H-shaped steel flatness prediction method, control method and system, medium and electronic equipment

Abstract

The present application relates to a kind of H-shaped steel flatness prediction method, control method and system, medium and electronic equipment, belong to H-shaped steel straightening technical field.Method includes data acquisition, initial flatness prediction, residual stress prediction, and input flatness prediction model after straightening of predicted initial flatness, predicted residual stress distribution data and second data, obtain the predicted final flatness of H-shaped steel finished product straightening after head and tail and multiple positions in the middle;According to the target flatness of production requirement, compare the difference degree of predicted final flatness and target flatness, and according to the difference degree, match corresponding control strategy from strategy library;According to the control strategy obtained by matching, control the working parameter of each component of rolling production line and / or cooling production line and / or straightening production line.The present application deeply fuses rolling, cooling and straightening process and flatness parameter, fully considers the influence of residual stress distribution on straightening process, realizes the high flatness control of H-shaped steel rolling piece.

Description

Technical Field

[0001] This invention belongs to the field of H-beam straightening technology, and relates to a method, control method and system, medium and electronic equipment for predicting the straightness of H-beams. Background Technology

[0002] H-beams are a core supporting material in grand projects such as green prefabricated buildings, large steel structure systems, skyscrapers, and bridges spanning rivers and seas. In particular, small and medium-sized H-beams are the mainstream products in the market, and their production scale continues to expand. According to statistics in 2020, the annual output reached 51.74 million tons, and exceeded 60 million tons the following year, accounting for more than 70% of the total output of structural steel, demonstrating the vigorous development trend of the industry.

[0003] With the expanding applications of H-beams and the surge in market demand, the requirements for diversification and precision in steel dimensions and specifications are increasing, posing a dual challenge to steel companies to improve both production efficiency and product quality. The production process of H-beams mainly includes three stages: rolling, cooling, and straightening, with straightening being the key step determining product quality. Straightening technology not only affects the final shape of the product but is also an important indicator of the advancement of production technology.

[0004] To address this challenge, domestic and international companies have adopted advanced multi-roll straightening equipment and continuous straightening technology, transforming traditional mechanical reduction adjustment devices into hydraulic ones, greatly improving adjustment accuracy and flexibility. Simultaneously, in-depth exploration of straightening theory has driven the optimization and upgrading of straightening equipment and the scientific formulation of process solutions. For example, the application of cantilever straighteners, with their convenient roll changing operation, good adaptability, and high production efficiency, has become a new favorite in the industry.

[0005] However, existing straightening processes mostly focus on the standardized application of rolling procedures, often neglecting the potential impact of residual stress generated during rolling and cooling on the straightness of H-beams. Residual stress can cause H-beams to exhibit unevenness such as bending, twisting, or warping along their length. These unevennesses not only affect the appearance quality of the components but may also reduce their load-bearing capacity and performance. During rolling, H-beams undergo intense plastic deformation, which can lead to uneven stress distribution within the material. Rolling parameters such as rolling temperature, rolling speed, and rolling force can all affect the distribution of residual stress. Properly controlling the rolling process can effectively reduce the generation of residual stress. During cooling, the temperature of H-beams drops rapidly, leading to thermal stress within the material. The choice of cooling rate and cooling method has a significant impact on residual stress. While residual stress generated during rolling and cooling processes is unavoidable, the subsequent straightening process uses mechanical methods to induce elastic or plastic deformation in the H-beams to eliminate or reduce internal residual stress. Existing straightening processes do not consider the influence of residual stress when selecting parameters such as roller diameter, roller type, and straightening force of the straightening machine, resulting in very limited control over the flatness of the finished product.

[0006] Furthermore, the analysis of straightening data and the setting of straightening parameters rely heavily on manual experience, lacking intelligent analysis and utilization. This leads to frequent process adjustments and makes it difficult to guarantee the stability of product straightness. In addition, existing control methods have limitations in universality, failing to comprehensively cover various H-beam specifications, exhibiting poor adaptability to new sample data, and requiring high parameter accuracy and data quality, thus increasing operational difficulty and cost. Summary of the Invention

[0007] In view of this, the purpose of this invention is to provide a method, control method and system, medium and electronic equipment for predicting the straightness of H-beams, comprehensively analyzing the influence of rolling, cooling and straightening processes on straightness, especially considering the influence and utilization of residual stress, and proposing control strategies for rolling, cooling and straightening processes to achieve high straightness production of H-beams.

[0008] To achieve the above objectives, this invention provides a method for predicting the straightness of H-beams, comprising the following steps:

[0009] Obtain the first data and the second data;

[0010] Input the first data into the initial straightness prediction model to obtain the predicted initial straightness of the H-beam before straightening;

[0011] Input the first data into the residual stress prediction model to obtain the predicted residual stress distribution data of the H-beam before straightening;

[0012] The predicted initial straightness, predicted residual stress distribution data, and the second data are input into the straightness prediction model after straightening to obtain the predicted final straightness at multiple positions at the head, tail, and middle of the finished H-beam after straightening.

[0013] The first data includes the first sub-data measured at the entrance position of the final rolling pass of the H-beam billet and the second sub-data measured at the exit position of the final rolling pass of the H-beam finished product; the first sub-data includes sample data of the external dimensions of the flange and web of the H-beam billet at multiple locations, the rolling reduction data of multiple locations, the running speed sample data, and the temperature field distribution sample data.

[0014] The second set of data includes sample data of the external dimensions of the finished H-beams at multiple locations on the flanges and webs, sample data of the operating speed, sample data of the temperature field distribution from the mill exit to the cooling bed inlet, and the chemical composition of the H-beams.

[0015] The second set of data includes dimensional sample data of the flanges and webs of the finished H-beams at multiple locations, various process parameters of the straightening production line, fixed or multiple lengths, and the chemical composition of the H-beams.

[0016] When performing initial straightness prediction, the first sub-data is input into the initial straightness prediction model to obtain the predicted initial straightness at multiple positions at the head, tail, and middle of the H-beam before straightening.

[0017] When predicting residual stress, the second sub-data is input into the residual stress prediction model to obtain the predicted residual stress distribution data at multiple positions at the head, tail, and middle of the H-beam before straightening.

[0018] The process of constructing the initial flatness prediction model includes:

[0019] Obtain the first historical production dataset, including rolling and cooling process parameters and the measured initial straightness of H-beams before straightening;

[0020] The first neural network was trained based on the first historical production dataset to obtain the initial flatness prediction model;

[0021] Correct the initial straightness prediction model and determine the output results.

[0022] The initial straightness prediction model was corrected, and the output results were determined, including:

[0023] The residual sequence of multiple sets of first initial straightness prediction values ​​output by the initial straightness prediction model and the measured initial straightness of the corresponding data sets is processed to obtain multiple sets of initial straightness deviation values.

[0024] Multiple sets of initial straightness deviation values ​​are input into the backpropagation neural network to obtain the initial straightness deviation correction value;

[0025] The initial straightness deviation correction value is compensated to the first initial straightness prediction value to obtain the second initial straightness prediction value;

[0026] The second initial flatness prediction value is output as the predicted initial flatness.

[0027] Optionally, the first historical production dataset can be the rolling and cooling process parameters and the measured initial straightness of H-beams before straightening for at least six months at the production site, or the rolling and cooling process parameters and the measured initial straightness of H-beams before straightening for an H-beam production line with an output of more than 500,000 tons.

[0028] Training the first neural network based on the first historical production dataset includes:

[0029] The first original model based on runtime time series is constructed using recurrent neural networks, stacked recurrent neural networks, and recurrent convolutional neural networks.

[0030] The first historical production dataset is divided into a first training set and a first test set according to a set ratio;

[0031] Multiple factors affecting the initial straightness of H-beams are selected from the first historical production dataset as input layers, and the initial straightness of H-beams is used as the output layer.

[0032] Select the training function of the first original model to obtain the initial flatness prediction model.

[0033] Optionally, the factors affecting the initial straightness of H-beams are the layout and operating parameters of each component of the rolling and cooling production lines, and the number of neurons in the input layer is the number of factors affecting the initial straightness selected.

[0034] If the number of data samples in the first training set is greater than 100,000, the squared loss function is selected as the training function. The squared loss function is L(f(x),y)=(f(x)-y). 2 If the number of data samples in the first training set is less than 100,000, the absolute value loss function is selected as the training function, and the absolute value loss function is L(Y,f(x))=|Yf(x)|.

[0035] Optionally, the factors that affect the initial straightness of H-beams in the first historical production data include: the position of the universal mill in the final rolling mill, the deformation amount of the universal mill pass, the rolling temperature of the universal mill, the position of the edge mill, the deformation amount of the edge mill pass, the position of the universal finishing mill, the deformation amount of the universal finishing mill pass, the position of the water tank after finishing, the water volume of the water tank after finishing, the water pressure of the water tank after finishing, the number of nozzles in the water tank after finishing, the water pressure of the nozzles in the water tank after finishing, the final rolling speed of the workpiece, the final rolling temperature of the workpiece, the section modulus of the workpiece, the distance from the mill exit to the cooling bed entrance, and the temperature of the workpiece on the cooling bed.

[0036] Optionally, the first historical production dataset can be divided into a first training set and a first test set in an 8:2 ratio. The first training set is used to train and optimize the initial flatness prediction model, and the first test set is used to verify the results of the initial flatness prediction model.

[0037] The process of constructing the residual stress prediction model includes:

[0038] A second historical production dataset is obtained, including rolling and cooling process parameters and predicted residual stress distribution data before H-beam straightening; a second neural network is trained based on the second historical production dataset to predict the residual stress.

[0039] Optionally, the second historical production dataset may be rolling and cooling process parameters and predicted residual stress distribution data of H-beams before straightening for at least six months at the production site, or rolling and cooling process parameters and predicted residual stress distribution data of H-beams before straightening for an H-beam production line with an output of more than 500,000 tons.

[0040] Training a second neural network based on a second historical production dataset includes:

[0041] A second original model based on runtime time series is constructed using recurrent neural networks, stacked recurrent neural networks, and recurrent convolutional neural networks.

[0042] The second historical production dataset is divided into a second training set and a second test set according to a set ratio;

[0043] Multiple factors affecting the residual stress of H-beams before straightening are selected from the second historical production dataset as input layers, and the residual stress of H-beams before straightening is used as the output layer.

[0044] By selecting the training function of the second original model, a residual stress prediction model is obtained.

[0045] Optionally, the factors affecting the residual stress of H-beams before straightening are the layout and operating parameters of each component of the rolling and cooling production lines, as well as the chemical composition of the H-beams. The number of neurons in the input layer is the number of factors selected that affect the residual stress of H-beams before straightening.

[0046] If the number of data samples in the first training set is greater than 100,000, the squared loss function is selected as the training function. The squared loss function is L(f(x),y)=(f(x)-y). 2 If the number of data samples in the first training set is less than 100,000, the absolute value loss function is selected as the training function, and the absolute value loss function is L(Y,f(x))=|Yf(x)|.

[0047] Optionally, the factors that affect the residual stress of H-beams before straightening in the second historical production data set include: the C, Si, Mn, S, P, Cr, Nb, V, and Ti content of the H-beams; the position of the universal mill in the final rolling mill unit; the deformation amount per pass of the universal mill; the rolling temperature of the universal mill; the position of the edging mill; the deformation amount per pass of the edging mill; the position of the universal finishing mill; the deformation amount per pass of the universal finishing mill; the position of the water tank after finishing; the water volume of the water tank after finishing; the water pressure of the water tank after finishing; the number of nozzles in the water tank after finishing; the water pressure of the nozzles in the water tank after finishing; the final rolling speed of the workpiece; the final rolling temperature of the workpiece; the cooling rate of the workpiece; the distance from the mill exit to the cooling bed entrance; the temperature of the upper cooling bed of the workpiece; and the temperature of the lower cooling bed of the mill.

[0048] The process of constructing the straightness prediction model after straightening includes:

[0049] Obtain a third historical production dataset that includes straightening process parameters and the final straightness of the H-beams after straightening;

[0050] A third neural network was trained based on the third historical production dataset to obtain a straightening flatness prediction model.

[0051] Correct the straightness prediction model after straightening and determine the output results;

[0052] The straightness prediction model after correction and straightening determines the following output results:

[0053] The residual sequence of multiple sets of first final flatness prediction values ​​output by the straightened flatness prediction model and the actual final flatness of the corresponding data sets is processed to obtain multiple sets of final flatness deviation values.

[0054] Multiple sets of final straightness deviation values ​​are input into the backpropagation neural network to obtain the final straightness deviation correction value;

[0055] The final straightness deviation correction value is compensated to the first final straightness prediction value to obtain the second final straightness prediction value;

[0056] The second final flatness prediction value is output as the final flatness.

[0057] Optionally, the third historical production dataset can be the straightening process parameters and the measured final straightness of H-beams after straightening for at least six months at the production site, or the straightening process parameters and the measured final straightness of H-beams after straightening for an H-beam production line with an output of more than 500,000 tons.

[0058] Training a third neural network based on a third historical production dataset includes:

[0059] A third original model based on runtime time series is constructed using recurrent neural networks, stacked recurrent neural networks, and recurrent convolutional neural networks.

[0060] The third historical production dataset is divided into a third training set and a third test set according to a set ratio;

[0061] Multiple factors affecting the final straightness of H-beams are selected from the third historical production dataset as input layers, and the final straightness of H-beams is used as the output layer.

[0062] Select the training function of the third original model to obtain the straightness prediction model after straightening.

[0063] Optionally, the factors affecting the final straightness of the H-beam are the layout and working parameters of each component of the straightening production line, the chemical composition content of the H-beam, and the number of fixed lengths or multiples of length. The number of neurons in the input layer is the number of factors selected that affect the final straightness of the H-beam.

[0064] If the number of data samples in the third training set is greater than 100,000, the squared loss function is selected as the training function. The squared loss function is L(f(x),y)=(f(x)-y). 2 If the number of data samples in the third training set is less than 100,000, the absolute value loss function is selected as the training function, and the absolute value loss function is L(Y,f(x))=|Yf(x)|.

[0065] Optionally, the factors that affect the straightness of the rolled piece after straightening in the third historical production data include: the C content, Si content, Mn content, S content, P content, Cr content, Nb content, V content, Ti content of the H-beam, straightening temperature, straightening speed, number of straightening rolls, reduction of each straightening roll, and number of fixed lengths or multiples of length.

[0066] This invention also provides a method for controlling the straightness of H-beams, comprising the following steps:

[0067] Using the above prediction method, the predicted final straightness of the H-beam is obtained;

[0068] Based on the target flatness required for production, compare the difference between the predicted final flatness and the target flatness.

[0069] Difference = (|Predicted final flatness - Target initial flatness| ÷ Target initial flatness) × 100%;

[0070] Match the appropriate control strategy from the strategy library based on the degree of difference;

[0071] Based on the control strategy obtained through matching, control the operating parameters of each component of the rolling production line and / or cooling production line and / or straightening production line.

[0072] Optionally, the strategy library includes multiple control strategies, each with its own priority based on its control efficiency. The level of control efficiency is negatively correlated with the time required to achieve the target flatness after adopting the control strategy, and positively correlated with the priority of the control strategy.

[0073] Optionally, the strategy library includes at least a first-class control strategy with higher priority and a second-class control strategy with lower priority. The first-class control strategy is a combined control strategy that controls at least two working parameters related to the rolling process and / or cooling process and / or straightening process. The second-class control strategy is a single control strategy that controls a working parameter related to the rolling process, or cooling process, or straightening process.

[0074] When matching control strategies, if the difference is greater than the preset difference, the first type of control strategy is matched; if the difference is less than the preset difference, the second type of control strategy is matched.

[0075] The present invention also provides an H-beam straightness control system, comprising:

[0076] The data acquisition module is used to acquire first data related to the rolling and cooling process of H-beams and second data related to the straightening process of H-beams.

[0077] The first prediction module is used to input the first data into the initial straightness prediction model to obtain the predicted initial straightness of the head, tail and middle of the H-beam before straightening.

[0078] The second prediction module is used to input the first data into the residual stress prediction model to obtain the predicted residual stress distribution data of the finished H-beam before straightening.

[0079] The third prediction module is used to input the predicted initial straightness, the predicted residual stress distribution data and the second data into the straightening straightness prediction model to obtain the predicted final straightness at the head, tail and middle of the H-beam after straightening.

[0080] The control module is used to compare the predicted final flatness with the target flatness required for production, obtain the difference between the two, match the corresponding control strategy from the strategy library based on the difference, and control the working parameters of each component of the rolling production line and / or cooling production line and / or straightening production line according to the corresponding control strategy.

[0081] Optionally, the data acquisition module is also used to acquire a first historical production dataset including rolling and cooling process parameters and measured initial straightness of H-beams before straightening; a second historical production dataset including rolling and cooling process parameters and predicted residual stress distribution data of H-beams before straightening; and a third historical production dataset including straightening process parameters and final straightness of H-beams after straightening.

[0082] Optionally, the first historical production dataset consists of rolling and cooling process parameters and measured initial straightness of H-beams before straightening for at least six months at the production site, or the rolling and cooling process parameters and measured initial straightness of H-beams before straightening for an H-beam production line with an output exceeding 500,000 tons; the second historical production dataset consists of rolling and cooling process parameters and predicted residual stress distribution data of H-beams before straightening for at least six months at the production site, or the rolling and cooling process parameters and predicted residual stress distribution data of H-beams before straightening for an H-beam production line with an output exceeding 500,000 tons; the third historical production dataset consists of straightening process parameters and measured final straightness of H-beams after straightening for at least six months at the production site, or the straightening process parameters and measured final straightness of H-beams after straightening for an H-beam production line with an output exceeding 500,000 tons.

[0083] Optionally, the first prediction module is also used to construct an initial straightness prediction model. The first prediction module trains a first neural network based on a first historical production dataset, takes multiple factors affecting the initial straightness of the H-beam in the first historical production dataset as input layers, takes the initial straightness of the H-beam as output layers, obtains an initial straightness prediction model, corrects the initial straightness prediction model, and determines the output result.

[0084] Optionally, the second prediction module is also used to construct a residual stress prediction model. The second prediction module trains a second neural network based on a second historical production dataset, using multiple factors in the second historical production dataset that affect the residual stress of the H-beam before straightening as the input layer and the residual stress of the H-beam before straightening as the output layer to obtain the residual stress prediction model.

[0085] Optionally, the third prediction module is also used to construct a straightening flatness prediction model. The third prediction module trains a third neural network based on the third historical production dataset, takes multiple factors affecting the final straightness of the H-beam in the third historical production dataset as input layers, and takes the final straightness of the H-beam as output layers to obtain a straightening flatness prediction model, and corrects the straightening flatness prediction model to determine the output result.

[0086] The present invention also provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor; the memory stores computer-executed instructions; the processor executes the computer-executed instructions stored in the memory to implement the aforementioned method for controlling the straightness of H-beams.

[0087] The present invention also provides a computer-readable storage medium storing computer-executable instructions, which are executed by a processor to perform the aforementioned H-beam straightness control method.

[0088] The present invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the aforementioned method for controlling the straightness of H-beams.

[0089] The beneficial effects of this invention are as follows:

[0090] The H-beam straightness prediction and control method provided by this invention obtains an initial straightness prediction model, a residual stress prediction model, and a post-straightening straightness prediction model by training a neural network based on historical production datasets, thereby obtaining the predicted initial and residual stress distributions. During the control process, the predicted initial straightness, predicted residual stress distribution data, and dimensional sample data from multiple locations on the flanges and web of the finished H-beam, along with various process parameters of the straightening production line, length (fixed or multiple length), and the chemical composition of the H-beam, are input into the post-straightening straightness prediction model to obtain the predicted final straightness at multiple locations at the head, tail, and middle of the finished H-beam after straightening. Further comparison of the difference between the predicted final straightness and the target straightness allows for the matching of appropriate control strategies to control and adjust the working parameters of each component of the rolling, cooling, and straightening production lines, thus achieving the control of the straightness of the H-beam rolled product.

[0091] Furthermore, this invention deeply integrates rolling, cooling, and straightening processes with flatness-related parameters, fully considering the impact of residual stress distribution in H-beams after rolling and cooling on the straightening process. This enables control over the flatness of H-beam rolls, demonstrating strong universality and coverage of a wide range of H-beam products. It also exhibits strong adaptability to newly emerging sample data and is easy to implement. Moreover, because this invention incorporates the dimensions and chemical composition of H-beams as influencing factors into each prediction model during the prediction and control process, its control method can flexibly address the production needs of various steel grades, from ordinary carbon structural steel to high-strength structural steel. It covers a full range of specifications (100mm×50mm to 400mm×200mm), including wide flange, medium flange, narrow flange, and thin-walled sections, demonstrating significant production adaptability.

[0092] Furthermore, this invention considers the change in straightness when the straightened H-beams are further cut into standard or multiple-length products during the construction of the post-straightening straightness prediction model. In other words, this invention takes into account the additional impact of standard or multiple-length products on straightness after the straightening process, thereby controlling the straightening process in advance to ensure high straightness of both standard and multiple-length products. This makes it less likely for the straightened H-beams to bend again after standard or multiple-length cutting, reducing cutting losses and improving metal yield and product quality.

[0093] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0094] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0095] Figure 1 A flowchart illustrating the H-beam straightness prediction method provided by this invention.

[0096] Figure 2 This is a schematic diagram of the process for constructing the initial straightness prediction model in the H-beam straightness prediction method provided by the present invention;

[0097] Figure 3 This is a schematic diagram of the process for constructing a residual stress prediction model in the H-beam straightness prediction method provided by the present invention;

[0098] Figure 4 This is a flowchart illustrating the process of constructing a straightening prediction model in the H-beam straightness prediction method provided by the present invention.

[0099] Figure 5 A flowchart illustrating the H-beam straightness control method provided by this invention;

[0100] Figure 6 This is a schematic diagram of the H-beam straightness control system provided by the present invention.

[0101] Figure label:

[0102] 100 - Data acquisition module; 200 - First prediction module; 300 - Second prediction module; 400 - Third prediction module; 500 - Control module. Detailed Implementation

[0103] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0104] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0105] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0106] Example 1

[0107] This invention provides a method for predicting the straightness of H-beams, such as... Figure 1 This includes the following steps:

[0108] Step S1, Data Acquisition: Obtain the first data related to the H-beam rolling process and the second data related to the H-beam straightening process;

[0109] Step S2, Initial Straightness Prediction: Input the first data into the initial straightness prediction model to obtain the predicted initial straightness of the head, tail and middle of the H-beam before straightening;

[0110] Step S3, Residual Stress Prediction: Input the first data into the residual stress prediction model to obtain the predicted residual stress distribution data of the finished H-beam before straightening;

[0111] Step S4, Final Straightness Prediction: Input the predicted initial straightness, predicted residual stress distribution data, and the second data into the straightening straightness prediction model to obtain the predicted final straightness at the head, tail, and multiple middle positions of the H-beam finished product after straightening.

[0112] The data obtained in step S1 includes:

[0113] The first data includes the first sub-data measured at the entrance position of the final rolling pass of the H-beam billet and the second sub-data measured at the exit position of the final rolling pass of the finished H-beam.

[0114] The first sub-data includes sample data of the external dimensions of H-beam billets at multiple locations on the flanges and webs, sample data of the rolling passes at multiple locations, sample data of the running speed, and sample data of the temperature field distribution.

[0115] The second sub-data includes sample data of the external dimensions of the finished H-beams at multiple locations on the flanges and webs, sample data of the operating speed, sample data of the temperature field distribution from the mill exit to the cooling bed inlet, and the chemical composition of the H-beams.

[0116] The second set of data includes dimensional sample data of the flanges and webs of the finished H-beams at multiple locations, various process parameters of the straightening production line, fixed or multiple lengths, and the chemical composition of the H-beams.

[0117] Step S2 further involves: when performing initial straightness prediction, inputting the first sub-data into the initial straightness prediction model to obtain the predicted initial straightness at multiple positions at the head, tail, and middle of the H-beam before straightening;

[0118] Step S3 further involves: during residual stress prediction, inputting the second sub-data into the residual stress prediction model to obtain the predicted residual stress distribution data of the finished H-beam before straightening.

[0119] Step S2 includes:

[0120] Step S21: Construct the initial flatness prediction model

[0121] Step S22: Input the first data into the residual stress prediction model to obtain the predicted residual stress distribution data of the finished H-beam before straightening.

[0122] like Figure 2 Step S21 includes:

[0123] Step S211: Obtain the first historical production dataset, which includes rolling process parameters and the measured initial straightness of the H-beam before straightening;

[0124] Step S212: Train the first neural network based on the first historical production dataset to obtain the initial flatness prediction model;

[0125] Step S213: Correct the initial flatness prediction model and determine the output results.

[0126] The first historical production dataset in step S211 consists of rolling process parameters and measured initial straightness of H-beams before straightening for at least six months at the production site, or rolling process parameters and measured initial straightness of H-beams before straightening for an H-beam production line with an output of more than 500,000 tons.

[0127] Table 1 shows the first part of the historical production data (for H-beams of H340×250 specification).

[0128]

[0129] Step S212 includes:

[0130] Step S2121: Construct the first original model based on the running time series using recurrent neural networks, stacked recurrent neural networks, and recurrent convolutional neural networks;

[0131] Step S2122: Divide the first historical production dataset into a first training set and a first test set according to a set ratio;

[0132] Step S2123: Select multiple factors affecting the initial straightness of H-beams from the first historical production dataset as input layers, and use the initial straightness of H-beams as the output layer;

[0133] Step S2124: Select the training function of the first original model to obtain the initial flatness prediction model.

[0134] In step S2122, the first historical production dataset is divided into a first training set and a first test set in a ratio of 8:2. The first training set is used to train and optimize the initial flatness prediction model, and the first test set is used to verify the results of the initial flatness prediction model.

[0135] In step S2123, the factors affecting the initial straightness of the H-beam are the layout parameters and operating parameters of each component of the rolling production line and the cooling production line, and the number of neurons in the input layer is the number of factors that affect the straightness selected.

[0136] The factors that affect the initial straightness of H-beams in the first historical production data include: the position of the universal mill in the final rolling mill, the deformation amount of the universal mill pass, the rolling temperature of the universal mill, the position of the edge mill, the deformation amount of the edge mill pass, the position of the universal finishing mill, the deformation amount of the universal finishing mill pass, the position of the water tank after finishing, the water volume of the water tank after finishing, the water pressure of the water tank after finishing, the number of nozzles in the water tank after finishing, the water pressure of the nozzles in the water tank after finishing, the final rolling speed of the workpiece, the final rolling temperature of the workpiece, the section modulus of the workpiece, the distance from the mill exit to the cooling bed entrance, and the temperature of the workpiece on the cooling bed.

[0137] In step S2123, the factors affecting the initial straightness of the H-beam are selected from the first historical production dataset as follows: universal mill position of the final rolling mill, universal mill pass deformation, universal mill rolling temperature, edge mill position, edge mill pass deformation, universal finishing mill position, universal finishing mill pass deformation, finishing water tank position, finishing water tank volume, finishing water tank pressure, finishing water tank nozzle number, finishing water tank nozzle pressure, final rolling speed of the workpiece, final rolling temperature of the workpiece, section modulus of the workpiece, distance from the mill exit to the cooling bed inlet, and temperature of the workpiece on the cooling bed. At this time, the number of neurons in the input layer is 17.

[0138] In step S2124, when selecting the training function, if the number of data samples in the first training set is greater than 100,000, the squared loss function is selected as the training function, and the squared loss function is L(f(x),y)=(f(x)-y). 2 If the number of data samples in the first training set is less than 100,000, the absolute value loss function is selected as the training function, and the absolute value loss function is L(Y,f(x))=|Yf(x)|.

[0139] Step S213 includes:

[0140] Step S2131: Perform residual sequence processing on the multiple sets of first initial straightness prediction values ​​output by the initial straightness prediction model and the measured initial straightness of the corresponding data sets to obtain multiple sets of initial straightness deviation values.

[0141] Step S2132: Input multiple sets of initial straightness deviation values ​​into the backpropagation neural network to obtain the initial straightness deviation correction value;

[0142] Step S2133: Compensate the initial straightness deviation correction value to the first initial straightness prediction value to obtain the second initial straightness prediction value;

[0143] Step S2134: Output the second initial flatness prediction value as the predicted initial flatness.

[0144] Step S3 includes:

[0145] Step S31: Construct a residual stress prediction model;

[0146] Step S32: Input the first data into the residual stress prediction model to obtain the predicted residual stress distribution data of the finished H-beam before straightening.

[0147] like Figure 3 Step S31 includes:

[0148] Step S311: Obtain the second historical production dataset, which includes rolling process parameters and predicted residual stress distribution data before H-beam straightening;

[0149] Step S312: Train a second neural network based on the second historical production dataset to predict residual stress.

[0150] In step S311, the second historical production dataset consists of rolling process parameters and predicted residual stress distribution data of H-beams before straightening for at least six months at the production site, or rolling process parameters and predicted residual stress distribution data of H-beams before straightening for an H-beam production line with an output of more than 500,000 tons.

[0151] Table 2 shows the second part of the historical production data (for H-beams of H340×250 specification).

[0152]

[0153] Step S312 includes:

[0154] Step S3121: Construct a second original model based on the running time series using recurrent neural networks, stacked recurrent neural networks, and recurrent convolutional neural networks;

[0155] Step S3122: Divide the second historical production dataset into a second training set and a second test set according to a set ratio;

[0156] Step S3123: Select multiple factors affecting the residual stress of H-beams before straightening from the second historical production dataset as the input layer, and use the residual stress of H-beams before straightening as the output layer.

[0157] Step S3124: Select the training function of the second original model to obtain the residual stress prediction model.

[0158] In step S3122, if the number of data samples in the first training set is greater than 100,000, the squared loss function is selected as the training function, and the squared loss function is L(f(x),y)=(f(x)-y). 2 If the number of data samples in the first training set is less than 100,000, the absolute value loss function is selected as the training function, and the absolute value loss function is L(Y,f(x))=|Yf(x)|.

[0159] In step S3123, the factors affecting the residual stress of the H-beam before straightening are the layout and operating parameters of each component of the rolling production line and the cooling production line, as well as the chemical composition of the H-beam. The number of neurons in the input layer is the number of factors selected that affect the residual stress of the H-beam before straightening.

[0160] The factors that affect the residual stress of H-beams before straightening in the second historical production data set include: the C, Si, Mn, S, P, Cr, Nb, V, and Ti content of the H-beams; the position of the universal mill in the final rolling mill unit; the deformation amount per pass of the universal mill; the rolling temperature of the universal mill; the position of the edging mill; the deformation amount per pass of the edging mill; the position of the universal finishing mill; the deformation amount per pass of the universal finishing mill; the position of the water tank after finishing; the water volume in the water tank after finishing; the water pressure in the water tank after finishing; the number of nozzles in the water tank after finishing; the water pressure of the nozzles in the water tank after finishing; the final rolling speed of the workpiece; the final rolling temperature of the workpiece; the cooling rate of the workpiece; the distance from the mill exit to the cooling bed entrance; the temperature of the upper cooling bed of the workpiece; and the temperature of the lower cooling bed of the mill.

[0161] In step S3123, the factors affecting the residual stress of the H-beam before straightening are selected as follows: the C content, Si content, Mn content, S content, P content, Cr content, Nb content, V content, Ti content of the H-beam; the position of the universal mill in the final rolling mill group; the deformation amount of the universal mill pass; the rolling temperature of the universal mill; the position of the edge mill; the deformation amount of the edge mill pass; the position of the universal finishing mill; the deformation amount of the universal finishing mill pass; the position of the water tank after finishing; the water volume of the water tank after finishing; the water pressure of the water tank after finishing; the number of nozzles of the water tank after finishing; the water pressure of the nozzles of the water tank after finishing; the final rolling speed of the workpiece; the final rolling temperature of the workpiece; the cooling speed of the workpiece; the distance from the mill exit to the cooling bed inlet; the upper cooling bed temperature of the workpiece; and the lower cooling bed temperature of the mill. At this time, the number of neurons in the input layer is 27.

[0162] Step S4 includes:

[0163] Step S41: Construct a straightness prediction model after straightening;

[0164] Step S42: Input the predicted initial straightness, the predicted residual stress distribution data, and the second data into the straightening straightness prediction model to obtain the predicted final straightness at multiple positions at the head, tail, and middle of the H-beam after straightening.

[0165] like Figure 4 Step S41 includes:

[0166] Step S411: Obtain the third historical production dataset, which includes straightening process parameters and the final straightness of the H-beam after straightening;

[0167] Step S412: Train the third neural network based on the third historical production dataset to obtain the straightening flatness prediction model;

[0168] Step S413: Correct the straightness prediction model after straightening and determine the output result.

[0169] Step S412 includes:

[0170] Step S4121: Construct a third original model based on the running time series using recurrent neural networks, stacked recurrent neural networks, and recurrent convolutional neural networks;

[0171] Step S4122: Divide the third historical production dataset into a third training set and a third test set according to a set ratio;

[0172] Step S4123: Select multiple factors affecting the final straightness of H-beams from the third historical production dataset as input layers, and use the final straightness of H-beams as the output layer;

[0173] Step S4124: Select the training function of the third original model to obtain the straightening flatness prediction model.

[0174] The third historical production dataset in step S411 consists of straightening process parameters and measured final straightness of H-beams after straightening for at least six months at the production site, or straightening process parameters and measured final straightness of H-beams after straightening for an H-beam production line with an output of more than 500,000 tons.

[0175] Table 3 shows the third part of the historical production data (for H-beams of H340×250 specification).

[0176]

[0177] In step S4123, the factors affecting the final straightness of the H-beam are the layout and working parameters of each component of the straightening production line, the chemical composition content of the H-beam, and the number of fixed lengths or multiple lengths. The number of neurons in the input layer is the number of factors that affect the final straightness of the H-beam selected.

[0178] The factors that affect the straightness of the rolled piece after straightening in the third historical production data set include: the C content, Si content, Mn content, S content, P content, Cr content, Nb content, V content, Ti content of the H-beam, straightening temperature, straightening speed, number of straightening rolls, reduction of each straightening roll, and number of fixed lengths or multiples of length.

[0179] In step S4123, the factors affecting the straightness of the rolled piece after straightening are selected from the third historical production dataset as follows: the C, Si, Mn, S, P, Cr, Nb, V, and Ti contents of the H-beam; straightening temperature; straightening speed; number of straightening rolls; the reduction amount of each straightening roll of the nine-roll straightener (8 reduction amount data for 9 straightening rolls); and the number of fixed lengths or multiple lengths. At this time, the number of input layer neurons is 21.

[0180] In step S4124, when selecting the training function, if the number of data samples in the third training set is greater than 100,000, the squared loss function is selected as the training function. The squared loss function is L(f(x),y)=(f(x)-y). 2 If the number of data samples in the third training set is less than 100,000, the absolute value loss function is selected as the training function, and the absolute value loss function is L(Y,f(x))=|Yf(x)|.

[0181] Step S413 includes:

[0182] Step S4131: Perform residual sequence processing on the multiple sets of first final flatness prediction values ​​output by the straightened flatness prediction model and the actual measured final flatness of the corresponding data sets to obtain multiple sets of final flatness deviation values.

[0183] Step S4132: Input multiple sets of final straightness deviation values ​​into the backpropagation neural network to obtain the final straightness deviation correction value;

[0184] Step S4133: Compensate the final straightness deviation correction value to the first final straightness prediction value to obtain the second final straightness prediction value;

[0185] Step S4134: Output the second final flatness prediction value as the final flatness.

[0186] Example 2

[0187] This invention also provides a method for controlling the straightness of H-beams, such as... Figure 5 This includes the following steps:

[0188] Step S1: Using the prediction method provided in Example 1, obtain the predicted final straightness of the H-beam;

[0189] Step S2: Based on the target flatness required for production, compare the predicted final flatness with the target flatness.

[0190] Step S3: Match the corresponding control strategy from the strategy library based on the degree of difference;

[0191] Step S4: Based on the control strategy obtained from the matching, control and / or adjust the working parameters of each component of the rolling and straightening production line.

[0192] The formula for calculating the difference in step S2 is: Difference = (|Predicted final flatness - Target initial flatness| ÷ Target initial flatness) × 100%;

[0193] The strategy library in step S3 includes multiple control strategies. Each control strategy has a priority based on its control efficiency. The level of control efficiency is negatively correlated with the time required to achieve the target flatness after adopting the control strategy, and positively correlated with the priority of the control strategy.

[0194] The strategy library in step S3 includes at least a first-class control strategy with higher priority and a second-class control strategy with lower priority. The first-class control strategy is a combined control strategy that controls at least two working parameters related to the rolling process and / or the straightening process. The second-class control strategy is a single control strategy that controls one working parameter related to the rolling process or the straightening process.

[0195] When matching control strategies in step S3, if the difference is greater than the preset difference, the first type of control strategy is matched; if the difference is less than the preset difference, the second type of control strategy is matched.

[0196] Example 3

[0197] This invention also provides an H-beam straightness control system, such as... Figure 6 The system includes a data acquisition module 100 for acquiring first data related to the H-beam rolling and cooling processes and second data related to the H-beam straightening process; a first prediction module 200 for inputting the first data into an initial straightness prediction model to obtain the predicted initial straightness of the H-beam at multiple positions at the head, tail, and middle before straightening; a second prediction module 300 for inputting the first data into a residual stress prediction model to obtain the predicted residual stress distribution data of the finished H-beam before straightening; a third prediction module 400 for inputting the predicted initial straightness, predicted residual stress distribution data, and second data into a post-straightening straightness prediction model to obtain the predicted final straightness of the finished H-beam at multiple positions at the head, tail, and middle after straightening; and a control module 500 for comparing the predicted final straightness with the target straightness required for production, obtaining the difference between the two, matching the corresponding control strategy from the strategy library based on the difference, and controlling the working parameters of each component of the rolling production line and / or cooling production line and / or straightening production line according to the corresponding control strategy.

[0198] The data acquisition module 100 is also used to acquire a first historical production dataset including rolling and cooling process parameters and measured initial straightness of H-beams before straightening; a second historical production dataset including rolling and cooling process parameters and predicted residual stress distribution data of H-beams before straightening; and a third historical production dataset including straightening process parameters and final straightness of H-beams after straightening.

[0199] The first historical production dataset consists of rolling and cooling process parameters and measured initial straightness of H-beams before straightening for at least six months at the production site, or the rolling and cooling process parameters and measured initial straightness of H-beams before straightening for an H-beam production line with an output exceeding 500,000 tons. The second historical production dataset consists of rolling and cooling process parameters and predicted residual stress distribution data of H-beams before straightening for at least six months at the production site, or the rolling and cooling process parameters and predicted residual stress distribution data of H-beams before straightening for an H-beam production line with an output exceeding 500,000 tons. The third historical production dataset consists of straightening process parameters and measured final straightness of H-beams after straightening for at least six months at the production site, or the straightening process parameters and measured final straightness of H-beams after straightening for an H-beam production line with an output exceeding 500,000 tons.

[0200] The first prediction module 200 is also used to construct an initial straightness prediction model. The first prediction module 200 trains a first neural network based on a first historical production dataset, takes multiple factors affecting the initial straightness of the H-beam in the first historical production dataset as input layers, takes the initial straightness of the H-beam as output layers, obtains an initial straightness prediction model, corrects the initial straightness prediction model, and determines the output result.

[0201] The second prediction module 300 is also used to construct a residual stress prediction model. The second prediction module 300 trains a second neural network based on a second historical production dataset, takes multiple factors in the second historical production dataset that affect the residual stress of the H-beam before straightening as the input layer, and takes the residual stress of the H-beam before straightening as the output layer to obtain a residual stress prediction model.

[0202] The third prediction module 400 is also used to construct a straightening flatness prediction model. The third prediction module 400 trains a third neural network based on the third historical production dataset, takes multiple factors affecting the final straightness of the H-beam in the third historical production dataset as input layers, and takes the final straightness of the H-beam as output layers to obtain a straightening flatness prediction model, corrects the straightening flatness prediction model, and determines the output result.

[0203] In some embodiments, the present invention also provides an electronic device, including a processor and a memory communicatively connected to the processor; the memory stores computer-executed instructions.

[0204] In this embodiment, the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. The processor mentioned above may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0205] In some embodiments, the present invention also provides a computer program product, including a computer program / instruction that, when executed by a processor, implements the H-beam straightness control method described in Embodiment 2.

[0206] In some embodiments, the present invention also provides a computer-readable storage medium storing computer-executable instructions, which are executed by a processor using the H-beam straightness control method described in Embodiment 2.

[0207] As will be understood by those skilled in the art, the computer-readable storage medium described in this embodiment allows for the implementation of all or part of the steps in the above method embodiments by computer program-related hardware. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0208] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting the straightness of H-beams, characterized in that: Includes the following steps: Obtain the first data and the second data; Input the first data into the initial straightness prediction model to obtain the predicted initial straightness of the H-beam before straightening; Input the first data into the residual stress prediction model to obtain the predicted residual stress distribution data of the H-beam before straightening; The predicted initial straightness, the predicted residual stress distribution data, and the second data are input into the straightened straightness prediction model to obtain the predicted final straightness at multiple positions at the head, tail, and middle of the H-beam. The process of constructing the initial flatness prediction model includes: Obtain the first historical production dataset, including rolling and cooling process parameters and the measured initial straightness of H-beams before straightening; The first neural network is trained based on the first historical production dataset to obtain the initial flatness prediction model; Correct the initial flatness prediction model and determine the output result; The process of correcting the initial flatness prediction model and determining the output results includes: The residual sequence processing is performed on the multiple sets of first initial straightness prediction values ​​output by the initial straightness prediction model and the measured initial straightness of the corresponding data sets to obtain multiple sets of initial straightness deviation values. The multiple sets of initial straightness deviation values ​​are input into the backpropagation neural network to obtain the initial straightness deviation correction values; The initial straightness deviation correction value is compensated to the first initial straightness prediction value to obtain the second initial straightness prediction value; The second initial flatness prediction value is output as the predicted initial flatness; The process of constructing the residual stress prediction model includes: Obtain a second historical production dataset that includes rolling and cooling process parameters and predicted residual stress distribution data before H-beam straightening; A second neural network is trained based on the second historical production dataset to obtain the residual stress prediction model; wherein... The second historical production dataset consists of rolling and cooling process parameters and predicted residual stress distribution data before H-beam straightening for at least six months at the production site, or rolling and cooling process parameters and predicted residual stress distribution data before H-beam straightening for an H-beam production line with an output of more than 500,000 tons. The process of constructing the straightness prediction model after straightening includes: Obtain a third historical production dataset that includes straightening process parameters and the final straightness of the H-beams after straightening; A third neural network is trained based on the third historical production dataset to obtain the straightened flatness prediction model. Correct the straightening flatness prediction model and determine the output result; The calibration of the straightened flatness prediction model and the determination of the output results include: The residual sequence processing is performed on the multiple sets of first final flatness prediction values ​​output by the straightened flatness prediction model and the actual final flatness of the corresponding data sets to obtain multiple sets of final flatness deviation values. The multiple sets of final straightness deviation values ​​are input into the backpropagation neural network to obtain the final straightness deviation correction value. The final straightness deviation correction value is compensated to the first final straightness prediction value to obtain the second final straightness prediction value; The second final flatness prediction value is output as the final flatness.

2. The prediction method according to claim 1, characterized in that: The first data includes the first sub-data measured at the entrance position of the final rolling pass of the H-beam billet and the second sub-data measured at the exit position of the final rolling pass of the finished H-beam. The first sub-data includes sample data of the external dimensions of H-beam billet flanges and webs at multiple locations, sample data of each pass pressing at multiple locations, sample data of running speed, and sample data of temperature field distribution; The second sub-data includes sample data of the external dimensions of the flanges and webs of the finished H-beams at multiple locations, sample data of the operating speed, sample data of the temperature field distribution from the mill exit to the cooling bed inlet, and the chemical composition of the H-beams; The second data includes sample data of the external dimensions of the flanges and webs of the finished H-beams at multiple locations, various process parameters of the straightening production line, fixed length or multiple length, and the chemical composition of the H-beams.

3. The prediction method according to claim 2, characterized in that: When performing initial straightness prediction, the first sub-data is input into the initial straightness prediction model to obtain the predicted initial straightness at multiple positions at the head, tail, and middle of the H-beam before straightening. When predicting the residual stress distribution, the second sub-data is input into the residual stress prediction model to obtain the predicted residual stress distribution data at multiple positions at the head, tail, and middle of the H-beam before straightening.

4. The prediction method according to claim 1, characterized in that: The first historical production dataset consists of rolling and cooling process parameters and measured initial straightness of H-beams before straightening for at least six months at the production site, or rolling and cooling process parameters and measured initial straightness of H-beams before straightening for an H-beam production line with an output exceeding 500,000 tons; among which, The factors that affect the initial straightness of H-beams in the first historical production data set include: the position of the universal mill in the final rolling mill, the deformation amount of the universal mill pass, the rolling temperature of the universal mill, the position of the edge mill, the deformation amount of the edge mill pass, the position of the universal finishing mill, the deformation amount of the universal finishing mill pass, the position of the water tank after finishing, the water volume of the water tank after finishing, the water pressure of the water tank after finishing, the number of nozzles in the water tank after finishing, the water pressure of the nozzles in the water tank after finishing, the final rolling speed of the workpiece, the final rolling temperature of the workpiece, the section modulus of the workpiece, the distance from the mill exit to the cooling bed entrance, and the temperature of the workpiece on the cooling bed.

5. The prediction method according to claim 4, characterized in that: Training the first neural network based on the first historical production dataset includes: The first original model based on runtime time series is constructed using recurrent neural networks, stacked recurrent neural networks, and recurrent convolutional neural networks. The first historical production dataset is divided into a first training set and a first test set according to a set ratio; Multiple factors affecting the initial straightness of H-beams are selected from the first historical production dataset as input layers, and the initial straightness of H-beams is used as the output layer. Select the training function of the first original model to obtain the initial flatness prediction model; in, The factors affecting the initial straightness of the H-beam are the arrangement parameters and operating parameters of each component of the rolling production line and the cooling production line, and the number of neurons in the input layer is the number of factors affecting the initial straightness selected. If the number of data samples in the first training set is greater than 100,000, the squared loss function is selected as the training function. The squared loss function is: ; If the number of data samples in the first training set is less than 100,000, the absolute value loss function is selected as the training function. The absolute value loss function is... .

6. The prediction method according to claim 1, characterized in that: Training the second neural network based on the second historical production dataset includes: A second original model based on runtime time series is constructed using recurrent neural networks, stacked recurrent neural networks, and recurrent convolutional neural networks. The second historical production dataset is divided into a second training set and a second test set according to a set ratio; Multiple factors affecting the residual stress of H-beams before straightening are selected from the second historical production dataset as input layers, and the residual stress of H-beams before straightening is used as the output layer. Select the training function of the second original model to obtain the residual stress prediction model; in, The factors affecting the residual stress of H-beams before straightening are the layout and operating parameters of each component of the rolling and cooling production lines, as well as the chemical composition of the H-beams. The number of neurons in the input layer is the number of factors selected that affect the residual stress of the H-beam before straightening; If the number of data samples in the second training set is greater than 100,000, the squared loss function is selected as the training function. The squared loss function is: ; If the number of data samples in the second training set is less than 100,000, the absolute value loss function is selected as the training function. The absolute value loss function is... .

7. The prediction method according to claim 6, characterized in that: The factors that affect the residual stress of H-beams before straightening in the second historical production data set include: the C, Si, Mn, S, P, Cr, Nb, V, and Ti content of the H-beams; the position of the universal mill in the final rolling mill group; the deformation amount per pass of the universal mill; the rolling temperature of the universal mill; the position of the edging mill; the deformation amount per pass of the edging mill; the position of the universal finishing mill; the deformation amount per pass of the universal finishing mill; the position of the water tank after finishing; the water volume of the water tank after finishing; the water pressure of the water tank after finishing; the number of nozzles in the water tank after finishing; the water pressure of the nozzles in the water tank after finishing; the final rolling speed of the workpiece; the final rolling temperature of the workpiece; the cooling rate of the workpiece; the distance from the mill exit to the cooling bed entrance; the temperature of the upper cooling bed of the workpiece; and the temperature of the lower cooling bed of the mill.

8. The prediction method according to claim 1, characterized in that: The third historical production dataset consists of straightening process parameters and measured final straightness of H-beams after straightening for at least six months at the production site, or straightening process parameters and measured final straightness of H-beams after straightening for an H-beam production line with an output exceeding 500,000 tons; among which... The factors that affect the straightness of the rolled piece after straightening in the third historical production data set include: the C content, Si content, Mn content, S content, P content, Cr content, Nb content, V content, Ti content of the H-beam, straightening temperature, straightening speed, number of straightening rolls, reduction of each straightening roll, and number of fixed lengths or multiples of length.

9. The prediction method according to claim 8, characterized in that: Training a third neural network based on the aforementioned third historical production dataset includes: A third original model based on runtime time series is constructed using recurrent neural networks, stacked recurrent neural networks, and recurrent convolutional neural networks. The third historical production dataset is divided into a third training set and a third test set according to a set ratio; Multiple factors affecting the final straightness of H-beams are selected from the third historical production dataset as input layers, and the final straightness of H-beams is used as the output layer. Select the training function of the third original model to obtain the straightened flatness prediction model; in, The factors affecting the final straightness of the H-beam are the layout and working parameters of each component of the straightening production line, the chemical composition content of the H-beam, and the number of fixed lengths or multiple lengths. The number of neurons in the input layer is the number of factors that affect the final straightness of the H-beam selected. If the number of data samples in the third training set is greater than 100,000, the squared loss function is selected as the training function. The squared loss function is: ; If the number of data samples in the third training set is less than 100,000, the absolute value loss function is selected as the training function, and the absolute value loss function is: .

10. A method for controlling the straightness of H-beams, characterized in that: Includes the following steps: Using the prediction method described in any one of claims 1 to 9, the predicted final straightness of the H-beam is obtained; Based on the target flatness required for production, the difference between the predicted final flatness and the target flatness is compared. Difference = (|Predicted final flatness - Target initial flatness| ÷ Target initial flatness) × 100%; Match the corresponding control strategy from the strategy library based on the degree of difference; Based on the control strategy obtained from the matching, control the operating parameters of each component of the rolling production line and / or cooling production line and / or straightening production line; in, The strategy library includes multiple control strategies, each with its own priority based on its control efficiency. The level of control efficiency is negatively correlated with the time required to achieve the target flatness after adopting the control strategy, and positively correlated with the priority of the control strategy.

11. The control method according to claim 10, characterized in that: The strategy library includes at least a first-class control strategy with higher priority and a second-class control strategy with lower priority; The first type of control strategy is a combined control strategy that controls at least two operating parameters related to the rolling process and / or cooling process and / or straightening process; The second type of control strategy is a single control strategy that controls an operating parameter related to the rolling process, cooling process, or straightening process; wherein, When matching control strategies, if the difference degree is greater than the preset difference degree, then the first type of control strategy is matched; When matching control strategies, if the difference degree is less than the preset difference degree, then the second type of control strategy is matched.

12. A straightness control system for H-beams, characterized in that: include: The data acquisition module is used to acquire first data related to the rolling and cooling process of H-beams and second data related to the straightening process of H-beams. The first prediction module is used to input the first data into the initial straightness prediction model to obtain the predicted initial straightness of the head, tail and middle of the H-beam before straightening. The second prediction module is used to input the first data into the residual stress prediction model to obtain the predicted residual stress distribution data of the finished H-beam before straightening. The third prediction module is used to input the predicted initial straightness, the predicted residual stress distribution data and the second data into the straightening straightness prediction model to obtain the predicted final straightness at the head, tail and middle of the H-beam after straightening. The control module is used to compare the predicted final flatness with the target flatness required for production, obtain the difference between the two, match the corresponding control strategy from the strategy library according to the difference, and control the working parameters of each component of the rolling production line and / or cooling production line and / or straightening production line according to the corresponding control strategy. The process of constructing the initial flatness prediction model includes: Obtain the first historical production dataset, including rolling and cooling process parameters and the measured initial straightness of H-beams before straightening; The first neural network is trained based on the first historical production dataset to obtain the initial flatness prediction model; Correct the initial flatness prediction model and determine the output result; The process of correcting the initial flatness prediction model and determining the output results includes: The residual sequence processing is performed on the multiple sets of first initial straightness prediction values ​​output by the initial straightness prediction model and the measured initial straightness of the corresponding data sets to obtain multiple sets of initial straightness deviation values. The multiple sets of initial straightness deviation values ​​are input into the backpropagation neural network to obtain the initial straightness deviation correction values; The initial straightness deviation correction value is compensated to the first initial straightness prediction value to obtain the second initial straightness prediction value; The second initial flatness prediction value is output as the predicted initial flatness; The process of constructing the residual stress prediction model includes: Obtain a second historical production dataset that includes rolling and cooling process parameters and predicted residual stress distribution data before H-beam straightening; A second neural network is trained based on the second historical production dataset to obtain the residual stress prediction model; wherein... The second historical production dataset consists of rolling and cooling process parameters and predicted residual stress distribution data before H-beam straightening for at least six months at the production site, or rolling and cooling process parameters and predicted residual stress distribution data before H-beam straightening for an H-beam production line with an output of more than 500,000 tons. The process of constructing the straightness prediction model after straightening includes: Obtain a third historical production dataset that includes straightening process parameters and the final straightness of the H-beams after straightening; A third neural network is trained based on the third historical production dataset to obtain the straightened flatness prediction model. Correct the straightening flatness prediction model and determine the output result; The calibration of the straightened flatness prediction model and the determination of the output results include: The residual sequence processing is performed on the multiple sets of first final flatness prediction values ​​output by the straightened flatness prediction model and the actual final flatness of the corresponding data sets to obtain multiple sets of final flatness deviation values. The multiple sets of final straightness deviation values ​​are input into the backpropagation neural network to obtain the final straightness deviation correction value. The final straightness deviation correction value is compensated to the first final straightness prediction value to obtain the second final straightness prediction value; The second final flatness prediction value is output as the final flatness.

13. The control system according to claim 12, characterized in that: The data acquisition module is also used to acquire a first historical production dataset including rolling and cooling process parameters and measured initial straightness of the H-beam before straightening; a second historical production dataset including rolling and cooling process parameters and predicted residual stress distribution data of the H-beam before straightening; and a third historical production dataset including straightening process parameters and final straightness of the H-beam after straightening; wherein, The first historical production dataset consists of rolling and cooling process parameters and measured initial straightness of H-beams before straightening for at least six months at the production site, or rolling and cooling process parameters and measured initial straightness of H-beams before straightening for an H-beam production line with an output of more than 500,000 tons. The second historical production dataset consists of rolling and cooling process parameters and predicted residual stress distribution data before H-beam straightening for at least six months at the production site, or rolling and cooling process parameters and predicted residual stress distribution data before H-beam straightening for an H-beam production line with an output of more than 500,000 tons. The third historical production dataset consists of straightening process parameters and measured final straightness of H-beams after straightening for at least six months at the production site, or straightening process parameters and measured final straightness of H-beams after straightening for an H-beam production line with an output exceeding 500,000 tons.

14. The control system according to claim 13, characterized in that: The first prediction module is also used to construct the initial straightness prediction model. The first prediction module trains a first neural network based on the first historical production dataset, takes multiple factors affecting the initial straightness of the H-beam in the first historical production dataset as input layers, takes the initial straightness of the H-beam as output layers, obtains the initial straightness prediction model, corrects the initial straightness prediction model, and determines the output result. The second prediction module is also used to construct the residual stress prediction model. The second prediction module trains a second neural network based on the second historical production dataset, takes multiple factors affecting the residual stress of the H-beam before straightening in the second historical production dataset as the input layer, and takes the residual stress of the H-beam before straightening as the output layer to obtain the residual stress prediction model. The third prediction module is also used to construct the straightening flatness prediction model. The third prediction module trains the third neural network based on the third historical production dataset, takes multiple factors affecting the final flatness of the H-beam in the third historical production dataset as input layers, and takes the final flatness of the H-beam as output layers to obtain the straightening flatness prediction model, corrects the straightening flatness prediction model, and determines the output result.

15. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the control method as described in any one of claims 10 to 11.

16. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the control method as described in any one of claims 10 to 11.

17. A computer program product comprising a computer program / instructions, characterized in that: When the computer program / instructions are executed by the processor, they implement the control method as described in any one of claims 10 to 11.

Images