CN-121997478-A - Method and device for generating temperature curve of hot rolled strip steel

CN121997478ACN 121997478 ACN121997478 ACN 121997478ACN-121997478-A

Abstract

The application provides a method and a device for generating a temperature curve of hot rolled strip steel, and relates to the technical field of steel rolling, wherein the method comprises the steps of inputting a sampling moment value into a trained target network model to generate a predicted temperature value corresponding to the sampling moment value through the processing of the target network model, wherein the total loss function of the target network model comprises a physical loss function, a boundary condition loss function and a data loss function, and the target network model corresponds to a hot rolling process interval; the method comprises the steps of determining a target sampling number corresponding to a hot rolling process interval according to mechanical property errors, determining a target sampling time value and a corresponding target predicted temperature value from the sampling time value and the corresponding predicted temperature value according to the target sampling number, and processing the target sampling time value and the target predicted temperature value to generate a hot rolling strip steel temperature curve, so that the accuracy and the reliability of hot rolling strip steel temperature prediction are effectively improved.

Inventors

WANG DEBIN
ZHAO ZHAOHONG
Zheng Hanghui
ZHANG BORUI
LIU CHENLAN
XU JINLONG
ZHOU YANJUAN
YANG XIONG
ZHANG BAOZHONG
LI ZHIWEI
ZHANG JUNXIA

Assignees

宁波钢铁有限公司

Dates

Publication Date: 20260508
Application Date: 20251203

Claims (10)

1. The method for generating the temperature curve of the hot rolled strip steel is characterized by comprising the following steps of: Inputting a sampling time value into a trained target network model to generate a predicted temperature value corresponding to the sampling time value through the processing of the target network model, wherein the total loss function of the target network model comprises a physical loss function, a boundary condition loss function and a data loss function, and the target network model corresponds to a hot rolling process interval; determining the target sampling quantity corresponding to the hot rolling process interval according to the mechanical property error; Determining a target sampling time value and a corresponding target predicted temperature value from the sampling time value and the corresponding predicted temperature value according to the target sampling quantity; And processing the target sampling moment value and the target predicted temperature value to generate a hot-rolled strip steel temperature curve.
2. The method of claim 1, wherein determining the target number of samples corresponding to the hot rolling process window based on the mechanical property error comprises: acquiring a first temperature value corresponding to the initial sampling number from a first data set; Processing each first temperature value by using a shallow regression model, and determining a first prediction error corresponding to the initial sampling number by combining a first Gaussian process agent model; under the condition that the first error corresponding to the initial sampling quantity is smaller than the first prediction error, processing each first temperature value by using a deep regression model, and determining an acquisition function value by combining a second Gaussian process agent model; And determining the sampling number corresponding to the current maximum acquisition function value as a new initial sampling number, and returning to the step of acquiring the first temperature value corresponding to the initial sampling number from the first data set until reaching the iteration termination condition, and determining the target sampling number.
3. The method of claim 2, wherein said processing each of said first temperature values using a shallow regression model in combination with a first gaussian process proxy model to determine a first prediction error corresponding to said initial sample number comprises: processing each first temperature value by using a shallow regression model to determine a first predicted value of the mechanical property of each first temperature value; determining a first mechanical property error of each first temperature value according to the difference between the first mechanical property predicted value and the actual mechanical property value of each first temperature value; determining the maximum value of the first errors of the mechanical properties as the first error corresponding to the initial sampling quantity; and acquiring a first prediction error corresponding to the initial sampling number according to a first prediction distribution under a first Gaussian process agent model.
4. The method of claim 2, wherein said processing each of said first temperature values using a deep regression model in combination with a second gaussian process proxy model to determine acquisition function values comprises: Processing each first temperature value by using a deep regression model to determine a mechanical property second predicted value of each first temperature value; Determining a mechanical property second error of each first temperature value according to the difference between the mechanical property second predicted value and the mechanical property actual value of each first temperature value; determining the maximum value of the second errors of the mechanical properties as the second error corresponding to the initial sampling quantity; adding the initial sampling number and the second error as new training data pairs into a second data set, and updating a current second Gaussian process agent model; And determining an acquisition function value according to the updated second prediction distribution and the current minimum error value under the second Gaussian process agent model.
5. The method of claim 2, further comprising, prior to said obtaining a first temperature value from the first dataset corresponding to the initial number of samples: Carrying out statistical treatment on the historical temperature data of the hot rolling process interval to determine a sampling number interval corresponding to the hot rolling process interval; sampling the sampling number interval to determine initial temperature values corresponding to each initial sampling group respectively; processing each initial temperature value by using a deep regression model to determine a mechanical property initial predicted value of each initial temperature value; determining the initial mechanical property error of each initial temperature value according to the difference between the initial mechanical property predicted value and the actual mechanical property value of each initial temperature value; Determining the maximum mechanical property initial error in each initial sampling group as an error value corresponding to the sampling number in the initial sampling group; And determining the sampling number corresponding to the maximum error value as an initial sampling number.
6. The method of claim 5, wherein the determining the number of samples corresponding to the maximum error value as the initial number of samples comprises: Training the initial Gaussian process proxy model by using a second data set to generate a second Gaussian process proxy model, wherein the second data set comprises data pairs formed by the sampling number and the error value; processing the initial prediction distribution under the second Gaussian process agent model and the minimum error value in the second data set to determine an acquisition function value; And determining the sampling number corresponding to the maximum acquisition function value as the initial sampling number.
7. The method of claim 1, wherein processing the target sample time values and target predicted temperature values to generate a hot rolled strip temperature profile comprises: processing the target sampling moment value and the target predicted temperature value to generate an initial temperature curve; Processing the initial temperature curve to determine the curvature of each node in the initial temperature curve; and updating the initial temperature curve according to the relation between the curvature of each node and a first curvature threshold value and a second curvature threshold value to generate a hot-rolled strip steel temperature curve, wherein the first curvature threshold value is smaller than the second curvature threshold value.
8. The method of claim 1, further comprising, prior to said inputting the sample time value into the trained target network model to generate the predicted temperature value corresponding to the sample time value via processing by the target network model: determining the total temperature difference value of the hot rolling process interval according to the thermodynamic model corresponding to the hot rolling process interval; Inputting the time value of the hot rolling process interval into an initial network model, and determining a temperature predicted value corresponding to each time value through the processing of the first initial network model; determining a data loss value according to the difference between the temperature predicted value and the temperature actual value of each time value; Inputting the inlet time value of the hot rolling process interval into an initial network model to determine an inlet temperature predicted value through the processing of the first initial network model; Determining a boundary condition loss value according to the difference between the inlet temperature predicted value and the inlet temperature actual value; Processing the temperature predicted value, the inlet temperature predicted value and the total temperature difference value to determine a physical loss value; Fusing the data loss value, the boundary condition loss value and the physical loss value to determine a total loss value; Training the initial network model based on the total loss value to generate a trained target network model.
9. A device for generating a temperature profile of a hot rolled strip, comprising: The first generation module is used for inputting the sampling moment value into a trained target network model so as to generate a predicted temperature value corresponding to the sampling moment value through the processing of the target network model, wherein the total loss function of the target network model comprises a physical loss function, a boundary condition loss function and a data loss function, and the target network model corresponds to a hot rolling process interval; The first determining module is used for determining the target sampling number corresponding to the hot rolling process interval according to the mechanical property error; the second determining module is used for determining a target sampling moment value and a corresponding target predicted temperature value from the sampling moment value and the corresponding predicted temperature value according to the target sampling quantity; And the second generation module is used for processing the target sampling moment value and the target predicted temperature value to generate a hot-rolled strip steel temperature curve.
10. An electronic device comprising a processor and a memory storing computer program instructions; The processor, when executing the computer program instructions, implements the method for generating a hot rolled strip temperature profile as defined in any one of claims 1 to 8.

Description

Method and device for generating temperature curve of hot rolled strip steel Technical Field The application relates to the technical field of steel rolling, in particular to a method and a device for generating a temperature curve of hot rolled strip steel. Background In the hot continuous rolling production process, the distribution of the strip steel temperature field directly determines the tissue evolution and the mechanical property of the strip steel. In the finish rolling and coiling stages, if the online measured temperature is not accurate enough, accidents such as aggravation of temperature difference between the inside and the outside of the finished product, uneven microstructure distribution, fluctuation of mechanical properties and the like easily occur, and even quality accidents are caused. The online temperature prediction precision is improved, the process parameters can be optimized, the energy consumption can be reduced, the product consistency is improved, and the method has important significance for intelligent manufacturing in the steel industry. In the related technology, the traditional numerical simulation or single data driving model is depended, the problems of high calculation cost and low real-time performance exist, the pure data driving method is high in calculation speed, but highly depends on dense and high-quality measuring point data, industrial field measuring points are sparse and high in noise, the model is easy to be over-fitted or unstable, and the constraint on basic physical laws such as heat conduction and convection is lacked, so that the temperature prediction precision of the hot-rolled strip steel is low. Therefore, in the hot continuous rolling production process, how to improve the accuracy of strip steel temperature prediction is very important. Disclosure of Invention The application provides a method and a device for generating a temperature curve of hot rolled strip steel. According to a first aspect of the present application, there is provided a method of generating a temperature profile of a hot rolled strip, the method comprising: Inputting a sampling time value into a trained target network model to generate a predicted temperature value corresponding to the sampling time value through the processing of the target network model, wherein the total loss function of the target network model comprises a physical loss function, a boundary condition loss function and a data loss function, and the target network model corresponds to a hot rolling process interval; determining the target sampling quantity corresponding to the hot rolling process interval according to the mechanical property error; Determining a target sampling time value and a corresponding target predicted temperature value from the sampling time value and the corresponding predicted temperature value according to the target sampling quantity; And processing the target sampling moment value and the target predicted temperature value to generate a hot-rolled strip steel temperature curve. According to a second aspect of the present application, there is provided an apparatus for generating a temperature profile of a hot rolled strip, comprising: The first generation module is used for inputting the sampling moment value into a trained target network model so as to generate a predicted temperature value corresponding to the sampling moment value through the processing of the target network model, wherein the total loss function of the target network model comprises a physical loss function, a boundary condition loss function and a data loss function, and the target network model corresponds to a hot rolling process interval; The first determining module is used for determining the target sampling number corresponding to the hot rolling process interval according to the mechanical property error; the second determining module is used for determining a target sampling moment value and a corresponding target predicted temperature value from the sampling moment value and the corresponding predicted temperature value according to the target sampling quantity; And the second generation module is used for processing the target sampling moment value and the target predicted temperature value to generate a hot-rolled strip steel temperature curve. According to a third aspect of the application, there is provided an electronic device comprising a processor and a memory storing computer program instructions, the processor implementing a method of generating a temperature profile of any one of the hot rolled strips described above when executing the computer program instructions. According to a fourth aspect of the present application, there is provided a computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement a method of generating a temperature profile of any one of the hot rolled steel strips described above. In summary, th