Search

CN-122021226-A - Automatic strip steel tension adjusting method in cold rolling annealing process based on neural network

CN122021226ACN 122021226 ACN122021226 ACN 122021226ACN-122021226-A

Abstract

A strip steel tension automatic adjusting method in a cold rolling annealing process based on a neural network comprises the steps of 1) extracting and analyzing historical operation data of strip steel tension of a cold rolling annealing furnace, classifying and summarizing parameters related to the historical operation data of the strip steel tension and tension adjustment as input of a neural network model, normalizing the historical operation data of the strip steel tension, 2) setting a tension automatic setting model based on a multi-layer neural network, simulating and learning the historical operation data of the strip steel tension of an operator by means of the neural network, forming a usable tension automatic setting model by means of offline training, and 3) checking the prediction effect of the tension automatic setting model by means of field actual production data, and training the tension automatic setting model by means of prediction errors. According to the invention, the burden of site operators can be effectively reduced, and the invention has higher application value.

Inventors

  • CHEN QIJUN
  • SUN YUEQING
  • YUAN WENZHEN
  • LI YAN
  • LIU DECHENG
  • PENG JUN
  • SHU SHAOLONG

Assignees

  • 宝山钢铁股份有限公司
  • 同济大学

Dates

Publication Date
20260512
Application Date
20241111

Claims (11)

  1. 1. A strip steel tension automatic adjustment method in a cold rolling annealing process based on a neural network is characterized by comprising the following steps of: The automatic adjusting method of the strip steel tension comprises the following steps: Step 1) extracting and analyzing historical operation data of strip steel tension of a cold-rolling annealing furnace, classifying and summarizing parameters related to the historical operation data of the strip steel tension and tension adjustment as input of a neural network model, and carrying out normalization processing on the historical operation data of the strip steel tension; Step 2) setting a tension automatic setting model based on a multi-layer neural network, simulating and learning historical operation data of strip steel tension of an operator by means of the neural network, and forming an available tension automatic setting model through offline training; Step 3) checking the prediction effect of the automatic tension setting model by means of on-site actual production data, training the automatic tension setting model by using the prediction error, The automatic adjustment method of the strip steel tension in the cold rolling annealing process based on the neural network is provided for adjusting and setting the strip steel tension in the cold rolling annealing process based on the neural network.
  2. 2. The method for automatically adjusting the tension of the strip steel in the cold rolling annealing process based on the neural network according to claim 1, wherein the method comprises the following steps: The historical operation data of the strip steel tension related to tension adjustment comprises the following steel plate specification data: Thickness, width, steel grade, plate shape, annealing furnace temperature, production speed and real-time plate temperature parameters of the steel plate.
  3. 3. The method for automatically adjusting the tension of the strip steel in the cold rolling annealing process based on the neural network according to claim 1, wherein the method comprises the following steps: In step 1), extracting and analyzing historical operation data of strip steel tension in the cold rolling annealing process, carrying out normalization processing on steel plate specification data and on-site actual production data, summarizing and forming steel plate thickness, width, steel grade, plate type and annealing furnace temperature, production speed and real-time plate temperature parameter steel plate specification data as model input through classifying and analyzing the historical operation data, and simultaneously providing a normalization processing method of the input data so as to avoid influence of different parameter dimension differences on training results.
  4. 4. A method for automatically adjusting the tension of a strip steel in a cold rolling annealing process based on a neural network as claimed in claim 1 or 3, wherein: In order to eliminate the dimension influence among data and solve the comparability among data indexes, the following normalization method is adopted for each item of input data: Where x= [ x 1 ,...,x n ] is the input data of the normalized neural network, X min ,x max is the minimum and maximum value of the original input data, respectively.
  5. 5. A method for automatically adjusting the tension of a strip steel in a cold rolling annealing process based on a neural network as claimed in claim 1 or 3, wherein: the full-connection neural network model is adopted to predict the tension, the neural network model constructed by adopting a 5-layer full-connection structure is determined through offline training and testing, the number of nodes of an hidden layer is determined by means of an empirical formula (2), the best performance of the model on a tension prediction task is ensured, Wherein h is the number of hidden layer nodes, the maximum is not more than 30, M is the number of nodes of the input layer, L is the number of nodes of the output layer, A is an adjustment constant between 1 and 10, The hidden layer activation function of the neural network adopts a Sigmoid function: The output layer activation function employs Relu functions: The input layer is used for receiving normalized input data x= [ x 1 ,...,x n ] and transmitting the normalized input data x= [ x 1 ,...,x n ] to a subsequent hidden layer, the hidden layer further processes the input layer data and transmits the input layer data to the output layer, and the output layer synthesizes all data sent by the hidden layer and then generates tension prediction data of each section of the annealing furnace.
  6. 6. The method for automatically adjusting the tension of the strip steel in the cold rolling annealing process based on the neural network according to claim 5, wherein the method comprises the following steps: input information x 1 ,...,x n is passed through the weighted connection, where w is the connection weight, the total input received by the neuron Will be compared to the neuron threshold b and then processed by a sigmoid activation function (3) to produce the output of the neuron.
  7. 7. The method for automatically adjusting the tension of the strip steel in the cold rolling annealing process based on the neural network according to claim 5, wherein the method comprises the following steps: the number of the nodes of the input layer is equal to the number of the input data, the number of the nodes of the output layer is equal to the number of the output data, the number of the nodes of the hidden layer is determined by an empirical formula (2), The hidden layer and the output layer neurons are functional neurons with an activation function, and the learning process of the neural network is a process of continuously adjusting the connection weight and the threshold value between the neurons.
  8. 8. The method for automatically adjusting the tension of the strip steel in the cold rolling annealing process based on the neural network according to claim 5 or 7, wherein the method comprises the following steps: The predicted mean square error is used as a neural network anti-transmission signal by comparing a network predicted result with an actual operator tension set value, and the weight among the nodes is continuously corrected to enable the predicted result to approach to the actual set value, and finally the neural network training work can be completed when the predicted error meets the performance requirement.
  9. 9. The method for automatically adjusting the tension of the strip steel in the cold rolling annealing process based on the neural network according to claim 1, 3 or 5, wherein the method comprises the following steps: Inputting the data preprocessed in the step 1 into the multi-layer neural network set in the step 2, and assuming that the predicted tension output of the neural network is Y i and the real tension data of the strip steel is O i , the accumulated mean square error of the neural network is as follows: Where n represents the number of training samples, |·| 2 represents the 2-norm of the vector.
  10. 10. The method for automatically adjusting the tension of the strip steel in the cold rolling annealing process based on the neural network according to claim 1, 8 or 9, wherein the method comprises the following steps: training the neural network by adopting error feedback learning to minimize an accumulated error E, so that the actual output of an output neuron in the network approximates to the target output; The learning process can be realized by a back propagation algorithm, and the weight and the threshold value are adjusted in the negative gradient direction of the gradient based on the gradient descent strategy; And after the accumulated mean square error E of the neural network meets the control accuracy requirement of the unit, the training of the neural network can be completed.
  11. 11. The method for automatically adjusting the tension of the strip steel in the cold rolling annealing process based on the neural network according to claim 10, wherein the method comprises the following steps: And inputting steel plate specification data in the cold rolling annealing process, namely the thickness, width, steel grade and plate shape of the steel plate, the temperature of an annealing furnace, the production speed and real-time plate temperature parameters in real time for the fully-connected neural network after training, so as to obtain a tension setting predicted value, and further setting the tension setting predicted value as a control parameter to a cold rolling unit, thereby completing the automatic adjustment of the strip steel tension in the cold rolling annealing process based on the neural network.

Description

Automatic strip steel tension adjusting method in cold rolling annealing process based on neural network Technical Field The invention relates to the technical field of metallurgical continuous annealing, in particular to a method for automatically adjusting the tension of (middle) strip steel in a cold rolling annealing process based on a neural network. Background The integral automation degree of the continuous annealing unit of the current domestic metallurgical cold rolling mill is higher, but the intelligent degree is slightly insufficient. Taking the continuous annealing process of the strip steel as an example, the strip steel tension in most annealing furnaces is still set according to a preset tension static table, the preset tension is distinguished according to the type, thickness, width and furnace section of the strip steel, the strip steel tension is irrelevant to the actual running condition of the strip steel, and the automatic optimal adjustment of the strip steel tension in the cold rolling process can not be realized, and the burden of an on-site operator can not be effectively reduced. When the set tension is not matched with the actual running condition of the strip steel, the strip steel in the furnace is easily deviated or buckled, so that the yield is low, and the shutdown and the production stop of the unit can be seriously caused. For this purpose, the machine set is usually provided with an operator to set the tension manually according to the actual situation, which requires a lot of manpower and material resources. Therefore, in order to realize intelligent production of the continuous annealing unit and ensure stable strip steel passing during the cold rolling annealing process, automatic setting and adjustment of strip steel tension in the annealing furnace are important. In order to realize the optimal adjustment of the strip steel tension in the cold rolling process, the existing solutions mainly comprise the following steps: Scheme 1, literature (Bai Zhenhua, cui Xiying, liu Yaxing, etc.) the technical study on tension optimization setting of a continuous annealing unit [ J ]. Yan Shanda school report, 2018,42 (02): 105-109.) the literature aims at solving the problems that the prior continuous annealing process is incomplete in tension setting target and cannot simultaneously consider strip deviation, thermal buckling, plate shape, stretching and the like, and on the basis of guaranteeing strip steel stable pass-through and comprehensive control indexes, the optimal pass-through stability in the strip continuous annealing process is ensured to be used as a control target, the problems of strip deviation, thermal buckling and the like in each process section are not taken as constraint conditions, and meanwhile the strip shape and the stretching problem are considered, so that a tension comprehensive optimization setting technology suitable for the continuous annealing process is established. Scheme 2. Literature (Tang Wei. Study on the coupling mechanism of cold rolling 2230 to the linewidth and the stable through plate [ D ]. Yan Shanda, 2020.) the literature inputs parameters such as furnace roller diameter, plate thickness, thickness difference, curling speed, strip steel temperature and the like as BP neural networks (BP (back propagation) neural networks are multilayer feedforward neural networks trained according to an error reverse propagation algorithm, are one of widely applied neural network models), and develops continuous annealing production speed forecasting models to realize optimal continuous annealing production speed forecasting under different initial plate shapes. However, the setting of continuous annealing process parameters based on the neural network is still in a theoretical research stage at present, the performance of the continuous annealing process parameters is far from reaching the application standard, and the automatic optimal adjustment of the strip steel tension in the cold rolling process can not be realized, and the burden of an on-site operator can not be effectively reduced. Scheme 3 literature (Bo Henan, iridium, a surname, text. Cold continuous rolling plate shape prediction based on neural network ensemble learning [ J ]. Rolled steel 2021,38 (01): 65-69.) the literature starts from cold rolled strip outlet plate shape prediction, and the plate shape prediction problem based on neural network is studied. In order to realize the prediction of the cold-rolled strip steel outlet plate shape, a wavelet neural network is optimized based on a particle swarm algorithm, the optimized network is used as a basic learner, an integrated learning prediction model is built through a bagging algorithm, and finally the prediction of the cold-rolled strip steel outlet plate shape is realized, but the optimal adjustment of strip steel tension in the cold rolling process cannot be realized. Disclosure of Invention In order to realize the optimal adj