method based on CNN algorithm and Lasso regression model according to claim 7,
in S15, the ratio of training data to test data in the sample data is determined, including,
modeling by using a regression method and a neural network method respectively, and comparing average relative errors of model predicted values and actual values in different proportions to obtain the proportion of training data and test data;
and in S16, three mechanical performance indexes of elongation at break, yield strength and tensile strength are selected as key output variables.
9. The CNN algorithm and Lasso regression model based hot rolled product quality prediction method according to claim 8, wherein the CNN is trained in step S2 to obtain a feature vector model, comprising,
the process of establishing the feature vector model includes both forward propagation and backward propagation,
A. forward propagation
Integrating one-dimensional input variables into 6 × 6 two-dimensional data, using 32 convolution kernels with the size of 3 × 3 for the first convolution layer, using 64 convolution kernels with the same size in the design of the second convolution layer, and extracting layer by layer through the two convolution layers to finally obtain representative local data characteristics, wherein the convolution output calculation formula is as follows:
wherein the content of the first and second substances,
representing the output of the convolutional layer after adding an activation function, the activation function being a ReLU function, F j The number of the characteristic diagrams is shown,
is the input of the convolution characteristic of the previous layer,
in order to convolve the kernel matrix with the desired pattern,
is an offset;
after the convolution, the characteristic sampling after the convolution is carried out again, all data characteristics that draw the convolution layer promptly carry out statistical processing, choose for use maximum value pooling mode to sample, and two pooling layer windows all set to 2 x 2 size, and every process pooling is handled, and the data size all can reduce half, and the computational formula is:
wherein the content of the first and second substances,
is the output of the pond layer after the action,
pooling weight, pooling is pooling operation;
B. counter-propagating
The back propagation is a process of carrying out error calculation on the predicted value and the actual value, carrying out training through back transmission, and updating the weight, and the error objective function is defined as follows:
wherein E represents a back propagation error, n is the number of samples, y n Which represents the actual output of the device,
representing the prediction output.
10. The CNN algorithm and Lasso regression model based hot rolled product quality prediction method according to claim 9, characterized in that the optimal regularization factor of the Lasso regression model determined in step S4 comprises,
let X i (x i1 ，x i2 ，…，x ip ) T Is an input value of the model, wherein i =1,2, …, N is the number of samples, and p is a characteristic number of the production variable; y is i Is the output value of the model; then the objective function optimized after the Lasso regression adds the L1 regularization term is:
wherein j =1,2, …, N, β = (β) 0 ，β 1 ，…，β j ) And λ is L1 regularization term coefficient for unknown parameters of dimension j × 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910538744.1A CN110264079B (en) | 2019-06-20 | 2019-06-20 | Hot-rolled product quality prediction method based on CNN algorithm and Lasso regression model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910538744.1A CN110264079B (en) | 2019-06-20 | 2019-06-20 | Hot-rolled product quality prediction method based on CNN algorithm and Lasso regression model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110264079A CN110264079A (en) | 2019-09-20 |
| CN110264079B true CN110264079B (en) | 2022-12-06 |
Family
ID=67919913
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910538744.1A Expired - Fee Related CN110264079B (en) | 2019-06-20 | 2019-06-20 | Hot-rolled product quality prediction method based on CNN algorithm and Lasso regression model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110264079B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110705807B (en) * | 2019-10-23 | 2023-11-03 | 哈工大机器人（合肥）国际创新研究院 | Semiconductor production quality prediction method and system based on field data |
| CN110991495B (en) * | 2019-11-14 | 2023-03-28 | 国机智能技术研究院有限公司 | Method, system, medium, and apparatus for predicting product quality in manufacturing process |
| CN111951037A (en) * | 2020-07-02 | 2020-11-17 | 西安丝路智慧科技有限公司 | Big data-based passenger flow prediction and travel marketing method |
| CN112034708A (en) * | 2020-08-26 | 2020-12-04 | 贵州和盛食品开发有限公司 | Noodle processing attachment self-adjusting system |
| CN112016252B (en) * | 2020-09-07 | 2022-07-01 | 重庆科技学院 | Method for acquiring precise take-off pressure of safety valve |
| CN112149254B (en) * | 2020-09-24 | 2023-05-02 | 上海电力大学 | Gas turbine starting process prediction method based on Lasso-RF model |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107330514A (en) * | 2017-07-10 | 2017-11-07 | 北京工业大学 | A kind of Air Quality Forecast method based on integrated extreme learning machine |
| CN109147878A (en) * | 2018-10-08 | 2019-01-04 | 燕山大学 | A kind of clinker free calcium flexible measurement method |
| JP2019028949A (en) * | 2017-08-03 | 2019-02-21 | 新日鐵住金株式会社 | Product state prediction device and method, manufacturing process control system, and program |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105205224B (en) * | 2015-08-28 | 2018-10-30 | 江南大学 | Time difference Gaussian process based on fuzzy curve analysis returns soft-measuring modeling method |
-
2019
- 2019-06-20 CN CN201910538744.1A patent/CN110264079B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107330514A (en) * | 2017-07-10 | 2017-11-07 | 北京工业大学 | A kind of Air Quality Forecast method based on integrated extreme learning machine |
| JP2019028949A (en) * | 2017-08-03 | 2019-02-21 | 新日鐵住金株式会社 | Product state prediction device and method, manufacturing process control system, and program |
| CN109147878A (en) * | 2018-10-08 | 2019-01-04 | 燕山大学 | A kind of clinker free calcium flexible measurement method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110264079A (en) | 2019-09-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110264079B (en) | Hot-rolled product quality prediction method based on CNN algorithm and Lasso regression model | |
| Gaspar-Cunha et al. | A multi-objective evolutionary algorithm using neural networks to approximate fitness evaluations. | |
| CN112861992B (en) | Wind power plant ultra-short term power prediction method based on independent sparse stacking self-encoder | |
| CN111461463A (en) | Short-term load prediction method, system and equipment based on TCN-BP | |
| CN112215353A (en) | Channel pruning method based on variational structure optimization network | |
| CN111985825A (en) | Crystal face quality evaluation method for roller mill orientation instrument | |
| CN114580934A (en) | Early warning method for food detection data risk based on unsupervised anomaly detection | |
| CN114298134A (en) | Wind power prediction method and device and electronic equipment | |
| CN115879369A (en) | Coal mill fault early warning method based on optimized LightGBM algorithm | |
| CN108509692B (en) | High-sulfur-content natural gas desulfurization process modeling method based on MiUKFNN algorithm | |
| CN111898799B (en) | BFA-Elman-based power load prediction method | |
| Phan et al. | Efficiency enhancement of evolutionary neural architecture search via training-free initialization | |
| CN113128666A (en) | Mo-S-LSTMs model-based time series multi-step prediction method | |
| CN111539558B (en) | Power load prediction method adopting optimization extreme learning machine | |
| CN109214513B (en) | Solid-liquid waste intelligent coupling treatment method based on adaptive deep belief network | |
| CN109146007B (en) | Solid waste intelligent treatment method based on dynamic deep belief network | |
| CN111476402A (en) | Wind power generation capacity prediction method coupling meteorological information and EMD technology | |
| CN116663745A (en) | LSTM drainage basin water flow prediction method based on PCA_DWT | |
| CN111369072A (en) | Nuclear minimum mean square time sequence online prediction model based on sparsification method | |
| CN116303386A (en) | Intelligent interpolation method and system for missing data based on relational graph | |
| CN117252285A (en) | Multi-index sewage water quality prediction method based on parallel CNN-GRU network | |
| CN115660038A (en) | Multi-stage integrated short-term load prediction based on error factors and improved MOEA/D-SAS | |
| CN113449912B (en) | Space load situation sensing method based on artificial intelligence technology | |
| CN114861815A (en) | Bearing fault diagnosis method and system based on improved DBN-MobileNet V2 | |
| CN114492007A (en) | Factor effect online identification method and device based on hierarchical error control |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20221206 |