CN115034437A - Hot rolled plate convexity prediction method based on improved XGboost - Google Patents
Hot rolled plate convexity prediction method based on improved XGboost Download PDFInfo
- Publication number
- CN115034437A CN115034437A CN202210515319.2A CN202210515319A CN115034437A CN 115034437 A CN115034437 A CN 115034437A CN 202210515319 A CN202210515319 A CN 202210515319A CN 115034437 A CN115034437 A CN 115034437A
- Authority
- CN
- China
- Prior art keywords
- convexity
- strip steel
- data
- value
- xgboost
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 44
- 229910000831 Steel Inorganic materials 0.000 claims abstract description 51
- 239000010959 steel Substances 0.000 claims abstract description 51
- 241000283153 Cetacea Species 0.000 claims abstract description 31
- 238000004519 manufacturing process Methods 0.000 claims abstract description 31
- 230000002159 abnormal effect Effects 0.000 claims abstract description 16
- 238000005457 optimization Methods 0.000 claims abstract description 15
- 238000010606 normalization Methods 0.000 claims abstract description 12
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 238000005096 rolling process Methods 0.000 claims description 21
- 238000012545 processing Methods 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 11
- 239000013598 vector Substances 0.000 claims description 11
- 238000012549 training Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 5
- 238000005452 bending Methods 0.000 claims description 4
- FFBHFFJDDLITSX-UHFFFAOYSA-N benzyl N-[2-hydroxy-4-(3-oxomorpholin-4-yl)phenyl]carbamate Chemical compound OC1=C(NC(=O)OCC2=CC=CC=C2)C=CC(=C1)N1CCOCC1=O FFBHFFJDDLITSX-UHFFFAOYSA-N 0.000 claims description 3
- 238000012804 iterative process Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 2
- 238000005098 hot rolling Methods 0.000 description 9
- 238000009776 industrial production Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000005299 abrasion Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000011217 control strategy Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 238000013000 roll bending Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- 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
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B21—MECHANICAL METAL-WORKING WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
- B21B—ROLLING OF METAL
- B21B37/00—Control devices or methods specially adapted for metal-rolling mills or the work produced thereby
- B21B37/28—Control of flatness or profile during rolling of strip, sheets or plates
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- General Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Mathematical Physics (AREA)
- Primary Health Care (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Manufacturing & Machinery (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mechanical Engineering (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a hot rolled plate convexity prediction method based on improved XGboost, which comprises the following steps: step 1: collecting production data of strip steel; step 2: preprocessing the collected production data, including removing abnormal values of the convexity of the rolled strip steel of F6, and carrying out normalization and Kmeans clustering on all the strip steel production data; and step 3: establishing an XGboost model by taking the convexity of the rolled strip steel of F6 as an output value and the production data of other strip steels as input values; and 4, step 4: the XGboost model is subjected to super-parameter optimization by adopting a whale optimization algorithm to obtain the optimal parameters, and the final XGboost model is determined.
Description
Technical Field
The invention relates to a hot rolled plate convexity prediction method based on improved XGboost, and belongs to the technical field of automatic control of a rolling process.
Background
As one of the important indexes of the strip shape quality of hot rolled strip steel products, the risk of the subsequent processing process is caused by the unsatisfactory plate convexity precision, so that the final products are unqualified and the raw materials are wasted. The plate convexity, also called as the transverse thickness difference of the strip steel, refers to the thickness difference of the plate strip along the width direction, reflects the section shape after the strip steel is rolled, and the calculation formula can be expressed as follows:
wherein C' represents the strip crown, h c Represents the thickness h at the center point of the cross section of the strip width L The thickness h of a left reference point on the cross section of the strip width R The thickness of the reference point on the right side of the cross section of the strip width is shown.
For hot continuous rolling strip steel, factors influencing the strip steel outlet plate convexity comprise the grinding convexity of a roller, the abrasion convexity and thermal expansion convexity of the roller in the rolling process, the roll shifting amount, the roll bending force and the like, which are mutually coupled, and the traditional hot rolling plate convexity prediction is that the establishment of a mechanism model in the hot rolling process often needs to meet certain assumed conditions due to the complexity of the process flow, for example, the boundary conditions established by a differential equation in an analytic model and the determination of material parameters in the finite element model establishing process are simplified, so that the model has certain limitations, the improvement of the precision of a plate convexity prediction model is very unfavorable, and a method for predicting the plate convexity in the hot continuous rolling process by using industrial large data is urgently needed.
In recent years, an artificial neural network has been widely used in many fields as a machine learning method which is widely accepted and used, and among them, it is also used for predicting the plate crown to obtain a good prediction effect.
Disclosure of Invention
The invention provides a hot rolled plate convexity prediction method based on improved XGboost, wherein in the rolling process, according to actually measured rolling data of a production line, an XGboost algorithm with high model prediction accuracy and high prediction efficiency is adopted to construct and optimize a model, so that the problems of low efficiency and large deviation of the existing plate convexity prediction technology are solved, and important support is provided for implementation of a subsequent plate shape control strategy.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
a hot rolled plate convexity prediction method based on improved XGboost comprises the following steps:
step 1: collecting production data of strip steel, including the thickness of an intermediate blank, the width of the rolled strip steel, the thickness of the rolled strip steel, the convexity of the rolled strip steel from F1 to F6, the thickness of a finish rolling inlet, the finish rolling temperature, the rolling force of F1 to F6 stands, the bending force of F1 to F6 stands and the roll shifting amount of F1 to F6 stands;
and 2, step: preprocessing the collected production data, including removing abnormal values of the convexity of the rolled strip steel of F6, and carrying out normalization and Kmeans clustering on all the strip steel production data;
and 3, step 3: establishing an XGboost model by taking the convexity of the rolled strip steel of F6 as an output value and the production data of other strip steels as input values;
and 4, step 4: and (4) carrying out super-parameter optimization on the XGboost model by adopting a whale optimization algorithm to obtain the optimal parameters, and determining the final XGboost model.
The technical scheme of the invention is further improved as follows: the specific method for preprocessing in the step 2 comprises the following steps:
step 2.1: calculating the standard deviation S of the convexity of the F6 rolled strip steel y And average value
Step 2.2: screening abnormal values of the rolled strip steel convexity data of F6 by adopting a Laplace criterion, and removing the abnormal values, wherein the Laplace criterion calculation formula is as follows:
wherein, y i The data of the convexity of the rolled strip steel of the ith F6,
is the average value of the convexity S of the rolled strip steel of F6 y The standard deviation of the convexity of the rolled strip steel is F6;
step 2.3: normalization processing is carried out on the rolled strip steel convexity and other strip steel production data of the F6 after the abnormal value is removed, and the calculation formula of the normalization processing is as follows:
wherein, x' i Is a normalized value, x i Is an initial value, x max 、x min Respectively, a maximum value and a minimum value in the data set;
step 2.4: performing Kmeans clustering on the data after the normalization processing, wherein the Kmeans clustering comprises the following steps:
step 2.4.1: selecting cluster center, μ 1 (0) ，μ 2 (0) ，μ 3 (0) ，···，μ k (0) ；
Step 2.4.2: defining a loss function:
wherein, c i Represents x' i In cluster, μ ci Representing the center of the cluster, and M represents the number of data after removing the abnormal value;
step 2.4.3: setting an iteration step number t;
step 2.4.4: for each set of data x' i It is assigned to the point closest to it:
c i t ＝argmin k ||x′ i -μ k t ||；
wherein, c i t Denotes x 'in the t iteration process' i In cluster, μ k t Representing the cluster center of the t-th iteration process;
step 2.4.5: and updating the cluster center for each cluster of data:
where b represents the data amount of k clusters, μ k t+1 Representing the updated cluster center;
step 2.4.6: repeating steps 2.4.4 and 2.4.5 until J converges to a minimum value;
step 2.5: and (3) carrying out clustering processing on the data according to the following steps of 7: a ratio of 3 is divided into a training set and a test set.
The technical scheme of the invention is further improved as follows: average value in said step 2.1
The calculation formula of (2) is as follows:
where m is the number of production data, y i The value of the i-th rolled strip convexity is F6.
The technical scheme of the invention is further improved as follows: standard deviation S in said step 2.1 y The calculation formula of (c) is:
where m is the number of production data, y i The value of the i-th rolled strip convexity is F6.
The technical scheme of the invention is further improved as follows: after the XGboost model is established in the step 3, model prediction is carried out by adopting default parameter combination; the default parameters include the maximum depth max _ depth of the tree, the minimum loss function reduction value gamma of the leaf node creation branch, the feature sampling proportion colsample _ byte of the tree, the sum min _ child _ weight of the minimum weight of the leaf node, the learning rate eta, and the subsampling proportion subsample of the training sample.
The technical scheme of the invention is further improved as follows: the specific method of the whale optimization algorithm in the step 4 comprises the following steps:
step 4.1: setting related parameters of a whale optimization algorithm, setting a constant b for adjusting the logarithmic spiral shape to be 1, and setting l as a random number between intervals of < -1,1 >;
step 4.2: setting the number N of whales and the maximum iteration number t max Initializing position information, wherein the position of each whale individual represents a parameter combination of an XGboost model;
step 4.3: designing an adaptability function, adopting MSE as a performance index, substituting the performance index into the XGboost model, calculating the adaptability of each whale individual in the population, and reserving the current optimal whale individual;
step 4.4: calculating parameters a and p and coefficient vectors A and C, judging whether p is less than 0.5, if so, performing the step 4.5, otherwise, adopting the bubble net to prey, and updating the trapping position of the bubble net as follows:
X(t+1)＝D′×e bl ×cos(2πl)+X * (t)
D′＝|X * (t)-X(t)|
wherein X * (t) is the position of the current optimal solution, X (t) is the current search individual position, b is the spiral shape parameter;
step 4.5: judging whether the absolute value of the coefficient vector A is smaller than 1, if so, surrounding a prey, otherwise, carrying out global search, and updating the position as follows:
D＝|C·X′(t)-X(t)|
X(t+1)＝X′(t)-A·D
wherein A is 2a × r 1 -a，C＝2×r 2 A is linearly reduced from 2 to 0, r in the iterative process 1 And r 2 Is [0,1]]X' (t) represents the position vector of the current optimal solution when surrounding a prey, and represents the position vector of a random individual when searching globally;
step 4.6: after the position updating is finished, calculating the fitness of each whale individual, and updating the optimal whale individual;
step 4.7: judging whether the maximum iteration times is reached, if so, ending to obtain an optimal solution, otherwise, returning to the step 4.3;
step 4.8: outputting an optimal parameter combination to obtain optimal parameters of the XGboost model;
step 4.9: and substituting the test sample into the determined model, and evaluating the accuracy of the model.
Due to the adoption of the technical scheme, the invention has the technical progress that:
the invention provides a hot rolled plate convexity prediction method based on improved XGboost, which is characterized in that a whale optimization algorithm is utilized to perform super-parameter optimization on an XGboost model, plate convexity prediction is realized, prediction accuracy is improved, and the defects that the traditional hot rolled plate convexity prediction depends on the principle that the assumed conditions of a model are easy to change and the prediction accuracy is low are overcome.
Meanwhile, the convexity standards of the plates produced on the current hot rolling site are numerous, and if the plates are put into the same model to be trained, a model with accurate data is difficult to obtain, so that the data is mined according to different target convexities, the data is processed by adopting a Kmeans method, each type of the data is modeled, the prediction precision of the model is improved, and the method is effectively applied to the hot rolling site.
In addition, the method provided by the invention adopts hot rolling production data, effectively processes the data, can be put into use through computer programming and has low cost.
Drawings
FIG. 1 is a schematic diagram of a hot rolling line serviced by the present invention;
FIG. 2 is a flow chart of a hot rolled plate convexity prediction method based on the improved XGboost of the invention;
fig. 3a and 3b are scatter-fit plots of cluster 0 and cluster 1 model predicted and measured values, respectively.
Detailed Description
The present invention will be described in further detail with reference to the following examples:
as shown in FIG. 1, the present invention is applied to a hot rolling line for predicting the crown of a sheet (i.e., the crown of a rolled strip of F6), and combines the prediction method with the hot rolling line. As shown in fig. 2, a flow chart of a hot-rolled plate convexity prediction method based on improved XGBoost includes the following specific steps:
step 1: the production data of the strip steel are collected, and the production data comprises coil number, production time, intermediate billet thickness, rolled strip width, rolled strip convexity of F1-F6, rolled strip convexity, finish rolling inlet temperature, finish rolling temperature, rolling force of F1-F6 stands, bending force of F1-F6 stands, roll shifting amount of F1-F6 stands and rolling speed of F1-F6 stands.
In this example, production data of a certain steel mill are collected, 2797 groups of data are counted, and part of sample data is shown in table 1. And selecting 34 characteristics with the largest influence on the crown of the plate, wherein the characteristics comprise the thickness of an intermediate blank, the width of the rolled strip steel, the thickness of the rolled strip steel, the crown of the rolled strip steel of F1-F5, the inlet temperature of finish rolling, the finish rolling temperature, the rolling force of F1-F6 stands, the rolling speed of F1-F6 stands, the bending force of F1-F6 stands and the roll shifting amount of F1-F6 stands. And taking the convexity of the rolled strip steel of F6 as an output value.
TABLE 1 partial sample data of a certain rolling mill
Step 2: preprocessing the acquired data, including removing abnormal values of the convexity of the rolled strip steel of F6, and normalizing and Kmeans clustering all the strip steel production data, and the specific steps are as follows:
step 2.1: and calculating the standard deviation and the average value of the convexity of the rolled strip steel of F6.
In this embodiment, the formula for calculating the mean and standard deviation is as follows:
wherein m is the number of production data, y i Is the ith production data.
Step 2.2: the method comprises the following steps of collecting rolled strip convexity data of F6 at a hot rolling site, screening abnormal values (abnormal values meeting the Lauder criterion) of the rolled strip convexity data of F6 by adopting the Lauder criterion, and removing the abnormal values, wherein the Lauder criterion calculation formula is as follows:
wherein x is i For the (i) -th production data,
average value of production data, S y Standard deviation of production data.
In this example, a total of 12 data points are culled.
Step 2.3: normalization processing is carried out on the rolled strip steel convexity and other strip steel production data of the F6 after the abnormal value is removed, and the calculation formula of the normalization processing is as follows:
wherein, x' i Is a normalized value, x i Is an initial value, x max 、x min Respectively, a maximum value and a minimum value in the data set;
step 2.4: performing Kmeans clustering processing on the data after the normalization processing, wherein the Kmeans clustering step comprises the following steps:
step 2.4.1: determining the number of clusters by combining an elbow method and a contour coefficient method, wherein the number of the clusters is 2 in the embodiment;
step 2.4.2: selecting cluster center, μ 1 (0) ，μ 2 (0) ；
Step 2.4.3: defining a loss function:
wherein, c i Represents x' i The cluster is located in the position of the cluster,
representing the cluster center;
step 2.4.4: setting the maximum iteration step number T equal to 20;
step 2.4.5: for each set of data x' i It is assigned to the point closest to it:
c i t ＝argmin k ||x′ i -μ k t ||；
wherein, c i t Denotes x 'in the t iteration process' i In cluster, μ k t Representing the cluster center of the t-th iteration process; step 2.4.6: and updating the cluster center for each cluster of data:
where b represents the data amount of k clusters, μ k t+1 Representing the updated cluster center;
step 2.4.7: repeating steps 2.4.5 and 2.4.6 until J converges to a minimum value;
in this embodiment, after the Kmeans clustering process, the original data set is divided into two data sets of 0 class and 1 class, which respectively include 1527 groups and 1258 groups of data.
Step 2.5: and (3) respectively carrying out 7: a ratio of 3 is divided into a training set and a test set.
And step 3: based on the preprocessed data, an XGboost model is established, and model prediction is carried out by adopting default parameter combination, wherein the default parameters comprise the maximum depth max _ depth of the tree, the minimum loss function descending value gamma of a leaf node creation branch, the feature sampling proportion colsample _ byte of the tree, the sum min _ child _ weight of the minimum weight of the leaf node, the learning rate eta and the sub-sampling proportion subsample of a training sample.
And 4, step 4: optimizing the XGboost model by adopting a whale optimization algorithm, and determining the hyper-parameters of the model, wherein the specific method comprises the following steps:
step 4.1: setting related parameters of a whale optimization algorithm, setting a constant b of a logarithmic spiral shape to be 1, and setting l as a random number between intervals of < -1 > and 1;
step 4.2: generating a population with the whale number N of 30, wherein the position of each whale individual represents a parameter combination of an XGboost model, and the maximum iteration time t max Initializing position information to 10, and determining the maximum depth max _ depth of the parameter value range tree to be [5, 50%]The leaf node creates a branch minimum loss function with a reduced value gamma of 0.001,100]The tree's characteristic sample ratio colsample _ byte is [0.5,0.9 ]]The sum of the minimum weights of the leaf nodes min _ child _ weight is [5, 50%]And the learning rate eta is [0.01,0.2 ]]The subsample ratio of the training sample is [0.5,0.9 ]]。
Step 4.3: designing an adaptability function, substituting a prediction Mean Square Error (MSE) of the convexity of the hot rolled plate as a performance index into an XGboost model, calculating the adaptability of each individual whale in the population, and reserving the current optimal individual whale, wherein the adaptability function calculation formula is as follows:
step 4.4: judging whether p is less than 0.5, wherein p is the probability of a predation mechanism and is a random number of a value range [0,1], if the p is less than 0.5, performing the step 4.5, otherwise, predating by adopting a bubble net, and updating the predation position of the bubble net as follows:
X(t+1)＝D′×e bl ×cos(2πl)+X * (t)
D′＝|X * (t)-X(t)|
wherein X * (t) is the current best solution position, X (t) is the current search individual position, b is the spiral shape parameter, l is [ -1,1]A random number in between;
step 4.5: and gradually reducing the parameter A and the convergence factor a along with the increase of the iteration times t, and if the absolute value of A is less than 1, gradually surrounding the current optimal solution by each whale individual and entering a local optimization stage. Judging whether the absolute value of the coefficient vector A is smaller than 1, if so, surrounding the prey, otherwise, searching the individual to be close to the random individual for global search, and updating the position as follows:
D＝|C·X′(t)-X(t)|
X(t+1)＝X′(t)-A·D
wherein A is 2a × r 1 -a，C＝2×r 2 A is linearly reduced from 2 to 0, r in the iterative process 1 And r 2 Is [0,1]]X' (t) represents the position vector of the current optimal solution when surrounding a prey, and represents the position vector of a random individual when searching globally;
step 4.6: after the position updating is finished, calculating the fitness of each whale individual, and updating the optimal whale individual;
step 4.7: and judging whether the maximum iteration times is reached, if so, ending to obtain the optimal solution, otherwise, returning to the step 4.3.
Step 4.8: and outputting an optimal parameter combination, wherein the maximum depth max _ depth of an optimal parameter tree of the XGboost model is 29, the drop value gamma of a leaf node creation branch minimum loss function is 20.0, the feature sampling proportion colsample _ byte of the tree is 0.9, the sum min _ child _ weight of the leaf node minimum weight is 17.64, the learning rate eta is 0.2, and the training sample sub-sampling proportion subsample is 0.9.
Step 4.9: and substituting the parameter values into the XGboost model for training, traversing the whale population scale and the maximum iteration times within [10,50], and finally determining the whale population scale N to be 25 and the maximum iteration times to be 10.
According to the requirements of the current industrial production, the actual convexity value is regarded as qualified within the range of the set value +/-10 mu m, and more than 98.45 percent of products meet the standard through a scattered point fitting graph of the model predicted value and the measured value shown in figure 3, so that the method can be effectively applied to the current industrial production.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, and not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those skilled in the art; the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; such modifications and substitutions do not depart from the spirit and scope of the corresponding technical solutions as defined in the appended claims.
Claims (6)
1. A hot rolled plate convexity prediction method based on improved XGboost is characterized by comprising the following steps: the method comprises the following steps:
step 1: collecting production data of strip steel, including the thickness of an intermediate blank, the width of the rolled strip steel, the thickness of the rolled strip steel, the convexity of the rolled strip steel from F1 to F6, the thickness of a finish rolling inlet, the finish rolling temperature, the rolling force of F1 to F6 stands, the bending force of F1 to F6 stands and the roll shifting amount of F1 to F6 stands;
and 2, step: preprocessing the collected production data, including removing abnormal values of the convexity of the rolled strip steel of F6, and carrying out normalization and Kmeans clustering on all the strip steel production data;
and step 3: establishing an XGboost model by taking the convexity of the rolled strip steel of F6 as an output value and the production data of other strip steels as input values;
and 4, step 4: and (4) carrying out super-parameter optimization on the XGboost model by adopting a whale optimization algorithm to obtain the optimal parameters, and determining the final XGboost model.
2. The hot-rolled plate convexity prediction method based on the improved XGboost as claimed in claim 1, wherein the method comprises the following steps: the specific method for preprocessing in the step 2 comprises the following steps:
step 2.1: calculating the standard deviation S of the convexity of the rolled strip steel of F6 y And average value
Step 2.2: screening abnormal values of the rolled strip steel convexity data of F6 by adopting a Laplace criterion, and removing the abnormal values, wherein the Laplace criterion calculation formula is as follows:
wherein, y i The data of the convexity of the rolled strip steel of the ith F6,
is the average convexity of the rolled strip steel of F6S y The standard deviation of the convexity of the rolled strip steel is F6;
step 2.3: normalization processing is carried out on the rolled strip steel convexity and other strip steel production data of the F6 after the abnormal value is removed, and the calculation formula of the normalization processing is as follows:
wherein x is i ' is a normalized value, x i Is an initial value, x max 、x min Respectively, the maximum value in the data set anda minimum value;
step 2.4: performing Kmeans clustering processing on the data after the normalization processing, wherein the Kmeans clustering step comprises the following steps:
step 2.4.1: selecting cluster center, μ 1 (0) ，μ 2 (0) ，μ 3 (0) ，···，μ k (0) ；
Step 2.4.2: defining a loss function:
wherein, c i Represents x' i The cluster is located in the position of the cluster,
representing the center of the cluster, and M represents the number of data after removing the abnormal value;
step 2.4.3: setting iteration step number t;
step 2.4.4: for each set of data x' i It is assigned to the point closest to it:
c i t ＝arg min k ||x′ i -μ k t ||；
wherein, c i t Denotes x 'in the process of the t iteration' i In cluster, μ k t Representing the cluster center of the t-th iteration process;
step 2.4.5: and updating the cluster center for each cluster of data:
where b represents the data amount of k clusters, μ k t+1 Representing the updated cluster center;
step 2.4.6: repeating steps 2.4.4 and 2.4.5 until J converges to a minimum value;
step 2.5: and (4) clustering the processed data according to the following steps of 7: a ratio of 3 is divided into a training set and a test set.
3. The hot-rolled plate convexity prediction method based on the improved XGboost as claimed in claim 2, characterized in that: the calculation formula of the average value y in the step 2.1 is as follows:
where m is the number of production data, y i The rolled strip crown value is the ith F6 rolled strip crown value.
4. The hot-rolled plate convexity prediction method based on the improved XGboost as claimed in claim 2, characterized in that: standard deviation S in said step 2.1 y The calculation formula of (2) is as follows:
where m is the number of production data, y i The value of the i-th rolled strip convexity is F6.
5. The hot-rolled plate convexity prediction method based on the improved XGboost as claimed in claim 1, wherein the method comprises the following steps: after the XGboost model is established in the step 3, model prediction is carried out by adopting default parameter combination; the default parameters include the maximum depth max _ depth of the tree, the minimum loss function reduction value gamma of the leaf node creation branch, the feature sampling proportion colsample _ byte of the tree, the sum min _ child _ weight of the minimum weight of the leaf node, the learning rate eta, and the subsampling proportion subsample of the training sample.
6. The hot-rolled plate convexity prediction method based on the improved XGboost as claimed in claim 1, wherein the method comprises the following steps: the specific method of the whale optimization algorithm in the step 4 comprises the following steps:
step 4.1: setting related parameters of a whale optimization algorithm, setting a constant b for adjusting the logarithmic spiral shape to be 1, and setting l as a random number between intervals of < -1,1 >;
step 4.2: setting the whale number N and the maximum iteration number t max Initializing position information, wherein the position of each whale individual represents a parameter combination of an XGboost model;
step 4.3: designing an adaptability function, adopting MSE as a performance index, substituting the performance index into the XGboost model, calculating the adaptability of each individual whale in the population, and reserving the current optimal individual whale;
step 4.4: calculating parameters a and p and coefficient vectors A and C, judging whether p is less than 0.5, if so, performing the step 4.5, otherwise, adopting the bubble net to prey, and updating the trapping position of the bubble net as follows:
X(t+1)＝D′×e bl ×cos(2πl)+X * (t)
D′＝|X * (t)-X(t)|
wherein X * (t) is the position of the current optimal solution, X (t) is the current search individual position, b is the spiral shape parameter;
step 4.5: and judging whether the absolute value of the coefficient vector A is less than 1, if so, enclosing a prey, otherwise, carrying out global search, and updating the position as follows:
D＝|C·X′(t)-X(t)|
X(t+1)＝X′(t)-A·D
wherein A is 2a × r 1 -a，C＝2×r 2 A is linearly reduced from 2 to 0, r in the iterative process 1 And r 2 Is [0,1]]X' (t) represents the position vector of the current optimal solution when surrounding a prey, and represents the position vector of a random individual when searching globally;
step 4.6: after the position updating is finished, calculating the fitness of each whale individual, and updating the optimal whale individual;
step 4.7: judging whether the maximum iteration times is reached, if so, ending to obtain an optimal solution, otherwise, returning to the step 4.3;
step 4.8: outputting an optimal parameter combination to obtain optimal parameters of the XGboost model;
step 4.9: and substituting the test sample into the determined model, and evaluating the accuracy of the model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210515319.2A CN115034437A (en) | 2022-05-11 | 2022-05-11 | Hot rolled plate convexity prediction method based on improved XGboost |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210515319.2A CN115034437A (en) | 2022-05-11 | 2022-05-11 | Hot rolled plate convexity prediction method based on improved XGboost |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115034437A true CN115034437A (en) | 2022-09-09 |
Family
ID=83120582
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210515319.2A Pending CN115034437A (en) | 2022-05-11 | 2022-05-11 | Hot rolled plate convexity prediction method based on improved XGboost |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115034437A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117753795A (en) * | 2024-02-07 | 2024-03-26 | 东北大学 | feedforward control method for hot rolled products with multiple steel types and specifications |
| CN117840232A (en) * | 2024-03-05 | 2024-04-09 | 东北大学 | Hot rolling process width prediction method based on incremental learning |
-
2022
- 2022-05-11 CN CN202210515319.2A patent/CN115034437A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117753795A (en) * | 2024-02-07 | 2024-03-26 | 东北大学 | feedforward control method for hot rolled products with multiple steel types and specifications |
| CN117753795B (en) * | 2024-02-07 | 2024-05-31 | 东北大学 | Feedforward control method for hot rolled products with multiple steel types and specifications |
| CN117840232A (en) * | 2024-03-05 | 2024-04-09 | 东北大学 | Hot rolling process width prediction method based on incremental learning |
| CN117840232B (en) * | 2024-03-05 | 2024-05-31 | 东北大学 | Hot rolling process width prediction method based on incremental learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115034437A (en) | Hot rolled plate convexity prediction method based on improved XGboost | |
| CN107967542B (en) | Long-short term memory network-based electricity sales amount prediction method | |
| CN111241657B (en) | Cold-rolled strip steel plate shape prediction method based on deep learning | |
| CN109902861B (en) | Order production progress real-time prediction method based on double-layer transfer learning | |
| CN112100745B (en) | Automobile girder steel mechanical property prediction method based on LDA theory | |
| CN114897227B (en) | Multi-steel mechanical property forecasting method based on improved random forest algorithm | |
| CN110929347A (en) | Hot continuous rolling strip steel convexity prediction method based on gradient lifting tree model | |
| CN103745273A (en) | Semiconductor fabrication process multi-performance prediction method | |
| CN108817103B (en) | Steel rolling model steel family layer classification optimization method | |
| CN109558677A (en) | A kind of hot rolling strip crown prediction technique based on data-driven | |
| CN109492748B (en) | Method for establishing medium-and-long-term load prediction model of power system based on convolutional neural network | |
| CN108805257A (en) | A kind of neural network quantization method based on parameter norm | |
| CN113128124B (en) | Multi-grade C-Mn steel mechanical property prediction method based on improved neural network | |
| CN111250548B (en) | Board convexity prediction method based on kernel partial least square combined support vector machine | |
| CN114548592A (en) | Non-stationary time series data prediction method based on CEMD and LSTM | |
| CN115470595A (en) | Hot-rolled strip steel convexity prediction method with data and mechanism fused | |
| CN111882114B (en) | Short-time traffic flow prediction model construction method and prediction method | |
| CN117840232B (en) | Hot rolling process width prediction method based on incremental learning | |
| CN113102516A (en) | Hot continuous rolling strip steel head width prediction method integrating rolling mechanism and deep learning | |
| CN115169453A (en) | Hot continuous rolling width prediction method based on density clustering and depth residual error network | |
| CN112308298A (en) | Multi-scenario performance index prediction method and system for semiconductor production line | |
| CN110910528B (en) | Method and device for predicting tensile strength of paper sheet | |
| CN116826745A (en) | Layered and partitioned short-term load prediction method and system in power system background | |
| CN116565850A (en) | Wind power ultra-short-term prediction method based on QR-BLSTM | |
| CN112966870B (en) | Cold rolled silicon steel quality defect prediction method based on neural network |
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 |