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

Abstract

The invention provides a method and device for generating a temperature curve of hot rolled strip steel, and relates to the technical field of steel rolling, and the method comprises the steps: inputting a sampling time value into a trained target network model, and generating a predicted temperature value corresponding to the sampling time value through the processing of the target network model, 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 and the hot rolling process interval are in a corresponding relation; according to the mechanical property error, the target sampling number corresponding to the hot rolling process interval is determined; 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 the target sampling time value and the target prediction temperature value are processed to generate a hot rolling strip steel temperature curve, and therefore the accuracy and reliability of hot rolling strip steel temperature prediction are effectively improved.

Description

Technical Field

[0001] This application relates to the field of steel rolling technology, and in particular to a method and apparatus for generating temperature curves of hot-rolled strip steel. Background Technology

[0002] In the hot strip rolling process, the temperature field distribution of the strip directly determines its microstructure evolution and mechanical properties. During the finishing rolling and coiling stages, insufficient accuracy in online temperature measurement can easily lead to unexpected issues such as increased internal and external temperature differences in the finished product, uneven microstructure distribution, and fluctuations in mechanical properties, potentially even causing quality accidents. Improving the accuracy of online temperature prediction can not only optimize process parameters but also reduce energy consumption and improve product consistency, which is of great significance to intelligent manufacturing in the steel industry.

[0003] In related technologies, many rely on traditional numerical simulations or single data-driven models, which suffer from high computational costs and low real-time performance. While pure data-driven methods offer faster computation speeds, they heavily depend on dense and high-quality measurement data. However, industrial field measurement points are often sparse and noisy, easily leading to model overfitting or instability. Furthermore, they lack constraints on fundamental physical laws such as heat conduction and convection, resulting in low accuracy in hot-rolled strip temperature prediction. Therefore, improving the accuracy of strip temperature prediction during hot continuous rolling production is crucial. Summary of the Invention

[0004] This application provides a method and apparatus for generating temperature profiles of hot-rolled strip steel.

[0005] According to a first aspect of this application, a method for generating a temperature profile of hot-rolled strip steel is provided, the method comprising: The sampled time value is input into the trained target network model, and after processing by the target network model, the predicted temperature value corresponding to the sampled time value is generated. The total loss function of the target network model includes a physical loss function, a boundary condition loss function, and a data loss function. The target network model corresponds to the hot rolling process range. The target sampling quantity corresponding to the hot rolling process range is determined based on the mechanical property error. Based on the target sampling quantity, the target sampling time value and the corresponding target predicted temperature value are determined from the sampling time value and the corresponding predicted temperature value. The target sampling time value and the target predicted temperature value are processed to generate the temperature curve of hot-rolled strip steel.

[0006] According to a second aspect of this application, an apparatus for generating a temperature profile of hot-rolled strip steel is provided, comprising: The first generation module is used to input the sampled time value into the trained target network model so that the target network model processes the sampled time value to generate the predicted temperature value. The total loss function of the target network model includes a physical loss function, a boundary condition loss function and a data loss function. The target network model is corresponding to the hot rolling process range. The first determining module is used to determine the target sampling quantity corresponding to the hot rolling process range based on the mechanical property error; The second determining module is used to determine the target sampling time value and the corresponding target predicted temperature value from the sampling time value and the corresponding predicted temperature value based on the target sampling quantity. The second generation module is used to process the target sampling time value and the target predicted temperature value to generate the temperature curve of hot-rolled strip steel.

[0007] According to a third aspect of this application, an electronic device is provided, comprising: a processor and a memory storing computer program instructions; the processor, when executing the computer program instructions, implements any of the above-described methods for generating a temperature profile of hot-rolled strip steel.

[0008] According to a fourth aspect of this application, a computer-readable storage medium is provided, on which computer program instructions are stored, which, when executed by a processor, implement any of the above-described methods for generating a temperature profile of hot-rolled strip steel.

[0009] In summary, the method and apparatus for generating the temperature curve of hot-rolled strip steel provided in this application have at least the following beneficial effects: First, the sampling time value can be input into a pre-trained target network model. After processing by the target network model, the predicted temperature value corresponding to the sampling time value is generated. The total loss function of the target network model includes a physical loss function, a boundary condition loss function, and a data loss function. The target network model corresponds to the hot-rolling process interval. Then, based on the mechanical property error, the target sampling quantity corresponding to the hot-rolling process interval can be determined. Next, based on the target sampling quantity, the target sampling time value and the corresponding target predicted temperature value are determined from the sampling time value and the corresponding predicted temperature value. Finally, the target sampling time value and the target predicted temperature value are processed to generate the temperature curve of the hot-rolled strip steel. Therefore, the target network model trained can generate predicted temperature values. Then, with mechanical performance error as the optimization target, the number of target samples, the corresponding target sampling time value, and the target predicted temperature value are determined in reverse. After smooth interpolation, the temperature curve of hot-rolled strip steel is generated. This process takes into account physical rationality, mechanical performance error, and data cost, so that the curve has both global smoothness and high fidelity of local details, effectively improving the accuracy and reliability of hot-rolled strip steel temperature prediction. Attached Figure Description

[0010] To more clearly illustrate the specific embodiments of this application or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A flowchart illustrating a method for generating a temperature profile of hot-rolled strip steel, provided for embodiments of this application; Figure 2 A schematic diagram of the structure of an initial network model provided for an embodiment of this application; Figure 3 A schematic diagram illustrating the process of generating a temperature profile for hot-rolled strip steel, provided for an embodiment of this application; Figure 4 A structural diagram of an apparatus for generating a temperature profile of hot-rolled strip steel, provided for an embodiment of this application; Figure 5 This is a structural diagram of an electronic device provided as an embodiment of the present application. Detailed Implementation

[0012] To make the above and other features and advantages of this application clearer, the application is further described below with reference to the accompanying drawings. It should be understood that the specific embodiments given herein are for the purpose of explanation to those skilled in the art, and are exemplary only, not restrictive.

[0013] In the following description, numerous specific details are set forth to provide a thorough understanding of this application. However, it will be apparent to those skilled in the art that the specific details are not required to practice this application. In other instances, well-known steps or operations have not been described in detail to avoid obscuring this application.

[0014] The method for generating the temperature profile of hot-rolled strip steel provided in this application embodiment can be executed by the device for generating the temperature profile of hot-rolled strip steel provided in this application embodiment, and the device can be configured in an electronic device.

[0015] refer to Figure 1 This application provides a method for generating a temperature profile of hot-rolled strip steel, the method comprising: Step 101: Input the sampled time value into the trained target network model so that the target network model can process the sampled time value to generate the predicted temperature value. The total loss function of the target network model includes the physical loss function, the boundary condition loss function and the data loss function. The target network model corresponds to the hot rolling process range.

[0016] The target network model can be generated by training an initial network model. The initial network model can be a physics-informed neural network (PINN) or any other trainable neural network model, and this application does not limit it.

[0017] Understandably, the physical loss function, by embedding thermodynamic mechanisms such as radiation, convection, and conduction as residual terms into the network model training, ensures that the network model strictly adheres to physical conservation laws at any sampling time, effectively avoiding the problem of prediction results not conforming to physical phenomena that may occur with purely data-driven approaches. Simultaneously, by using the measured temperature value at the boundary inlet as a boundary penalty term, the network model ensures strict consistency with the actual measured values ​​at the interval endpoints, preventing endpoint drift and improving prediction reliability. Furthermore, by using the data loss constrained by the measured temperatures of sparse sampling points, the network model ensures that the prediction results do not deviate from the true observation points, achieving high-precision fitting with small samples. Therefore, model training combining these three loss mechanisms allows the generated target network model to possess both data authenticity and physical plausibility during temperature prediction, significantly improving the accuracy and reliability of predicted temperature values.

[0018] Understandably, the initial network model can be trained in advance to generate the target network model.

[0019] Optionally, the total temperature difference of the hot rolling process interval can be determined based on the thermodynamic model corresponding to the hot rolling process interval.

[0020] There are various thermodynamic models, such as radiation heat transfer model, convection heat transfer model, workpiece deformation temperature rise model, workpiece friction temperature rise model, workpiece-roll contact temperature drop model, laminar flow cooling heat transfer model, etc.

[0021] Among them, the temperature difference dT generated by the radiation heat transfer model g It can be represented as follows: ; in, Thermal emissivity; This is the Stefan Boltzmann constant, typically taken as 5.69 × 10⁻⁶. -11 kJ / m·s·℃ 4 ; This refers to the density of the strip steel, expressed in kg / m³. 3 C p (T) represents the specific heat of the strip steel, expressed in kJ / kg·℃; T z The measured finishing rolling temperature of the workpiece is given in °C (°C); T yThis refers to the ambient temperature, expressed in °C. This refers to the strip width, in mm. The time the rolled piece is in an air-cooled state, measured in seconds (s); h n The current strip thickness is in mm.

[0022] In addition, the temperature difference dT generated by the convective heat transfer model h It can be represented as follows: ; Among them, T z The measured finishing rolling temperature of the workpiece is given in °C (°C); T x f represents the temperature of the cooling medium, expressed in °C. s P represents the nozzle water flow rate, in L / min. s The nozzle water pressure is expressed in MPa; k T These are the coefficients for the water-cooling model; The time for heat exchange, measured in seconds (s); h n The current strip thickness is expressed in mm; C p (T) represents the specific heat of the strip steel, expressed in kJ / kg·℃; This refers to the density of the strip steel, expressed in kg / m³. 3 .

[0023] In addition, the temperature difference dT generated by the rolling deformation temperature rise model o It can be represented as follows: ; Wherein, dT3 is the temperature rise of the rolled piece caused by deformation heat, in °C; K is the deformation thermal gain coefficient. m H is the deformation resistance, in MPa; H is the thickness of the rolled piece at the mill inlet, in mm; h is the thickness of the rolled piece at the mill outlet, in mm; C p (T) represents the specific heat of the strip steel, in kJ / kg·℃; This refers to the density of the strip steel, expressed in kg / m³. 3 .

[0024] In addition, the temperature difference dT generated by the frictional temperature rise model of the rolled piece k It can be represented as follows: ; Among them, dT k The temperature rise of the rolled piece due to frictional heat is expressed in °C. This is the frictional heat gain coefficient; K is the coefficient of friction; m This is the resistance to deformation, expressed in MPa. The rolling time is expressed in seconds (s). This refers to the density of the strip steel, expressed in kg / m³. 3 C p (T) represents the specific heat of the strip, in kJ / kg·℃; H represents the thickness of the rolled piece at the mill inlet, in mm; h represents the thickness of the rolled piece at the mill outlet, in mm; V represents the speed of the rolled piece at the mill outlet, in m / s; f s f is the forward slip ratio; b This is the backslip rate.

[0025] In addition, the temperature difference generated by the temperature drop model of the contact between the workpiece and the rolls It can be represented as follows: ; in, Temperature drop of the rolled piece caused by contact, in °C; The contact temperature drop gain coefficient; T is the temperature drop coefficient; v T represents the working roll temperature, expressed in °C. u H represents the mill inlet temperature of the rolled piece, in °C; H represents the mill inlet thickness of the rolled piece, in mm; h represents the mill outlet thickness of the rolled piece, in mm. The rolling time is expressed in seconds (s). The thermal conductivity of the working roll is expressed in kJ / m·hr·℃. The thermal conductivity of the work roll is expressed in m. 2 / hr; The thermal conductivity of the rolled piece is expressed in kJ / m·hr·℃. The thermal conductivity of the rolled product is expressed in meters (m). 2 / hr.

[0026] In addition, the temperature difference t generated by laminar flow cooling heat transfer CT(AIM) It can be represented as: ; in, The water-cooled temperature drop between the temperature measuring instruments installed before and after the laminar flow cooling section, in °C; t CT(AIR) This represents the temperature difference of the rolled piece after air cooling, in °C. The range of error is defined as negative values. The temperature drop is measured in °C (°C) for laminar flow cooling. The temperature drop is measured in °C from the side-spray water cooling system.

[0027] Among them, the temperature difference t of the rolled piece generated by the air cooling model CT(AIR) It can be represented as: ; in, Thermal emissivity; It is the Stefan Boltzmann constant; The time between the temperature measuring instrument after finishing rolling and the CT temperature measuring instrument on the coiler is expressed in seconds. The density of the rolled product, in kg / m³ 3 ;c represents the specific heat of the rolled product, in kJ / kg·℃;h n The change in thickness at the finish rolling exit is expressed in meters (m); t FC The measured finishing rolling temperature is given in °C.

[0028] In addition, the temperature difference generated by the laminar flow cooling model It can be represented as: ; Among them, l B The length cooled by a single manifold, measured in meters (m); q B Specific heat flux for laminar flow cooling; The density of the rolled product, in kg / m³ 3 c represents the specific heat of the rolled product, in kJ / kg·℃; v n h represents the change in the finishing mill exit speed. n This represents the thickness (variation) at the finish rolling exit.

[0029] In addition, the temperature difference generated by the side-spray water cooling model It can be represented as follows: ; in, The time involved in the side spraying of water is measured in seconds (s); q s The specific heat flux carried by the side-sprayed water is important for heat transfer. ; t is the heat transfer coefficient; t is the current stage temperature of the rolled piece; t w Water temperature; The density of the rolled product, in kg / m³ 3 ; c is the specific heat of the rolled piece, in kJ / kg·℃; h7 is the set thickness of the finished product, in meters.

[0030] It is understandable that the total temperature difference between different hot rolling process zones may be the same or different, which can be expressed as... Let m represent the total temperature difference for each hot rolling process interval, where m = 1, 2, 3, 4, 5. These correspond to the total temperature difference for each hot rolling process interval: from the furnace outlet to the roughing mill inlet, the roughing mill interval (inlet-outlet), the roughing mill outlet to the finishing mill inlet, the finishing mill interval (inlet-outlet), and the finishing mill outlet to the coiler inlet. It can be represented as follows: .

[0031] Therefore, in this embodiment of the application, the total temperature difference of each hot rolling process interval can be determined according to the above formula, so that the result fully reflects the thermodynamic characteristics of each hot rolling process interval, thereby improving the accuracy and reliability of the total temperature difference of each process interval.

[0032] Then, the time values ​​of the hot rolling process interval can be input into the initial network model. After processing by the first initial network model, the temperature prediction value corresponding to each time value can be determined. Then, the data loss value can be determined based on the difference between the temperature prediction value and the actual temperature value at each time value.

[0033] The first initial network model can be a PINN model, or any other trainable neural network model, etc., and this application does not limit it.

[0034] In addition, the data loss function can be expressed as: ; Among them, L dmj For data loss values, T is the predicted temperature value corresponding to the existing data points. mj S represents the actual temperature value, and S represents the total amount of data for this hot rolling process range.

[0035] Then, the entry time value of the hot rolling process zone can be input into the initial network model. After processing by the first initial network model, the predicted entry temperature value is determined. Then, based on the difference between the predicted entry temperature value and the actual entry temperature value, the boundary condition loss value is determined.

[0036] Specifically, regarding the boundary conditions, at the entrance of each hot rolling process section... The actual inlet temperature T of the strip steel bm Since this is known, it can provide an initial condition for the model, ensuring that the network-predicted temperature at the inlet of the hot rolling process zone matches the known value. Therefore, by inputting the inlet time value of the hot rolling process zone into the initial network model, and processing it, the predicted inlet and outlet temperatures can be determined. ,Right now: .

[0037] Furthermore, the boundary condition loss function can be expressed as: ; Among them, L bmi This represents the boundary condition loss value. T is the predicted inlet temperature. bm This is the actual inlet temperature.

[0038] The predicted temperature, inlet temperature, and total temperature difference can then be processed to determine the physical loss value. The data loss value, boundary condition loss value, and physical loss value can then be fused to determine the total loss value.

[0039] The physical loss function in the hot rolling process range can be expressed as: ; in, T is the predicted temperature value. bm This is the actual inlet temperature. This represents the total temperature difference within the hot rolling process range.

[0040] Furthermore, the total loss function in the hot rolling process range can be expressed as: ; in, These are the weight values ​​corresponding to the physical loss function. These are the weight values ​​corresponding to the boundary condition loss function. These are the weight values ​​corresponding to the data function.

[0041] Finally, the initial network model is trained based on the total loss value to generate the trained target network model.

[0042] In the case where the initial network model is the PINN model, the structure of the PINN model can be as follows: Figure 2 As shown, the initial network model includes an input layer, hidden layers, and an output layer. The total loss function can include the physical loss function, the boundary condition loss function, and the data loss function. Furthermore, the input data for this initial network model can be the heat dissipation time of each hot rolling process. Feature extraction can be performed using two fully connected hidden layers, outputting predicted temperature values ​​corresponding to different times. The hidden layers can use the Tanh activation function, or other activation functions, etc., which are not limited in this application.

[0043] Therefore, in this embodiment, the total loss function value, i.e., the total loss value, can be obtained by weighted fusion of the physical loss function, boundary loss function, data loss function, and their respective weight values. Then, the initial network model can be trained based on the total loss value until the initial network model converges or reaches the iteration stopping condition. The initial network model at this point is then determined as the target network model that has been successfully trained.

[0044] It is understandable that for each hot rolling process interval, a target network model corresponding to each hot rolling process interval can be trained and generated in the above manner. The predicted temperature value obtained by using the target network model is more accurate and reliable.

[0045] Step 102: Determine the target sampling quantity corresponding to the hot rolling process range based on the mechanical property error.

[0046] The hot-rolled strip steel production process includes multiple hot rolling process sections, including five hot rolling process sections: from the heating furnace outlet to the roughing mill inlet, the roughing mill section, from the roughing mill outlet to the finishing mill inlet, the finishing mill section, and from the finishing mill outlet to the coiler inlet. The roughing mill section includes the roughing mill inlet to the outlet, and the finishing mill section includes the finishing mill inlet to the outlet.

[0047] It is understandable that the temperature variation amplitude and process criticality differ across hot rolling process intervals. Therefore, the number of sampling points selected for each hot rolling process interval during temperature curve generation typically varies. For example, in hot rolling process intervals with large temperature variation amplitudes, drastic curvature changes, and significant impacts on the final properties of the strip, such as the finishing mill interval, a larger number of samples can be set to ensure the temperature curve captures detailed features. Conversely, for intervals with relatively gentle temperature changes and less impact on the final product properties, a relatively smaller number of samples can be set, thereby reducing the number of samples and computational costs while maintaining accuracy. Therefore, in this embodiment, to improve the accuracy of temperature prediction, mechanical property errors can be processed to determine the target number of samples corresponding to each hot rolling process interval.

[0048] Optionally, a first temperature value corresponding to the initial number of samples can be obtained from the first dataset.

[0049] The first dataset includes multiple first temperature values, or it may include multiple time values ​​and data pairs consisting of first temperature values. The first temperature value may be an actual temperature value, or it may be a predicted temperature value generated by the target network model, etc. This application does not limit this.

[0050] For example, with an initial sampling quantity of 30, 30 first temperature values ​​can be obtained by uniform sampling in the first dataset, or by random sampling, etc. This application does not limit this.

[0051] In addition, there are several ways to determine the initial number of samples. For example, it can be determined based on experience, or it can be determined by performing a series of processing on historical temperature data.

[0052] Optionally, the historical temperature data of the hot rolling process range can be statistically processed first to determine the sampling quantity range corresponding to the hot rolling process range.

[0053] This involves statistically processing historical temperature data for each hot rolling process interval to determine the maximum, minimum, and average historical temperature values ​​for each interval. Then, considering factors such as temperature variation and process criticality, the minimum number of samples N required to cover that hot rolling process interval can be determined. min and the maximum number of samples N max This is to form a corresponding sampling quantity range. Typically, the temperature range and degree of change may differ for each hot rolling process interval, and the corresponding sampling quantity range may also be different. It can be determined separately for each hot rolling process interval, etc., and this application does not limit this.

[0054] Then, sampling can be performed on the sampling quantity range to determine the initial temperature value corresponding to each initial sampling group.

[0055] In the process of sampling within a range of sample quantities, uniform sampling or stratified sampling may be used, and this application does not limit the specific methods used.

[0056] For example, multiple initial sampling groups can be obtained by stratified sampling within the sampling range. Then, different numbers of initial temperature values ​​can be determined in each initial sampling group by uniform sampling or random sampling. For example, initial sampling group 1 can select 10 initial temperature values, initial sampling group 2 can select 25 initial temperature values, initial sampling group 3 can select 100 initial temperature values, and so on. This application does not limit this.

[0057] Then, a deep regression model can be used to process each initial temperature value to determine the initial predicted value of the mechanical properties for each initial temperature value.

[0058] The deep regression model can be a fully configured regression model, such as an extreme gradient boosting (XGBoost) model, a random forest (RF) model, a light gradient boosting machine (LightGBM) model, or other regression models, etc. This application does not limit it in this regard.

[0059] In addition, mechanical properties may include yield strength YS, tensile strength TS, and elongation EL, etc., which are not limited in this application.

[0060] Understandably, the deep regression model can be a pre-trained and directly usable model. Therefore, by inputting each initial temperature value into the deep regression model and processing it, the model can generate initial predicted values ​​of mechanical properties associated with each initial temperature value. In other words, it can determine the initial predicted values ​​of yield strength, tensile strength, and elongation associated with each initial temperature value.

[0061] Then, the initial error of the mechanical properties at each initial temperature can be determined based on the difference between the initial predicted value and the actual value of the mechanical properties at each initial temperature.

[0062] The actual values ​​of yield strength, tensile strength, and elongation associated with each initial temperature value can be determined through laboratory tensile tests, or other methods can be used. For example, for hot-rolled strip steel of the same material and specification, a portion of the hot-rolled strip steel can be sent to the laboratory for tensile testing, while the remaining portion can undergo further processing. This application does not limit this approach.

[0063] The difference between the initial predicted mechanical property value and the actual mechanical property value at each initial temperature can be determined as the initial error of the mechanical property at each initial temperature value; or the absolute difference between the initial predicted mechanical property value and the actual mechanical property value at each initial temperature value can be determined as the initial error of the mechanical property at each initial temperature value; or the relative error between the initial predicted mechanical property value and the actual mechanical property value at each initial temperature value can be determined as the initial error of the mechanical property at each initial temperature value, etc. This application does not limit this.

[0064] Optionally, the yield strength error value, the tensile strength error value, and the elongation error value can be determined first for each initial temperature value, corresponding to the initial predicted yield strength value and the actual yield strength value. For each initial temperature value, the maximum value among the yield strength error value, tensile strength error value, and elongation error value is determined as the initial mechanical property error corresponding to that initial temperature value.

[0065] Then, the maximum initial mechanical performance error in each initial sampling group can be determined as the error value of the number of samples in the corresponding initial sampling group, and the number of samples corresponding to the maximum error value can be determined as the initial sampling number.

[0066] In this process, after determining the initial mechanical property error for each initial temperature value, for each initial sampling group, the maximum value of the initial mechanical property error in the initial sampling group can be further determined as the error value corresponding to the number of samples included in the initial sampling group. After determining the error value corresponding to the number of samples in each initial sampling group, the number of samples corresponding to the maximum error value can be determined as the initial sampling number.

[0067] Optionally, the initial second Gaussian process surrogate model can be trained using the second dataset to generate the second Gaussian process surrogate model. Then, the initial prediction distribution under the second Gaussian process surrogate model and the minimum error value in the second dataset can be processed to determine the acquisition function value, and the sampling number corresponding to the maximum acquisition function value can be determined as the initial sampling number.

[0068] The second dataset may include data pairs consisting of the number of samples and the error value.

[0069] In addition, the acquisition function can satisfy the following relationship: ; in, Let be the expected function. This represents the expected value of a random variable, also known as a probability-weighted average. `max(•)` indicates taking the maximum value. min E(N) represents the minimum error value in the current second dataset, and E(N) represents the initial prediction distribution under the current second Gaussian process surrogate model.

[0070] Therefore, in this embodiment, after determining the initial prediction distribution under the second Gaussian process surrogate model and the minimum error value in the second dataset, the acquisition function value can be further calculated using the acquisition function. Then, the acquisition function value corresponding to each N value can be obtained. Then, the number of samples corresponding to the largest acquisition function value among all acquisition function values ​​can be determined as the initial number of samples. That is, in the process of determining the initial number of samples, the mechanical performance error of each initial temperature value is fully considered, so that the determined initial number of samples can be more accurate and reliable.

[0071] Then, a shallow regression model can be used to process each first temperature value, and combined with the first Gaussian process surrogate model, the first prediction error corresponding to the initial number of samples can be determined.

[0072] Optionally, a shallow regression model can be used to process each first temperature value to determine the first predicted value of the mechanical properties of each first temperature value. Then, based on the difference between the first predicted value of the mechanical properties of each first temperature value and the actual value of the mechanical properties, the first error of the mechanical properties of each first temperature value can be determined. Finally, the maximum value of the first error of the mechanical properties can be determined as the first error corresponding to the initial sampling quantity.

[0073] Shallow regression models can be understood as having fewer parameters than deep regression models, allowing for rapid estimation. For example, they can be XGBoost models, RF models, LightGBM models with fewer trees and shallower depths, or other regression models, etc. This application does not limit them in this regard.

[0074] Understandably, the shallow regression model can be a pre-trained and directly usable model. Therefore, by inputting each first temperature value into the shallow regression model and processing it, a first predicted value of mechanical properties associated with each first temperature value can be generated. In other words, the first predicted value of yield strength, the first predicted value of tensile strength, and the first predicted value of elongation associated with each first temperature value can be determined.

[0075] Furthermore, the difference between the first predicted value of the mechanical properties at the first temperature and the actual value of the mechanical properties can be a difference, an absolute error, a relative error, etc., and this application does not limit it.

[0076] Next, we can first determine the first error value of yield strength (corresponding to the first predicted yield strength value and the actual yield strength value), the first error value of tensile strength (corresponding to the first predicted tensile strength value and the actual tensile strength value), and the first error value of elongation (corresponding to the first predicted elongation value and the actual elongation value) for each first temperature value. For each first temperature value, the maximum value among the first error values ​​of yield strength, tensile strength, and elongation can be determined as the first error of the mechanical property corresponding to that first temperature value. Then, the maximum value among the first errors of each mechanical property is determined as the first error corresponding to the initial sampling quantity.

[0077] Then, based on the first prediction distribution under the first Gaussian process surrogate model, the first prediction error corresponding to the initial sampling quantity can be obtained.

[0078] The first Gaussian process surrogate model can be a low-fidelity Gaussian process surrogate model, which can be expressed as: ; in, The prediction error of the first Gaussian process surrogate model, It is a Gaussian process. Let be the mean function of a Gaussian process. The kernel function of the Gaussian process is given by the equation, which shows that... The probability distribution can be derived from a Gaussian process. As described.

[0079] Next, the first error corresponding to the initial sampling quantity can be compared with the first prediction error. If the first error is less than the first prediction error, then the initial sampling quantity and the potential value of the first error are considered to be large, and further processing can be performed. If the first error is less than or equal to the first prediction error, then the initial sampling quantity and the potential value of the first error are considered to be small, and they can be discarded without further processing. In this case, the initial sampling quantity can be re-determined. The specific implementation method can refer to the description of determining the initial sampling quantity above, which will not be repeated here.

[0080] If the first error is less than the first prediction error, a deep regression model is used to process each first temperature value, and a second Gaussian process surrogate model is combined to determine the acquisition function value.

[0081] It is understandable that since the first error is smaller than the first prediction error, its potential value can be considered to be large. At this time, a deep regression model can be used to process each first temperature value to determine the second predicted value of the mechanical properties of each first temperature value. Then, based on the difference between the second predicted value of the mechanical properties of each first temperature value and the actual value of the mechanical properties, the second error of the mechanical properties of each first temperature value can be determined, and the maximum value of the second error of the mechanical properties can be determined as the second error corresponding to the initial sampling quantity.

[0082] The process involves inputting a first temperature value into a deep regression model. This model processes the data to determine a second predicted value for the mechanical properties at that first temperature. Next, the differences between the predicted and actual mechanical properties for each of the three mechanical property indices at the first temperature value are determined. These differences can be expressed as differential values, relative errors, or absolute errors. The largest difference among the three indices is then defined as the second error for the mechanical properties at that first temperature value. Finally, after determining the second error for each first temperature value, the largest of all second error values ​​is defined as the second error corresponding to the initial sampling quantity.

[0083] Then, the initial sampling number and the second error can be added to the second dataset as new training data pairs, and the current second Gaussian process surrogate model can be updated. After that, the acquisition function value can be determined based on the second prediction distribution under the updated second Gaussian process surrogate model and the current minimum error value.

[0084] Specifically, after adding the initial sampling number and the second error as new training data pairs to the second dataset, the updated second dataset can be used to update the current second Gaussian process surrogate model. Correspondingly, the prediction distribution corresponding to this second Gaussian process surrogate model will also be updated. Then, based on the updated second prediction distribution and the minimum error value in the current second dataset, the acquisition function value corresponding to each N value under this prediction distribution can be determined in conjunction with the acquisition function.

[0085] Then, the number of samples corresponding to the current maximum acquisition function value can be determined as the new initial number of samples, and the process of obtaining the first temperature value corresponding to the initial number of samples from the first dataset can be returned until the iteration termination condition is met, and the target number of samples is determined.

[0086] The iteration termination condition is reaching the maximum number of iterations or the convergence of the acquisition function value. At this time, the current number of samples can be determined as the target number of samples, which can also be called the optimal number of samples for this hot rolling process range.

[0087] Understandably, if the maximum number of iterations has not been reached or the acquisition function value has converged, the sampling quantity corresponding to the current maximum acquisition function value can be determined as the new initial sampling quantity. Then, the first temperature value corresponding to the new initial sampling quantity can be obtained from the first dataset. A shallow regression model is then used to process each temperature value, combined with a first Gaussian process surrogate model, to determine the first prediction error corresponding to the initial sampling quantity. If the first error is less than the first prediction error, a deep regression model can be used to process each first temperature value, combined with a second Gaussian process surrogate model, to determine the new acquisition function value. Then, it is determined whether the iteration termination condition has been met. If the iteration termination condition has not been met, the above process can be repeated based on the new initial sampling quantity until the iteration termination condition is met, and the target sampling quantity is determined. If the first error is greater than or equal to the first prediction error, the second-maximum acquisition function value can be found from the acquisition function values ​​of the previous round, and its corresponding sampling quantity is determined as the new initial sampling quantity. The above process is then repeated until the iteration termination condition is met, and the target sampling quantity is determined.

[0088] Therefore, in this embodiment, the above-described method can be used to determine the target sampling quantity for each hot rolling process interval. During this process, a hierarchical proxy modeling strategy is employed. In low-fidelity processes, a large-scale, rapid search narrows the candidate interval. In high-fidelity processes, the error function is accurately fitted, and the acquisition function is optimized, thereby significantly reducing evaluation costs while maintaining accuracy. In other words, by combining the efficiency of multi-fidelity proxy modeling with the robustness of the maximum error principle, prediction performance and data overhead can be balanced across different hot rolling process intervals, thereby significantly improving the modeling efficiency and reliability of mechanical property predictions for the hot continuous rolling process.

[0089] Step 103: Based on the target sampling quantity, determine the target sampling time value and the corresponding target predicted temperature value from the sampling time value and the corresponding predicted temperature value.

[0090] In the process of determining the target sampling time value and the corresponding target predicted temperature value from the sampling time value and the corresponding predicted temperature value, the selection can be random, uniform, or in other ways, etc., and this application does not limit this.

[0091] For example, in the target sampling number N * When the value is 100, 100 data pairs consisting of sampling time values ​​and corresponding predicted temperature values ​​can be randomly selected from a large number of data pairs consisting of sampling time values ​​and corresponding predicted temperature values ​​as the target sampling time value and target predicted temperature value; alternatively, 100 data pairs consisting of sampling time values ​​and corresponding predicted temperature values ​​can be selected evenly according to time intervals as the target sampling time value and target predicted temperature value; alternatively, 100 data pairs consisting of sampling time values ​​and corresponding predicted temperature values ​​can be selectively selected as the target sampling time value and target predicted temperature value according to the rate of change of the predicted temperature value, etc. This application does not limit this.

[0092] Optionally, after determining the target sampling quantity for a certain hot rolling process range, a new sampling time value can be determined according to the target sampling quantity, and the new sampling time value can be input into the target network model. After processing by the target network model, a target predicted temperature value corresponding to the target sampling time value can be generated. This application does not limit this.

[0093] Step 104: Process the target sampling time value and the target predicted temperature value to generate the temperature curve of hot-rolled strip steel.

[0094] It is understandable that, since the number of target sampling time values ​​and target predicted temperature values ​​is the target sampling number for each hot rolling process interval, i.e. the optimal sampling number, the generated hot-rolled strip temperature curve can more realistically reflect the temperature change of the strip during the hot rolling process after processing the target sampling time values ​​and target predicted temperature values, thereby improving the reliability of the hot-rolled strip temperature curve.

[0095] There are various ways to process the target sampling time value and the target predicted temperature value. For example, cubic spline interpolation can be used to determine that the curve has a continuous second-order reciprocal at the sampling point, or polynomial interpolation can be used, etc. This application does not limit the specific methods.

[0096] Optionally, the target sampling time value and the target predicted temperature value can be processed first to generate an initial temperature curve. Then, the initial temperature curve can be processed to determine the curvature of each node in the initial temperature curve. Then, based on the relationship between the curvature of each node and the first curvature threshold and the second curvature threshold, the initial temperature curve can be updated to generate the hot-rolled strip temperature curve.

[0097] The first curvature threshold is less than the second curvature threshold. The first curvature threshold and the second curvature threshold can be preset values ​​or can be adjusted according to actual needs. This application does not limit this.

[0098] For example, we can first create a node set consisting of the target sampling time value and the target predicted temperature value. As initial nodes, an initial temperature curve is constructed using cubic spline interpolation. Then, the curvature of the initial temperature curve is calculated at each node. It can satisfy the following relationship: ; in, The function representing the initial temperature curve. The first derivative, It is the second derivative.

[0099] Then, the curvature of each node can be compared with the first curvature threshold. Second curvature threshold The relationship between the two is used to classify the curvature and update the initial temperature profile. For example, for... In flat areas, a larger node spacing can be maintained; for In mutation regions, additional nodes can be inserted within these regions to capture details; for In the transition region, the nodes can be appropriately densified. Then, some nodes in the smooth regions can be removed, and new target sampling time values ​​and target predicted temperature values ​​can be selected as new nodes for the transition or abrupt change regions, but the total number of nodes must remain unchanged from the target sampling number. After updating the node set, cubic spline interpolation is performed again to ensure that the function itself, its first derivative, and its second derivative are continuous throughout the entire interval.

[0100] Therefore, in this embodiment of the application, an adaptive cubic spline interpolation algorithm can be used for curve reconstruction. By dynamically monitoring the curvature of the interpolation curve at each node and comparing it with a preset curvature threshold, additional nodes can be adaptively inserted in temperature abrupt changes or key transition regions, thereby improving the global smoothness of the reconstructed curve and the ability to restore local physical details.

[0101] Optionally, the above adaptive interpolation results can be compared with the dense uniform interpolation results to calculate the maximum fitting error and the required computation time. The results show that the method provided in this application can significantly reduce the amount of computation while keeping the maximum fitting error controllable.

[0102] The following is combined Figure 3 The generation process of the temperature curve of hot-rolled strip provided in this application is briefly explained.

[0103] This can be achieved by first obtaining multiple time values ​​within each hot rolling process interval. Multiple actual temperature values ​​T mj This data forms a dataset. Then, by combining the thermodynamic equations, boundary conditions, and the dataset, the data loss function, boundary condition loss function, and physical loss function can be determined. These loss functions are then weighted and fused to determine the total loss function for each hot rolling process interval. The total loss function can be used to train the initial network model to generate a target network model, such as the PINN model, for each hot rolling process interval. Then, for each hot rolling process interval, the corresponding target network model can be used to generate predicted temperature values ​​corresponding to multiple sampling time points. The sampling time points and predicted temperature values ​​can form data pairs.

[0104] Then, the target number of samples for each hot rolling process interval can be determined using the Bayesian-XGBoost model.

[0105] Specifically, we can first use the actual data of each hot rolling process interval, the predicted data obtained by the PINN model, and historical temperature data to determine the sampling quantity interval corresponding to each hot rolling process interval. Then, we can perform sampling processing within the sampling quantity interval to determine the initial temperature value corresponding to each initial sampling group. We can then use a deep regression model to determine the initial predicted value of mechanical properties for each initial temperature value. Based on the difference between the initial predicted value of mechanical properties and the actual value of mechanical properties, we can determine the error value of the sampling quantity in each initial sampling group, and thus determine the initial sampling quantity.

[0106] After determining the initial sampling quantity, the first temperature value corresponding to the initial sampling quantity can be obtained. Then, a shallow regression model is used to process each first temperature value to determine the first error corresponding to the initial sampling quantity. Combined with a first Gaussian process surrogate model, the first prediction error corresponding to the initial sampling quantity is determined. Next, the relationship between the first error corresponding to the initial sampling quantity and the first prediction error is compared. If the first error is less than the first prediction error, a deep regression model can be used to process each first temperature value. Combined with a second Gaussian process surrogate model, the acquisition function value is determined. The sampling quantity corresponding to the current maximum acquisition function value is then determined as the new initial sampling quantity. The above process is repeated based on the new initial sampling quantity until the iteration termination condition is met, thus determining the target sampling quantity. If the first error is greater than or equal to the first prediction error, a new initial sampling quantity can be determined, and the above process is repeated based on the new initial sampling quantity until the target sampling quantity is determined.

[0107] Then, based on the target sampling quantity, data points can be selected to determine the target sampling time value and the corresponding target predicted temperature value. After that, the target sampling time value and the target predicted temperature value are processed, such as fitting the curve through adaptive spline interpolation reconstruction, to generate the hot-rolled strip temperature curve, which can then be output and displayed.

[0108] Therefore, in this embodiment, an initial network model can be constructed for each hot rolling process interval first, and then jointly trained by combining thermodynamic equation residuals, boundary condition penalties, and data loss to obtain a highly reliable target network model whose prediction results both conform to physical laws and closely approximate measured data. Then, a multi-fidelity Bayesian-XGBoost optimization framework is used, combined with mechanical performance errors, to determine the target sampling number for each hot rolling process interval, i.e., the optimal number of data points, taking into account temperature change characteristics, mechanical performance errors, and data costs. Next, the target sampling time value and target predicted temperature value corresponding to the target sampling number are obtained. Adaptive cubic spline interpolation is used to reconstruct the temperature-time curve. Based on the relationship between the curvature of each node and the first and second curvature thresholds, additional nodes are inserted in abrupt or transitional regions to improve curve smoothness and local detail restoration capabilities. The final generated temperature-time curve achieves high precision, high smoothness, and complete restoration of local details, thereby improving the accuracy and reliability of hot-rolled strip temperature curve generation. Subsequently, based on the temperature curve of hot-rolled strip steel, process parameters such as laminar cooling water volume and mill speed can be adjusted and controlled in real time during feedforward control and adaptive optimization processes to achieve precise control of strip steel performance, providing technical support for realizing intelligent, high-precision and high-quality hot continuous rolling production.

[0109] In this embodiment, the sampling time values ​​are first input into a trained target network model. The target network model processes these values ​​to generate predicted temperature values. The total loss function of the target network model includes a physical loss function, a boundary condition loss function, and a data loss function. The target network model corresponds to a hot-rolling process interval. Then, based on the mechanical property error, the target sampling quantity corresponding to the hot-rolling process interval is determined. Next, based on the target sampling quantity, the target sampling time value and the corresponding target predicted temperature value are determined from the sampling time values ​​and the corresponding predicted temperature values. Finally, the target sampling time value and the target predicted temperature value are processed to generate the hot-rolled strip temperature curve. Thus, the trained target network model can generate predicted temperature values. Then, using the mechanical property error as the optimization objective, the target sampling quantity and the corresponding target sampling time value and target predicted temperature value are determined in reverse. After smooth interpolation, the hot-rolled strip temperature curve is generated. This process balances physical rationality, mechanical property error, and data cost, ensuring that the curve simultaneously possesses global smoothness and high fidelity in local details, effectively improving the accuracy and reliability of hot-rolled strip temperature prediction.

[0110] According to this application, a device 400 for generating temperature profiles of hot-rolled strip steel is provided, such as... Figure 4 As shown, the device 400 includes a first generation module 410, a first determination module 420, a second determination module 430, and a second generation module 4400.

[0111] The first generation module 410 is used to input the sampled time value into the trained target network model so that the target network model processes the sampled time value to generate the predicted temperature value. The total loss function of the target network model includes a physical loss function, a boundary condition loss function and a data loss function. The target network model corresponds to the hot rolling process range.

[0112] The first determining module 420 is used to determine the target sampling quantity corresponding to the hot rolling process range based on the mechanical property error.

[0113] The second determining module 430 is used to determine the target sampling time value and the corresponding target predicted temperature value from the sampling time value and the corresponding predicted temperature value based on the target sampling quantity.

[0114] The second generation module 440 is used to process the target sampling time value and the target predicted temperature value to generate a hot-rolled strip temperature curve.

[0115] Optionally, the first determining module 420 includes: The acquisition unit is used to acquire the first temperature value corresponding to the initial sampling quantity from the first dataset; The first determining unit is used to process each of the first temperature values ​​using a shallow regression model and combine it with a first Gaussian process surrogate model to determine the first prediction error corresponding to the initial sampling quantity. The second determining unit is used to process each of the first temperature values ​​using a deep regression model and, in conjunction with a second Gaussian process surrogate model, determine the acquisition function value when the first error corresponding to the initial sampling quantity is less than the first prediction error. The third determining unit is used to determine the number of samples corresponding to the current maximum acquisition function value as the new initial number of samples, and return to execute the above steps of obtaining the first temperature value corresponding to the initial number of samples from the first dataset until the iteration termination condition is reached, and determine the target number of samples.

[0116] Optionally, the first determining unit is specifically used for: A shallow regression model is used to process each of the first temperature values ​​to determine the first predicted value of the mechanical properties for each of the first temperature values; Based on the difference between the first predicted value of mechanical properties and the actual value of mechanical properties for each first temperature value, a first error of mechanical properties for each first temperature value is determined; The maximum value in the first error of mechanical properties is determined as the first error corresponding to the initial sampling number; Based on the first prediction distribution under the first Gaussian process surrogate model, the first prediction error corresponding to the initial sampling quantity is obtained.

[0117] Optionally, the second determining unit is specifically used for: A deep regression model is used to process each of the first temperature values ​​to determine a second predicted value of the mechanical properties for each of the first temperature values; Based on the difference between the second predicted value of mechanical properties and the actual value of mechanical properties for each first temperature value, a second error of mechanical properties for each first temperature value is determined; The maximum value in the second error of mechanical properties is determined as the second error corresponding to the initial sampling quantity; The initial sampling number and the second error are added to the second dataset as new training data pairs, and the current second Gaussian process proxy model is updated. The acquisition function value is determined based on the second prediction distribution under the updated second Gaussian process surrogate model and the current minimum error value.

[0118] Optionally, the first determining module 420 further includes: The fourth determining unit is used to perform statistical processing on the historical temperature data of the hot rolling process interval to determine the sampling quantity interval corresponding to the hot rolling process interval; The fifth determining unit is used to perform sampling processing on the sampling quantity range to determine the initial temperature value corresponding to each initial sampling group; The sixth determining unit is used to process each of the initial temperature values ​​using a deep regression model to determine the initial predicted value of the mechanical properties for each of the initial temperature values. The seventh determining unit is used to determine the initial error of the mechanical properties of each initial temperature value based on the difference between the initial predicted value of the mechanical properties and the actual value of the mechanical properties. The eighth determining unit is used to determine the maximum initial mechanical property error in each initial sampling group as the error value of the number of samples in the corresponding initial sampling group; The ninth determining unit is used to determine the number of samples corresponding to the maximum error value as the initial number of samples.

[0119] Optionally, the ninth determining unit is specifically used for: The initial Gaussian process surrogate model is trained using the second dataset to generate a second Gaussian process surrogate model, wherein the second dataset includes data pairs consisting of the number of samples and the error value; The initial prediction distribution under the second Gaussian process surrogate model and the minimum error value in the second dataset are processed to determine the acquisition function value; The number of samples corresponding to the maximum acquisition function value is determined as the initial number of samples.

[0120] Optionally, the second generation module 440 is specifically used for: The target sampling time value and the target predicted temperature value are processed to generate an initial temperature curve; The initial temperature curve is processed to determine the curvature of each node in the initial temperature curve; The initial temperature curve is updated based on the relationship between the curvature of each node and the first curvature threshold and the second curvature threshold to generate a hot-rolled strip temperature curve, wherein the first curvature threshold is less than the second curvature threshold.

[0121] Optionally, the first generation module 410 is further configured to: The total temperature difference of the hot rolling process interval is determined based on the thermodynamic model corresponding to the hot rolling process interval. The time values ​​of the hot rolling process interval are input into the initial network model, and after processing by the first initial network model, the temperature prediction value corresponding to each time value is determined. The data loss value is determined based on the difference between the predicted temperature value and the actual temperature value at each time point. The inlet time value of the hot rolling process zone is input into the initial network model, and the inlet temperature prediction value is determined after processing by the first initial network model. The boundary condition loss value is determined based on the difference between the predicted inlet temperature and the actual inlet temperature. The predicted temperature value, the predicted inlet temperature value, and the total temperature difference value are processed to determine the physical loss value; The data loss value, boundary condition loss value, and physical loss value are fused together to determine the total loss value; The initial network model is trained based on the total loss value to generate the trained target network model.

[0122] The device for generating the temperature curve of hot-rolled strip steel provided in this application can first input the sampling time value into a pre-trained target network model. After processing by the target network model, the predicted temperature value corresponding to the sampling time value is generated. The total loss function of the target network model includes a physical loss function, a boundary condition loss function, and a data loss function. The target network model corresponds to the hot-rolling process interval. Then, based on the mechanical property error, the target sampling quantity corresponding to the hot-rolling process interval is determined. Then, based on the target sampling quantity, the target sampling time value and the corresponding target predicted temperature value are determined from the sampling time value and the corresponding predicted temperature value. Finally, the target sampling time value and the target predicted temperature value are processed to generate the temperature curve of hot-rolled strip steel. Thus, the trained target network model can generate the predicted temperature value. Then, with the mechanical property error as the optimization target, the target sampling quantity and the corresponding target sampling time value and target predicted temperature value are determined in reverse. After smooth interpolation, the temperature curve of hot-rolled strip steel is generated. This process takes into account physical rationality, mechanical property error, and data cost, so that the curve has both global smoothness and high fidelity of local details, effectively improving the accuracy and reliability of hot-rolled strip steel temperature prediction.

[0123] It should be understood that the specific features, operations, and details described herein with respect to the methods of this application can also be similarly applied to the apparatus and system of this application, or vice versa. Furthermore, each step of the methods of this application described above can be performed by a corresponding component or unit of the apparatus or system of this application.

[0124] It should be understood that the various modules / units of the device of this application can be implemented wholly or partially through software, hardware, firmware, or a combination thereof. Each module / unit can be embedded in the processor of the electronic device in hardware or firmware form or independent of the processor, or it can be stored in the memory of the electronic device in software form for the processor to call to execute the operation of each module / unit. Each module / unit can be implemented as an independent component or module, or two or more modules / units can be implemented as a single component or module.

[0125] like Figure 5 As shown, this application provides an electronic device 500, which includes a processor 501 and a memory 502 storing computer program instructions. The processor 501 executes the computer program instructions to implement the steps of the method for generating the temperature profile of hot-rolled strip steel described above. This electronic device 500 can be broadly categorized as a server, terminal, or any other electronic device with the necessary computing and / or processing capabilities.

[0126] In one embodiment, the electronic device 500 may include a processor, memory, network interface, communication interface, etc., connected via a system bus. The processor of the electronic device 500 can be used to provide necessary computing, processing, and / or control capabilities. The memory of the electronic device 500 may include non-volatile storage media and internal memory. The non-volatile storage media may store an operating system, computer programs, etc. The internal memory can provide an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface and communication interface of the electronic device 500 can be used to connect and communicate with external devices via a network. When the computer program is executed by the processor, it performs the steps of the method of this application.

[0127] This application provides a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the above-described method for generating the temperature curve of hot-rolled strip steel.

[0128] Those skilled in the art will understand that the method steps of this application can be performed by a computer program instructing related hardware, such as electronic device 500 or a processor. The computer program can be stored in a non-transitory computer-readable storage medium, and its execution causes the steps of this application to be performed. Depending on the context, any reference herein to memory, storage, or other media may include non-volatile or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, etc. Examples of volatile memory include random access memory (RAM), external cache memory, etc.

[0129] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification, provided that such combination does not contain contradictions.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for generating a temperature profile of hot-rolled strip steel, characterized in that, include: The sampled time value is input into the trained target network model, and after processing by the target network model, the predicted temperature value corresponding to the sampled time value is generated. The total loss function of the target network model includes a physical loss function, a boundary condition loss function, and a data loss function. The target network model corresponds to the hot rolling process range. The target sampling quantity corresponding to the hot rolling process range is determined based on the mechanical property error. Based on the target sampling quantity, the target sampling time value and the corresponding target predicted temperature value are determined from the sampling time value and the corresponding predicted temperature value. The target sampling time value and the target predicted temperature value are processed to generate the temperature curve of hot-rolled strip steel.

2. The method as described in claim 1, characterized in that, The step of determining the target sampling quantity corresponding to the hot rolling process range based on mechanical property errors includes: Obtain the first temperature value corresponding to the initial number of samples from the first dataset; Each of the first temperature values ​​is processed using a shallow regression model, and combined with a first Gaussian process surrogate model, the first prediction error corresponding to the initial number of samples is determined. If the first error corresponding to the initial sampling quantity is less than the first prediction error, a deep regression model is used to process each of the first temperature values, and a second Gaussian process surrogate model is combined to determine the acquisition function value. The number of samples corresponding to the current maximum acquisition function value is determined as the new initial number of samples, and the process of obtaining the first temperature value corresponding to the initial number of samples from the first dataset is repeated until the iteration termination condition is met, and the target number of samples is determined.

3. The method as described in claim 2, characterized in that, The process of using a shallow regression model to process each of the first temperature values, and combining it with a first Gaussian process surrogate model, to determine the first prediction error corresponding to the initial sampling quantity includes: A shallow regression model is used to process each of the first temperature values ​​to determine the first predicted value of the mechanical properties for each of the first temperature values; Based on the difference between the first predicted value of the mechanical properties and the actual value of the mechanical properties for each first temperature value, a first error of the mechanical properties for each first temperature value is determined; The maximum value in the first error of mechanical properties is determined as the first error corresponding to the initial sampling quantity; Based on the first prediction distribution under the first Gaussian process surrogate model, the first prediction error corresponding to the initial sampling quantity is obtained.

4. The method as described in claim 2, characterized in that, The process of using a deep regression model to process each of the first temperature values, and combining it with a second Gaussian process surrogate model to determine the acquisition function value, includes: A deep regression model is used to process each of the first temperature values ​​to determine a second predicted value of the mechanical properties for each of the first temperature values; Based on the difference between the second predicted value of mechanical properties and the actual value of mechanical properties for each first temperature value, a second error of mechanical properties for each first temperature value is determined; The maximum value in the second error of mechanical properties is determined as the second error corresponding to the initial sampling quantity; The initial sampling number and the second error are added to the second dataset as new training data pairs, and the current second Gaussian process proxy model is updated. The acquisition function value is determined based on the second prediction distribution under the updated second Gaussian process surrogate model and the current minimum error value.

5. The method as described in claim 2, characterized in that, Before obtaining the first temperature value corresponding to the initial sampling quantity from the first dataset, the method further includes: Statistical processing is performed on the historical temperature data of the hot rolling process range to determine the sampling quantity range corresponding to the hot rolling process range; The sampling quantity range is sampled to determine the initial temperature value corresponding to each initial sampling group; A deep regression model is used to process each of the initial temperature values ​​to determine the initial predicted values ​​of the mechanical properties for each of the initial temperature values; The initial error of mechanical properties for each initial temperature value is determined based on the difference between the initial predicted value and the actual value of mechanical properties for each initial temperature value. The maximum initial mechanical property error in each initial sampling group is determined as the error value of the number of samples in the corresponding initial sampling group; The number of samples corresponding to the maximum error value is determined as the initial number of samples.

6. The method as described in claim 5, characterized in that, The step of determining the number of samples corresponding to the maximum error value as the initial number of samples includes: The initial Gaussian process surrogate model is trained using the second dataset to generate a second Gaussian process surrogate model, wherein the second dataset includes data pairs consisting of the number of samples and the error value; The initial prediction distribution under the second Gaussian process surrogate model and the minimum error value in the second dataset are processed to determine the acquisition function value; The number of samples corresponding to the maximum acquisition function value is determined as the initial number of samples.

7. The method as described in claim 1, characterized in that, The step of processing the target sampling time value and the target predicted temperature value to generate a hot-rolled strip temperature curve includes: The target sampling time value and the target predicted temperature value are processed to generate an initial temperature curve; The initial temperature curve is processed to determine the curvature of each node in the initial temperature curve; The initial temperature curve is updated based on the relationship between the curvature of each node and the first curvature threshold and the second curvature threshold to generate a hot-rolled strip temperature curve, wherein the first curvature threshold is less than the second curvature threshold.

8. The method as described in claim 1, characterized in that, Before inputting the sampled time value into the trained target network model for processing and generating the predicted temperature value corresponding to the sampled time value, the method further includes: The total temperature difference of the hot rolling process interval is determined based on the thermodynamic model corresponding to the hot rolling process interval. The time values ​​of the hot rolling process interval are input into the initial network model, and after processing by the first initial network model, the temperature prediction value corresponding to each time value is determined. The data loss value is determined based on the difference between the predicted temperature value and the actual temperature value at each time point. The inlet time value of the hot rolling process zone is input into the initial network model, and the inlet temperature prediction value is determined after processing by the first initial network model. The boundary condition loss value is determined based on the difference between the predicted inlet temperature and the actual inlet temperature. The predicted temperature value, the predicted inlet temperature value, and the total temperature difference value are processed to determine the physical loss value; The data loss value, boundary condition loss value, and physical loss value are fused together to determine the total loss value; The initial network model is trained based on the total loss value to generate the trained target network model.

9. A device for generating temperature profiles of hot-rolled strip steel, characterized in that, include: The first generation module is used to input the sampled time value into the trained target network model so that the target network model processes the sampled time value to generate the predicted temperature value. The total loss function of the target network model includes a physical loss function, a boundary condition loss function and a data loss function. The target network model is corresponding to the hot rolling process range. The first determining module is used to determine the target sampling quantity corresponding to the hot rolling process range based on the mechanical property error; The second determining module is used to determine the target sampling time value and the corresponding target predicted temperature value from the sampling time value and the corresponding predicted temperature value based on the target sampling quantity. The second generation module is used to process the target sampling time value and the target predicted temperature value to generate the temperature curve of hot-rolled strip steel.

10. An electronic device, characterized in that, The electronic device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the method for generating the temperature curve of hot-rolled strip steel as described in any one of claims 1-8.

Images