CN112036626B - Online forecasting method for edge linear defects of hot rolled strip steel - Google Patents
Online forecasting method for edge linear defects of hot rolled strip steel Download PDFInfo
- Publication number
- CN112036626B CN112036626B CN202010852476.3A CN202010852476A CN112036626B CN 112036626 B CN112036626 B CN 112036626B CN 202010852476 A CN202010852476 A CN 202010852476A CN 112036626 B CN112036626 B CN 112036626B
- Authority
- CN
- China
- Prior art keywords
- training
- data
- strip steel
- rolled strip
- neural network
- 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.)
- Active
Links
- 230000007547 defect Effects 0.000 title claims abstract description 101
- 229910000831 Steel Inorganic materials 0.000 title claims abstract description 82
- 239000010959 steel Substances 0.000 title claims abstract description 82
- 238000013277 forecasting method Methods 0.000 title claims abstract description 10
- 238000012549 training Methods 0.000 claims abstract description 70
- 238000013528 artificial neural network Methods 0.000 claims abstract description 50
- 238000004519 manufacturing process Methods 0.000 claims abstract description 31
- 238000000034 method Methods 0.000 claims abstract description 17
- 238000007781 pre-processing Methods 0.000 claims abstract description 10
- 238000010438 heat treatment Methods 0.000 claims description 59
- 230000006870 function Effects 0.000 claims description 20
- 238000012360 testing method Methods 0.000 claims description 17
- 230000004913 activation Effects 0.000 claims description 13
- 238000010586 diagram Methods 0.000 claims description 12
- 238000002791 soaking Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 11
- 238000009826 distribution Methods 0.000 claims description 10
- 210000002569 neuron Anatomy 0.000 claims description 9
- 238000012795 verification Methods 0.000 claims description 9
- 238000010079 rubber tapping Methods 0.000 claims description 7
- 238000005457 optimization Methods 0.000 claims description 6
- 238000007599 discharging Methods 0.000 claims description 4
- 238000010200 validation analysis Methods 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000007639 printing Methods 0.000 claims description 2
- 230000001537 neural effect Effects 0.000 claims 1
- 238000004458 analytical method Methods 0.000 abstract description 3
- 230000004044 response Effects 0.000 abstract description 2
- 238000005096 rolling process Methods 0.000 description 10
- 230000008569 process Effects 0.000 description 4
- 229910000976 Electrical steel Inorganic materials 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000015572 biosynthetic process Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 239000002253 acid Substances 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001447 compensatory effect Effects 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000005098 hot rolling Methods 0.000 description 1
- 230000001939 inductive effect Effects 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 238000005272 metallurgy Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 239000002436 steel type Substances 0.000 description 1
- 230000037303 wrinkles Effects 0.000 description 1
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"
-
- 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
- G06N3/045—Combinations of 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- General Business, Economics & Management (AREA)
- Mathematical Physics (AREA)
- Marketing (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Tourism & Hospitality (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Development Economics (AREA)
- Primary Health Care (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Manufacturing & Machinery (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Control Of Heat Treatment Processes (AREA)
- Control Of Metal Rolling (AREA)
Abstract
The invention discloses an online forecasting method for edge linear defects of hot rolled strip steel, which mainly comprises the following steps: 1. collecting hot rolled strip steel production data as sample data and preprocessing the sample data; 2. establishing a hot-rolled strip steel edge linear defect prediction model according to the GA-DNN neural network; 3. training and verifying a hot rolled strip steel edge linear defect prediction model; 4. and (3) carrying out online forecasting and analysis on the edge linear defect forecasting model of the hot rolled strip steel by using the GA-DNN neural network. The method has the characteristics of high prediction precision, high response speed, capability of participating in control on line in real time and the like, and has important significance for controlling the surface quality of the hot-rolled strip steel.
Description
Technical Field
The invention belongs to the field of surface quality control of hot-rolled strip steel in the metallurgical rolling technology, and particularly relates to an intelligent online forecasting method for edge linear defects of the hot-rolled strip steel.
Background
With the high-speed development of the industry in China, the use requirement of high-quality hot rolled strip steel is larger and larger, and the surface quality is one of important indexes affecting the quality of hot rolled strip steel products. The edge linear defect is one of the surface quality defects of the hot rolled strip steel, and is expressed as a skin-turning or black line according to the severity. Such defects not only seriously affect the yield but may also affect the production process of the downstream hot rolling process (e.g., acid mill train).
There are several documents related to control and analysis of linear defects in hot rolled strip. For example: the SSP module is utilized to improve the edge linear defect of the hot rolled strip steel (metallurgical equipment 2011, stage 2: 28-30+49), the formation mechanism of the edge defect of the hot rolled strip steel and the research status quo (Henan metallurgy 2008, volume 16, stage 3: 1-4+27) are considered that the lateral fold of the intermediate blank is the main reason for forming the edge linear defect, the severity of the narrow fold is directly related to the heating temperature and the heating time, and corresponding compensatory solving measures are provided; the research of the formation mechanism of the edge linear defects in the rolling process (university of Chongqing university journal 2018, volume 32, 4 th period: 107-110) considers that the temperature distribution of the width direction of the plate blank in the rough rolling process is uneven, the temperature of the edge of the plate blank is too low, the plate blank easily enters into a two-phase region in advance, the deformation resistance is reduced, the deformation amount exceeding other positions is caused in the region, in the subsequent rolling process, the side metal of the intermediate blank is continuously flattened to the upper surface and the lower surface, and finally wrinkles and linear defects are formed; "analysis of hot rolled sheet surface skin-lifting defect" ("physical test" 2009, volume 27, stage 1: 46-51) considers that the occurrence of edge linear defects is directly related to the original defects in the steel region.
From the current research results, the influence factors of the edge linear defects are complex and variable, but the edge linear defects can be divided into a steel area and a rolling area. Regarding the rolling zone, although the knowledge of the specific generation mechanism of the edge linear defects is not exactly the same, it is generally considered that the heating temperature, the heating time and the rough rolling temperature are key influencing factors. In addition, although various solving measures have been proposed at home and abroad, equipment or process parameter optimization is basically performed empirically, and no online prediction of edge linear defects has been performed so far.
In summary, the invention uses the heating temperature, heating time and rough rolling outlet temperature of the rolling zone as main influencing factors, establishes an intelligent online prediction method for the edge linear defects of the hot rolled strip steel by adopting an intelligent method according to actual production data, can predict whether the defects occur in real time according to current process parameters, and performs parameter optimization adjustment based on the intelligent online prediction method, thereby having important significance for controlling the surface quality of the hot rolled strip steel in the production process.
Disclosure of Invention
The invention aims to provide an online forecasting method for edge linear defects of hot rolled strip steel. According to the deep neural network model, the intelligent online prediction model of the hot rolled strip steel edge linear defect neural network is established based on actual production data. The model mainly considers three aspects of heating temperature, heating time and rough rolling outlet temperature, and specifically comprises the following steps: the method comprises the steps of feeding furnace temperature, preheating section temperature, first heating temperature, second heating temperature, third heating temperature, discharging temperature, R2 feedback temperature, preheating time, first heating time, second heating time, third heating time, soaking time and 13 variable parameters of product steel types, so that the occurrence condition of the edge linear defects of the hot rolled strip steel is predicted. The method has the characteristics of high prediction precision, high response speed, real-time online participation in control and the like, and has important significance on actual production.
An online forecasting method for edge linear defects of hot rolled strip steel comprises the following steps:
s1, collecting hot rolled strip steel production data as sample data and preprocessing the sample data:
s11, collecting on-site actual production data and establishing an original data set, wherein the production data comprises the furnace charging temperature T f Preheating section temperature T p A heating temperature T 1 Two heating temperatures T 2 Three heating temperatures T 3 Tapping temperature T o R2 feedback temperature T R Preheating time S R One time of addition S 1 Two times of addition S 2 Three times S 3 Soaking time S a The number of the steel m and the corresponding linear defect occurrence condition O is selected as n groups, and the obtained original data set dataset is as follows:
{(T f ,T p ,T 1 ,T 2 ,T 3 ,T o ,T R ,S R ,S 1 ,S 2 ,S 3 ,S a ,m)|O} i (i=1,2,3···n);
s12, preprocessing an original data set, namely, performing linear defect data uneven distribution processing and abnormal data eliminating processing to obtain n 'groups of data sets dataset1, wherein the n' groups of data sets dataset1 are as follows:
{(T f ,T p ,T 1 ,T 2 ,T 3 ,T o ,T R ,S R ,S 1 ,S 2 ,S 3 ,S a ,m)|O} j (j=1,2,3···n′);
s13, performing One-Hot independent encoding treatment on a steel grade m of a non-digital type in a dataset1 to convert the steel grade m into a digital type;
s14, dividing the data set dataset1 into a training sample and a test set, randomly extracting 86% of data to be used as the data set dataset2 of the training sample, and the rest data to be used as the test set testset;
s2, establishing an online forecasting model of the edge linear defects of the hot-rolled strip steel according to the GA-DNN neural network:
s21, determining a DNN neural network structure, which specifically comprises the following steps:
s211, firstly determining that the input layer variable of the neural network has the furnace-in temperature T f Preheating section temperature T p A heating temperature T 1 Two heating temperatures T 2 Three heating temperatures T 3 Tapping temperature T o R2 feedback temperature T R Preheating time S R One time of addition S 1 Two times of addition S 2 Three times S 3 Soaking time S a And steel class m, so the number of neuron nodes of the input layer is A 1 =13;
S212, the output of the neural network is the occurrence condition O of strip steel linear defects, so the node number of the output layer of the neural network is A k =1;
S213, determining the layer number of the hidden layer and the node number A of each layer 2 ,A 3 …A k-1 ;
S214, selecting an activation function activation, an error loss function loss, an optimizer and a matrix function of each layer;
s215, setting a learning rate lr;
s216, determining a small-batch training sample batch and a training step number Epoch;
s22, setting GA algorithm parameters, wherein the GA algorithm parameters comprise an initialized population scale Q, a maximum iteration number N and a crossover probability p 1 Probability of variation p 2 ;
S3, training an online forecasting model of the edge linear defects of the hot rolled strip steel:
s31, dividing a data set dataset2 of a training sample into a training set and a verification set, randomly selecting 80% of the data set dataset2 as a training set trainset of the GA-DNN neural network, and the remaining 20% as a verification set validation set;
s32, training the GA-DNN neural network, and stopping training when the network model reaches the training step number;
s33, after model training is finished, an error loss diagram and an accuracy diagram of a training set trainset and a verification set validlation set are made, and whether the average loss error of the network model is smaller than 0.5 and whether the accuracy can meet the production requirement of 85% are judged;
s34, utilizing the GA-DNN neural network after training, and according to the parameters T in the training set trainset and the test set testset f ,T p ,T 1 ,T 2 ,T 3 ,T o ,T R ,S R ,S 1 ,S 2 ,S 3 ,S a Comparing the m predicted linear defect occurrence condition data value with the data values of the real linear defect occurrence conditions in the training set and the test set and making a difference value, wherein the difference value of 0 indicates that the prediction is correct, and the difference value of + -1 indicates that the prediction is wrong, so as to obtain an actual error distribution diagram of the training set and the test set, and judging whether the 85% precision production requirement is met;
s35, judging whether the model meets the precision requirement, if so, storing the GA-DNN network model as an online prediction model M of the edge linear defects of the hot-rolled strip steel, and printing the weight threshold coefficient value of each layer of the model; if one of S33 and S34 is not satisfied, returning to S2 to adjust the network structure and the optimization parameters, and retraining the network;
s4, intelligent online forecasting and analyzing of the linear defects at the edge of the hot rolled strip steel:
s41, firstly loading the online forecasting model M of the edge linear defect of the hot-rolled strip steel saved in the step S35, and embedding the model M into a control system of a hot-rolled strip steel production site;
s42, in the production process, according to the current rolled steel grade, according to the values of the furnace feeding temperature, the preheating section temperature, the first heating temperature, the second heating temperature, the third heating temperature, the furnace discharging temperature, the R2 feedback temperature, the preheating time, the first heating time, the second heating time, the third heating time and the soaking time which are automatically fed back by the system, predicting the occurrence condition of the linear defects of the current strip steel in real time;
s43, analyzing numerical intervals of linear defects generated by the hot rolled strip steel according to the online prediction model M of the linear defects at the edge of the hot rolled strip steel, and guiding actual production according to the numerical intervals of the linear defects generated by the hot rolled strip steel induced by various factors.
Preferably, the preprocessing of the dataset in step S12 specifically includes the following steps:
s121, adopting an automatic screening method, and processing the acquired original data set dataset according to the occurrence of the linear defect and the non-occurrence of the linear defect;
s122, randomly extracting linear defect data and non-linear defect data according to the ratio of 1:1; carrying out disordered merging treatment on the two;
s123, eliminating the data containing the missing items, and obtaining a new available data set dataset1 after finishing.
Preferably, in step S35, the optimization and adjustment of the GA-DNN neural network structure and parameters specifically includes the following steps:
s351, adjusting the number of hidden layers of the neural network and the number of nodes of the neural network of each hidden layer;
s352, adjusting activation function activation of the output layer of each neural network layer;
s353, adjusting a network training optimizer;
s354, adjusting a small batch of training samples batch and training step number Epoch;
s355, introducing regularization treatment into each hidden layer of the GA-DNN neural network;
s356, conducting dropout processing on the input layer of the GA-DNN neural network.
Compared with the prior art, the invention has the following beneficial effects:
the online forecasting method for the edge linear defects of the hot rolled strip steel can forecast the occurrence of the edge linear defects of the strip steel in real time according to actual production data. The method has high accuracy, strong generalization, high execution speed and high portability, meets the industrial production requirement, can be well embedded in a production field control system to participate in guiding production, has better effect on the improvement of the edge linear defects of the hot-rolled strip steel, and has important significance on reducing the cutting rate, improving the yield and improving the core competitiveness of the product.
Drawings
FIG. 1 is a flow chart of a GA-DNN neural network for hot rolled strip edge linear defects according to the present invention;
FIG. 2 is a block diagram of a GA-DNN neural network;
FIG. 3 is a graph of error loss for a training set and a validation set of a neural network according to an embodiment of the present invention;
FIG. 4 is a precision graph of a neural network training set and a validation set in accordance with an embodiment of the present invention;
FIG. 5 is a diagram of the actual error distribution of the training set according to an embodiment of the present invention;
FIG. 6 is a graph showing the actual error distribution of a test set according to an embodiment of the present invention; and
FIG. 7 is a flow chart of the method for online forecasting of edge linear defects of hot rolled strip according to the present invention.
Detailed Description
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
As shown in FIG. 1, the flow chart of the intelligent online forecasting method for the edge linear defects of the hot rolled strip steel comprises the following steps:
s1, collecting hot rolled strip steel production data as sample data and preprocessing the sample data:
s11, collecting field actual data and establishing an original data set, wherein the method specifically comprises the following steps: temperature T of entering furnace f Preheating section temperature T p A heating temperature T 1 Two heating temperatures T 2 Three heating temperatures T 3 Tapping temperature T o R2 feedback temperature T R Preheating time S R One time of addition S 1 Two times of addition S 2 Three times S 3 Soaking time S a And steel grade m and corresponding linear defect occurrence condition O, wherein 0 represents that linear defect is generated, and 1 represents that linear defect is not generatedThe number of which is n=12538 groups, the raw dataset is:
{(T f ,T p ,T 1 ,T 2 ,T 3 ,T o ,T R ,S R ,S 1 ,S 2 ,S 3 ,S a ,m)|O} i i=1,2,3…12538。
s12, preprocessing an original data set dataset, wherein the preprocessing specifically comprises the following steps: the linear defect data uneven distribution processing and abnormal data eliminating processing comprise the following specific processes:
s121, adopting an automatic screening method, and processing the acquired original data set dataset according to the occurrence of the linear defect and the non-occurrence of the linear defect;
s122, randomly extracting linear defect data and non-linear defect data according to the ratio of 1:1; carrying out disordered merging treatment on the two;
s123, deleting the data containing the missing items, and finishing to obtain a data set dataset1, wherein the number of the dataset is n' =4200 groups, namely { (T) f ,T p ,T 1 ,T 2 ,T 3 ,T o ,T R ,S R ,S 1 ,S 2 ,S 3 ,S a ,m)|O} j j=1, 2,3 … 4200, the data structure form of which is shown in table 1.
S13, performing One-Hot independent encoding treatment on the non-digital type of the steel grade m in the dataset1, so that the non-digital type is converted into a digital type.
S14, dividing the data set dataset1 into training samples and test sets, randomly extracting 86% of the data set dataset2 serving as model training samples, rounding hundred bits, adding up to 3600 groups, and taking the remaining 600 groups as model test set sample data testset.
TABLE 1 partial data in dataset1
S2, establishing an intelligent online prediction model of the edge linear defects of the hot-rolled strip steel according to the GA-DNN neural network:
s21, determining a DNN neural network structure, which specifically comprises the following steps:
s211, firstly determining that the input layer variable of the neural network has the furnace-in temperature T f Preheating section temperature T p A heating temperature T 1 Two heating temperatures T 2 Three heating temperatures T 3 Tapping temperature T o R2 feedback temperature T R Preheating time S R One time of addition S 1 Two times of addition S 2 Three times S 3 Soaking time S a And steel class m, so the number of neuron nodes of the input layer is A 1 =13;
S212, because the number of the input variables is only 13, the number of the input layer neuron nodes is small, the number of the hidden layer neuron nodes cannot be too large, otherwise, the fitting phenomenon can occur, after the result comparison of multiple model training and the model structure adjustment, the final decision is made to adopt a k=5 layer neural network, and the number A of the hidden layer neuron nodes is set 2 =200,A 3 =80,A 4 =50, the activation functions of the three hidden layers are all Relu functions;
s213, the output of the neural network is the occurrence of strip steel linear defect O, so the node number of the output layer of the neural network is A k =1, the activation function of the output layer selects Softmax function;
s214, because the output layer is the strip steel linear defect occurrence condition, only two possibilities exist, the loss function selects the Binary-cross sentropy, the optimizer adopts Adam, the matrix function adopts metrics, the learning rate of the optimizer is set to be 0.001, a small batch of training samples batch is determined to be 32, the training times Epoch is determined to be 200 times, L2 regularization is introduced into the neural network hidden layer, the setting value is 0.005, the drop function is introduced into the neural network hidden layer, and the setting value is 0.02.
S22, setting GA algorithm parameters, specifically comprising the following steps:
s221, setting a population scale Q as 80 according to a weight threshold scale of the neural network, and coding a randomly generated initial value;
s222, taking the randomly generated initial value as an initial weight and a threshold value of the neural network, predicting output after training the neural network, taking the predicted output and an error between the predicted output and the expected output as an individual fitness value F,where y is the expected output of the ith node of the neural network; o (O) i A prediction output for the i-th node; k is a coefficient;
s223, performing crossover and mutation operations on the selected individuals with higher fitness values, and setting crossover probability p 1 Is 0.4 and variation probability p 2 0.2;
and S224, calculating the fitness value of the individual after the reproduction, setting the maximum iteration number N as 100 times, transmitting the finally optimized weight threshold value to the neural network after the iteration number is reached, otherwise, executing S223.
S3, training an intelligent online forecasting model of the edge linear defects of the hot rolled strip steel:
s31, dividing a training sample dataset2 into a training set and a verification set, randomly selecting 80% of the training samples as GA-DNN neural network training set sample trainset to share 2880 groups of data, and using the rest 720 groups of data as verification set sample validization set.
S32, training the GA-DNN neural network, and setting model training times;
s321, the variable of the data is put into the furnace temperature T f Preheating section temperature T p A heating temperature T 1 Two heating temperatures T 2 Three heating temperatures T 3 Tapping temperature T o R2 feedback temperature T R Preheating time S R One time of addition S 1 Two times of addition S 2 Three times S 3 Soaking time S a Inputting the steel class m into an input layer of the model, randomly distributing each layer of neurons to an initial weight coefficient between each input and each neuron when training is started, and then calculating allThe sum of the products of the input and the weight coefficient is calculated and transmitted through an activation function, the last output is used as the input of the next layer to transmit the variable to the next hidden layer and the activation function, and the like, and the variable is transmitted to the output layer and the activation function of the output layer all the time, so that the first round of training process is completed;
s322, after the forward propagation of the first round is finished, the model takes the occurrence condition O of the linear defect as the output of the network, then compares the linear defect occurrence condition O with the true defect occurrence condition in the training data, and calculates the loss error of the linear defect prediction and the true occurrence condition of the strip steel trained in the round through the set loss function;
s323, setting an optimizer, carrying out back propagation through errors obtained by the loss function, updating the weight coefficient of each layer of neurons through chained derivation to enable the loss value to be minimum, and continuously transmitting training data into the model to carry out next training after parameter updating is finished until the set training times are reached, and stopping training.
S33, after model training is finished, a training set trainset and verification set validlation set error loss diagram is made, a precision diagram is shown in FIG. 4, and the result shows that the average loss error of the network model is smaller than 0.5, the average precision of the model can reach 90%, and the production requirement is met.
S34, utilizing the GA-DNN neural network after training, and according to the parameters T on the training set trainset and the test set testset f ,T p ,T 1 ,T 2 ,T 3 ,T o ,T R ,S R ,S 1 ,S 2 ,S 3 ,S a The m predicted linear defect occurrence data value is compared with the real linear defect occurrence data value on the training set and the test set and is subjected to difference, the value of 0 indicates that the prediction is correct, the value of + -1 indicates that the prediction is wrong, so that an actual error distribution diagram of data of 3600 groups of the training set is obtained, the actual error distribution diagram of data of 600 groups of the test set is shown in fig. 5, and the result shows that the model precision reaches 90% and meets the production requirement.
S35, the model meets the requirements of both S33 and S34, so that the GA-DNN neural network model is saved as an intelligent online prediction model M for the edge linear defects of the hot-rolled strip steel, and the weight threshold coefficient values of all layers of the model are printed.
S4, online forecasting and analyzing the linear defects at the edge of the hot rolled strip steel:
s41, loading the intelligent online forecasting model M of the edge linear defects of the hot-rolled strip steel stored in S35, and embedding the intelligent online forecasting model M into a control system of a hot-rolled strip steel production site.
S42, in the production process, according to the current rolled steel grade, the linear defect occurrence condition of the current strip steel is predicted in real time according to the parameter values of the furnace feeding temperature, the preheating section temperature, the first heating temperature, the second heating temperature, the third heating temperature, the furnace discharging temperature, the R2 feedback temperature, the preheating time, the first heating time, the second heating time, the third heating time and the soaking time which are automatically fed back by the system.
S43, analyzing numerical intervals of linear defects generated by the hot rolled strip steel according to the intelligent online prediction model M of the linear defects at the edge of the hot rolled strip steel. Taking silicon steel 270 as an example, the result is obtained by an intelligent online forecasting model M of the linear defects at the edge of the hot rolled strip steel, and the factors inducing the linear defects to be generated are one heating temperature, two heating temperatures, three heating temperatures, one heating time, three heating times and soaking times, and the forecasting numerical intervals are shown in Table 2, so that the silicon steel 270 is optimized according to the heating process in actual production.
TABLE 2 parameter interval forecast for the generation of line defects at the edge of silicon steel 270
The above examples are only illustrative of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention, and various modifications and improvements made by those skilled in the art to the technical solution of the present invention should fall within the scope of protection defined by the claims of the present invention without departing from the spirit of the present invention.
Claims (3)
1. An online forecasting method for edge linear defects of hot rolled strip steel is characterized by comprising the following steps:
s1, collecting hot rolled strip steel production data as sample data and preprocessing the sample data:
s11, collecting on-site actual production data and establishing an original data set, wherein the production data comprises the furnace charging temperature T f Preheating section temperature T p A heating temperature T 1 Two heating temperatures T 2 Three heating temperatures T 3 Tapping temperature T o R2 feedback temperature T R Preheating time S R One time of addition S 1 Two times of addition S 2 Three times S 3 Soaking time S a The number of the steel m and the corresponding linear defect occurrence condition O is selected as n groups, and the obtained original data set dataset is as follows:
{(T f ,T p ,T 1 ,T 2 ,T 3 ,T o ,T R ,S R ,S 1 ,S 2 ,S 3 ,S a ,m)|O} i (i=1,2,3···n);
s12, preprocessing an original data set, namely, performing linear defect data uneven distribution processing and abnormal data eliminating processing to obtain n 'groups of data sets dataset1, wherein the n' groups of data sets dataset1 are as follows:
{(T f ,T p ,T 1 ,T 2 ,T 3 ,T o ,T R ,S R ,S 1 ,S 2 ,S 3 ,S a ,m)|O} j (j=1,2,3···n′);
s13, performing One-Hot independent encoding treatment on a steel grade m of a non-digital type in a dataset1 to convert the steel grade m into a digital type;
s14, dividing the data set dataset1 into a training sample and a test set, randomly extracting 86% of data to be used as the data set dataset2 of the training sample, and the rest data to be used as the test set testset;
s2, establishing an online forecasting model of the edge linear defects of the hot-rolled strip steel according to the GA-DNN neural network:
s21, determining a DNN neural network structure, which specifically comprises the following steps:
s211, firstly determining an input layer of a neural networkThe variable is the temperature T of the furnace f Preheating section temperature T p A heating temperature T 1 Two heating temperatures T 2 Three heating temperatures T 3 Tapping temperature T o R2 feedback temperature T R Preheating time S R One time of addition S 1 Two times of addition S 2 Three times S 3 Soaking time S a And steel class m, so the number of neuron nodes of the input layer is A 1 =13;
S212, the output of the neural network is the occurrence condition O of strip steel linear defects, so the node number of the output layer of the neural network is A k =1;
S213, determining the layer number of the hidden layer and the node number A of each layer 2 ,A 3 …A k-1 ;
S214, selecting an activation function activation, an error loss function loss, an optimizer and a matrix function of each layer;
s215, setting a learning rate lr;
s216, determining a small-batch training sample batch and a training step number Epoch;
s22, setting GA algorithm parameters, wherein the GA algorithm parameters comprise an initialized population scale Q, a maximum iteration number N and a crossover probability p 1 Probability of variation p 2 ;
S3, training an online forecasting model of the edge linear defects of the hot rolled strip steel:
s31, dividing a data set dataset2 of a training sample into a training set and a verification set, randomly selecting 80% of the data set dataset2 as a training set trainset of the GA-DNN neural network, and the remaining 20% as a verification set validation set;
s32, training the GA-DNN neural network, and stopping training when the network model reaches the training step number;
s33, after model training is finished, an error loss diagram and an accuracy diagram of a training set trainset and a verification set validlation set are made, and whether the average loss error of the network model is smaller than 0.5 and whether the accuracy can meet the production requirement of 85% are judged;
s34, utilizing the GA-DNN neural network after training, and according to the training set trainset and the test set testsetParameter T of (2) f ,T p ,T 1 ,T 2 ,T 3 ,T o ,T R ,S R ,S 1 ,S 2 ,S 3 ,S a Comparing the m predicted linear defect occurrence condition data value with the data values of the real linear defect occurrence conditions in the training set and the test set and making a difference value, wherein the difference value of 0 indicates that the prediction is correct, and the difference value of + -1 indicates that the prediction is wrong, so as to obtain an actual error distribution diagram of the training set and the test set, and judging whether the 85% precision production requirement is met;
s35, judging whether the model meets the precision requirement, if so, storing the GA-DNN network model as an online prediction model M of the edge linear defects of the hot-rolled strip steel, and printing the weight threshold coefficient value of each layer of the model; if one of S33 and S34 is not satisfied, returning to S2 to adjust the network structure and the optimization parameters, and retraining the network;
s4, intelligent online forecasting and analyzing of the linear defects at the edge of the hot rolled strip steel:
s41, firstly loading the online forecasting model M of the edge linear defect of the hot-rolled strip steel saved in the step S35, and embedding the model M into a control system of a hot-rolled strip steel production site;
s42, in the production process, according to the current rolled steel grade, according to the values of the furnace feeding temperature, the preheating section temperature, the first heating temperature, the second heating temperature, the third heating temperature, the furnace discharging temperature, the R2 feedback temperature, the preheating time, the first heating time, the second heating time, the third heating time and the soaking time which are automatically fed back by the system, predicting the occurrence condition of the linear defects of the current strip steel in real time;
s43, analyzing numerical intervals of linear defects generated by the hot rolled strip steel according to the online prediction model M of the linear defects at the edge of the hot rolled strip steel, and guiding actual production according to the numerical intervals of the linear defects generated by the hot rolled strip steel induced by various factors.
2. The method for online forecasting of edge linear defects of hot rolled steel strip according to claim 1, wherein the preprocessing of the dataset in step S12 specifically comprises the following steps:
s121, adopting an automatic screening method, and processing the acquired original data set dataset according to the occurrence of the linear defect and the non-occurrence of the linear defect;
s122, randomly extracting linear defect data and non-linear defect data according to the ratio of 1:1; carrying out disordered merging treatment on the two;
s123, eliminating the data containing the missing items, and obtaining a new available data set dataset1 after finishing.
3. The online forecasting method of hot-rolled strip edge linear defects according to claim 1, wherein the optimization and adjustment of the GA-DNN neural network structure and parameters in step S35 specifically comprises the following steps:
s351, adjusting the number of hidden layers of the neural network and the number of nodes of the neural network of each hidden layer;
s352, adjusting activation function activation of the output layer of each neural network layer;
s353, adjusting a network training optimizer;
s354, adjusting a small batch of training samples batch and training step number Epoch;
s355, introducing regularization treatment into each hidden layer of the GA-DNN neural network;
s356, conducting dropout processing on the input layer of the GA-DNN neural network.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010852476.3A CN112036626B (en) | 2020-08-21 | 2020-08-21 | Online forecasting method for edge linear defects of hot rolled strip steel |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010852476.3A CN112036626B (en) | 2020-08-21 | 2020-08-21 | Online forecasting method for edge linear defects of hot rolled strip steel |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112036626A CN112036626A (en) | 2020-12-04 |
CN112036626B true CN112036626B (en) | 2023-08-22 |
Family
ID=73580555
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010852476.3A Active CN112036626B (en) | 2020-08-21 | 2020-08-21 | Online forecasting method for edge linear defects of hot rolled strip steel |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112036626B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110097543A (en) * | 2019-04-25 | 2019-08-06 | 东北大学 | Surfaces of Hot Rolled Strip defect inspection method based on production confrontation network |
CN110110839A (en) * | 2019-04-19 | 2019-08-09 | 燕山大学 | Cold rolling of strip steel lateral thickness difference forecasting procedure based on PSO-LM-BP neural network |
EP3561614A1 (en) * | 2018-04-23 | 2019-10-30 | Sensima Inspection Sàrl | Process for training a self-learning machine in a defect monitoring system |
-
2020
- 2020-08-21 CN CN202010852476.3A patent/CN112036626B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP3561614A1 (en) * | 2018-04-23 | 2019-10-30 | Sensima Inspection Sàrl | Process for training a self-learning machine in a defect monitoring system |
CN110110839A (en) * | 2019-04-19 | 2019-08-09 | 燕山大学 | Cold rolling of strip steel lateral thickness difference forecasting procedure based on PSO-LM-BP neural network |
CN110097543A (en) * | 2019-04-25 | 2019-08-06 | 东北大学 | Surfaces of Hot Rolled Strip defect inspection method based on production confrontation network |
Non-Patent Citations (1)
Title |
---|
遗传算法优化神经网络的热轧带钢弯辊力预报模型;王振华;龚殿尧;李广焘;张殿华;;东北大学学报(自然科学版)(12);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN112036626A (en) | 2020-12-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2023130666A1 (en) | Strip steel plate convexity prediction method based on data-driving and mechanism model fusion | |
CN102937784A (en) | Artificial neural network based method for controlling online prediction of casting billet quality | |
CN105740467A (en) | Mining method for C-Mn steel industry big data | |
CN102540879A (en) | Multi-target evaluation optimization method based on group decision making retrieval strategy | |
CN113849020B (en) | Billet heating curve design method and device based on artificial intelligence algorithm | |
CN115608793B (en) | Finish rolling temperature regulation and control method for mechanism fusion data | |
CN116852665A (en) | Injection molding process parameter intelligent adjusting method based on mixed model | |
CN116048028A (en) | Technological parameter optimization method based on reinforcement learning | |
CN114871000B (en) | Flotation dosing self-adaptive regulation and control method | |
CN113589693A (en) | Cement industry decomposing furnace temperature model prediction control method based on neighborhood optimization | |
CN113095484A (en) | Stock price prediction method based on LSTM neural network | |
CN112036626B (en) | Online forecasting method for edge linear defects of hot rolled strip steel | |
CN107716560A (en) | A kind of new Hot Strip Rolling load distribution method | |
CN110110447B (en) | Method for predicting thickness of strip steel of mixed frog leaping feedback extreme learning machine | |
CN106886799A (en) | A kind of continuous annealing band steel quality online test method based on hybrid integrated study | |
Li et al. | Dual ensemble online modeling for dynamic estimation of hot metal silicon content in blast furnace system | |
CN113901725A (en) | Intelligent prediction system and method for continuous rolling process of seamless steel pipe based on Matlab | |
CN112598026B (en) | Continuous annealing production process fault diagnosis method based on unbalanced data deep learning | |
CN117000780A (en) | Self-adaptive cold-rolled strip multichannel shape forecasting method | |
LU501336B1 (en) | Method for predicting acrylamide in fried dough based on an optimized bp neural network | |
CN111222627A (en) | Air knife distance data driving prediction method for plating control | |
CN116108932A (en) | Method for establishing fusion model of steel production process data and mechanism | |
Qazani et al. | Multiobjective optimization of roll-forming procedure using NSGA-II and type-2 fuzzy neural network | |
CN109142167B (en) | Method for establishing online robust regular soft measurement model of ore grinding granularity | |
CN111861041A (en) | Method for predicting dynamic recrystallization type flowing stress of Nb microalloyed steel |
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 | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20240223 Address after: Unit 901-01, 8th Floor, Unit 1, Building 1, No. 3 Changtong Road, Chaoyang District, Beijing, 100000 Patentee after: Beijing Mitian Technology Co.,Ltd. Country or region after: China Address before: 066004 No. 438 west section of Hebei Avenue, seaport District, Hebei, Qinhuangdao Patentee before: Yanshan University Country or region before: China |