A method for real-time prediction of crown in hot strip rolling process

Abstract

The application discloses a convexity real-time prediction method for a hot strip rolling process, and belongs to the technical field of industrial process performance index prediction, and the method comprises the following steps: obtaining historical data of the hot strip rolling process; performing data alignment in the spatial dimension on the obtained historical data; establishing a distributed convexity prediction model, and training the distributed convexity prediction model by using the aligned historical data; deploying the trained distributed convexity prediction model; obtaining real-time data of the hot strip rolling process, performing data alignment in the spatial dimension on the obtained real-time data; inputting the aligned real-time data into the deployed distributed convexity prediction model, and completing convexity prediction. The application can improve the convexity prediction precision, and provides reliable technical support for production process optimization and quality control.

Description

Technical Field

[0001] This invention relates to the field of industrial process performance index prediction technology, and in particular to a method for real-time prediction of crown in hot strip rolling process. Background Technology

[0002] As a key material widely used in industries such as automobiles, home appliances, and construction, the quality of strip steel directly affects the performance and safety of downstream products. In the rolling process of steel manufacturing, strip crown control is a crucial step, affecting not only the final shape of the rolled strip but also directly determining the yield of the finished product. Although strip crown is one of the key indicators of the quality of hot-rolled strip steel at the exit, closed-loop management of crown control has not yet been achieved. In the finishing rolling process, the exit crown of strip steel is often not effectively controlled, and the size of the exit crown of each slab is difficult to control precisely. Therefore, accurate prediction of strip steel exit crown is a necessary prerequisite for its control.

[0003] In current steel production processes, convexity detection technology primarily focuses on the time dimension, meaning that convexity values ​​are obtained based on data sampling at specific points in time. Typically, this data sampling is achieved through machinery, allowing for periodic or intermittent measurement of strip convexity during production. However, relying solely on the time dimension for convexity detection has limitations, as it cannot comprehensively reflect the spatial distribution of convexity in the slab.

[0004] In actual production, the crown of a strip slab is not uniformly distributed across its entire surface, but rather exhibits spatial variations. These variations can be caused by factors such as material properties, rolling processes, and equipment wear. Therefore, understanding and accurately measuring the spatial crown distribution of the slab is crucial for predicting and controlling crown variations. Through spatial crown data, we can better understand the overall morphology and deformation of the slab, thereby more accurately predicting future crown trends.

[0005] Another challenge is the high-speed data flow in steel production. The amount of data generated on the production floor is enormous, and its flow is extremely rapid. Capturing and processing this data in real time for convexity prediction is a significant technical challenge. Effectively achieving real-time convexity prediction in an environment of real-time data flow requires a combination of efficient data acquisition systems, real-time data processing algorithms, and reliable data transmission and storage facilities. Only under these conditions can timely monitoring and adjustment of convexity changes during production be ensured, thereby achieving real-time monitoring and management of the industrial process.

[0006] In summary, the development of convexity prediction technology needs to overcome measurement discrepancies in both time and space dimensions, and address the challenges of high-speed data flow in steel production. Only by fully understanding and addressing these challenges can the stability and efficiency of the production process be effectively improved, enabling real-time monitoring and control of industrial processes. Overall, the following issues still need to be addressed in convexity prediction technology:

[0007] First, the hot rolling process of strip steel is a complex process, involving multiple steps such as heating furnace, roughing mill, flying shear, finishing mill, laminar flow cooling, and coiler. The finishing mill typically consists of seven stands used to precisely control the size and shape of the strip steel. In the entire hot rolling process, the crown of the finished slab is a crucial quality indicator, directly affecting the forming and performance of downstream products. To improve product quality and production efficiency, crown prediction algorithms are constantly being updated and iterated. However, most current crown prediction models still employ centralized prediction methods, processing and analyzing all data in a single center. These centralized models face numerous challenges, including low prediction accuracy. Because centralized models cannot fully consider the data distribution characteristics and variation patterns during the finishing rolling process, their prediction results often deviate significantly from the actual situation. This problem seriously affects the stability of the production process and the effectiveness of quality control, urgently requiring a more accurate and reliable crown prediction method to address this challenge. To better meet the needs of actual production processes, a real-time crown prediction model that more closely aligns with the automated flow of on-site data needs to be designed. This model should be able to combine the data distribution characteristics during the finishing rolling process to more accurately predict the crown changes of strip slabs, thereby providing a more reliable quality control and optimization scheme for the production process.

[0008] Secondly, the crown setting value of slabs is not fixed. Typically, slab specifications are divided according to exit thickness, and slabs with different exit thicknesses often exhibit significant crown differences. Therefore, effectively controlling the exit crown of slabs of different specifications after rolling is often challenging. Although data-driven prediction models perform relatively accurately when predicting slabs of the same specification, their accuracy is often severely affected when dealing with slabs of various specifications and steel grades. This is because slabs of different specifications and steel grades differ in material properties, production processes, and environmental conditions, leading to significant uncertainty and complexity in their crown variation patterns. Traditional data-driven models may fail to fully capture these differences, resulting in a substantial decrease in accuracy when predicting the crown of slabs of different specifications. To address this issue, selecting a more suitable deep learning algorithm is crucial. Deep learning algorithms possess powerful nonlinear modeling and adaptive learning capabilities, enabling them to better handle complex data relationships and features, thereby improving the model's accuracy in predicting the crown of diverse slabs. By fully utilizing large amounts of production data and advanced deep learning technology, more flexible and efficient prediction models can be built to adapt to the crown prediction needs of slabs of different specifications, and to provide more reliable quality control and optimization solutions for the production process. Therefore, improving the accuracy of crown prediction for slabs of different specifications by combining data-driven approaches with more suitable deep learning algorithms is of great significance for optimizing production processes, improving product quality, and reducing production costs.

[0009] Third, the raw data extracted using the database software ibaAnalyzer provided by steel production suffers from temporal misalignment. Since rolling operations from stand 1 to stand 7 are sequential, and the sampling rates of relevant variables differ significantly between stands, effectively processing multi-stand data to adapt it for model input becomes a crucial issue in both offline training and real-time prediction. Data alignment is paramount in both offline training and real-time prediction, directly impacting model accuracy and stability. To address this challenge, an effective alignment method is needed to ensure data alignment before model input in both offline training and real-time prediction. During data acquisition, data collected by sensors often exhibits temporal alignment. However, our goal is to predict the convexity of the slab in the spatial dimension. Therefore, an innovative alignment method is required to align the raw sampled data in the spatial dimension. This alignment design is critical, requiring consideration not only of temporal and spatial data alignment but also of data sampling differences between stands and the sequential nature of the process flow. By employing carefully designed alignment methods, it can be ensured that data from each stage of the production process can be fully utilized, thereby improving the model's predictive accuracy and the stability of the production process.

[0010] Fourth, traditional convexity prediction models are mostly trained and validated offline based on historical data. However, in actual production processes, data flows in real time. Therefore, it is of great significance to perform real-time predictions on a prototype system using the trained model. This requires addressing the challenge of high-speed data flow in the steel production process. Summary of the Invention

[0011] This invention provides a method for real-time prediction of crown during hot strip rolling, which at least partially solves the aforementioned technical problems existing in the prior art.

[0012] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0013] On one hand, the present invention provides a method for real-time prediction of crown during hot strip rolling, comprising:

[0014] Acquire historical data of the hot strip rolling process; wherein, the historical data includes the crown of the slab after finishing rolling and process variable data generated by each finishing mill stand that will affect the crown of the slab after finishing rolling;

[0015] The acquired historical data is spatially aligned to obtain aligned historical data.

[0016] A distributed convexity prediction model is established and trained using aligned historical data. The distributed convexity prediction model consists of a seven-layer network, with each layer corresponding to a finishing mill stand. Each layer takes the process variable data generated by the corresponding stand as input and extracts the feature data of the corresponding process variable data. Finally, the feature data extracted by each layer are combined to achieve convexity prediction.

[0017] Deploy the trained distributed convexity prediction model;

[0018] Real-time data of the hot strip rolling process is acquired, and the acquired real-time data is aligned in spatial dimension to obtain aligned real-time data; wherein, the real-time data includes process variable data generated by each finishing mill stand; the aligned real-time data is input into the deployed distributed crown prediction model to complete crown prediction.

[0019] Furthermore, the data alignment in the spatial dimension includes:

[0020] Calculate the percentage of position of the data corresponding to each stand during the entire stand rolling process;

[0021] The data for each rack is aligned in real time based on the percentage of the data's location.

[0022] Furthermore, the formula for calculating the percentage of position of the data corresponding to each stand during the entire stand rolling process is as follows:

[0023]

[0024] in, The starting length refers to the entry length before the first finishing mill stand is rolled, and the starting thickness refers to the entry thickness before the first finishing mill stand is rolled.

[0025] Furthermore, each layer of the distributed convexity prediction model is composed of a variational autoencoder;

[0026] By inputting the process variable data generated by each rack into the variational autoencoder of the corresponding layer, the latent variables of the process variable data of the corresponding rack are obtained. In the first six layers of the distributed convexity prediction model, an auxiliary prediction value is generated by a decoder. The goal of the auxiliary prediction value is to guide the learning by minimizing the mean square error between the predicted exit convexity and the actual exit convexity. At the same time, the seventh layer generates its own latent variables and receives the latent variables generated by the first six layers for decoding and prediction to achieve the final convexity prediction.

[0027] Furthermore, the loss function of the first six layers of the distributed convexity prediction model is expressed as follows:

[0028]

[0029] In the formula, y represents the auxiliary prediction value constructed by the decoder for the i-th instance; i denoted as the actual exit convexity value corresponding to the i-th instance; N is the number of samples.

[0030] Furthermore, the loss function of the seventh layer of the distributed convexity prediction model is expressed as:

[0031]

[0032] In the formula, MSE loss7 D represents the mean squared error of the seventh-level prediction. KL (q(z|x)||p(z)) represents the KL divergence between the approximate posterior distribution and the prior distribution of the latent variables; β is the adjusted MSE and D KL (q(z|x)||p(z)) is a hyperparameter of the weights.

[0033] Furthermore, the real-time prediction method for the crown of the hot strip rolling process is implemented using a cloud-edge-end collaborative architecture; the cloud-edge-end collaborative architecture includes a cloud side, an edge side, and an end side; wherein, the end side is equipped with a data acquisition system; the edge side is equipped with a real-time database, a real-time feature collaboration database, a model parameter database, and a server; and the cloud side is equipped with a historical database and a model training system.

[0034] The end-side data acquisition system is used to collect historical and real-time data of the hot strip rolling process, storing the collected historical data in the cloud-side historical database and the collected real-time data in the edge-side real-time database. The cloud-side model training system is used to train the distributed convexity prediction model based on the historical data in the historical database. The trained distributed convexity prediction model is deployed on the edge-side server. The edge-side model parameter database is used to store the model parameters of the distributed convexity prediction model. The server is used to read the real-time data in the real-time database and realize convexity prediction through the deployed distributed convexity prediction model. The feature data extracted by the first six layers of the distributed convexity prediction model is stored in the real-time feature collaboration database. The seventh layer of the distributed convexity prediction model extracts the feature data extracted by the first six layers from the real-time feature collaboration database and combines it with its own extracted feature data to realize convexity prediction.

[0035] Furthermore, upstream, midstream, and downstream servers are deployed on the side. Each server is responsible for processing data from different mill stands. The upstream server processes data from the first, second, and third finishing mill stands. It reads the data from these mill stands in real time from the data acquisition system and performs feature extraction and writing operations through the network of the corresponding layer of the distributed convexity prediction model. The midstream server processes data from the fourth, fifth, and sixth finishing mill stands. It reads the data from these mill stands in real time from the data acquisition system and performs feature extraction and writing operations through the network of the corresponding layer of the distributed convexity prediction model. The downstream server reads the data from the seventh mill stand collected by the data acquisition system every 50 milliseconds, as well as the relevant feature data written by the upstream and midstream servers in the real-time feature collaboration database. It then performs convexity prediction based on the real-time acquired data, performing a convexity prediction operation every 0.5% of the slab length to ensure timely tracking and convexity prediction of the slab.

[0036] Furthermore, the model parameter database is implemented using a MySQL database to read and write model parameters on different servers, ensuring real-time synchronization and updates of model parameters. Meanwhile, the historical database, real-time database, and real-time feature collaboration database are all implemented using a pSpace database to enable the playback of historical and real-time data, as well as the writing and reading of feature data, providing data support for the model's real-time prediction.

[0037] Furthermore, after the distributed convexity prediction model is deployed, the method further includes:

[0038] The deployed distributed convexity prediction model is used to predict the convexity of slabs of different specifications and steel grades.

[0039] The prediction results are displayed dynamically, and the mean square error of the prediction results for each slab is calculated.

[0040] Slabs with a mean square error exceeding a preset threshold are selected for partial parameter retraining and model updates; wherein, the partial parameters include the final output weights and biases of the seventh layer network of the distributed convexity prediction model.

[0041] In another aspect, the present invention also provides an electronic device comprising a processor and a memory; wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the above-described method.

[0042] In another aspect, the present invention also provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the above method.

[0043] The beneficial effects of the technical solution provided by this invention include at least the following:

[0044] 1. This invention designs a distributed crown prediction model tailored to the data distribution characteristics of the finishing mill process. This model utilizes a variational autoencoder (VAE) to achieve regression prediction of the exit crown of multiple stands. This invention leverages the variational autoencoder to achieve deep mining of feature data. By extracting and fusing latent variables, the accuracy of the final prediction is effectively improved. This deep mining technique enables the model to better understand the inherent characteristics and correlations of the data, thereby more accurately predicting the exit crown and providing strong support for quality control and optimization of the production process. Simultaneously, based on the actual conditions of the production site, this invention divides the variables into stands, establishes a distributed crown prediction model, and achieves accurate crown prediction results. The innovation of this method lies in comprehensively considering the data distribution characteristics of the finishing mill process and the needs of the actual production scenario. Through distributed modeling and deep feature mining, it achieves a significant improvement in the accuracy of crown prediction, bringing a major technological breakthrough to quality management and efficiency improvement in the steel production process.

[0045] 2. This invention designs an alignment method based on position percentage. Based on the entry length and thickness of the workpiece, and the exit thickness of each workpiece after rolling through different stands, the position percentage of real-time data during the entire stand rolling process is calculated using the real-time transmission length variable. Finally, during prediction, real-time alignment is performed based on the position percentage of data from different stands. This position percentage-based alignment method considers the positional changes of the workpiece throughout the production process and determines the position percentage of real-time data during rolling accordingly. By performing real-time alignment based on the position percentage of data from different stands, the convexity of each stand can be predicted more accurately, achieving spatial alignment of data. This innovative method allows the model to better adapt to the data distribution characteristics in actual production scenarios, improving the accuracy and stability of the convexity prediction model and providing effective technical support for real-time convexity prediction.

[0046] 3. This invention achieves distributed deployment of the program by deploying the distributed model on three servers: upstream, midstream, and downstream. Furthermore, it utilizes a MySQL database on different servers to read and write model parameters, ensuring real-time synchronization and updates of these parameters. Simultaneously, two pSpace databases are used to enable real-time data playback and feature writing / reading, providing crucial data support for real-time model prediction. This approach also better aligns with the data distribution in production processes. The combined application of these measures enables the method of this invention to effectively address the real-time nature and diversity of data in actual production processes, ensuring efficient model operation and accurate prediction, and providing reliable technical support for production process optimization and quality control. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart of the real-time prediction method for crown of hot strip rolling process provided in the embodiments of the present invention;

[0049] Figure 2 This is a schematic diagram of the spatial dimension data alignment method provided in an embodiment of the present invention;

[0050] Figure 3 This is a distributed convexity prediction model framework based on variational autoencoders provided in the embodiments of the present invention;

[0051] Figure 4 This is an offline distributed model convexity prediction error diagram provided in an embodiment of the present invention;

[0052] Figure 5 This is the convexity prediction graph of the offline distributed model provided in the embodiments of the present invention;

[0053] Figure 6 This is a comparison diagram of local parameter updates on the side provided in an embodiment of the present invention;

[0054] Figure 7 This is a schematic diagram illustrating the use of pSpace and the MySQL database provided in an embodiment of the present invention;

[0055] Figure 8 This is the rolling schedule of the prototype system provided in the embodiments of the present invention;

[0056] Figure 9 This is a schematic diagram of the overall architecture for real-time convexity prediction cloud-edge-device collaboration provided in an embodiment of the present invention;

[0057] Figure 10 This is a system block diagram of the electronic device provided in the embodiments of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0059] First, it should be noted that in the embodiments of the present invention, the words "exemplarily," "for example," etc., are used to indicate that they are examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the term "exemplarily" is intended to present the concept in a specific manner. Furthermore, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either one or the other.

[0060] First Embodiment

[0061] To address the problems existing in the prior art, this embodiment provides a method for real-time prediction of crown during hot strip rolling, so as to at least partially solve the aforementioned technical problems existing in the prior art.

[0062] To address the first and second issues, this method designs a distributed crown prediction model tailored to the data distribution characteristics of the finishing rolling process. This model utilizes a variational autoencoder (VAE) to achieve regression prediction of multi-stand exit crown. This method leverages the VAE to achieve deep mining of feature data. By extracting and fusing latent variables, the accuracy of the final prediction is effectively improved. This deep mining technique enables the model to better understand the inherent characteristics and correlations of the data, thereby more accurately predicting exit crown and providing strong support for quality control and optimization of the production process. Furthermore, in actual production, the slab needs to pass through seven finishing rolling stands to reach the exit for crown measurement. Therefore, the process variables affecting crown are not generated simultaneously but are distributed across different stands. Traditional crown prediction algorithms do not differentiate between these variables but instead combine variables such as rolling force and bending roll force from multiple stands for model training and index prediction, which does not reflect reality. Therefore, this study, based on the actual production conditions, divides the variables into stand segments, establishes a distributed crown prediction model, and achieves accurate crown prediction results. The innovation of this method lies in its comprehensive consideration of the data distribution characteristics of the finishing rolling process and the actual production scenario requirements. Through distributed modeling and deep feature mining, it has achieved a significant improvement in the accuracy of convexity prediction, bringing a major technological breakthrough to the quality management and efficiency improvement of the steel production process.

[0063] To address the third issue, this method employs a position percentage-based alignment approach. Based on the workpiece's entry length and thickness, and the exit thickness after each stand's rolling, the method calculates the position percentage of real-time data throughout the entire stand's rolling process using the real-time transfer length variable. During final prediction, real-time alignment is performed based on the position percentages of data from different stands. This position percentage-based alignment considers the workpiece's positional changes throughout the production process and determines the position percentage of real-time data during rolling accordingly. By performing real-time alignment based on the position percentages of data from different stands, the convexity of each stand can be predicted more accurately, achieving spatial alignment of the data. This innovative method allows the model to better adapt to the data distribution characteristics in actual production scenarios, improving the accuracy and stability of the convexity prediction model and providing effective technical support for real-time convexity prediction.

[0064] To address the fourth issue, this method employs multiple measures. First, distributed deployment of the program is achieved by deploying the distributed model on three servers: upstream, midstream, and downstream. Second, MySQL databases are used to read and write model parameters across different servers, ensuring real-time synchronization and updates of model parameters. Simultaneously, two pSpace databases are used to implement real-time data playback and feature writing / reading, providing crucial data support for real-time model prediction. Among these, using the pSpace database for real-time data playback and feature writing / reading is the most important and fundamental step. Parameter reading and local parameter updates are performed using the MySQL database, which better aligns with the data distribution in the production workflow.

[0065] During the prediction process, the real-time data is aligned by calculating the position percentage of each data point, and then convexity is predicted. Simultaneously, the prediction results are dynamically displayed, and the mean square error of each slab prediction result is calculated. If the deviation is large, some parameters are retrained to ensure the model's accuracy and stability. The retrained parameters are then written back to the MySQL database to ensure real-time updates of the model parameters.

[0066] The combined application of these measures enables this method to effectively address the real-time and diverse nature of data in actual production processes, ensuring the efficient operation and accurate prediction of the model, and providing reliable technical support for the optimization of production processes and quality control.

[0067] The real-time prediction method for the crown of the hot strip rolling process is implemented using a cloud-edge-end collaborative architecture, such as... Figure 9As shown, the cloud-edge-device collaborative architecture adopted in this embodiment includes a cloud side, an edge side, and a device side; wherein, the device side is equipped with a data acquisition system; the edge side is equipped with a real-time database, a real-time feature collaboration database, a model parameter database, and a server; and the cloud side is equipped with a historical database and a model training system.

[0068] The end-side data acquisition system is used to collect historical and real-time data of the hot strip rolling process, storing the collected historical data in the cloud-side historical database and the collected real-time data in the edge-side real-time database. The cloud-side model training system is used to train the distributed convexity prediction model based on the historical data in the historical database. The trained distributed convexity prediction model is deployed on the edge-side server. The edge-side model parameter database is used to store the model parameters of the distributed convexity prediction model. The server is used to read the real-time data in the real-time database and realize convexity prediction through the deployed distributed convexity prediction model. The feature data extracted by the first six layers of the distributed convexity prediction model is stored in the real-time feature collaboration database. The seventh layer of the distributed convexity prediction model extracts the feature data extracted by the first six layers from the real-time feature collaboration database and combines it with its own extracted feature data to realize convexity prediction.

[0069] Furthermore, the servers deployed on the side include upstream servers, midstream servers, and downstream servers; each server is responsible for processing data from different racks. The upstream server processes data from the first, second, and third racks, reading the data from these racks in real-time and performing feature extraction and writing operations through the corresponding layer of the distributed convexity prediction model network. The midstream server processes data from the fourth, fifth, and sixth racks, reading the data from these racks in real-time and performing feature extraction and writing operations through the corresponding layer of the distributed convexity prediction model network. The downstream server's task is to read data from the seventh rack every 50 milliseconds and the relevant feature data written by the upstream and midstream servers into the real-time feature collaboration database, and then perform convexity prediction to ensure timely tracking and convexity prediction of the slab.

[0070] The research data in this embodiment comes from the strip rolling process of a steel company. Based on the above, the execution flow of the real-time prediction method for the crown of the hot strip rolling process is as follows: Figure 1 As shown, it includes the following steps:

[0071] S1, acquire historical data of the hot strip rolling process;

[0072] The historical data includes the crown of the slab after finishing and the process variable data generated by each finishing mill stand that will affect the crown of the slab after finishing; for example, the process variable data may include the rolling force, bending roll force, wear, thermal expansion, incoming material thickness, incoming material length, roll diameter, rolling speed, roll shifting amount, etc. of each mill stand.

[0073] S2, perform spatial alignment on the acquired historical data to obtain aligned historical data;

[0074] It should be noted that data processing is performed before model design. This embodiment processes historical data, aligning it by position percentage. Each stand is divided according to whether rolling force has started. The position percentage of each data point is obtained by dividing the pass length of each stand by the exit length. Then, the data from different stands are aligned based on the position percentage. The finally obtained offline aligned data is then processed and input into the model. The pass length is a known variable that gradually increases from the stand inlet length to the total outlet length, equivalent to the slab length at that moment. Therefore, it is used to calculate the position percentage. Specifically, the formula for calculating the position percentage of each data point is as follows:

[0075]

[0076] The aforementioned position percentage calculation formula is based on the ratio of the actual transfer length of each rolling stand to the exit length of the entire rolling process. This calculation determines the relative position of each data point throughout the rolling process, thereby enabling the alignment of data from different stands.

[0077] In real-time prediction, the data exit length is not a directly obtainable variable. Instead, it needs to be estimated using the inlet thickness and length, as well as the exit thickness of each stand, to predict the rolling length through each stand. The calculation formula is shown below:

[0078]

[0079] The starting length and starting thickness are determined by using the inlet length before F1 rolling as the initial length and thickness. Based on the target exit thickness of stands F1 through F7, the theoretical total length of the steel plate passing through stands F1 through F7 is calculated. Then, the percentage index of the position corresponding to each data point is calculated by real-time detection of the transmitted length variable.

[0080] Furthermore, since the various endpoints cannot communicate during real-time prediction, a positioning percentage needs to be pre-set so that the upstream and midstream endpoints can write the required data features into the pSpace database, which is then read by the downstream endpoint. The fixed percentage ranges from 5% to 95%, with equal intervals of 0.5%. Considering that the transmission length of variables at the current time and previous time may not always be exactly the same as the set percentage, a tolerance of 0.25% is left.

[0081] In addition to the corresponding data received from the receiving end, the downstream server will also read real-time features extracted from the upstream and midstream from the pSpace database and align them according to the percentage index calculated for each rack. This completes the alignment of real-time data and real-time prediction. The specific alignment method is as follows: Figure 2 As shown.

[0082] S3. Establish a distributed convexity prediction model and train the distributed convexity prediction model using aligned historical data; wherein, the distributed convexity prediction model contains a seven-layer network, each layer of the network corresponds to a finishing mill stand, each layer of the network takes the process variable data generated by the corresponding stand as input, and extracts the feature data of the corresponding process variable data; finally, the feature data extracted by each layer of the network are combined to realize convexity prediction.

[0083] It should be noted that the above steps describe the process of selecting model parameters, training the distributed model, and evaluating performance metrics to establish a distributed convexity prediction model. To address the challenge of convexity prediction data distribution, this embodiment designs a distributed VAE framework specifically tailored for convexity prediction in the strip steel finishing rolling process. The entire framework is systematically divided into seven distinct layers, each corresponding to one of the seven mill stands involved in the finishing rolling process. The ultimate goal of this method is to extract latent variables associated with each specific layer. The latent variables obtained from each layer are then provided to the final decoding and prediction stages. This method ensures efficient distributed convexity prediction across continuous mill stands during the rolling process.

[0084] This embodiment designs a distributed prediction model based on a variational autoencoder. The model is improved to better meet our target requirements and enhance prediction accuracy. The model improvement focuses on two main aspects:

[0085] Decoder Improvement: In the first six layers, this improvement aims to generate auxiliary predictions through the decoder. Its purpose is to supervise and guide the learning process, ensuring that the latent variable z extracted from each layer of the VAE model substantially contributes to the accurate prediction of the final convexity. Simultaneously, the seventh layer model can generate its own latent variables, receiving the latent variables from the first six layers for decoding and prediction.

[0086] The loss function has been redefined: This improvement relates to the design of the final loss function. The original reconstruction error of the VAE is discarded, focusing instead on regression prediction. The loss function ultimately combines the mean squared error (MSE) of the label loss with the KL (Kullback-Leibler) divergence, maintaining the effectiveness of the latent space while achieving regression prediction.

[0087] The finishing rolling process includes seven different mill stands, each corresponding to a layer in the VAE model. For the first six layers, each layer's model is configured to use the operating variables of that mill stand as input and extract latent variables z through an encoding process. To supervise the learning process between these layers and ensure that the generated latent variables z are meaningful for predicting the final exit crown, an auxiliary prediction value can be formulated using a decoder. The goal of this auxiliary value is to guide the learning by minimizing the mean squared error between the predicted and actual exit crown. The loss function for the first six layers is calculated in the same way, which is the mean squared error (MSE) of the auxiliary prediction. Specifically, the loss function for the first six layers can be expressed as:

[0088]

[0089] In the formula, y represents the auxiliary prediction value constructed by the decoder for the i-th instance; i is the corresponding actual convexity value; N is the number of samples.

[0090] For the seventh-layer VAE model, the process involves encoding the operational variables of the seventh layer to learn the latent variable z, and decoding it by fusing the latent variables from the first six layers together as the latent variable for prediction. The final loss function combines the label loss MSE. loss7 The KL divergence is used to ensure that the final fused latent variable z effectively reflects the cumulative effect of all racks. Therefore, the loss function can be expressed as:

[0091]

[0092] In the formula, MSE loss7 This represents the mean squared error of the predictions for the seventh layer, and its calculation formula is the same as that for the first six layers, namely: D KL (q(z|x)||p(z)) represents the KL divergence between the approximate posterior distribution and the prior distribution of the latent variable, and β is the adjusted MSE. loss7 and D KL The hyperparameters for the weights of the two components (q(z|x)||p(z)) are...

[0093] In general, VAE models can generate latent variables through learning. These latent variables contain deep relationship features between variables. Applying the VAE algorithm to each rack, by inputting the relevant variables of that rack, the VAE model is trained to obtain the latent variables for that layer. Finally, the latent variables z obtained from each layer (i.e., each rack) are combined for final decoding and prediction. This achieves distributed convexity prediction across multiple racks. The overall framework is as follows: Figure 3 As shown.

[0094] In the verification phase, this embodiment uses the trained model to predict the crown of slabs of 10 different specifications and steel grades. It can be seen that there is no significant overall deviation. The offline training and prediction results are as follows: Figure 4 , Figure 5 As shown. However, the prediction accuracy still needs improvement. Therefore, in subsequent deployment of the prototype system for real-time prediction, to improve prediction accuracy, local parameter retraining was performed on the slab with large prediction deviations. Specifically, 256 parameters, including the final output weights and biases of the last layer of the VAE model, were selected for retraining and model updating. Offline verification showed that this measure significantly improved the model's prediction accuracy, as detailed in the following figures. Figure 6 As shown in the figure. The results show that after updating the local parameters of the model, the prediction accuracy for the two new slabs is improved. Therefore, this improvement measure can be applied to the real-time prediction process.

[0095] S4, Deploy the trained distributed convexity prediction model;

[0096] It should be noted that, in the actual deployment of the prototype system, this embodiment mainly uses the pSpace database and the MySQL database to complete the construction and communication of the entire system. In this embodiment, the MySQL database is used to read model parameters and update some parameters on multiple servers; the pSpace database is used to realize real-time reading and writing of slab data, and to complete feature extraction and final convexity prediction.

[0097] Specific usage examples are as follows: Figure 7 As shown, the entire deployment process consists of the following steps.

[0098] First, pSpace was used to implement real-time playback of historical data. The purpose of this step was to simulate real-time data flow, providing a foundation for the deployment of the prototype system's edge model. By writing program code, historical data was written cyclically every 50 milliseconds to ensure the continuity and realism of the data flow.

[0099] Based on real-time data flow, this embodiment divides the data from the seven racks into three servers (upper, middle, and lower) for edge deployment. Each server is responsible for processing data from its respective rack and simulating and predicting the rolling process in real time according to a specific schedule. The rolling schedule and retraining strategy for these servers are as follows: Figure 8 As shown, it details the data reading and prediction process for each rack at different time points, as well as the local parameter retraining for slabs with large prediction deviations.

[0100] Specifically, the upstream server is responsible for processing data from the first, second, and third racks. It reads the data from these racks in real time and performs feature extraction and writing operations using the corresponding layer's model. Feature extraction involves extracting key feature information—latent variables—from the corresponding rack data using the VAE model at each layer, and then using it for subsequent convexity prediction. These features are written to the pSpace database at intervals of 0.5% of the slab length so that the downstream server can perform real-time predictions.

[0101] The midstream server is also responsible for real-time data processing, but it processes data from the fourth, fifth, and sixth racks. It performs data reading and feature writing operations in the same way to ensure the continuity of the entire rolling process and the integrity of data flow.

[0102] The downstream server's task is to read data from rack seven and relevant features written by upstream and midstream servers in the pSpace library every 50 milliseconds, and then perform convexity prediction. It uses real-time acquired data for prediction, performing a convexity prediction operation every 0.5% of the slab length. This frequency ensures timely tracking and prediction of the slab.

[0103] After real-time prediction is completed, the data is collected to calculate the mean absolute error (MAE) of the overall prediction results for each slab. If the MAE is large, the slab data is used to retrain some parameters of the last layer of the model, and this retrained data is applied to the real-time prediction of the next slab. Simultaneously, the retrained parameters are written back into the MySQL database.

[0104] The final cloud-edge-device collaborative architecture is as follows: Figure 9 As shown, it covers the entire process from data processing to prediction, ensuring the stability and effectiveness of the entire system. The hardware and software configuration information for the cloud-edge-device system is shown in Table 1.

[0105] Table 1 Prototype System Hardware and Software Configuration

[0106]

[0107]

[0108] S5: Acquire real-time data of the hot strip rolling process, perform spatial dimension data alignment on the acquired real-time data to obtain aligned real-time data; wherein, the real-time data includes process variable data generated by each finishing mill stand; input the aligned real-time data into the deployed distributed crown prediction model to complete crown prediction.

[0109] In summary, this embodiment provides a real-time crown prediction method for hot strip rolling processes based on cloud-edge-device collaboration. A distributed crown prediction model is designed to address the data distribution characteristics of the finishing rolling process. A Variational Autoencoder (VAE) is used to achieve regression prediction of multi-stand exit crown, effectively improving the accuracy of the final prediction. Simultaneously, this method designs an alignment method based on position percentage. Based on the entry length and thickness of the rolled piece, and the exit thickness of each piece after rolling on different stands, the position percentage of real-time data throughout the entire stand rolling process is calculated using the real-time transmission length variable. During final prediction, real-time alignment is performed based on the position percentage of data from different stands. By performing real-time alignment based on the position percentage of data from different stands, the crown of each stand can be predicted more accurately, achieving spatial alignment of the data. Furthermore, this method achieves distributed deployment of the program by deploying the distributed model on three servers: upstream, midstream, and downstream. Moreover, the model parameters are read and written using a MySQL database on different servers, ensuring real-time synchronization and updates of the model parameters. Meanwhile, two pSpace databases were used to enable real-time data playback and feature writing and reading, providing crucial data support for the model's real-time prediction. This also better aligns with the data distribution in the production process.

[0110] Second Embodiment

[0111] This embodiment provides an electronic device, such as... Figure 10 As shown, the electronic device includes a processor and a memory; wherein the processor and the memory can be connected via a communication bus; the memory stores at least one instruction, which is loaded and executed by the processor to implement the method of the first embodiment described above. Furthermore, the electronic device may also include a transceiver, the processor and the transceiver can be connected via a communication bus, and the transceiver is used to communicate with other devices.

[0112] Below, in conjunction with Figure 10 A detailed introduction to each component of this electronic device is provided below:

[0113] The processor is the control center of the electronic device. The electronic device may include multiple processors, each of which can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The term "processor" can refer to a single processor or a collective term for multiple processing elements. For example, a processor can be one or more central processing units (CPUs), other general-purpose processors, application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), one or more field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor can perform various functions of the electronic device by running or executing software programs stored in memory and by calling data stored in memory.

[0114] In a specific implementation, as one example, the processor may include one or more CPUs, for example... Figure 10 CPU0 and CPU1 shown are, of course, merely illustrative examples.

[0115] The memory is used to store the software program that executes the solution of the present invention, and the processor controls its execution. For specific implementation methods, please refer to the above method embodiments, which will not be repeated here.

[0116] Optionally, the memory may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory may be integrated with the processor or may exist independently, and may be accessed through the interface circuit of the electronic device (…). Figure 10 (Not shown in the image) is coupled to the processor; however, this embodiment of the invention does not impose specific limitations on this.

[0117] The transceiver may include a receiver and a transmitter. Figure 10 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function. The transceiver can be integrated with the processor or exist independently, and is connected through the interface circuit of the electronic device (…). Figure 10 (Not shown in the image) is coupled to the processor, and this embodiment of the invention does not specifically limit this.

[0118] In addition, it should be noted that, Figure 10 The structure of the electronic device shown is not intended to limit the device. Actual devices may include more or fewer components than shown, or combine certain components, or have different component arrangements. Furthermore, the technical effects achieved by this electronic device when performing the method of the first embodiment described above can be referenced to the technical effects described in the first embodiment; therefore, they will not be repeated here.

[0119] Third Embodiment

[0120] This embodiment provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the method of the first embodiment described above. The computer-readable storage medium may be a ROM, random access memory, CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc. The instruction stored therein can be loaded and executed by a processor in a terminal.

[0121] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely or partially hardware embodiment, a completely or partially software embodiment, or an embodiment combining software and hardware aspects. Moreover, when implemented in software, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any usable medium accessible to a computer or a data storage device such as a server or data center containing one or more sets of usable media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive (SSD).

[0122] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0123] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0124] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element. Furthermore, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone, where A and B can be singular or plural. Additionally, the character " / " in this text generally indicates an "or" relationship between the preceding and following objects, but it can also indicate an "AND / OR" relationship. Please refer to the context for specific interpretations. "At least one" refers to one or more items, while "more than" refers to two or more items. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can be represented as: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0125] Furthermore, it is understood that in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0126] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0127] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of functional modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms. Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs. Additionally, the functional units in the various embodiments of this invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0128] If the method is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0129] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments of the present invention have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make several improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A method for real-time prediction of crown during hot strip rolling, characterized in that, include: Acquire historical data of the hot strip rolling process; wherein, the historical data includes the crown of the slab after finishing rolling and process variable data generated by each finishing mill stand that will affect the crown of the slab after finishing rolling; The acquired historical data is spatially aligned to obtain aligned historical data. A distributed convexity prediction model is established and trained using aligned historical data. The distributed convexity prediction model consists of a seven-layer network, with each layer corresponding to a finishing mill stand. Each layer takes the process variable data generated by the corresponding stand as input and extracts the feature data of the corresponding process variable data. Finally, the feature data extracted by each layer are combined to achieve convexity prediction. Deploy the trained distributed convexity prediction model; Real-time data of the hot strip rolling process is acquired, and the acquired real-time data is aligned in spatial dimension to obtain aligned real-time data; wherein, the real-time data includes process variable data generated by each finishing mill stand; the aligned real-time data is input into the deployed distributed crown prediction model to complete crown prediction; Each layer of the distributed convexity prediction model consists of a variational autoencoder. By inputting the process variable data generated by each rack into the variational autoencoder of the corresponding layer, the latent variables of the process variable data of the corresponding rack are obtained. In the first six layers of the distributed convexity prediction model, an auxiliary prediction value is generated by a decoder. The goal of the auxiliary prediction value is to guide the learning by minimizing the mean square error between the predicted exit convexity and the actual exit convexity. At the same time, the seventh layer generates its own latent variables and receives the latent variables generated by the first six layers for decoding and prediction to achieve the final convexity prediction. The loss function of the first six layers of the distributed convexity prediction model is expressed as follows: ； In the formula, This represents the auxiliary prediction value constructed by the decoder for the i-th instance; The actual exit convexity value corresponding to the i-th instance; N is the number of samples; The loss function of the seventh layer of the distributed convexity prediction model is expressed as: ； In the formula, This represents the mean squared error of the prediction at the seventh layer; This represents the KL divergence between the approximate posterior distribution and the prior distribution of the latent variables. It is an adjustment and The hyperparameters of the weights.

2. The method for real-time prediction of crown during hot strip rolling as described in claim 1, characterized in that, The spatial dimension data alignment includes: Calculate the percentage of position of the data corresponding to each stand during the entire stand rolling process; The data for each rack is aligned in real time based on the percentage of the data's location.

3. The method for real-time prediction of crown during hot strip rolling as described in claim 2, characterized in that, The formula for calculating the percentage of position of the data corresponding to each stand during the entire stand rolling process is as follows: ； in, The starting length refers to the entry length before the first finishing mill stand is rolled, and the starting thickness refers to the entry thickness before the first finishing mill stand is rolled.

4. The method for real-time prediction of crown during hot strip rolling as described in claim 1, characterized in that, The real-time prediction method for the crown of the hot strip rolling process is implemented using a cloud-edge-end collaborative architecture. The cloud-edge-end collaborative architecture includes a cloud side, an edge side, and an end side. The end side is equipped with a data acquisition system. The edge side is equipped with a real-time database, a real-time feature collaboration database, a model parameter database, and a server. The cloud side is equipped with a historical database and a model training system. The end-side data acquisition system is used to collect historical and real-time data of the hot strip rolling process, storing the collected historical data in the cloud-side historical database and the collected real-time data in the edge-side real-time database. The cloud-side model training system is used to train the distributed convexity prediction model based on the historical data in the historical database. The trained distributed convexity prediction model is deployed on the edge-side server. The edge-side model parameter database is used to store the model parameters of the distributed convexity prediction model. The server is used to read the real-time data in the real-time database and realize convexity prediction through the deployed distributed convexity prediction model. The feature data extracted by the first six layers of the distributed convexity prediction model is stored in the real-time feature collaboration database. The seventh layer of the distributed convexity prediction model extracts the feature data extracted by the first six layers from the real-time feature collaboration database and combines it with its own extracted feature data to realize convexity prediction.

5. The method for real-time prediction of crown during hot strip rolling as described in claim 4, characterized in that, Upstream, midstream, and downstream servers are deployed on the side. Each server is responsible for processing data from different mill stands. The upstream server processes data from the first, second, and third finishing mill stands. It reads data from these mill stands in real-time from the data acquisition system and performs feature extraction and writing operations through the corresponding layer of the distributed convexity prediction model network. The midstream server processes data from the fourth, fifth, and sixth finishing mill stands. It reads data from these mill stands in real-time from the data acquisition system and performs feature extraction and writing operations through the corresponding layer of the distributed convexity prediction model network. The downstream server reads data from the seventh mill stand collected by the data acquisition system every 50 milliseconds, as well as relevant feature data written by the upstream and midstream servers in the real-time feature collaboration database. It then performs convexity prediction based on the real-time acquired data, performing a convexity prediction operation every 0.5% of the slab length to ensure timely tracking and convexity prediction of the slab.

6. The method for real-time prediction of crown during hot strip rolling as described in claim 5, characterized in that, The model parameter database is implemented using a MySQL database and is used to read and write model parameters on different servers to ensure real-time synchronization and updates of model parameters. Meanwhile, the historical database, real-time database, and real-time feature collaboration database are all implemented using a pSpace database to enable the playback of historical and real-time data as well as the writing and reading of feature data, providing data support for real-time prediction of the model.

7. The method for real-time prediction of crown during hot strip rolling as described in claim 1, characterized in that, After the distributed convexity prediction model is deployed, the method further includes: The deployed distributed convexity prediction model is used to predict the convexity of slabs of different specifications and steel grades. The prediction results are displayed dynamically, and the mean square error of the prediction results for each slab is calculated. Slabs with a mean square error exceeding a preset threshold are selected for partial parameter retraining and model updates; wherein, the partial parameters include the final output weights and biases of the seventh layer network of the distributed convexity prediction model.

Images